Anal. Chem. 1988, 60,1953-1958
1953
Determination of Drug Substances in Biological Fluids by Direct Injection Multidimensional Liquid Chromatography with a Micellar Cleanup and Reversed-Phase Chromatography J o s e p h V. Posluszny* a n d Robert Weinberger
Applied Biosystems, Inc., 170 Williams Drive, Ramsey, New Jersey 07446
A multldlmenslonai high-pressure liquid chromatographic procedure has been developed for the determlnation of drug substances In blood serum or plasma employing direct injection. The first chromatographic dlmension, which provides for sample extraction and cleanup, employs a micellar mobile phase wlth sodlum dodecyi sulfate as the modifier. The second dlmenslon, coupled on-line to the first, utliizes conventional reversed-phase medla and modiflers for the analytlcal separatlon. The general appllcabillty of this method is illustrated wlth applications for determining carbamazepine, chioramphenlcoi, and procainamide wlth absorption detection. Fluorescence detection is employed for determining furosemide, qulnidlne, and propranolol. Ail of these drugs gave recoveries of 96-104% and relative standard deviations of 1-6 %. Blind correlation studies comparing this method to manual liquld chromatographic methods for propranolol and chloramphenicolshowed no slgnlficant dlfferences. The llmlts of detection were 50-100 ng/mL for UV applications and 4-10 ng/mL with fluorescence detection. Determinations can be completed at a rate of about three to five per hour. Column longevlty is excellent.
chromatographic approach improves column lifetime and provides a mechanism for release of strongly protein bound drugs (15, 16). Micellar chromatography shows moderate efficiency due to poor mass transfer ( I 7-19). With large injection volumes, the inefficiency is amplified. As with any singular dimension approach with a small injection volume and the lack of an enrichment step, the sensitivity of the determination is inadequate for many applications. The multidimensional approach reported here makes use of the advantages of micellar chromatography (direct serum/plasma injection, extended column life and drug recovery) operating on a consumable “extraction column” that can last for several hundred injections. Unlike the ISRP approach, this technique employs the full range of conventional bonded-phase supports. Since we have shown that micellar mobile phases form homogeneous solutions with many reversed-phase solvents and buffers, the two chromatographic dimensions are mutually compatible. The entire process is automated and initiated by either manual or automated injection. By combination of the micellar mobile phase with a complementary set of silica-based bonded phases a protocol for performing methods development of the extraction cycle becomes straightforward.
The determination of drugs in biological fluids has been greatly enhanced through the development of liquid chromatography (LC) technology (1). The most straightforward method for sample preparation typically uses liquid/liquid extraction to recover and concentrate the drug and remove endogenous proteins that can be harmful to the column packing. In recent years the use of solid phase extraction on disposable cartridges has done much to increase laboratory throughput and replace many liquid-liquid extraction procedures (2,3). Both liquid-liquid and solid phase extraction are labor-intensive operations. Ideally, the resolving power of LC could be employed to prepare or cleanup the sample and perform the analytical separation. The difficulties with direct injection procedures are primarily due to column degradation from irreversible protein adsorption resulting in a decrease in column performance and an increase in back pressure. Specialty packings such as the internal surface reversed-phase support (ISRP) ( 4 , 5 ) have been employed to eliminate protein adsorption. The lack of an enrichment step precludes simple one-dimensional chromatographic separations from providing the required limits of detection for many determinations. Adopting a multidimensional approach to perform direct injection analysis improves sensitivity by allowing large injection volumes and yet permits trace enrichment and peak compression through solvent focusing (4-8). Recently, the use of micellar chromatography has been reported to allow the direct injection of serum or plasma (9-12). Micellar chromatography or “pseudophase chromatography” employs surfactant as mobile phase modifiers (13, 14). For direct serum injection (DSI) the micellar
EXPERIMENTAL SECTION Reagents. Chloramphenicol, carbamazepine, procainamide, and quinidine were obtained from Sigma Chemical Co. (St. Louis, MO). Furosemide and propranolol were obtained from United States Pharmacopoeial Convention, Inc. (Rockville, MD). HPLC grade methanol, acetonitrile, and tetrahydrofuran were from J. T. Baker (Phillipsburg, NJ). Water was purified through an in-house reverse-osmosis system (Hydro Service, Durham, NC). Sodium dodecyl sulfate (SDS),obtained from Aldrich Chemical Co. (Milwaukee, WI), was used as received. Mono- and dibasic sodium phosphate salts were obtained from Sigma, and concentrated phosphoric acid was obtained from J. T. Baker. Blank serum was obtained from Gibco Laboratories (Grand Island, NY). Apparatus. The high-performance liquid chromatography (HPLC) system was a modular system from Applied Biosystems, Inc. (formerly Kratos Analytical),Ramsey, NJ. The configuration of equipment can be seen in Figure 1. Solvent delivery was provided by two Spectroflow 400 pumps fitted with three-way solenoid valves to perform step changes in solvent composition. The pulse dampeners were removed to reduce void volume in the system, thereby facilitating efficient solvent switching. The detectors were a Spectroflow 783 for UV applications and a Spectroflow 980 for fluorescence applications. A six-port column switching valve with electrical actuator was manufactured by Valco (Houston, TX). Data were acquired with either an Applied Biosystems DS650 data system or a Spectra Physics 4290 integrator. Chromatography. All columns used in this study were cartridge columns manufactured by Applied Biosystems, Inc. (formerly Brownlee Laboratories), Santa Clara, CA. The analytical column was a RP-18, Spheri-5,packed with 5-pm spherical porous particles in 4.6 mm i.d. X 220 mm long cartridges. The extraction columns were a variety of Applied Biosystems silica-basedbonded phase packings specially packed in a 10 mm i.d. X 30 mm long
0003-2700/88/0360-1953$01.50/00 1988 American Chemical Society
1954
ANALYTICAL CHEMISTRY, VOL. 60, NO. 18, SEPTEMBER 15, 1988
Table I. Chromatography Conditions Summary drug
extraction phase
carbamazepine chloramphenicol furosemide procainamide propranolol quinidine
C8 C8
F l
valve configuration
CN CN CN
I
WASTE
I MP
r--1
6)
c
I \
detector/settings ABS; 287 nm ABS; 278 nm fluor; 270 nm ex, 400 nm em ABS; 280 nm fluor; 230 nm ex, 340 nm em fluor; 238 nm ex, 340 nm em
C8
1
micellar extraction solvent
I
C
Q Mppp+J M Figure 1. Alternative valve configurations: pump 1, loading pump; pump 2, analytical pump; column 1, extraction column; column 2, analytical column; (M) micellar extraction solvent, (W) extraction column wash solvent, (MP) analytical mobile phase.
cartridge. The phases investigated were standard 5-pm porous materials with an 80-A pore size. The phase used for each application is summarized in Table I. The micellar extraction eluants used were as follows: eluant A, 8/92 acetonitrile/water that contained 40 mM SDS and 40 mM sodium phosphate, monobasic (SPM); eluant B, 4/96 acetonitrile/water containing 20 mM SDS and 20 mM SPM; and eluent C, 14/86 acetonitrilelwater containing 40 mM SDS and 40 mM SPM. The pH of the extraction eluents was adjusted to 7 with a concentrated NaOH solution except for the furosemide application, which was adjusted to 3.5 with phosphoric acid. The wash solvent for the extraction column was 70130 acetonitrile/ water containing 40 mM SDS and 40 mM SPM. The flow rate was maintained at 2.0 mL/min, except for the quinidine application where it was 1.5 mL/min. The analytical mobile phases used for each application were as follows: carbamazepine, 55/45 methanol/water containing 40 mM SDS and 40 mM SPM (pH 3.0); chloramphenicol, 28/72 methanol/water containing 20 mM SPM (pH 4.6); furosemide, 30/70 methanol/water containing 20 mM SDS and 20 mM SPM (pH 3.5); procainamide, 40/60 acetonitrilelwater containing 40 mM SDS and 40 mM SPM (pH 3.0); propranolol, 78/22 methanol/water containing 40 mM SDS, 40 mM SPM, and 0.4 mL of morpholine (pH 3.0); and quinidine, 34/66 tetrahydrofuran/ water with 0.6 mL of acetic acid and 0.15 mL morpholine per liter of solvent (pH 3.5). The pH was adjusted with a concentrated NaOH solution or phosphoric acid as required before the organic
component was added. The flow rate was set at 2.0 mL/min, except for the quinidine application which ran a t 1.5 mL/min. The detector conditions for the UV and fluorescence applications are listed in Table I. For the fluorescence applications, cutoff filters were used with the exception of the furosemide application. The 400-nm band-pass filter used in the furosemide application had a 25-nm bandwidth. Procedure. Stock standard solutions were prepared by dissolving the drug in HPLC grade acetonitrile, methanol, or water. Aliquots of these solutions were diluted in water to prepare the working standard solutions at the necessary concentrations. The same stock solutions were used as a source of the drug for preparing spiked serum samples. The appropriate aliquot of the solution was taken and evaporated to remove the solvent to prevent protein denaturing. A volumetric portion of serum was added to give prnper dilution and concentration of drug in serum After being spiked, the sample was filtered through a 0.2-pm filter. For recovery studies, aqueous standards were prepared in an identical manner for direct comparison to the drug level in serum. Calibration curves for propranolol and chloramphenicol were generated from dilute aqueous standards derived from stock solutions of each drug. Validation studies were performed for the drugs propranolol and chloramphenicol on samples obtained from different clinical laboratories where they were first analyzed by traditional liquid extraction methodology (20,21). A limited amount of sample was obtained frozen from the clinical lab and thawed just prior to analysis. These samples were well centrifuged at the lab, had good clarity, and were free from particulate matter. The propranolol samples were from a varied adult patient population undergoing treatment with the drug. Following clinical analysis approximately 1 to 2 mL of plasma was available for direct injection analysis. Injection volumes of 100 pL were used for the propranolol application. The chloramphenicol samples were from neonates, and a limited amount of sample was available from the start. After the clinical analysis, 100-200 pL of plasma remained for the direct approach. The direct injection method required 25-50-pL sample injections for the necessary sensitivity. Although filtering is strongly recommended, the sparse amount of sample available for this study required running the samples as obtained. For both studies the samples were decoded for comparison to the clinical results after the direct injection results were reported.
RESULTS AND DISCUSSION Extraction Column Chromatography. The extraction column chromatogram had a characteristic profile during the course of a drug extraction cycle. Irregardless of the type of bonded phase, the excluded components always occupied the first several minutes of the chromatogram with the micellar mobile phases listed in Table I. Therefore, the appropriate bonded phase was selected to give the desired retention characteristics of the drug relative to the excluded components. In each application the drug was eluted with the micellar mobile phase and cut onto the analytical column. Recovering the drug was investigated by both fore flush and back flush operation of the extraction column. T o minimize the time the two columns were in series, the micellar solvent was adjusted to maximize peak efficiency. Optimized combinations of the micellar mobile phase and the stationary phase gave solute retention times of 4-8 min. A 2-3 min heartcut was employed depending on the specific application. Examination of the micellar solvents shows that only minor modification
ANALYTICAL CHEMISTRY, VOL. 60, NO. 18, SEPTEMBER 15, 1988 3
1
I
I
I
0
3
I
I
I
9
1
2
I
1
5
Cycle Time (minl
Figure 2. Extraction column profile for propranolol application: 100 pL injection of plasma spiked at 400 nglmL level; 1, excluded proteinaceous materlal; 2, propranolol; 3, wash solvent front; 4, reequilibration front of micellar solvent. Solvent event program was as follows: 2-6.5 min, micellar solvent; 6.5-8.5 min, wash solvent; 8.5-15 min, micellar solvent.
was necessary for a variety of drugs. Considering eluent A as the standard extraction solvent, eluent B was a 1:l dilution of eluent A with water, and eluent C derived from eluent A by increasing the amount of acetonitrile from 80 to 140 mL per liter of solvent. The pH modification was only necessary to induce ion suppression of the carboxylate group of furosemide. Ion pairing or ionic suppression was generally used to maximize the dimensionality of the method, the goal being to have two separate mechanisms of retention on both the cleanup and the analytical column. After the drug was eluted from the extraction column, wash solvent supplied to pump 1by a solenoid valve for 2 min was used to elute highly retained endogenous components. The wash was then followed by a 7-10-min equilibration step in the extraction solvent. To illustrate the extraction cycle (Figure 21, a serum sample was spiked with propranolol at the 400 ng/mL level. While this was at the upper end of the therapeutic range, the high level was useful to clearly distinguish the peak and set the cut window for the multidimensionalprocess. The cut window was set to begin at approximately 0.3 min prior to the start of elution to 0.3 min after elution of the drug (Figure 2, peak 2). After the drug was eluted, the effects of the wash solvent were evident by peak 3. As this peak returned on scale, the wash was completed and the micellar solvent was delivered to the column for equilibration. A characteristic peak (peak 4) was observed when the equilibration was completed. At this point, the extraction column was ready for the next injection. A specified amount of organic modifier was usually added to the micellar extraction mobile phase. The purpose of this solvent was to increase the eluting power of the otherwise relatively weak micellar solution. While addition of solvent decreases the surfactant aggregation number, alters the critical micelle concentration, and increases micellar ionization, the fundamental role of the micellar solution in preventing protein precipitation still persists. To maintain saturation of the
1955
stationary phase of the extraction column, surfactant was added to the wash solution. Analytical Chromatography. The analytical column was operated under conventional reversed-phase conditions. With the proteinaceous material removed in the extraction step, any reversed-phase packing material is compatible with the system. All of the applications in this report used RP-18 reversed-phase media for the analytical phase. To fully appreciate the multidimensional system, the extraction colunin and analytical column had different functionality. The analytical phase was more retentive for the drug component than the extraction phase with the eluent used during the cut. The drug was then transferred from the extraction column with a solvent that focused and enriched the drug on the analytical column. Focusing the drug allowed it to be chromatographed in the analytical dimension with typical HPLC efficiency. It is advantageous to change the mechanism of retention to fully exploit the multidimensional system, e.g. the use of ionic suppression followed by ion pairing or vice versa. In some cases SDS was added to the analytical mobile phase to minimize artifacts from the switching process or to act as an ion pair reagent for the analyte. At the high levels of organic modifiers present in these solutions, micelles, if they are still present, are severely altered compared to more aqueous media. MultidimensionalSystem. With both the extraction and analytical column chromatography developed independently, the multidimensional system was integrated with a column switching valve. Three common valve configurations were considered in this study to evaluate their influence on the system. Two of these configurations can be seen in Figure 1. In the first configuration (A), the extraction column was between the injection valve and the column switching valve. Solvent delivery to the extraction column was performed by pump 1. During drug transfer from the extraction column to the analytical column, pump 1 delivered solvent to both columns. With this configuration the extraction column was always operated in the forward flush mode. In the B configuration, the extraction column was set across the switching valve. When the valve was actuated, solvent delivery to both columns was performed by the analytical pump (pump 2). As in the "A" configuration the extraction column was forward flushed for elution of the drug. The final configuration (C not pictured) was derived from Figure 2B. Again the extraction column was set across the switching valve. A t the switching valve, the line from the analytical pump and the line to the analytical column had their positions switched. Similar to the B configuration, when the valve was actuated, solvent delivery to both columns was provided by the analytical pump. However, the flow path along the extraction column is reversed thereby back flushing the extraction column to elute the component of interest. The applicationsfor propranolol, furosemide, and quinidine, were generated with the "A" configuration. The chromatogram of propranolol (Figure 3) is representative of this group. Evident here is the solvent front from the cut of the micellar solvent to the analytical column. The base-line disruption was minimized by having components common to both solvents present at the same concentration in both the extraction and the elution mobile phases, excepting of course, the principal modifier, methanol. The cycle time scale on Figure 3, indicates the extraction plus analysis time. The analysis time started when the two columns were placed in series mimicking an injection onto the analytical column. As seen in Figure 3, when data acquisition was started at a cycle time of 0, nonchromatographic time is added to the analysis portion of the cycle. When optimizing
1956
ANALYTICAL CHEMISTRY, VOL. 60, NO. 18, SEPTEMBER 15, 1988
Table 11. Drug Recovery
serum Level, drug
carbamazepine chloramphenicol furosemide procainamide
PgImL
inj vol, PL
1.0
100
8.0
100
1.0 1.0 0.1
propranolol
0.09
100 100 400 150
quinidine
1.2
200
therapeutic
LOD, ng/mL
%
recovery
range,” jtg/mL
4-12
100
98 99
96
50 9
10-20 NA
104
70
4-8
102 103 103
6 5
0.05-0.1
4
2-5
From ref 1. A
B
t
3
Q
N 0
x
1
1
%
L -
I
9
18
9
18
Cycle Time (min) 0
14
Cycle Time (mini Flgure 3. Direct injection analysis, propranolol application: 100 pL injection of spiked serum at 90 ng/mL concentration: fluorescence detection, 230-nm excitation and emission collected with 340-nm cutoff filter; valve configuration A.
sample throughput this time was better used as chromatographic time during the previous injection. Subsequent applications initiated data acquisition simultaneously with valve switch. Peak compression played an important role in this methodology. In the analytical dimension, the peak volume was reduced by focusing the drug on the head of the column. The peak volume for propranolol was 1.6 mL coming off the extraction column (Figure 2). Without focusing, the peak volume realized from the combination of dimensions should have been larger. The peak volume observed for propranolol with the combination of both chromatographic steps was 0.9 mL. Focusing on the analytical column allowed for the system to make-up for the low efficiency of the micellar chromatographic dimension. Four assays per hour could be run for any of these applications. The “B” configuration was used to develop the applications for chloramphenicol (Figure 4)and procainamide. Unlike the previous applications the drug was eluted from the extraction column with micellar solvent delivered by pump 2. If the analytical mobile phase was used to strip the drug from the extraction column, the chromatography suffered since the focusing/enrichment step was less efficient. While the drug contained in the sample injection volume was compressed and enriched, the benefit of weak mobile phase elution was not realized. Additionally, the strong analytical solvent performed a less selective recovery from the extraction column. Reduced selectivity in the drug recovery step increased the number of unidentified components in the analytical chromatography. In the B configuration, an additional step was needed to prime the analytical pump with the micellar solvent thereby
Flgure 4. Direct injection analysis, chloramphenicol application: A, 100-wL injection of spiked serum at 1 pg/mL concentration: B, blank serum; absorbance detection, 278 nm at 0.01 AUFS; valve configuration 8. B
A
i
- -
8
8 Cycle Time (minl
20
20
Figure 5. Direct injection analysis, carbamazepine application: A, 100-pL injection of spiked serum at 1 Kg/mL concentration: B, blank serum; absorbance detection, 287 nm at 0.01 AUFS; valve configuration C.
allowing a more selective recovery and a focus/enrichment of the drug. This nonchromatographic step was not visible to the data acquisition system as it occurred before the heart-cut was initiated. The cycle time (extraction plus analysis) reflects this extra step. The final application for carbamazepine (Figure 5) was performed with the drug eluted in the back-flush mode with valve configuration C. The need for selective recovery and priming of the analytical pump with the extraction solvent was also necessary here. The back-flush elution appeared to
ANALYTICAL CHEMISTRY, VOL. 60, NO. 18, SEPTEMBER 15, 1988
generate a significant amount of unwanted peaks to the analytical chromatogram and is not recommended for trace determinations. Analytical Figures of Merit. Percent recovery, limits of detection (LOD), and normal therapeutic range for the six drugs studied are given in Table 11. Recoveries ranged from 96% to 104% for spiked serum samples. This scatter about 100% recovery for this varied population does not appear to be related to any individual drug entity but merely represents the normal relative standard deviation (RSD) of the method. The recovery studies performed here were not extensive but serve only to indicate the general utility of the method. The limits of detection (LODs) for all of the drugs studied were well below the normal therapeutic range expected for each drug. The application of this technique toward therapeutic drug monitoring should be relatively straightforward. For pharmacokinetic studies, it appears that the LOD’s obtained permit drug determinations of 4-10 half-lives from the high end of the therapeutic ranges listed in Table 11. The injection size naturally impacts on the overall sensitivity and detectability of the method. For determination of our sampling of drugs in serum at the microgram-per-millilter level, 100-pL injection volumes provided sufficient sensitivity. For high sensitivity fluorescence applications, injection volumes in the 100-200-pL range gave LOD’s in the 4-10 ng/mL range. Early precision studies on 11 sets of data of varying concentrations of the same population of drugs gave RSD’s ranging from 0.6% to 6.1% (n = 2-4 per set of data). A more extensive study of propranolol a t 10 concentration levels (10.1-800 ng/mL, 100-pL injection, n = 3 for each level) gave RSD’s ranging from 1.2% to 4.2% for each data set. Normalizing these data (peak height/concentration) gave an RSD of 3.1% (n = 30). The calibration curve constructed from these data was linear ( r = 0.9998) and had a Y intercept equivalent of 1 ng/mL. During the development of these applications, the extraction column provided several hundred injections of serum in the 100-200 pL range. Under micellar cleanup conditions 30 mL of serum could be injected before any chromatographic deterioration of the extraction column was noticeable. The analytical column did not deteriorate during the course of these studies. Selectivity. For all of these applications, the detectability was artifact limited and not detector limited. The artifacts became significant below the 100 ng/mL level for UV applications and below 10 ng/mL for fluorescence applications. The LOD’s for each drug were calculated based on 3X the artifact interference. Artifacts were introduced from both the system and the matrix. System artifacts came from the valve switching, solvent changes, and reagent impurities. Reagent artifacts are troublesome in any multidimensional system involving an enrichment step. Any solvents and reagent impurities will also enrich with the components of interest. Although system artifacts may be dealt with through optimized chromatography, the optimization time can be greatly diminished by reducing system artifacts. These can be overcome to some degree through the cycle development and the sample injection volume to increase the drug/background response ratio. Matrix artifacts were a true limitation of the system since increasing the injection volume does not increase the ratio of druglartifact response. Matrix artifacts were dealt with through better optimized chromatography. After the analytical chromatography was fully exploited, the micellar dimension offered another degree of selectivity enhancement through minor modification of the extraction solvent or a change of the extraction phase.
.--
IC0
I
1957
/.I
J
0 ,
, 200
0
ya
ua
.p/d
Figure 6. Comparison of direct injection and liquid extraction methodology for propranolol.
Propranolol Validation Study. Samples from patients undergoing propranolol therapy were first determined in a clinical laboratory by a manual extraction procedure. The same samples were accumulated and analyzed as a batch using the direct injection approach described here. Standard and sample analyses were interspersed and all calculations were completed prior to decoding the samples. A calibration curve was generated by using aqueous standards from 80 to 500 ng/mL Cy = 0.248~+ 1.3; r = 0.999). Samples were analyzed in either singlet or duplicate runs by the direct injection method. Reproducibility data on these samples from the clinical analyses were not available. Regression analysis was performed comparing the two sets of results (Figure 6). Not included was the clinical data for four samples reported as