Strategies for the Assessment of Matrix Effect in ... - ACS Publications

Practical, experimental approaches for studying, identifying, and eliminating the effect of matrix on the results of quantitative analyses by HPLC−M...
0 downloads 7 Views 157KB Size
Anal. Chem. 2003, 75, 3019-3030

Strategies for the Assessment of Matrix Effect in Quantitative Bioanalytical Methods Based on HPLC-MS/MS B. K. Matuszewski,* M. L. Constanzer, and C. M. Chavez-Eng

Merck Research Laboratories, West Point, Pennsylvania 19486

In recent years, high-performance liquid chromatography (HPLC) with tandem mass spectrometric (MS/MS) detection has been demonstrated to be a powerful technique for the quantitative determination of drugs and metabolites in biological fluids. However, the common and early perception that utilization of HPLC-MS/MS practically guarantees selectivity is being challenged by a number of reported examples of lack of selectivity due to ion suppression or enhancement caused by the sample matrix and interferences from metabolites. In light of these serious method liabilities, questions about how to develop and validate reliable HPLC-MS/MS methods, especially for supporting long-term human pharmacokinetic studies, are being raised. The central issue is what experiments, in addition to the validation data usually provided for the conventional bioanalytical methods, need to be conducted to confirm HPLC-MS/MS assay selectivity and reliability. The current regulatory requirements include the need for the assessment and elimination of the matrix effect in the bioanalytical methods, but the experimental procedures necessary to assess the matrix effect are not detailed. Practical, experimental approaches for studying, identifying, and eliminating the effect of matrix on the results of quantitative analyses by HPLC-MS/MS are described in this paper. Using as an example a set of validation experiments performed for one of our investigational new drug candidates, the concepts of the quantitative assessment of the “absolute” versus “relative” matrix effect are introduced. In addition, experiments for the determination of, the “true” recovery of analytes using HPLC-MS/ MS are described eliminating the uncertainty about the effect of matrix on the determination of this commonly measured method parameter. Determination of the matrix effect allows the assessment of the reliability and selectivity of an existing HPLC-MS/MS method. If the results of these studies are not satisfactory, the parameters determined may provide a guide to what changes in the method need to be made to improve assay selectivity. In addition, a direct comparison of the extent of the matrix effect using two different interfaces (a heated nebulizer, HN, and ion spray, ISP) under otherwise the same sample preparation and chromatographic conditions was made. It was demonstrated that, for the investigational drug under study, the matrix effect was clearly observed when ISP interface 10.1021/ac020361s CCC: $25.00 Published on Web 06/04/2003

© 2003 American Chemical Society

was utilized but it was absent when the HN interface was employed. High-performance liquid chromatography (HPLC) with tandem mass spectrometric (MS/MS) detection is now considered the method of choice for the quantitative determination of drugs and metabolites in biological fluids. The methodology achieved its preferred status because it has been perceived that MS/MS detection was highly selective and thus effectively eliminated interference by endogenous impurities. Even without any cleanup or extraction of samples and with very little or no chromatographic separation, endogenous impurities from biofluids were not detected, and the only MS/MS signal observed in control biofluids originated from the desired analyte. Therefore, a common perception was that utilization of HPLC-MS/MS practically guaranteed method selectivity and both sample extraction and chromatography could be simplified or even eliminated. Chromatographic run times of 0-3 min using short (e2 cm) HPLC columns were commonly utilized, allowing high-throughput (20-40 samples/ h) determination of analytes in complex biological matrixes. Contrary to this common belief, the reliability of quantitative assays using HPLC-MS/MS and the integrity of resulting pharmacokinetic (PK) data may not be absolute. Results may be adversely affected by lack of selectivity due to ion suppressions caused by the sample matrix, interferences from metabolites, and “cross-talk” effects. Remarkable examples of the importance of eliminating matrix effect and ion suppression during the development of quantitative methods based on HPLC-MS/MS for two compounds studied in our laboratories were reported by us earlier.1,2 Coeluting, undetected matrix components may reduce or enhance the ion intensity of the analytes and affect the reproducibility and accuracy of the assay. The degree of ion suppression for an analyte and an internal standard may be different in different lots of the same biofluid (for example, urine or plasma), originating from different subjects and over a prolonged period of time required, for example, for completion of multiple dose clinical study, adversely affecting the reliability of determination and the integrity of PK data. Some additional examples are available in the literature * Corresponding author. Fax: (215) 652-8548. E-mail: bogdan_matuszewski@ merck.com. (1) Fu, I.; Woolf, E. J.; Matuszewski, B. K. J. Pharm. Biomed. Anal. 1998, 18, 347-357. (2) Matuszewski, B. K.; Constanzer, M. L.; Chavez-Eng, C. M. Anal. Chem. 1998, 70 (5), 882-889.

Analytical Chemistry, Vol. 75, No. 13, July 1, 2003 3019

illustrating the need for careful assessment of HPLC-MS/MS assay selectivity3 including evaluation of selectivity in postdose biological fluids in the presence of metabolites4-6 and the need for an efficient extraction of analytes from biological materials2,7 and chromatographic separation.2 The matrix effect phenomenon was originally described by Kebarle and Tang,8 who showed that electrospray responses of organic bases decreased with an increase in concentrations of other organic bases. However, in the context of quantitative bioanalysis of drugs and metabolites, present in the same type of matrix (i.e., human urine or plasma) but originating from different sources (subjects), the matrix effect issue was not sufficiently studied and addressed. In addition, the recently issued U.S. Food and Drug Administration’s (FDA) Guidance for Industry on Bioanalytical Method Validation9 and a Conference Report from the workshop held on the same subject in Arlington, VA, in January 2000,10 clearly indicate the need for the assessment of matrix effect during development and validation of HPLC-MS/MS methods “to ensure that precision, selectivity, and sensitivity will not be compromised”.9,10 However, in both of these documents, the experiments necessary to demonstrate the presence or absence of matrix effect in a given bioanalytical method are not described or suggested. Qualitatively, experiments confirming the presence of matrix effect in biological matrixes in comparison with the MS/MS response in neat solvents or HPLC mobile phases were proposed,11-14 but they do not provide a guidance of how to evaluate and determine whether an existing analytical method or a method being developed is selective or suffers from the lack of selectivity due to the effect of matrix. Therefore, a need exists to develop an experimental protocol to demonstrate during assay development and validation the absence or presence of matrix effect in a newly developed bioanalytical method and use this information as guidance for making changes and corrections, if any, to the original method that would allow the establishment of a truly selective method free of matrix effect interferences. Experimental strategies that allow this type of method evaluation are described in this paper. These strategies will be illustrated using as an example the experimental data obtained during development of bioanalytical methods for a selected drug candidate studied recently in our laboratories. (3) Clarke, S. D.; Hill, H. M.; Noctor, T. A. G.; Thomas, D. Pharm. Sci. 1996, 2, 203-207. (4) Constanzer, M. L.; Chavez, C. M.; Matuszewski, B. K.; Carlin, J.; Graham, D. J. Chromatogr., B 1997, 693, 117-129. (5) Matuszewski, B. K.; Chavez, C. M.; Constanzer, M. L. J. Chromatogr., B 1998, 716, 195-208. (6) Jemal, M.; Xia, E.-Q. Rapid Commun. Mass Spectrom. 1999, 13, 97-106. (7) Buhrman, D.; Price, P.; Rudewicz, P. J. Am. Soc. Mass Spectrom. 1996, 7, 1099-1105. (8) Kebarle, P.; Tang, L. Anal. Chem. 1993, 65, 972A-986A. (9) Department of Health and Human Services, Food and Drug Administration, Guidance for Industry on Bioanalytical Method Validation. Fed. Regist. 2001, 66 (100), 28526 (Docket No. 98D-1195). (10) Shah, V. P.; Midha, K. K.; Findlay, J. W. A.; Hill, H. M.; Hulse, J. D.; McGilveray, I. J.; McKay, G.; Miller, K. J.; Patnaik, R. N.; Powell, M. L.; Tonelli, A.; Viswanathan, C. T.; Yacobi, A. Pharm. Res. 2000, 17 (12), 15511557. (11) Bonfiglio, R.; King, R. C.; Olah, T. V.; Merkle, K. Rapid Commun. Mass Spectrom. 1999, 13, 1175-1185. (12) King, R.; Bonfiglio, R.; Fernandez-Metzler, C.; Miller-Stein, C.; Olah, T. J. Am. Soc. Mass Spectrom. 2000, 11 (11), 942-950. (13) Choi, B. K.; Hercules, D. M.; Gusev, A. I. J. Chromatogr., A 2001, 907, 337-342. (14) Choi, B. K.; Hercules, D. M.; Gusev, A. I. Fresenius’ J. Anal. Chem. 2001, 369, 370-377.

3020

Analytical Chemistry, Vol. 75, No. 13, July 1, 2003

The mechanism and the origin of the matrix effect is not fully understood,8,12 but it may originate from the competition between an analyte and the coeluting, undetected matrix components reacting with primary ions formed in the HPLC-MS/MS interface. Depending on the environment in which the ionization and ion evaporation processes take place, this competition may effectively decrease (ion suppression) or increase (ion enhancement) the efficiency of formation of the desired analyte ions present at the same concentrations in the interface. To determine an analyte by HPLC-MS/MS, the uncharged molecules of this analyte need to be transformed to ions that are later analyzed by MS/MS according to their mass-to-charge (m/z) ratios. The HPLC-MS/MS interface can be considered as a “chemical reactor” in which primary ions react with analyte molecules in a very complex series of charge-transfer and ion-transfer reactions. The rate and efficiency of these reactions are highly dependent on the relative ionization energies, proton affinities, or both of the molecules present in the “reactor” at any given time. It is intuitively clear that the efficiency of formation of the desired ions must be very much matrix-dependent due to the competition between the molecule of interest and a number of other undetected but coeluting molecules present in the system that are capable of reacting with primary ions. This effect may reduce or increase the intensity of analyte ions and affect the reproducibility and accuracy of the assay. Unfortunately, most of the HPLC-MS/MS methods published in the literature do not address the matrix effect issue although eliminating this effect is critical in establishing reliable methods. Ignoring this effect may adversely affect the reliability of determination of analyte concentrations and the integrity of PK data generated. Our earlier report2 and observations of others12 indicated that the extent of matrix effect may be dependent on the HPLC-MS interface employed in a given method (atmospheric pressure chemical ionization, APCI, vs electrospray ionization, ESI). The ionization mechanism is different when these different interfaces are used, which may affect the efficiency of formation of the desired ions in the presence of the same coeluting compounds. To address this issue, a detailed comparison of the matrix effect under otherwise the same sample extraction and HPLC conditions was made using a heated nebulizer (HN) versus ion spray (ISP) interface that are utilized in the Sciex HPLC-MS/MS systems commonly used for quantitative bioanalysis. EXPERIMENTAL SECTION Materials. Compound 1 and the internal standard (IS, 2, Figure 1) were synthesized at Merck Research Laboratories (Rahway, NJ). All solvents and reagents were of HPLC or analytical grade and were purchased from Fisher Scientific (Fair Lawn, NJ). The different lots of drug-free human heparinized plasma originated from Biological Specialties Corp. (Lansdale, PA). Nitrogen (99.999%) was purchased from West Point Supply (West Point, PA). Instrumentation. A Perkin-Elmer (PE) Sciex (Thornhill, ON, Canada) API 3000 tandem mass spectrometer equipped with a heated nebulizer or an ion spray interface, a PE 200 autoinjector, and a PE 200 quaternary pump were used for all HPLC-MS/MS analyses. The data were processed using MacQuan software (PE Sciex) on a MacIntosh Quadra 900 microcomputer.

Figure 1. Chemical structures of compound 1 and an internal standard 2.

Standard Solutions. A stock solution of 100 µg/mL for standards 1 and 2 were prepared in the mobile phase. A 10 µg/ mL stock solution containing 1 was then prepared by serial dilution. This solution was then diluted further with the mobile phase to give a series of working standards of 0.005-5.0 µg/mL. The 100 µg/mL stock solution of the internal standard 2 was serially diluted with the mobile phase to yield a working standard of 0.3 µg/mL. Chromatographic Conditions. Chromatographic separation of 1 and 2 was performed on a Keystone Scientifics Hypersil BDS C-18 (50 × 4.6 mm 3 µm, Keystone Scientific, Bellefonte, PA) analytical column with a mobile phase consisting of 80% acetonitrile and 20% water containing 0.1% formic acid, pumped at a flow rate of 1 mL/min. The total run time was 6 min. Both analytes were baseline separated. The retention times of 1 and 2 were about 2.4 and 1.2 min corresponding to capacity factors (k′) of 3.8 and 1.4, respectively. When the HN interface was utilized, the total eluent from the column (1 mL/min) was directed to the interface, whereas in the case of the ISP interface, the flow was split 95:5; the flow directed to the ISP interface was equivalent to 50 µL/min. HPLC-MS/MS Conditions. A PE Sciex triple quadrupole mass spectrometer (Sciex API 3000) was interfaced via a Sciex HN or ISP probe with the HPLC system. The HN probe was maintained at 500 °C, and gas-phase chemical ionization was effected by a corona discharge needle (+4 µA) using positive ion APCI. The nebulizing gas (N2) pressure was set for the HN and ISP interfaces at 80 and 40 psi, respectively. The auxiliary flow was 2.0 (HN) and 0.0 L/min (ISP), the curtain gas flow (N2) was 0.9 L/min, and the sampling orifice potential was set at +50 V, for both HN and ISP interfaces. The dwell time was 400 ms, and mass analyzers Q1 and Q3 were operated at unit mass resolution. The mass spectrometer was programmed to admit the protonated molecules [M + H]+ at m/z 394 for 1 and m/z 358 for 2 via the first quadrupole filter (Q1). Collision-induced fragmentation at Q2 (collision gas N2, 275 × 1013 atoms cm-2) yielded the product ions at Q3 of m/z 326 and 290 for 1 and 2, respectively. Peak area ratios (1/2) obtained from selective reaction monitoring of the analytes (m/z 394 f 326)/(m/z 358 f 290) were utilized for the construction of calibration lines, using weighted (1/x2) linear leastsquares regression of the plasma concentrations and measured peak area ratios. Data collection, peak integration, and calculations were performed using MacQuan PE-Sciex software.

Sample Preparation. Three sets of five standard lines were prepared to evaluate the assay accuracy, precision, recovery, and absence or presence of matrix effect. The first set of five standard lines (set 1) was prepared to evaluate the MS/MS response for neat standards of two analytes (1 and 2) injected in the mobile phase. The second set (set 2) was prepared in plasma extracts originating from five different sources and spiked after extraction. The third set (set 3) was prepared in plasma from the same five different sources as in set 2, but the plasma samples were spiked before extraction. By comparing the absolute areas of peaks 1, 2, peak areas ratios, and slopes of the standard lines between these three different sets of standard lines, the absence or presence of matrix effect on the quantification of 1 and 2 was assessed. In addition, precision and accuracy of the method and recovery of analytes were also determined. Set 1. Five standard lines were constructed using neat solutions of 1 and 2 in the mobile phase. The samples were prepared by placing 100 µL of the appropriate standards of 1, 100 µL of the 0.3 µg/mL stock solution of 2, and 100 µL of the mobile phase (total volume 300 µL) into 15-mL centrifuge tubes. After mixing, the solutions were transferred into autosampler vials and 50 µL was injected directly into the HPLC-MS/MS system. Set 2. Five standard lines were constructed in five different lots of plasma by placing 1 mL of plasma in 15-mL centrifuge tubes followed by the addition of 200 µL of the mobile phase (to simulate the addition of standard solutions of 1 and 2 that were each added in set 3 in 100 µL of the mobile phase). After vortexing, the plasma was basified with pH 9.8 carbonate buffer (1 mL) and extracted with 7 mL of methyl tert-butyl ether. The tubes were capped with Teflon-lined caps, rotate-mixed for 15 min, centrifuged at 3000 rpm (3056g) for 5 min, and placed in a dry ice-acetone mixture. The organic layer was separated, placed in clean 15-mL centrifuge tubes, and evaporated to dryness under a stream of nitrogen in a 50 °C water bath, and the residue was reconstituted in 100 µL of the mobile phase, 100 µL of the 0.3 µg/mL stock solution of 2, and 100 µL of the appropriate standards of 1 (total volume 300 µL). The extracts from the control samples (blanks) were reconstituted in 300 µL of the mobile phase. A total of 50 µL of the extracts was injected into the HPLC-MS/MS system. In set 2, the analytes were spiked after extraction into different plasma extracts, whereas in set 3 (below), the analytes were spiked into different plasmas before extraction. Set 3. Five standard lines were constructed in five different lots of plasma (same plasmas as in set 2) by placing 1 mL of plasma in 15-mL centrifuge tubes to which 100 µL of the appropriate standards of 1 and 100 µL of the 0.3 µg/mL stock solution of 2, both in the mobile phase, were added before extraction. The control (blank) tubes had 1 mL of plasma to which 200 µL of the mobile phase was added. After vortexing, the plasma was basified and analytes were extracted in the same manner as in set 2. The residues were reconstituted in 300 µL of the mobile phase, and 50 µL was injected into the HPLC-MS/MS system. Samples from sets 1-3 were analyzed using ISP interface first, and as soon as all samples were analyzed, the same samples were injected into the same HPLC-MS/MS system equipped with the HN interface. The order of injection was as follows: samples from set 1 were injected first (five standard lines, each line from a low to a high concentration), followed by sets 2 and 3 injected in the Analytical Chemistry, Vol. 75, No. 13, July 1, 2003

3021

same order (low to high concentrations) for plasma lot 1, followed by lot 2, etc. Precision, Accuracy, and Recovery. The precision of the method was determined by the replicate analyses (n ) 5, set 3) of human plasma containing 1 at all concentrations utilized for the construction of calibration curves. The linearity of each standard curve was confirmed by plotting the peak area ratio of 1 to 2 versus drug concentration. The sample concentrations were calculated from the equation y ) mx + b, as determined by weighted (1/x2) linear regression of the standard line. The accuracy of the method was expressed by [(mean observed concentration)/(spiked concentration] × 100. The recovery was determined by comparing the mean peak areas of 1 and 2 obtained in set 3 to those in set 2. Assessment of Matrix Effect. The assessment of matrix effect and assay reliability is critical when homologues rather than stable isotope-labeled analytes are utilized as internal standards. By comparing the peak areas of the analyte standards, standards spiked before and after extraction into different lots of plasma, and the peak area ratios of analytes to an IS, the recovery and ion suppression or enhancement associated with a given lot of plasma were assessed. Assessment of Assay Selectivity. The assay selectivity was assessed by analyzing extracts from five lots of plasma from different sources. Endogenous peaks at the retention time of the analytes of interest were not observed in any of the plasma lots evaluated. In addition, the “cross-talk” between MS/MS channels used for monitoring 1 and 2 for both analytes was assessed by the following: (1) separately injecting 1 at the highest concentration on the standard line (200 ng/mL) and monitoring the response in the IS channel and (2) by injecting a plasma sample spiked only with the IS (2) and monitoring the response in the drug channel at the sensitivity (y-axis) required for monitoring 1 at the lowest limit of quantification. No “cross-talk” was observed. RESULTS Evaluation of the Matrix Effect and Assay Validation Using HN Interface. The matrix effect and the possibility of ionization suppression or enhancement for 1 and 2 was evaluated by comparing the results of analysis of three sets of samples (set 1, set 2, set 3) prepared as described in the Experimental Section. These three sets corresponded to three types of system evaluation. In the first set (set 1), standards of the analytes present in the neat reconstitution solvent (HPLC mobile phase used in the assay) were analyzed directly at seven concentrations and analyses were repeated five times at each concentration (35 samples). The results of analyses of set 1 provided a good insight into the overall HPLCMS/MS system reproducibility in measuring the absolute peak areas on consecutive injections, the performance of the detector, and the chromatographic system as a whole. In the second set (set 2), plasma samples from five different plasma lots were first extracted and spiked after extraction with the analytes 1 and 2 in the same solvent (mobile phase) as in set 1. Any additional variability of the peak areas for the analytes than those observed in set 1, as demonstrated by an increase in the coefficients of variation (CV) at each concentration, would be indicative of an effect of sample matrix since analytes at the same concentrations were spiked into plasma extracts. In set 3, analytes were spiked before extraction into plasma samples originating from five different 3022

Analytical Chemistry, Vol. 75, No. 13, July 1, 2003

Table 1. Precisiona (CV, %) of Determination of Peak Areas of 1, Internal Standard (2), and the Peak Area Ratios (1/2) in Sets 1,b 2,c and 3d Using Heated Neublizer Interface precision (CV, %) peak area 1 nominal concn set set set (ng/mL) 1 2 3

peak area 2

peak area ratio 1/2

set set set 1 2 3

set 1

set 2

set 3

accuracye (%) 99 101 103 101 100 99 97

0.5 1.0 5.0 10.0 50.0 100.0 200.0

4.0 5.0 4.7 4.2 3.1 2.0 3.2

7.9 10.2 4.0 5.2 9.3 6.0 8.5 3.2 4.4 7.7 5.0 5.2 5.3 2.3 4.7 6.0 5.6 5.2 4.1 6.8 5.3 4.3 3.0 4.0 4.7 6.9 3.9 3.1 4.7 3.6 5.2 6.8 4.1 4.1 5.1

0.9 2.5 1.6 1.6 0.7 2.7 2.0

4.7 3.3 4.1 4.4 3.9 3.4 2.7

2.9 2.6 2.9 1.9 1.3 2.6 1.9

column

A

B

G

H

I

C

D

E

F

J

a n ) 5. b 1 and 2 standards in mobile phase. c 1 and 2 spiked after extraction into extracts from five different plasma lots. d 1 and 2 spiked before extraction into extracts from five different plasma lots. e Expressed as the[(mean observed concentration)/(nominal concentration)] × 100.

sources as in set 2. The variability in CV values here would reflect a combined effect of a sample matrix and potential differences in recovery of analytes from different plasma lots. In all three cases (sets 1-3), five standard lines were constructed (total of 3 × 35 ) 105 samples). In a typical, conventional method validation, only set 3 samples (35 samples) with analytes spiked before extraction into a single lot of a biological fluid are usually analyzed. The results of the analyses for sets 1-3 are summarized in Tables 1 and 2. The results obtained in this manner allow determination of the matrix effect (ME), recovery (RE) of the extraction procedure, and overall “process efficiency” (PE) by comparing the absolute peak areas for 1 obtained in sets 1-3 (Table 2). If one depicts the peak areas obtained in neat solution standards in set 1 as A, the corresponding peak areas for standards spiked after extraction into plasma extracts as B (set 2), and peak areas for standards spiked before extraction as C (set 3), the ME, RE, and PE values can be calculated as follows:

ME (%) ) B/A × 100

(1)

RE (%) ) C/B × 100

(2)

PE (%) ) C/A × 100 ) (ME × RE)/100

(3)

The terms “process efficiency”, “extraction efficiency”, and “ion suppression” were originally introduced by Buhrman et. al.7 In their study, ion suppression was defined as (100 - B/A × 100), and the potential for ion enhancement was not considered. To account for both ion suppression and ion enhancement and to avoid negative values in the case of ion enhancement, the ratio (B/A × 100) is defined here generally as a matrix effect. The ME calculated in this manner may be referred to as an “absolute” matrix effect since the signal response of the standard present in the plasma extract is compared to the response of a standard made directly in a neat mobile phase. Although the presence of this absolute matrix effect may be of some concern (vide infra), the more important parameter in the evaluation and validation of a

Table 2. Matrix Effect (ME), Recovery (RE), and Process Efficiency (PE) Data for 1 and 2 in Five Different Lots of Human Plasma Using HN Interface mean peak areaa nominal concn (ng/mL) 0.5 1.0 5.0 10.0 50.0 100.0 200.0

set 1

1 set 2

set 3

1.94 3.78 19.16 39.66 194.29 390.24 756.25

2.56 5.03 25.87 52.62 252.38 553.14 1046.41

2.57 4.88 25.93 49.70 251.35 498.21 978.84

set 1

2 set 2

set 3

49.16 48.23 48.96 50.36 48.34 48.07 48.99

61.77 59.44 61.25 62.22 60.53 66.07 64.23

58.81 55.44 58.01 57.01 57.92 58.13 58.19

mean column

A

B

C

D

E

F

MEb (%) 1 2

REc (%) 1 2

PEd (%) 1

2

132 133 135 133 130 142 138

126 123 125 124 125 137 131

101 97 100 94 100 90 94

95 93 95 92 96 88 91

133 129 135 125 130 128 129

120 115 118 113 120 121 119

135

127

97

93

130

118

G

H

I

J

K

L

a In arbitrary units, ×104, n ) 5. b Matrix effect expressed as the ratio of the mean peak area of an analyte spiked postextraction (set 2) to the mean peak area of the same analyte standards (set 1) multiplied by 100. A value of >100% indicates ionization enhancement, and a value of 100% indicates an ionization enhancement and a value of 100%, a clear indication that matrix effect was not taken into account in the calculations, and that the ratio C/A × 100 (eq 3) rather than C/B × 100 (eq 2) was utilized. In the case of 1 and 2, due to the observed ionization enhancement for both 1 and 2 (135 and 127%, respectively, Table 2, columns G and H), the (C/A × 100) values were >100% (130 and 118% for 1 and 2, respectively, Table 2, columns K and L). The “true” recovery values, free from matrix effect contributions, and calculated using the ratio (C/B × 100) (eq 2), were 97 and 93% for 1 and 2, respectively (Table 2, columns I and J) when the HN interface was utilized. The recovery values calculated according to eq 2 should be independent of the HPLC-MS/MS interface employed. However, they calculated lower when the ISP interface was utilized (77 and 83% for 1 and 2, respectively, Table 5, columns I and J). The origin for this apparent discrepancy is, at present, unknown. However, since absolute peak areas were used for calculations, any instruAnalytical Chemistry, Vol. 75, No. 13, July 1, 2003

3027

ment or interface instability during analyses of samples in set 2 (that were analyzed first) and set 3 (analyzed at the end of the run) may have led to the decrease of the absolute peak areas of C in comparison with B lowering the RE values. Such a relative decrease in response may be especially noticeable when the ISP interface is utilized. In this case, 95:5 splitting of the flow from the column to the interface was employed, and any partial clogging of the splitter may have decreased the absolute response for both analytes at the end of the run when samples from set 3 were analyzed. The matrix effect may also affect the reliability of determination of recovery values in conventional bioanalytical methods. Based on the data presented here, the determination of recoveries in all of these methods should be performed using eq 2 instead of eq 3. Method Validation and Matrix Effect. Evaluation of the Assay Precision and Accuracy in Single versus Multiple Sources of a Biofluid. The importance and the necessity of evaluating the matrix effect during method development and validation is best illustrated by the comparison of the results of a typical validation experiments performed in a single lot of plasma versus the same validation experiment performed in five different plasma lots. The precision and accuracy values obtained in a single plasma lot using the ISP interface were highly satisfactory and ranged from 4.2 to 8.5%, and 96 to 108%, respectively, at all concentrations utilized for constructing the standard line ((Table 4, columns L and N). In this case, even the precision values for the determination of the absolute peak areas for both 1 and 2 were excellent and varied from 3.3 to 8.5% for 1 and 2.1-6.0% for 2 (data not shown in the tables). Based on these results, the ISP method would have been considered as valid and adequate for supporting PK studies. However, when the same validation was attempted in five different plasma lots, the precision values were unacceptably high (11.127.8%, Table 4, column K) under otherwise identical extraction and chromatographic conditions as in the validation experiment utilizing a single plasma lot. The reason for this significant method imprecision (Figure 4) was the presence of high relative matrix effect when the ISP interface was employed, as indicated by the high CV values for peak areas of 1 and 2 in five different plasma lots (11.6-23.8% and 4.6-11.3%, respectively, Table 4, columns B and F). In a single plasma lot, the analogous CV values were in the range of 3.3-8.5% for 1 and 2.1-6.0% for 2. Despite this high relative matrix effect in five different lots of plasma for the drug, the method would have been validated if the relative matrix effect for the IS had the same pattern as for the drug. In such a case, the 1/2 ratio would not have been affected. However, this was not the case for 1 and 2, and the high CV values of the ratio of 1/2 (11.1-27.8%, Table 4, column K) had its origin in a significant variability in the MS/MS response for the same concentrations of 1 in different plasma lots accompanied by relatively the same responses for the IS (Figure 2). Contrary to the results obtained using the ISP interface, the precision and accuracy of the method using the HN interface in five different lots of plasma was excellent, as illustrated by the precision and accuracy values ranging from 1.3 to 2.9% and 97 to 103%, respectively (Table 1, columns I and J). The CVs of the peak area ratios of 1/2 were smaller (Table 1, column I) than the CVs of the peak areas of 1 and 2 (Table 1, columns C and F, 3028

Analytical Chemistry, Vol. 75, No. 13, July 1, 2003

respectively), confirming the desired compensating effect of the presence of internal standard on the precision and reliability of quantification of 1. When such highly satisfactory data are obtained in five different lots of a biofluid, there is no need to assess the overall system and assay performance in a single lot of a biofluid, and such experiment was not repeated here. Evaluation of Slopes of Standard Lines. The lot-to-lot variability in the determination of 1/2 ratios, a measure of drug concentration, is best evaluated by comparing the slopes of standard lines constructed in different plasma lots. The high variability (CV) of these slopes is indicative of the overall effect of the sample matrix on the drug/IS ratio rather than on the individual responses of 1 and 2, when slopes obtained in set 2 (spiked plasma extracts) are compared. Any variability in the analogous slopes obtained in set 3 in five different lots of plasma may reflect the combined differences in the effect of matrix and a variable recovery in different plasma lots (Figure 3). In the case of the HN interface, the variability in standard line slopes (3.5% for set 2 and 1.0% for set 3, Table 3, columns B and C) were very small and comparable to the variability in similar five slopes constructed directly from standards (set 1, CV ) 1.3%, Table 3 column A). The very small variability of slopes obtained in set 1 was a direct indicator of an excellent overall HPLC-MS/MS system reproducibility and performance. Together, these results indicated that the matrix effect, if any, had no effect on the determination of 1/2 ratios in different plasma lots. A small improvement in the CV values obtained in set 3 versus set 2 (1.0 vs 3.5%) may be indicative of a possible and favorable compensating effect of the small difference in recoveries for 1 and 2 in different plasma lots on the 1/2 ratios. Contrary to the HN interface, the analysis of similar slope data for sets 1-3 obtained using the ISP interface (Table 3) clearly indicated that due to a significant matrix effect the variability in slopes between different plasma lots was quite significant (for example, ∼44% increase in slope was observed between lots 3 and 4 in set 2; data not shown). The CV values for sets 2 and 3 were high and comparable (14.9 and 13.2%, Table 3, columns B and C), indicating that the overall high variability of the method was due to the matrix effect rather than any potential differences in recoveries between different plasma lots for both 1 and 2. The variability of slopes obtained in set 1 (0.9%, Table 3, column A) was negligible and was similar to analogous values obtained for set 1 when the HN interface was employed (1.3%, Table 3, column A), confirming that the overall ISP system reproducibility and performance was maintained. To additionally confirm that inadequate assay precision and accuracy of the ISP method was due to the effect of matrix and not, for example, due to the inadequate overall system performance, the experiments similar to those performed in set 3 but using a single instead of five different lots of plasma were repeated and five slopes of standard lines in the same single plasma lot were determined. The small CV value (2.4%) obtained for slopes in set 3a (Table 3, column D) clearly indicated that the overall system performance was highly adequate and the variability in slope values (13.2%) observed when plasma from five different sources was utilized (Table 3, column C) was due primarily to the effect of matrix. Simplified Approaches for Assessing the Matrix Effect. The approach described in this paper allows a detailed quantitative

assessment of the matrix effect during method validation and a full examination of the validity of the bioanalytical method. However, some simplified, alternative approaches for the assessment of matrix effect listed below may be considered. (1) Since the knowledge of the absolute matrix effect is not necessary to establish the method validity, analyses of samples in set 1 may not be required. In addition, the data obtained in set 3 are reflective of the combination of two effects: the effect of sample matrix and recoveries of analytes. To determine the relative matrix effect only, samples in set 2 prepared in a biofluid originating from five different sources need only to be analyzed. Careful examination of the peak areas (heights) of the drug and an IS spiked into different biofluid extracts, the degree of variability (CV) of the absolute responses, and drug/IS ratios are all indicative of the presence or absence of matrix effect. Although the knowledge of an absolute matrix effect is, in principle, not necessary to establish method validity, it is important to have an idea about the extent of detector sensitivity on the matrix in which analytes are injected since the possibility of a more pronounced relative matrix effect may increase with the increase in the absolute matrix effect. If a large absolute matrix effect is observed, the likelihood for greater variability in the detector response from a biofluid originating from large number of different subjects (instead of just five plasma lots studied in a typical validation experiment) participating in long-term clinical studies may increase. The elimination of a relative matrix effect in such cases may be especially important and may require much more attention than in cases where the absolute matrix effect is relatively small or negligible. (2) Instead of analyses of a full set 2 samples (7 × 5 ) 35 samples), repeat analyses of standards spiked into a biofluid extract from five different sources but only at two or three concentrations (2 × 5 or 3 × 5 ) 10 or 15 samples) may be performed and data analyzed as in case 1 above. (3) Both the absolute and relative matrix effects can be determined by analyzing samples in sets 1 and 2 at two or three concentrations (2 × 2 × 5 ) 20 or 2 × 3 × 5 ) 30 samples) instead at all seven concentrations. Again, the presence of the absolute and even a relative matrix effect for the individual analytes (drug and an IS) does not preclude the method to be valid. If the pattern of variability for the drug and the IS is the same, the ratio of the drug/IS, a measure of drug concentration, may not be affected, and the CVs of the ratio would be much smaller than the CVs of the individual responses for the drug and the IS. Such decrease in the CV values would confirm the compensating effect of the presence of internal standard on the precision and reliability of quantification of the drug. (4) To assess absolute and relative matrix effect and recoveries of analytes in an abbreviated fashion, samples in sets 1-3 but only at two or three concentrations instead at all concentrations (seven) on the standard line need to be analyzed. The total number of samples would be 2 × 3 × 5 ) 30 or 3 × 3 × 5 ) 45. Assuming the absence of the matrix effect is demonstrated, the validation experiments (set 3, 35 samples) need to be later performed. (5) Slopes of five standard lines obtained in set 3 may be compared (vide infra). If these slopes in five different sources of a biofluid are practically the same (CV < 4-5%), the absence of any significant matrix effect on quantification may be considered

as being confirmed. Recommendations for Bioanalytical Method Validation. The current common practice in method validations involves the analyses of samples in set 3 only and in a single source of a biofluid. As clearly demonstrated in this paper, the validation data obtained in this manner (Table 4, column L and Figure 4) may be totally misleading since they do not take into account the possibility of a severe matrix effect on quantification. When the same method was attempted to be validated in the same biofluid (plasma) but originating from five different sources, unacceptable precision values were obtained (Table 4, column K and Figure 4) and the method could not be considered as valid. The best way to perform method validation and to assess the combined effect of sample matrix and variable recoveries on assay results for both the drug and an IS is to determine precision and accuracy of the method (set 3) in biofluid samples originating from at least five different sources instead from a single source. The slopes of the standard lines constructed in these different biofluid lots may then be compared. These slopes should be practically the same, which would indicate that sample matrix and any potential differences in recoveries do not affect the precision and accuracy of the method. The method may be considered valid and no further evaluation of the matrix effect may be necessary when validation is performed in five different lots of a biofluid, and the precision and accuracy values for set 3 and slopes of the standard lines are as good as those illustrated in Table 1 (column I) and Table 3 (column C) (HN interface). Similar data obtained in a single lot of a biofluid are not sufficient for the method to be considered valid, and the absence of matrix effect needs to be demonstrated. A question may be asked why validation experiments are recommended in five instead of in any other number of different biofluid lots. The only reason for this recommendation is to take into account the practical aspects of method validation procedures that usually involve repeat analyses of five samples at each concentration. Without increasing overall number of samples analyzed, the matrix effect in five different lots of a biofluid may then be assessed simultaneously with the determination of the precision and accuracy of the method. It would be highly desirable to assess matrix effect in as many different biofluid sources as possible, but this is difficult experimentally and is highly impractical. General Comments and Other Possible Implications of Matrix Effect on the Results of Quantitative Bioanalysis. 1. Matrix Effect and Drug Interaction Studies. A different form of matrix effect may be encountered during drug interaction studies. In these studies, a three-way crossover experiment is usually performed in which subjects are treated with drug A in the first part (1), with drug B, that may potentially interact with A, in the second part (2) of the study and, finally, in the third part (3), with drugs A + B combined. The bioanalytical and PK data obtained in these drug interaction studies are usually obtained based on the analyses of biofluid samples (for example, plasma) generated in parts 1 and 3 from a number of subjects using a bioanalytical method developed for drug A in a control plasma. However, matrix effect may also originate from the presence of B and its metabolites that are present only in samples from part 3 of the study but not in samples from part 1. Therefore, compound A Analytical Chemistry, Vol. 75, No. 13, July 1, 2003

3029

may be considered as being present in two different matrixes in plasma samples from part 1 and part 3 of the study. To confirm the selectivity of the method for A in the presence of B and its metabolites, one can consider pooling postdose plasma samples from part 2 of the study and constructing a standard line for A in this pooled plasma containing B and its metabolites. When the slopes of the standard line constructed in a control plasma and in pooled plasma from part 2 are practically the same, the matrix effect from B and its metabolites on the quantification of A may be considered as negligible. This procedure is routinely utilized in our laboratories to assess matrix effect in drug interaction studies when samples from part 2 of the study are available. 2. Matrix Effect and Assay of Multiple Analytes. The matrix effect issues in quantitative HPLC-MS/MS may be especially complex when bioanalysis of multiple analytes in the same analytical run are considered. The absence of matrix effect for all individual analytes may need to be demonstrated. The development of methods free from matrix effect for multiple analytes may require some considerable effort in designing proper chromatographic conditions, sample extraction procedure, proper choice of HPLCMS interface, and internal standards. The decision about performing simultaneous assays for a number of analytes using HPLCMS/MS should not be made easily since the data generated may not be as reliable as it is commonly perceived. 3. Elimination of Matrix Effect. Matrix effect may be eliminated or minimized by the following: (1) changing and improving sample extraction procedure and by eliminating undetected matrix interferences, (2) performing the assay under more efficient chromatographic conditions to separate analytes of interest from undetected endogenous compounds that may affect the efficiency of ionization of analytes, and (3) evaluating and changing the HPLC-MS interface and the mechanism of ionization of analytes. Whenever any change in the above parameters is made, the matrix effect should be reevaluated and its absence should be confirmed before analysis of “real” samples is undertaken. 4. Matrix Effect and the Use of Stable Isotope-Labeled Internal Standards. Use of stable isotope-labeled analogues as internal standard is highly recommended since matrix effect should not (15) Chavez-Eng, C. M.; Constanzer, M. L.; Matuszewski, B. K. J. Chromatogr., B 2002, 767, 117-129.

3030

Analytical Chemistry, Vol. 75, No. 13, July 1, 2003

affect the relative efficiency of ionization of the drug and its stable isotope-labeled IS. There are a number of analytical issues connected with the use of labeled internal standards in bioanalysis.15 They include the problems with isotopic purity of compounds, “cross-contamination” or “cross-talk” between MS/MS channels used for monitoring the drug and IS, isotopic integrity of the label in biological fluid and during sample processing, etc. These issues need to be carefully addressed and require separate studies. CONCLUSIONS Careful assessment of matrix effect should constitute an integral and important part of validation of any quantitative HPLCMS/MS method utilized for supporting PK studies. The precision and accuracy of the method should be assessed using biofluids from different sources (subjects), and a relative matrix effect should be evaluated by analyzing biofluid extracts from different sources (lots) spiked with analytes after extraction. The extent of matrix effect seems to be highly dependent on the mechanism of ionization in the HPLC-MS interface. Under otherwise identical sample extraction and chromatographic conditions, the relative matrix effect for compounds studied in this paper was not observed when the APCI (HN) interface was utilized but was very significant when the ESI (ISP) interface was employed. The validity and integrity of quantitative data obtained using HPLC-MS/MS should be carefully verified by demonstrating the absence of matrix effect, interference from metabolites in postdose samples, and absence of “cross-talk” effect. The strategies described in this paper may provide guidance for establishing selective, quantitative bioanalytical methods based on HPLC-MS/ MS. ACKNOWLEDGMENT The authors thank Mr. M. Schwartz for his help in graphical presentations of some data. Thanks are also due to Dr. E. J. Woolf for the critical review of the manuscript. Received for review May 31, 2002. Accepted April 10, 2003. AC020361S