Identifying, Evaluating, and Controlling Bioanalytical Risks Resulting

Nov 1, 2010 - Therefore, the risk of nonuniform matrix ion suppression/ enhancement is usually deemed low when an SIL-IS is used. Here, we have identi...
0 downloads 0 Views 1MB Size
Download PDF
Anal. Chem. 2010, 82, 9671–9677

Identifying, Evaluating, and Controlling Bioanalytical Risks Resulting from Nonuniform Matrix Ion Suppression/Enhancement and Nonlinear Liquid Chromatography-Mass Spectrometry Assay Response Guowen Liu,* Qin C. Ji,* and Mark E. Arnold Bioanalytical Sciences, Bristol-Myers Squibb Company, Route 206 and Province Line Road, Princeton, New Jersey 08543, United States Matrix ion suppression/enhancement is a well-observed and discussed phenomenon in electrospray ionization mass spectrometry. Nonuniform matrix ion suppression/ enhancement across different types of samples in an analytical run is widely believed to be well compensated for by using a stable isotope-labeled internal standard (SIL-IS) in bioanalysis using liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS). Therefore, the risk of nonuniform matrix ion suppression/ enhancement is usually deemed low when an SIL-IS is used. Here, we have identified, evaluated, and proposed solutions to control bioanalytical risks from nonuniform matrix ion suppression/enhancement even with an SILIS through a case study using omeprazole. Two lots of human blank urine were tested, and ion enhancement of about 500% for omeprazole was observed in one lot but not in the other. When a quadratic regression model had to be used, the assay failed the industry acceptance criteria due to unacceptable positive bias for the middle and high quality control (QC) samples. The failure was attributed to different extents of matrix ion enhancement between the standards (STDs) and QCs, which resulted in the misaligned results from the regression model. It was concluded that, for the same amount of drug, nonuniform ion enhancement for different types of samples (STD or QC) resulted in different ion intensities, therefore leading to different response behaviors (linear or nonlinear) at the mass spectrometer detector. A simplified mathematical model was used to evaluate the risk when unmatched response models occurred for different types of samples. A diagnostic factor Q (Q ) XULOQ(-A/B)) was proposed to monitor the risks, where XULOQ is the upper limit of quantitation of the assay, A is the quadratic slope of the curve, and B is the linear slope of the curve. The potential maximum errors were estimated on the basis of the mathematical model for different scenarios, and Q values were given to control * Corresponding authors. E-mail: [email protected] (G.L.); Qin.Ji@ bms.com (Q.C.J.). 10.1021/ac1013018  2010 American Chemical Society Published on Web 11/01/2010

the risks under these conditions for bioanalysis using LC-MS/MS. Liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) has been the dominant technology for small molecule bioanalysis for at least two decades. Matrix ion suppression/enhancement of a specific analyte during ionization, due to competition or enhancement from coeluting matrix molecules, is a widely observed phenomenon in modern mass spectrometry.1-3 The impacts of matrix ion suppression/enhancement on the quantitation of small molecules using LC-MS/MS have been well discussed and documented.4-8 The use of a stable isotope-labeled internal standard (SIL-IS) is believed to be the optimal solution to these problems. SIL-ISs, such as deuterium-, 13C-, and 15Nlabeled analogues, are very close to their corresponding analytes in both physical and chemical properties. Although there are limited cases9,10 reported in which an SIL-IS was used but failed to compensate matrix effects due to a slight difference in the LC elution time between the analyte and the IS, the effect of matrix ion suppression/enhancement is expected to be effectively canceled out by the SIL-IS as long as they are coeluted from the LC column. Therefore, an SIL-IS is highly recommended for regulated bioanalysis.11,12 (1) Fu, I.; Woolf, E. J.; Matuszewski, B. K. J. Pharm. Biomed. Anal. 1998, 18, 347–57. (2) Matuszewski, B. K.; Constanzer, M. L.; Chavez-Eng, C. M. Anal. Chem. 1998, 70, 882–889. (3) Kebarle, P.; Tang, L. Anal. Chem. 1993, 65, 972A–986A. (4) Hernandez, F.; Sancho, J. V.; Pozo, O. J. Anal. Bioanal. Chem. 2005, 382, 934–946. (5) Jemal, M.; Xia, Y. Q. Curr. Drug Metab. 2006, 7, 491–502. (6) Van Eeckhaut, A.; Lanckmans, K.; Sarre, S.; Smolders, I.; Michotte, Y. J. Chromatogr., B: Anal. Technol. Biomed. Life Sci. 2009, 877, 2198–2207. (7) Jemal, M.; Ouyang, Z.; Xia, Y. Q. Biomed. Chromatogr. 2010, 24, 2–19. (8) Remane, D.; Wissenbach, D. K.; Meyer, M. R.; Maurer, H. H. Rapid Commun. Mass Spectrom. 2010, 24, 859–867. (9) Wang, S.; Cyronak, M.; Yang, E. J. Pharm. Biomed. Anal. 2007, 43, 701– 707. (10) Jemal, M.; Schuster, A.; Whigan, D. B. Rapid Commun. Mass Spectrom. 2003, 17, 1723–1734. (11) Matuszewski, B. K. J. Chromatogr., B: Anal. Technol. Biomed. Life Sci. 2006, 830, 293–300. (12) Matuszewski, B. K.; Constanzer, M. L.; Chavez-Eng, C. M. Anal. Chem. 2003, 75, 3019–3030.

Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

9671

Figure 1. Flowchart of the bioanalysis process using LC-MS/MS.

It is known that mass spectrometers become a nonlinear device when the concentration or the ion intensity generated from a target compound exceeds a certain level.13 Saturation during ionization is one of the main reasons for the nonlinear response,14 and detector saturation is believed to be another factor.15 Using an SIL-IS in a bioanalytical method is expected to compensate for the nonlinear effect in the ion source. As shown in Figure 1, when an SIL-IS is used, it is expected to track the analyte throughout the whole sample preparation process and in some analysis steps such as dilution, pipetting, extraction, drying, reconstituting, injection, chromatography, and the ionization process. The concentration ratio of an analyte to its IS for a specific sample is expected to be unchanged from the start of the sample processing to the point it enters the ion source. During the ionization process, the ratio of the ions generated from the analyte and its coeluting SIL-IS is the same as their concentration ratio, no matter whether there is significant matrix ion suppression/ enhancement or not. However, the absolute amount of ions generated could be significantly different due to matrix ion suppression/enhancement. When the subsequent steps are nonlinear, the observed ratio of the analyte and its IS may change. This can be easily understood from Figure 1, where the ions generated from the analyte and the IS are separated after ionization. The nonlinear effects of the MS detector can no longer be compensated for even by an SIL-IS, which may lead to a nonlinear calibration curve. Due to this nature of MS, quadratic regression models are sometimes used to address the nonlinear response behavior.16,17 For a set of samples, matrix ion suppression/enhancement may happen to one group of samples but not to another group of (13) Tang, K.; Page, J. S.; Smith, R. D. J. Am. Soc. Mass Spectrom. 2004, 15, 1416–1423. (14) Enke, C. G. Anal. Chem. 1997, 69, 4885–4893. (15) Kim, T.; Tang, K.; Udseth, H. R.; Smith, R. D. Anal. Chem. 2001, 73, 4162– 4170. (16) Pruvost, A.; Levi, M.; Zouhiri, F.; Menier, I.; Benech, H. J. Chromatogr., B: Anal. Technol. Biomed. Life Sci. 2007, 850, 259–266. (17) Xue, Y. J.; Yan, J. H.; Arnold, M.; Grasela, D.; Unger, S. J. Sep. Sci. 2007, 30, 1267–1275.

9672

Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

samples, or the degree of matrix ion suppression/enhancement may vary among different groups of samples. This can be defined as “nonuniform matrix ion suppression/enhancement”. This phenomenon is not uncommon when samples are generated from different matrix sources and sample preparation steps do not remove matrix components responsible for the matrix effect, for example, when standards (STDs) are prepared in one lot of biological matrix and quality controls (QCs) are prepared in another lot of biological matrix, or when incurred samples are collected from distinct patient populations or from different dose vehicle groups, etc. This phenomenon was observed before, but its impact on bioanalysis has not been fully discussed or evaluated, especially when an SIL-IS was used. In addition to its main function to track the analyte, the IS has also been used to diagnose assay problems.18 When the method is executed properly, the IS responses are a very good indicator of nonuniform matrix ion suppression/enhancement throughout an analytical run. Tan et al.18 have summarized multiple cases regarding internal standard response variations and the implication of these variations. In this study, bioanalytical risks from nonuniform matrix ion suppression/ enhancement using an SIL-IS and a nonlinear response are evaluated through an example where dramatic differences in the matrix ion enhancement were observed between the STD and QC samples, which were prepared in two different lots of human urine. A validation run failed due to unacceptable bias for the middle- and high-concentration QCs, which was not caused by sample preparation errors. It is worth mentioning that the nonuniform matrix ion enhancement across different lots of human urine was initially observed on Bristol-Myers Squibb Co. (BMS) proprietary compounds in a real life study. However, the structures for the BMS compound and its SIL-IS cannot be disclosed. Therefore, here we use omeprazole and d3-omeprazole (SIL-IS) to demonstrate this observation and discuss its impacts. (18) Tan, A.; Hussain, S.; Musuku, A.; Masse, R. J. Chromatogr., B: Anal. Technol. Biomed. Life Sci. 2009, 877, 3201–3209.

EXPERIMENTAL DETAILS Chemicals, Reagents, Materials, and Apparatus. HPLC grade acetonitrile was purchased from J.T. Baker (Phillipsburg, NJ). Deionized water was generated using the NANOpure Diamond ultrapure water system from Barnstead International (Dubuque, IA). Formic acid (SupraPur grade) was purchased from EMD Chemicals (Gibbstown, NJ). Human blank urine was collected in house (lots A and B). Omeprazole and d3-omeprazole were obtained from Sigma-Aldrich (St. Louis, MO) and C-D-N Isotopes (Quebec, Canada), respectively. The specific chemical structures for omeprazole and d3-omeprazole are shown in Figure 1s in the Supporting Information. Vials used for standard and quality control samples were 2.0 mL polypropylene vials from VWR International (Bridgewater, NJ). A JANUS automated workstation from Perkin-Elmer (Waltham, MA) was used for adding or transferring solutions during sample preparation. LC-MS/MS Equipment. All sample analyses were performed on a Sciex API 4000 triple-quadrupole mass spectrometer (Applied Biosystems/MDS SCIEX, Concord, Ontario, Canada), which was controlled by Analyst 1.4.2 software. This software was also used for data acquisition and processing. The mass spectrometer was directly coupled with an HPLC system from Leap Technology (Carrboro, NC), which consisted of a Leap HTC-PAL autosampler, a Flux 4x Ultra mobile-phase delivery pump, and a HotDog 5090 column heater from Prolab (Reinach, Switzerland). The separation was achieved on a Luna silica column (5 µm, 100 × 2 mm) from Phenomenex (Torrance, CA). Sample Preparation and Methods. Sample Preparation. STDs and QCs were prepared in either blank human urine lot A or blank human urine lot B, which were collected in house at different times. In general, the curve range was from 1 to 1000 ng/mL. Eight levels of standard solution in human urine were prepared with concentrations at 1, 2.5, 10, 50, 200, 500, 750, and 1000 ng/mL. Five levels of QCs, lower limit of quantitation (LLOQ), low QC (LQC), geometric QC (GMQC), middle QC (MQC), and high QC (HQC), in human urine were prepared with concentrations at 1, 3, 40, 500, and 850 ng/mL, respectively. Three sets of samples (R, β, γ) containing STDs and QCs were prepared. STDs were prepared in urine lot A and QCs prepared in urine lot B for set R. Both STDs and QCs were prepared in urine lot A for set β and in urine lot B for set γ. Another set of samples (δ) were prepared the same way as set R, but the concentration of omeprazole in each sample was only one-sixth of that in set R for both STDs and QCs. All samples were processed by dilution. Specifically, 50 µL of human urine sample was mixed with 50 µL of internal standard working solution (100 ng/mL in acetonitrile for sets R, β, and γ and 16.7 ng/mL for set δ). The mixture was then diluted 5-fold with acetonitrile before injection of 10 µL of the diluted samples onto the LC-MS/MS system for analysis. All the final diluted samples before injection contained 10% urine matrix. A deuterium-labeled internal standard (d3-ompeprazole) was used throughout the analysis. LC-MS/MS Method. Chromatographic separation was achieved with isocratic elution and a quick step gradient wash. Specifically, mobile phase A was 0.1% formic acid in deionized water, and mobile phase B was 0.1% formic acid in acetonitrile/water (95/5, v/v). The HPLC gradient started with 90% B for 0.80 min, changed to 0% B in 0.05 min, stayed at 0% B for 0.5 min, switched back to

90% B in 0.05 min, and then stayed at 90% B until the next injection. The flow rate was 600 µL/min, and the retention time for omeprazole and d3-omeprazole was 0.67 min. The mass spectrometer was used in positive ion electrospray selective reaction monitor (SRM) mode for detection. The critical parameters for the mass spectrometer were set as the following. The transitions monitored for omeprazole and d3-omeprazole were 346 > 198 and 349 > 198, respectively. The turbo ion spray voltage was set at 4000 V. The declustering potential was set at 50 V, and the collision energy used for both compounds was 18 eV. The temperature for the ion source was 650 °C. The total run time was 2.0 min. RESULTS AND DISCUSSION Nonuniform Matrix Effects Were Observed for Samples Prepared in Different Human Urine Lots. Although the same levels of matrix ion enhancement were observed for both omeprazole and d3-omeprazole within one sample, different matrix ion enhancement effects were observed across different samples prepared in two different lots of human urine. Parts A and B of Figure 2 show the IS responses for samples in sets R and β, respectively. As shown in Figure 2A, the IS responses for STDs in urine lot A were more than 5 times that from the QC in urine lot B. As expected, the IS responses were consistent in Figure 2B across all samples. The dramatic difference observed in Figure 2A was believed to be caused by a matrix effect. Following the experiment setup by King et al.19 for matrix effect evaluation, a postcolumn infusion experiment was carried out to identify the type of matrix effect (ion enhancement or suppression). The same LC-MS/MS conditions for the quantitation of omeprazole were used. Specifically, 100 ng/mL omeprazole was continuously infused through a postcolumn T-connection to the mass spectrometer at a flow rate of 150 µL/min. At the same time, three samples (A, B, and C) were injected onto the LC-MS/MS system. Samples A and B were urine lots A and B diluted 10-fold using acetonitrile. Sample C was a neat control sample containing acetonitrile only. As shown in Figure 3, extensive ion enhancement regions were found in the chromatograms for both samples A and B, with some difference between these two samples. A typical LC-MS/MS chromatogram of omeprazole was also added to Figure 3 to show its retention time. A close look at Figure 3 revealed that, at the retention time (∼0.67 min) where omeprazole was eluted, there was significant ion enhancement (∼500%) for omeprazole from urine lot A. However, there was no ion enhancement effect but some degree of ion suppression from urine lot B. The results clearly explained the observed IS response difference in Figure 2A. Positive Bias for High Concentration Levels of QCs Were Observed in a Set of Samples in Which STDs and QCs Were Prepared in Different Lots of Urine. A quadratic calibration curve with 1/X2 weighting had to be selected to best fit the STDs in set R due to their quadratic responses on the LC-MS/ MS system. The quadratic equation obtained for STDs in set R is Y ) (-2.99E-6)X2 + 0.00852X + 0.000559 (r ) 0.9995), where Y is the relative instrument response of omeprazole to d3-omeprazole and X is the concentration of omeprazole. As (19) Bonfiglio, R.; King, R. C.; Olah, T. V.; Merkle, K. Rapid Commun. Mass Spectrom. 1999, 13, 1175–1185.

Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

9673

Figure 2. IS peak area plots of two sets of samples showing nonuniform matrix effects: (A) set R, STDs prepared in urine lot A and QCs prepared in urine lot B, (B) set β, both STDs and QCs prepared in urine lot A (points near the x-axis are double blank samples, which contain no IS or analyte).

Figure 3. Postcolumn infusion chromatograms of omeprazole after injection of different lots of diluted human blank urine or acetonitrile with a superimposed chromatogram of omeprazole.

shown in Table 1 for set R, the accuracy for all standards, LLOQ, LQC, and GMQC was within 10%. However, unacceptable positive biases were observed for the MQCs and HQCs. Specifically, for MQCs, an average positive bias of 24.5% was observed with very good precision (CV ) 3.0%). For HQCs, their values were all EQB (exceed quadratic boundary). No concentration value can be calculated on the basis of the quadratic equation. The quadratic boundary corresponds to an accuracy of HQC of 167.4%. This indicates that the positive bias for all HQCs was more than 67.4%. Sample preparation error was excluded after an initial troubleshooting. A matrix effect or quadratic response was therefore suspected to be the reason for this observation. Since an SIL-IS was used throughout the sample analysis and no separation was observed between omeprazole and d3-omeprazole, the matrix ion enhancement/suppression was initially believed to be well compensated on the basis of our previous understanding of the SIL-IS. Quadratic Response Alone Was Not the Cause of the Positive Bias. An experiment was carried out to test the quadratic response effect. As described in the section “Sample Preparation and Methods”, a set of samples (set β) were prepared and tested, in which both STDs and QCs were prepared in urine lot A. A 9674

Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

Table 1. Accuracy and Precision for STDs and QCs in Two Different Sample Sets STD accuracy (%) STD

1

2

STD1 STD2.5 STD10 STD50 STD200 STD500 STD750 STD1000

102 97.6 103 102 99.8 93.2 102 107

98 99.9 102 102 96.3 95.4 97.8 109

STD1 STD2.5 STD10 STD50 STD200 STD500 STD750 STD1000

99.1 95.9 102 103 98.7 93.5 107 102

102 100 102 101 97.5 96.2 98.8 105

QC accuracy and precision (n ) 6) QC level

average precision (%)

CV (%)

Set R LLOQ LQC GMQC MQC HQC

104.3 98.0 98.7 124.5 >167.4

4.7 2.6 2.0 3.0 N/A

Set β LLOQ LQC GMQC MQC HQC

99.8 98.7 102.8 96.9 101.4

1.7 2.3 2.7 2.0 3.1

quadratic regression model also had to be selected for the STDs due to their quadratic responses on the LC-MS/MS system for

Figure 4. Accuracy and precision of QCs at each level (n ) 6) from four sets of samples: (A) set R, STDs in urine lot A and QCs in urine lot B with quadratic responses on LC-MS for STDs and linear responses for QCs, (B) set β, STDs and QCs in urine lot A with quadratic responses on LC-MS/MS for both STDs and QCs, (C) set γ, STDs and QCs in urine lot B with linear responses on LC-MS/MS for both STDs and QCs, (D) set δ, STDs in urine lot A and QCs in urine lot B with linear responses on LC-MS/MS for both STDs and QCs. The broken line shows the (15% acceptance limits. Note (*) that all HQCs in set R were EQB (exceed quadratic boundary). No concentration value can be calculated on the basis of the quadratic equation. The quadratic boundary equals an accuracy of 167.4% for HQC. This indicates that the average accuracy for HQCs in set γ is larger than 167.4%.

omeprazole. A good quadratic curve (Y ) (-2.91E-6)X2 + 0.00844X + 0.000708, r ) 0.9995) was obtained for the STDs. As shown in Table 1 for set β, the accuracy for both STDs and QCs is within 10% and the CV for all QCs is within 3.1%. These results clearly suggested that a quadratic response alone did not lead to the unacceptable positive biases for MQCs and HQCs in set R. Therefore, the nonuniform matrix effect observed in set R was to be evaluated for the unacceptable positive biases. It is also worth mentioning here that this experiment also indicates that the quadratic regression model is a valid model to use when there is no nonuniform matrix effect. Nonuniform Matrix Ion Enhancement Was Not the Only Cause of the Positive Bias. Another experiment was therefore carried out to test the nonuniform matrix effect excluding the quadratic response factor on the observed positive bias. As described in the section “Sample Preparation and Methods”, a set of samples (set δ) including STDs in urine lot A and QCs in urine lot B were prepared and tested. To avoid signal saturation on the LC-MS/MS system, the concentration of omeprazole in STDs and QCs for set δ was only one-sixth of that for set R. A good linear calibration curve (Y ) 0.00811X + 0.00123, r ) 0.9981) was obtained for STDs in set δ. Meanwhile, the same level of nonuniform matrix effect was observed in set δ (Figure 2s; see the Supporting Information for details) as in set R (Figure 2A). Positive biases were no longer observed in MQCs and HQCs (Figure 4D). It is worthwhile to mention that the variation of the quantitation of the LLOQ and LQC has increased significantly for set δ (Figure 4D) due to their weak instrument responses. On the basis of the results of this experiment, the nonuniform matrix effect alone was not the cause for the unacceptable positive bias either.

Figure 5. Calibration curves of standards from sets β and γ: (A) set β, standards prepared in urine lot A, (B) set γ, standards prepared in urine lot B.

Combination of Nonuniform Matrix Ion Enhancement and Nonlinear Response (Detector Saturation) Caused the Assay Failure. We have demonstrated that neither the nonuniform matrix effect or nonlinear responses alone led to the unacceptable positive response for the MQC and HQC in set R. Combination of both factors could be the real cause of the problem. As shown in Figure 1, although the ratios of the ions generated from the analyte to those generated from its IS for samples at the same concentration level are the same, the absolute ion intensities of the analyte and its IS for different samples are dramatically different due to nonuniform matrix ion enhancement. If the subsequent event (detection) is not linear, the observed ratio of the analyte to the IS at the detector may change. Especially when the detector was saturated for the analyte but not for the IS (the IS was at a lower concentration and therefore lower absolute ion abundance), the IS was no longer tracking and compensating for the analyte. In our experiments shown in Figure 5, the responses of omeprazole for the high-concentration STDs prepared in urine lot A (set β) fell into the nonlinear dynamic range of the detector due to dramatic ion enhancement but the d3-omeprazole responses were always in the linear dynamic range due to a lower concentration; therefore, a quadratic calibration curve (Y ) (-2.91E-6)X2 + 0.00844X + 0.000708, r ) 0.9995) had to be used to fit the STDs in set β. However, the responses of omeprazole were always in the linear dynamic range of the detector for all the STDs prepared in urine lot B (set γ), which had no ion enhancement but rather a small amount of ion suppression (see Figure 3). Therefore, a good linear calibration curve (Y ) 0.00808X + 0.000984, r ) 0.9993) was obtained for the STDs in set γ. In this case, samples prepared in different urine lots had different response behaviors on the LC-MS/MS system. The mismatched response behaviors, resulting from a nonuniform matrix effect and nonlinear response, caused the unacceptable biases for MQCs and HQCs in set R. As shown in Figure 5, two calibration curves from urine lots A and B converge at the low end but diverge at the high end. This explains not only the increasing unacceptable positive biases for MQC and HQC but also no bias at the low end (LLOQ, LQC, and GMQC). This problem may be overlooked during our daily work. Only when one set of the samples, either the STDs or the QCs, exhibit detector saturation will scientists be aware of this issue due to the failure of an analytical run. However, if this were to occur with incurred study samples, where their performance is different from that of the standards and QCs, inaccurate quantitation of Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

9675

Figure 6. Theoretical errors derived from two unmatched regression models. (A) STDs were fitted to a quadratic curve, but incurred samples were better fitted to a linear curve. (B) STDs were fitted to a linear curve, but incurred samples were better fitted to a quadratic curve.

the incurred samples would occur. Without vigilant evaluation of the IS responses among all samples, the problem could then be overlooked by the scientist, as the analytical run will still meet the acceptance criteria. Even the incurred sample reanalysis (ISR) test, which is used to confirm the bioanalysis results, will likely pass, but the actual results obtained for the incurred sample will be incorrect. In our opinion, this could have a significant impact on bioanalysis. It will have a bigger impact on the highconcentration samples than on the low-concentration samples, and in some cases, the high-concentration samples will have a much bigger impact on the readout of the drug exposure. This issue has to be mitigated and addressed appropriately during the method development and validation process. Risk Management and Recommendations. IS Response Monitoring and Control. As discussed the section “Combination of Nonuniform Matrix Ion Enhancement and Nonlinear Response (Detector Saturation) Caused the Assay Failure”, nonuniform matrix ion suppression/enhancement between samples may lead to the responses of the analyte and IS falling into different response regions (linear or nonlinear) of the mass spectrometer. The matrix ion suppression/enhancement, especially as it applies to significant differences seen among samples, of an analytical method should be monitored during method development and validation and throughout sample analysis. When the method is executed properly, the IS response is a very good indicator of nonuniform matrix ion suppression/enhancements.18 One practice across the pharmaceutical industry in bioanalysis today is to monitor and control the IS response variations, e.g., establish acceptance criteria for IS response variations for each analytical run. It is important to point out that one can minimize the sample to sample differences in matrix effects by proper sample cleanup and thereby achieve more consistent results for all samples. Thus, proper method development and testing is always favored and will reduce the chances that different types of samples (STD, QC, 9676

Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

or incurred samples) fall into different calibration/response categories. This will certainly help to minimize the risk of overor undercalculating the results of incurred samples. Quadratic Curve Parameters and Instrument Absolute Response Monitoring. Both quadratic and linear calibration curves are currently being used across the industry for LC-MS/MS assays. There are always some debates/discussions about which model is a better model to use. The current FDA guidance does not rule out the use of either one, but does recommend going for the simplest if possible. The linear regression model is favored due to its simplicity. However, for a given set of data, a quadratic model will fit the data better than a simple linear model because it takes into account more variables. A statistical tool should be used to assess the fit of the curve to the data sets and the selection of a weighting function. However, no matter which model is selected, the risk of different instrument responses resulting from nonuniform matrix ion suppression/enhancement always exists. As shown in Figure 6A, when the responses of the standards are quadratic but the responses for the incurred samples are in the linear response region, the results of the incurred samples at high concentration levels will be overestimated. As shown in Figure 6B, when the responses of the standards are linear but the responses for the incurred samples are quadratic, the results of the incurred samples at high concentration levels will be underestimated. Therefore, no matter which regression model is selected, the risks have to be evaluated and controlled when nonuniform matrix ion suppression/enhancement is suggested by the IS response variation. Here, we use our case as an example to discuss some risks related to different regression models. To simplify our discussion, we have made the following assumptions. First, we assume two extreme cases; one group of samples (STDs or incurred samples) are in the quadratic response region, and the other group of samples (STDs or incurred samples) are in the linear response

region. Second, we use the equation of the quadratic model as Y ) AX2 + BX + C and the equation for the linear model as Y ) BX + C, where Y is the area ratio of analyte to IS and X is the concentration of the analyte. We assume the parameters B and C are the same in both cases, which should be reasonable since two curves converge at the low end. As shown in Figure 6A, the error can be expressed as -AX2/(AX2 + B), with the biggest error of -AXULOQ/(AXULOQ + B) (when X2 ) XULOQ, the upper limit of quantitation (ULOQ)). If we give an error tolerance of +15%, then we can derive that XULOQ(-A/B) should not exceed 0.13. Here, we propose to use XULOQ(-A/B) as a diagnostic factor for risk evaluation and control. For convenience, we define this as the Q factor. When this is applied to our case for sample set R, the calculated Q value is 0.34. When Q exceeds 0.13, scientists usually have two options to decrease it: (1) reduce the instrument sensitivity (decrease A) or (2) reduce the upper limit of quantitation (decrease XULOQ). Similarly, in case B, the biggest percentage error is XULOQ(-A/ B). If we give an error tolerance of -15%, then Q should not exceed 0.15. (Please see Supporting Information Figure 3s for the details of the calculations.) Therefore, when a quadratic calibration curve is selected and nonuniform matrix ion suppression/enhancement is observed, the Q value can be used to evaluate the potential risk. As shown in Figure 6B, a linear regression model is not always a safe haven, but the problem is even more difficult to identify because this may only affect the unknown samples, so the STDs and QCs are always fine. To minimize this risk, we recommend examining the absolute response for the analyte of all samples in an analytical run when nonuniform matrix ion suppression/enhancement is suggested. The absolute response of any accepted sample (not including samples that have a value larger than the ULOQ) should not be significantly higher than the absolute response of the ULOQ sample of the standards. If in doubt, the affected samples should be reanalyzed with proper dilution. CONCLUSIONS Nonuniform matrix ion enhancement was observed when STDs and QCs were prepared in different urine lots. Ion enhancement, ∼500%, was observed in the STDs compared to QC samples. This is due to different levels of endogenous components within the different lots of biological matrix being used for STD and QC preparation. A quadratic regression model was selected and fit the STDs very well. The assay failed the acceptance criteria due to unacceptable positive bias for the middle and high QCs. The failure of the assay was attributed to a combination effect of a nonuniform matrix effect and a nonlinear response over the curve range. Here, we have demonstrated that an SIL-IS is not always

a safe haven and does not compensate for all aspects of MS/MSbased bioanalytical assays. Specifically, the nonlinear response generated by detector saturation cannot be compensated for by an SIL-IS. Therefore, matrix ion suppression/enhancement (especially nonuniform matrix ion suppression/enhancement across different samples) should be more carefully evaluated not only for its impact on source conditions but also for the potential for downstream impacts at the detector. Different regression models were discussed regarding the impacts on the final results. A diagnostic factor Q (Q ) XULOQ(-A/B)) was proposed to evaluate the potential risk. When nonuniform matrix ion suppression/enhancement is suggested by the IS response variation and a quadratic regression model is selected for the standard curve, it is recommended to examine the Q value. A Q value of 0.13 could give an error up to 15%. While a linear regression model is selected for the standards, it is suggested to monitor the instrument absolute responses of the incurred samples, especially for those samples with unusually higher IS response. When the IS response is unusually high, it could likely be caused by the matrix ion enhancement. Also, the analyte response could also be elevated and fall into the nonlinear response region if the concentration of the analyte in the sample is on the high end of the curve. It is worth mentioning that this observation was timely identified during method development because the standard and quality control samples were prepared in different lots of blank matrix. Although preparing standard and quality control samples in different lots of biological matrix is not required by the current Food and Drug Administration Bioanalytical Method Validation Guidance (2001), we strongly recommend adopting this practice to identify any risk as early as possible. It is also worth mentioning that when good sample cleanup is performed during sample preparation, the chance for matrix effects or nonuniform matrix effects across different types of samples (STDs, QCs, or incurred samples) during LC-MS/MS analysis will be fairly low. Therefore, a good sample cleanup is always desired, even when an SIL-IS is used. ACKNOWLEDGMENT We thank Heidi Snapp, Dr. Anne-Francoise Aubry, and Dr. Mohammed Jemal for helping review the manuscript. SUPPORTING INFORMATION AVAILABLE Additional information as noted in the text. This material is available free of charge via the Internet at http://pubs.acs.org. Received for review May 19, 2010. Accepted October 13, 2010. AC1013018

Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

9677