An isotope dilution-liquid chromatography-tandem mass spectrometry (ID-LC-MS/MS)-based candidate reference measurement procedure for the quantification of zonisamide in human serum and plasma

Chemicals and reagents

LC-MS grade methanol (CAS 67-56-1) was purchased from Biosolve (Valkenswaard, The Netherlands). Dimethyl sulfoxide (DMSO) (CAS 67-68-5, ACS reagent, ≥99.99 %) and acetic acid (CAS 64-19-7) were bought from Sigma Aldrich (Taufkirchen, Germany). Isopropanol (CAS 67-63-0, high-performance liquid chromatography [HPLC] grade) was purchased from Riedel-de Haën (Seelze, Germany). Water was purified using a Millipore Milli Q 3 UV system from Merck. Native human serum (Art No. S1-Liter) was obtained from Merck (Darmstadt, Germany), TDM-free human serum (surrogate matrix) was obtained from Roche Diagnostics GmbH (ID No.12095432001, Mannheim, Germany), native plasma matrix (Li-heparin, dipotassium (K₂)-ethylene diamine tetraacetic acid (EDTA), and tripotassium (K₃) EDTA) was obtained from anonymized leftover patient samples. Pools were prepared from them in accordance with the Declaration of Helsinki.

Zonisamide (CAS 68291-97-4, Art. No. MM1379.00, Lot No. W997903) was bought from LGC Mikromol (Luckenwalde, Germany) and its isotope labeled ISTD [²H₄,¹⁵N]-zonisamide (Art. No. C3209, Lot No. MS-ALS-17-128-P3) from Alsachim (Illkirch Graffenstaden, France). NMR solvent DMSO-d₆ and the qNMR internal standard 1,3,5-trimethoxybenzene (TraceCert qNMR certified reference material, Art No. 74599, Lot No. BCBW3670) were obtained from Sigma Aldrich Germany. Five millimeter 10-inch NMR tubes were purchased from Sigma Aldrich and/or Euroisotop GmbH (Hadfield, United Kingdom).

qNMR for determination of the purity of the standard materials

qNMR measurements were performed on a Jeol 600 MHz NMR with a He-cooled Cryoprobe. Single-Pulse-¹HNMR (Supplementary Figure 1) was utilized for the quantitation (Methylene protons, CH ₂ SO₂NH₂, 2H) with an inter-scan delay of 70 s. These methylene protons are best suited as the quantifying signals since the 166 most susceptible functional group, to hydrolysis, –NH₂ functionalization, cyclization reactions, etc., is the –SO₂NH₂ group and any chemical modification here would have direct influence on the chemical shift value of this singlet resonance. The qNMR internal standard, 1,3,5-trimethoxybenzene (Content: 99.96 ± 0.065 (k=1)), is a TraceCert CRM available from SigmaAldrich and as per the accompanying certificate of analysis (CoA), this material is directly traceable to NIST PS1 (primary qNMR standard). Additional details about NMR acquisition and FID processing parameters can be found in the Supplementary Material 2.

Preparation of calibrators and quality control (QC) samples

Two calibrator stock solutions were used to prepare eight matrix-based calibrators. Per stock solution, 60 mg of zonisamide was weighed in tin boats on a microbalance (XPR2, Mettler Toledo, Columbus, Ohio, USA). The substance was then dissolved in 5 mL DMSO using a volumetric flask to reach a concentration of 12.0 mg/mL. The concentrations of the stock solutions were calculated based on the purity of the reference material (99.4 ± 0.1 %, determined by qNMR) and the amount weighed. From these stock solutions, another two working solutions were prepared.

Two working solutions were prepared from these stock solutions. Eight spike solutions were prepared by diluting the stock and working solutions, which were then further used to prepare final matrix-based calibrator levels. Therefore, calibrator spike solutions were volumetrically diluted 1 + 99 (v/v) in native human serum matrix to produce matrix-based calibrators uniformly distributed from 1.50 to 60.0 μg/mL (7.07–283 μmol/L) (see Figure 1).

Figure 1:

Schematic overview of zonisamide calibrator and control levels chosen to allow optimal coverage of measurement and therapeutic reference range. Black circles, calibration samples 1–8; black triangles, QC levels 1–4; black diamond, alert level; black line, measurement range; dotted black line, therapeutic reference range. Conversion factor µg/mL to µmol/L: 4.7.

In addition, four matrix-based QC levels were prepared by weighing 50 mg zonisamide in tin boats. Final QC levels were then prepared in the same way as described for the calibrator levels. The concentrations of the QC levels were set at four critical control points: above the lower limit of quantification (2.60 μg/mL), below (8.00 μg/mL) and within (20.0 μg/mL) the therapeutic reference range, and at the laboratory alert level (40.0 μg/mL) [10] (see Figure 1).

ISTD solution

[²H₄,¹⁵N]-Zonisamide was dissolved in the appropriate amount of DMSO to give a 1 mg/mL stock solution, by volumetric addition of DMSO. The stock solution was stored at −20 °C until further use for a maximum of eight months. The ISTD working solution was prepared freshly on each day of sample preparation by a two-fold dilution of the stock solution: 60 µL DMSO was mixed with 40 µL ISTD stock solution followed by the addition of 3,900 µL Milli-Q-water to obtain a concentration of 10.0 μg/mL.

Sample preparation

Native human serum, TDM free serum (surrogate matrix) and plasma (Li-heparin, K₂-EDTA and K₃-EDTA) were used as sample specimen. First, 100 µL ISTD working solution was mixed with 50 µL sample specimen (native sample/calibrator/QC) in a 2 mL tube (Eppendorf, Hamburg, Germany). Afterwards, 1,000 µL precipitation solution (75 % methanol [v/v]) was added to precipitate proteins. Then, 100 µL of the supernatant was first diluted 1 + 9 (v/v) followed by a second dilution (1 + 39 [v/v]) with mobile phase A.

Liquid chromatography mass spectrometry

An Agilent 1290 Infinity II LC system (Santa Clara, CA, USA), equipped with a binary pump, a vacuum degasser, an autosampler at 7 °C and a column compartment, was used for the chromatographic separation. Zonisamide was separated using an Agilent Zorbax Eclipse XDB-C8 column (100 × 3 mm, 3.5 µm, Santa Clara, CA, USA) which was maintained at 40 °C in the column compartment. The mobile phases consisted of 10 % methanol in Milli-Q-water (v/v) with 0.1 % acetic acid (A) and 95 % methanol in Milli-Q-water (v/v) (B).

All measurements were performed at a flow rate of 0.6 mL/min within a total runtime of 10 min. The injection volume was 5 μL. Contamination of the mass spectrometer was reduced using a divert valve, switching the eluent flow until 0.8 min and from 6.5 min to the waste.

Zonisamide was detected by multiple reaction monitoring (MRM) using an AB Sciex Q-Trap 6500+ mass spectrometer (Framingham, Massachusetts, USA) with a Turbo V ion source operating in negative electrospray ionization mode (ESI – mode). An optimized ion spray voltage of −3500 V and a temperature of 600 °C were applied. Nitrogen gas was used as curtain gas, collision gas, ion gas source 1, and ion gas source 2 and was set at 35, 10, 70 and 60 psi, respectively.

Several analyte-specific mass transitions were tested and optimized for intensity and reproducibility, resulting in the following transitions: m/z 211.1–119.1 (quantifier) and m/z 211.1–147.1 (qualifier). Monitoring of the quantifier/qualifier ratio in native matrix samples compared to the quantifier/qualifier ratio of neat system suitability test (SST) samples allowed checking for interfering substances in matrix samples, which should not differ by more than 20 %. Table 1 shows an overview of the selected reaction transitions as well as the remaining compound-dependent MS settings.

System suitability test

An SST was performed prior to each analysis to verify the system sensitivity, chromatographic performance, and potential carry-over effects. For this purpose, two SST samples, SST sample 1 and SST sample 2, were prepared corresponding to the analyte concentration within the processed calibrator level 1 and 8, respectively.

To pass the SST, the signal-to-noise ratio of the quantifier transition had to be ≥200 for SST sample 1 and retention times for SST sample 1 and 2 had to be within 3.8 min ± 0.5 min. Moreover, carry-over was assessed by injecting SST sample 2 followed by two blank injections. The analyte peak area in the first blank had to be ≤20 % of the analyte peak area of SST sample 1 to pass.

Calibration structure of analytical series and data processing

At the beginning and the end of the analytical series, the calibrators were measured in increasing concentrations. Both calibration functions were used to generate the final calibration function by quadratic regression of the area ratios of the analyte and the ISTD (y) against the analyte concentration (x) resulting in the function, y=a ₀  + a ₁ x + a ₂ x ² . Data processing was performed using Analyst software (version 1.6.3 or higher). Analyte and ISTD peaks were integrated within a retention time window of 3.8 ± 0.5 min using a smoothing and peak splitting factor of three, a noise percentage of 90 %, and a base sub-window of 0.5 min.

Method validation

Assay validation and determination of measurement uncertainty were performed based on existing validation guidelines such as the Clinical & Laboratory Standards Institute’s C62A Liquid Chromatography-Mass Spectrometry Methods [36], the International Council of Harmonization’s (ICH) guidance document Harmonised Tripartite Guideline Validation of Analytical Procedures: Text and Methodology Q2 (R1) [37] and the Guide to the expression of uncertainty in measurement (GUM) [38].

Traceability to SI units

Traceability to the SI unit of mass (kilogram), the most important parameter for a reference measurement method, has been established by the utilization of 1,3,5-trimethoxybezene as a qNMR internal standard which is directly traceable to NIST PS1 (primary qNMR standard) as per the accompanying certificate of analysis (CoA). Furthermore, traceability to the SI unit of amount of substance (mole) is also included owing to NIST PS1. Since Planck’s constant is now the main parameter for defining kilogram and Avogadro’s constant for mole, qNMR methodology is best suited to fulfill both the traceability chains. Additionally, as per the latest IUPAC Technical Report, qNMR has been stated as a potential primary reference measurement procedure ideally suited for the characterization of primary reference materials [29]. Six individual experiments (Supplementary Material 2 Table 1 and Figure 2), involving six individual weightings of the analyte and 1,3,5-trimethoxybenzene (qNMR ISTD) yield a final content value of 99.4 ± 0.1 % (k=1).

Selectivity

The developed gradient combined with a reversed phase column (Agilent Zorbax Eclipse XDB-C8) separated zonisamide well from polar and apolar matrix components with a retention time of 3.8 min (Figure 2). Selectivity was determined by analyzing sample pools of analyte-free native human serum, surrogate serum and native Li-heparin plasma. No signals were observed within the expected retention time window (3.8 ± 0.5 min). Moreover, no residual zonisamide was observed in the ISTD ion traces. Furthermore, the samples measured in the method comparison study showed no interference within the retention time window of zonisamide.

Figure 2:

Zonisamide LC-MS/MS derived analytical readouts. (A) Chromatogram of a matrix blank showing the analyte SRM ion trace (left) and ISTD SRM ion trace (right); (B) chromatogram of calibrator level 1 with a concentration of 1.50 μg/mL spiked in native serum (left) and ISTD (right); (C) pooled patient sample with a concentration of 5.65 μg/mL (left) and ISTD (right).

Matrix effect/specificity

Matrix-dependent effects were avoided by a high dilution of the sample after protein precipitation. This was demonstrated by the post-column infusion experiment, as no change in the ionization field was observed at the expected retention time of zonisamide, independent of the matrix tested.

Furthermore, a comparison of slopes and coefficients of determination in different matrices was performed. Calibrations showed mean slopes (n=6, sample preparations) of 0.0545 (95 % CI 0.0542–0.0549) for the neat solution, 0.0546 (95 % CI 0.0543–0.0550) for the native serum, 0.0547 (95 % CI 0.0544–0.0550) for the surrogate serum and 0.0549 (95 % CI 0.0545–0.0554) for the native Li-heparin plasma. The CIs of the slopes overlap, indicating that they are not significantly different. Moreover, mean r² values were ≥0.99 independent of the matrix used for calibration. Both results confirm the absence of a ME. Furthermore, matrix-based samples, when calibrated against neat, showed no significant bias ranging from −0.9 to 1.3 % with CVs of less than 1.9 %.

In addition, MEs were evaluated based on the approach of Matuszewski et al. [42]. The recoveries of the analyte peak areas ranged from 101 to 108 % and the ISTD peak areas ranged from 101 to 109 % (Table 2). The recoveries are slightly above 100 % indicating an ion enhancement. However, the area ratio recoveries were between 99 and 100 %. This confirms the compensating effect of the labeled ISTD.

Linearity

The linearity of the method was demonstrated by analyzing calibration curves (n=6, sample preparations) extended by ±20 % of the assay measuring range (1.20–72.0 μg/mL). The residuals were randomly distributed in a quadratic regression model and were therefore selected for assay calibration. The correlation coefficients were 1.00 for all individual calibrations. Based on these results, samples 1–11 were evaluated and showed a linear dependency with a correlation coefficient of 1.00. The relative deviation ranged from 0.1 to 1.2 % and the CV was determined to be ≤0.9 %.

Lower limit of the measuring interval and limit of detection

The LLMI was determined using spiked samples at the lowest calibrator level. The relative bias showed a deviation of 0.3 % and the CV was 1.7 % at a concentration of 1.50 μg/mL. The LOD was estimated to be 0.317 μg/mL with an intensity corresponding to approximately one-fifth of the average peak height of calibrator 1.

Precision and accuracy

Precision, accuracy, and trueness were determined using spiked serum and plasma samples at four concentration levels (2.60, 8.00, 20.0 and 40.0 μg/mL). Each level was prepared in triplicate and injected twice by two different operators (n=12) on five different days (n=60). Prior to the sample preparation, two highly concentrated samples (64.0 and 80.0 μg/mL) were diluted with matrix to determine accuracy and trueness for these samples (n=6, two operators).

Precision was evaluated in a multi-day validation experiment. To assess the overall variability of the method, variability components such as variability between injections, between preparations, between calibrations, and between days were determined using an ANOVA-based variance component analysis. Intermediate precision was found to be less than 1.4 % independent of the matrix. The repeatability CV ranged from 0.7 to 1.2 % over all concentration levels (Table 3). The evaluation of trueness within this experimental design was evaluated by comparing the measured data to the calculated sample concentrations. Therefore, data from the first day were evaluated (n=6, two operators). The relative mean bias ranged from 0.0 to 0.8 % for native serum levels and from 0.2 to 2.0 % for Li-heparin plasma levels. Highly concentrated native serum levels showed a bias between 0.9 and 1.1 % (Table 4).

Stability

Processed samples were found to be stable for six days at 7 °C with recoveries ranging from 98 to 102 %. Additionally, the stability of spiked frozen serum calibrator and QC material stored at −20 °C was evaluated after 28 days. Therefore, freshly prepared calibrators and QC levels were used to evaluate the frozen samples. The recoveries ranged from 98 to 101 % compared with the original value and are within the TE. Consequently, samples can be stored at −20 °C for 27 days.

Equivalence of results between independent laboratories

The method comparison study containing 117 samples (77 native patient serum samples, 10 native patient sample pools and 30 spiked serum samples) showed that two samples were below the lower limit of quantitation and five samples were above the calibration range. Dilution of these samples was not possible due to lack of additional sample volume. Moreover, one spiked sample was highlighted as outlier using the LORELIA (local reliability) outlier test [43]. Therefore, these samples were excluded from the evaluation.

Passing-Bablok regression analysis showed a good agreement between the two laboratories and resulted in a regression equation with a slope of 1.02 (95 % CI 1.02–1.03) and an intercept of 0.02 (95 % CI −0.06 to 0.08). Moreover, the Pearson correlation coefficient was ≥0.999 (Figure 3A). The Bland Altman analysis plot showed a data scatter which is independent of the analyte concentration with a mean bias of 2.6 % (95 % CI interval from 2.2 to 3.1 %) and a 2S agreement of 4.8 % (lower limit CI interval from −3.0 to −1.4 %, upper limit CI interval from 6.6 to 8.2 %) (Figure 3B).

Figure 3:

Results from the patient sample-based zonisamide method comparison study performed between two independent laboratories. (A) Passing–Bablok regression plot including the Pearson regression analysis for the method comparison study of the RMP (n=109) samples between the independent laboratories (Laboratory 1: Risch site, and Laboratory 2: Roche site). The regression analysis resulted in a regression equation with a slope of 1.02 (95 % CI 1.02–1.03) and an intercept of 0.02 (95 % CI −0.06 to 0.08). The Pearson correlation value was ≥0.999. (B) Bland–Altman plot for the method comparison study of the RMP (n=109 samples) between two independent laboratories (Laboratory 1: Risch site, and Laboratory 2: Roche site). The inter-laboratory measurement bias was 2.6 % (95 % CI interval from 2.2 to 3.1 %) and the 2S interval of the relative difference was 4.8 % (lower limit CI interval from −3.0 to −1.4 %, upper limit CI interval from 6.6 to 8.2 %).

The y-axis within the Bland Altman Plot shows the difference of laboratory 2-laboratory 1, scaled by their mean. Naturally, both measurements come along with a certain measurement uncertainty. Thus, the uncertainty of the calculated difference of both measurements is respectively larger. Based on the uncertainty from calibrator preparation of ≤0.86 % an allowable bias for the RMP B_ref≤1.72 % can be assumed. The acceptable CV of the difference between two laboratories can be estimated based on the exact mathematical derivation for the CV of the ratio (lab2 − lab1)/0.5*(lab1 + lab2). It can be approximated by CV(lab2 − lab1)=√(CV_(lab2)² + CV_(lab1)²). The performance evaluation in the laboratory 2 showed an intermediate precision of ≤2.6 % which is slightly higher to the intermediate precision of laboratory 1 of ≤1.2 %.

Thus, following the TE idea, a maximum allowed difference between the two laboratories can be calculated as 2 * B _ref  + 1.96 * √(CV _{(lab 2)} ²  + CV _{(lab 1)} ² ), leading to 9.1 %. This means that based on the uncertainty calculation of the calibrator preparation and the precision of the two laboratories, we can expect that most of all differences of the measurements of the individual samples fall between ±9.1 %, which is in good agreement with the observed data.

Uncertainty of results

The total measurement uncertainties for single measurements of serum samples ranged from 1.1 to 1.4 %, and the total measurement uncertainties for target value assignment (n=6, three measurements on two days) ranged from 0.8 to 1.0 %, independent of concentration level and sample type (Tables 5 and 6).

Consequently, the expanded measurement uncertainties are between 2.2 and 2.9 % for single measurements and between 1.6 and 2.1 % for multiple measurements. The derived total uncertainty is multiplied by a coverage factor of k=2 to obtain an expanded uncertainty.