An isotope dilution-liquid-chromatography-tandem mass spectrometry-based candidate reference measurement procedure for the quantification of total and free valproic acid in human serum and plasma

Chemicals and reagents

LC-MS grade methanol (CAS No. 67-56-1) was purchased from Biosolve (Valkenswaard, The Netherlands). Dimethyl sulfoxide (DMSO) (CAS No. 67-68-5) (ACS reagent, ≥ 99.99 %), isopropanol HPLC grade (CAS No. 67-63-0), ammonium acetate LC-MS grade (CAS No. 631-61-8), sodium acetate ACS reagent (CAS No. 127-09-3) and acetic acid (CAS No. 64-19-7) were purchased from Sigma Aldrich (Taufkirchen, Germany). The standard of valproic acid (CAS No. 99-66-1, Cat. No. PHR1061, Lot No. LRAC5650 and LRAD048) was bought from Sigma Aldrich (Taufkirchen, Germany), its deuterated ISTD, [²H₆]-valproic acid (CAS No. 87745-18-4, Cat. No. V094752, Lot No. 4-LDO-89-3) was bought from Toronto Research Chemicals (North York, Canada).

NMR solvent DMSO‑d₆ (CAS Nr.: 2206-27-1) and the qNMR ISTD 1,3,5-trimethoxybenzene (Lot No. BCBW3670) were obtained from Sigma Aldrich, Germany. 5 mm 10-inch NMR tubes were purchased from Sigma Aldrich and/or Euroisotop GmbH (Hadfield, United Kingdom).

Analyte-free human serum was obtained by pooling anonymized, leftover patient samples (n ≥ 5). Therapeutic drug monitoring (TDM) free serum used as surrogate matrix (ID. No. 12095432001) was obtained from Roche Diagnostics GmbH (Mannheim, Germany), native plasma matrix (lithium [Li-heparin], dipotassium ethylene diamine tetraacetic acid [K₂-EDTA] and tripotassium [K₃-EDTA]) was obtained from anonymized, leftover patient samples (n ≥ 5). Water was produced in-house with a Millipore Milli-Q 3 UV system from Merck (Darmstadt, Germany) and aqua ad iniectabilia B. Braun (Cat No. 530103) was purchased from B. Braun Medical AG (Sempach, Switzerland). All pools prepared from patient samples were in accordance with the Declaration of Helsinki. For filtration two devices were used: Amicon Ultra-15, PLTK Ultracel-PL Membrane, 30 kDa (Cat No. UFC903024) and Centrifree Ultrafiltration Device with Ultracel PL membrane (Cat. No. 4104), both obtained from Merck (Darmstadt, Germany).

General requirements for laboratory equipment

Certified and calibrated equipment was used. The minimum sample weight for the microbalance used (XPR2, Mettler Toledo, Columbus, OH, USA) was determined according to United States Pharmacopeial Convention (USP) guidelines (USP Chapters 41 and 1,251). Positive displacement pipettes were used to measure organic solvents and serum. Volumetric glassware (Class A volumetric flasks) that fulfilled the criteria of ISO 1042 and USP were used to prepare stock and spike solutions.

qNMR for determination of the purity of the standard materials

The qNMR measurements were performed on a JEOL 600 MHz NMR equipped with a He-Cryo probe (4–5 times the sensitivity compared to RT probes). For Valproic acid, single-pulse-¹HNMR was utilized for the quantitation (Supplementary Material 2, Figure 1; 1H; δ=2.22 ppm); 1,3,5-trimethoxybenzene as qNMR ISTD, 3H; DMSO‑d₆ as solvent with an inter-scan delay of 70 s. The identity of this methine quantitative resonance was established by 2D-HSQC spectrum.

Figure 1:

Total valproic acid 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 1 with a concentration of 2.40 µg/mL spiked in analyte-free human serum (left) and ISTD (right). (C) Patient sample with a concentration of 4.85 µg/mL (left) and ISTD (right) showing baseline separation between valproic acid and an unknown interference. ISTD, internal standard; LC-MS/MS, liquid chromatography-tandem mass spectrometry.

Additionally, we also performed quantum chemical calculations to ascertain the conformer distribution of valproic acid in solution because the Boltzmann average of these structures constitute the NMR spectrum. All the information is presented in the Supplementary Material 3.

Preparation of calibrators and quality control samples

In brief, two individual calibrator stock solutions were prepared which were further used to prepare spike solutions and final calibrator levels. Therefore, per stock solution 500 mg valproic acid was transferred into a 10 mL volumetric flask on an analytical balance (AT261 Delta Range, Mettler Toledo). Then, the flasks were filled up with DMSO to reach concentrations of 50.0 mg/mL. For free valproic acid, 200 mg of valproic acid was weighed and dissolved in 10 mL DMSO, resulting in stock solutions with concentrations of 20.0 mg/mL. The exact concentrations of each stock solution were calculated based on the purity of the analyte (99.7 ± 0.4 % (k=2) and 98.9 ± 0.6 % (k=2), determined by qNMR) and the amount weighed.

These stock solutions were used to prepare working and spike solutions, which were further used to prepare the final calibrator levels. These were prepared by a 1+99 dilution (v+v) into human serum matrix for valproic acid and into mobile phase A for free valproic acid, resulting in concentrations ranging from 2.40 to 145 μg/mL for valproic acid and from 1.60 to 42.0 μg/mL for free valproic acid.

Additionally, four quality control (QC) levels were prepared for each assay. QC levels and calibrators were prepared identically using a third independent stock solution. For free valproic acid however, the spike solutions were diluted into analyte-free human serum ultrafiltrate (obtained by Amicon filter device centrifuged at 1,000 rcf and 25 °C for 180 min). The concentrations of the QC samples were 4.00, 25.0, 75.0 and 120 μg/mL for total valproic acid and 2.00, 6.00, 15.0 and 24.0 μg/mL for free valproic acid.

Internal standard solution

For the preparation of the ISTD stock solutions, [²H₆]-valproic acid was dissolved in the appropriated amount of DMSO directly in the manufacturer’s container to achieve a 1,000 μg/mL stock solution. The working solution was prepared by a dilution of the stock solution with ultrapure water to obtain concentrations of 12.5 and 7.50 μg/mL for total and free valproic acid, respectively.

Sample preparation: total valproic acid assay

A total of 100 µL of the ISTD working solution was pipetted into a 2 mL tube (Eppendorf Safe-Lock Tubes; Eppendorf, Hamburg, Germany) and 50 µL of the sample specimen (native sample/calibrator/QC) was added. 1,000 µL precipitation solution (75 % methanol in ultrapure water [v+v]) was added to precipitate proteins and after centrifugation 250 µL of the supernatant was diluted 1+4 (v+v) using mobile phase A.

Sample preparation: free valproic acid assay

A total of 500 µL of native sample was pipetted into a 1 mL Centrifree filter device. The ultrafiltration device was centrifuged at 1,000 rcf and 25 °C for a period of 30 min. In a separate 2 mL tube, a 100 μL of ISTD working solution was mixed with 50 μL of the filtrated sample (calibrator/QC/sample ultrafiltrate) and mixed with 1,000 μL of mobile phase A. Finally, the solution was further diluted with mobile phase A at a ratio of 1+3 (v+v) and stored at 7 °C until analysis.

Liquid chromatography mass spectrometry

Measurements for total and free valproic acid were performed on an Agilent 1,290 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.

Valproic acid was separated using an Agilent Zorbax Eclipse XDB-C18 column (50 × 2.1 mm, 3.5 µm, Santa Clara, CA, USA), which was kept in the column compartment at 50 °C. For total valproic acid analysis an additional Phenomenex Security Guard Cartridge Polar (Art. No. AJ0-7600-S, C18, 4 × 2 mm Lane Cove, Australia) with a Phenomenex Security Guard Cartridge Kit (Art. No. KJ0-4282) was used. The mobile phase A consisted of 2 mM ammonium acetate in ultrapure water/methanol 95+5 (v+v) with 0.05 % acetic acid and 0.025 mM sodium acetate. Mobile phase B contained 2 mM ammonium acetate in methanol/ultrapure water 95+5 (v+v) with 0.05 % acetic acid and 0.025 mM sodium acetate. The LC was performed with a flow rate of 0.4 mL/min and a gradient profile over 13 min. Contamination of the MS was reduced using a divert valve switch, switching the eluent flow either to the MS or to the waste. The injection volume was set to 8 μL.

The detection was performed on an AB Sciex Triple Quad 6,500+ and Q-Trap 6,500+ mass spectrometer (Framingham, Massachusetts, USA) with a Turbo V ion source using a negative electro spray ionization. The quantifier transition (valproate sodium acetate adduct [m/z 225.0–143.0]) sets the basis for the quantitative method and is associated with a correspondent transition of the ISTD [²H₆]-valproic acid (valproate sodium acetate adduct [m/z 231.0–149.0]). Supplementary Materials 1 and 2 include detailed LC and MS methods.

System suitability test

A SST was performed prior to each analysis to check the sensitivity of the system, the chromatographic performance and possible carry-over effects. Therefore, a valproic acid stock solution (10 mg/mL valproic acid in DMSO) was used to prepare two concentration levels, SST1 and SST2 with concentrations corresponding to the processed calibrator levels 1 and 8, respectively. The analytical data from experiments were only collected if the SST has been passed. To pass the SST, the signal-to-noise ratio of the quantifier transition for SST1 had to be ≥ 30 and the retention time for both samples had to be within 4.5 ± 0.5 min [72]. Furthermore, two solvent blanks were injected after the SST2 to check possible carry-over effects. The analyte peak area observed in the first blank had to be ≤ 20 % of the analyte peak area of SST1 to pass the SST. This purity criterion applied also to all further blanks in the measurement campaign.

Calibration and structure of analytical series and data processing

At the beginning and end of the analytical series, calibrators (cal 1 – cal 8) with increasing concentrations were measured. Both calibration sets were used to generate the final calibration curve through quadratic regression of the area ratios of the analyte to the ISTD (y) against the analyte concentration (c), 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 4.5 ± 0.5 min using a smoothing factor of three and peak splitting factor of two, a noise percentage of 80 %, and a base sub-window of 0.7 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 [72], the International Council of Harmonization’s guidance document, Harmonised Tripartite Guideline Validation of Analytical Procedures: Text and Methodology Q2 (R1) [73] and the Guide to the expression of uncertainty in measurement (GUM) [74].

Traceability to SI units

Traceability to the SI unit of mass (kilogram) and the SI unit of the amount of substance (mole) has been established using 1,3,5-trimethoxybenzene as the qNMR ISTD, which is directly traceable to NIST PS1 (the primary qNMR standard). Since the kilogram is now defined by Planck’s constant and the mole by Avogadro’s constant, the qNMR methodology is ideally suited to satisfy both traceability chains. For valproic acid six individual experiments involving six individual weighings (Supplementary Material 3; Table 1 and Figure 2) were performed, yielding a final absolute content value of 99.7 ± 0.4 % (k=2).

Figure 2:

Free valproic acid LC-MS/MS derived analytical readouts. (A) Chromatogram of a matrix blank showing the analyte SRM ion trace (left) and ISTD (right). (B) Chromatogram of calibrator level 1 with a concentration of 1.60 µg/mL spiked in mobile phase A (left) and ISTD (right). (C) Pooled patient sample (n ≥ 5) with a concentration of 8.86 µg/mL (left) and ISTD (right). ISTD, internal standard; LC-MS/MS, liquid chromatography-tandem mass spectrometry.

Selectivity/specificity

The developed gradient combined with a reversed-phase column (Agilent Zorbax Eclipse XDB-C18, 50 × 2.1 mm, 3.5 μm) showed a baseline separation of valproic acid and an unknown impurity. The separation of the unknown interference was managed by applying isocratic step elution; details of the elution gradient are found in Supplementary Materials 1 and 2. To get an even baseline during this step, the column needs to be equilibrated to the start conditions for 5.5 min with mobile phase A.

Using the valproic acid pseudo-selected reacting monitoring (SRM) (m/z 143.0–143.0) as described in literature, loss of sensitivity in matrix and shifting retention time was observed. Switching to acetate buffered mobile phases [45] led to stabilized analyte retention times. Selecting valproic acid/sodium acetate or valproic acid/acetic acid clusters in the first quadrupole increased ionization yields.

Under the chosen chromatographic conditions, no signals were detected at the expected retention times for valproic acid for both assays when analyzing matrix blank samples. An uncharacterized impurity was detected at the retention time of 5.5 min and showed a resolution of ≥ 3.3 for both assays. No residual unlabeled analyte was observed when ISTD solutions were analyzed (Figures 1 and 2). While measuring native patient samples within the method comparison study, an additionally unknown signal was detected in a very few samples at a retention time of 3.6 min when total valproic acid was measured. The peaks were baseline separated and have therefore no impact on the quantification of valproic acid in patient derived materials (Figure 1C).

Matrix effect

Matrix dependent effects caused by salts, proteins and phospholipids were avoided by ultrafiltration and sample preparation with a dilution after precipitation and ultrafiltration. The post column infusion experiment showed that there was no visual matrix effect (ME) in the ionization field at the expected retention time of valproic acid.

The relative ME, comparing absolute areas of the analyte and ISTD for total valproic acid, ranged from 93 to 102 % for the analyte peak area and from 93 to 101 % for the ISTD peak area across all tested matrices, with the exception of the lowest level in analyte-free serum (Table 1). At this level, the relative ME was found to be 105 % for the analyte peak area and 93 % for the ISTD peak area, indicating a potential ME. This variance is attributed to the fact that, out of the five preparations, two exhibited a higher analyte peak area while the ISTD area remained stable. Since all concentration levels were derived from the same matrix pool, a ME can be excluded. All other measurements remained unaffected, and the ISTD effectively compensates for any potential fluctuations in the MS signals in the remaining measurements.

The relative ME for free valproic acid ranged from 99 to 102 % for the analyte peak area and from 100 to 102 % for the ISTD peak area (see Table 2).

Quantitative analysis of total valproic acid showed mean slopes (n=2, sample preparations) of 0.059 (95 % CI from 0.058 to 0.060) for neat solution, 0.059 (95 % CI from 0.058 to 0.059) for analyte-free human serum, 0.061 (95 % CI from 0.060 to 0.061) for TDM-free serum and 0.059 (95 % CI from 0.057 to 0.060) for Li-heparin plasma. Quantitative analysis of free valproic acid showed mean slopes (n=2, sample preparations) of 0.078 (95 % CI from 0.077 to 0.079) for neat solution, 0.078 (95 % CI from 0.078 to 0.078) for analyte-free human serum ultrafiltrate, 0.078 (95 % CI from 0.077 to 0.080) for TDM-free serum ultrafiltrate and 0.081 (95 % CI from 0.078 to 0.083) for Li-heparin plasma ultrafiltrate. The absence of a ME can be confirmed based on the overlapping CIs of the different experiments. Independent of the matrix used for calibration, coefficients of determination were better than 0.999.

Linearity

Coefficients of determination for all individual calibrators were ≥ 0.999 and therefore linearity was proven for both assays. The serially diluted samples 1 to 11 showed a linear relationship with a correlation coefficient of 1.000 for both total and free valproic acid with recoveries ranging from 101 to 105 % and 95 to 101 % for total and free valproic acid, respectively.

Lower limit of the measuring interval

The relative bias showed a deviation of −0.9 % and the CV was 2.0 % for total valproic acid at a concentration of 2.40 μg/mL. For free valproic acid, the relative bias was found to be −1.6 % with a CV of 1.6 % at a concentration of 1.60 μg/mL.

Accuracy, trueness, and precision

The total method variability was estimated using an ANOVA based variance component analysis. Repeatability CV, including variability components such as between-injection variability and between-preparation variability. Intermediate precision was found to be less than 3.6 % for total valproic acid (n=60) and the repeatability ranged from 1.3 to 3.2 % (Table 3).

The precision experiment for free valproic acid consisted of 48 measurements, with data from one day excluded due to technical issues and errors in sample preparation. This resulted in an intermediate precision of ≤ 4.0 % (n=48) for free valproic acid and a repeatability ranging from 1.3 to 3.5 % (Table 4).

Total valproic acid showed an accuracy for spiked analyte-free human serum samples between −0.4 and 4.1 % and for spiked Li-heparin plasma samples between −0.3 and 3.5 %. Further, the mean of the high concentrated samples ranged from 1.9 to 2.3 % (Table 5).

To determine the free fraction of valproic acid in serum, total and free fraction was determined. Hence, the following concentrations of the serum samples were used for the accuracy: 3.67 ± 0.08, 10.1 ± 0.1, and 17.2 ± 0.2 μg/mL. A mean bias from −2.9 to −0.9 % for spiked analyte-free human serum (see Table 6) was observed.

For spiked native serum and native Li-heparin plasma ultrafiltrates, a mean bias from 0.5 to 3.4 % and 0.5–4.0 % were determined (Table 7). Additionally, the high concentrated spiked analyte-free human serum samples showed a linear dependence (r²=1.000) after serial dilution with CVs of ≤ 1.3 %.

Stability

Calibrator and QC levels of total valproic acid were found to be stable at 7 °C for 6 days for valproic acid with a mean recovery of 103 %. For free valproic acid, the mean recovery was 100 % after 15 days. Stability of spiked frozen native serum calibrators and QC levels stored at −20 °C were found to be stable for 20 days for total and free valproic acid with mean recoveries of 100 and 98 %, respectively.

Equivalence of results between independent laboratories

The method comparison study of valproic acid containing 150 native anonymized patient samples showed that four samples were below the LLMI, two samples were found to be above the calibration range and six samples were excluded since they were mixed up. Further, two samples had not enough sample volume for the preparation. These samples were therefore excluded from evaluation.

Passing-Bablok regression analysis showed agreement for total valproic acid between the two laboratories. Total valproic acid resulted in a regression equation with a slope of 1.00 (95 % CI 0.99–1.02) and an intercept of 0.10 (95 % CI −0.87–0.65). The Pearson correlation coefficient was 0.994 (Figure 3A). Bland-Altman analysis showed agreement between the two laboratories for total valproic acid. The data scatter is independent of the analyte concentration with a mean bias of 1.1 % (95 % CI 0.34–1.88 %) and a 2S agreement of 8.9 % (lower limit of CI interval from −9.1 to −6.5 % and upper limit of the CI interval from 8.7 to 11.3 %) for total valproic acid (Figure 3B).

Figure 3:

Results from the patient sample-based total valproic acid method comparison study. (A) Passing-Bablok regression plot including the Pearson regression analysis for the method comparison study of the RMP (n=136) between the independent laboratories (site 1: Risch; site 2: Roche). Passing-Bablok resulted in a regression equation with a slope of 1.00 (95 % CI 0.99 to 1.02) and an intercept of 0.10 (95 % CI −0.90 to 0.65). The Pearson correlation value was ≥ 0.99. (B) Bland-Altman plot for the method comparison study of the RMP (n=136 patients) between two independent laboratories (laboratory 1: Risch site, and laboratory 2: Roche site). The interlaboratory measurement bias was 1.1 % (95 % CI interval from 0.3 to 1.9 %) and the 2SD interval of the relative difference was 8.9 % (lower limit CI interval from −9.1 to −6.5 %, upper limit CI interval from 8.7 to 11.3 %).

Additionally, a precision experiment was performed within the second laboratory. The resulting CVs (n=36) for intermediate precision and repeatability were ≤ 4.3 % and ≤ 3.4 %, respectively, for total valproic acid.

For free valproic acid, 70 native patient samples were used for evaluation. Passing-Bablok regression analysis showed agreement for free valproic acid between the two laboratories. Free valproic acid resulted in a regression equation with a slope of 0.99 (95 % CI 0.98–1.00) and an intercept of 0.07 (95 % CI 0.00–0.18). The Pearson correlation coefficient was 0.999 (Figure 4A). Bland-Altman analysis showed agreement between the two laboratories for free valproic acid. The data scatter is independent of the analyte concentration with a mean bias of −0.5 % (95 % CI −1.1–0.0 %) and a 2S agreement of 4.6 % (lower limit of CI interval from −6.0 to −4.1 % and upper limit of the CI interval from 3.1 to 5.0 %) for free valproic acid (Figure 4B).

Figure 4:

Results from the patient sample-based free valproic acid method comparison study. (A) Passing-Bablok regression plot including the Pearson regression analysis for the method comparison study of the RMP (n=70) between the independent laboratories (site 1: Risch; site 2: Roche). Passing-Bablok resulted in a regression equation with a slope of 0.99 (95 % CI 0.98 to 1.00) and an intercept of 0.07 (95 % CI 0.00 to 0.18). The Pearson correlation value was ≥ 0.999. (B) Bland-Altman plot for the method comparison study of the RMP (n=70 patients) between two independent laboratories (laboratory 1: Risch site, and laboratory 2: Roche site). The interlaboratory measurement bias was −0.5 % (95 % CI interval from −1.1 to 0.0 %) and the 2SD interval of the relative difference was 4.6 % (lower limit CI interval from −6.0 to −4.1 %, upper limit CI interval from 3.1 to 5.0 %).

Additionally, a precision experiment was performed within the second laboratory. The resulting CVs (n=36) for intermediate precision and repeatability were ≤ 1.5 % and ≤ 1.6 %, respectively, for free valproic acid. The results are comparable to Site 1, and it could be shown that the method is transferable and meets the requirements independent of the laboratory equipment and personnel.

Uncertainty of results

Measurement uncertainties for single measurements of serum samples for total valproic acid ranged from 1.9 to 3.4 % and for target value assignments (n=6, three measurements on two days) from 0.9 to 1.3 % (Tables 8 and 9). Consequently, the expanded measurement uncertainties (k=2) are between 3.9 and 6.9 % and 1.9 and 2.6 % for single measurements and multiple measurements, respectively.

For free valproic acid, the measurement uncertainty for single measurements of serum samples ranged from 2.0 to 4.1 % and for target value assignments (n=6, three measurements on two days) from 1.1 to 1.4 % (Tables 10 and 11). Consequently, the expanded measurement uncertainties (k=2) are between 4.0 and 8.1 % and 2.2 and 2.9 % for single measurements and multiple measurements, respectively.