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

Objectives: Phenobarbital serves as an antiepileptic drug (AED) and ﬁ nds application in the treatment of epilepsy either as monotherapy or adjunctive therapy. This drug exhibits various pharmacodynamic properties that account for its bene ﬁ cial e ﬀ ects as well as potential side e ﬀ ects. Accurate measurement of its concentration is critical for optimizing AED therapy through appropriate dose adjustments. Therefore, our objective was to develop and validate a new reference measurement procedure (RMP) for the accurate quanti ﬁ cation of phenobarbital levels in human serum and plasma. Methods: A sample preparation protocol based on protein precipitation followed by a high dilution step was established in combination with a liquid chromatography-tandem mass spectrometry (LC-MS/MS) method using a C8 column to separate target analytes from known and unknown interferences. Assay validation and determination of measurement uncertainty were performed based on current guidelines. Selectivity and Speci ﬁ city were assessed using spiked serum and plasma samples; to investigate possible matrix e ﬀ ects (MEs) a post-column infusion experiment and a comparison of standard line slopes was performed. Precision and accuracy were determined within a multiday precision experiment. Results: The RMP was shown to be highly selective and speci ﬁ c, with no evidence of matrix interferences. It can be used to quantify phenobarbital in the range of 1.92 to 72.0 μ g/mL. Intermediate precision was less than 3.2 %, and repeatability coe ﬃ cient of variation (CV) ranged from 1.3 to 2.0 % across all concentration levels. The relative mean bias ranged from − 3.0 to − 0.7 % for native serum levels, and from − 2.8 to 0.8 % for Li-heparin plasma levels. The measurement uncertainties (k=1) for single measurements and target value assignment were 1.9 to 3.3 % and 0.9 to 1.6 %, respectively. Conclusions: A novel LC-MS/MS-based candidate RMP for the quanti ﬁ cation of phenobarbital in human serum and plasma is presented which can be used for the standardization of routine assays and the evaluation of clinically relevant samples.


Introduction
Barbiturates are a class of depressant drugs that derived from barbituric acid.Despite the synthesis of approximately 2,500 derivatives, a subset of 50 has been marketed as anticonvulsants, anesthetics, and hypnotics [1].Notably, most of these agents have been replaced by safer, nonaddictive and non-controlled drugs.Nevertheless, phenobarbital (C 12 H 12 N 2 O 3 , molecular weight=232.24Da) is still widely used.Especially in developing countries, it is the drug of choice to treat epilepsy since it is inexpensive, effective, and usually taken with good compliance [2][3][4].Phenobarbital prevents seizures by facilitating γ-aminobutyric acid (GABA)-mediated inhibition through allosteric interaction with the GABA A receptors [5,6].Mean channel open time is prolonged without affecting open frequency or conductance [7].Sedative properties of phenobarbital are due to its ability to directly activate GABA A receptors [8].Phenobarbital is metabolized slowly, resulting in a half-life of 70 to 140 h and is inhibited by other drugs (e.g., phenytoin and valproic acid), resulting in higher plasma concentrations [9].Approximately 25 % of phenobarbital is excreted unchanged in the urine [10].The remainder is metabolized in the liver to p-hydroxyphenobarbital and 9-D-glucopyranosylphenobarbital, both of which are pharmacologically inactive [11,12].The therapeutic range is from 10.0 to 40.0 μg/mL and the toxic level is reached at 50.0 μg/mL [13] and can lead to respiratory and cardiovascular depression, and death [14,15].
Immunoassays are widely used for (therapeutic drug monitoring) TDM of phenobarbital [16], although they may be non-specific and may react with metabolites, thus leading to confounding results in plasma [17].In contrast, spectrometric techniques (e.g., LC-MS/MS) are highly specific and sensitive, and the methods are based on state of the art analytical principles.However, LC-MS/MS methods lack standardization with many individual laboratories developing 'in-house' methods [18][19][20][21][22][23].Therefore, RMPs play a crucial role in the standardization and harmonization of patient results.Currently, there is an RMP listed by the Joint Committee for Traceability in Laboratory Medicine (JCTLM) which was also used for target value assignment of the reference material "Antiepilepsy Drugs in Frozen Human Serum, SRM 900a" published by NIST [24][25][26].Unfortunately, this material is no longer commercially available.Since the status of the NIST RMP was unclear and we were not aware of reference laboratories that have instituted the NIST RMP, it was decided to establish and validate a new candidate RMP according to the requirements of the ISO 15193 guideline [27].In this regard, a commercially available SI-traceable reference material was utilized, the target value assignment of which was established by qNMR methodology as per the accompanying certificate of analysis.
Based on the desired phenobarbital routine measurement uncertainty (k=1), the desired target measurement uncertainty (k=1) for the RMP was estimated as previously described in our published RMPs [28][29][30][31].It ranged from 0.6 to 7.5 % depending on which estimation modeleither based on the therapeutic range (Glick [32], Burnett [33]) or on the half lifetime of the drug (Fraser [34]) was used as described by Steele et al. [35].Whereas the Glick and Burnett figures of merit are feasible, the measurement uncertainty demanded by the Fraser model seems unreasonable low.It is an effect of dosing phenobarbital in rather short intervals, although it has a very slow clearance from the circulation.This therapeutic approach leads to very shallow exposure troughs over the time leading to a very stable steady state drug exposure.Investigations into proficiency data by Steele and coworkers showed that the between laboratory variance of phenobarbital measurements is 8.4 % thus suggesting an actual (real life) measurement uncertainty of 6.7 % (k=1) when the average inter-laboratory/intra-laboratory variance component factor of 0.8 was also applied [35].Pauwels critically questioned the model approach of Fraser by assessing the intra-individual variability of drug levels in routine.They corrected their data for the analytical variability (CV a =3.7 %) and could show from 20 measurement pairs, that in clinical TDM use, the intra-individual (CV i ) phenobarbital drug level variability is 19.8 %.From that data they derived desirable routine performance criteria of 9.9 % imprecision (CV a ) and 8.3 % bias summing up to a total error of analysis (TEA) allowance of 24.6 % [36].If the total error (TE) calculation model based on biological variation is set to the "optimal" analytical performance specification goal [37], a CV a of 5.0 % and a bias of 4.1 % must not be exceeded to meet the associated TEA goal of 12.3 %.The found CV a matches well withand is consequently confirmed bythe CV a data previously derived from the proficiency data based top-down analysis performed by Steele.Since the goal of phenobarbital TDM is the avoidance of overexposure, e.g., due to drug-drug interactions it seems feasible to target the measurement uncertainty based on the therapeutic range considerations or on measurement requirements based on biological variability data derived from patients in medical care.The analytical goals derived from the Fraser model based on pure pharmacokinetic considerations must be critically discussed as not in match with the clinical needs.
Taken together, it can be assumed that a phenobarbital assay, to serve state of the art phenobarbital TDM as RMP, must perform at a CV a level of approximate 1.7 % or better.This is to meet the performance target criterion by Braga et al. stating that analytical uncertainties of less than one third of routine services should be achieved by an RMP [38].
We present here a novel LC-MS/MS-based candidate RMP for the quantification of phenobarbital in human serum and plasma.This manuscript does only cover phenobarbital, a RMP for primidone will be covered in a separate report.To facilitate the reproduction of the candidate RMP by other laboratories, details are described in two supplementary documents focusing on the technical implementation of the test procedure and the calculation of measurement uncertainty.

Materials and methods
A full description of the methods, materials and equipment used is provided in Supplementary Material 1.

Preparation of calibrators and quality control samples
Two individual calibrator stock solutions were prepared, which were used for the preparation of working and spike solutions.These solutions were than used for the preparation of the final matrix-based calibrator levels.
Per stock solution, 75.0 mg of phenobarbital were weighed in tin boats on a micro balance (XPR2, Mettler Toledo, Columbus, Ohio, USA) and dissolved in 5 mL DMSO using a volumetric flask to reach concentrations of 15.0 mg/mL.The exact concentration of the stock solutions was calculated based on the purity of the reference material (99.7 ± 0.4 %, based on the certificate) and the amount of weighed phenobarbital.These stock solutions were further diluted with DMSO to obtain working solutions with concentrations of 1.50 mg/mL.Both, stock and working solutions were then used to prepare eight calibrator spike solutions in DMSO.Final matrix-based calibrators, uniformly distributed from 1.92 to 72.0 μg/mL (8.27 to 310 μmol/L) (see Figure 1), were prepared by a 1+99 dilution (v/v) into human serum matrix.
Matrix-based quality control (QC) samples were prepared similarly to the calibrators using a third independent stock solution with a concentration of 20.0 mg/mL.The concentrations for the control levels were set at four critical control points: above the limit of quantification (2.60 μg/mL), below (8.00 μg/mL) and within (25.0 μg/mL) the therapeutic reference range, and at the laboratory alert level (50.0 μg/mL) (Figure 1).

ISTD solution
To prepare the ISTD stock solution, 900 µL [ 2 H 5 ]-phenobarbital dissolved in methanol (1.00 mg/mL) was mixed with 8,100 µL of DMSO (v/v) to obtain a 100 μg/mL ISTD stock solution, which was stored at −20 °C for up to 8 months.The ISTD working solution was freshly prepared on each day of sample preparation by diluting the stock solution.Therefore, 100 µL of ISTD stock solution was added to 3,900 µL Milli-Q water to reach a final concentration of 2.50 μg/mL.

Sample preparation
The following sample matrices can be used: native human serum, surrogate serum, and plasma (Li-heparin plasma, K 2 -EDTA, and K 3 -EDTA).Fifty microliter of sample specimen (native sample/calibrator/QC) was mixed with 100 µL of ISTD working solution in a 2 mL tube (Eppendorf Safe-Lock Tube, Eppendorf, Hamburg, Germany).Then, 1,000 µL precipitation solution (75% methanol in Milli-Q water [v/v]) was added and the sample centrifuged.Subsequently, 10 µL of the supernatant was diluted with mobile phase A in two steps (step 1: 1+99

Liquid chromatography mass spectrometry
Chromatographic separation was accomplished with an Agilent 1290 Infinity II LC system (Santa Clara, CA, USA) equipped with a binary pump, a vacuum degasser, an autosampler and a column compartment.Chromatographic separation of phenobarbital and the prodrug primidone was achieved on an Agilent Zorbax Eclipse XDB-C8 column (100 × 3 mm, 3.5 µm, Santa Clara, CA, USA) which was kept in the column compartment at 40 °C.The mobile phases consisted of water/ methanol 90+10 (v/v) with 0.1 % acetic acid (A) and methanol/water 95+5 (v/v) (B).The LC was performed with a flowrate of 0.6 mL/min using a gradient over 10 min.The injection volume was set at 5 μL.Switching the eluent flow to the waste until 0.5 min and from 6.5 min, by a divert valve, reduced the contamination of the MS system.Phenobarbital was detected in multiple reaction monitoring (MRM) mode using an AB Sciex Triple Quad 6500+ and Q-Trap 6500+ mass spectrometer (Framingham, Massachusetts, USA) with a Turbo V ion source in negative electrospray ionization mode (ESImode).The ion source was optimized with a spray voltage of −3,500 V and a temperature of 600 °C.Nitrogen gas was used as curtain gas, collision gas, ion gas source 1, and ion gas source 2 and were set at 35, 10, 60, and 50 psi, respectively.The collision cell entrance potential, collision exit potential, declustering potential and dwell time was set at −10 V, −15 V, −30 V and 50 ms, respectively, for all mass transitions.
The quantifier transition (m/z 231.1 to 42.1) serves as the basis of the quantification method.An additional specific mass transition (qualifier, m/z 231.1 to 188.1) was monitored to exclude interfering substances in native matrix samples.The quantifier/qualifier area ratios of native matrix samples and neat system suitability test (SST) samples were compared and should not differ by more than 20 %.Table 1 provides an overview of the SRM transitions and the remaining compounddependent MS settings.

Testing for system suitability
Prior to each analysis, the SST is performed to determine the system's sensitivity, chromatographic performance, and carry-over effects.Concentration levels of the SST sample 1 and SST sample 2 corresponded to the analyte concentration within the processed calibrator level 1 and 8, respectively.For passing the SST, the signal-to-noise ratio of the quantifier transition had to be ≥50 for SST sample 1 and both SST samples had to show a retention time within 4.6 ± 0.5 min.For the examination of potential carry-over effects, the high concentrated SST sample 2 was injected, followed by two blank injections.The analyte peak area observed in the first blank after the injection of the SST sample 2 had to be ≤20 % of the analyte peak area of calibrator 1.The purity criterion for the first sample SST was applied to all further blanks in the measurement campaigns.

Calibration, structure of analytical series and data processing
Calibration is performed using calibrators prepared as described in the section 'Preparation of calibrators and quality control samplesʼ.To generate the final calibration function, both calibrations were measured in increasing order at the beginning and at the end of an analytical series.The final calibration function was determined by a linear regression of the area ratios of phenobarbital and its ISTD (y) vs. the corresponding phenobarbital concentration (x), resulting in the function, y=ax × b.
Data evaluation was performed using the Analyst software, Version 1.6.3,employing the Intelli Quant algorithm (ABSciex).Phenobarbital and its ISTD showed a retention time of 4.6 min and were integrated within a 30 s window.A smoothing and peak splitting factor of 3 were used for peak integration.The noise percent was set to 90 % with a base sub window of 0.8 min.

Method validation
Assay validation and determination of measurement uncertainty were performed according to the Clinical & Laboratory Standard Institute (CLSI) Guidelines C62A Liquid Chromatography-Mass Spectrometry Methods [39], the International Conference on Harmonization guidance document Harmonised Tripartite Guideline Validation of Analytical Procedures: Text and Methodology Q2 (R1) [40] and the Guide to the expression of uncertainty in measurement (GUM) [41].Further details are published in Taibon et al. [42].
Selectivity: To evaluate the selectivity of the method, native human serum, surrogate serum, and Li-heparin plasma were spiked with phenobarbital, primidone, and the ISTD [ 2 H 5 ]-phenobarbital at a concentration of 9.00 μg/mL.Analyte-free matrices were additionally checked for possible interfering matrix signals for the analyte quantifier and qualifier transitions.The residual amount of unlabeled analyte in the ISTD was determined to demonstrate the suitability of the ISTD used.Therefore, analyte-free matrices were spiked with ISTD, and the amount of residual unlabeled analyte must not exceed 20 % of the lower limit of the measuring interval (LLMI), which corresponded to the concentration of the lowest calibrator level.
Matrix effect and specificity: To investigate possible matrix effects (MEs), a qualitative post column infusion experiment was performed.For this, a solution of 25 ng/mL phenobarbital in mobile phase A/mobile phase B 1+1 (v/v) was infused into the HPLC column effluent through a T-piece at a flow rate of 7 μL/min.Processed matrix samples (native human serum, surrogate serum, and native plasma [Liheparin, K 2 -EDTA, and K 3 -EDTA]) were then injected into a stable analyte background signal and the change in the background signal was recorded.Any shift in the MRM analyte signal at the expected retention time would indicate a matrix component mediated effect on the ionization.
To further demonstrate the matrix independence of the method, the slopes of the standard lines (n=6 sample preparations) were compared in neat solution (mobile phase A), native human serum, drugfree human serum, and Li-heparin plasma [43].The confidence intervals (CI) of the slopes must overlap and the coefficient of determination must In addition, absolute areas of analyte and ISTD as well as area ratios were compared [44].Therefore, analyte and ISTD solutions were spiked into native human serum, surrogate serum, Li-heparin plasma, and neat solution (75% methanol) after protein precipitation for three levels (8.00, 25.0, and 50.0 μg/mL) spread over the working range.To evaluate the ME, the mean peak areas of the analyte and ISTD were compared with those of the neat samples, not exceeding a percent deviation of ±10 %.All samples were prepared in five replicates (n=5 sample preparations).
Linearity: To determine the linearity, the calibrators were prepared twice a day for 3 days (n=6 sample preparations).Additionally, to extend the calibration range by ±20 % two serum samples were prepared at the lower and upper end of the measurement range (1.40 to 90.0 μg/mL, n=6 sample preparations).Residuals and coefficient of determination were determined and had to be randomly distributed and ≥0.999, respectively.
Furthermore, the linearity of the method was demonstrated based on the recovery of serially diluted samples using the preferred regression model for calculation.Using calibrator level 1 as sample 1 and calibrator level 8 as sample 11 nine additional samples were diluted as follows: Measurement results must show a linear relationship with a correlation coefficient ≥0.999.Recovery will be reported as the percentage of recovery of the measured concentration relative to the nominal concentration of the sample pools.
Lower limit of the measuring interval and limit of detection: Accuracy, trueness, and precision of the LLMI were determined by measuring spiked native human serum samples whose concentration corresponded to the lowest calibrator level (1.92 μg/mL).Sample preparation was performed five times, and recovery, bias, and precision were determined.
To estimate the limit of detection (LOD), the approach of Armbruster et al. [45] was used.Therefore, the limit of blank was calculated using 10 independent matrix blank samples as follows: LOB=mean blank + 1.645(SD blank ).The LOD was then estimated using 10 replicates of calibrator level 1, which was used as the low concentration sample: LOD=LOB + 1.645(SD low concentration sample ).
Precision and accuracy: To evaluate the precision and accuracy of the developed method, a 5 days validation experiment was performed as previously described [28-31, 42, 46].An ANOVA-based variance component analysis was used to evaluate the total method variability, which included variability components such as between-injection variability, between-preparation variability, between-calibration variability, and between-day variability.
In brief, four spiked native serum and native Li-heparin plasma samples (2.60, 8.00, 25.0 and 50.0 μg/mL) covering the measurement range and two native patient serum samples were prepared in triplicate and injected twice.This scheme was performed by two individuals in parallel, defined as validation part A and B, over 5 days (n=12 measurements per day, n=60 measurements per 5 days).For each part and day, independent calibration curves were generated and used for quantitative analysis.To further vary conditions, two batches of columns and two different sample weightings were used for these measurements.Biowarp, an internal statistic program based on the VCA Roche Open Source software package in R [47], was used to evaluate the data.
Accuracy was assessed using four spiked native human serum and Li-heparin plasma samples (2.60, 8.00, 25.0 and 50.0 μg/mL) as well as two spiked high concentrated serum samples (80.0 and 100 μg/mL; dilution 1 and 2).Three sample preparations were prepared for each part A and B on one day (n=6 sample preparations) and the percentage of recovery of the measured concentration relative to the spiked concentration determined.

Sample stability:
The autosampler stability of processed samples (calibrators and QC levels) stored at 7 °C was evaluated by comparison to freshly prepared calibrators and QC levels after 7 days.The stability of matrix-based calibrators and controls stored at −20 °C was evaluated after 28 days.Therefore, frozen samples were compared to freshly prepared calibrators and QC levels.The total error (TE) was used as an acceptance criterion and estimated based on the results of the precision and accuracy experiment, resulting in a maximum TE of ±8 %.Stability can be ensured for a measurement interval of 2 to 28 days (x) for x−1 day, and for a measurement interval of >4 weeks (y) for y−1 week.

Equivalence of results between two independent laboratories:
To assess the agreement of the candidate RMP between two independent laboratories (laboratory 1 Dr. Risch Ostschweiz AG, Buchs SG and laboratory 2 Roche Diagnostics GmbH, Penzberg) a method comparison study was performed using 127 samples (70 anonymized native serum samples, 27 native plasma samples, and 30 spiked samples) provided by laboratory 2. In addition, a 3 days precision experiment was carried out in laboratory 2 using the same spiked and native serum samples as in laboratory 1.The LC-MS system and laboratory equipment used were similar in both laboratories with the following modifications for site 2: an ultra-microbalance XP6U/M (Mettler Toledo) and aluminum weighing boats were used for the preparation of stock solutions.Calibrators were prepared independently in both laboratories as described in Supplementary Material 1.
Uncertainty of measurements: Measurement uncertainty was determined according to the GUM [41] and Taibon et al. [42], where the following parameters were considered: purity of the reference material based on the certificate, weighing of the analyte, preparation of stock, working, spike and calibrator solutions, preparation of the ISTD solution, sample preparation of the calibrators, measurement of the calibrators and generation of the calibration curve, preparation and measurement of the unknown samples as well as evaluation of the sample results.To assess the uncertainty in the preparation of the calibrators, a type B evaluation was performed, while all other aspects were evaluated as type A. The total measurement uncertainty was then estimated by combining the type A and type B uncertainties.
For the assignment of reference or target values, multiple sample preparations for each sample were performed on at least 2 different days and the result was calculated as the arithmetic mean (n=x).Further details can be found in Supplementary Material 2.

Selectivity
The goal was to develop an assay with the highest possible selectivity, which was achieved by optimizing the gradient in combination with the use of a reversed phase column (Agilent Zorbax Eclipse XDB-C8).Thus, phenobarbital and primidone, which can be at least partially metabolized to phenobarbital (see Figure 2), were baseline separated within 10.0 min.The chromatographic resolution (R) was ≥4.1 for all matrices tested.Selectivity was further determined by analyzing sample pools of analyte-free native human serum, surrogate serum, and native Li-heparin plasma which showed no signals at the expected retention time.In addition, no other polar or apolar matrix components were detected as interferences in the retention time window of phenobarbital in the samples measured during the method comparison study.The ISTD used fulfilled the requirements and did not contain any unlabeled analyte.

Matrix effect and specificity
To reduce matrix-dependent effects caused by salts, proteins, and phospholipids, a protein precipitation protocol followed by a high dilution step was introduced.During the post-column infusion experiment, no significant shift in ionization to the expected retention time was observed in serum and plasma matrix (K 2 -EDTA plasma, K 3 -EDTA plasma, and Li-heparin plasma).MEs were further investigated by comparing the slopes and coefficients of determination of calibrations in different matrices (native serum, surrogate serum, and native Li-heparin plasma) with the calibration in neat solution (mobile phase A).The mean slopes (n=6 sample preparations) were found to be 0.215 (95% CI 0.214 to 0.217) for neat solution, 0.215 (95% CI 0.214 to 0.216) for native serum, 0.218 (95% CI 0.216 to 0.220) for surrogate serum and 0.217 (95% CI 0.215 to 0.219) for native Liheparin plasma.The CIs of the slopes overlap, indicating that they are not significantly different from each other and supporting the absence of MEs.Mean r 2 values were ≥0.99 independent of the matrix used for calibration.When matrix samples were measured against the neat matrix, the relative bias for all matrices and levels ranged from −2.2 to 3.7 % with CVs of less than 4.5 %.
In addition, analyte peak areas, ISTD peak areas, and area ratios of spiked matrix samples were compared with spiked neat samples (75% methanol [v/v]) after protein precipitation.Recoveries of the analyte and ISTD areas were determined for all matrices and levels and ranged from 100 to 107 % and 100 to 105 %, respectively (Table 2).The corresponding area ratios were between 100 and 102 %, confirming the absence of a ME and compensating effect of the labeled ISTD.Overall, these data show that the ME of this method is negligible.

Linearity
The linearity of the method was demonstrated by analyzing six native serum calibration curves covering ±20 % of the measurement range a (1.40 to 90.0 μg/mL).The residuals were randomly distributed in a linear regression model and were therefore selected for assay calibration.Coefficient of determination were 0.999 for all individual calibrations.Based on these results, the serially diluted samples 1 to 11 were evaluated and showed a linear relationship with a coefficient of determination of 0.999.The relative deviation ranged from −0.7 to 2.4 % and the CV was determined to be ≤2.7 %.

Lower limit of the measuring interval and limit of detection
The LLMI was defined using spiked samples at the lowest calibrator level (1.92 μg/mL).Relative bias showed a deviation of 2.3 % and the CV was found to be 0.5 %.The LOD was estimated to be 0.697 μg/mL.

Accuracy and precision
Accuracy and precision were evaluated within a multi-day validation experiment using four spiked serum and plasma samples (2.60, 8.00, 25.0 and 50.0 μg/mL) and two native patient samples.Two different operators prepared each level in triplicate on 1 day over 5 different days and the samples were injected two times (n=60, measurements).Dilution integrity of the method was demonstrated using highconcentrated samples (80.0 and 100 μg/mL).
Variability components as between injections, between preparations, between calibrations, and between days were determined using an ANOVA-based variance component analysis.Intermediate precision including variances as between-day and -calibration was found to be less than 3.2 %.The repeatability CV including between-preparation and -injection ranged from 1.3 to 2.0 % across all concentration levels and regardless of the type of sample (Table 3).The trueness was determined using spiked samples by comparing the mean measured and calculated concentrations.The relative mean bias (n=6 sample preparations) within the measurement interval ranged from −3.0 to −0.7 % for serum samples and from −2.8 to 0.8 % for plasma samples.High concentrated samples showed a bias between −4.1 and −1.1 % (Table 4).

Stability
Samples were found to be stable at 7 °C for 6 days on the autosampler with recoveries ranging from 96 to 104 % when compared to freshly prepared serum samples.Stability of spiked frozen serum calibrator and QC materials stored at −20 °C was shown for 27 days.The recoveries ranged from 94 to 102 % compared to freshly prepared serum calibrators.

Equivalence of results between independent laboratories
The method comparison study containing 127 samples (native serum and plasma patient samples as well as spiked samples) was performed between site 1 (Dr.Risch Ostschweiz AG, Buchs SG) and site 2 (Roche Diagnostics GmbH, Penzberg).Four samples were found to be below the lower limit of measuring interval and therefore excluded from the study.In addition, three samples were highlighted as outliers using the LORELIA outlier test [48].Two of the samples were most likely mixed up.Passing-Bablok regression analysis showed a good agreement between the two laboratories and resulted in a regression equation with a slope of CI, confidence interval.Conversion factor µg/mL to µmol/L: ..
Schierscher et al.: An ID-LC-MS/MS-based candidate RMP for the quantification of phenobarbital 1.00 (95% CI 0.99 to 1.01) and an intercept of 0.15 (95% CI 0.00 to 0.27).The Pearson correlation coefficient was ≥0.998 (Figure 3A).The Bland-Altman analysis showed a mean bias of 1.0 % with a 95% CI interval from 0.4 to 1.6 (Figure 3B).The data scatter was independent of the analyte concentration, ranging from −5.6 to 7.6 % (lower limit CI interval from −6.6 to −4.6 %, upper limit CI interval from 6.5 to 8.6 %).Precision performance evaluation within the second laboratory showed CVs for intermediate precision and repeatability of ≤2.4 % and ≤1.8 %, respectively.The results showed that the suggested RMP is transferable between independent laboratories and meets the requirements.

Uncertainty of results
The estimation of uncertainty of the calibrators was done as a type B evaluation.In the precision experiment all other aspects, e.g., calibration, sample preparation, measurement, and evaluation of the sample result, were evaluated as type A uncertainty.Measurement uncertainties (k=1) for single measurements of serum samples ranged from 1.9 to 3.3 % and was reduced to 0.9 to 1.6 % for target value assignment (n=6, three measurements on 2 days) (Tables 5 and 6).With the repeated measurements scheme in target value assignment the CV a figure of merit derived from routine data (see Introduction) can be met [32,36].Consequently, the expanded measurement uncertainties (k=2) ranged from 3.8 to 6.6 % for single measurements and from 1.9 to 3.3 % for target value assignment.

Discussion
The presented RMP candidate based on ID-LC-MS/MS allows the measurement of phenobarbital levels in human serum and plasma.The design was developed based on routine measurements and targeted therapeutic analyte concentrations in human serum and plasma specimens.To prepare samples for the selected analytical method, tandem MS with ESI in the negative mode and reversed-phase chromatography, protein precipitation followed by dilution was found to be more effective than liquid-liquid extraction or solid-phase extraction.
The uncertainty contribution of the measurement was reduced by designing a single analyte RMP.This ensured that LC-MS/MS instrument settings, such as the MS source parameters and the LC mobile phase compositions, were tailored to the specific needs of phenobarbital based on its physicochemical target analyte properties.One may argue that an experimental limitation of the presented procedure might be, that the quantifier transition relies on the detection of a rather small fragment with a mass of 42 Da.This fragment (an amide moiety) however showed high, stable intensities and was not affected by interferences.Since it is also commonly used in research and routine practices, we envisioned that this fragment was a good fit for setting up a phenobarbital candidate RMP [23,[49][50][51][52].
Optimization of the presented sample preparation included fluid handling, selection of optimal pipettes; protein precipitation with equilibration times and dilution into Passing-Bablok regression plot including the Pearson regression analysis for the method comparison study of the RMP (n=120 patients) between the independent laboratories (site 1 Dr. Risch Ostschweiz AG and site 2 Roche Diagnostics GmbH).Passing-Bablok regression analysis resulted in a regression equation with a slope of 1.00 (95% CI 0.99 to 1.01) and an intercept of 0.15 (95% CI 0.00 to 0.27).The Pearson correlation value was ≥0.998.(B) Bland-Altman plot resulted in an inter-laboratory measurement bias of 1.0 % (95% CI interval from 0.4 to 1.6 %) and a 2S interval of the relative difference of 6.6 % (lower limit CI interval from −6.6 to −4.6 %, upper limit CI interval from 6.5 to 8.6 %).
the linear range of the MS detectora point to remember when the point is transferred to an instrument with different absolute ion yield.Designing the RMP as single analyte method allowed to maximize the number of data points across the recorded chromatography peaks, which lead to a better (less fluctuating) peak area determination by the integration software.
The validation study confirmed that the developed analytical method met the sensitivity, selectivity, and reproducibility requirements for an RMP.In addition, the absence of MEs was evaluated using a calibration slope comparison and an ion yield attenuation experiment.The measurement uncertainty for multiple measurements (n=6) was found to be within the desired target measurement uncertainty of 3.4 % (k=2).
The successful transfer of the method to a second independent laboratory demonstrated that such a transfer can be performed without significant inter-laboratory bias.This indicates that the calibration solution preparation and sample preparation protocols are robust.In addition, the method comparison study showed that the method is capable of processing a high volume of patient samples in a CV, coefficient of variation.Conversion factor µg/mL to µmol/L: ..The measurement uncertainty of the whole approach for a target value assignment estimated as a combination of the uncertainty of calibrator preparation (type B uncertainty) and uncertainty of the precision experiment (type A uncertainty) are given in bold.
relatively short time, making it suitable for taking a leading role in the traceability chain, conducting method comparison studies and resolving problematic routine samples.Although the JCTLM database stated that the phenobarbital reference measurement procedure is still present, the activity status of the NIST RMP was unclear.In addition, there is no measurement service listed [53].Since we received no positive answer of NIST on the current availability of the JCTLM procedure C10RMP6, a formal comparison of this candidate RMP to the existing RMP was not possible.Due the lack of NIST based reference material (see introduction) not even an indirect comparison was feasible.
Another limitation of the performed method transfer might be that the instrumentation was rather similar in both laboratories.We are however very confident that changing equipment to other vendors will not influence the central figures of merit of the RMP outlined here.We did prove this in the past [54], that method transfer to other LC-MS/MS setups is unproblematic.However, we encourage the readership that if such undertaking is carried out, that central parameters as the LLMI, the measurement range, the phenobarbital retention time and the recorded SRM must be closely monitored to be comparable to the method presented here.

Conclusions
In this paper a modernized LC-MS/MS-based candidate RMP for phenobarbital in human serum and plasma is presented which serves as traceable and reliable platform for the standardization of routine assays and evaluation of clinically relevant samples.

Figure 1 :
Figure 1: Schematic overview of set calibrator and control levels, which were chosen to allow an optimal coverage of measurement and therapeutic reference ranges.Black circles indicate calibrator spike solution, black triangles are QC spike solutions 1-4, the solid black line is the measurement range, the dashed black line is the therapeutic reference range, and the black diamond is the alert level.QC, quality control.Conversion factor µg/mL to µmol/L: 4.3.

Figure 3 :
Figure 3: Results from the patient sample-based phenobarbital 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=120 patients) between the independent laboratories (site 1 Dr. Risch Ostschweiz AG and site 2 Roche Diagnostics GmbH).Passing-Bablok regression analysis resulted in a regression equation with a slope of 1.00 (95% CI 0.99 to 1.01) and an intercept of 0.15 (95% CI 0.00 to 0.27).The Pearson correlation value was ≥0.998.(B) Bland-Altman plot resulted in an inter-laboratory measurement bias of 1.0 % (95% CI interval from 0.4 to 1.6 %) and a 2S interval of the relative difference of 6.6 % (lower limit CI interval from −6.6 to −4.6 %, upper limit CI interval from 6.5 to 8.6 %).

Table  :
MS/MS parameters of phenobarbital and [  H  ]-phenobarbital.Schierscher et al.: An ID-LC-MS/MS-based candidate RMP for the quantification of phenobarbital be ≥0.99 to exclude a ME.Recovery of calibrator samples (n=6 sample preparations) in surrogate matrix, native human serum, and plasma was evaluated using neat calibrators as standards.

Table  :
Matrix effect (ME) data of native serum, surrogate serum and plasma compared to neat analyte solution.Analyte peak areas, ISTD peak areas, and analyte/ISDT area ratios as used in analyte quantification were investigated.Means from five-fold analysis were used as data input.The relative matrix effect (ME) was calculated as ME (%)=set /set  × , where set  corresponds to the respective matrix samples and set  to the neat samples.

Table  :
Precision performance parameters for phenobarbital quantification using the candidate RMP (n= measurements).RMP, reference measurement procedure; Conversion factor µg/mL to µmol/L: ..The coefficients of variation for repeatability and intermediate precision, which were determined from the individual variances, are printed in bold.

Table  :
Bias and % CI of native serum and native Li-heparin plasma samples (n=).The mean bias and corresponding confidence intervals were calculated using the individual sample biases of n= preparations.

Table  :
Overview of measurement uncertainty for phenobarbital quantification with the candidate RMP in serum samples for single measurements.Patient sample  Patient sample  . µg/mL . µg/mL . µg/mL . µg/mL . µg/mL . µg/mL CV, coefficient of variation; RMP, reference measurement procedure.Conversion factor µg/mL to µmol/L: ..The measurement uncertainty of the whole approach for a single measurement estimated as a combination of the uncertainty of calibrator preparation (type B uncertainty) and uncertainty of the precision experiment (type A uncertainty) are given in bold.