Based on the results of a previous survey of members of the Korean Society of Clinical Chemistry [14], 10 clinical laboratories performing immunosuppressive drug TDM by LC-MS/MS were asked to participate in this study. After enrollment, the test samples were prepared in Asan Medical Center, Seoul, Korea, dispatched to the participating laboratories on dry ice in a Styrofoam box on 23 July 2018, and reached the laboratories within two days. Participating laboratories were asked to perform the test within three days of receiving the test samples. For each test sample, participating laboratories were asked to make duplicate aliquots for testing before the sample preparation step. Test results and method information, including the dates of reception and test, calibrator, internal standard, LC manufacturer, MS manufacturer and model, and extraction method, were collected in a formatted Excel file. Additional surveys regarding calibrator reconstitution protocols were conducted by e-mail. The data was compiled and analyzed at Asan Medical Center. The Institutional Review Board of Asan Medical Center exempted the approval for this study (reference number: 2018-1070).

Surplus samples from patients undergoing routine TDM were pooled to achieve three different concentrations of test samples for each drug. The samples included EDTA whole blood for TAC, SIR, EVE, and CSA and EDTA plasma for MPA. The samples were collected for seven days and stored at 4°C up to seven days before pooling. The samples were pooled based on their reported concentrations to achieve concentrations below, within, and above the therapeutic range of each drug. Each 1 mL pooled sample was distributed in a glass vial after mixing on a roller mixer for 1 hour. The minimum number of vials for each test sample was 10; for TAC concentration 2 (TAC-2), 30 vials were prepared to evaluate homogeneity and short-term stability. The test samples were stored at -70°C before delivery. Thus, the samples underwent only one freeze-thaw cycle before being tested at the participating laboratories. For TAC-2, 10 vials were tested in triplicate on the day of production to evaluate homogeneity. To evaluate short-term stability, two vials were stored at 5°C for five days and then tested for TAC in duplicate. These four results were compared with the homogeneity values. In the homogeneity assessment, a between-sample standard deviation < 0.3$S_{\stackrel{-}{x}}$ was considered acceptable. In the stability assessment, a difference from the homogeneity mean < 0.3$S_{\stackrel{-}{x}}$ was considered acceptable.

All test samples were in a frozen state on arrival at each laboratory. Details of the methods used in each laboratory are shown in Table 1. All laboratories conducted simultaneous detection with the following combinations: TAC+SIR+EVE+CSA (six laboratories) and TAC+SIR+EVE (three laboratories); one laboratory (Lab F) that was supposed to perform a CSA test with a TAC+SIR+EVE+CSA combination did not receive the CSA test samples due to a communication error. In total, TAC detection was performed in 10 laboratories, SIR detection in nine laboratories, EVE detection in nine laboratories, CSA detection in eight laboratories, and MPA detection in three laboratories. All participating laboratories used commercial calibrator sets from Chromsystems Instruments & Chemicals GmbH (Gräfelfing, Germany).

Based on the survey results of the calibrator reconstitution methods of participating laboratories, we conducted an experiment on calibrator reconstitution and storage conditions. To evaluate the impact of the calibrator reconstitution process and prolonged storage after reconstitution, we used the lyophilized calibrator 6PLUS1 Multilevel Whole Blood Calibrator Set (Lot no. 4917; Chromsystems Instruments & Chemicals GmbH) with six different concentrations plus one blank value. The calibrator contained TAC, SIR, EVE, and CSA; however, CSA could not be measured in the experimental test system.

The calibrators were stored at -70°C until use and then thawed at room temperature (20–25°C) for 30 minutes before adding 2.0 mL of distilled water to each vial. The contents of the vials were then mixed under two different conditions: static incubation (sitting) for 30 minutes at room temperature (20–25°C) after the addition of water followed by mixing on a roller for 60 minutes (Condition A0) or mixing on a roller for 20 minutes without static incubation (Condition B0). Aliquots from these vials were tested in four replicates. Aliquots from Condition A0 vials were initially frozen at -20°C and then thawed on different days, resulting in four storage conditions: 42 days at -20°C (Condition A1), 35 days at -20°C and seven days at 5°C (Condition A2), 28 days at -20°C and 14 days at 5°C (Condition A3), and 14 days at -20°C and 28 days at 5°C (Condition A4). These aliquots were tested in duplicate on the same day.

Simultaneous quantification of TAC, SIR, and EVE was performed for the calibrators in the same manner as for patient samples at Asan Medical Center. Briefly, 40 µL samples were prepared by protein precipitation using 80 µL of aqueous 0.1 M zinc sulfate and 200 µL of acetonitrile containing the internal standards. The internal standard solution consisted of ascomycin (Cerilliant, Round Rock, TX, USA), SIR-d₃ (IsoSciences, Ambler, PA, USA), and EVE-d₄ (Cerilliant). HPLC was conducted using an ACQUITY UPLC I-Class system (Waters Corporation, Milford, MA, USA) equipped with an ACQUITY UPLC HSS SB C18 column (30 mm × 2.1 mm, 4 µm; Waters Corporation). The system was operated at a flow rate of 0.2 mL/min with a column temperature of 55°C. The mobile phase for isocratic elution consisted of 50% methanol (HPLC grade; Merck, Darmstadt, Germany) and 50% ammonium acetate buffer (2 mM, pH 2.7, with ≥ 98% acetate ammonium; Sigma-Aldrich, St. Louis, MO, USA). MS was detected on a Xevo TQ-S LC-MS/MS system (Waters Corporation) using positive electrospray ionization and argon as the collision gas. Peak areas were obtained by multiple reaction monitoring of the following mass transitions for quantification: TAC (m/z 821.5 > 768.5), SIR (m/z 931.6 > 864.6), EVE (m/z 975.7 > 908.6), ascomycin (m/z 809.5 > 756.5), SIR-d₃ (m/z 934.6 > 864.6), and EVE-d₄ (m/z 979.7 > 912.6). Response values were calculated using the resultant peak areas (area of analyte/area of internal standard). LC-MS/MS analysis and data acquisition were managed using Masslynx software version 4.1 (Waters Corporation). TargetLynx software version 4.1 (Waters Corporation) was used for chromatographic peak detection and baseline determination.

Data from the participating laboratories were used to estimate test reproducibility. The repeatability standard deviation (Sr) is the root mean square value of all differences in w_i, where w_i is the difference between the duplicate measurement results in the i^th laboratory and the number of participating laboratories is p:

The standard deviation of the sample averages The standard deviation of the sample averages ($S_{\stackrel{-}{x}}$) is:

where $d_i={\stackrel{-}{x}}_i-\stackrel{=}{x}$ and the average of the duplicate measurement results in the i^th laboratory is ${\stackrel{-}{x}}_i$ and the average of the $\stackrel{-}{x}$ for one test sample is $\stackrel{=}{x}$.

When ${(S_{\stackrel{-}{x}})}^2>{(S_r)}^2/2,$ the reproducibility standard deviation ($S_R$) is :

The reproducibility coefficient of variation (CV_R) is :

Robust statistical methods based on the medians were also used to describe the result distribution. The robust repeatability standard deviation (S_r^*) was calculated using Algorithm S (ISO13528:2015) [15, 16]. The robust standard deviation of the sample averages ($S_{\stackrel{-}{x}}^{*}$) is :

which is the scaled median absolute deviation (MADe), where $D_i={\stackrel{-}{x}}_i-\mathrm{med}(\stackrel{-}{x})$. The percent difference (D%) was used to describe the deviation in the individual laboratory results, calculated as $\%D_i=100D_i/\mathrm{med}(\stackrel{-}{x})$. The robust reproducibility standard deviation (S_R*) and the robust reproducibility CV (CV_R*) were calculated as $S_R^{*}=\sqrt{{(S_{\stackrel{-}{x}})}^2+{(S_r^{*})}^2/2}$ and ${\%\mathrm{RSD}}_R^{*}=100S_R^{*}/\mathrm{med}(\stackrel{-}{x}),$ respectively. Spearman’s rank correlation coefficient (rho, ρ) was calculated to determine the relationship between the D% values of different test drugs paired with similar drug concentration measurements.

For the result analysis, a multiple linear regression model was used to compare the slopes of the calibration curves (non-weighted):

where the response value is Y, the assigned value of calibrator is X, and the dummy variable of the conditions is D_C. The t-test on coefficient β₂ was performed to determine whether the slopes obtained from the reconstitution or the storage conditions were different from those at the baseline onditions (A0 or A1), with P ≤ 0.05. Data integration, analysis, and visualization were performed using R version 3.6.3 (R Foundation for Statistical Computing, Vienna, Austria).

In the homogeneity test, the overall average of TAC-2 was 8.99 ng/mL. Because the variance of sample averages was < 1/3 of the within-sample variance, the between-sample standard deviation was estimated as 0. Therefore, the homogeneity was considered sufficient. In the short-term stability test, the overall average of TAC-2 was 8.88 ng/mL. Its difference from the homogeneity mean was 0.11 ng/mL, which was lower than the calculated check value (0.17, 0.3$S_{\stackrel{-}{x}}$ of TAC-2). Therefore, the stability was considered acceptable.

Statistical summary of the laboratory measurement results is shown in Table 2. The robust reproducibility CV_S (CV_R*) were > for SIR-1, SIR-2, EVE-1, and EVE-2. A pattern suggesting a positive correlation among test samples was identified when the distribution of D% values was visualized using a heatmap table (Fig. 1A). For correlation analysis between the results for different drugs, the test samples were paired in similar concentrations: SIR-1 and EVE-2 and SIR-2 and EVE-3 (N=18); TAC-1 and SIR-2 (N=9); and TAC-1 and EVE-3 (N=9). The ρ values between SIR and EVE, TAC and SIR, and TAC and EVE results were 0.334 (P=0.175), 0.383 (P=0.308), and -0.150 (P=0.700), respectively. SIR and EVE results showed a weak positive correlation (Fig. 1B).

Additional survey results for the calibrator reconstitution protocols are shown in Table 3. Based on the survey results, Conditions A0 (similar to Lab A protocol) and B0 (similar to Lab H protocol) were selected. The slopes of the calibration curves did not significantly differ between these two conditions (Table 4).

In the experiment on storage conditions, the calibration curve slopes of SIR from Conditions A2 and A3 (P < 0.001 and P = 0.030) and of EVE from Conditions A3 and A4 (P < 0.001 and P = 0.009) were significantly different from those of Condition A1. When the response value corresponding to a concentration, for example, of 6 ng/mL (similar to the mean concentrations of SIR-2) from the baseline condition calibration curve was entered into the experimental condition equation, positive biases were present in conditions A1–A3; however, the percent differences were expected to be in the range of 0.217–0.598% (Table 4).