Candidate reference measurement procedure based on isotope dilution-two dimensional-liquid chromatography–tandem mass spectrometry for the quantification of androstenedione in human serum and plasma

Chemicals and reagents

LC-MS grade methanol (CAS 67-56-1), acetonitrile (CAS 75-05-8) and formic acid (99 % ULC/MS grade CAS 64-18-6) were purchased from Biosolve (Valkenswaard, The Netherlands). Ammonium fluoride (CAS 12125-01-8) was purchased from Sigma Aldrich (Taufkirchen, Germany). Isopropanol (CAS 67-63-0) and zinc sulfate heptahydrate (CAS 7446-20-0) were purchased from Honeywell (Charlotte, NC, USA).

Water was purified using a Millipore Milli-Q IQ 7000 system from Merck (Darmstadt, Germany). Steroid-free human serum (ID 11717863001) and albumin RPLA 1 (Surrogate matrix, Assay Quality, ID 11726536001) were obtained from Roche Diagnostics GmbH (Mannheim, Germany). Mixed-gender native plasma pools (Li-Heparin [CUST-BB-17022020-4C], dipotassium (K₂) ethylene diamine tetraacetic acid (EDTA) [CUST-BB-17022020-4E], and tripotassium (K₃) EDTA [CUST-BB-17022020-4D]) were obtained from Biotrend (Cologne, Germany). As surrogate matrix an albumin matrix with a concentration of 1 g/dL in phosphate-buffered saline was used. Native sample pool was composed of samples from 5 individual patients with endogenous ASD levels. All native sample pools were prepared in accordance with the Declaration of Helsinki. The primary reference material ASD (CAS 63-05-8) was purchased from NMIA (M955), and its isotope-labeled internal standard (ISTD), ¹³C₃-ASD (CAS 327048-86-2), was obtained from Supelco™ (Merck). 6-Dehydrotestosterone (CAS 2484-30-2) was obtained from the European Directorate for the Quality of Medicines (EDQM). 5β-Androstan-3,17-dione (CAS 1229-12-5), 4-androsten-3α-ol-17-one (CAS 2791-99-3), and 4-androsten-3β-ol-17-one (CAS 571-44-8) were purchased from Steraloids, Inc. (Newport, RI, USA). 1-Testosterone (CAS 65-06-5) was sourced from Cayman Chemical (Ann Arbor, MI, USA). Testosterone (CAS 58-22-0) and epitestosterone (CAS 481-30-1) were obtained from Supelco^® (Merck) and dehydroepiandrosterone (DHEA, CAS 53-43-0) was obtained from Sigma-Aldrich (Merck).

Preparation of calibrators and quality control (QC) samples

Three calibration stock solutions were independently prepared by weighing 1 mg (± 5 %) of the reference material NMIA (M955 Batch 04-S-02, 99.6 ± 0.3 % [k=2]) using an ultra-microbalance and dissolving it in acetonitrile to 100 mL, resulting in a 10 μg/mL (34.9 μmol/L) solution. The exact concentration was based on the material’s purity and weight. Two stock solutions were further diluted with methanol/water (1 + 1, v + v) to prepare 100 ng/mL (349 nmol/L) working solutions. These were used to prepare eight calibrator spike solutions of varying concentrations in methanol/water (1 + 1, v + v). Each calibrator spike solution was diluted in an albumin surrogate matrix (1 + 49, v + v), resulting in a range of 0.00800–12.0 ng/mL (0.0279–41.9 nmol/L).

Quality control samples (QCs) were prepared using the third independent primary stock solution (10 μg/mL, ± 5 %). This solution was diluted with methanol to create a working solution (0.1 μg/mL) used to prepare respective spike solutions. Final matrix based QC concentrations were 0.0240, 2.00, and 10.0 ng/mL (0.0838, 6.98, and 34.9 nmol/L). Additionally, two native patient samples were used as QCs to establish a control chart.

ISTD solution

The commercially available stock solution, ¹³C₃-ASD 100 μg/mL (15 µL), was transferred into a volumetric flask (250 mL) and diluted with methanol/water (1 + 1, v + v) up to the calibration mark, to obtain an ISTD solution with a concentration of 6 ng/mL (20.7 nmol/L).

Sample preparation

Native serum, plasma (lithium-heparin, K₂EDTA, K₃EDTA), and albumin surrogate matrix (1 g/dL) can be used as sample matrices. A 200 µL sample was transferred into a 1.5 mL screw cap, and 40 µL of internal standard working solution was added. Samples were incubated on an overhead shaker for 15 min at room temperature. Then, 600 µL of precipitation reagent was added, followed by shaking for 10 min on a thermomixer (Eppendorf 5,382 thermomixer C) at 1,400 rpm and centrifuging for 20 min at 15,000 rcf (centrifuge 5,430 R, Eppendorf). The supernatant (600 µL) was loaded onto 1 cc Support Liquid Extraction (SLE) Waters HLB Oasis Prime cartridges. Cartridges were washed with 1,000 µL Milli-Q water. Extraction was performed by adding 200 µL acetonitrile (twice), evaporating the extract, and reconstituting the residue in 100 µL methanol/water (1 + 1, v + v). The solution was transferred into a high-performance liquid chromatography (HPLC) vial with a 100 µL insert.

Liquid chromatography-tandem mass spectrometry (LC-MS/MS)

A two-dimensional heart-cut LC approach was used for the chromatographic purification. This approach involves the use of two orthogonal stationary phases: a Waters Acquity Ethylene Bridged Hybrid (BEH) C18 column (50 × 2.1 mm, 1.7 µm, Waters Corporation, Milford Massachusetts, USA) and a Raptor Biphenyl column (150 × 2.1 mm, 2.7 µm, Restek, Bellefonte, Pennsylvania, USA). The mobile phases used in this method consist of water (A1) and acetonitrile (B1) in the first dimension, and 0.2 mM ammonium fluoride in water (A2) and methanol/acetonitrile (7 + 3, v + v) (B2) in the second dimension. ASD was detected in multiple reaction monitoring (MRM) mode using an AB Sciex QTrap 6,500 + mass spectrometer operating in electrospray positive ionization mode. The quantifier ion transition (ASD m/z 287.1 → 97.0) and the corresponding internal standard transition (¹³C₃-ASD m/z 290.1 → 100.0) served as the basis for ASD quantitation. Supplementary Material 1 includes detailed LC and MS methods.

System suitability test (SST)

To ensure long-term traceability, an SST was established to check sensitivity and chromatographic resolution before each sequence. Two levels (SST1 and SST2) in methanol/water (1 + 1, v + v) contained ASD at 0.00800 ng/mL and 12.0 ng/mL (0.0279 nmol/L and 41.9 nmol/L), respectively.

SST1 was injected to check sensitivity, with the signal-to-noise (S/N) ratio calculated using Analyst software (AB Sciex). An S/N ratio of ≥10 was required. SST2 was injected to check for carryover, followed by a double blank injection. The peak observed in the first blank was required to be≤20 % of the peak area of calibrator 1 to pass the SST. A retention time of 7.9 ± 0.5 min for ASD was required for both SST1 and SST2.

Calibration data processing and structure of analytical series

Calibrator levels were measured at increasing concentrations at the beginning and end of every sequence, using 8 levels to generate the final calibration function through linear regression (y=ax + b) of area ratios (analyte/internal standard).

For processing the raw data file, the Analyst software (v1.6.2 or higher) was used with Intelli Quan as Quantitation Integration Algorithm. ASD and its internal standard showed a retention time of 7.9 min and were integrated with a 30 s window. Raw data were processed with a smoothing factor of 3. The calibration curve was linear with a 1/x weighting.

The sequence setup varied based on measurement requirements. For reference value assignments, sample preparations (n=x) depended on the desired measurement uncertainty, with samples measured on at least two different days. For method comparison studies or complaint sample measurements, samples were prepared with n=1.

Method validation

Assay validation and determination of MU were performed following established guidelines for validation. These guidelines include the Clinical & Laboratory Standards Institute’s C62A Liquid Chromatography-Mass Spectrometry Methods [13], and the Guide to the expression of uncertainty in measurement (GUM) [14].

Analytical performance specifications (APS) have been calculated based on the European Federation of Clinical Chemistry and Laboratory Medicine biological variation (BV) database [15]. For ASD, several entries can be found in the database for the desirable specifications. Based on these entries the recommended specification for imprecision (CV), bias (B), maximum expanded allowable measurement uncertainty (MAU), and total error (TE) range between 2.9 and 5.8 % CV_RMP, 2.7–5.0 % B_RMP, 3.8–7.7 % MAU_RMP and 9.1–14.5 % TE_RMP [15], 16]. Therefore, these criteria are established as the acceptance criteria for the candidate RMP.

Selectivity

Effective separation of ASD from 6-dehydrotestosterone, 4-androsten-3α-ol-17-one and 5β-androstan-3,17-dione was achieved in the first dimension (Figure 2A). ASD, testosterone, DHEA, 4-androsten-3β-ol-17-one, 1-testosterone and epitestosterone were transferred to the second dimension and successfully separated from ASD with a chromatographic resolution R (FWHM) ≥ 3.2 (Figure 2B).

Figure 1:

Chromatographic separation of androstenedione from potential interfering compounds. Concentration ASD in neat solution (methanol/water [1 + 1, v + v]) 5.00 ng/mL (17.5 nmol/L), concentration of all potential interfering compounds 10.0 ng/mL (34.7–34.9 nmol/L). (A) Chromatogram (overlay) of the 1st dimension. (B) Chromatogram (overlay) for the 2nd dimension. Interferences are separated from ASD with a chromatographic resolution (FWHM) ≥ 5.4.

Figure 2:

Androstenedione LC-MS/MS derived analytical readouts. (A) Chromatogram of calibrator level 1 with a concentration of 0.00800 ng/mL (0.0279 nmol/L) spiked in surrogate matrix (left) and ISTD (right); (B) pooled native patient human serum with a concentration of 1.48 ng/mL (5.17 nmol/L), analyte (left) and ISTD (right); (C) fortified Li-heparin patient sample with a concentration of 5.52 ng/mL (19.3 nmol/L), analyte (left) and ISTD (right). ISTD, internal standard; SRM, standard reference material.

During the analysis, no interfering signals were observed for the quantifier transition (m/z: 287.1 -> 97.0) and the corresponding internal standard transition (m/z: 290.1-> 100.0) within the expected retention time window (7.9 min ± 0.5 min) (Figure 1). This observation was consistent in both the analyte-free surrogate matrix and the native serum pool.

The internal standard used in this method was evaluated for any residual unlabeled analyte, and no significant analyte signal was observed. Therefore, it can be confidently used as the internal standard for our novel RMP.

Matrix effects (ME)

The results of the post column infusion experiment indicated that none of the tested matrices exhibited ion suppression or enhancement during the retention time of the analyte of interest.

The slopes of the standard lines were determined to be 0.770 (95 % CI 0.768–0.773) for the neat solution, 0.762 (95 % CI 0.758–0.766) for the surrogate matrix, 0.761 (95 % CI 0.748–0.774) for the native matrix, and 0.773 (95 % CI 0.759–0.778) for the Li-Heparin plasma. The comparable slopes and overlapping corresponding confidence intervals suggest that they are not significantly different from each other, confirming the absence of a matrix effect. Additionally, excellent correlation was observed (r≥0.999), regardless of the matrix used. Calibrator samples in the surrogate matrix, native serum, and Li-Heparin plasma were evaluated as controls using the neat calibration as the standard. The average recovery for the surrogate matrix was 99 % (95 % CI 98–101), for the native serum matrix was 99 % (95 % CI 98–100), and for the Li-Heparin plasma was 98 % (95 % CI 97–100), thereby further proving matrix independence.

Linearity

The residuals of the linear regression model, which utilized a 1/x weighting, were randomly and uniformly distributed. Excellent correlation (r≥0.999) was achieved for all individual calibration curves and acceptance criteria were fulfilled.

Additionally, the linearity of the method was confirmed by analyzing serially diluted samples. The measurement results exhibited a linear relationship with an r≥0.999. Recoveries for the serial dilutions ranged from 99 to 102 %.

Lower limit of the measuring interval (LLMI)

The LLMI was determined using a spiked matrix sample with a concentration of 0.00800 ng/mL (0.0279 nmol/L), which corresponds to the concentration of the lowest calibrator level. The relative deviation (n=6) was 3.5 % and CV was determined at 3.1 %, thus fulfilling the defined APS.

Precision, trueness and accuracy

The intermediate precision, which includes variances related to between-day calibration, preparation, and injection, ranged between 1.2 and 1.9 %. The repeatability CV ranged between 0.9 and 1.6 % across all concentration levels (Table 1).

Trueness and accuracy were assessed by analyzing spiked surrogate matrix, steroid-free human serum, and native Li-Heparin plasma pools. The relative mean bias ranged from 0.6 to 2.2 %, regardless of the matrix and concentration level.

Dilution integrity was evaluated using two spiked samples at concentration levels of approximately 13.0 ng/mL and 20.0 ng/mL (45.4 nmol/L and 69.8 nmol/L). The mean deviation was found to be within −0.5 % and −1.0 % (Table 2). Overall, the method demonstrated no statistically significant bias and showed independence from the matrix used. The data presented fulfill the APS as described above.

Sample stability

Processed samples remained stable for 14 days at 8 °C, with average recoveries between 98 and 101 %. Furthermore, the stability of spike solutions and spiked frozen control materials stored at −80 °C was assessed over a 12-week period. Average recoveries for spike solutions ranged from 107 to 109 %, while for spiked matrix based materials, they ranged from 96 to 106 %.

Equivalence of results between independent laboratories

The method comparison study involved 76 samples that were measured at two separate laboratories (Site 1: Roche Diagnostics GmbH, Penzberg and Site 2: Department of Pediatrics and Adolescent Medicine, University Hospital Erlangen). These included 36 unaltered native leftover patient samples, 20 pools, 10 spiked native samples and 10 diluted samples. One pooled sample was identified as an outlier using the LORELIA outlier test [18] and was therefore excluded from the analysis.

The regression analysis resulted in a regression equation with a slope of 0.98 (95 % Cl 0.97 to 0.99) and an intercept of 0.000 (95 % Cl −0.003 to 0.006). The Pearson correlation value was r≥0.999. The interlaboratory measurement bias was −0.9 % (95 % CI interval from −2.8 to 1.0 %) and the 2S interval of the relative difference was ± 15.9 % (lower limit CI interval from −20.0 to −13.6 %, upper limit CI interval from 11.8 to 18.2 %) (Figure 3). The 3-day precision experiment at Site 2 showed comparable coefficient of variations (CVs) to those observed at Site 1, with intermediate precision of 1.8–3.8 % and repeatability of 0.5–3.7 %. Two results had to be excluded due to measurement issues for the highest level. Both the data scatter and data bias prove that the proposed RMP is transferable between independent laboratories.

Figure 3:

Results from the patient sample-based androstenedione 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=75) samples between the independent laboratories (Laboratory 1: Roche Diagnostics GmbH, Penzberg and Laboratory 2: Department of Pediatrics and Adolescent Medicine, University Hospital Erlangen). The regression analysis resulted in a regression equation with a slope of 0.98 (95 % Cl 0.97 to 0.99) and an intercept of 0.000 (95 % Cl −0.003 to 0.006). The Pearson correlation value was≥0.999. (B) Bland–Altman plot for the method comparison study of the RMP (n=75) between two independent laboratories (Laboratory 1 and Laboratory 2). The interlaboratory measurement bias was −0.9 % (95 % CI interval from −2.8 to 1.0 %) and the 2S interval of the relative difference was ± 15.9 % (lower limit CI interval from −20.0 to −13.6 %, upper limit CI interval from 11.8 to 18.2 %). CI, confidence interval; RMP, reference measurement procedure; SD, standard deviation.

A second method comparison study was conducted between the RMP (performed at Site 1) and an in-house developed LC-MS/MS assay performed at Site 3 (Institute of Laboratory Medicine at Leipzig University Hospital) [9]. In total, 76 samples were measured, whereas 13 samples were found below the LOQ of the in-house developed LC-MS/MS assay. Therefore, the analysis included 63 samples.

Performing a Passing Bablok regression, the method yielded a regression equation with a slope of 0.92 (95 % CI 0.90 to 0.98) and an intercept of 0.002 (95 % CI −0.065 to 0.035). The correlation coefficient was r≥0.993. Furthermore, the Bland-Altman analysis demonstrated a bias of −6.9 % (95 % CI −9.6 to −4.2 %), which was found to be statistically different from zero (Figure 4). This indicates the presence of systematic error or bias in the measurements, contributing to the overall imprecision of the study. Additionally, the 2S interval of the relative difference, with a range of ± 21.2 % (95 % CI −32.7 to −23.4 % and 9.6–18.9 %).

Figure 4:

Results from the patient sample-based androstenedione method comparison study performed between the candidate RMP and a well-established routine RMP. (A) Passing–Bablok regression plot including the Pearson regression analysis for the method comparison study of the RMP (n=63) samples between the independent laboratories (Laboratory 1: Roche Diagnostics GmbH, Penzberg and Laboratory 3: Site 3: Institute of Laboratory Medicine at Leipzig University Hospital). The regression analysis resulted in a slope of 0.92 (95 % CI 0.90 to 0.98) and an intercept of 0.002 (95 % CI −0.065 to 0.035). The correlation coefficient was r≥0.993. Bland-Altman analysis demonstrated a bias of −6.9 % (95 % CI −9.6 to −4.2 %) and the 2S interval of the relative difference was ± 21.2 % (95 % CI −32.7 to −23.4 % and 9.6 to 18.9 %

Equivalence of results between the candidate and the JCTLM listed RMP

Equivalence between the candidate RMP and the JCTLM-listed RMP was shown by measuring 20 anonymized native patient samples covering the measuring range. During the analysis, one sample was found to be below the quantification limit of the JCTLM-listed RMP and was subsequently excluded from the dataset. The subsequent data analysis using Passing-Bablok revealed a high level of agreement, as evidenced by the regression equation with a slope of 0.98 (95 % CI 0.93 to 1.05) and an intercept of −0.01 (95 % CI −0.06 to 0.03), along with a strong correlation coefficient of r≥0.997. The measurement bias was −2.8 % (95 % CI interval from −5.2 to −0.4 %) and the 2S interval of the relative difference was ± 9.8 % (lower limit CI interval from −16.7 to −8.4 %, upper limit CI interval from 2.7 to 11.0 %) (Figure 5).

Figure 5:

Results from equivalence testing between the JCTLM listed and the candidate RMP. (A) Passing–Bablok regression plot including the Pearson regression analysis for the method comparison study of the RMP (n=19) samples between the JCTLM listed and the candidate RMP. The regression analysis resulted in a regression equation with a slope of 0.98 (95 % Cl 0.93 to 1.05) and an intercept of −0.01 (95 % Cl −0.06 to 0.03). The correlation coefficient was r≥0.997. (B) Bland–Altman plot for the method comparison study of the RMP (n=19) between the JCTLM listed and the candidate RMP. The measurement bias was −2.8 % (95 % CI interval from −5.2 to −0.4 %) and the 2S interval of the relative difference was ± 9.8 % (lower limit CI interval from −16.7 to −8.4 %, upper limit CI interval from 2.7 to 11.0 %). CI, confidence interval; JCTLM, joint committee on traceability in laboratory medicine RMP, reference measurement procedure; SD, standard deviation.

Naturally, both measurement procedures include inherent measurement errors, and the error in the calculated difference between the two procedures is consequently larger. Each RMP must meet the stricter performance specifications of CV_ref≤2.9 % and B_ref≤2.7 %. According to the total error concept, the maximum allowable difference is calculated as 2B_ref + 1.96 * sqrt(CV_ref ² + CV_ref ²), resulting in 13.4 %. The presented results align with the pre-determined acceptance criteria and the accuracy and intermediate precision data from individual methods.

Estimation of measurement uncertainty (MU)

The measurement process for single measurements exhibited an overall uncertainty ranging from 1.4 to 2.1 % regardless of the sample concentration, as indicated in Table 3. To account for a confidence level of ∼95 % assuming a normal distribution, the resulting overall uncertainty was multiplied by a coverage factor of k=2. As a result, the expanded measurement uncertainties fall within the range of 2.7–4.1 %.

For the establishment of reference or target values, multiple sample preparations were carried out for each sample on at least two different days. The average of these results, calculated using n=6, was used. The overall uncertainties were found to range from 0.8 to 1.6 %, with expanded uncertainties ranging from 1.7 to 3.1 % (k=2), as shown in Table 4.