An isotope dilution-liquid chromatography-tandem mass spectrometry-based candidate reference measurement procedure for the quantification of 17β-estradiol in human serum and plasma

Chemicals and reagents

As primary reference material, E2 (CAS 50-28-2) from the National Metrology Institute of Japan (NMIJ, CRM 6004-a) purchased from LGC (Wesel, Germany) was used. ISTD solutions in acetonitrile, 17β-estradiol-2,3,4–¹³C₃ (¹³C₃-E2, CAS 1261254-48-1, Supelco™) and 17β-estradiol-D₅ (D₅-E2, CAS 221093-45-4, Supelco™) were both obtained from Merck (Darmstadt, Germany). Three levels of E2 in human serum (BCR-576, BCR-577 and BCR-578) were sourced from the Joint Research Centre (Brussels, Belgium).

17α-Estradiol (CAS 57-91-0), estrone (CAS 53-16-7), ethyl acetate (CAS 141-78-6), n-heptane (CAS 142-82-5), dansyl chloride (CAS 605-65-2), and ammonium fluoride (CAS 12125-01-8) were obtained from Merck. ULC/MS grade acetonitrile (CAS 75-05-8), methanol (CAS 67-56-1) and formic acid (CAS 64-18-6) were purchased from Biosolve (Valkenwaard, Netherlands).

Steroid-free human serum was used as calibrator matrix and was purchased from Roche (Penzberg, Germany) or Golden West Diagnostics (Temecula, CA, USA). As an alternative calibrator matrix, a surrogate matrix composed of 1 % bovine serum albumin (Albumin BPLA 1 from Roche Penzberg, Germany) in phosphate-buffered saline (PBS, from VWR, Ismaning, Germany) was used.

Mixed gender native plasma pools (Li-heparin [CUST-BB-31072020-1], dipotassium (K₂)-ethylene diamine tetraacetic acid (EDTA) [CUST-BB-17022020-4E], and tripotassium (K₃) EDTA [CUST-BB-17022020-4D]) were obtained from Biotrend (Colonge, Germany). The native human serum pool consisted of at least five samples. According to the Declaration of Helsinki, all native serum and plasma patient samples were anonymized samples.

qNMR for determining the absolute purity of the standard materials

qNMR experiments were performed on a Jeol 600 MHz NMR equipped with a Helium-cryoprobe. Single-pulse-¹HNMR was utilized for the quantitation (aromatic proton meta to phenolic group; δ=7.10 ppm; 1H; dimethyl terephthalate as qNMR ISTD; CD₃CN as solvent) with an inter-scan delay of 70 s (Supplementary Material 2). The mass fraction value of our in-house method and that of the NMIJ 6004-a matches exactly, both being 98.4 %. Further details relevant to this manuscript are available in the Supplementary Material 2.

Additionally, we also performed quantum chemical calculations to ascertain the conformer distribution of estradiol in solution because the Boltzmann average of these structures constitute the NMR spectrum. Computational calculations to determine the most reactive chemical bond towards deprotonation were also performed to find the optimal resonance for quantitation. Finally DFT single-point calculations were performed at the wB97X-V/def2TZVP level of theory and all the information is presented in the Supplementary Material 2.

Matrix based calibrator and quality control (QC) preparation

Two independent primary stock solutions were prepared by weighing 5 mg E2 (NMIJ, CRM 6004-a) in a disposable weighing boat on an ultra-microbalance (XP6U/M, Mettler Toledo, readability of 0.0001 mg) and dissolving it in a 50 mL volumetric flask containing acetonitrile. The purity specified in the certificate of the reference material CRM 6004-a (98.4 % ± 0.3 %) was incorporated to yield the final stock solution concentrations. Based on these stock solutions four working and eight spike solutions were prepared in 50 % methanol (v + v) which were used for the generation of steroid-free matrix-based calibrator levels (1 + 49 v + v) with concentrations from 5.00 to 5,000 pg/mL (18.4 to 18,357 pmol/L) for the standard range (SR) method (See Supplementary Material 1, Section 3.2.3). Based on the same calibrator preparation concept the calibrator levels for the high sensitivity (HS) method were prepared with individual sample weighings and preparation of working and spike solutions. Final matrix based calibrator levels ranged from 0.400 to 5.00 pg/mL (1.47–18.4 pmol/L) (See Supplementary Material 1, Section 4.2.3).

A third independent weighing was used for the production of quality control samples (QCs) for both measuring ranges, respectively. To ensure accuracy and enable long-term monitoring, two spiked QC levels (15.0 pg/mL and 3,500 pg/mL [55.1 pmol/L and 12,850 pmol/L]), three levels of a certified secondary reference material (BCR-576 31.1 pg/mL [114 pmol/L], BCR-577 188 pg/mL [690 pmol/L], and BCR-578 365 pg/mL [1,340 pmol/L]), and one native patient sample (∼112 pg/mL [411 pmol/L]) were used (See Supplementary Material 1, Section 3.2.4). As there were no certified reference materials available for the concentration range of the HS method the certified material BCR-576 was diluted with steroid-free serum to a final concentration of 3.11 pg/mL (11.4 pmol/L) and used as a control sample. In addition, two spiked QC levels (steroid-free serum) with concentrations of 1.00 pg/mL (3.67 pmol/L) and 4.00 pg/mL (14.7 pmol/L) were prepared (See Supplementary Material 1, Section 4.2.4).

Sample preparation

Sample preparation was optimized and validated for serum matrix and plasma (lithium-heparin, K₂-EDTA, K₃-EDTA) matrices. To prepare the samples using the SR method a total volume of 450 µL of sample was pipetted into a screw cap vial. Then, 50 µL of the ISTD spiking solution (¹³C₃-E2, 1.00 ng/mL [3,671 pmol/L] in 50 % methanol) was added, and the sample incubated for 30 min at room temperature. The sample was then diluted with deionized water and subjected to a supported liquid extraction (Novum SLE 6cc, Phenomenex) using ethyl acetate. The obtained extract was evaporated to dryness in a N₂-evaporator, and the residue was reconstituted in 100 µL 100 mM Na₂CO₃–NaHCO₃ buffer. After a final filtration step, the purified sample is ready for measurement.

To achieve the required sensitivity, it was necessary to optimize the sample preparation protocol using derivatization for the HS method. The process involved pipetting 900 µL of the sample into a screw cap vial and adding 100 µL of the ISTD spiking solution (D_5-E2, 25.0 pg/mL [91.8 pmol/L] in 50 % methanol) followed by an incubation step. For sample clean-up, a liquid-liquid-extraction (LLE) was performed using a mixture of heptane and ethyl acetate (9 + 1, v + v), and the resulting organic layer was evaporated to dryness in a N₂-evaporator. The residue was then reconstituted in 100 µL 100 mM Na₂CO₃–NaHCO₃ buffer. For derivatization, 100 µL of dansyl chloride in acetonitrile (1 mg/mL) was added to the sample, which was then vortexed and incubated at 65 °C for 3 min. After this step, the sample is ready for measurement and can be stored at 8 °C until analysis.

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

To adjust for the differences in sample preparation and the requirement for a derivatization step in the HS method, two distinct LC-MS/MS methods were optimized. Both, the SR method and the HS method required a heart-cut two-dimensional (2D) LC-MS/MS analysis. This analysis was performed on an Agilent 1290 Infinity LC system coupled to an AB Sciex QTrap 6500 triple quad mass spectrometer. For the heart-cut approach, the analyte transfer between columns was accomplished by connecting them using a 100 µL sample loop. For specific details regarding the 2D LC-MS/MS analysis, including the gradient, source, and multi-reaction monitoring (MRM) parameters, please refer to Supplementary Material 1.

In the case of the SR method, the first dimension employed an Acquity BEH C18 column from Waters (2.1 × 50 mm, 1.7 µm, Milford, MA, USA), while the second dimension utilized a Raptor Biphenyl column from Restek (2.1 × 100 mm, 2.7 µm, Centre County, PA, USA). In the first dimension, chromatographic separation was carried out at a temperature of 50 °C, with a flow rate of 0.300 mL/min, while in the second dimension, the flow rate was set at 0.500 mL/min. Water (A1) and acetonitrile (B1) were used as mobile phases in the first dimension, and 0.2 mM NH4F in water (A2) and methanol (B2) in the second dimension. The retention time of E2 and ¹³C₃-E2 was 7.6 min. Electrospray ionization (ESI) was performed in negative mode, and MRM was employed using the following transitions: E2 m/z 271.2>145.0 (quantitation ion (QI)) and m/z 271.2>182.9 (confirmation ion (CI)), ¹³C₃-E2 m/z 274.2>147.9 (ISTD QI) and m/z 274.2>186.0 (ISTD CI). The total run time per sample was 9 min.

Correction note: Correction added November 6, 2025 after online publication May 26, 2025: The numbers 4 and 7 in 147.9 were mistakenly swapped, resulting in an incorrect value. The last sentence should correctly read:

Electrospray ionization (ESI) was performed in negative mode, and MRM was employed using the following transitions:E2 m/z 271.2>145.0 (quantitation ion (QI)) and m/z 271.2>182.9 (confirmation ion (CI)), ¹³C₃-E2 m/z 274.2>147.9 (ISTD QI) and m/z 274.2>186.0 (ISTD CI).

In the HS method, the first dimension utilized an Acquity BEH C8 column from Waters (2.1 × 50 mm, 1.7 µm), while the second dimension employed a Raptor Biphenyl column from Restek (2.1 × 100 mm, 2.7 µm). Chromatographic separation was performed at a temperature of 50 °C in the first dimension, with a flow rate of 0.300 mL/min. In the second dimension, the flow rate was set at 0.500 mL/min. In the first dimension, the mobile phases used were water (A1) and acetonitrile (B1) both containing 0.1 % formic acid. In the second dimension, the mobile phases consisted of water (A2) and a mixture of methanol and acetonitrile (7 + 3, v/v) (B2) containing 0.1 % formic acid, respectively. The retention time of dansyl-E2 and dansyl-E2-d₅ was 5.8 min. ESI was performed in positive mode. MRM was performed using the following transitions: dansyl-E2 m/z 506.1>171.1 (QI) and m/z 506.1>156.1 (CI), D₅-dansyl-E2 m/z 511.1>171.1 (ISTD QI) and m/z 511.1>156.1 (CI). The total run time per sample was 9 min.

Structure of analytical series, calibration and data analysis

Prior to the start of any analytical series, a system suitability test was performed, which had to be passed. This included checking the instrument sensitivity, the chromatographic retention time, and potential analyte carry-over after injection of high concentrated samples. Further details can be found in the Supplementary Material 1.

For both the SR and HS method, an individual set of calibrator and quality control samples (QCs) was prepared for each measurement sequence. In each sequence, the calibrators and QCs were injected twice – once at the beginning and once at the end of a sequence. The peak area ratio (AR) of the analyte QI to ISTD QI was used for quantitation. The calibration functions used both injections of all calibration levels without forcing the functions through the origin. For the SR method a quadratic model was used in the form AR=a*cA² + b*cA + c (cA=analyte concentration) and 1/cA weighting, while for the HS method, a linear model was used in the form AR=a*cA + b 1/cA weighting. Further information regarding the software utilized and the specific parameters for peak integration can be found the Supplementary Material 1.

Method validation

The method was validated as previously described in detail by Taibon et al. [41] and according to the Clinical & Laboratory Standard Institute Guidelines C62A: “Liquid Chromatography-Mass Spectrometry Methods” [42], the International Conference on Harmonization guidance document “Harmonized Tripartite Guideline Validation of Analytical Procedures: Text and Methodology Q2 (R1)” [43] and the GUM [44].

Selectivity/specificity

To identify potential interfering steroids (17α-estradiol, estrone, 1,3,5(10)-estratrien-16β, 17β-epoxy-3-ol, 5(10)-estren-3,17-dione, 4,16-androstadien-3β-ol, 5,16-androstadien-3β-ol, 16,(5α)-androsten-3-one, 2,(5α)-androsten-17-one, 3-deoxydehydroepiandrosterone, and 5a-dihydrotestosterone) in the first dimension of the SR method, individual standards at concentrations of 1,000 ng/mL were separately injected. Measurements were conducted using both ESI(+) and ESI(−) modes on an UPLC-QToF instrument (Synapt G2-Si with an Acquity UPLC, Waters Inc.). Column materials, HPLC gradient, and eluents were used as specified for the first dimension of the LC method parameters. As a single modification, ESI-additives have been added to the fluidics after HPLC separation and directly before the ESI-Probe, using a syringe pump (5 μL/mL continuous injection in “combined” fluidics mode with the HPLC-flow of 0.3 mL/min; Additive in ESI(+): 6.1 % Formic acid in deionized H₂O (v/v); Additive in ESI(−): 6.3 mM Ammonium fluoride in deionized H₂O). This ensured optimal electrospray ionization conditions in both positive and negative ion modes while maintaining the eluents from the first dimension of separation, which are free of ESI additives. ESI-MS settings and data acquisition were performed with the following parameters: capillary voltage at 3.0 kV, source temperature at 120 °C, desolvation temperature at 500 °C, mass range from m/z 10 to m/z 1,000, and full-scan centroid mode. Data analysis was conducted using MassLynx V4.2 SCN983 (Waters Inc.). The chromatographic separation of E2 and potential interferents was then assessed to ensure that any interferences were baseline-separated. Interferents that were not separated in the first dimension were analysed using the 2D-LC setup, and their chromatographic separation was evaluated.

Dansyl chloride was used as a derivatizing reagent for the high-sensitivity method. This reagent is commonly applied in fluorescence detection and mass spectrometry (MS) analysis of phenols and amines. Given the chemical structure of the potential interferents, 17α-estradiol, estrone, and 1,3,5(10)-estratrien-16β, 17β-epoxy-3-ol were selected for interference testing in the high-sensitivity method. Individual standards at concentrations of 225 pg/mL were injected separately, and measurements were performed using the 2D-LC setup. Chromatographic separation was assessed in both the first and second dimensions.

Steroid-free human serum spiked with the isotope labeled ISTD (with the specified concentration used for the SR and HS protocol) and was analyzed to determine the amount of residual E2 in the ISTD. The amount of unlabeled analyte in the ISTD must not exceed 20 % of the amount of the lower limit of the measuring interval (LLMI), corresponding to the concentration level of the lowest calibrator.

In addition, both the steroid-free human serum and the native serum pool were evaluated to identify potential interferences that may occur at the expected analyte retention time.

Matrix effects

For qualitative evaluation of possible signal enhancement or suppression effects, a post-column infusion experiment was performed. To achieve a consistent MS signal, a 50 % methanol (aq) (v + v) solution containing E2 and ¹³C₃-E2 (c=150 ng/mL [550,701 pmol/L]) for the SR and D₅-E2 (c=0.250 ng/mL [918 pmol/L]) for the HS method was infused into the post-column-flow (flow rate 10 μL/min). The SR was evaluated using different samples, such as a neat solution (50 % methanol v + v), the surrogate matrix, the steroid-free human serum, the native human serum pool, K₂-EDTA plasma, K₃-EDTA plasma, and lithium-heparin plasma. On the other hand, the HS method was tested using the neat solution, surrogate matrix, steroid-free human serum, and a native matrix pool diluted with steroid-free human serum to decrease the endogenous E2 concentration.

In addition, to quantitatively assess potential matrix effects, calibrator levels were spiked in the same matrices except the K_2/3EDTA plasma matrices for the SR method and native matrix pool for the HS method. Calibrator levels were prepared once and injected twice. The slopes (x-term) of the standard lines were compared and confidence intervals (CIs) determined. Matrix effects were quantitatively assessed according to the formula: ME%=B/A × 100 (B=area ratio in matrix, A=area ratio in neat solution) based on Botelho et al. [37].

Precision, trueness and accuracy

To determine the precision of the method, a five-day validation experiment was performed, as previously described by Taibon et al. [41]. For data analysis, an internal statistical program (Biowarp), based on the VCA Roche Open Source software package in R, was utilized. An ANOVA-based variance component analysis (VCA) performed to estimate the total variability of the method (type A uncertainty) [45].

The experiment involved the analysis of seven samples, which covered the entire calibration range of the SR method. These samples consisted of three levels of certified secondary reference materials (BCR-576 [31.1 pg/mL [144 pmol/L]], BCR-577 [188 pg/mL [690 pmol/L]], and BCR-578 [365 pg/mL [1,340 pmol/L]]). Additionally, three levels (5.00 pg/mL [18.4 pmol/L], 15.0 pg/mL [55.1 pmol/L], and 3,500 pg/mL [12,850 pmol/L]) were spiked in steroid-free serum and one level of a native patient serum sample (approx. 30 pg/mL [110 pmol/L]) was included. For the HS method, three levels were used, with two samples spiked into steroid-free serum at concentrations of 1.00 pg/mL (3.67 pmol/L) and 5.00 pg/mL (18.4 pmol/L), and one sample prepared by diluting secondary reference material (BCR-576) with steroid-free serum to achieve a final concentration of 3.11 pg/mL (11.4 pmol/L). All samples were prepared six-fold at each day (three-fold in two separate measurement sequences named as part A and part B) and injected twice, resulting in n=12 measurements per day and n=60 measurements over five days. For each part A and B an independent calibration curve was generated preparing individual calibrator levels.

Trueness and accuracy for both methods was assed using the same samples as for the precision experiment with the exception of the native sample. Additionally, the accuracy of the SR method was evaluated by fortifying the lithium heparin plasma matrix with E2 concentrations of 15.0 pg/mL (55.1 pmol/L) and 3,500 pg/mL (12,850 pmol/L). To assess the dilution integrity of both methods, a highly concentrated sample at a concentration of 6,000 pg/mL (22,028 pmol/L) for the SR method and 5.50 pg/mL (20.2 pmol/L) for the HS method was used. Therefore, samples were diluted 1 + 1 with steroid-free human serum. A total of six sample preparations were performed, with each sample prepared in triplicate for both part A and part B on a single day. Accuracy was evaluated as closeness of agreement between the test result and the accepted reference value, whereas trueness was evaluated as closeness of agreement between the average value obtained from a series of test results and an accepted reference value.

Linearity

To evaluate linearity of both methods (0.400 pg/mL [1.47 pmol/L] to 5.00 pg/mL [18.4 pmol/L] HS and 5.00 pg/mL [18.4 pmol/L] to 5,000 pg/mL [18,357 pmol/L] SR), expanded (±20 %) calibrator sets in steroid-free serum were prepared in triplicate including individual weighings. Thus, the concentrations ranged from 4.00 pg/mL (14.7 pmol/L) to 6,000 pg/mL (22,028 pmol/L) for the SR range method and from 0.320 pg/mL (1.17 pmol/L) to 6.00 pg/mL (22.0 pmol/L) for the HS method. All samples of each calibrator set were prepared once and injected twice. Calibration curves with linear and quadratic fit (1/cA weighting) were created and the calibrations equation parameters were evaluated.

Additionally, the linearity of the method was validated by examining the recovery of serially diluted samples (n=1) using the preferred regression model for calculation. Measurement results have to show a linear dependency. Therefore, for the SR method 11 samples were prepared by serial dilution of a high concentrated sample (approximately 5,000 pg/mL [18,357 pmol/L]) with steroid free-serum following specific ratios (10 + 0, 9 + 1, 8 + 2, 7 + 3, etc.). To show linearity across both measuring ranges, a sample with an approximate concentration of 16.0 pg/mL (58.7 pmol/L) was prepared the same way, however samples >5.00 pg/mL (>18.4 pmol/L) were measured using the SR method and samples ≤5.0 pg/mL (≤18.4 pmol/L) were measured using the HS method.

Lower limit of the measuring interval (LLMI)

The LLMI was set to the concentration of the lowest calibrator level. Precision and accuracy at concentration of 5.00 pg/mL (18.4 pmol/L) for the SR method, as well at 0.400 pg/mL (1.47 pmol/L) for the HS method, were determined by six-fold preparation.

Uncertainty of measurements

Standard measurement uncertainty was evaluated following the GUM guidelines [44]. The overall standard uncertainty in a single measurement was determined by considering two main factors: the standard uncertainty in calibrator preparation (unc_cal) and the calculated standard uncertainty in the precision experiment (unc_prec). To ensure accurate estimation of the error, the calibrator level with the highest standard uncertainty (unc_cal) was selected for each sample concentration level when combining uncertainties. Reference or target values were established by averaging multiple sample preparations conducted on at least two different days. The standard measurement uncertainty was obtained by combining the standard uncertainty in calibrator preparation (unc_cal) with the standard uncertainty (standard deviation) of the mean of the measurement results (unc_mean). The resulting expanded measurement uncertainty was then multiplied by a coverage factor of k=2 for a confidence level of 95 %, assuming a normal distribution. Detailed information can be found in Supplementary Material 3.

Equivalence to JCTLM listed reference measurement procedures (RMPs)

To establish equivalence with the JCTLM-listed Reference Measurement Procedures (RMPs), the CDC Hormones Standardization Program’s Phase 1 Panel was utilized. This panel, provided by the Centers for Disease Control and Prevention (CDC), includes 40 non-pooled human serum samples with assigned target values determined using either the Ghent University’s ID-GC-MS reference method [38] or the CDC ID-LC-MS/MS reference method [37]. Since these samples only encompass the concentration range of the SR method, an additional set of 10 single donor human serum samples were measured using the HS method. These samples had assigned reference values provided by the CDC and covered concentrations ranging from 1.10 to 4.70 pg/mL (4.04 to 17.3 pmol/L). The purpose of this measurement was to assess the equivalence of the HS method to the JCTLM RMPs for concentrations below 5.00 pg/mL (18.4 pmol/L). To prove the robustness of the candidate RMP and enable ongoing method comparison and bias assessment in accordance with the CLSI EP09C (formerly CLSI EP09-A2) guideline [39], 40], regular participation in the Phase 2 CDC HoST program was carried out from Q2 2023 to Q1 2024.

Equivalence of results between independent laboratories

To assess the agreement of the RMP between two independent laboratories (Laboratory 1: Roche Diagnostics GmbH, Penzberg, Germany; Laboratory 2: Department of Pediatrics and Adolescent Medicine, University Hospital Erlangen, Erlangen, Germany), a method comparison study was performed on n=94 native residual, anonymized patient samples. In addition, a three-day precision experiment was performed at Laboratory 2. Spiked samples were provided by Laboratory 1. Both laboratories prepared their own calibrator levels using the primary reference material.

For SR measurements adaptations to the sample preparation protocol were made in the second laboratory in order to use the 96-well plate format. All steps before the sample extraction were done in one single 96-well deep well plate from polypropylene. In brief, 400 µL sample and 40 µL ISTD were incubated before adding 360 µL water. The sample was mixed and equally distributed onto two 96-well plates (Phenomenex Novum SLE) for extraction. The eluate of both parts was collected in one set of glass inserts (Hirschmann, Eberstadt, Germany). After evaporation, the samples were reconstituted, mixed, sealed and measured.

A second method comparison study (n=51) was carried out by Laboratory 1 and Laboratory 3 (Institute of Laboratory Medicine, Clinical Chemistry, and Molecular Diagnostics, University of Leipzig Medical Center, Germany) to compare a high-throughput, in-house developed LC-MS/MS routine assay to the candidate RMP performed at Laboratory 1. The routine assay utilized commercially available calibration material and employed online solid phase extraction for sample preparation [46].

Traceability to SI units

The availability and use of the JTCLM listed primary reference material NMIJ-6004a from NMIJ enables the metrological traceability to the SI unit of mass (kilogram), the most important parameter for an RMP.

Selectivity

To assess the chromatographic separation of E2 and known interferences in the SR method (Figure 1) and HS method (Figure 2), neat solutions were prepared and injected individually. All interferences were effectively separated from E2 in the first chromatographic dimension and were not transferred to the second dimension. The compound 1,3,5(10)-Estratriene-16β,17β-epoxy-3-ol contained an unidentified impurity probably due to insufficient purification. This impurity was not fully separated from E2 in the first dimension of the HS method, leading to its transfer to the second dimension, where it was successfully separated. However, it has to be considered that the concentration used for the interference testing was about 45 times higher than the highest calibrator level concentration.

Figure 1:

Chromatographic separation of E2 and potential interferents in the first dimension of the SR method.

Figure 2:

Chromatographic separation of E2 and potential interferents in the first dimension of the HS method.

For ISTD spiked steroid-free serum samples, in both methods, no peaks were observed in the analyte SRM traces. This indicates the absence of residual E2 in the used ISTDs. Additionally, the steroid-free serum and native serum samples did not show any potential interferences at the expected analyte retention times.

Matrix effects specificity

A post-column infusion experiment was performed to evaluate ion suppression or enhancement effects across various matrices for both methods. Neither method showed significant ion suppression or enhancement at the analyte and ISTD retention times. Quantitative evaluation of matrix effects was conducted by comparison of the standard line slopes of different matrices. Due to a pipetting error, calibrator level four had to be excluded for the steroid-free human serum matrix in the SR method. As a result, the following slopes (x-term) and CIs were obtained: 0.0087 (95 % CI 0.0086 to 0.0089) for the neat solution, 0.0086 (95 % CI 0.0085 to 0.0087) for the surrogate matrix, 0.0087 (95 % CI 0.0085 to 0.0089) for the steroid-free human serum, 0.0087 (95 % CI 0.0085 to 0.0090) for the native human serum pool, and 0.0086 (95 % CI 0.0084 to 0.0088) for lithium-heparin plasma. For the HS method slopes and CIs were found to be 0.37 (95 % CI 0.35 to 0.38) for the neat solution, 0.40 (95 % CI 0.36 to 0.44) for the surrogate matrix, 0.40 (95 % CI 0.39 to 0.41) for the steroid-free human serum. Based on the comparable slopes and overlapping estimated CIs for both methods, it can be concluded that there is no matrix effect present.

The evaluation of sample matrix effects, showed no effect for the SR method and HS method with matrix effects ranging from 96 to 102 % with the exception of the steroid-free human serum for HS method indicated a slight ion enhancement resulting in a mean ME% of 107 %. However, the comparison of standard line slopes along with our evaluations on the accuracy and precision of the HS method showed no significant influence from this effect, thereby proving the method to be matrix independent.

Precision, trueness and accuracy

To evaluate the overall variability of the RMP, a validation experiment was conducted over multiple days. The intermediate precision of the SR method ranged from 1.3 to 3.1 %, and the repeatability consistently remained ≤2.9 % across different concentration levels, except for the lowest level of 5.00 pg/mL (18.4 pmol/L), where the intermediate precision and repeatability were found to be 8.7 % and 7.8 %, respectively. Due to technical issues, two measurements for BCR-576 and one for BCR-577 were excluded (Table 1). When employing the HS method for levels ≤5.00 pg/mL (≤18.4 pmol/L), the ANOVA analysis revealed an intermediate precision ≤5.9 % and a repeatability ≤5.6 % (Table 2).

The accuracy and trueness were assessed by preparing six samples at each concentration level in a single day (Tables 3 and 4). The relative mean bias for the SR method ranged from −2.4 to 1.9 % for all levels including secondary reference material as well as self-spiked steroid-free human serum samples. The two evaluated levels for the lithium-heparin plasma matrix were found with a relative mean bias of −0.8 % and −0.4 %, respectively. In addition one sample >ULMI (approximately 6,000 pg/mL [22,028 pmol/L]) was diluted 1 + 1 (v + v) with steroid-free human serum and prepared six-fold resulting in a relative mean bias of 0.0 %. For the HS method the mean bias of all three levels ranged from −3.0 to 2.9 %. A sample with a concentration >ULMI (approximately 5.50 pg/mL [20.2 pmol/L]) was diluted 1 + 1 (v + v) with steroid-free serum and prepared six-fold resulting in a relative mean bias of 2.4 %.

Overall these results are in a very good agreement with the RMP performance specifications defined based on the biological variation as well as the suggested performance criteria for total estradiol based on Vesper et al. [24].

Linearity

Linearity was evaluated over a concentration range of 0.320–6,000 pg/mL (1.17–22,028 pmol/L), which was divided into the two separate measurement ranges (SR and HS method). Both methods were assessed using linear regression, employing a linear model and a quadratic model.

All prepared calibrator sets demonstrated correlations coefficients of ≥0.999 for the SR method and ≥0.996 for the HS method, indicating a good fit for both models. Considering the smaller residuals, the linear regression model with a quadratic fit (AR=a*cA² + b*cA + c) and a 1/cA weighting was selected for calibration of the SR method. For the HS method the residuals of the linear and quadratic model were found to be similar, wherefore the linear model with a 1/cA weighting was chosen for calibration.

In addition, the linearity of both measuring ranges was proven through the measurement of serially diluted samples, resulting in correlation coefficients of ≥0.999. The average recovery for the SR method was determined to be 98 %, while the average recovery for samples bridging the SR and HS method was found to be 94 %.

Lower limit of the measuring interval (LLMI)

The LLMI was determined for both LC-MS/MS methods by six-fold preparation of the lowest calibration level (Figures 3 and 4). For the SR method, a relative deviation of −5.4 % and a CV of 2.8 % was determined at 4.94 pg/mL (18.1 pmol/L). For the HS method, the LLMI was determined at a concentration of 0.400 pg/mL (1.47 pmol/L), with a relative deviation of 1.5 % and a CV of 9.9 %. Based on the criteria established performance criteria [24], 37], the RMP allows for an imprecision of 5.7 % and a bias of ±0.8 pg/mL (±2.94 pmol/L) for samples with concentrations ≤20 pg/mL (≤73.4 pmol/L). As a result, these findings fall within the anticipated range of values.

Figure 3:

Estradiol LC-MS/MS chromatographic readouts. For SR method: (A) calibrator level 1 with a concentration of 5.00 pg/mL spiked in steroid-free serum matrix, analyte (left) and ISTD (right). (B) Secondary reference material BCR-578 (lyophilized native human serum) with a concentration of 365 pg/mL, analyte (left) and ISTD (right). (C) Fortified Li-heparin patient sample with a concentration of 59.8 pg/mL, analyte (left) and ISTD (right). SR, standard range method; HS, high sensitivity method; ISTD, internal standard.

Figure 4:

Estradiol LC-MS/MS chromatographic readouts. For HS method: (A) calibrator level 1 with a concentration of 0.400 pg/mL spiked in steroid-free serum matrix, analyte (left) and ISTD (right). (B) Quality control sample with a concentration of 4.02 pg/mL in steroid-free serum analyte (left) and ISTD (right). (C) Native serum sample with a concentration of 1.02 pg/mL, analyte (left) and ISTD (right). SR, standard range method; HS, high sensitivity method; ISTD, internal standard.

Uncertainty of measurements

The overall standard uncertainty for a single measurement for the SR method varied between 1.5 % and 3.2 %, with the exception of the lowest level, where it was found to be 8.8 %, as presented in Table 5. By multiplying the overall standard uncertainty by a coverage factor of k=2 (corresponding to a confidence level of 95 % assuming a normal distribution), the corresponding expanded uncertainty ranged from 3.0 to 6.3 %, and was determined to be 17.5 % for the lowest level. These uncertainties can be further reduced by performing multiple measurements. Therefore, for each level multiple sample preparations and measurements were performed on two different days, and the arithmetic mean (n=6) was calculated. As shown in Table 6, the overall standard uncertainties were reduced to 0.8–2.2 %, resulting in expanded uncertainties ranging from 1.7 to 4.4 %.

For the HS method standard uncertainty for a single measurement ranged from 5.9 to 6.2 %, as shown in Table 7. By multiplying the standard uncertainty by a coverage factor of k=2 the expanded uncertainty ranged from 11.7 to 12.4 %. To assign target values, samples were prepared on a minimum of two different days, and the arithmetic mean (n=6) was determined. The standard uncertainties were found to be ≤3.1 % with corresponding expanded uncertainties ≤6.1 % (Table 8).

Equivalence to JCTLM listed reference measurement procedures (RMPs)

To show equivalence between the established candidate RMP and two of the JCTLM listed RMPs from the University of Ghent [38] and CDC [37] the CDC HoSt Phase 1 panel, which consisted of 40 individual human serum samples, was analyzed with n=1 preparations. For samples with concentrations ≤5.00 pg/mL (≤18.4 pmol/L), the HS method was used (n=4 samples), while samples with concentrations >5.00 pg/mL (>18.4 pmol/L) were measured using the SR method (n=36 samples). Passing-Bablok regression showed an excellent correlation with a slope of 0.99 (95 % CI 0.97 to 1.01) and an intercept of 0.12 (95 % CI −0.51 to 0.59) (Figure 5A). The Bland-Altman plot showed a mean bias of −0.8 % (95 % CI −2.9 to 1.3) and the 2S interval of the relative difference was 12.8 % (Figure 5B). This implies a strong agreement between the measured concentrations using the candidate RMP and the reference values assigned with JCTLM listed RMPs [37], 38]. In addition, as the phase 1 panel covers mostly the SR method, 10 individual human serum samples (1.10–4.70 pg/mL [4.04–17.3 pmol/L]) were measured using the HS method. These measurements showed a maximum absolute bias ranging from −0.17 to 0.22 pg/mL compared to the reference values (Table 9).

Figure 5:

Equivalence testing results between the cRMP and JCTLM-listed methods [37], 38] were analyzed using the CDC HoSt Phase 1 reference panel, which included 40 individual human serum samples ranging from 1.15 to 268 pg/mL (n=1 preparation). The Passing-Bablok regression (A) showed an excellent correlation, with a slope of 0.99 (95 % CI 0.97 to 1.01) and an intercept of 0.12 pg/mL (95 % CI −0.51 to 0.59 pg/mL). The Bland-Altman plot (B) revealed a mean bias of −0.8 % (95 % CI −2.9 to 1.3) and a 2S interval of the relative difference at 12.8 %. CI, confidence interval; RMP, reference measurement procedure; SD, standard deviation.

In order to demonstrate the robustness of the RMP candidate method, participation in the CDC HoSt Phase 2 program was carried out on a regular basis. Data evaluation was done based on the official report provided by CDC. Findings on measurement bias and imprecision were compared against the suggested performance criteria for E2 of ±12.5 % bias for samples >20 pg/mL (>73.4 pmol/L), ±2.5 pg/mL bias for samples ≤20 pg/mL (≤73.4 pmol/L), and ≤11.4 % for imprecision at all concentrations [24]. Linear regression analysis resulted in a slope of 1.01 (95 % CI 0.99 to 1.03) and an intercept of −1.82 (95 % CI −3.48 to −0.15 pg/mL) (Figure 6). Bias evaluation showed a relative mean bias of −3.4 % (95 % CI −5.4 to −1.4 %) for samples >20 pg/mL (>73.4 pmol/L) and an absolute mean bias of −1.25 (95 % CI −2.92 to 0.41) for samples <20 pg/mL (<73.4 pmol/L). The results demonstrate that the method is accurate and robust, and the results remain consistent throughout the year, between lots, and over the reported measurement range.

Figure 6:

To demonstrate the robustness of the RMP candidate RMP, regular participation in the CDC HoST Phase 2 program was undertaken. Linear regression analysis resulted in a slope of 1.01 (95 % CI 0.99 to 1.03) and an intercept of −1.82 pg/mL (95 % CI −3.48 to −0.15 pg/mL). Bias evaluation showed a relative mean bias of −3.4 % (95 % CI −5.4 to −1.4 %) for samples >20 pg/mL and an absolute mean bias of −1.25 pg/mL (95 % CI −2.92 to 0.41 pg/mL). CI, confidence interval; RMP, reference measurement procedure; SD, standard deviation.

Equivalence of results between independent laboratories

The analysis of anonymized native patient samples (n=94) was conducted at Laboratories 1 and 2. Out of the total samples n=1 sample was identified as outlier based on the LORELIA outlier test and was not included in the analysis [47]. Passing-Bablok analysis revealed a good agreement between the two laboratories, resulting in a regression equation with a slope of 0.97 (95 % CI 0.97 to 0.98) and an intercept of −0.01 (95 % CI −0.13 to 0.21) (Figure 7A). Additionally, Pearson’s correlation coefficient was ≥0.999. Relative Bland-Altman analysis was conducted for samples greater than 20 pg/mL (73.4 pmol/L) (n=54), while absolute Bland-Altman analysis was performed for samples with a concentration of 20 pg/mL (73.4 pmol/L) or less (n=39). The relative Bland-Altman analysis revealed a mean bias of −1.2 % (95 % CI: −3.5 to 1.1 %), with the ±1.96*SD range of relative differences spanning from −17.4 to 15.0 % (lower limit 95 % CI: −21.3 to −13.6 %, upper limit 95 % CI: 11.2–18.9 %) (Figure 7B). The absolute Bland-Altman analysis indicated a mean bias of −0.4 pg/mL (95 % CI: −0.7 to −0.1 pg/mL), with the ±1.96*SD range of differences between −2.2 and 1.5 pg/mL (lower limit 95 % CI: −2.7 to −1.7 pg/mL, upper limit 95 % CI: 0.9–2.0 pg/mL) (Figure 7C).

Figure 7:

Method comparison study results between two independent laboratories. (A) Passing-Bablok regression plot with Pearson regression analysis for the RMP method comparison (n=93) between Laboratory 1 (Roche Diagnostics GmbH, Penzberg) and Laboratory 2 (Department of Pediatrics and Adolescent Medicine, University Hospital Erlangen). The regression equation had a slope of 0.97 (95 % CI 0.97 to 0.98) and an intercept of −0.01 (95 % CI −0.13 to 0.21), with a Pearson correlation value of ≥0.999. (B) Relative Bland-Altman plot for the RMP method comparison (n=54) between the two laboratories. The interlaboratory measurement bias was −1.2 % (95 % CI −3.5 to 1.1), with a ±1.96*SD range of relative differences from −17.4 to 15.0 % (lower limit 95 % CI −21.3 to −13.6, upper limit 95 % CI 11.2 to 18.9). (C) Absolute Bland-Altman analysis (n=39) showed a mean bias of −0.4 pg/mL (95 % CI −0.7 to −0.1), with a ±1.96*SD range of differences between −2.2 and 1.5 pg/mL (lower limit 95 % CI −2.7 to −1.7, upper limit 95 % CI 0.9 to 2.0). CI, confidence interval; RMP, reference measurement procedure; SD, standard deviation.

At Laboratory 2, the three-day precision experiment shows SR method intermediate precision ranging from 1.9 to 4.9 % and repeatability ranging from 1.3 to 3.9 % across all levels. For the HS method (Levels ≤5.00 pg/mL [≤18.4 pmol/L]), intermediate precision was ≤5.8 % and repeatability ≤5.0 %. All CVs were comparable to those at Laboratory 1.

In order to establish comparability with an in-house developed high-throughput LC-MS/MS based assay [46], an additional method comparison study was conducted (Figure 8). This assay used a commercially available kit for calibration. A total of n=51 native patient samples were analysed, with two samples exceeding the upper limit of the measuring interval in the in-house developed assay and one sample confirmed as outlier performing the LORELIA outlier test [47]. The Passing-Bablok regression analysis resulted in a regression equation with a slope of 1.11 (95 % CI 1.07 to 1.15) and an intercept of −3.4 (95 % CI −8.1 to −0.9) with r=0.998. Furthermore, the Bland-Altman analysis revealed a mean bias of 7.2 % (95 % CI 2.7–11.6 %) and a corresponding 2S interval ranging from −23.0 to 37.4 %.

Figure 8:

Results from the method comparison study performed between the candidate RMP (Site 1, Roche Diagnostics GmbH, Penzberg) and an in-house developed high-throughput LC-MS/MS based assay (Site 3, Institute of Laboratory Medicine, Clinical Chemistry, and Molecular Diagnostics, University of Leipzig Medical Center). (A) Passing-Bablok regression analysis resulted in a regression equation with a slope of 1.11 (95 % CI 1.07 to 1.15) and an intercept of −3.4 (95 % CI −8.1 to −0.9) with r=0.998. (B) Bland-Altman analysis revealed a mean bias of 7.2 % (95 % CI 2.7 to 11.6) and a corresponding 2S interval ranging from −23.0 to 37.4 %. CI, confidence interval; RMP, reference measurement procedure; SD, standard deviation.