International collaborative study to evaluate and calibrate two recombinant L chain Ferritin preparations for use as a WHO International Standard

Materials for the collaborative study

The codons for human FLC were separately optimized for bacterial and mammalian expression, cloned into pET-21a for E. coli BL21 (DE3) expression or pcDNA™3.4 for ExpiCHO-S^TM expression (Supplementary Figure 2) and the sequences verified by Sanger sequencing. FLC was expressed in each cell type following the manufacturer’s instructions {pET-21a(+) DNA - Novagen (merckmillipore.com), MAN0014337_expicho_expression_system.pdf (thermofisher.com)} and the cell pellets harvested by centrifugation. E. coli cell pellets were lysed, on ice, with 1% Lysozyme (Merck, Dorset, UK) and 0.1% Benzonase nuclease (Merck, Dorset, UK) in PBS, sonicated (Soniprep 150, Henderson Biomedical, London, UK) and the supernatant containing the crude FLC extract isolated by centrifugation. ExpiCHO-S^TM cell pellets were lysed in ice cold deionized water containing 0.1% Benzonase nuclease (Merck, Dorset, UK) and the supernatant containing the crude FLC extract again isolated by centrifugation. Both crude FLC extracts were further purified as described by Levi et al. [11, 12], dialyzed into PBS, aliquoted (1 mL at 6 mg/mL), stored at −80 °C and thawed in a water bath at +37 °C prior to lyophilization or further analysis. The purified FLC amino acid sequence from both expression systems was confirmed by liquid chromatography with tandem mass spectrometry (LC-MS/MS) using Swiss-Prot for data analysis. The concentrations of the purified FLC preparations were determined by immunoassay using the 3rd IS as the assay calibrant. Three commutability (clinical) serum samples containing low, normal and high levels of Ferritin were purchased from the Wales External Quality Assurance Scheme (WEQAS) and stored at −80 °C prior to lyophilization.

Participants were allocated three ampoules of each of the coded materials shown in Table 1.

Sample A consisted of FLC expressed in E. coli BL21 (DE3) transformed with plasmid pET-21a containing the full FLC sequence. At the National Institute for Biological Standards and Control (NIBSC), purified FLC protein was diluted to ∼10 μg/mL in human serum (TCS Biosciences, Buckingham, UK) containing 40 mM HEPES (Merck, Dorset, UK) and 1% Trehalose (Merck, Dorset, UK), dispensed into ampoules (∼1 mL/ampoule), lyophilized and coded 19/118. The mean weight of the dispensed solution in 220 ampoules was 1.0089 g. The imprecision of the filling (CV) was 0.171%, and the residual moisture was 0.38%. Sample B consisted of FLC expressed in ExpiCHO-S^TM cells transfected with Plasmid pcDNA™ 3.4 containing the full FLC sequence. At the NIBSC purified FLC protein was diluted to ∼7 μg/mL in human serum containing 40 mM HEPES and 1% Trehalose, dispensed into ampoules (∼1 mL/ampoule), lyophilized and coded 19/162. The mean weight of the dispensed solution in 143 ampoules was 1.0094 g. The imprecision of the filling (CV) was 0.160%, and the residual moisture was 0.42%. Samples C, D and E consisted of human serum from patients with high, normal and low levels of Ferritin respectively. At the NIBSC the serum samples were formulated with the addition of 40 mM HEPES and 1% Trehalose, prior to dispensing into ampoules (∼1 mL/ampoule) and lyophilization. The mean weight of the dispensed solutions in 9 ampoules was 1.0285 g. The imprecision of the filling (CV) was 0.130%, and the residual moisture was 0.15%. Note, samples C, D and E were only for distribution in the collaborative study. Samples D and E are intended to be assayed neat.

For all samples the sera were tested and found negative for anti-HIV, anti-HCV, HCV RNA and HBsAg.

Participants

A total of 15 laboratories in 10 countries across the globe agreed to participate in the study, 12 laboratories from nine countries returned results (detailed in Table 2), each of which has been assigned a code number. This coding does not reflect the order of listing.

Ferritin quantitation methods

A WHO International Standard should be suitable for use in as many different assay methods as possible. Therefore, all participating laboratories were requested to perform their usual method of analysis (Table 3). Where a laboratory performed more than one assay method, or an additional variation of the same assay method, each was treated as if performed by different laboratories.

Study design

Each participant was provided with a protocol and three ampoules of each study samples (A–E) detailed above. For each assay method, participants were requested to perform two assays on each of three days resulting in six assay estimates for each study sample relative to their in-house calibrant. Regardless of the assay method used, participants were requested to prepare two independent sets of fresh dilutions using their usual assay diluent to fall within the measurable range of their assay for each preparation in each assay run.

Raw assay data including pre-dilutions and dilutions were requested together with a summary of the participants’ own estimates of potency for the six Ferritin samples.

Statistical analysis

The potencies of coded samples A-E were calculated relative to the 3rd International Standard, 94/572. Individual assay potency estimates were calculated using parallel line analysis with a log transformation of assay response. Calculations were performed using the R software program [13]. Non-linearity and non-parallelism of dose-response relationships were considered in the assessment of assay validity. Linearity was assessed by calculating the coefficient of determination R² and parallelism was assessed by calculation of the ratio of fitted slopes for the test and reference sample. Instances where the R² value was <0.99, or where the ratio of fitted slopes was outside the range 0.90–1.11, were considered invalid and no estimates were reported in such cases.

Results from all valid assays were combined to generate unweighted geometric mean (GM) estimates for each laboratory and these laboratory means were used to calculate overall unweighted geometric means for each sample. Variability between assays (within laboratories) and between laboratories has been expressed using the geometric coefficient of variation (GCV = {10^s − 1} × 100% where s is the standard deviation of the log₁₀ transformed estimates).

Individual assay estimates of relative potencies were log transformed and a variance components analysis was performed in order to determine the intra-lab and inter-lab components of variation. These were used to determine standard uncertainty estimates for samples A (19/118) and B (19/162). The expanded uncertainty was obtained by multiplying the standard uncertainty by the coverage factor k=2.23, taken to correspond to a 95% level of confidence.

Stability

To predict stability on extended storage, ampoules of samples A (19/118) and B (19/162) were stored at elevated temperatures (+4, +20, +37, +45 and + 56 °C) in temperature mapped equipment for longer than one year before analysis. The relative potencies of the accelerated thermal degradation samples were used to fit an Arrhenius equation relating degradation rate to absolute temperature assuming first-order decay [14], and hence predict the degradation rates when stored at a range of temperatures.

Assay data

The 12 participants contributed data from a total of 75 assays, deviations from the study protocol and other anomalies were as follows: (i) Laboratory 5 provided two data sets from separate operators. (ii) Laboratory 9 reported results from one assay on each of three days due to staff shortage.

Assay validity

The majority of assays gave valid potency estimates when assessed using the validity criteria described in Statistical analysis, which were intended for use in the analysis of data from this study only and should not be interpreted as suitable for routine use in the assessment of assay validity within all collaborating laboratories. Around 10% of cases were excluded due to non-linearity or non-parallelism. Where the number of valid assays was less than three (n<3) a GCV was not calculated.

Laboratory reported potencies and potencies relative to the WHO 3rd International Standard for Ferritin (Human Recombinant), 94/572

Potency estimates from individual assays for each laboratory, calculated by the NIBSC and reported by the laboratories are respectively presented in Supplementary Tables 1 and 2. Details of the individual laboratory geometric mean (GM) potencies of samples S, A–E based on laboratory reported estimates are shown in Table 4. Individual laboratory GM potencies of samples A–E relative to sample S, the 3rd IS for Ferritin (Human Recombinant) based on the NIBSC calculated estimates are shown in Table 5. For some manufacturers, the IS used to calibrate an assay test kit varies dependent on the region where the assay test kit is marketed. Seven of the 12 laboratories use the 3rd IS for Ferritin (Human, Recombinant) to calibrate their test kits directly or indirectly, with the remaining five laboratories claiming traceability to either the 1st or 2nd IS for Ferritin (see Table 3 for details). The GM laboratory reported results and GM NIBSC calculated results relative to sample S are presented as histograms in Figure 1A–K. Based on an overall normal serum Ferritin concentration range of 15–200 ng/mL [5], all participating laboratories identified commutability (clinical) samples C, D and E as containing high, normal, and low levels of Ferritin respectively.

Figure 1:

Concentrations based on laboratory-reported estimates and potency estimates relative to sample S (WHO 3rd IS for Ferritin). (A) Concentrations of sample S based on laboratory-reported estimates. (B) Concentrations of sample A based on laboratory-reported estimates. (C) Potency estimates of sample A relative to sample S. (D) Concentrations of sample B based on laboratory-reported estimates. (E) Potency estimates of sample B relative to sample S. (F) Concentrations of sample C based on laboratory-reported estimates. (G) Potency estimates of sample C relative to sample S. (H) Concentrations of sample D based on laboratory-reported estimates. (I) Potency estimates of sample D relative to sample S. (J) Concentrations of sample E based on laboratory-reported estimates. (K) Potency estimates of sample E relative to sample S. Estimates are expressed on a log scale. The number in the square denotes the laboratory code. Each square represents the geometric mean estimate from that laboratory.

Intra-laboratory variability

For laboratory reported results, GCV’s ranged from 0.4 to 7.2% (median 2.8%) for samples S, A and B; 0.5–10.1% (median 2.6%) for commutability samples C and D and 2.0–21.7% (median 6.6%) for commutability sample E (Table 4). Using the NIBSC calculations relative to sample S (Table 5), GCV’s for samples A and B were ≤5% (representing good repeatability); ranged from 1.1 to 13.1% (median 3.9%) for commutability samples C and D and 1.2–42.8% (median 11.3%) for commutability sample E. Sample E is below the limit of quantitation (LoQ) for the dilution series used for samples S, A and B, thereby contributing to the higher intra-laboratory variability for this sample. For three laboratories (08, 09 and 10) no %GCV’s were available for one or more samples due to insufficient valid assays.

Inter-laboratory variability

Inter-laboratory variability is much improved when potencies for samples A and B are determined relative to the 3rd IS for Ferritin (Human Recombinant) as shown in Table 5 and histograms in Figure 1B–E. Inter-laboratory variability is greater for clinical (heterogeneous) samples C–E which is reduced to around 20% for samples C and D and 29% for sample E when excluding laboratories 09 and 10 due to insufficient valid results. Again, sample E is below the LoQ for the dilution series used for samples S, A and B, thereby contributing to the higher intra- and inter-laboratory variability for this sample. Laboratories 09 and 10 have been excluded from the final inter-laboratory GM and GCV calculations because of insufficient valid results. For comparison, Tables 4–6 present the data including and excluding Laboratories 09 and 10.

Table 7 summarizes inter-laboratory variability and shows inter-laboratory variability is independent of the standard used for samples C and D. Sample E has been excluded because it is below the LoQ.

Potencies of samples A and B relative to the 3rd IS

There was better overall agreement between all laboratories and between assay methods for the potency of sample A (19/118), the candidate 4th International Standard, than for sample B (19/162). Summary statistics of concentration estimates for samples A and B relative to sample S are shown in Supplementary Table 3. The overall geometric mean potency for sample A is 10.5 µg/ampoule with expanded uncertainty limits 10.2–10.8 µg/ampoule.

Potencies for samples B–E relative to the candidate 4th International Standard for Ferritin (Human Recombinant), 19/118

Table 6 shows the concentration estimates for samples B–E relative to sample A, based on the potency of 10.5 µg/ampoule. We suggest that sample B can be distributed and used as an internal assay/run control.

Stability

Estimates of the potency for ampoules of samples A (19/118) and B (19/162) stored at elevated temperatures for a period of 2.0 and 1.7 years respectively are summarised in Supplementary Tables 4 and 5. The analysis has shown a predicted loss of potency per year of 0.01% for 19/118 and 0.02% for 19/162 when stored at −20 °C. We can be confident that 19/118 and 19/162 will be stable for decades when stored at −20 °C and are suitable for shipment at ambient temperature.