Development of orthogonal NISTmAb size heterogeneity control methods

The NISTmAb is a monoclonal antibody Reference Material from the National Institute of Standards and Technology; it is a class-representative IgG1κ intended to serve as a pre-competitive platform for harmonization and technology development in the biopharmaceutical industry. The publication series of which this paper is a part describes NIST’s overall control strategy to ensure NISTmAb quality and availability over its lifecycle. In this paper, the development of a control strategy for monitoring NISTmAb size heterogeneity is described. Optimization and qualification of size heterogeneity measurement spanning a broad size range are described, including capillary electrophoresis-sodium dodecyl sulfate (CE-SDS), size exclusion chromatography (SEC), dynamic light scattering (DLS), and flow imaging analysis. This paper is intended to provide relevant details of NIST’s size heterogeneity control strategy to facilitate implementation of the NISTmAb as a test molecule in the end user’s laboratory. Graphical abstract Representative size exclusion chromatogram of the NIST monoclonal antibody (NISTmAb). The NISTmAb is a publicly available research tool intended to facilitate advancement of biopharmaceutical analytics. HMW = high molecular weight (trimer and dimer), LMW = low molecular weight (2 fragment peaks). Peak labeled buffer is void volume of the column from L-histidine background buffer. Electronic supplementary material The online version of this article (10.1007/s00216-017-0819-3) contains supplementary material, which is available to authorized users.

(Eppendorf, PN 89166-278 for 0.5 mL). To this was added 1 µL of 20% (w/v) SDS stock, 2 µL of 10 kDa internal standard, and 10 µL (or as indicated) of 0.5 mol/L iodoacetamide in water, prepared fresh. The sample was mixed by vortexing, then incubated for 5 min at 70 °C in a water bath (or as otherwise indicated). The sample was cooled to room temperature, vortexed, centrifuged briefly, and then transferred to sample vials for CE-SDS analysis.
For analysis under reducing conditions, 100 µg PS 8670 (in 10 µL) was diluted with 85 µL SDS sample buffer, 1 µL 20% (w/v) SDS, 2 µL 10 kDa internal standard protein, and 5 µL 2mercaptoethanol. The sample was vortexed, then incubated for 10 min at 70 °C in a water bath, or as otherwise indicated. The sample was cooled to room temperature, and then vortexed, centrifuged, and transferred to an appropriate sample vial. The instrument qualification standard (IQ) was prepared by diluting 10 µL of MW Marker protein mix with 85 µL SDS sample buffer spiked with 2 µL 10 kDa internal standard. The sample was vortexed, and then incubated for 10 min at 70 °C. The sample was cooled to room temperature, vortexed, centrifuged, and transferred to a sample vial. Blanks were prepared by substituting formulation buffer in the place of PS 8670 in the appropriate volume.

Method Linearity/LOD/LOQ Sample Preparation. A dilution series of RM 8670 under
reducing conditions was prepared in triplicate as follows. A stock solution (2.0 mg/mL) of reduced PS 8670 was prepared by diluting 70 µL 10 mg/mL RM 8670 with 250 µL citratephosphate/SDS sample buffer, 6 µL 10 kDa internal standard, and 16 µL 2-mercaptoethanol. A stock blank solution was prepared by mixing 140 µL L-His buffer, 12 µL 10 kDa internal standard, 32 µL 2-mercaptoethanol, and 500 µL citrate-phosphate/SDS sample buffer. The 2.0 mg/mL stock and the blank were vortexed and then incubated 10 min at 70 °C. After the samples cooled to room temperature, the PS 8670 sample was serially diluted using the blank sample to yield the following dilution series: 1.5, 1.0, 0.5, 0.25, and 0.025 mg/mL. Samples (100 µL each) were analyzed by CE-SDS as described below in "Instrumental Method".
Preparation of Stressed Samples. PS 8670 (1 mg in 100 µL formulation buffer) was dispensed into a thin-walled polypropylene 0.2 mL PCR tube with cap (Fisher Scientific PN 14230205) and placed on a sheet of aluminum foil, reflective side up, in the chamber of a Stratalinker 2400 UV source equipped with 5 UVC bulbs (Eiko, F15T8/BL, 365 nm, 15 W). An identical sample was prepared, wrapped in aluminum foil to exclude light, and placed in the chamber as a thermal degradation control. The samples were irradiated for 21 h.
Retention times and corrected peak areas were recorded for the non-reduced and reduced sample analysis as listed in Table S1.  Quality parameters were then calculated from recorded measurands using the equations below. Monomeric purity of 8670 was calculated according to equation S1: (S1) Glycan occupancy of the heavy chain was calculated using equation S2: Where H = heavy chain and NGH = aglycosylated/non-glycosylated heavy chain. Thioether relative abundance (RA) was calculated using equation S3: Where thio = thioether, non-reducible species; H = heavy chain, L = light chain; and NGH = aglycosylated heavy chain.
A plot of IQ marker protein migration time versus the base 10 logarithm of nominal marker molecular weight was sometimes plotted for characterization purposes (see Figure S1 for representative electropherogram). This plot is not included in the quality parameters but may be useful in assay troubleshooting. Historical experience has been of an approximately linear curve, a linear fit of which has yielded R-squared values greater than or equal to 0.98 and relative residual standard deviations less than 5%.

Calculation of statistics including intermediate precision was performed in Microsoft
Excel using the Analyse-it® plug-in (Analyse-it Software, Ltd., Leeds, UK) as discussed in [1].
Briefly, the precision for a given quality parameter was calculated by performing an ANOVA to estimate the total variance of the dataset and to model the components of the variance due to within-day variability (repeatability) and between-day variability (encompassing multiple columns, instrument drift, etc). This analysis was accomplished using the Analyse-it® measurement system analysis (MSA) precision tool and setting the model to "Y with 1 random factor", where the factor was the date of analysis. The estimator was set to be standard deviation with a two-sided 95% confidence interval. The method was chosen to be "Exact/MLS", and the "ANOVA" option was checked.
Method Linearity Regressions. Linear regression (LINEST function in Excel) of corrected area versus loading concentration was performed using each individual data point (as opposed to means shown in Figure S2) to allow a statistical fit evaluation. Residuals were calculated based on the linear fit and residual standard deviation (rSD) and relative residual standard deviation (rrSD) were calculated for each fit. The rSD was calculated using Equation S4.
where Y calc is the theoretical Y value calculated from the line of best fit, Y meas is the measured Y value, and n is the number of data points in the curve. The rrSD was calculated using Equation S5.
(S5) Table S2 gives the goodness-of-fit parameters for each linear regression. relative abundance (RA) of the aglycosylated heavy chain (NGH) peak were recorded at the target loading concentration to be 9.1 (1.1) and 0.51 (0.03) % (SD), respectively. The aglycosylated heavy chain peak was chosen for this because it is at low abundance with a 6sigma SNR close to the LOQ limit at the target loading concentration. The LOD and LOQ were calculated using Equations S6 and S7.
where C inj is concentration of total protein loaded in the experiment (mg/mL) and V inj is the injection volume (mL). An electrokinetic injection was utilized for the CE experiments; therefore, V inj (mL) was calculated based on equation S8.
where µ mv is the mean apparent mobility of the minor variant peak in ; r cap is the internal radius of the capillary in µm; E EKI is the electrokinetic injection field in V/cm; and t inj is the EKI time in seconds. The apparent mobility of the aglycosylated heavy chain (mean of 15 measurements) was used for these calculations (µ mv = 3.6 x10 -5 cm 2 V -1 s -1 ). The mean apparent mobility was calculated using the following equation S9: where l d is the capillary length to detector in cm; l t is the total capillary length in cm; V app is the separation voltage in volts; and t mv is the migration time of the minor variant in seconds. The mass-based LOD and LOQ can be converted to a percent relative abundance corresponding to the experiment run at the target concentration using equations S10 and S11.
(S10) (S11) In the case of the CE assays discussed herein, C inj = C target because a minor variant (e.g. NGH) was present at appropriate SNR for this type of determination. This may not be true for all analytes and all assay types (as will be seen for SEC below). The two step calculation method described allows for mass-based LOD/LOQ to be calculated at a C inj smaller than the C target (but still within the linear range) and later converted to a percent-based LOD/LOQ at the target loading concentration (C target ) of the optimized assay.
Specificity. Method specificity and carryover was evaluated as described in the main text.
Intermediate Precision. For determination of intermediate precision, samples were prepared according to the optimized methods. Each day, a fresh vial of PS 8670 was thawed from ─80 °C to room temperature, inverted 5 times to mix, centrifuged briefly, and subjected to the appropriate sample preparation (reduced or non-reduced). One non-reduced sample preparation was performed per day for four days; this design was repeated for the reduced sample. Blank samples and instrument qualification (IQ) samples were prepared also as described. Samples were analyzed by CE-SDS as in Table S3. The sequences were of the format Blank─IQ─(PS 8670 × 3)─IQ─Blank. Quality parameters were calculated and subjected to statistical analysis as described above. For analysis under reducing conditions and quantitation of glycan occupancy and thioether relative abundance: 1. Dilute 10 µL of 10 mg/mL mAb with 85 µL 0.04 mol/L citrate-phosphate/1% SDS sample buffer (pH 6.7) in a Protein LoBind tube.
3. In the chemical safety cabinet, add 5 µL 2-mercaptoethanol to the tube. Cap tightly.
Vortex briefly to mix.
4. Incubate the mAb solution for 10 min in a 70 °C water bath.
5. Remove the vial from the water bath and allow to reach room temperature.
6. Briefly centrifuge the vial to collect any condensate.
7. Transfer the vial contents to a 0.2 mL sample vial and analyze by CE-SDS.
Performance Criteria. The performance criteria for the method under non-reducing and reducing conditions were set for each parameter based on the measured intermediate precision.
These criteria are useful for ensuring that the analytical method is in control, thus establishing confidence in the data acquired using the method. The criteria for the IQ are as follows:  Visually conforms to expectation (expected peak shape and pattern)  10 kDa internal standard migration time falls within ±3u c of the mean: (12.09 min to 12.81 min).
 100 kDa marker peak migration time falls within ±3u c of the mean: (21.34 min to 22.60 min).
The criteria for injections of PS 8670 under non-reducing conditions are as follows:  Visually conforms to expectation (expected peak shape, no new peaks above LOD). The criteria for injections of PS 8670 under reducing conditions are as follows:  Visually conforms to expectation (expected peak shape, no new peaks above LOD). Blank injections should contain only the 10 kDa internal standard peaks (main peak + known impurities) and no new peaks above the LOD.

CE-SDS Method Parameters
The detailed time programs for the CE-SDS instrument methods in 32Karat are given below in Tables S4-S6.   The initial conditions for the PA800 plus instrument used for all CE-SDS methods are given below in Tables S7 and S8.  The integration parameters employed for analysis of CE-SDS data in this work are given below in tables S9-S11.    (resolution dimer and monomer (R s ) number of theoretical plates for the monomer (N), and monomer peak asymmetry (A s ) were fit with a quadratic model [2,3]. The number of theoretical plates (column efficiency) of the monomer peak was determined using the statistical momentbased peak integration in Chromeleon 7 software instead of the USP integration method in order to fully monitor the peak tailing in calculation of the second peak moment (peak variance). The USP method (which uses a 5% peak height to evaluate peak fronting and tailing) was used as the default method for all other integration including determination of R s and A s .

Buffer
Sample Preparation. Samples were removed from the ─80°C freezer and placed on the bench to thaw. Samples were thawed at least 30 min at room temperature, and inverted 5 times with brief centrifuging in between each inversion to ensure homogeneity of the sample. Samples were then transferred to sample vial inserts and then to vials prior to analysis. Each sample was injected neat, with no dilution or buffer exchange. The concentration of each sample was approximately 10 µg/µL. The instrument qualification standard (IQ) was prepared by reconstituting the contents of the vial in 0.5 mL DIUF water and gently mixed. The vial was placed in ice for several minutes and mixed again. The IQ standard was aliquoted into a microcentrifuge tube and briefly centrifuged to remove any fine particulates before injecting onto the column. The IQ standard was stored at 2 o C to 8 °C and used within two weeks of preparation. Preparation of Stressed Samples. UV-Stressed Primary Standard 8670 was prepared as described above for CE-SDS.

Method
Data Analysis. Raw chromatograms were processed with Thermo Scientific Dionex Chromeleon 7 Chromatography Data System using optimized integration parameters (Table   S17). Retention times and peak areas were recorded for SEC sample analysis as listed in Table   S13. A plot of IQ marker protein retention time versus the base 10 logarithm of nominal marker molecular weight was sometimes plotted for characterization purposes (see Figure S3 for representative chromatogram). This plot is not included in the quality parameters but may be useful in assay troubleshooting. Historical experience has been of an approximately linear curve, a linear fit of which has yielded R-squared values greater than or equal to 0.99 and relative residual standard deviations less than 1%.

Calculation of statistics including intermediate precision was performed in Microsoft
Excel using the Analyse-it® plug-in (Analyse-it Software, Ltd., Leeds, UK) as discussed above in the CE-SDS section.

Method Linearity Regressions. Linear regression (LINEST function in Excel) of
corrected area versus loading concentration was performed using each individual data point (as opposed to means shown in Figure S4) to allow a statistical fit evaluation. Residuals were calculated as discussed in the CE-SDS section above and are shown in Table S14.  Chromeleon software-reported SNR values were utilized, which are calculated using an equation of peak height/noise (mAU). The LOD and LOQ were calculated using Equations S6 and S7 (with LMW used in place of NGH). The mass-based LOD and LOQ were then converted to a percent relative abundance corresponding to the experiment run at the target loading concentration using equations S10 and S11.
Specificity. The method specificity with respect to potential matrix interferents and carryover was assessed as described in the main text. Percent recovery was determined from linearity injection data described. The extinction coefficient and absorbance at 280 nm are used to predict the total mass of protein eluting from the column according to equation S15; and equation S16 allows an estimation of the recovery based on the known content of protein injected.
(S15) (S16) Where F is the flow rate (0.300 mL/min), ε is the extinction coefficient (1.42 mg/mL × cm at 280 nm), and l is the path length (cm). The total amount of "µg injected" was based on measured PS 8670 concentration of 10.014 mg/mL as determined in [1].

Summary of Qualified SEC Method
Performance Criteria. The performance criteria for the method were set for each parameter based on the measured intermediate precision. These criteria are useful for ensuring that the analytical method is in control, thus establishing confidence in the data acquired using the method. The criteria for IQ are as follows:  Visually conforms to expectation (expected peak shape and pattern). The criteria for injections of PS 8670 are as follows:  Visually conforms to expectation (expected peak shape and presence of 4 distinct peaks). Blank injections should contain no new peaks above the LOD.
The integration parameters employed for analysis of SEC data in this work are given below in Table S17.

DLS Materials and Method
See main text for details. Data analysis. Python 2.7 (Python Software Foundation, https://www.python.org/) was used to write a script that extracts raw data obtained from exported FI files and applies filters to remove edge particles, air bubbles, stuck particles, and particles smaller than 2 µm. Edge particles, specified by FI, were not considered in the analysis because they encounter the border of the image frame precluding accurate sizing of the particles. Air bubbles were defined as all particles with an equivalent circular diameter (ECD) greater than or equal to 5 µm and an aspect ratio greater than 0.9. Particles that possess both characteristics were removed by the script.

Buffer
To remove stuck particles from the raw data, the field of view (1280 pixels by 1024 pixels) was divided into 10 pixels by 10 pixel bins to create a 2D histogram. Each bin value represents the number of particles found in a given spatial location over the entire run. For a normal run, bin values vary from 0 to 5 particles per bin; bins with a stuck particle displayed a much higher particle number. If a bin contained a number of particles six standard deviations or more greater than the median of unique values in the histogram, all particles in that bin were rejected and were not included in the mass calculation. Once these filters were applied, the mass of the protein within the particles was calculated based on Equation S17 and S18 as described by Kalonia et al. [4]. The mass was calculated for each particle using the assumption that the density of dry protein was approximately 1.41 g/mL [5] and that 20% of protein particles were made up of protein (the remaining 80% was composed of water) [4]. The total mass of protein within all of the particles in a sample was obtained by summing the masses for each particle. The concentration of protein in the particles (Equation S18) was calculated by dividing the mass obtained from Equation S17 by the volume analyzed by the FI. Based on these two equations, it is possible to approximately describe how much protein was in the particles relative to the protein in solution. This was obtained by taking a ratio of the concentration of protein in the particles (Equation S18) to the total protein concentration in solution of the unstressed sample.    3   1000   41  1  30  1   070638  0  45  0  1  70864  1  70763  1

 
Concentration of protein in particles (ng/mL) = (S18) C = circularity; represents the squate of the ratio of the circumference of an equivalent area circle over the measured perimeter of the particle (0 = most fibrous to 1 = spherical). ECD = equivalent circular diameter (µm); defined as the diameter of a polystyrene microsphere with the same image area as the observed particle Sample preparation for PS 8670. Four vials of the PS 8670 material were obtained for assessing precision of the flow imaging method. The PS 8670 samples were removed from the -80 C freezer and placed at room temperature on the bench to thaw for thirty minutes. Prior to analysis, the vials were inverted gently 5 times.
Method Development and Optimization. The effect of degassing, effect of purge and prime volumes, effect of different solutions on the optimization step, and effect of sample handling on observed particle concentrations were assessed. For these studies, that required large volumes of samples, a commercially available polyclonal IgG solution, which was intentionally aggregated as described above, was used.
To study the effect of degassing, the polyclonal IgG sample was either analyzed on the instrument immediately or degassed for 10 minutes under vacuum prior to analysis. To study the effect of prime volume (volume required to clean the flow cell and tubing of the previous sample) on particle concentrations, the protein sample was analyzed as above (with no degassing), but either 2 mL or 0 mL of water was primed between each run. To study the effect of purge volume on particle concentration, the polyclonal IgG sample was analyzed with purge volumes of 0 µL, 30 µL, or 200 µL of water. Purge volume was composed of the sample and the flushing fluid remaining in the fluid path. This sample/fluid mix must be purged before sample analysis can begin. To study the effect on particle concentration of the fluid choice for instrument optimization (when the instrument determines the threshold pixel intensity to be used for particle isolation during analysis), each polyclonal IgG sample, prior to analysis, was optimized with either PBS buffer, or with the actual polyclonal IgG solution. To study the effect of sample handling, two analysts measured the particle concentration in the aggregated polyclonal IgG samples. Prior to dispensing the sample for analysis, they either tilted the protein solution 5 times or swirled it for 10 s. All control experiments were performed in at least triplicates, unless otherwise noted.
The results from the method variation experiments are shown in Table S18. Particle concentration represents the cumulative particle concentration equal to or above a given equivalent circular diameter (ECD). The particle distribution is separated into the five size bins: shows better reproducibility at counting larger particles and slightly worse repeatability at counting smaller particles. However, due to limited sample availability, it was determined that optimization with buffer or even water was the most convenient choice. From these studies, the following variables were chosen to be used on the PS 8670 and RM 8671 analysis: no degas, prime volume = 2 mL, purge volume = 200 µL, optimization with water.
The results from the sample handling experiments are shown in  Table S20. Initially, the flow cell was purged with 10 mL of deionized ultrafiltered (DIUF) water to obtain a low particle count. After this, the histidine formulation buffer was run to obtain initial particle concentrations. Since detergents can readily remove adsorbed protein, 0.02% (w/v) of Tween 20 was added to the histidine buffer to aid in desorbing protein if adsorption was occurring. If the protein is not being adsorbed, particle concentrations in buffers will remain low, even after running the PS 8670 material. If protein adsorption is a problem, but the protein is readily washed off the surface of the flow cell or tubing with buffer, there will be an observable increase in particle concentration in the histidine buffer that is run after the PS 8670 material. If the protein is less likely to come off in buffer alone, a histidine buffer containing a small amount of detergent Tween (buffer + Tween) is run after the PS 8670 sample. If the protein comes off in this solution, higher particle concentrations will be observed during the buffer + Tween run.
From Table S20, after run # 4 (PS 8670 run), the particle concentrations in all of the subsequent buffer runs are still low. There is a slight increase in concentration in run 6 (histidine + Tween) compared to the run # 3 (initial run of histidine + Tween). Similarly, after two more runs of PS 8670, the concentrations in the buffers (run # 8, 9, 11, and 12) are not significantly higher than their initial runs. There appears to be only a minor increase in subvisible particle concentrations in the buffers after the PS 8670 material has been run, indicating adsorption is not a problem.
Repeatability, Intermediate Precision, Count Accuracy, and Sizing Accuracy. The optimized method identified from the above section was used to assess the reproducibility, count accuracy, size accuracy, and precision that can be obtained with this technique. Count-Cal microspheres of nominally 5 µm diameter and 3000 mL 1 concentration were first vigorously shaken for 20 s and sonicated for 20 s. Then 0.7 mL was gently pipetted out and inserted into the sample port of the flow imaging system for analysis. Triplicates of these standards were analyzed each day over a period of 5 days. Initial Count-Cal concentration measurements were higher than manufacturer's specifications. A similar observation was made with multiple batches of the Count-Cal microspheres; this suggested that the FI flow cell was slightly larger than expected. To remedy this, a primary microsphere standard, composed of nominally 4 µm latex microspheres, was run multiple times on the instrument. These microspheres were tightly calibrated for concentration on a highly characterized light obscuration instrument. The experimental concentrations obtained on the DPA-4200 were compared to the precisely calibrated concentrations of the primary microsphere standard to obtain a concentration correction factor. This correction factor was applied to all of the raw concentration data to adjust for the larger-than-expected flow-cell.
Suspensions of 2 µm and 10 µm polystyrene microspheres in DIUF water were also analyzed. ETFE particles, prepared as described above, were also analyzed in triplicates over a period of three days. For every sample analyzed in all of these studies (unless otherwise noted), the FI optimize illumination step was performed with water, and 0.7 mL of the sample was loaded into the FI [6]. Of this 0.7 mL, 0.2 mL was used for purging the instrument at a flow rate of 0.17 mL/min. Between each run, 2 mL of water was used for priming at a flow rate of 6 mL/min.
The results of these studies are shown in Table S21. Over the span of 5 days, the Count-Cal solution, corrected for FI cell thickness, showed little variability with a mean particle concentration of 3095 (87) mL -1 (SD) between 3 µm and 8 µm in size. The intra-day measurements ranged from 2977 mL -1 to 3128 mL -1 with the CV values fluctuating from 1% to 4%, which is within the acceptable range of the FI (concentration repeatability is ± 5%, according to the instrument manual). The concentration accuracy prescribed by the instrument manufacturer is ± 10% with the Count-Cal bottle label stating that the concentration should fall in the range 3000 mL -1 ± 300 mL -1 . The particle size distribution for each sample was analyzed to determine the mode value in size in the samples; all size measurements are in ECD. The sizing of the nominally 5 µm microspheres ranged from 4.89 µm to 4.96 µm, with the CVs ranging from almost 0% to 3%, which is within the acceptable range (sizing repeatability = ± 5%) as given by the instrument manual. The Count-Cal bottle label states that the microspheres have a mean diameter of 5.010 µm ± 0.035 µm in size. The observed size is within the manufacturer's specifications.
The 2 µm and 10 µm microspheres were also run to study the precision in the concentration and size measurements. The concentration measured for both microsphere sizes was reproducible over three runs, with CVs below or equal to 2%. The microspheres were sized accurately and precisely (1.88 µm and 9.64 µm) with low CV values (≤ 0.5%). According to the instrument manual, sizing accuracy for spherical polystyrene microspheres is ± 0.5 µm for particles < 5 µm, and ±5 % for particles ≥ 5 µm.

Table S18
Effect of method variations on repeatability of sizing and counting of subvisible particles in an aggregated polyclonal IgG solution. The concentrations of the polyclonal IgG particles are separated into five size bins (ECD ≥ 2 µm, ≥ 4 µm, ≥ 8 µm, ≥ 12 µm, ≥ 20 µm). Each value is a mean of at least 3 separate runs (n = 3) with the uncertainty expressed as (SD).