Abstract
A quantitative spectral method has been developed to precisely measure the color of protein solutions. In this method, a spectrophotometer is utilized for capturing the visible absorption spectrum of a protein solution, which can then be converted to color values (L*a*b*) that represent human perception of color in a quantitative three-dimensional space. These quantitative values (L*a*b*) allow for calculating the best match of a sample's color to a European Pharmacopoeia reference color solution. In order to qualify this instrument and assay for use in clinical quality control, a technical assessment was conducted to evaluate the assay suitability and precision. Setting acceptance criteria for this study required development and implementation of a unique statistical method for assessing precision in 3-dimensional space. Different instruments, cuvettes, protein solutions, and analysts were compared in this study. The instrument accuracy, repeatability, and assay precision were determined. The instrument and assay are found suitable for use in assessing color of drug substances and drug products and is comparable to the current European Pharmacopoeia visual assessment method.
LAY ABSTRACT: In the biotechnology industry, a visual assessment is the most commonly used method for color characterization, batch release, and stability testing of liquid protein drug solutions. Using this method, an analyst visually determines the color of the sample by choosing the closest match to a standard color series. This visual method can be subjective because it requires an analyst to make a judgment of the best match of color of the sample to the standard color series, and it does not capture data on hue and chroma that would allow for improved product characterization and the ability to detect subtle differences between samples. To overcome these challenges, we developed a quantitative spectral method for color determination that greatly reduces the variability in measuring color and allows for a more precise understanding of color differences. In this study, we established a statistical method for assessing precision in 3-dimensional space and demonstrated that the quantitative spectral method is comparable with respect to precision and accuracy to the current European Pharmacopoeia visual assessment method.
Introduction
Color is an important quality attribute of protein drug solutions in pharmaceutical development (ICH Q6B) (1). Color in a protein solution may come from the protein itself due to light scattering as well as from impurities. The variation of color in a protein drug solution can be an indication of inconsistency of a manufacturing process and may reflect varying impurity levels in that protein drug solution. The current color measurement method used in pharmaceutical manufacturing described by the European Pharmacopoeia (Ph. Eur.) is a visual assessment method (2, 3). It requires an analyst to select the best color match of a sample to a Ph. Eur. color reference solution. This visual assessment method has potential variability from the preparation of the color reference solutions and requires a subjective choice when an analyst picks the best match (3). Being limited to 37 Ph. Eur. color reference solutions this method does not capture full data on hue and chroma that would allow for improved product characterization and the ability to detect subtle differences between samples.
To overcome these challenges, a quantitative spectral method was developed for color determination that reduces the variability in determination of the color and allows for a more precise understanding of color differences. With this quantitative method, information about the color of a protein solution is captured in its visible absorption spectrum, which can then be converted to color values defined as International Commission on Illumination (CIE) L*a*b* color values (2, 3). These quantitative L*a*b* values allow for calculating the best match of a sample's color to a color standard. This method correlates to the visual assessment method but with improved reproducibility. It can be employed for both low and high concentration liquid formulations and allows for enhanced measurement of the color of protein solution during development and throughout its shelf life. Moreover, implementation of this method can lead to an improved understanding of the impact of product- and process-related impurities on the color of a protein solution, and it can help to determine the impact of process parameters on the color of that protein solution during the course of process development.
The HunterLab UltraScan VIS is a visible-range color measurement spectrophotometer that measures a sample's absorption spectrum and correlates it to human color perception. In this study, we compare the quantitative spectral method with the current visual assessment method and determine assay precision and accuracy as well as instrument repeatability (4) per the guidance in ICH Q2 (R1). As color is a bulk solution property and not a specific measure for purity or any individual impurity, certain validation characteristics in ICH Q2 (R1) are not applicable for a color assay such as specificity and linearity. Additionally, an assessment of the linearity of the assay is not applicable as the assay correlates to human visual perception of color (5). In this quantitative spectral method color is represented in a 3-dimensional (3D)color space (L*a*b*) (3). Determining assay precision within a 3D space is not obvious and as such we present here implementation of a measure that for color differences allows for statistical evaluation of assay precision. Assay range is covered within the Ph. Eur. reference color solutions. The assessment described here demonstrated that the method was ready for full validation with acceptance criteria based on this study to enable its use in a quality control (QC) clinical laboratory.
Materials and Methods
All measurements were carried out using a HunterLab UltraScan VIS spectrophotometer (HunterLab, Inc., Reston, VA). The system includes a spectrophotometer coupled to EasyMatch QC-ER software (version 4.73). In the UltraScan VIS optical system, the diffuse (sphere) geometry is configured for measurements in the total transmission (TTRAN) mode. The optical system has an effective bandwidth of 10 nm, and spectral data is reported every 10 nm. The conversion from absorption spectrum to color perception involves tristimulus color calculations, which are performed from 360 nm to 780 nm as recommended by the CIE (5). The software uses CIE L*a*b* values to represent different colors under daylight D65 and 10° standard observer conditions. The instrument was standardized for 100% transmission (cuvette filled with water) and 0% transmission (light path blocked using a black light blocker; HunterLab, A04-1011-745A). All measurements were carried out using a 1 cm path length cuvette and the HunterLab micro cell holder. The algorithm for color matching to Ph. Eur. standards followed that outlined in the accompanying manuscript (3). Briefly, best match is the closest interpolated reference color solution and report value correlates to visual assessment readout and is the next darker reference color solution within that series (3).
The didymium solid wavelength standard and 0.2 optical density (OD) neutral density filter were provided by HunterLab for system suitability testing (verification of wavelength and photometric performance) of the instrument. The various sizes of cuvettes used for the measurements were either from Starna Cells, Inc. (Atascadero, CA) or Brand Tech Scientific, Inc. (Essex, CT).
The Ph. Eur. reference color solutions were made of yellow, red, and blue primary solutions from Ricca Chemical Company (Arlington, TX) and were prepared gravimetrically (3).
System Suitability Standards
System suitability must be checked on a daily basis per good laboratory practice and good manufacturing practice (GMP) requirements to ensure the instrument can perform accurately in accordance with vendor specifications. A solid didymium color standard and a 0.2 Optical Density (OD) neutral density filter were employed for this purpose. The transmission spectrum of a didymium color standard and a 0.2 OD neutral density filter collected by a HunterLab UltraScan VIS spectrophotometer are shown in Figure 1. Unlike standardization for sample testing, for system suitability testing the instrument was standardized against air (100% transmission).
Transmission spectrum of a didymium color standard (solid line) and a 0.2 OD neutral density filter (dashed line). The two filters were used for system suitability testing.
The didymium filter standard was used to verify the wavelength accuracy and the photometric response. The transmission values of the didymium filter were factory-baselined at two specific wavelengths (430 nm and 570 nm). Using these baseline values we were able to verify both the wavelength accuracy and the photometric response signal to ensure that the instrument was within the tolerance range that the vendor specified. The neutral density filter was used to verify the midrange photometric response over the whole visible range (360 nm to 780 nm). The spectrum was then converted by the EasyMatch QC-ER software to XYZ values (precursors of L*a*b*), which correspond to the color intensity of the blue, green, and red portions of the spectrum, respectively. Instrument suitability can be confirmed if the measured XYZ values are within the tolerance range of the factory-baselined target values.
In order to evaluate the reproducibility and robustness of the proposed suitability standards, two sets of didymium and neutral density filters were evaluated for six replicates on two instruments with different orientations (cuvette facing frontward and facing backward). The system suitability standards were removed and reinserted into the cuvette holder for each measurement.
Assay Precision
Assay precision was evaluated using representative monoclonal antibody (mAB) protein solutions mAB X, mAB Y, and mAB Z at concentrations of 25, 50, and 200 mg/mL, respectively, as low, medium, and high protein concentration samples.
Instrument Repeatability:
Instrument repeatability was determined by one analyst, who measured each of the three protein solutions six times on the same instrument without removing the cuvette from the sample holder between readings. The data from this study reflect only the instrument repeatability without the assay preparation variability.
Assay Repeatability:
Assay repeatability was determined by one analyst, who measured each of the three protein solutions six times on the same instrument by removing the cuvette and refilling the sample between readings. The data from this study reflect the assay variability including both the sample preparation variability and the instrument variability.
Intermediate Precision:
Analyst-to-analyst and instrument-to-instrument variability (intermediate precision) were evaluated by comparing two analysts measuring the three protein solutions six times on two instruments. Cuvettes were refilled between sample readings.
Comparability with the Visual Assessment Method (Accuracy)
There are no published L*a*b* values for Ph. Eur. color reference solutions or other certified standards and, therefore, it is a challenge to check the accuracy of the quantitative spectral method by using certified color standards directly. However, by comparing the spectral method with the current visual assessment method, we can demonstrate the comparability of the two methods and thus, indirectly demonstrate the accuracy of the spectral method.
We first tested six gravimetrically prepared Ph. Eur. color reference solutions as unknown samples (blinded to the analysts). Five analysts were asked to visually compare them with thirty-seven Ph. Eur. color reference solutions using the visual assessment method and report the best match (3). The same samples were then read on the HunterLab instrument.
Results and Discussion
Evaluation of the System Suitability Standards
In order to evaluate the reproducibility and robustness of the suitability standards proposed for system suitability testing, two sets of the didymium standards and two sets of 0.2 OD neutral density filters were tested on two instruments. The relative standard deviations to the mean (RSDM) of six replicates for each standard on each instrument were calculated. The data were compared against the vendor's specification.
Table I summarizes the percent transmittance of the didymium standards at 430 nm and 570 nm. Both sets of standards showed good reproducibility and robustness on both instruments. All measurements were within the tolerance range specified in the vendor's certificate (%T430 = Target ± 2.0, %T570 = Target ± 3.1).
Reproducibility and Robustness of the Didymium Standards
The transmission spectra of the neutral density filters were collected and then converted by the EasyMatch QC-ER software to the XYZ values that correspond to the color of the standards. As shown in Table II, both filter sets showed good reproducibility and robustness on both instruments. All measurements met the vendor's specification (X, Y, Z = Target ± 0.8).
Reproducibility and Robustness of the Neutral Density Filters
The robustness of the standards was also evaluated by comparing the orientation of the standards in the cuvette holder facing frontward and backward. There was a negligible difference in the readings between the different orientations, and both passed vendor's specifications (data not shown).
Assay Precision
Measurement of Precision of Color Differences in 3D L*a*b* Color Space:
Typically, during the course of a method validation such as spectroscopic determination of protein concentration, precision is expressed with respect to standard deviation (SD) and RSDM of the measured values. Due to the nature of the 3D CIE L*a*b* color space, the variability of the L*a*b* values cannot be represented by their RSDM directly (for example, analyzing each dimension separately), as that does not correlate to color perception; thus, they are not presented here.
Within the CIE L*a*b* color space, total color difference between two points (two different colors) is expressed as a distance (Euclidean distance, ΔE*) between points within the 3D L*a*b* color space (2, 3); this distance correlates to human perception (2, 3). For example a larger ΔE* will correlate to two color samples that are visually perceived to be farther apart as compared to a smaller ΔE* for two color samples that would be perceived to be closer in color. As such a measure of precision requires expression of distances to a mean within 3D L*a*b* color space.
It is important to note that due to minor nonlinear differences between human visual perception and the L*a*b* color space, the ΔE* calculation has been updated from a Euclidean calculation to a more complex elliptical equation (ΔE2000), which corrects for these minor differences (6). The work described here utilizes the ΔE2000 calculation method to estimate overall color difference between two colors.
The mean or center point (in L*a*b*) of a number of measurements is determined by separately averaging the L*, a*, and b* values independently and the center point or mean is the average L*, average a*, and average b*. The differences or distance between a measured and the mean value is therefore the calculated ΔE2000 between each measured point and the calculated mean. The ΔE2000 distance between each measurement to the mean of the replicates can be calculated, and the closer (smaller ΔE2000 value) they are, the better the repeatability and precision. The maximum value of ΔE2000 between replicates and the mean is reported as a measure of variability in the repeatability and precision sections below. Within the L*a*b* color space, a ΔE2000 of 1 unit between two points (two colors) is considered a small but perceptible difference in color (7). Two solutions with a ΔE2000 of less than 0.5 between them will perceptibly be considered the same color by most analysts (7). Within this study we rely on these values of color that a human can perceive to set precision acceptance criteria.
Instrument Repeatability:
Instrument repeatability was determined using representative protein solutions as shown in Table III. The instrument gave the same reported color values for six replicates of each sample. The maximum ΔE2000 to the mean was calculated for each protein solution: 0.10, 0.05, and 0.23 for mAB X, mAB Y, and mAB Z, respectively. These low values of ΔE2000 indicate that all of the measured values are tightly clustered around the mean, demonstrating very good repeatability. This is consistent with our observations in validating other spectral methods that if one measures only instrument variability by reading the same sample in the same cuvette without removing it from the sample holder the resultant measured values are very close to each other. In addition, it should be noted that as mentioned previously, a ΔE2000 elliptical color difference equal to 1 is a prediction of a visible color difference between two colors for the average human observer. Because human perception can only note a color difference when ΔE2000 ≥1 (7), instrument repeatability is considered acceptable as compared with the visual assessment method.
Instrument Repeatability (Instrument A)
Assay Repeatability:
Assay repeatability is summarized in Table IV. The maximum ΔE2000 between each measurement to the mean of the six replicates are 0.03, 0.08, and 0.13 for mAB X, mAB Y, and mAB Z, respectively. Again, as with the instrument repeatability, these measures of assay repeatability are quite low and the difference between instrument and assay repeatability are not meaningful. It should be noted that all samples are measured neat so there is no additional error incurred from sample dilution, which is commonly observed for spectral measurement of protein concentration. The measurement of neat solutions does, however, required careful washing of cuvettes between sample readings, especially when working with more-viscous high concentration protein solutions. It is recommended to rinse the cuvette thoroughly with water and then with the solution to be measured at least once before adding the actual solution for measurement to the cuvette. As with instrument repeatability, the assay variability measured is well below that observable by human perception (ΔE2000 = 1) and these results are considered acceptable.
Assay Repeatability (Instrument A)
Intermediate Precision:
Table V and Table VI show the L*a*b* values of the three protein solutions obtained from two different analysts and two instruments. The L*a*b* values used to calculate the ΔE2000 in the tables are the averages of six measurements from each protein solution. For instrument-to-instrument variability the ΔE2000 between the mean of instrument 1 and the mean from instrument 2 (six measurements each, Appendix Table SII–SIII) are 0.04, 0.05, and 0.06 for mAB X, mAB Y, and mAB Z, respectively. For the analyst-to-analyst variability the ΔE2000 between the mean of analyst 1 and the mean of analyst 2 (six measurements each, Appendix Table SI–SII) are 0.10, 0.05, and 0.11for mAB X, mAB Y, and mAB Z, respectively. These ΔE2000 values for intermediate precision are quite low and demonstrate the excellent precision of the instrument. Again, as a ΔE2000 = 1 is a prediction of a just-perceptible color difference by the average human observer, these intermediate precision results are considered well within an acceptable range compared with the visual assessment method. It should be noted that these measures of intermediate precision cannot be directly compared to the previous results as the repeatability numbers represent the farthest distance of a replicate from the mean whereas the intermediate precision represents the distance between two averages, that is, between two instruments or two analysts.
Instrument-to-Instrument Variability (Same Analyst)
Analyst-to-Analyst Variability (Same Instrument)
Comparability with the Visual Assessment Method
Table VII shows the best color match for the five unknown Ph. Eur. color reference solutions determined by the visual assessment method and the quantitative spectral method. The same results were obtained from all five analysts for the test samples, and their results matched the instrument readings. Comparability between the spectral and visual assessment methods using fifteen protein solutions of varying concentrations (20–200 mg/mL) and turbidities (ref I–IV) is shown in the Swartz et. al. paper published in this issue (3){Swartz, 2015 #11;Swartz, 2016 #11}. There is very good agreement between the quantitative spectral method and visual assessment method, with greatly improved reproducibility in the spectral method described herein.
Comparison of Visual Assessment Method and Quantitative Spectral Method Using Ph. Eur. Reference Color Solutions as Test Samples
Evaluation of Various Cuvettes
Standard 3.5 mL clear-walled quartz cuvettes with a 1 cm fixed path length (Starna #1-Q-10) are recommended for use with the HunterLab UltraScan VIS spectrophotometer. A minimum fill volume of 1.2 mL is required to ensure robust reproducibility (data not shown). A semi-micro 1.4 mL cuvette (Starna #9B-SOG-10) can also be used for HunterLab measurement with a minimum fill volume of 0.5 mL (data not shown). Inconsistent data were obtained from a 0.7 mL semi-micro cuvette (Starna #9B9-SOG-10) with multiple measurements of the same sample (data not shown), which was likely due to the narrow window of this cuvette (2 mm) that would allow any small bubbles or minor positional changes of the cuvette to have a significant impact on the readings. Therefore, this cuvette is not suitable for use with the HunterLab instrument. For high concentration proteins and antibody-drug conjugates, where a disposable cuvette would be highly desirable, acrylic cuvettes (Brand Tech Scientific Inc. #759080D) were evaluated (data not shown) and deemed acceptable.
Conclusions
This study establishes ΔE2000 as a statistical measure of precision for analysis of data in 3D space and demonstrates that the HunterLab UltraScan VIS spectrophotometer is suitable for quantitatively measuring the color of a protein solution in a GMP testing environment. All of the repeatability and precision measurements yielded ΔE2000 results of <0.25, which, as noted above, is considered well below what is a visually perceptible color difference. This study supports the use of this spectral quantitative method with protein solutions that report to the Ph. Eur. color standards. For protein solutions having a color outside of the range of the Ph. Eur. color standards, the method may still be employed, but should report CIE L*a*b* values instead of the Ph. Eur. standard best match. This assessment demonstrated that the method was ready for full validation with an acceptance criteria for precision and repeatability set at ΔE2000 to the mean ≤0.5. The subsequent QC validation study, using two analysts on two instruments measuring three protein solutions on three separate days for intermediate precision, was successfully executed with comparable results to those reported in this article. It should be noted that in this study all three proteins reported to the B color series. In the QC validation study one protein reported to the R color series and had very comparable results. When proteins are encountered that report to other color series, one would need to do some qualification of the assay in that color series. In our experience most proteins report to the B series, and we believe it would be a relatively rare event for a protein to report to the Y or GY series using the instrumental method. However, based on our studies and the theoretical basis of the method we fully expect the assay to perform comparably no matter to which color series the protein reports. This now enables the use of the quantitative spectral determination of color within a GMP environment for batch testing and stability. This current study also demonstrated that a didymium standard and a 0.2 OD neutral density filter can produce reproducible and robust results on different instruments and, thus, are suitable for system suitability checks. The quantitative spectral method is comparable to the current visual assessment method, but it can produce more precise color values.
Conflict of Interest Declaration
The authors declare that they have no competing interests.
Acknowledgments
The authors would like to thank Margaret Tang, Travis Horst, and Inna Notkin for their contributions to this study.
Appendix
Analyst 1 on Instrument A
Analyst 2 on Instrument A
Analyst 2 on Instrument B
- © PDA, Inc. 2016
