November 15, 2015 (Vol. 35, No. 20)
Evaluating Raman Spectroscopy as a Promising Technique for Cell-Culture Applications
Raman spectroscopy, which is a noninvasive technique, can be conducted using either small benchtop or portable instruments. Combined with chemometric analysis, it has attracted significant attention as a possible rapid, low-cost alternative method for evaluating cell-culture media used in biopharmaceutical manufacturing.
To date, several studies on the Raman analysis of cell-culture media have suggested that this technique is indeed promising. However, these studies generally involved the use of cell-culture media with widely differing compositions dissolved in aqueous solutions, or were quantifying high-level raw materials in media.
SAFC manufactures a number of dry-powder cell-culture media products, many of which are only slightly different from one another. Thus, to evaluate the capabilities of Raman spectroscopy for dry-powder media identification and raw material quantitation for similar media products, a comprehensive analysis of raw material spectral contributions to dry-powder media spectra was performed. Spectra of pure solid raw materials and any known raw material polymorphs were obtained and used for classical least squares (CLS) analysis of chemically defined dry-powder media spectra.
Unlike Fourier transform infrared (FTIR) spectroscopy, which is a direct vibrational absorbance technique, Raman spectroscopy is an indirect measurement of molecular vibrations. In Raman spectroscopy, the sample is exposed to monochromatic light and some of the molecules are promoted to a higher vibrational energy state. Lower energy scattered light is then emitted and detected.
Raman scattering is typically more intense for higher-energy light, but typically there is also more background fluorescence. Handheld Raman instruments often suffer from increased background fluorescence because they typically use 785 nm light. Newer 1064 nm instruments are now available that overcome this issue.
Materials and Methods
Two Raman spectrometers were purchased for identification of raw materials, the Bruker MultiRAM FT-Raman and the Rigaku FirstGuard 1064 handheld instrument. Both spectrometers were evaluated, but in general, the Rigaku FirstGuard 1064 was two to four times less sensitive to raw materials in dry-powder media (depending on the raw material). Thus, only the data from the Bruker instrument are discussed in this article.
Solid samples of raw materials (amino acids, vitamins, polymers, etc.) and formulations (six total: complete media and feeds) were analyzed by placing the solid samples into vials or tubes and placing them in the 1064 nm laser beam from a Bruker MultiRAM FT-Raman instrument. The laser power was 500 mW, the resolution was 2 cm-1 and 128 scans were obtained for each sample.
The formulations were rotated during analysis to cover more surface area and obtain more reproducible spectra. Six spectra of each raw material were obtained at multiple spots on the solid, and all spectra were averaged before being used in the analysis.
For each sample, the averaged data was truncated (700–1800 cm-1), the first derivative was obtained and the data was smoothed using a Savitzky-Golay algorithm. The two polymorph spectra for HEPES were separated and used as individual raw materials for the analyses. The spectra from the raw materials and the spectra from the formulations were then used to predict the mass percent of each raw material using a CLS model, which is depicted on the left-hand side of Figure 1.
According to the model, each raw material spectrum (RM #x spectrum) is multiplied by its respective mass fractions (MF RM #x) to generate a prediction for the dry-powder media spectrum (DPM spectrum prediction). Note that normalization of the data was not performed, so the Raman coefficients were directly proportional to the percent mass of the raw material in the dry-powder media. All analyses were performed using Matlab version R2014b.
Figure 1. A classical least squares (CLS) approach was used to analyze dry-powder media (DPM). Left: CLS model, RM (raw material), MF (mass fraction), and DPM. Right: Spectra of individual ingredients and the calculated spectrum of the corresponding mixture.
Results and Discussion
Raw material sensitivity. The CLS model was used to fit the DPM spectra for the six different media products, and the principal component analysis results for a media formulation containing several ingredients, including many at levels below 1–2%, are shown in Figure 2.
On the left-hand side of the figure, the pie chart illustrates the mass percents for each raw material. The pie chart on the right-hand side of the figure depicts the percent contribution for each raw material obtained from the Raman analysis. It can be clearly seen that while many raw materials were detected, the practical detection limit for the raw materials was 1–2% mass.
Notably, L-asparagine and dextrose compose greater than 95% of the dry-powder media Raman spectrum, and mixture of only those two components could be identified as the dry-powder media formulation shown in Figure 2 using a simple percent match identification test. It should also be noted that ingredients with broad and weak Raman bands, such as Lutrol and polyvinyl alcohol, are generally undetected even when present at concentrations > 2% mass.
Detection of polymorphs. Many pure solid compounds can adopt different crystalline structures that have different physical properties and different Raman spectra. This issue complicates identification of solid dry-powder media, because differences in the vibrational spectra of solid raw materials with the same primary chemical structure, but different crystalline or amorphous solid forms, can lead to false negatives when distinguishing between DPM spectra using percent matching or principal component analysis (PCA).
Figure 2. Mass percent and Raman spectral contri-butions for a complex cell-culture media formulation
HEPES buffer is one such compound with two different polymorphs, and its structure often depends on the supplier (Figure 3). Notably, the Raman spectra of these two samples are equivalent when obtained in aqueous solution (data not shown). Buffers, such as HEPES, are a high-concentration component in SAFC dry-powder media products; thus, polymorphism presents a large risk for false negatives in identity testing.
Dry-powder media identification. To determine whether Raman spectroscopy can be used for accurate identification of DPM formulations, six chemically defined media were evaluated. Medium 1 was a feed, Media 2–4 and 6 are complete media, and medium 5 is a concentrated medium. Media 2 and 4 are the most similar of all media products evaluated.
PCA was conducted for each medium, and the results are presented in Figure 3. The ovals represent the 95% confidence intervals for each medium. Media 3 and 6 had very large confidence intervals due to HEPES polymorphism, while media 2 and 4 were indistinguishable due to the lack of sensitivity to differences in low mass percent raw materials.
Recent reports of new analysis methods, however, suggest that the detection limits can be decreased by increasing the complexity of sampling. These results clearly demonstrate that polymorphism and lack of raw material sensitivity cause false negatives and false positives, respectively, when performing dry-powder media identification using Raman spectroscopy with minimal sample preparation. The six chemically defined feeds could not be sufficiently distinguished by simple percent matching of the DPM spectra. It should also be noted that polymorphism and lack of sensitivity also cause problems for multivariate analysis of this data, as seen in Figure 3.
Figure 3. Left: Raman spectra of solid HEPES obtained from two different vendors. The primary chemical structure is the same (LC/MS not shown). Right: Principal component analysis (PCA) scores plot for the DPM lot spectra from all media evaluated. The ovals are the 95% confidence intervals.
Conclusions
While Raman spectroscopy offers the significant advantages of rapid, noninvasive analysis for some raw materials, the results obtained in the present study suggest that it is an insufficient technique for the identification of complex chemically defined dry-powder cell-culture media. The unique identification of dry-powder media should be limited to products that have differences in high mass percent Raman active components, unless more complex sampling methods are pursued.
Even in this case, however, it is important to address polymorphism of high mass percent Raman active materials through modeling or sample preparation. In addition, the potential lower sensitivity of portable Raman spectrometers must also be taken into consideration when conducting analyses of dry-powder media.
Kevin Patrick Kent, Ph.D. ([email protected]), is senior scientist, Manisha Sahni works as associate scientist, and Chandana Sharma, Ph.D., serves as principal scientist at SAFC.
