(M1430-08-55) Quantitative Determination of Microcrystalline Cellulose in Drug Products by Chemometrics Analysis of Raman Spectroscopy

Purpose: Microcrystalline cellulose (MCC) is an excipient widely used in pharmaceutical preparations as disintegrant, adhesive, diluent. Thus, the quality of MCC should be controlled. However, when there are several excipients with lower solubility just like MCC in most formulations, it is hard to quantitatively analysis of MCC with interference from other insoluble excipients by traditional methods. To solve this difficulty, MCC is alternatively determined by chemometrics analysis of Raman spectroscopy.

Methods: Firstly, twenty-seven batches of testing formulation prototype (Test) were produced with nine concentrations of MCC, 80%-120% of the MCC concentration level in reference listed drug (RLD) while whole preparation process and other formula composition were kept consistent. Next, the spectra of all Test and two batches of RLDs were acquired by Raman spectrometer. The spectra of Test were divided into two types, one was calibration set to establish Partial Least Squares Regression (PLSR) model and evaluate root mean square error of cross validation (RMSECV), the other was prediction set to calculate root mean square error of prediction (RMSEP). Suitable spectra pretreatment process could significantly reduce the interference signal and improve the accuracy of the PLSR model(1). Therefore, design of experiment (DOE) was introduced into the spectra pretreatment process. The crucial parameters were assessed to ensure the accuracy of the model by predictable index of model (Q²), RMSECV, RMSEP and ratio of RMSEP to RMSECV (Ratio). Finally, the established model was used to predict the concentration of MCC in the RLD.

Results: A wonderful or good-fit model needs Q² close to 1, RMSECV and RMSEP as small as possible, and Ratio close to 1 to ensure the stability of the internal and external sample predictions. The spectra preprocessing improved the accuracy of model by maximizing the elimination of useless signals and retaining the main information of features(2). To reduce the number of experiment and find critical factor, some spectra pretreatment algorithms were selected to orthogonal experiment (Fig. 1). The results showed that baseline correction is the most important factor. The model of S9 with better Q² and S1 without spectra preprocessing were selected to evaluate model, as examples. The original spectra (Fig. 2A) was compared with the baseline corrected spectra (Fig. 2B), it clearly revealed the necessity of baseline correction. Although, these samples had same kinds of excipients with same concentrations and same preparation process, there were a big difference in the spectra due to different sources of the excipients and other potential environmental factors. These interferences, which were not related to concentrations of MCC, could be eliminated by baseline correction. Moreover, the spectra preprocessing did not interfere with the recognition of MCC with different concentrations (Fig. 2C). The S-plot of S9 (Fig. 2D) showed that most of the Raman shifts that greatly contribute to spectra modeling were attributed to the characteristic bands of MCC.The prediction abilities of the model were also proved by the score plots (Fig. 3) with significance level of 0.05. The sample marked with red dots indicated that their concentration of MCC were within 10% of the standard Test (100% MCC). Although, no discrete values appeared in calibration set of S1(Fig. 3A), RLDs in S1 were out of bounds (Fig. 3C). This phenomenon indicated that model of S1 did not have the ability to predict RLD. And the sample of calibration set (Fig. 3B) marked with red dots and RLD in S9 (Fig. 3D) clustered together in the same red circle. This result not only showed that the model could predict both Test and RLD, but also confirmed that the concentration of MCC was close to between RLD and Test. The same analytical process was carried out for other models of orthogonal experiment. And S9 had the most satisfactory accuracy, precision and prediction ability.

Conclusion: In this study, the PLSR models of MCC were established by reasonable DOE. An optimal model was selected among amounts of data analysis. MCC in RLD and Test were compared through the model. At present, the focus of chemometrics modeling is on the identification of natural products and process analysis in pharmaceutical sciences. However, this experiment shows that chemometrics modeling also provides a new way to measure insoluble excipients in complex formulations.

References: 1. Hu Y, Erxleben A, Ryder AG, McArdle P. Quantitative analysis of sulfathiazole polymorphs in ternary mixtures by attenuated total reflectance infrared, near-infrared and Raman spectroscopy. J Pharm Biomed Anal. 2010;53(3):412-20. doi: 10.1016/j.jpba.2010.05.002.
2. Morais CLM, Santos MCD, Lima KMG, Martin FL. Improving data splitting for classification applications in spectrochemical analyses employing a random-mutation Kennard-Stone algorithm approach. Bioinformatics. 2019;35(24):5257-63. doi: 10.1093/bioinformatics/btz421.

Acknowledgements: Thank Chemmind Co., Ltd for help on Modellab software and thank Sartorius. Co, Ltd. for help on Simca software. And all authors declare no conflict of interest.

Fig. 1 Orthogonal design table and their results

Fig.2 (A) original spectra of RLD and Test with 100% MCC (B) spectra of RLD and Test with 100% MCC after baseline correction (C) spectra of Test with different concentration MCC after baseline correction (D) S-plot of S9

Fig.3 Score plot of calibration set in (A) S1, (B) S9 and RLDs in (C) S1, (D) S9