(M1330-02-09) Computational Prediction of Initial Dissolution Performance after Co-milling with PVP

Purpose: Absorption limited by slow dissolution in the gastrointestinal tract is a major problem in developing new drug candidates in the research pipeline (1). One technique to increase the dissolution rate of drug molecules is co-milling (2). However, guidelines linking drug properties with improvements in dissolution rates by co-milling are lacking, resulting in a ‘trial-and-error’ formulation approach. Computational tools could be used to determine influential variables which are predictive of dissolution performance after co-milling. Therefore, the aim of this study was to quantitatively predict the initial dissolution performance after application of the co-milling technique, identifying the most influential drug properties.

Methods: Structurally diverse poorly water-soluble drugs (n = 25) were ball-milled with Polyvinylpyrrolidone (PVP) K25 for 15 minutes to prepare binary co-milled mixtures (COM). To prepare binary physical mixtures (PM), the drug substances and PVP were blended for 10 minutes. Dissolution tests of COM and PM were performed in USP 2 apparatus (n = 3) under non-sink conditions in FaSSIF. Measurements with in-line UV probes facilitated quantification of dissolved drug concentrations every 30 seconds. Further, the particle size of drug powders were determined and used as a model parameter. The linear initial slope ratios (LISR) were calculated by including the average values of the first seven measurement points of each dissolution profile. A linear fit was applied and the slope of was used to calculate the LISR using the following equation: Linear initial slope ratio (LISR) = 3 min LIS, COM / 3 min LIS, PM where 3 min LIS is the linear slope of the dissolution rate [ug/mL/min] of the first three minutes from the fitted line of the COM or PM. A partial least squares (PLS) model was developed to select the most influential drug properties on LISR in the training set (n = 19), which was validated with the test set (n = 6) to obtain reliable quantitative prediction.

Results: The model was able to accurately predict the initial dissolution gains after application of the Co-milling technique by only including four variables (R² = 0.76, RMSE_tr = 0.125, RMSE_te = 0.148, table 1). All measurement points are within the prediction interval of the training set (figure 1). Selected variables for inclusion in the PLS model are the measured powder parameter D10 and the predicted molecular descriptors, Kappa 3, ShaaCH and HBDnCH. D10 is a particle size parameter of pure drug powders measured before their use in co-milling, and Kappa 3 is a variable related to the complexity of the molecule; both factors are positively related to the ratio of the LISR. This indicates that a larger original particle size of API powders and increased complexity of the molecule’s structure have a positive effect on initial dissolution rate after co-milling. HBDnCH corresponds to the number of hydrogen bond donors in the drug molecule. Since PVP K25 is a hydrogen bond acceptor, a possible explanation is that hydrogen bonding between the drug molecule and the excipient during the co-milling process may lead to an increase in the initial dissolution rate.

Conclusion: The developed model was successfully able to quantitatively predict the increase in initial dissolution rate due to co-milling with PVP. The particle size, the complexity of the molecule and the number of hydrogen bond donors appeared to be influential factors when considering the likely increase of the initial dissolution after co-milling. Studies with additional excipients could show which of these properties are generally applicable to predict dissolution performance after co-milling, and which properties are excipient specific i.e., related to PVP K25.

References: 1. Loftsson T, Brewster ME. Pharmaceutical applications of cyclodextrins: basic science and product development. J Pharm Pharmacol. 2010 Nov;62(11):1607-21. doi: 10.1111/j.2042-7158.2010.01030.x. PMID: 21039545.
2. Vogt M, Kunath K, Dressman JB. Dissolution improvement of four poorly water soluble drugs by cogrinding with commonly used excipients. Eur J Pharm Biopharm. 2008 Feb;68(2):330-7. doi: 10.1016/j.ejpb.2007.05.009. Epub 2007 May 21. PMID: 17574401.

Acknowledgements: This work has received funding from the European Union´s Horizon 2020 research and innovation program under the Marie Skłodowska-Curie grant agreement No 955756 (InPharma).

Figure 1: PLS model for LogLISR.
Following abbreviations are used: Ratio of the linear initial slope for the first three minutes dissolution of drugs in co-milled mixtures and in physical mixtures (LISR), decadic logarithm of LISR (LogLISR).

Following abbreviations are used: RMSE = Root mean square error, RMSECV = Root mean square error of cross validation, Te = Test set, Tr = Training set, D10 = Portion of particles smaller than D10 [µm] is 10%, Kappa 3 = third order simple Kier-Hall shape descriptor, ShaaCH = Atom-type hydrogen E-state index for aCHa groups (aromatic carbons), HBDnCh = Sum of NPA partial atomic charges on nitrogen based HB donor hydrogen atoms.