(233ao) Predicting the In Vivo Performance of Amorphous Solid Dispersions Based on Molecular Descriptors and Statistical Analysis
- Conference: AIChE Annual Meeting
- Year: 2016
- Proceeding: 2016 AIChE Annual Meeting
- Group: Pharmaceutical Discovery, Development and Manufacturing Forum
Monday, November 14, 2016 - 3:15pm-5:45pm
To retard drug precipitation, the selection of the stabilizing polymer (POL) is crucial. From a screening perspective, it is of interest to early evaluate different polymers in terms of their supersaturation potential and precipitation inhibition effect, in order to reduce time, risk and costs during formulation development.
Current screening methodologies to predict ASDs performance in solution generally consist in the implementation of miniaturized kinetic solubility experiments based on pH- or solvent-shift methods, using smaller volumes apparatus or in 96-well plates . An alternative is the use of computational tools particularly the existing physiologically-based pharmacokinetic (PBPK) models, such as GastroPlusâ?¢ or SimcypÂ®. However these models need to be combined with accurate in vitro / in vivodissolution experiments as input parameters. These data is typically obtained at advanced stages of formulation development. Thus, from an early screening perspective, there is lack of a theoretical basis for the ranking of the best polymers with precipitation inhibition effect.
Today, there is a significant number of research papers demonstrating the in vivo bioavailability of ASDs when compared with the reference products . Taking advantage of the data available in the literature, the purpose of this work was to develop a statistical model, based on multivariate data analysis tools that could help on guiding ASD formulation design to obtain the desired in vivoperformance.A database with 37 observations (ASDs) and 35 XY variables was constructed as schematically shown in Figure 1. The X variables include simple molecular descriptors of the APIs and POLs (e.g. MW, Log P, Tm, Tg, number of H-bond donors/acceptors, etc) and interactions thereof, while the Y variables correspond to experimental data (e.g. Cmax, tmax, AUCin vivo) obtained from the literature. Principal component analysis (PCA) and partial least square (PLS) models were developed using SIMCA-P+ 13.0 (Umetrics, Sweden). All variables from the dataset were mean centered and scaled to unit variance before effective analysis, in order to give variables equal weight.
Figure 1.Schematic representation of the dataset.
First, an outlierâ??s identification stage was conducted using PCA to handle variability in data from disparate sources. An amorphous dispersion was identified as an outlier, which was then removed from the dataset.
After outlierâ??s identification, a two-component PLS model with multiple responses was developed - Log (AUCin vivo, ASD/AUCin vivo, REF) and Log (Cmax in vivo, ASD/Cmax in vivo, REF). These Y-variables were identified as being correlated from the PCA analysis (data not shown). Figure 2 shows the correlations between the observed Log (AUCin vivo, ratio) and Log (Cmax in vivo, ratio) and the respective predicted values, together with the predicted test set. These correlations were obtained after variable selection to maximize model performance/minimize prediction error.
Figure 2 â?? Observed data versus predicted data by the PLS model: A â?? Log AUCratio response; B - Log Cmax, ratio; training set (red circles); prediction set (blue circles).
The R2 and Q2 values for each Y-variable were around 0.7 and 0.5, respectively. Given the existing uncontrolled variability in the data - in vivodata obtained disparate sources and different animal models - the obtained model is considered adequate for interpretation purposes. Figure 3 shows the loading plot together with the most important variables (VIP) plot of the model. The loading plot shows the relationships between the inputs and output variables simultaneously, while the VIP shows the variables by descending order of influence in the model.
Figure 3â?? PLS loading plot (A) and VIP plot (B) of model with two PCs, based on the training set.
Â As can be seen from Figure 3, the most important variables for the model included general API-related molecular descriptors, POL-related molecular descriptors and API-POL interaction variables mainly associated with the capacity of the POL to establish hydrogen bonding with the APIs. These results confirm the key role of polymers for the kinetic stabilization of drugs in solution.
Figure 4 shows two scatter plots of two of the most important variables for the model. The size of each point/observation corresponds to the AUCin vivo and Cmax,in vivogains which were categorized in three different levels (low, medium and high).
Figure 4 â?? Scatter plots of two of the most important variables for the model. The size of the points/observations represent the AUCin vivo Â gain (A) and the Cmax,in vivogain (B).
The importance of hydrogen bonding for improving the in vivo performance of ASDs was in line with the observation that polymers with higher solubility parameters also tend to contribute for higher AUCs. In general, cellulose-based polymers (i.e. HPMCAS, HPMC, HPMCP) seem to provide better precipitation inhibition across different classes of drugs, when compared with other polymer families.
To conclude, chemometric tools based on multivariate data analysis can be applied to assess correlations between molecular descriptors of the formulation ingredients and performance related output variables, namely AUCin vivo and Cmax, in vivo. These models are useful not only for narrowing the scope of formulations variables to be further explored, but also to rapidly identify suitable systems with synergistic interactions for subsequent clinical evaluation.<span">
 Bevernage J. et al., Evaluation of gastrointestinal drug supersaturation and precipitation: Strategies and issues, International Journal of Pharmaceutics, 2013, 453 (1), pp. 25-35;