From Mini to Micro Scale – Feasibility of Raman spectroscopy as PAT tool

Background: Active coating is an important unit operation in the pharmaceutical industry. The quality, stability, safety and performance of the final product largely depend on the amount and uniformity of coating applied. Active coating is challenging regarding the total amount of coating and its uniformity. Consequently, there is a strong demand for tools, which are able to monitor and determine the endpoint of a coating operation. In previous work, it was shown that Raman spectroscopy is an appropriate process analytical tool (PAT) to monitor an active spray coating process in a pan coater [1]. Using a multivariate model (Partial Least Squares) the Raman spectral data could be correlated with the coated amount of the API diprophylline. While the multivariate model was shown to be valid for the process in a mini scale pan coater (batch size: 3.5kg cores), the aim of the present work was to prove the robustness of the model by transferring the results to tablets coated in a micro scale pan coater (0.5kg). Method: Coating experiments were performed in both, a mini scale and a micro scale pan coater. The model drug diprophylline was coated on placebo tablets. The multivariate model, established for the process in the mini scale pan coater, was applied to the Raman measurements of tablets coated in the micro scale coater for 6 different coating levels. Then, the amount of coating, which was predicted by the model, was compared with reference measurements using UV spectroscopy. Results: For all 6 coating levels the predicted coating amount was equal to the amounts obtained by UV spectroscopy within the statistical error. Thus, it was possible to predict the total coating amount with an error smaller than 3.6%. The root mean square of error for calibration and prediction (RMSEC and RMSEP) was 0.335mg and 0.392mg, respectively, which means that the predictive power of the model is not dependent on the scale or the equipment. Conclusion: The scale-down experiment showed that it was possible to transfer the PLS model developed on a mini scale coater to a micro scale coater.


Introduction:
Pan coating as an important unit operation for the formulation of solid oral dosage forms can crucially influence the performance of a medicinal product.Film coatings are applied to pharmaceutical solid dosage forms for a number of purposes and may be categorized on the basis of their influence on drug release, namely: Immediate release coatings which are often used to improve product stability, to facilitate product identification and to improve in product organoleptic characteristics.Modified release coatings can be subdivided into delayedrelease (e.g.enteric) coatings and extended-release (sustained-release) coatings and are used to achieve a desired therapeutic effect.Active coating (where the active ingredient is coated onto the core) can be classified as an immediate release coating and upgrades an oral dosage form regarding the drug release profile.Fixed dose combinations in the core and the coating, respectively, are a typical field of application.Active coating is challenging regarding the total amount of coating and its uniformity.Consequently, there is a strong demand for tools, which are able to monitor and determine the endpoint of a coating operation.
In previous work, Romero-Torres et al. [2,3] and El Hagrasy et al. [4,5] implemented Raman spectroscopy as a PAT tool for coating processes by correlating Raman spectra of coated tablets with coating time, respectively Raman spectra of coated tablets with the average weight gain of tablets.Moreover, Raman spectroscopy had also been previously found to be suitable for the calibration of the coating thickness [6].For the first time Mueller et al. [1] focused on the determination of the amount and uniformity of active coating.Using a multivariate model (Partial Least Squares) the Raman spectral data could be correlated with the coated amount of the API diprophylline.While the multivariate model has been shown to be valid for the process in a mini scale pan coater (batch size: 3.5kg cores), the aim of the present work was to prove the robustness of the model by transferring the results to tablets coated in a micro scale pan coater (0.5kg).

Drug:
The water-soluble caffeine derivative diprophylline (BASF, Ludwigshafen, Germany) was used for the active coating as model drug because it is detectable well by both, Raman and UV spectroscopy.

Coating Solution:
An aqueous-based coating solution was prepared, consisting of 12 % w/w diprophylline, 6 % w/w hydroxypropyl methylcellulose and 2% w/w polyethylene glycol 1500.The polyethylene glycol and diprophylline were added to 2/3 of distilled water.After the API and polyethylene glycol were dissolved, the hydroxypropyl methylcellulose which was dispersed in 1/3 of distilled water of 90°C was slowly added.Then the solution was mixed for 20 minutes to disperse the polymer.The solution was allowed to rest for at least 12 hours to hydrate the polymer before use.

Tablet Coating
Both, lab and micro-scale batches of tablet cores were coated with the same film coating formulation.The lab-scale batch was coated using a BFC5, Bohle Film Coater (L.B.Bohle, Ennigerloh, Germany).A Pro-C-epT 4M8 Film Coater (Pro-C-epT, Zelzate, Belgium) was used to coat the micro-scale batch (table 1).The process parameters are shown in tables 2 and 3.Over a period of 150 min 20 tablets at a time were collected every 30 minutes.At the end of the micro scale coating process an average of (11±0.5)mg diprophylline as measured by UV spectroscopy was coated onto each tablet.

Raman measurements and calculations:
For the Raman measurements a RamanRXN2 analyzer of Kaiser Optical Systems (Ann Arbor, USA) with a laser wavelength of 785 nm was used.The spectrometer was equipped with a non-contact optic sampling device (PhAT probe).The excitation laser (785 nm diode laser) was introduced and magnified to form a circular illumination area of 6 mm diameter (area: 28.3 mm 2 ) to cover a large sample area, which improves the reliability and reproducibility of Raman measurements.

Reference analysis:
To determine the amount of coated diprophylline an UV spectroscopic method was applied (Lambda-2, Perkin-Elmer, Ueberlingen, Germany).The calibration covered a range between 1.00mg/500ml and 20.03mg/500ml.Three measurements at 273nm were performed for each calibration level and for each tablet after being dissolved in distilled water.

Calibration and validation development:
Tablets were collected from the mini scale coater at different stages of coating (tablet sets from 0-12.7mg diprophylline).For the off-line quantitative calibration development these tablets (n=42) were measured by Raman spectroscopy with a scanning time of 15 seconds for each tablet.Then the multivariate model was built up with this data set.As reference analytical method UV spectroscopy was applied to the same tablets in order to obtain the amount of coated diprophylline.
The same multivariate model, derived from the coating process at mini scale coater, was applied to tablets coated in the micro scale coater.Again, tablets were collected at six coating stages on the micro scale run (tablet sets from 0-11.8mg diprophylline).Then, the amount of coating, which was predicted by the model, was compared with measurements using UV spectroscopy.

Results:
A Partial Least Squares (PLS) predictive model was constructed using the Standard Normal variate (SNV) spectral data from the tablet sets with 0 to 12.7mg diprophylline (Table 4).

Table 4: Parameter set of the multivariate model
Figure 1 shows the selected wavenumber range (1200 -1400cm -1 ) the multivariate model (mini scale) is based on.Both peaks (1290 and 1330cm -1 ), which can be assigned to diprophylline (CN-stretch and imidazole ring stretch), gain intensity with coating time, reflecting the increasing amount of diprophylline.
Best predictive results for this model were obtained by using three principal components which explain 94.4% of the X-Variance (spectral data) and 99.4% of the Y-Variance (coated diprophylline amount).The root mean square of errors for calibration and prediction (RMSEC and RMSEP) are direct estimates of the modelling error and the prediction error, expressed in original measurement units.In this work the RMSEC of 0.335mg describes the error of the model built up at the mini scale coater.The RMSEP of 0.392mg, being almost equal to the RMSEC, shows that the model developed for the mini scale coater was successfully applied to the micro scale coater.Consequently, the predictive power of the multivariate model is not dependent on the scale or the equipment, since the predicted amount of coating was equal to the amount obtained by UV spectroscopy within the statistical error of 3.6% (calculated as RMSEP related to the average coating amount at the end of the coating run).

Conclusion:
In the present work, it was shown that Raman spectroscopy used as a PAT tool can be implemented in different scales of a coating process.Especially, it was demonstrated that multivariate models developed to correlate the API's coating amount to the Raman spectra for a coating process in the mini scale coater can be transferred to micro scale processes.Thus, there is no need to develop a new multivariate model when dealing with scale up, scale down or switching of the equipment.Since this work focused on the robustness of multivariate models regarding off-line measurements, it will be highly desirable for future investigations to deal with the transfer of inline measurements between different scales.

Figure 1 :
Figure 1: Baseline-corrected spectra of tablets during the coating run (off-line)

Figure 2 :
Figure 2: Calibration (left) and validation (right) variance regression model for SNV Raman spectra