Monitoring processing-induced crystallinity changes

Some of the procedures used to process pharmaceutical solids, such as milling, spray drying or lyophilization, can disrupt the crystalline structure and lead to the formation of amorphous regions. As the physicochemical properties of pharmaceuticals are influenced by the form of their solid state, such phase transformations can profoundly impact both processing behaviour¹ and the bioavailability of the active ingredient in the finished product (Figure 1).^2,3 Changes in dissolution rate and chemical stability, for example, influence bioavailability, while changes in compressability, compactability and hygroscopicity influence processability.

The influence of phase transformations on product performance is by no means limited to APIs; changes in the excipients in a dosage form during processing also have to be considered.

Figure 1: The amorphous state: properties and significance.

To establish the integrity of the finished product it is important, therefore, to determine the existence and amount of amorphous material within a crystalline matrix. A number of different methods are available, but they are best divided into four groups: gravimetric, calorimetric, spectroscopic and density methods.

Gravimetric methods are based on differences in the absorption of water between the crystalline and amorphous states⁴. Although such methods can determine the amorphous content quite accurately, they are time-consuming and can destroy the sample (e.g., dynamic vapour sorption). Consequently, they are unsuitable for on-line measurements. The same is true of calorimetric methods⁴, which measure the enthalpy changes during crystallization. Density measurements, although nondestructive, tend to be time-consuming and not very accurate.

This leaves spectroscopic methods, where differences in the spectra generated by amorphous and crystalline materials are evaluated. Fourier transform Raman spectroscopy (FT-RS)⁵ and Fourier transform Raman near infrared spectroscopy (FT-NIRS)⁶, are particularly fast, accurate and nondestructive and, therefore, well-suited to on-line analysis.

X-ray powder diffraction (XRPD) is another potentially useful nondestructive spectroscopic method. Previously, it has been unable to compete with FT-RS and FT-NIRS in terms of measurement time and accuracy. Detection limits (DLs) cited in the literature have shown considerable variation. Recent refinements in hardware and software, however, have significantly increased the technique's speed, accuracy and versatility.

Experimental techniques

XRPD. This is a versatile, nondestructive technique. By measuring the angular positions and intensities of the diffracted radiation a wealth of structural, physical and chemical information can be obtained about the material investigated.

NIRS. This is a nondestructive, spectroscopic method utilizing the NIR region of the electromagnetic spectrum (wavelength from 800–2500 nm). This spectral region is particularly suitable for the excitation of the vibration of functional groups in organic compounds, such as OH-, NH- and CH.

Materials

Alpha-lactose monohydrate was used as a model substance. In the pharmaceutical industry, this material is widely employed as a filler and diluent in tablets and capsules, as a carrier substance in dry powder inhalers and as a lyoprotectant during the lyophilization of proteins.

To ensure complete recrystallization of any amorphous regions present, samples of 100% crystalline alpha-lactose monohydrate were stored at a relative humidity of 65% at 30 °C for 1 week before use. Samples of amorphous lactose were prepared by lyophilization of a saturated lactose solution.

Spectra were obtained of the pure crystalline and amorphous form and the binary mixtures at intervals of 10% (10–90%) and 1% (1–9% and 90–100%).

Instrumentation and methods

XRPD patterns were obtained using an X'Pert PRO MPD system (PANalytical B.V., The Netherlands) with nickel-filtered copper Kα radiation (40 kV, 40 mA), Bragg-Brentano focusing optics and the X'Celerator detector (PANalytical B.V.) Measurement conditions were as follows:

• Mixture 0–100% crystallinity: 0.008° 2θ step size; 3 min/measurement; scan range 4–40° 2θ.

• Mixture 1–9% amorphous content: 0.016° 2θ step size; 2 min/measurement, scan range 7–45° 2θ.

• Mixture 0–10% crystallinity: 0.008° 2θ step size, 6 min/measurement, scan range 12–13° 2θ.

Data were evaluated using PANalytical's X'Pert HighScore software for phase identification. For low amorphous content samples (Figure 2), the relative intensity between reflections and background (I_net/I_obs) was used for the evaluation. All scans were also evaluated by partial least squares regression (PLSR) using either baseline data or the baseline and reflections in the region of 10–35° 2θ using Unscrambler software (Camo ASA, Norway). For low crystalline content samples data were evaluated using the peak area of the 12.4° 2θ reflection and by PLSR of the recorded scan.

Figure 2: XRPD patterns of crystalline lactose with varying amounts of amorphous lactose.

NIRS data were acquired using a Bruker Optics Vector 22/N FT NIR spectrometer (Bruker Optik GmbH, Germany) with fibre optics. The mean spectrum of 32 scans for each sample over a wave number region of 3800–12000 cm^-1 was recorded (taking 30 s for each measurement). The regression model for the samples was calculated by PLSR after data preprocessing by multiplicative scatter correction and first derivative. Low amorphous content samples were evaluated over wavelengths of 5342–4865 cm^-1, and low crystalline samples were evaluated over wavelengths of 5992–5693 cm^-1.

Key points

Results

For an increasing crystalline content (0–100%) at 10% intervals, calibration curves for both methods showed excellent linearity with a correlation coefficient of better than 0.99.

Figure 3: Results of the cross validation for samples with low amorphous content (1–9%). (a) NIRS data after PLS regression; (b) XRPD data with classical evaluation of Inet / Iobs; and (c) XRPD data after PLS regression on the background of the XRPD scans.

Results for low amorphous content (1–9% amorphous) samples are shown in Figure 3 and Table 1. The classical linear regression gave a good correlation of 0.994 for both NIRS and XRPD. NIRS gave a root mean square error of prediction (RMSEP) value of 0.36 and a limit of detection (LOD) of 0.22%. With the standard I_net/I_obs method of analysis for XRPD the RMSEP was 0.52% and the DL 0.99%. With a multivariant evaluation (PLSR method) of the whole XRPD diffraction pattern in the 10–35° 2θ range, the RMSEP could be reduced to 0.38% and the DL to 0.63. Where the background only was used for the multivariant evaluation, the model simplified to two principal components and RMSEP and LOD could be further improved to 0.33 and 0.30%, respectively.

Table 1: Comparison of reliability parameters for NIRS and XRPD measurements on the samples with low amorphous content.

Results for low crystalline content (0–10%) samples are shown in Figure 4 and Table 2. In contrast to the results for low amorphous content, those for NIRS were not as impressive as those for XRPD. With NIRS, linear regression had a correlation of 0.990, a RMSEP value of 0.540 and a DL of 0.82%. Using the peak area of the 12.4° 2θ reflection, XRPD had a correlation of 0.999, a RMSEP value of 0.30 and a DL of 0. 50%. Further evaluation using PLSR reduced the RMSEP to 0.14 and the DL of crystalline material to 0.30%.

Figure 4: Results of the cross validation for samples with low crystalline content (0–10%). (a) NIRS data after PLS regression; (b) XRPD data using the peak area under the reflection at 12.4° 2Theta; and (c) XRPD data after PLS regression using the scan between 12° and 13° 2Theta.

Discussion

Both XRPD and NIRS proved accurate and precise, with DLs below 1%.

Table 2: Comparison of reliability parameters for NIRS and XRPD measurements on the samples with low crystalline content.

For samples with a low amorphous content (1–9%), XRPD gave data of a similar quality to NIRS. XRPD showed a slightly smaller method error (RMSEP), especially when the background information was used, but it also showed a higher variation within the data sets.

For samples with a low crystalline content, XRPD was more accurate and precise, and had lower DLs than NIRS. This can be explained by the lower sensitivity of XRPD to the presence of water. XRPD also has the advantage of offering a simple means of determining whether or not a sample has recrystallized and, if it has, which phase is involved. Taking all this into account, XRPD appears both qualitatively and quantitatively superior to NIR for monitoring crystallization and recrystallization processes.

Conclusion

XRPD and NIRS are powerful analytical techniques capable of correlating the physicochemical structure of pharmaceutical solids to observed stability and drug release profiles. Both techniques are characterized by high precision and accuracy, low DLs, nondestructive analysis, and rapid production of results. Despite slight differences in accuracy between the analysis of partially amorphous and partially crystalline samples of lactose, both methods were considered sufficiently accurate for the purposes of process control. With the use of PLSR for data analysis, XRPD appears both qualitatively and quantitatively superior to NIRS for samples with a low crystalline content and is, therefore, best suited to the monitoring of crystallization and recrystallization processes.

Detlef Beckers is market segment manager pharmaceuticals, food and life science at PANalytical B.V. He has responsibilities for applications and business development and the co-ordination of the pharmaceutical, food and life sciences market activities within the company.

Ingeborg Lehrach gained her PhD at the University of Bonn (Germany) in 2004. She now works at research-based pharmaceutical company Grünenthal GmbH in Aachen (Germany).

Klaus-Jürgen Steffens is professor of pharmaceutical technology at the University of Bonn. Following studies, PhD and habilitation at Marburg University (Germany), he was appointed as professor of pharmaceutical technology — first at Braunschweig University (Germany) and now in Bonn.

References

1. P. York, Int. J. Pharm., 14(1), 1–28 (1983).

2. B.A. Hendrikson, Int. J. Pharm., 60(3), 243–252 (1990).

3. B.C. Hancock and G. Zografi, J. Pharm. Sci. 86(1), 1–12 (1997).

4. A. Saleki-Gerhardt, C. Ahlneck and G. Zografi, Int. J. Pharm. 101(3), 237–247 (1994).

5. L.S. Taylor and G. Zografi, Pharm. Res., 15(5), 755–761 (1998).

6. A. Gombás et al., Int. J. Pharm., 256(1), 25–32 (2003).

7. I. Fix and K.J. Steffens, Drug Dev. Ind. Pharm., 30(5), 513–523 (2004).