1 In situ monitoring of powder blending by non-invasive Raman 2 spectrometry with wide area illumination 3 Pamela Allan,a Luke J. Bellamy,a,1 Alison Nordon,a* David Littlejohn,a* John Andrewsb 4 and Paul Dallinb 5 a WestCHEM, Department of Pure and Applied Chemistry and CPACT, University of 6 Strathclyde, 295 Cathedral Street, Glasgow, G1 1XL, UK 7 b Clairet Scientific Ltd., 17/18 Scirocco Close, Moulton Park Industrial Estate, 8 Northampton, NN3 6AP, UK 9 1 Present address: GlaxoSmithKline, Priory Street, Ware, SG12 0DJ, UK. 10 11 * denotes authors to whom correspondence should be sent 12 (Email: [email protected] and [email protected]) 13 14 Abstract 15 A 785 nm diode laser and probe with a 6 mm spot size were used to obtain spectra of 16 stationary powders and powders mixing at 50 rpm in a high shear convective blender. 17 Two methods of assessing the effect of particle characteristics on the Raman sampling 18 depth for microcrystalline cellulose (Avicel), aspirin or sodium nitrate were compared: (i) 19 the information depth, based on the diminishing Raman signal of TiO in a reference 2 20 plate as the depth of powder prior to the plate was increased, and (ii) the depth at which a 21 sample became infinitely thick, based on the depth of powder at which the Raman signal This is an accepted author manuscript, accepted by Journal of Pharmaceutical and Biomedical Analysis, 04 December 2012. 1 22 of the compound became constant. The particle size, shape, density and/or light 23 absorption capability of the compounds were shown to affect the “information” and 24 “infinitely thick” depths of individual compounds. However, when different sized 25 fractions of aspirin were added to Avicel as the main component, the depth values of 26 aspirin were the same and matched that of the Avicel: 1.7 mm for the “information” 27 depth and 3.5 mm for the “infinitely thick” depth. This latter value was considered to be 28 the minimum Raman sampling depth when monitoring the addition of aspirin to Avicel in 29 the blender. Mixing profiles for aspirin were obtained non-invasively through the glass 30 wall of the vessel and could be used to assess how the aspirin blended into the main 31 component, identify the end point of the mixing process (which varied with the particle 32 size of the aspirin), and determine the concentration of aspirin in real time. The Raman 33 procedure was compared to two other non-invasive monitoring techniques, near infrared 34 (NIR) spectrometry and broadband acoustic emission spectrometry. The features of the 35 mixing profiles generated by the three techniques were similar for addition of aspirin to 36 Avicel. Although Raman was less sensitive than NIR spectrometry, Raman allowed 37 compound specific mixing profiles to be generated by studying the mixing behaviour of 38 an aspirin – aspartame – Avicel mixture. 39 < Take in Figure 1> 40 Keywords 41 Raman spectrometry; process analytical technologies (PAT); powder blending; sampling 42 depth; real-time monitoring; pharmaceuticals. 43 2 44 1. Introduction 45 Raman spectrometry is proving to be a useful monitoring technique in the pharmaceutical 46 industry, especially in secondary manufacturing [1, 2]. Considerable advantages have 47 been demonstrated for analysis of tablets [3-18] and capsules [11, 19-23], particularly 48 when transmission mode measurements were used [16-18, 20-24]. To ensure that 49 pharmaceutical dosage forms contain the appropriate amount of active ingredient(s), the 50 constituents must be blended to a homogeneous state. While transmission Raman 51 spectrometry is suited for the analysis of tablets and capsules, the backscatter mode of 52 measurement is more amenable for in situ analysis of larger unit operations such as 53 powder blending. However, there are relatively few reports describing the use of 54 backscatter Raman spectrometry for this purpose [25-27]; in contrast, use of in situ near 55 infrared (NIR) spectrometry is far more common [28-33]. The mixing of diltiazem 56 hydrochloride pellets and paraffinic wax was investigated by Vergote et al. using non- 57 invasive Raman spectrometry [26]. The process was monitored via a glass window in the 58 side of the vessel using a laser spot of approximately 2 mm diameter. There was no 59 significant difference in the intensity of the Raman signal when the diltiazem pellets were 60 stationary or mixing at 50 rpm. Therefore, spectra could be recorded without stopping the 61 mixing procedure and the Raman signal became constant when homogeneity had been 62 reached. De Beer et al. [25] monitored the blending of Avicel PH 102, lactose DCL 21, 63 dilitazem hydrochloride, and silicium dioxide using an invasive Raman probe; the end of 64 the probe was flush with the inner surface of the mixing vessel wall. The mixing end- 65 point identified from the Raman measurements was comparable to that obtained using 66 NIR spectrometry. Hausman et al. also reported the successful implementation of in-line 3 67 Raman spectrometry to monitor blending of azimilide dihydrochloride at low dose 68 (1% w/w) [27]. 69 One reason for the limited application of Raman spectrometry to powder 70 processes is that conventional backscatter Raman systems typically employ optics that 71 produce laser spot sizes smaller than 500 µm diameter, which results in the measurement 72 of only a small volume of the sample. More representative sampling has been achieved 73 by continuously rotating a sample during measurement [5, 6, 9, 10, 13, 14, 34] or by 74 scanning at several positions [8, 34, 35], thereby increasing the sampling area. To 75 overcome some of the sub-sampling limitations of conventional backscatter Raman 76 systems, defocused probes or wide area sampling optics with laser beam diameters of 3 – 77 7 mm have also been investigated [15, 36-38]. Backscatter Raman measurements also 78 exhibit a strong bias towards the upper surface layers of the sample. Monte Carlo 79 simulations of the backscattered Raman signal from a 4 mm thick tablet using a 4 mm 80 diameter laser spot have shown that 88% of the signal is generated in the top 1 mm layer 81 of the tablet [24]. However, the increased sampling depth and area of wide area 82 illumination probes results in significantly larger sampling volumes compared to 83 conventional backscatter Raman probes. For example, it has been shown that Raman 84 signals can originate from material 13 mm below the surface of a sample with a PhAT 85 probe with a 7 mm spot size [38]. The sampling volume of a PhAT probe (3 mm spot 86 size) was found to be approximately 1300 times larger than that of a non-contact Raman 87 probe (150 µm spot size) and approximately 16700 times larger than that of an 88 immersion probe (60 µm spot size) [36]. The larger sampling volume associated with the 89 PhAT probe resulted in comparable results to NIR spectrometry for quantification of 4 90 mixtures of anhydrous and hydrated polymorphs; the sampling volume of the PhAT 91 probe was found to be comparable to that for the NIR probe [39]. When mixtures of 92 polymorphs of flufenamic acid (forms I and III) with different particle sizes were 93 analysed using large and small spot size Raman probes, the wide area illumination probe 94 was found to be less sensitive to particle size owing to the larger sampling volume 95 measured with this probe [37]. 96 This report describes the first use of a PhAT probe for in situ Raman monitoring 97 of powder mixing in a high shear blender. The study included evaluation of the effect of 98 particle size variations on the sampling depth that can be achieved with the PhAT probe 99 to allow comparison with previously reported results obtained with probes of different 100 optical configurations. In addition, the effect of particle size on the Raman signal was 101 investigated; this was important as the effect of particle size on the backscatter Raman 102 signal is highly dependent on the optical configuration of the probe employed [40]. 103 Mixing profiles based on non-invasive Raman spectrometry have been assessed to 104 compare the information they provide with mixing profiles produced by non-invasive 105 NIR spectrometry and broadband acoustic emission (AE) spectrometry. Some advantages 106 of Raman measurements over NIR spectrometry were identified for powder monitoring. 107 5 108 2. Experimental 109 2.1. Instrumentation 110 2.1.1. Raman spectrometer 111 A Kaiser Raman RXN1 spectrometer with a PhAT probe (Kaiser Optical Systems Inc., 112 Ann Arbor. MI, USA) was used for all experiments except where stated below. The 113 785 nm Invictus diode laser was operated at 400 mW at source. The laser beam was 114 optically expanded to give a 6 mm spot size, a working distance of 254 mm and a depth 115 of field of 50 mm. The probe to sample distance was 203 mm. For the sampling depth 116 studies, spectra were acquired using an exposure time of 0.5 s and 1 accumulation. In situ 117 Raman measurements were made through the glass wall of the mixing vessel during the 118 powder blending experiments. A spectrum was acquired every 5 s with an exposure time 119 of 2.5 s and 1 accumulation. The exposure time for each set of experiments was selected 120 such that the largest Raman signal occupied approximately 60 – 70% of the dynamic 121 range of the CCD detector. 122 For the experiment comparing Raman, NIR and AE, and the monitoring of the 123 blending of a three component mixture, a different PhAT probe system was used. In this 124 case, the 785 nm laser beam was optically expanded to give a 3 mm spot size and a 125 working distance of 100 mm. The probe to sample distance was 100 mm. Spectra were 126 acquired every 3 s, through the wall of the vessel, with an exposure time of 2 s and 1 127 accumulation. 128 A spectrum of aspartame was acquired using a Kaiser RXN1 spectrometer with a 129 MR probe equipped with a non-contact optic (0.4 inch working distance and a laser spot 6 130 size of approximately 100 µm). An Invictus diode laser with a wavelength of 785 nm was 131 employed, and was operated at 350 mW at the source. Aspartame was contained within a 132 glass sample vial, which was placed in the off-line sample compartment, and the 133 spectrum was acquired through the side of the vial using an exposure time of 4.0 s and 1 134 accumulation. 135 Data were acquired using HoloGRAMS software (Kaiser Optical Systems). A 136 dark current spectrum was obtained before each set of experiments and subtracted 137 automatically from each subsequent spectrum acquired. Data were exported into 138 GRAMS/32 and converted to text files, which were subsequently imported into Matlab 139 7.0 (Mathworks Inc., Natick, Massachusetts, USA) for analysis using the PLS_Toolbox 140 version 3.04 (Eigenvector Research Inc., Manson, Washington, USA). As spectra of 141 samples containing Avicel exhibited a sloping baseline arising from fluorescence, second 142 derivative spectra were calculated throughout using the Savitzky – Golay function with a 143 second order polynomial and a 13 or 25 point filter width for data acquired using the 6 or 144 3 mm laser spot size, respectively. 145 2.1.2. Near infrared reflectance spectrometer 146 Some measurements of powder mixing were obtained simultaneously by non-invasive 147 Raman spectrometry, NIR spectrometry and broadband AE spectrometry. In situ NIR 148 measurements were made through the glass wall of the vessel as described previously 149 [31, 32] using a Zeiss Corona 45 NIR spectrometer (Carl Zeiss, Heidenheim, Germany). 150 Measurements were acquired using Aspect software (Carl Zeiss) package and spectra 151 stored as log(1/R) where R is the reflectance and was calculated from the intensity of 152 light reflected by the sample relative to that for a reflectance standard. The integration 7 153 time was 32 ms and 10 scans were co-added for each acquired spectrum allowing 154 measurements to be taken every 0.5 s. Data were exported as text files into Matlab for 155 further analysis using PLS_Toolbox. First derivative spectra were calculated using the 156 Savitzky – Golay function with a 5 point filter width and second order polynomial. 157 2.1.3. Acoustic emission 158 The broadband AE monitoring equipment has been described previously [41]. A Nano 30 159 transducer (Physical Acoustics Ltd, Cambridge, UK) was attached to the glass wall of the 160 mixing vessel using a silicone-based vacuum grease (Dow Corning) and adhesive tape. 161 The Nano 30 transducer was attached to a 2/4/6 series pre-amplifier (Physical Acoustics 162 Limited). The pre-amplifier required a 28 V power supply (Physical Acoustics Limited) 163 and the gain of the pre-amplifier was set to 60 dB. The pre-amplifier was connected to an 164 Agilent 54642A oscilloscope using a 5 m length cable, which was linked to a computer via a 165 GPIB to USB interface (Agilent Technologies). 166 A data capture program, written in C++ by Douglas McNab and Robbie Robinson 167 from the Centre of Ultrasonic Engineering (CUE) at the University of Strathclyde, 168 enabled the cyclic sampling of the acoustic signals displayed on the oscilloscope and 169 signals were saved as comma separated variable (CSV) files. Signals were acquired using 170 a sampling rate of 2 MHz and the time interval between collection of signals was 2 s. A 171 total of 450 signals were collected with each signal consisting of 4000 points. All 172 acoustic signals were imported into Matlab for analysis. A power spectrum was 173 calculated of each signal and three spectra were co-added to give a composite spectrum 174 every 6 s. Signal areas were calculated by summing the intensities of the signals between 175 0 and 400 kHz. 8 176 177 2.2. Powders 178 Microcrystalline cellulose (Avicel PH-101; FMC, Cork, Ireland), aspirin, sodium nitrate 179 (both from Sigma-Aldrich, Dorset, UK) and aspartame (provided by GSK, UK) were 180 used. The Avicel particles have a tap density of 0.45 g cm-3, are granular in shape and 181 have an average particle size of 50 μm. Aspirin particles are low aspect ratio needles with 182 an average particle size of 192 µm. Sodium nitrate particles are granular with an average 183 particle size of 275 µm. Aspartame particles are high aspect needles (particle size was not 184 measured). Average particle size information was obtained by laser diffraction, while 185 density and shape information was obtained from the literature [42]. The powders were 186 sieved through 10 cm diameter brass pan sieves (Endecotts Ltd, UK) to obtain different 187 particle size ranges: <38, 38 – 53, 53 – 106, 150 – 212, 212 – 250, 250 – 300, 300 – 355 188 and 355 – 425 µm for Avicel; <106, 106 – 150, 212 – 250, 250 – 300, 300 – 355, and 425 189 – 500 µm for aspirin; 150 – 212, 212 – 250, 250 – 300, 300 – 355, 355 – 425, 425 – 500 190 and 500 – 800 µm for sodium nitrate. The mid-point of the sieve range has been used in 191 plots to represent each particle size fraction. 192 193 2.3. Procedures 194 2.3.1. Raman sampling depth studies 195 A 2.68 mm thick layer of glass was placed on top of the inverted PhAT probe with a 196 203 mm spacer attached (Figure 1a). A series of plastic plates, with a 30 mm diameter 197 hole, were placed on top of the glass plate (Figures 1b and 1c). For investigation of 9 198 powder depths of 0 – 4.48 mm, up to 16 × 0.28 mm thick plates were stacked on top of 199 each other. A depth of 8.48 mm was obtained by placing an additional 4 mm thick plastic 200 plate on top of the 16 plates. This combination of plates enabled sampling depths of 201 ≤4.48 mm to be measured. The powder was carefully placed into the hole in the plastic 202 plates and levelled off using a razor blade, ensuring minimal compaction of the solid, and 203 the mass of solid was recorded. A TiO reference layer (solid particles of TiO sealed 2 2 204 between two microscope slides) was placed over the top layer of the powder (Figures 1b 205 and 1c) with aluminium foil used as the final layer to reduce loss of laser and Raman 206 photons from the upper surface of the TiO reference layer. Each corner of the 2 207 microscope slide was labelled to ensure consistency in positioning, so that the same part 208 of the TiO was analysed each time. A spirit level was placed on top of the glass above 2 209 the spectrometer to ensure the radiation was pointing vertically upwards (see Figure 1a). 210 Analysis of the variance of the Avicel signal at 1095 cm-1 revealed that the repeatability 211 (expressed as the relative standard deviation (RSD) for n = 6) of the Raman 212 measurements, positioning of the sample holder, the weighing procedure and the powder 213 levelling process were 0.22, 0.34, 0.66, and 0.26%, respectively. 214 The information depth was estimated using the TiO Raman signal at 397 cm-1. As 2 215 the depth of Avicel, aspirin or sodium nitrate increased, the TiO signal decreased until it 2 216 became zero; hence, the information depth was defined as the point where the exciting 217 laser can no longer penetrate the powder to generate a detectable Raman signal of the 218 TiO reference layer. This type of approach has been used previously for information 2 219 depth measurements in NIR reflectance spectrometry [31, 43, 44]. A second set of 220 measurements was made based on the change in the intensity of the Raman signals of 10