934 Volume 59, Number 7, 2005 APPLIED SPECTROSCOPY0003-7028 / 05 / 5907-0934$2.00 / 0q 2005 Society for Applied Spectroscopy

spectroscopic techniques

Comparison of Sampling Techniques for In-Line MonitoringUsing Raman Spectroscopy

HAKAN WIKSTROM, IAN R. LEWIS, and LYNNE S. TAYLOR*Department of Industrial and Physical Pharmacy, School of Pharmacy, Purdue University, West Lafayette, Indiana 47907 (H.W.,L.S.T.); and Kaiser Optical Systems, Inc., 371 Parkland Plaza, Ann Arbor, Michigan 48103 (I.R.L.)

Raman spectroscopy is currently of interest as a process monitoringtool for pharmaceutical unit operations. In this study, the perfor-mance characteristics of Raman spectrometers with different sam-pling optics have been investigated in the context of process moni-toring, with emphasis being placed on assessing homogeneity inpowder blends and following changes in solid-state form during wetgranulation. A novel large spot non-contact Raman sampling devicewas compared with a traditional small spot size non-contact sam-pling device and an immersion probe. The large spot non-contactoptics provided significant advantages over the standard systemsboth as a result of the enhanced sampling volume and because ofthe greater robustness of the system to fluctuations in the samplingdistance during the wet granulation process.

Index Headings: Sampling; In-line monitoring; Process inducedtransformations; High shear wet granulation; Raman spectroscopy.

INTRODUCTION

As analytical instrumentation advances in sophistica-tion, there is growing interest in interfacing such instru-ments to pharmaceutical unit operations with the aim ofenhancing process understanding and ultimately monitor-ing and controlling the process.1 Optical spectroscopictechniques are especially suitable for in-line process mon-itoring because remote sampling can be achieved bytransmitting radiation to and from the sample via fiber-optic cables.2 For example, near-infrared spectroscopy(NIR) is now widely used as an in-line technique forpharmaceutical processes, in particular to extract infor-mation about moisture content.3–5 However, until recently,Raman spectroscopy has mostly been used for monitoringother industrial processes, for example, in the polymerindustry6 and during manufacturing of inorganic materi-al7 even though solid-state active pharmaceutical ingre-dients (APIs) are generally good Raman scatterers. Fur-thermore, Raman spectroscopy can provide certain ad-vantages due to the fact that water is a poor Raman scat-terer8 making it possible to study changes in theproperties of API in water-rich environments, which arefrequently encountered during pharmaceutical unit oper-ations such as wet granulation. Due to the limited inten-sity of the Raman scattering, the exciting radiation is gen-

Received 9 February 2005; accepted 23 March 2005.* Author to whom correspondence should be sent.

erally focused into a high intensity cone of light in orderto facilitate the collection and detection of sufficient ine-lastically scattered photons.9 Consequently, for staticsamples, only the small portion of the sample that inter-acts with the focused beam is sampled. For many com-mercial systems, the incident radiation is typically fo-cused to a spot size of less than 500 mm in diameter. Thiscan be a significant disadvantage in process monitoring,since such focusing can lead to two significant problems,i.e., sub-sampling and loss of signal intensity if the sam-ple moves out of the focus of the incident radiation. Forexample, if blend homogeneity is of interest, a 100 mmspot must travel over a considerable area before a rep-resentative sample volume can be measured in a binarymixture composed of 200 mm particles, and any move-ment of the sample relative to the sampling device mustbe such that a reasonable interaction of the material withthe incident radiation is maintained. In addition, since theRaman effect is weak, an integration time of at least fiveseconds is typically required in order to obtain a Ramanspectrum of sufficient quality. Consequently, althoughRaman spectroscopy has been used to monitor pharma-ceutical processes where it is possible to maintain a goodcontact with the sample, such as chemical reaction10 andcrystallization monitoring,11 there are limited reports ofthe use of this technique for measuring processes that arespatially more dynamic and heterogeneous.12

Recently, Wikstrom et al. demonstrated that Ramanspectroscopy could be used for in-line monitoring of an-hydrate to hydrate phase transformations during highshear wet granulation using an immersion probe coupledto fiber optics to interface the radiation with the sample.13

For the immersion optics utilized in this study, the irra-diating radiation is focused close to the window sealingthe probe. The probe was placed into the granulation bedand material that was swept past the probe was analyzed.However, the authors noted that there were a number ofproblems associated with this approach, including the po-tential for the material to stick to the probe window, andit was concluded that invasive probes could not be con-sidered optimum for monitoring these types of pharma-ceutical processes.

Having previously established the utility of Ramanspectroscopy for obtaining information about phase trans-

APPLIED SPECTROSCOPY 935

formations during wet granulation, the objective of thecurrent study was to evaluate different sampling devicesfor in-line monitoring. The immersion optics used pre-viously was compared with non-contact optics. In addi-tion, the importance of sampling volume and depth offocus was investigated by comparing a sampling devicewith a standard spot size and focusing optics with a largespot size system.

EXPERIMENTAL

Several experiments were designed to determine theoptimum Raman sampling approach for bulk analysis ofpharmaceutical processes. The experimental sub-set in-cluded studies of the effect of sub-sampling and estima-tions of the sampling volume and focus tolerance for eachsampling methodology. Finally, wet granulation experi-ments were performed, where the results of the afore-mentioned experiments were used to interpret the differ-ences observed when using the different sampling meth-odologies.

Materials. Theophylline anhydrous (AT) was obtainedfrom Rhodia (Cranbury, NJ) and placed in an oven at100 8C for at least one hour prior to use to ensure thatno monohydrate was present and was used immediatelyafter cooling. Theophylline monohydrate (MT) was pre-pared by placing AT in a desiccator with 100% relativehumidity (RH) for two weeks prior to use to ensure thatno anhydrate was present. Avicel-PH-101 microcrystal-line cellulose (MCC) was obtained from FMC Corpora-tion (Newark, DE). Mannitol was obtained from RugerChemical Company (Irvington, NJ) and screened througha 20-mesh screen to reduce agglomeration before use.Polyvinyl pyrrolidone K-29/32 (PVP) was obtained fromISP Technologies, Inc. (Wayne, NJ). The binder solutionwas prepared by slowly adding 100 g of PVP to 800 mLof double-distilled water.

Instrumention. Raman spectra were collected with785 nm excitation using two RamanRxn1-785 Ramanspectrometers (Kaiser Optical Systems, Inc., Ann Arbor,MI) and a prototype PhAT System (Kaiser Optical Sys-tems, Inc., Ann Arbor, MI). The first RamanRxn1 wasequipped with a ¼-in. diameter immersion optic (KaiserOptical Systems, Inc., Ann Arbor, MI) attached to an MRProbe (Kaiser Optical Systems, Inc., Ann Arbor, MI),yielding a spot size at the sample of 60 mm. The secondRamanRxn1 was identical to the first but instead of at-taching an immersion optic to the MR Probe the MRProbe was equipped with a nominal 63.5 mm (2.5 in.)working distance non-contact optics (NCO, Kaiser Op-tical Systems, Inc., Ann Arbor, MI). This sampling ar-rangement led to a spot size at the sample of 150 mm.The laser power at the sample for both optics was mea-sured to be around 100 mW using a Coherent LaserCheckpower meter (Auburn, CA). The third system, the pro-totype 785 nm PhAT System Raman spectrometer (KaiserOptical Systems, Inc., Ann Arbor, MI), possesses a non-contact optic sampling device (Kaiser Optical Systems,Inc., Ann Arbor, MI). The nominal working distance ofthis sampling device was 100 mm (with a 5 cm spacerattached to the probe head), with a spot size of about 3mm. The laser power at the sample was comparable withthe other systems. All of the systems used fiber optics to

interface the spectrometer to the sampling device, en-abling remote measurements.

Software. HoloGRAMS software (version 4.0, KaiserOptical Systems, Inc., Ann Arbor, MI) was used to con-trol the Raman spectrometer. Excel 2000 (build 9.0.2720,Microsoft Corporation, Seattle, WA) was used for cali-bration calculations and graph plotting. SIMCA-P1 (ver-sion 10.5, Umetrics AB, Umea, Sweden) was used forprincipal components analysis (PCA), and Sigma Plot(version 8.02, SSPS, Inc., Chicago, IL) was used forcurve fitting.

Sub-sampling. Sub-sampling was defined as misrep-resentation of the spectrum of a mixture due to inade-quate sampling volume. In order to test sub-sampling,powder blends were prepared by geometrical dilution ofAT and MT in 20% (mol/mol) increments starting at 10:90 MT/AT (mol/mol) to 90:10 MT/AT (mol/mol). In ad-dition, pure samples of both forms were measured. Thesamples analyzed by the small spot sampling deviceswere only analyzed using the immersion optic since pre-liminary results indicated that it was not sufficient to onlysample from the top layer of the sample due to severesub-sampling problems. Two approaches were used to an-alyze the samples using the immersion optic. Firstly, theimmersion optic was manually moved around in the sam-ple for the entire acquisition time, with 2 s integrationtime and 64 accumulations. Secondly, in order to furtherincrease the volume of material sampled, a dynamic pow-der-mixing device was constructed. This device consistedof an electrical motor that rotated the sample vial. Theimmersion probe was fitted with a blade and inserted intothe sample bed. An additional blade was inserted into thevial, positioned so that it cleaned the sides of the vialfrom sticking material. The blade fitted to the immersionprobe was positioned so that it moved material from thebottom of the vial. The acquisition time used for thissetup was over 4 min, with 2 s integration time and 128accumulations. For the PhAT System, samples were pre-pared as above and measured both stationary and withthe samples rotating. No additional mixing was necessaryto obtain repeatable measurements with the PhAT Sys-tem. The acquisition time for the PhAT System was justunder ten seconds, with an integration time of 1 s andeight accumulations.

Estimation of Sampling Volume. AT compacts of dif-ferent thickness were compressed using an IR press (In-ternational Crystal Laboratories, Inc., Garfield, NJ) witha 13 mm die and an applied force of 2000 psi with adwell time of 5 s. The compact thickness was determinedusing a micrometer (Mitutoyo America Corp., Aurora,IL). A 525 mm thick silicon wafer with an etched back-side (WaferNet GmbH, Eching, Germany) was placed un-derneath the compact and the maximum penetration wasestimated by determination of the thickness of the thin-nest compact that gave the maximum peak height of the1707 cm21 peak from anhydrous theophylline. Spot sizeestimates were provided by the instrument manufacturerand were reported to be 60 mm for the immersion optics,150 mm for the standard NCO, and 3 mm for the PhATSystem. The apparent sampling volume was then esti-mated by multiplying the maximum penetration depthwith the area of illumination, thus assuming the samplevolume to be cylindrical.

936 Volume 59, Number 7, 2005

TABLE I. Batch information and experimental parameters for thewet granulation experiments.

Amount of theophylline anhydrousAmount of MCCAmount of mannitolAmount of PVP (added in aqueous solution)Amount of water

90 g105 g105 g

14.5 g115 g

Mixing speed (tip speed)Chopper speed

100 rpm (0.9 m/s)1200 rpm

Dry mixing timeBinder solution addition timeWet massing time

120 s37 s

600 s

FIG. 1. Positioning of sampling devices relative to chopper in the gran-ulation bowl for simultaneous monitoring of the wet granulation process.

Evaluation of Focus Tolerance. An assessment of theeffect of sampling distance on signal intensity was per-formed by analyzing a sample of anhydrous theophyllinepowder at different distances from the tip of the samplingdevice. Three measurements were carried out at eachsampling distance and the optimum sampling distancewas revisited three times. The integration time was 5 swith one accumulation taken for each Raman system. Theaverage peak height of the peak at 1707 cm21 was usedto determine the sampling efficiency. The baseline at1707 cm21 was interpolated based on a linear fit of thebaseline between 1740 cm21 and 1780 cm21.

Wet Granulation. Dry material was weighed andplaced in a Diosna P 1/6 high shear mixer-granulator(Dierks & Sohne GmbH, Osnabruck, Germany) equippedwith a 2 L stainless steel bowl. All granulations werecarried out according to Table I, unless otherwise noted.The binder solution was sprayed onto the mix using aMasterflex Quick Load Model 7021-24 pump (Cole-Palmer Instrument Co., IL), and the amount of bindersolution added was determined gravimetrically using aMettler PC 8000 balance (Mettler Toledo, Inc., Hights-town, NJ).

To compare the three sampling devices, several gran-ulation experiments were performed and monitored si-multaneously by all sampling devices. With the chopperof the granulation apparatus at 12 o’clock, the granulationliquid was added at 10 o’clock. The sampling device at-tached to the probe of the PhAT System was situatedabout 100 mm above the bed at about 5 o’clock. The 2.5-in. NCO sampling device was positioned about 6 cmabove the bed at about 7 o’clock and the immersion opticwas immersed at about 3 o’clock in a 458 angle facingthe movement of the bed. This setup is illustrated in Fig.1. The transformation of the initial anhydrous theophyl-line to the monohydrate form during the wet granulationprocedure was found to be insensitive to the positioningof the sampling devices, as verified by monitoring thetransformation with individual sampling devices placedin different positions from those described above. Bothof the NCO sampling devices were fitted with a thin poly-ethylene film to protect the optics against rising powderfouling from the granulation bed. All sampling deviceswere verified to be free from interference from the othertwo before the granulation experiments were started. Datacollection on the system attached to the immersion opticwas started 90 s before the mixing started, while the 2.5-in. NCO sampling device and PhAT System were started60 and 30 s before the onset of mixing, respectively. The

final spectrum for each system was acquired once thegranulation had been stopped.

Each spectrum was the average of four scans, eachwith an integration time of 5 s for the RamanRxn1 sys-tems and an integration time of 1 s per scan for the PhATSystem. Spectra were obtained every 30 s over the du-ration of the wet granulation experiment for the standardsystems and every 10 s for the PhAT System. The im-mersion optic was placed in the mixing bowl just abovethe impeller and angled towards the movement of the bedto reduce the risk of sample adhesion.

RESULTS AND DISCUSSION

Sub-sampling. Sub-sampling was assessed by measur-ing different blends of AT and MT of known composition.The data were analyzed by PCA in order to determine theeffect of sampling methodology on the observed samplehomogeneity and consequently the ability to determine theconcentration of each solid-state form. The data matrixwas pretreated using a standard normal variate (SNV)transformation and then mean centered before the PCAmodels were determined. The significant principal com-ponents from a principal component analysis of mixturesat different weight ratios should be related to spectralchanges occurring due to composition variations. In thiscase, the first principal component (which models thegreatest variation in the system) was found to resemble thespectral difference between monohydrate and anhydroustheophylline (Fig. 2a), as expected and in agreement withprevious reports.14 Since the PCA was performed on SNVtransformed data, the aforementioned spectral differencewas calculated on appropriately weighted spectra to ac-count for the intensity differences between the two com-pounds. The second principal component resembles thespectrum for a weighted 50:50 (mol/mol) mixture, i.e., thesum of the two compounds weighted so that they are onthe same scale (Fig. 2b). This principal component is in-terpreted as a correction for nonlinearity effects due to theintensity difference in the spectra between the anhydrousand monohydrate theophylline and an effect of the SNV

APPLIED SPECTROSCOPY 937

FIG. 2. Loading interpretation of PCA models describing mixing: (a)First principal component loadings compared with the weighted spectraldifference of the two forms; (b) second principal component loadingscompared with the weighted spectral sum of the two forms.

FIG. 3. Sub-sampling evaluation based on score maps of homogeneitydetermination using four different sampling methodologies. Ideal mix-ing is a parabolic fit in principal component space: (a) manual move-ment of immersion optic in sample during spectral collection (64 3 2s acquisition time); (b) automatic mixing throughout spectral collectionusing immersion optic (128 3 2 s acquisition time); (c) spectra collectedfrom stationary sample using the PhAT System (8 3 1 s acquisitiontime); and (d) spectra collected from rotating sample using the PhATSystem (8 3 1 s acquisition time).

transformation. All models explained more than 99% ofthe variation in the data, with the first PC explaining about98% in all four cases.

The PCA model generated from data collected using theRamanRxn1 system is shown in Fig. 3a, where the im-mersion probe was manually moved in the powder blendsin order to increase the volume of powder analyzed. Thesecond model (Fig. 3b) was also performed with theRamanRxn1 system but with dynamic mixing as describedabove and double the collection time. The third and fourthPCA models (Figs. 3c and 3d, respectively) used data col-lected with the PhAT System on stationary and rotatingsamples, respectively. It should also be noted that the spec-tra for these models were collected in less then a tenth ofthe time of the first model. Each plot in Fig. 3 shows thescore values for the first and second principal componentsof the model, whereby each spectrum yields a score value.Score values distant in space are expected for samples ofdifferent compositions, and for an ideal range of mixturesthe score values should lie on a curved line starting at thescore values representing pure component A and endingat the score values representing pure component B in theprincipal component space. The score values for a mixtureof A and B should place the mixture on the line connectingthe pure components, and the position on that line is de-

pendent on the observed composition of A and B in themixture. Score values that fall outside this line are notrepresentative of perfect mixtures, which in this case canbe attributed to sub-sampling. The advantage of this ap-proach as opposed to an ordinary calibration is that theresults are not affected by variations in the dependent var-iable, i.e., the estimated ratio of the two components in

938 Volume 59, Number 7, 2005

the mixture, and thus only variations between individualspectra are being analyzed, enabling the effect of sub-sam-pling to be more clearly seen.

Firstly, it is apparent from comparing Figs. 3a and 3bthat the observed homogeneity is significantly improvedby increasing the sampling time and introducing constantmixing during the spectral collection for the standardRamanRxn1 system. Thus, the points derived from repeatmeasurements on the same composition form much tight-er clusters in Fig. 3b and are closer to the ideal line. Ascan be seen from Fig. 3c, the observed homogeneity issignificantly improved by utilizing the larger spot size ofthe PhAT System despite the reduced sampling time. Theobserved homogeneity could be further improved by ro-tating the samples, thereby increasing the volume of sam-ple exposed to the sampling device, as seen from Fig.3d, where the majority of the points now fall on the idealline and are more tightly clustered in the principal com-ponent space.

Estimation of Sampling Volume. In order to explainthe results observed above, an assessment of the samplingvolume for each of the probes was made. In order toapproximate the sampling volume for the experimentalsystem employed, several compacts of anhydrous the-ophylline were prepared with different thicknesses. Thesecompacts were analyzed as described above, using thepeak height of the peak at 1707 cm21 to approximate thedepth of penetration. The penetration depth was definedas the thickness when no change in peak height could beobserved. In this study, the penetration depths were de-termined to be about 300 mm for the immersion optic,about 600 mm for the standard 2.5-in. NCO, and about2000 mm for the PhAT System. The latter is in agreementwith the penetration depth for an equivalent PhAT Sys-tem that was previously determined to be approximately1700 mm.15 This penetration depth was determined ac-cording to the methodology previously described byClarke et al.16

The spot size of the three sampling devices is 60 mm,150 mm, and 3000 mm, respectively, as provided by theinstrument manufacturer. Using these numbers, the ap-parent sampling volume for each system could be esti-mated as approximately 0.85 nL, 10.6 nL, and 14.1 mLfor the immersion optics, standard 2.5-in. NCO, andPhAT System, respectively. Thus, the sampling volumeof the PhAT System is approximately 1300 times largerthan the standard 2.5-in. NCO and approximately 16 700times larger than the immersion optics. The dramatic in-crease in the sampling volume with the PhAT Systemexplains the results shown in Fig. 3. Because of the mag-nitude of difference in sampling volume between theprobes, increasing the volume of sample measured byusing dynamic mixing and an increased analysis timecannot result in the same reduction in sub-sampling ten-dency. Obviously, the importance of sub-sampling willbe system dependent and will depend on the domain sizeof heterogeneous regions relative to the volume sampledby the sampling device.

Evaluation of Focus Tolerance. Both the focal posi-tion and the depth of focus is quite different for the threesampling devices and thus any movement of the sampleduring process monitoring will have different effects onspectrum quality for each probe. For the immersion op-

tics, the light is focused at the tip face of the sapphirewindow at the end of the optic, whereas for the standard2.5-in. NCO the focus is around 64 mm from the end ofthe sampling device. The PhAT System has optimal col-lection efficiency 100 mm from the end of the optic win-dow on the 2-in. spacer. For the pharmaceutical processunder consideration, it was of interest to perform an as-sessment of the magnitude of sample movement duringgranulation, i.e., fluctuations in the height of the powderbed due to application of rotational mixing and additionof the granulation liquid, and the response of each probeto changes in sample position.

Bed height variations throughout the granulation ex-periments were examined. Initially, the dry powder loadleft a 7.5 cm head space in the granulator bowl whenevened out prior to mixing. This meant that the immer-sion optic was placed into the bed, while both the NCOsampling devices could be positioned outside of the bowl.During the dry mixing stage, the bed reached as far upas 4 cm from the top of the bowl and as far down as 8cm from the bowl edge, thus, a total vertical movementof approximately 4 cm. During the wet massing stage,i.e., after water has been added to the powder with con-stant mixing, the maximum distance from the granules tothe bowl edge was determined to be 8 cm and the min-imum distance was found to be 5 cm, thus, a verticalmovement of approximately 3 cm. Once the granulationwas complete and the granules were evened out the dis-tance from the top edge of the bowl to the granule bedwas 7 cm.

The results of the focusing tolerance determination forthe three sampling techniques are shown in Fig. 4. Asexpected, the optimum sampling distance of the immer-sion optics is in absolute contact with the probe tip, with-in the resolution of our experiment, as shown in Fig. 4a.The peak height dropped to 45% of the optimum valueat a distance of 8.2 mm from the probe tip, as determinedby fitting the peak height data shown in Fig. 4a with adouble exponential curve fit (Eq. 1):

H 5 Ae2ax 1 Be2bx (1)

where H is the peak height for the peak at 1707 cm21, Aand B are y-axis intercepts, a and b are rate constants,and x is the sampling distance in centimeters. During thewet granulation, it is unlikely that the granules will be inintimate contact with the probe window for the entiresampling period as material is constantly being sweptpast the probe tip, which will reduce the signal-to-noiseratio and require a longer integration time to acquire areasonable spectrum. This observation, in conjunctionwith the observed sticking of material to the probe win-dow (discussed below), makes the immersion optics apoor choice for granulation monitoring. The immersionoptic is much better suited for reaction monitoring in so-lutions where the appropriate contact with the sample iseasy to maintain.

The optimum sampling distance for the standard 2.5-in. NCO was found to be 76.6 mm (Fig. 4b), rather thanthe expected 64 mm, as determined by curve fitting thepeak intensity data to a Lorentzian peak fit:

AH 5 H 1 (2)0 2x 2 x01 1 1 2a

APPLIED SPECTROSCOPY 939

FIG. 4. Focus tolerance for the three sampling devices (changes in bedheight during a wet granulation experiment is illustrated by dashed bar).The intensity at 1707 cm21 is plotted against the sampling distance: (a)immersion optics; (b) 2.5-in. NCO; and (c) PhAT System.

TABLE II. Effect of sampling on observed transformation kinetics.Experiments carried out according to Table I.

Immersion optics

Mid [min] % RSD

2.5-in. NCO

Mid [min] % RSD

PhAT System

Mid [min] % RSD

3.60 22.35 2.91 3.84 3.18 3.70

k [min21] % RSD k [min21] % RSD k [min21] % RSD

5.15 30.53 6.08 12.26 5.55 9.72

where H is the peak height for the peak at 1707 cm21,H0 is the minimum peak height observed, A is the max-imum peak height increase, a is the rate constant, x0 isthe optimum sampling distance in centimeters, and x isthe sampling distance in centimeters. The peak heightdropped to 45% of the optimum value 11 mm from theoptimal working distance as determined by curve fit tothe Lorentzian peak.

The effect of sampling distance for the PhAT System

can be seen in Fig. 4c. The optimum signal was deter-mined to be around 18 cm from the end of the samplingdevice; however, at no time was the signal less than 45%of the optimum signal. In the context of the granulationprocess under investigation, at a distance of 62 cm fromthe optimum distance, the signal will display a minimaldependence on distance as can be seen from the plateauregion of Fig. 4c. This indicates that the PhAT System isquite insensitive to the changes in powder bed height thatoccur during granulation and is thus likely to be the bestchoice for monitoring of the granulation process.

Wet Granulation. During aqueous wet granulation,anhydrous theophylline undergoes a solvent-mediatedphase conversion to the monohydrate form.17 This trans-formation occurs sufficiently rapidly to make off-linemonitoring impracticable as a method of probing the ki-netics of this process. It is therefore of interest to comparethe ability of the different Raman systems to quantita-tively follow the kinetics of this transformation. In orderto determine the amount of theophylline transformed ateach time point during the wet granulation experimentsit was necessary to use calibration models. For the stan-dard spot size systems, a previously described calibrationcurve was used,13 and for the PhAT System the rotatedhomogeneity samples (Fig. 3d) were used to construct abivariate calibration as previously described.13,18 The ob-served versus predicted plots of the bivariate calibrationbased on the PhAT System data was found to be signif-icantly better than the previously published calibration,with an R2 value for the PhAT System calibration of0.994 compared to an R2 value of 0.982 for the standardRamanRxn1 system. Both calibrations are considered ad-equate for the determination of the amount of theophyl-line transformed at different time points in the course ofthe wet granulation experiments.

A repeatability test of the granulation process and thesampling devices was performed by running three repeatgranulations using the exact conditions described in TableI. As a method of comparing the different sampling de-vices, data were fit to the empirical Eq. 3 in order toremove differences in the sampling times:

Max 2 MinC 5 Min 1 (3)MT 2kt

1 1 1 2Mid

where CMT is the fraction of monohydrate, Min is theinitial level of CMT (i.e., 0), Max is the final level of CMT

(i.e., 1), Mid is the time when 50% has been transformed,k is the rate of transformation, and t is time in minutes.The mean midpoint (n 5 3) and the rate of transformationare summarized in Table II, as well as the relative stan-dard deviation of the values.

940 Volume 59, Number 7, 2005

FIG. 5. Monitoring the transformation of theophylline anhydrous to mono-hydrate during wet granulation. Percent monohydrate formed plotted versustime. (a) Repeatability of transformation profile during wet granulation oftheophylline anhydrous as observed using data collected by the PhAT Sys-tem. (b) Effect of sticking on observed transformation kinetics of the con-version of anhydrous theophylline when using immersion optics during wetgranulation. Experiments carried out according to Table I.

FIG. 6. Effect of mixing speed on the transformation kinetics of theconversion of anhydrous theophylline during wet granulation. (a) Per-cent monohydrate formed plotted versus mixing speed (tip speed). (b)Relationship between both tip speed (v) and rate of transformation (k)and inverse of onset time (1/t0) for the conversion of anhydrous the-ophylline during wet granulation. The linear fit to the inverse onset time(1/t0 5 0.4271v 1 0.4605) had an R2 value of 0.991 and the linear fitof the rate constant (k 5 0.2245v 1 0.1592) had an R2 value of 0.996.The PhAT System was used for monitoring of the transformation andthe experiments were carried out according to Table I, with the excep-tion of the changing mixing speed.As can be seen from Table II, the time to the midpoint

is similar for the PhAT System and the standard 2.5-in.NCO, but it is slightly delayed for the immersion optic.Additionally, the repeatability of the midpoint values aresimilar for the two NCO systems, but significantly dif-ferent for the immersion optic system. These two differ-ences are believed to be due to sticking of material to theprobe, making an accurate determination of the midpointdifficult. However, the determination of the rate constantis quite similar for all three systems, although the varia-tion is significantly decreased for the NCO systems ascompared with the immersion optics. This effect is alsoattributed to the problems with sticking of material to theend of the immersion optics. Additionally, the variationin determination of the rate constant is further decreasedfor the PhAT System relative to the standard NCO. Thiseffect is ascribed to the increase in sampling rate withthe PhAT System, which makes the rate constant morereproducible due to more accurate description of thetransformation kinetics. This increase in sampling rate ispossibly due to an increased signal-to-noise ratio withthis system compared to the standard systems, which can

be attributed to the greater constancy of the signal duringsample movement as described above.

The repeatability of the transformation when monitoredwith the PhAT System can be seen visually in Fig. 5a. Acomparison of the transformation profile as measured withthe different systems is shown in Fig. 5b. It is apparentthat both of the non-contact optic sampling devices yieldessentially identical transformation profiles, although thePhAT System can sample at rates three times that of thestandard NCO system and produce spectra with the samesignal-to-noise ratio, resulting in a more complete trans-formation profile. In contrast, the immersion optic basedprobe sometimes measures a slightly slower transforma-tion rate. This is most likely due to intermittent stickingof the material on the optic window at some point in theexperiment, resulting in non-representative sampling.Sometimes this sticking was a transient phenomenon, ascan be seen in the Immersion 1 trace of Fig. 5b, but mostof the time the material stuck permanently to the end ofthe optic, as seen in the Immersion 2 trace of Fig. 5b. The

APPLIED SPECTROSCOPY 941

FIG. 7. Illustration of the effect of sampling rate on the observed trans-formation kinetics for the conversion of anhydrous theophylline. Percentmonohydrate formed plotted versus time. Experiments carried out ac-cording to Table I with the exception of the mixing speed (200 rpm).

former phenomenon is potentially the most problematicissue, since it would provide no warning that stickingmight have occurred during the process, leading to mis-interpretations of the data. Clearly non-contact optics ispreferable in this situation.

Effect of Mixing Speed. Earlier work has shown thatthe onset time and the transformation rate appeared to beinfluenced by the mixing speed used during the wet gran-ulation process.13 However, due to the slow sampling rate,which led to poor resolution between different mixingspeeds and the problems with sticking, this phenomenoncould only be described qualitatively. In order to furtherinvestigate the influence of process parameters on thetransformation kinetics, granulations were performed us-ing different mixing speeds and the transformation wasmonitored using the PhAT System, which can collect aRaman spectrum in one third the time of the system uti-lized previously (for the reasons discussed above). Mix-ing speeds of 50, 100, 200, and 400 rpm were employed.A plot of the transformation profiles for different mixingspeeds as determined by the PhAT System is shown inFig. 6a. For the PhAT System data, the amount trans-formed versus the time (commencing from the onsettime) was fitted to the empirical Eq. 4. The onset timecould be accurately determined because of the fast sam-pling rate of the PhAT System:

2k(t2t )0C 5 Max[1 2 e ] (4)MT

where the variables are similar to those described in Eq.3 with the addition of t0 being the onset time. The resultsclearly show that mixing speed influences transformationkinetics and that for the 50–400 rpm range of speeds thetransformation time decreases with increasing mixingspeed. Furthermore, linear relationships between both tipspeed (v) and rate of transformation (k) and the inverseof onset time (1/t0) could be established, as shown in Fig.6b. The improved sampling capability of the PhAT Sys-tem leading to improved data analysis thus facilitated thedetermination of the relationship between processing pa-rameters and transformation kinetics.

At very fast mixing speeds the advantage of the PhATSystem is even more pronounced. The standard systemsrequire a sampling period of every 30 s in order to obtain

a spectrum of sufficient quality, and hence few datapoints are obtained across the most rapid parts of thetransformation, which can lead to inaccuracies. However,the PhAT System could collect spectra every 10 s, whichmade it possible to accurately follow the transformation,even at high mixing speeds. This difference in spectralcollection speed is illustrated in Fig. 7, in which the trans-formation curves are displayed as individual data pointsfor a mixing speed of 200 rpm. Thus, the results fromthe focus tolerance evaluation explain why a much short-er sampling time could be used with the PhAT Systemin order to achieve a comparable spectral quality, withthis in turn leading to better repeatability, description ofthe kinetic profile, and the ability to monitor more rapidlyoccurring transformations.

CONCLUSION

For process monitoring where it is difficult to maintaina constant sampling distance due to the dynamics of theprocess, the tightly focused radiation employed by mostRaman spectrometers is disadvantageous. A comparisonof different sampling optics clearly shows the benefits ofa Raman system that employs a large spot size in com-bination with a long focal length optic, both in terms ofspectral quality and minimization of sub-sampling ef-fects. This study also shows the potential of Raman spec-troscopy as a noninvasive pharmaceutical process moni-toring tool, in particular to evaluate solid-state transfor-mations during wet granulation.

ACKNOWLEDGMENTS

Dr. Jukka Rantanen and Francis E. Rhea are gratefully acknowledgedfor assistance with data analysis and experimental support. Sharon Der-am, SBI Analytical, Inc., and Kaiser Optical Systems, Inc., are acknowl-edged for their assistance with instrumentation. The NSF-IUCRC Cen-ter for Pharmaceutical Process Research and AstraZeneca R&D Mol-ndal are acknowledged for funding.

1. J. Workman, M. Koch, and D. J. Veltcamp, Anal. Chem. 75, 2859(2003).

2. Y. Qiu, X. Z. Qu, X. Q. Song, J. X. Chen, and F. T. Chau, Instrum.Sci. Technol. 24, 143 (1996).

3. S. Watano, K. Terashita, and K. Miyanami, Bull. Univ. Osaka Pref.,Series A 39, 187 (1990).

4. F. Gonzalez and R. Pous, J. Pharm. Biomed. Anal. 13, 419 (1995).5. J. Rantanen, S. Lehtola, P. Ramet, J. P. Mannermaa, and J. Yliruusi,

Powder Technol. 99, 163 (1998).6. C. Wang, T. J. Vickers, J. B. Schlenoff, and C. K. Mann, Appl.

Spectrosc. 46, 1729 (1992).7. I. M. Clegg, N. J. Everall, B. King, H. Melvin, and C. Norton,

Appl. Spectrosc. 55, 1138 (2001).8. T. W. Collette and T. L. Williams, J. Environ. Monit. 4, 27 (2002).9. I. R. Lewis and H. G. M. Edwards, Handbook of Raman Spectros-

copy (Marcel Dekker, New York, 2001), p. 1080.10. D. E. Pivonka and J. R. Empfield, Appl. Spectrosc. 58, 41 (2004).11. L. X. Yu, R. A. Lionberger, A. S. Raw, R. D’Costa, H. Q. Wu, and

A. S. Hussain, Adv. Drug Deliv. Rev. 56, 349 (2004).12. T. Vankeirsbilck, A. Vercauteren, W. Baeyens, G. Van der Weken,

F. Verpoort, G. Vergote, and J. P. Remon, Trac 21, 869 (2002).13. H. Wikstrom, P. J. Marsac, and L. S. Taylor, J. Pharm. Sci. 94, 209

(2005).14. J. Rantanen, H. Wikstrom, R. Turner, and L. S. Taylor, Anal. Chem.

77, 556 (2005).15. F. C. Clarke, Pfizer Ltd., personal communication (2004).16. F. C. Clarke, S. V. Hammond, R. D. Jee, and A. C. Moffat, Appl.

Spectrosc. 56, 1475 (2002).17. E. Rasanen, J. Rantanen, A. Jorgensen, M. Karjalainen, T. Paakkari,

and J. Yliruusi, J. Pharm. Sci. 90, 389 (2001).18. L. S. Taylor and G. Zografi, Pharm. Res. 15, 755 (1998).