Vibrational Spectroscopy 100 (2019) 48–56 Contents lists available at ScienceDirect Vibrational Spectroscopy journal homepage: www.elsevier.com/locate/vibspec The influence of particle size on the intensity and reproducibility of Raman T spectra of compacted samples ⁎ Diego A. Gómez, Jordi Coello, Santiago Maspoch Departament de Química, Facultat de Ciències, Universitat Autònoma de Barcelona, 08193, Bellaterra, Barcelona, Spain A R T I C LE I N FO A B S T R A C T Keywords: Given the growing interest in the application of Raman spectroscopy for quantitative purposes in solid phar- Raman spectroscopy maceutical preparations, a revision of the effect of particle size on Raman spectra of compacted samples is Particle size presented. For this purpose, a set of tablets of potassium hydrogen phthalate (KHP) of different particle size were Peak intensity prepared. KHP was used because of its purity and stability, which allow to consider that samples will not be Reproducibility of a raman spectrum altered during measurements; but also because of its chemical structure (aromatic ring and carboxylic groups), that are present in many active pharmaceutical ingredients (API). The latter makes possible to consider KHP as a model pseudo-API. As KHP tablets only contain a pure compound, the mapping strategy that was considered for measuring our samples will not be affected by subsampling issues. The spectra variance can be attributed to the intrinsically reproducibility in recording the spectra (which mainly depends on the instrument set-up) and the site-to-site differences in elastic scattering properties. Two different instrumental optics have been studied: a macro-Raman system and a Raman microscope (500 μm and 50 μm laser spot size, respectively). The effect of the spectra preprocessing is also evaluated. The overall results demonstrate raw Raman intensity increases with particle size up to a value that depends on tablet width and that the applied spectral preprocessing (baseline correction and a unit vector normalization), reduces the differences in Raman intensities due to the particle size, but does not completely eliminate it for the lower particle sizes (< 20 μm). For tablets containing particles with predefined sizes, it corrects the mapping site-to-site differences in elastic scattering. 1. Introduction to the physical effect is particle size and, in this sense, it has been re- ported that the accuracy of a Raman method is affected when particle Raman spectroscopy is becoming a popular technique for the de- sizes of the calibration and prediction samples are different [10,11]. velopment of analytical methods applicable to several industrial fields. The theoretical relation between Raman signal and powder prop- Among them, and probably because of the implantation of new analy- erties was described by Schrader et al. [12,13], as an extension of the tical procedures in the framework of PAT (Process Analytical Kubelka-Munk theory of the optical properties of crystal powders [14]. Technology) initiative [1], a relatively high amount of the new proce- The radiation balances involved are explained by a set of differential dures are focused on pharmaceutical analysis, where Raman spectro- equations, in which one of the terms corresponds to the coefficient for scopy is being used for monitoring of different unit operations in the the elastic scattering of the sample, which is inversely proportional to production chain [2–5]. When considering to analyze a solid sample the diameter of the particles. This coefficient determines, as expected, using this technique, the spectrum obtained is a combination of the the flux of the excitation radiation, but also determines the magnitude chemical and physical properties of the sample being measured: the of the flux of the inelastically scattered radiation (Raman radiation). first represents the main contribution to the appearance of the spec- Two different scales regarding particle size must be considered: trum, which allows for example, the building of spectral libraries [6,7], nano- and micro-scale. In the case of nanoparticles, as particle size re- or the development of quantitative methods for the determination of duces, subtle changes in the vibrational spectrum of the material are API(Active Pharmaceutical Ingredient) content in a solid pharmaceu- noticeable; these are justified in the confinement of phonons effect, and tical preparation [8,9]; but the latter is also important and, to some in practice presents useful information for size characterization at na- extent, its contribution is less intuitive. The main property contributing noscale [15–17]. ⁎ Corresponding author. E-mail address: santiago.maspoch@uab.cat (S. Maspoch). https://doi.org/10.1016/j.vibspec.2018.10.011 Received 2 October 2018; Received in revised form 26 October 2018; Accepted 26 October 2018 Available online 01 November 2018 0924-2031/ © 2018 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/). D.A. Gómez et al. Table 1 Experimental set-ups and main conclusions reported in previous studies regarding the influence of particle size on Raman spectra. Ref. Compound(s) analyzed Raman instrument Irradiation diameter Main findings (particle sizes) [18] Ba(NO3)2, NH4Cl, Backscattering FT-Raman, benchtop (1064 nm laser). 300 μm. Raman intensities decrease as particle size increases. K2CrO4, Li2CO3 (105-205 μm, powders). [19] NaClO3, NaBrO3, Backscattering dispersive Raman (514.5 nm laser). A) 1 mm distance = 500 μm. Raman intensities decrease as particle size increases. K4Fe(CN)6 · 3H2O, A non-contact fiber optic probe was used, acquiring spectra B) 10 mm distance = 4600 μm. The dependence of the intensities with particle size diminishes. A CuSO4 · 5H2O, in two predefined distances regarding sample position. maximum is observed in the intermediate fractions. A compacted K3Fe(CN)6, NaNO3 sample produces less intense signals than the same sample before (76-605 μm, powders, NaNO3 was also measured as a compaction. tablet). [20] Flufenamic acid, two polymorphs. (65-215 μm, powders). Backscattering dispersive Raman (785 nm laser). A) Immersion optic: 60 μm. Each polymorph led to a different result: with one of them the Two types of fiber optic probes were used. variation of intensities with particle size is quite low, with the other polymorph intensities increase as particle size decreases until a maximum is reached. 49 B) Non-contact sampling device Intensities do not change when the different particle size fractions are (PhAT probe): 3000 μm. measured. [21] Chlorpheniramine maleate (2 % w/w) - microcrystalline Backscattering and transmission dispersive PhAT probe 3000 μm. The backscattering setting is relatively insensitive to the differences cellulose (MCC) tablets (three sizes of MCC: D50 = 16.3, Raman in MCC particle size, in comparison to the transmission mode. 61.7 and 178.0 μm). (785 nm laser). [22] Binary mixtures of K2CrO4 and Ba(NO3)2 (9 concentration Backscattering dispersive Raman. Confocal microscope, 6900 μm. Mixtures with the same mass fraction, but different particle size, levels), each compound in 10X objective (632 nm laser). produce spectra whose peak intensities are more different, than in the 8 particles sizes. Laser beam was focused inside the sample for illuminating case of measuring two mixtures with different mass fraction but the (75-425 μm, powders, mixtures were measured at two the whole sample surface, that was contained in a recipient same particle size. compaction levels). with a diameter of 6.9 mm. A firmly packed powder sample produces more intense peaks than a slightly compacted powder. [23] CaCO3 Backscattering dispersive portable Raman (785 nm laser). 100 μm. Signal increases with particle size up to the 24 μm fraction, then (2-121 μm, powders). decreases for the larger sizes. [24] Ibuprofen and mannitol tablets, 5 levels of concentration, Transmission dispersive Raman, benchtop 4300 μm. A Principal Component Analysis, including all the types of tablets, 3 compaction pressures (785 nm laser) shows that the content and particle size of the drug substance, the (each compound in two sizes: ibuprofen, D50 = 71, particle size of the filler and, in a less extent, the compression force 154 μm; mannitol, D50 = 91, 450 μm) affect Raman spectra. [25] Alpha quartz Backscattering dispersive Raman. Confocal microscope, 2 μm. Raman intensities increase with particle size up to a 20-30 μm (< 11 μm-250 μm, powders). 50X objective (488 nm laser). fraction, then stop to increase for larger size fractions. Vibrational Spectroscopy 100 (2019) 48–56 D.A. Gómez et al. Vibrational Spectroscopy 100 (2019) 48–56 Considering materials in the micro-scale, which is the order of size chemical stability in laboratory conditions [32]. The structure of the magnitude of the compounds used for the preparation of a solid phar- compound also mimics a common API because of the aromatic ring of maceutical form, several studies in which particle size was controlled the phthalate; this functional group produces strong signals in the have been described in literature, testing different inorganic salts and Raman spectra, and commonly represents the most selective analytical organic compounds and using instruments built both for transmission signal for the determination of many APIs. Results on both raw and and/or backscattering measurements, but with some differences in the preprocessed spectra are compared. excitation source-sample arrangement. A brief summary of the experi- Raman spectra have been recorded using a FT-Raman instrument mental set-ups used, and the main conclusions reported is shown in that incorporate non-contact optic sampling devices, in which the Table 1 [18–25]. sample is placed on a software-controlled x-y-z mapping stage and il- More specifically, the relation between sample particle size and the luminated via a parabolic mirror objective (macro-Raman system, intensity of Raman signal has been mainly studied for slightly tamped 500 μm spot size), or a 10X objective (Raman microscope, 50 μm spot powders, using instruments with different laser spot diameter (ranging size). Both optics include a video camera, to visually focus the laser from 2 μm up to more than 4000 μm). In general, the major conclusion onto the surface and collect spectra of the measuring positions. Sets of reported has been that signal increases as particle size decreases spectra were acquired considering two acquisition modes: several suc- [18–20]. However, a detailed examination of the plots of Raman in- cessive spectra at the same measuring point (repeatability), and from tensity versus the different particle sizes tested shows that this trend is different points of the measured tablets following a predefined grid not totally fulfilled along the whole interval considered on each of these (mapping). The estimation of a repeatability value is related to the best studies, being noticeable in some cases that, for the smaller sizes, in- precision that can be achieved using a specific spectrometer and re- tensities increase with particle size. And actually, this latter trend is cording conditions, with prefixed values for spectral resolution and what Schrader and Bergmann predict in their theoretical study. Chio number of scans averaged. On the other hand, mapping is a practical et al. [25] reported an increase of the intensity starting from a fraction solution to overcome the subsampling problem derived from the limited containing particles of less than 11 μm up to 20–30 μm, then intensities spot size [33] and sample heterogeneity. The differences found between stop increasing for larger sizes. A very small irradiation spot was used in the two recording procedures, in a situation where the subsampling risk this work and, given that measurements were performed using a con- is null (tablets are composed by a single compound), may be useful to focal instrument, very few particles were irradiated and a sampling measure the effect of the particle size and distribution for a given deepness delimitation was possible. On the other hand, Wang et al. [19] sampled area. performed Raman measurements on solid particles dispersed in dif- ferent organic liquids; the experimental findings allowed them to prove 2. Theory that diffuse reflectance is an important mechanism in understanding the effect of particle size on Raman intensities. Nonetheless, their findings The propagation of a Raman radiation in a particulate solid sample regarding the evolution of Raman intensities with particle size are op- was studied by Schraeder and Bergmann, as an extension of the posed to the theoretically expected trend. Kubelka-Munk theory for diffuse reflectance radiation. From the defi- More recently, Kristova et al. [23] reported the effect of particle size nition of an elastic (r ) and an inelastic (s ) scattering coefficients, and on the fundamental vibrations of the carbonate in calcite, a mineral after solving the radiative transfer equations, one of their findings was which internal modes have been well studied [26]. They found another that the intensity of the backscattered Raman radiation could be de- change of trend in their measurements, with Raman intensities in- scribed by the following equation creasing with particle size up to 20 μm, and then decreasing for larger sizes. Additionally, they also found that the rate of increasing was not ⟵ → s k ⎧ k sinh2 kd + (a + r ) sinh kd cosh kd − krd ⎫ ∅R = ∅0 constant but wavenumber-dependent. a ⎨⎩ [(a + r ) sinh kd + cosh kd]2 ⎬ ⎭ (1) Hence, the bibliographical review of the previous experimental ⟵ works shows that there are considerable differences in the main con- Where ∅R is the Raman radiation emerging from the irradiated face, → clusions reported on them, even though the particle sizes of the frac- ∅0 is the irradiation flux, and, s and r , are respectively, the linear tions that were used in each case were, to some extent, overlapped. Napierian absorption, elastic scattering and Raman scattering coeffi- Particle size is not an arbitrary property of the solid pharmaceutical cients. k is a parameter related to a and r. forms currently available in the market, in fact it is an important at- By considering that the elastic coefficient is proportional to the re- tribute that influences some characteristics of the final product such as ciprocal of the particle diameter, and after assigning values to a , s and the dissolution rate, drug release rate for controlled release formula- k, the main conclusions Schraeder and Bergmann obtained were that tions, and the dosage unit content uniformity [27,28]. Tablets are the backscattered Raman intensity should: most popular of all the medicinal preparations intended for oral use [29], and Raman spectroscopy is becoming a very useful tool for the i) Increase with particle diameter quantitative determination of API [30]. Since for this purpose, cali- ii) Increase with sample thickness but should arrive to a plateau after a bration samples must be prepared, it is necessary to go deeper in the certain penetration into the sample. The depth of this penetration understanding of the effect of particle size not only on Raman intensity, increases with particle size, and for particles of low and medium but also in the attainable precision. On the other hand, raw Raman diameter (10–100 μm) should be in the range of 1–4 mm spectra are scarcely used. It is nowadays well-known that, in order to obtain an adequate precision in a quantitative analysis, Raman spectra As pointed out by McCreery [34], for a FT spectrometer, the signal must be preprocessed before applying any regression model; in that magnitude for a given Raman shift value is proportional to the photons sense, the standard protocol includes a baseline correction and spectral observed for that Raman shift only, while the noise has contributions normalization [31]. The performance of this standard preprocessing from all Raman shift values. This leads to the condition that the ex- procedure in the correction of the effect of particle size must be also pected standard deviation should be constant at all wavenumbers, and considered. function of the instrument characteristics and recording set-up. This is With the aim of a better understanding of the effect of particle size true if spectra are measured exactly at the same point of the sample, but in compacted tablets on Raman intensities and their reproducibility, a if spectra are recorded at different positions on tablet surface, the dif- set of tablets of potassium hydrogen phthalate (KHP), previously ferences in the elastic scattering must also be considered. The elastic grounded and sieved from < 20 μm to > 100 μm, have been prepared. scattering coefficient mainly depends on particles size distribution, but KHP was chosen because it is a primary standard, what assures its also on differences in refraction indexes, surfaces orientation, etc. 50 D.A. Gómez et al. Vibrational Spectroscopy 100 (2019) 48–56 Another contribution that will affect the reproducibility of spectra ac- Nd:YAG laser as excitation source (1064 nm) with a spot size of 500 μm quisition is the slight differences in focusing because tablets are not and 50 μm respectively, and a liquid N2 cooled Ge diode detector. A always placed in a perfect horizontal position. software-controlled x-y-z stage is available in both instruments, which Consequently, the expected variance (s 2 ) in mapping measurements allows controlled and reproducible mapping measurements. The Raman can be described as: microscope was also used as an optical microscope to visually confirm 2 the fractions obtained after sieving. Opus 7.1 was used for system s 2 = sa2 + ss2 + sf2 + s (2) control and to export data. All the spectra were preprocessed using m Unscrambler 10.3 (Trondheim, Norway). sa2 Where is the intrinsic analytical variance when recording a spectrum (function of the instrumental set-up: spectral resolution, number of 3.3. Spectral acquisition scans averaged); ss2 is the variance produced by the differences in scattering between different positions of the same sample; sf2 is the The spectra were recorded in two different ways: 10 measurements variance due to focus variability; and sm2 accounts for the differences in at the same point (repeatability) and 10 measurements at different composition of the sampled volume when analyzing complex mixtures. points of the tablet surface (mapping). One of the fractions was also Since ss2 and sf2 cannot be differentiated separately, both terms can be measured before compaction. Each spectrum was acquired from 50 to joined and expressed simply as ‘scattering effect’. In the case of re- 3600 cm−1 with a spectral resolution of 1 cm−1, and after averaging 32 cording Raman spectra of a sample containing a single compound sm2 scans. The laser power was set at 400 mW. Previously it was confirmed can be neglected and (2) simplifies to: that this relatively high laser power did not have any effect on KHP s 2 = sa2 + ss2 (3) stability. For mapping acquisition, the system was optically focused on a central point of the tablet, and a grid of 10 point was recorded. To From a practical point of view, for a given instrumental set-up, sa avoid tablet displacements, they were fixed on an aluminum disk fitted can be estimated recording the spectra at the same measuring position. on the x-y-z stage. Along the text, this recording mode will be referred as a repeatability measurement. 3.4. Spectral preprocessing 3. Materials and methods Baseline corrections (baseline offset and slope of the spectrum, ab- breviated BLC), and a normalization to a unit vector length (UNorm) 3.1. Preparation of tablets were applied to study the effect of these pretreatments on the intensities and reproducibility of the peaks, in comparison to the raw spectra. The Potassium hydrogen phthalate (KHP) (Fluka, Switzerland) was in- experimental findings related to the Raman intensities are discussed itially dried at 105 °C, and then grounded by hand with a mortar and considering three wavenumbers (band assignments according Ref. pestle. Subsequently, six fractions of different particle diameter were [35]): 1675 cm−1 (C]O stretching), 1601 cm−1 (CCt stretching) and obtained by dry sieving using a RP.15 laboratory sieve shaker (CISA, 1039 cm−1 (CH bending in plane). Barcelona, Spain). The nominal aperture of the available sieves (CISA, Barcelona, Spain) was 20 μm, 25 μm, 50 μm, 75 μm and 100 μm (frac- 4. Results and discussion tions will be denoted in figures as: < 20 μm = 10 μm, 20–25 μm = 22.5 μm, 25–50 μm = 37.5 μm, 50–75 μm = 62.5 μm, 4.1. Effect of compacting powder 75–100 μm = 87.5 μm, 100–150 μm = 125 μm). The powder from each fraction was used to prepare a compacted sample in the form of a disc Since most of the previous published studies have been performed using a Perkin Elmer 15.001 hydraulic tablet press (Waltham, MA, on powders, initially, a set of spectra were acquired on the 20–25 μm USA): 300 mg of each fraction was used for the preparation of the ta- particle size fraction without compaction: the powder was only gently blets; the resulting disks have a diameter of 13.4 mm and a thickness of flattened using a laboratory spatula over the measuring platform in the 1.4 mm for all the fractions except for the > 100 μm fraction, which sample compartment of the spectrometer, employing the proper thickness was 1.5 mm. The compaction pressure was 140 MPa for all the amount of sample to obtain approximately the same thickness that in samples. compacted samples, and 10 spectra were recorded as mapping on the macro-Raman system. 3.2. Instruments and software The average of 10 spectra for the loosely packed and compacted sample are overlaid in Fig. 1. As can be seen, the measurement of a Raman spectra were acquired using two different instruments: a compacted sample results in a significantly more intense spectrum conventional macro-Raman system (Brüker MultiRam spectrometer, (Fig. 1A). In this first comparison, only the compaction level introduces Billerica, MA, USA) and a Raman microscope equipped with a 10X a difference between samples, as both possess the same particle size. objective (RamanScope III, Billerica, MA, USA); both instruments use a Hence, the difference between spectra reflects the difference in two Fig. 1. Overlapped Raman spectra of the loosely packed (grey line) and compacted (black line) for the 22.5 μm fraction. A) Raw spectra, B) Spectra after baseline corrections and unit vector normalization. The arrows denote the wavenumbers selected for the intensity measurements (1674, 1601,1039 cm−1). 51 D.A. Gómez et al. Vibrational Spectroscopy 100 (2019) 48–56 physical characteristics of the measured sample: the packing degree 4.2.1. Repeatability results between particles, and a resulting porosity. A high-power laser is irra- Fig. 2 depicts the results found. As can be seen, the trend is very diated on the sample, and in each case, there is a number of scattering similar for both optical systems: an increment in signal intensity with molecules that will produce a Raman signal. As the intensities are particle size is clearly observed in raw spectra for the fractions of stronger in the compacted sample, this suggests that a higher number of smaller size, up to a value around 40 μm, from which the intensity molecules are involved in the resulting response. This can be attributed values stabilize and are independent of particle size (Fig. 2A and C). to an increase in the density of the sampled material and a decrease of The evolution of Raman intensities for the smaller fractions, agrees with the elastic scattering coefficient, in perfect concordance with what it is what would be expected for a system controlled by the diffuse re- expected from Kubelka-Munk theory, and has been previously reported flectance effect: Raman signal increases when particle size increases. for Near Infrared (NIR) reflectance measurements [36]. Wang et al. The repeatability is quite good, with RSD% values (n = 10) in the range [19] reported a similar comparison using NaNO3, however in their of 0.2–1 %, and 1–5 %, for the macro-Raman system and Raman mi- experiments a compacted sample produced less intense signals than the croscope respectively, producing error bars that are hardly detected at corresponding powder; they mentioned that this result was counter- the graphical scale of the figure. intuitive, but consistent with the other results found using the same The stabilization of the Raman intensity from a determined particle optical system, formed by parallel optical fibers. Chen et al. [22] also size can be understood if we consider the depth (volume) of sample presented the powder/compacted sample spectra for a K2CrO4/Ba from which the Raman radiation collected in backscattering mode (NO3)2 mixture, in this case a compacted sample resulted in more in- comes from. In their theoretical approach, assuming specific values for tense peaks than those of a loosely packed sample. A confocal Raman a and k (Eq. 1), Schrader and Bergmann found that the intensity of the microscope was used in this study. backscattered Raman radiation should arrive to a plateau at a sample In summary, it is clear that instrument optics determines the ex- depth of around 1.5 mm for a particle size of 10 μm and 4 mm for perimental findings and must be taken into account when the results 100 μm. This result is in a good agreement with the conclusions re- from previous studies are interpreted. This is made more evident in ported by Matousek and Parker [37], who found, from a simulation Table 1, where the results reported of the effect of particle size on model where elastic and inelastic radiations propagate in the medium Raman intensity have been summarized, showing different conclusions in random-walk-like fashion within a three dimensional space, that for depending on the experimental set-ups used in each case. a particle size of 10 μm, the 88% of the backscattered Raman radiation After applying the spectral preprocessing (Fig. 1B), it is no longer was produced in the first millimeter of sample depth, and less than 3% possible to distinguish the samples by their compaction degree. Table 2 after 1.5 mm. presents the intensity values for the wavenumbers selected, and the Going back to our results, Fig. 3 depicts the intensity recorded at corresponding standard deviation for the 10 positions measured. The 1601 cm−1 for KHP tablets containing particles corresponding to the absolute standard deviation values for the measurements on the com- 22.5 and 125 μm fractions, as a function of tablet thickness. In a good pacted and loosely packed samples are of the same order (variances are agreement with the theoretical findings described above, the intensity homogeneous according to an F-test). However, and because of the increases with the particle size, arrives to a plateau after a defined different signal intensity, the percent relative standard deviation (RSD sample thickness, and that this stabilization thickness increases with %) of the measurements performed on the tablet (2.5–3 %), is almost particle size (1.1–1.2 mm for 22.5 μm fraction and 1.4–1.5 mm for half of the RSD% value obtained from the powder sample (5%) when 125 μm). Despite that our results show a lower increment than the raw spectra are compared. After spectral preprocessing, the RSD% is theoretical prediction in the variation of penetration with particle size, practically the same for the compacted and powdered sample, and it is the relevant fact is that the trend predicted by the theory is experi- reduced by a factor of 4–5, yielding RSD% values in the range of mentally corroborated. 0.3–1.6 %. It must be kept in mind that the tablets prepared for the experi- mental design shown in Fig. 2 had a thickness around 1.4–1.5 mm; in this manner, once arrived at the particle size for which this is the 4.2. Evolution of Raman intensities with particle size in compacted samples maximum depth of penetration, it should be expected that the intensity would remain constant, even for increasing particle sizes. The latter is Powders were compacted at 140 MPa, a standard compaction what it is actually observed. And finally, it should be pointed out that pressure in pharmaceutical tablet manufacturing. In a previous study, it even for a relatively large particle size (125 μm), the sample depth from was found that considering tablet compaction pressures ranging from which the Raman spectrum is acquired is relatively small, emphasizing 140 to 700 MPa are not distinguish in the acquired Raman spectra after the fact that, using a backscattering Raman spectrometer, the acquired applying a spectra normalization [8]. spectra mainly possess information from only the surface irradiated. In the next sections, the evolution of Raman peak intensities with After the spectral preprocessing (Fig. 2B and D), the effect of par- particle size will be explained considering the strategy in which spectra ticle size on peak intensities is considerably reduced, however an were acquired; the results of the measurements in the same position ANOVA test (α = 0.05) shows that for the 1039 cm−1 and 1601 cm−1 (repeatability) will be presented first, and then the results found in peaks, there are still significant differences between fractions. These mapping mode. Thereafter, a joint discussion of the repeatability and differences arise from the fraction of smaller particle size (< 20 μm) mapping measurements will be presented. which intensity is slight but statistically different of the intensity for other particle sizes. Table 2 Effect of compaction on Raman intensities for the sample of 20–25 μm measured by mapping (n = 10) in the macro-Raman system. Intensity, standard deviation in brackets and RSD%, before and after spectral preprocessing. Raw data Preprocessed Wavenumber/cm−1 Powder Tablet Powder Tablet 1675 0.102 (0.005) 4.9 % 0.163 (0.005) 3.1 % 0.0320 (0.0005) 1.6 % 0.0322 (0.0003) 0.9 % 1601 0.487 (0.022) 4.5 % 0.776 (0.020) 2.6 % 0.1532 (0.0005) 0.3 % 0.1529 (0.0008) 0.5 % 1039 0.764 (0.037) 4.8 % 1.220 (0.032) 2.6 % 0.2402 (0.0015) 0.6 % 0.2406 (0.0015) 0.6 % 52 D.A. Gómez et al. Vibrational Spectroscopy 100 (2019) 48–56 Fig. 2. Evolution of peak Raman intensity and standard deviation of repeatability measurements (n = 10 spectra) of compacted KHP tablets with particle size. Macro- Raman system: A) raw spectra; B) after spectral preprocessing. Raman microscope: C) raw spectra; D) after spectral preprocessing. In C) and D) the vertical line indicates the laser spot diameter. attributed to the well-known subsampling problem of Raman spectro- scopy, and must be interpreted in terms of the differences in elastic light scattering (Eq. 3). A similar trend (variance increases with particle size) is widely known for NIR spectroscopy [38]. Even though the effect is lower in Raman spectroscopy than in NIR, it must be highlighted that in both cases the same trend is found. Moreover, the different relations between the laser diameter (500 μm), and the size of the particles being measured, must be kept in mind; as a result the major amplitude of the error bars for the larger fractions (Fig. 4A), can be justified in the more significant impact of the individual arrangement of the particles in comparison when the total number of particles irradiated diminished. The relative higher increase of the standard deviation with particle size when using a microscope (Fig. 4C) can be attributed to the smaller sampled area. In this case, the scattering effect is enhanced because of the smaller number of particles being irradiated. Having a laser spot size of 50 μm, it seems that when the diameter of particles exceeds this Fig. 3. Average and standard deviation of Raman intensity at 1601 cm−1 value, the specific arrangement of the individual particles will sig- measured on KHP tablets of different thickness using the macro-Raman system. nificantly determine the intensity of the resulting spectrum acquired in Particle size 22.5 μm (red squares) and 125 μm (yellow dots). Spectra acquired each of the sampling positions. at a laser power of 400 mW, averaging 128 scans, 6 measurements in different The spectral preprocessing minimizes the particle size-dependent positions of the same tablet. Lines drawn only to visualize the trends. variations in Raman intensities (Fig. 4B and D), however, an ANOVA test shows again that there are still significant differences for the 4.2.2. Mapping results 1039 cm−1 and 1601 cm−1 bands, similarly to the previously discussed Fig. 4 shows the average peak height and its standard deviation observation for repeatability measurements: the intensity for the versus particle size, when recording the spectra by mapping, using a smaller fraction (< 20 μm) is significantly different from the other grid of 10 points. The same trend described above is found: an increase sizes. in Raman intensities from < 20 μm up to around 50 μm, is again ob- served. 4.3. Comparison between mapping and repeatability measurements As expected, given that tablets are solely constituted by a pure component, the average intensity for each particle size is the same than For a better visualization of the different trend in peak reproduci- in repeatability measurements. It is interesting to note that the mag- bility, the evolution of standard deviations and the corresponding RSD nitude of the standard deviation increases for the larger fractions. Since % for the Raman intensity at 1601 cm−1 before and after the spectral the object being measured is a homogeneous sample, this fact cannot be preprocessing is depicted in Fig. 5 as a function of the particle size. It is clear that, while there is a difference between mapping and 53 D.A. Gómez et al. Vibrational Spectroscopy 100 (2019) 48–56 Fig. 4. Evolution of the average peak Raman intensity and standard deviation of mapping measurement (n = 10 spectra) of compacted KHP tablets with particle size. Macro-Raman system: A) raw spectra; B) after spectral preprocessing. Raman microscope: C) raw spectra; D) after spectral preprocessing. In C) and D) the vertical line indicates the laser spot diameter. repeatability measurements when considering raw spectra, this differ- measurements are significantly higher than the variances of repeat- ence disappears after the spectral preprocessing. An F test (one-tailed, ability measurements. After the spectral preprocessing, variances be- α = 0.05) confirms that, for raw spectra, variances of mapping come homogeneous. This fact can be interpreted in terms described in Fig. 5. Evolution of standard deviation (SD) and percent relative standard deviation (RSD %) of Raman intensity at 1601 cm−1 with particle size, both for mapping and repeatability measurements. Macro-Raman system: A) SD, B) RSD %. Raman microscope: C) SD, D) RSD %. In C) and D) the vertical line indicates the laser spot size. 54 D.A. Gómez et al. Vibrational Spectroscopy 100 (2019) 48–56 Eq. 3. The variance recorded in mapping mode is the combination of References two sources of variability, the intrinsic repeatability and the differences in elastic scattering between different points of the tablet surface. The [1] U.S.D. Of H. And H.S. Fda, Guidance for Industry PAT — a Framework for fact that the relative standard deviation (RSD%) after spectral pre- Innovative Pharmaceutical Development, Manufacuring, and Quality Assurance, (2004), p. 16 http://www.fda.gov/downloads/Drugs/ processing overlap with the values obtained in repeatability conditions GuidanceComplianceRegulatoryInformation/Guidances/ucm070305.pdf. (Fig. 5B and D), strongly suggest that the elastic scattering variability is [2] J. Kim, J. Noh, H. Chung, Y.-A. Woo, M.S. Kemper, Y. Lee, Direct, non-destructive practically removed after a normalization step. quantitative measurement of an active pharmaceutical ingredient in an intact capsule formulation using Raman spectroscopy, Anal. Chim. Acta 598 (2007) 280–285, https://doi.org/10.1016/j.aca.2007.07.049. [3] T. Vankeirsbilck, A. Vercauteren, W. Baeyens, G. Van der Weken, F. Verpoort, 5. Conclusion G. Vergote, J.P. Remon, Applications of Raman spectroscopy in pharmaceutical analysis, Trends Analyt. Chem. 21 (2002) 869–877, https://doi.org/10.1016/ S0165-9936(02)01208-6. When irradiating a particulate sample to acquire a Raman spectrum, [4] S.C. Pinzaru, I. Pavel, N. Leopold, W. Kiefer, Identification and characterization of the excitation flux moves throw the sample by multiple elastic processes pharmaceuticals using Raman and surface-enhanced Raman scattering, J. Raman Spectrosc. 35 (2004) 338–346, https://doi.org/10.1002/jrs.1153. of scattering, and a small fraction of the scattered photons suffer Raman [5] G. Févotte, In situ raman spectroscopy for in-line control of pharmaceutical crys- scattering. Thereafter, the Raman scattered radiation also travels inside tallization and solids elaboration processes: a review, Chem. Eng. Res. Des. 85 the sample through a diffuse reflectance scattering mechanism. This (2007) 906–920, https://doi.org/10.1205/cherd06229. mechanism is the qualitative basis of the Schrader and Bergmann the- [6] J.D. Rodriguez, B.J. Westenberger, L.F. Buhse, J.F. Kauffman, Standardization of Raman spectra for transfer of spectral libraries across different instruments, Analyst oretical approach [19,20]. The effect of particle size in the diffuse re- 136 (2011) 4232, https://doi.org/10.1039/c1an15636e. flectance mechanism has been studied and demonstrated in a large [7] C.M. Gryniewicz-Ruzicka, J.D. Rodriguez, S. Arzhantsev, L.F. Buhse, J.F. Kauffman, number of works in NIR Spectroscopy [39,40]. From a qualitative point Libraries, classifiers, and quantifiers: a comparison of chemometric methods for the analysis of Raman spectra of contaminated pharmaceutical materials, J. Pharm. of view, Raman intensity should be affected by particle size in a com- Biomed. Anal. 61 (2012) 191–198, https://doi.org/10.1016/j.jpba.2011.12.002. parable way than in NIR. Hence, an increment of intensity with particle [8] J. Arruabarrena, J. Coello, S. Maspoch, Raman spectroscopy as a complementary size should be expected. tool to assess the content uniformity of dosage units in break-scored warfarin ta- blets, Int. J. Pharm. 465 (2014) 299–305, https://doi.org/10.1016/j.ijpharm.2014. Our results are in a good agreement with this assumption. Raman 01.027. intensities are lower for the smaller particle size fractions and increase [9] D.A. Gómez, J. Coello, S. Maspoch, Raman spectroscopy for the analytical quality up to a size of approximately 40–50 μm. For larger particle sizes in- control of low-dose break-scored tablets, J. Pharm. Biomed. Anal. 124 (2016) 207–215, https://doi.org/10.1016/j.jpba.2016.02.055. tensity arrives to a plateau. An intuitive explanation to the intensity [10] T.R. De Beer, W.R. Baeyens, Y.V. Heyden, J.P. Remon, C. Vervaet, F. Verpoort, stabilization could be that the amount of sample that produces a Raman Influence of particle size on the quantitative determination of salicylic acid in a scattering able to leave the sample in a backscattering mode becomes pharmaceutical ointment using FT-Raman spectroscopy, Eur. J. Pharm. Sci. 30 (2007) 229–235, https://doi.org/10.1016/j.ejps.2006.11.009. constant. In fact, as predicted by theory [19], calculated by simulations [11] N. Townshend, A. Nordon, D. Littlejohn, J. Andrews, P. Dallin, Effect of particle [37] and experimentally verified in this work, Raman spectra recorded properties of powders on the generation and transmission of Raman scattering, in backscattering mode, comes from a thin sample layer which thick- Anal. Chem. 84 (2012) 4665–4670, https://doi.org/10.1021/ac203446g. ness depends on the particle size. In this work, the laboratory prepared [12] B. Schrader, G. Bergmann, Die Intensität des Ramanspektrums polykristalliner Substanzen, Fresenius’ Zeitschrift Für Anal. Chemie. 225 (1967) 230–247. tablets had a thickness of 1.4–1.5 mm. This value is approximately the [13] B. Schrader, A. Hoffmann, S. Keller, Near Infrared Fourier transform Raman spec- estimated penetration depth for compacted particles with a diameter in troscopy, facing absorption and background.pDf, Spectrochim. Acta Part A Mol. the range 40–50 μm. Consequently, the Raman intensity becomes con- Spectrosc. 47 (1991) 1135–1148. [14] P. Kubelka, F. Munk, Ein Beitrag zur Optik der Farbanstriche, Zeit. Für Tekn. Phys. stant for higher particle sizes, just because the whole tablet has been 12 (1931) 593–601. sampled. Taking into account the particle sizes used in pharmaceutical [15] S. Singh, H. Zeng, C. Guo, W. Cai (Eds.), Nanomaterials: Processing and tablets, a penetration depth of 1–1.5 mm has to be expected, and ac- Characterization With Lasers, Wiley-YCH, 2012. [16] A.K. Arora, M. Rajalakshmi, T.R. Ravindran, V. Sivasubramanian, Raman spectro- tually this should be considered in the design of the calibration sample scopy of optical phonon confinement in nanostructured materials, J. Raman set, as well as in the interpretation of the quantitative results obtained. Spectrosc. 38 (2007) 604–617, https://doi.org/10.1002/jrs.1684. The spectral preprocessing used in this work considerably reduces [17] S. Osswald, V.N. Mochalin, M. Havel, G. Yushin, Y. Gogotsi, Phonon confinement effects in the Raman spectrum of nanodiamond, Phys. Rev. B 80 (2009) 075419, , the effect of particle size on signal intensities, but it must be empha- https://doi.org/10.1103/PhysRevB.80.075419. sized that a small but significant difference has been still found for the [18] M.V. Pellow-Jarman, P.J. Hendra, R.J. Lehnert, The dependence of Raman signal smallest particle size (< 20 μm). Measurements reproducibility is also a intensity on particle size for crystal powders, Vib. Spectrosc. 12 (1996) 257–261, https://doi.org/10.1016/0924-2031(96)00023-9. function of particle size. RSD% in repeatability conditions (i.e. same [19] H. Wang, C.K. Mann, T.J. Vickers, Effect of powder properties on the intensity of position in tablet) represents the best possible value attainable and it is raman scattering by crystalline solids, Appl. Spectrosc. 56 (2002) 1538–1544. a function of the spectrometer set up (mainly spectral resolution and [20] Y. Hu, H. Wikström, S.R. Byrn, L.S. Taylor, Analysis of the effect of particle size on number of scans averaged). RSD% of values obtained in mapping mode polymorphic quantitation by raman spectroscopy, Appl. Spectrosc. 60 (2006) 977–984. are significantly higher than those obtained when measuring in the [21] N. Townshend, A. Nordon, D. Littlejohn, M. Myrick, J. Andrews, P. Dallin, same position, but in this case the spectral preprocessing practically Comparison of the determination of a low-concentration active ingredient in eliminates the elastic scattering variations, producing mapping var- pharmaceutical tablets by backscatter and transmission raman spectrometry, Anal. Chem. 84 (2012) 4671–4676, https://doi.org/10.1021/ac203447k. iances comparable to those obtained in repeatability conditions. Similar [22] Z.P. Chen, L.M. Li, J.W. Jin, A. Nordon, D. Littlejohn, J. Yang, J. Zhang, R.Q. Yu, trends are found both for the macro Raman system and the microscope, Quantitative analysis of powder mixtures by raman spectrometry: The influence of however for the latter, the lower amount of particles sampled, and the particle size and its correction, Anal. Chem. 84 (2012) 4088–4094, https://doi.org/ 10.1021/ac300189p. lower magnitude of the acquired signals, produce higher RSD% values. [23] P. Kristova, L.J. Hopkinson, K.J. Rutt, The effect of the particle size on the funda- mental vibrations of the [CO 3 2–] anion in calcite, J. Phys. Chem. A 119 (2015) 4891–4897, https://doi.org/10.1021/acs.jpca.5b02942. [24] A. Sparén, M. Hartman, M. Fransson, J. Johansson, O. Svensson, Matrix effects in Acknowledgements quantitative assessment of pharmaceutical tablets using transmission raman and near-infrared (NIR) spectroscopy, Appl. Spectrosc. 69 (2015) 580–589, https://doi. This research was funded by MINECO (Spain) through the project org/10.1366/14-07645. [25] C.H. Chio, S.K. Sharma, P.G. Lucey, D.W. Muenow, Effects of particle size and laser- CTQ 2016-79696-P (AEI/FEDER, EU). Raman spectrophotometer was induced heating on the raman spectra of alpha quartz grains, Appl. Spectrosc. 57 partly acquired with the financial support of AQU (Agencia per a la (2003) 774–783, https://doi.org/10.1366/000370203322102852. Qualitat del Sistema Universitari de Catalunya) through the project [26] M. De La Pierre, C. Carteret, L. Maschio, E. André, R. Orlando, R. Dovesi, The 2009 SGR 1470. Raman spectrum of CaCO3 polymorphs calcite and aragonite: a combined experi- mental and computational study, J. Chem. Phys. 140 (2014) 164509, , https://doi. 55 D.A. Gómez et al. Vibrational Spectroscopy 100 (2019) 48–56 org/10.1063/1.4871900. [34] R.L. McCreery, Raman Spectroscopy for Chemical Analysis, John Wiley & Sons, Inc., [27] B.Y. Shekunov, P. Chattopadhyay, H.H.Y. Tong, A.H.L. Chow, Particle size analysis 2000. in pharmaceutics: principles, methods and applications, Pharm. Res. 24 (2007) [35] B. Orel, D. Hadži, F. Cabassi, Infrared and Raman Spectra of potassium hydrogen 203–227, https://doi.org/10.1007/s11095-006-9146-7. phthalate, Spectrochim. Acta Part A Mol. Spectrosc. 31 (1975) 169–182. [28] B.R. Rohrs, G.E. Amidon, R.H. Meury, P.J. Secreast, H.M. King, C.J. Skoug, Particle [36] J. Arruabarrena, J. Coello, S. Maspoch, Enhancing sensitivity and precision on NIR size limits to meet USP content uniformity criteria for tablets and capsules, J. reflectance determination of an API at low concentration: application to an hor- Pharm. Sci. 95 (2006) 1049–1059, https://doi.org/10.1002/jps. monal preparation, J. Pharm. Biomed. Anal. 60 (2012) 59–64, https://doi.org/10. [29] Z.I. Hanan, J.M. Durgin, Pharmacy Practice for Technicians, 5th ed., Cencage 1016/J.JPBA.2011.10.026. Learning, New York, 2015. [37] P. Matousek, A.W. Parker, Bulk raman analysis of pharmaceutical tablets, Appl. [30] S. Šašić (Ed.), Pharmaceutical Applications of Raman Spectroscopy, John Wiley & Spectrosc. 60 (2006) 1353–1357, https://doi.org/10.1366/000370206779321463. Sons, Inc., 2008. [38] M. Blanco, J. Coello, H. Iturriaga, S. Maspoch, F. Gonzalez, R. Pous, Effect of the [31] S.J. Fraser, K.C. Gordon, Raman spectroscopy in the study of pharmaceuticals: the sample cell module and particle size on the NIRS determination of piracetam, Near problems and solutions to sub-sampling and data analysis, Riv. Eur. Sci. Med. Infrared Spectroscopy: Bridging the Gap Between Data Analysis and NIR Farmacol. 19 (2014) 3–8. Applications, Ellis Horwood, Chichester, 1992. [32] M. Csuros, Environmental Sampling and Analysis. Lab Manual, CRC Press, 1997. [39] D.J. Dahm, K.D. Dahm, Interpreting Diffuse Reflectance and Transmittance. A [33] S.E.J. Bell, J.R. Beattie, J.J. McGarvey, K.L. Peters, N.M.S. Sirimu, S.J. Speers, Theoretical Introduction to Absorption Spectroscopy of Scattering Materials, NIR Development of sampling methods for Raman analysis of solid dosage forms of Publications, 2007. therapeutic and illicit drugs, J. Raman Spectrosc. 35 (2004) 409–417, https://doi. [40] P. Williams, K. Norris (Eds.), Near-Infrared Technology in Agricultural and Food org/10.1002/jrs.1160. Industries, Am Assoc. Cereal Chemists, St Paul, Minnesota, 1987. 56