- Email: [email protected]
Influence of moisture variation on the performance of Raman spectroscopy in quantitative pharmaceutical analyses
Accepted Manuscript Title: Influence of Moisture Variation on the Performance of Raman Spectroscopy in Quantitative Pharmaceutical Analyses Authors: Md Nayeem Hossain, Benoˆıt Igne, Carl A. Anderson, James K. Drennen III PII: DOI: Reference:
S0731-7085(18)31529-2 https://doi.org/10.1016/j.jpba.2018.10.022 PBA 12271
To appear in:
Journal of Pharmaceutical and Biomedical Analysis
Received date: Revised date: Accepted date:
27-6-2018 12-10-2018 12-10-2018
Please cite this article as: Hossain MN, Igne B, Anderson CA, Drennen JK, Influence of Moisture Variation on the Performance of Raman Spectroscopy in Quantitative Pharmaceutical Analyses, Journal of Pharmaceutical and Biomedical Analysis (2018), https://doi.org/10.1016/j.jpba.2018.10.022 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Influence of Moisture Variation on the Performance of Raman Spectroscopy in Quantitative Pharmaceutical
U
SC R
Md Nayeem Hossaina, Benoît Igneb, Carl A. Andersona,b and James K. Drennen IIIa,b,*
IP T
Analyses
a
A
N
Duquesne University Graduate School for Pharmaceutical Sciences, Pittsburgh, Pennsylvania 15282
b
Corresponding author at: Duquesne University Center for Pharmaceutical Technology, Duquesne
PT
*
ED
M
Duquesne University Center for Pharmaceutical Technology, Pittsburgh, Pennsylvania 15282
CC E
University, Pittsburgh, Pennsylvania 15282
A
Tel.: 1-412-396-1172, Fax: 1-412-396-4660, E-mail: [email protected]
Graphical abstract
IP T SC R U N A M
Raman spectral variations of the pharmaceutical materials were observed due to the
PT
ED
Highlight
presence of water.
Unwanted spectral baseline change due to moisture variation caused spectral
CC E
A
inconsistency across calibration and test samples which degraded quantitative model
performance.
Including moisture variations into the calibration set significantly improved Raman spectroscopic model performance.
Abstract
To develop a robust quantitative calibration model for spectroscopy, different sources of variability that
IP T
are not directly related to the components of interest should be included in the calibration samples; this variability should be similar to that which is anticipated during validation and routine operation. Moisture
SC R
content of pharmaceutical samples can vary as a function of supplier, storage conditions, geographic
origin or seasonal variation. Additionally, some pharmaceutical operations (e. g., wet granulation) cause
N
U
exposure of excipients and API to water. Although water is a weak Raman scatterer, moisture variability
A
has an indirect effect on analytical model performance. Because many pharmaceutical components have
M
intrinsic fluorescent characteristics (with broad spectral features), moisture variability may cause spectral
ED
artifacts in the form of baseline variation associated with fluorescence quenching. This work investigates the deleterious effects of water quenching on quantitative prediction accuracy of a multivariate
PT
calibration algorithm for Raman spectroscopy. To demonstrate this, a formulation composed of
CC E
acetaminophen, lactose, microcrystalline cellulose, HPMC and magnesium stearate was used. Tablets were manufactured using laboratory scale equipment. A full-factorial design was used to vary
A
acetaminophen (5 levels), and excipient ratios (3 levels) to generate tablets for calibration and testing. Tablet moisture variation was introduced by placing samples in different humidity chambers. Significant spectral effects arising from fluorescence were identified in the Raman spectra and due to moisture variation: the fluorescence related spectral variability caused substantial degradation of the prediction
performance for API. The work demonstrated that accounting for moisture variation during method
IP T
development reduced the prediction error of the multivariate prediction model.
SC R
Keywords: Raman Spectroscopy; Pharmaceutical; Quantitative analysis; Robustness; Moisture;
M
A
N
U
Fluorescence; PLS modelling.
PT
ED
1. Introduction
CC E
In the nearly seventeen years since the start of the PAT (Process Analytical Technology) initiative, spectroscopic techniques have become more widely used for designing, monitoring and controlling pharmaceutical unit operations. Such PAT methods have become an important component of a
A
comprehensive quality management system[1]. Among these, Raman spectroscopy has emerged as a promising process analytical tool due to its sharp peak features offering inherently good chemical specificity, its non-destructive nature, and the flexibility in sampling interfaces offering various opportunities for in-line, on-line or off-line analysis of solid oral solid dosage forms. Raman spectroscopy
has proven valuable for quantitative prediction of critical components of pharmaceutical products[2], managing critical process parameters[3-5], and for assuring that products meet real-time release criteria[6, 7].
IP T
Before implementing a quantitative spectroscopic prediction model, calibration model development is required. While developing Raman methods, it is necessary to ensure that models meet the required
SC R
performance standards; in particular, the accuracy, precision, and robustness of the model must meet predetermined validation criteria. A variety of multivariate calibration techniques have been used to
U
create models that were considered fit-for-purpose [8-10]. Though these methods allow the analyst to
A
N
extract relevant chemical and physical information about samples, numerous sources of spectral
M
variability that occur during the method life-cycle may confound measurements. To ensure long-term
ED
model performance, model robustness may be enhanced by including such sources of spectral variability in the calibration model. These sources of variability may be identified during a risk assessment.
PT
Understanding the impact of such variability on the spectra and the effect of these spectral changes on
CC E
model performance is a critical part of a sound model development process [10, 11].
Many sources of spectral variability arise during the pharmaceutical manufacturing process, including
A
sampling, analytical instrument instability, sample properties and environmental fluctuation. Sample variability may be introduced through raw material or process variability or as a result of sampling procedure: the time, location and method of sampling is often at fault. Typically, lot-to-lot differences of incoming raw materials introduce new physical and chemical variations [12, 13]. The consequence of
these variations on spectra may depend on the sampling configuration, spectral collection modalities, etc. [14-16]. As an example, the effect of particle size variation in tablets has been compared for backscatter and transmission Raman spectroscopy[17]. Because backscatter Raman generates most of the signal
IP T
from very near the tablet surface [18], particle size had a reduced impact relative to the transmission mode. In the case of transmission measurements, photon propagation characteristics were more
SC R
influenced by particle size variation due to the effect of light scattering on the effective optical path
length. When particle size variability was not included in the calibration set, model accuracy suffered.
U
Spectral processing methods such as derivative, normalization, etc., were applied to remove the baseline
N
effect of particle size variation, thereby improving model performance. Also, by matching particle size
M
A
variability for both API and excipients in the calibration, adequate predictions were obtained for independent test sets [17, 19-21]. A change of size, shape or density of tablets due to changes in the
ED
manufacturing process can also introduce spectral variability [22, 23]. By adjusting laser power and
PT
accumulation time, or enhancing density range in the calibration model, robustness can be enhanced
CC E
against tablet size and density variations [20, 22].
While the application of Raman in pharmaceutical applications has increased dramatically over the past
A
decade due to the improvement of instrumentation [24-26], one continuing challenge, and a major source of spectral variability causing interference with the Raman spectral signal, is fluorescence. Produced by emission of photons from low-lying excited electronic states, fluorescence often masks the vibrational shift of Raman spectra. This effect occurs in a sample whenever the frequency of the excitation laser coincides with transition energy from the ground to an electronically excited state.
Different experimental strategies and instrument modifications have been developed to overcome this problem[27]. By using excitation sources in the near-infrared region coupled with multichannel detectors, modern dispersive Raman spectrometers decrease the fluorescence background and provide adequate
IP T
Raman intensity[28]. Multivariate regression and spectral preprocessing methods have been used to reduce the impact of fluorescence signal on prediction results [29-32]. Moreover, model performance
background without introducing other spectral disturbances[33].
SC R
should be adequate as long as both calibration and validation sets contain similar fluorescence
U
One of the challenging sources of variability that analysts often encounter during spectroscopic model
A
N
development and deployment is environmental variability; for example, relative humidity will often vary
M
from the warehouses where raw materials are stored to the facilities where products are manufactured.
ED
This can cause significant variability in the amounts of water sorbed by pharmaceutical samples [34-36]. Moreover, different pharmaceutical unit operations, for example high shear granulation or film-coating
PT
processes, introduce significant amounts of water which can create batch to batch moisture variability
CC E
[23, 37]. Most often, Raman analytical methods ignore the effect of moisture because Raman presents relatively weak water scattering. However, because pharmaceutical samples exhibit fluorescence from
A
Active Pharmaceutical Ingredient (API) [38] , excipient [30, 31], dyes [32], capsule shells [39], etc., the effect of moisture as a fluorescence quencher needs to be considered [40]. This can introduce unwanted spectral inconsistency across calibration and test samples.
The objective of this paper is to investigate the effect of moisture and fluorescence on Raman spectroscopic analysis for pharmaceutical solid oral dosage forms. The effect of these changes on Raman spectra and multivariate quantitative model performance was assessed, with the aim to provide strategies
IP T
for the development of robust calibration models.
2. Material and Method
SC R
2.1. Samples and Environment conditions
Granules of acetaminophen (APAP; Mallinckrodt Inc., Raleigh, NC, USA), hydroxypropyl methylcellulose
N
U
(HPMC; Pharmacoat 606, Shin-Etsu Chemical Co. LTD, Tokyo, Japan) and lactose (modified spray-dried;
A
Foremost Farms USA, Rothschild, WI, USA) were manufactured using a fluid bed granulator (model WSG
M
5, Glatt, Binzen, Germany). The active containing granules were mixed with extra-granular MCC (MCC;
ED
Avicel PH 200, FMC Biopolymer, Mechanicsburg, PA, USA), lactose and magnesium stearate (Fisher Scientific, Waltham, MA, USA) in a bin blender. Tablets comprised of the aforementioned components
PT
were manufactured at laboratory scale using a Carver Automatic Tablet Press (Model 3887. 1SD0A00,
CC E
Wabash, IN, USA) and 13 mm die and flat-faced punches. Target tablet weight was 700 mg. HPLC was performed on the tablets to ensure uniformity of blend and tablets.
A
A five by three level, two factor [active content (19. 11%, 23. 21%, 27. 30%, 31. 40% and 35. 49% w/w; and excipient ratio levels 2, 1, 0. 5)] full factorial design was created. The use of a full-factorial experimental design provided orthogonality between the active ingredient and the excipient ratios. While it would have been possible to have all other components vary independently from the active ingredient,
the use of excipient ratio helped to reduce the number of design points. Table 1 summarizes the design. Three replicate tablets per level were created from the same blend for a total of 45 samples. From each design point, one tablet was randomly chosen to create a test set. Therefore, a total of 30 tablets were
IP T
used for calibration and 15 tablets were included in the test set.
After allowing the tablets to undergo viscoelastic relaxation (two weeks), they were placed in
SC R
environmental chambers at room temperature. The four chambers equilibrated with a relative humidity (RH) of 11% (saturated lithium chloride), 32% (saturated magnesium chloride), 52% (saturated
U
magnesium nitrate) and 75% (saturated sodium chloride). Tablets were left to equilibrate in each
A
N
chamber, beginning with the lowest moisture level. Once the mass of tablets had equilibrated in the 11%
M
RH chamber, they were analyzed and moved to the 32% RH chamber. The equilibration and analysis was
ED
repeated for each chamber. Tablets were moved from low to high relative humidity to simplify the experimental design and reduce the experimental period.
PT
2.2. Fluorescence Quenching Effect
CC E
To understand the nature of spectral change due to fluorescence quenching, spectra of pure components were collected in the presence of water and molecular gas, oxygen (O2) and nitrogen (N2). To investigate
A
the impact of water, pure component compacts were created by compressing approximately 700 mg of each material in the Carver press. Then, these pure component compacts were stored in the four aforementioned humidity conditions to observe the effect of moisture on Raman spectra.
However, spectra were collected in the presence of O2 and N2 using custom gas chambers rather than a humidity chamber.
2.3. Raman Data Collection
IP T
Calibration and test sets of Raman spectra were collected on a TRS 100 (Cobalt Light Systems Ltd., Abington, Oxfordshire, UK) for analysis of compacts. The transmission spectrometer is equipped with a
SC R
diode laser operating at 830 nm as the radiation source, a long pass spectral filter and a thermoelectrically cooled CCD camera for the detector. All compacts were scanned over the wavenumber range of 38-2400
N
U
cm-1. The measurements were conducted at room temperature (approximately 24oC). Parameters were
A
tuned according to the tablet scale: exposure time for lab scale and manufacturing scale samples was 0. 5
M
sec and 1.3 seconds, respectively. Five accumulations were collected per spectrum with a laser power of
ED
0. 6W.
PT
The effect of O2 and N2 gasses on the spectra of powdered samples was evaluated with a RamanRxn2TM backscattering spectrometer (Kaiser Optical Systems, Inc, Ann Arbor, MI, USA) and iCRaman software
CC E
(version 4. 1, Mettler Toledo, Columbus, OH, USA). A fiber-coupled PhAT probe (HoloGRAMS version 4. 0, Kaiser Optical Systems, Inc, Ann Arbor, MI, USA) was equipped with a 400 mW laser at 785 nm. The
A
diameter of laser spot size was 6 mm. Accumulation time was 3 seconds, and the spectral range was 382400 cm-1.
2.4. Reference Testing
Acetaminophen reference values for all compacts were determined using high-pressure liquid chromatography (HPLC; Waters Alliance 2790, Milford, MA, USA), followed by ultraviolet detection (Waters 2487). A method, modified from USP29-NF24 (‘‘Acetaminophen Tablets’’), was employed [41].
IP T
The mobile phase was water: methanol: acetic acid (80: 17: 3), and the stationary phase was a 15 cm by 4. 6 mm C18 column with 3 µm packing. The detection wavelength was 243 nm. The error of the reference
SC R
measurement was estimated at 0. 83%. Reference testing was performed after all laboratory tablets had been equilibrated and scanned at all four RH conditions.
N
U
2.5. Model Evaluation
A
In order to develop a calibration model for the prediction of API concentration and to determine the
M
potential impact of moisture variation, partial least-squares regression (PLS) was used with the nonlinear
ED
iterative partial least squares (NIPALS) algorithm [9]. The capability of a model developed at one moisture level to predict the active ingredient concentration of compacts stored at other relative humidity levels
PT
was tested. The prediction error was used to evaluate the effect of moisture variability on model
CC E
performance. Various diagnostic tests were employed to understand the underlying causes of model performance.
A
2.6. Performance measurement by Prediction error
The root-mean-square error (RMSE) was used as a performance criterion. The choice of latent variables was determined by comparing the evaluation of the root mean-square error of calibration (RMSEC) and of cross-validation (RMSECV). The ability of a model developed at one moisture level to predict accurately
an active ingredient concentration of tablets stored at other relative humidity conditions was tested by prediction error (RMSEP):
(1)
IP T
∑𝑛 (𝑦̂𝑖 − 𝑦𝑖 )2 𝑅𝑀𝑆𝐸𝑃 = √ 𝑖=1 𝑛
SC R
where 𝑦̂ and 𝑦 represent vectors of predicted and reference values, respectively, and 𝑛 the number of samples.
N
U
2.7. Diagnostic Tests
A
Statistics such as the Hotelling’s 𝑇 2 and the 𝑄-residuals (𝑄𝑟 ) can be used to ensure the suitability of a
M
model for predicting any sample. Hotelling’s 𝑇 2 tests the membership of an observation to the
ED
population forming the calibration set. 𝑄𝑟 determines whether a sample spectrum is different due to an
failure, etc. ) [42].
PT
un-modelled source of variance, including a possible instrumental fault ( e. g., sampling error, component
𝑇𝑖2 = 𝑡𝑖 (
(𝑇 𝑇 𝑇) −1 ) 𝑡𝑖 𝑛
(2)
A
CC E
Hotelling’s 𝑇 2 was defined as follows:
where 𝑡𝑖 is the latent variable scores for 𝑡𝑖 th sample spectrum and 𝑇 is the matrix of calibration sample model scores.
𝑄𝑟 was calculated as:
𝑄𝑟 = 𝑠𝑢𝑚(𝑥𝑖 − 𝑡𝑖 𝑃𝑇 )2
IP T
(3)
where 𝑄𝑟 represents sum of squared reconstruction error, 𝑥𝑖 is the sample spectrum, 𝑡𝑖 is the latent
SC R
variable scores for the 𝑖 𝑡ℎ sample spectrum, and P is the model loading. More information about these statistics and their use for process monitoring can be found in the literature[42].
N
U
4. Results
A
4.1. Effect of moisture variability on Raman spectroscopy
M
Figure 1 shows spectral baseline changes when tablets were exposed to four different moisture levels
ED
(11% (red), 32% (blue), 52% (green) and 75% (magenta) RH). It can be observed that increasing the water
PT
content in the tablet leads to a decrease in the fluorescent background. Mechanistically, the decrease in fluorescence intensity was caused by the deactivation of the excited-state fluorophores in contact with
CC E
certain molecules (e.g., water in this study) in the sample. These molecules are referred to as quenchers [39]. This spectral change, due to the presence of water, was much more significant than the effect of
A
photobleaching (due to successive exposure of the sample to the laser). The effect of photobleaching is illustrated in Figure 2, where a tablet was stored at a relative humidity 52% and scanned repeatedly over eight weeks. With an increasing number of scans and storage time, there was a small decrease in the baseline due to photobleaching.
To understand the source of spectral variation in the presence of moisture, a variety of compacts were prepared using pure components; the compacts were then exposed to different moisture levels. Figure 3 reveals that MCC and HPMC compacts demonstrated a significant decrease in the baseline with increasing
IP T
water adsorption, which may be due to fluorescence quenching (Figure 3). However, APAP, Lactose and Magnesium Stearate spectra demonstrated very small changes in the different conditions. This is likely
SC R
due to the absence of fluorescence with these three pure components. The results indicate that the
primary source of spectral variation in the prepared samples was from MCC and HPMC, with water acting
U
as a quencher.
A
N
In Figure 4, fluorescence quenching for MCC and HPMC was demonstrated in the presence of oxygen, a
M
known quenching agent in cellulosic materials [42]. Previous studies demonstrated that, due to the
ED
presence of lignin in the fiber component, a large number of pulp and paper samples exhibit laser induced fluorescence (LIF). However, in the presence of molecular oxygen, lignin caused LIF contributions to
PT
Raman spectra were significantly decreased[43]. Figure 4 shows that due to the presence of O2, the
CC E
spectral baseline of MCC changed over time. However, when spectra were collected in the presence of N2 or air, quenching was reduced. This result suggests that fluorescence decreases dramatically in the
A
presence of the quenching agent O2. A small spectral change was observed with MCC in the presence of air due to the continuous laser exposure. This provides another demonstration that the effect of photobleaching was less than that of quenching associated with moisture. Acetaminophen and Lactose spectra demonstrated limited spectral change due to the lack of fluorophores.
The effect of fluorescence quenching on spectral baseline has been shown to vary with the content of both fluorophores and moisture. To characterize the impact of these two factors, spectral slope and intensity was observed for raw spectra. The slope of the spectra was calculated from a simple least-
IP T
squares regression between Raman shift and Raman intensity. Three Raman shifts (276, 750, 1404 cm -1) were chosen from the spectra to calculate this slope. The slope values are displayed in Table 2, where
SC R
results from four moisture levels and three excipient ratios (2:1, 1:1 and 0. 5:1 of MCC: Lactose) are presented.
U
First, the amount of fluorescent components affected the baseline intensity, as anticipated. Fluorescence
A
N
intensity decreased with a decreasing amount of MCC (for a given moisture level). This change was
M
reflected in the slope. Second, the moisture content had a significant impact on the baseline: the Raman
ED
intensity decreased with increasing water content (for a given compact composition), causing a dramatically reduced slope with increasing moisture content (Table 2). These results emphasize the
PT
quenching effect of water on Raman spectra.
CC E
4.2. Effect of sample moisture variability on quantitative model performance
The observed spectral variations due to the presence of water had a significant impact on quantitative
A
model performance. Performance was evaluated by developing four models (one per relative humidity condition). These models were then used to predict four test sets corresponding to the four relative humidity conditions.
A number of preprocessing methods such as baseline weighted least squares, normalization, mean centering, derivative, and detrending were evaluated to remove the baseline changes of the full spectral range described previously. Random block cross-validation (five blocks) was used to determine the most
IP T
appropriate spectral pre-treatments and model complexity (number of latent variables). An optimization was independently conducted based on the minimization of cross-validation error. Savitzky-Golay first
SC R
derivative (window size-31, polynomial order-2) followed by normalization to unit area and mean-
centering was chosen for all models. Derivative methods were helpful to remove baseline variations and
U
normalization decreased the intensity of variations introduced due to the fluorescence quenching or any
N
physical differences between calibration and test sets. The responses were auto-scaled (zero mean, unit
M
A
variance).
ED
As shown in Table 3, models offered acceptable error when predicting API content for samples from the same moisture levels, where the preprocessing method was successful at removing fluorescent baseline
PT
variability. However, poor results were achieved for prediction of samples from unique moisture levels;
CC E
error statistics increased with increasing range between the moisture levels of calibration and test sets. The prediction error was largely due to bias, as seen in Figure 5A. These results may be interpreted as a
A
lack of robustness of the calibration model to moisture variability.
4.3. Robustness enhancement methods
Different strategies are applied to develop a model that is robust against different sources of variations. In the present study, a global modeling approach was used as a means to improve model accuracy.
Global models include all of the expected variability in the calibration set. In this case, calibration sets from four relative humidity levels were combined to facilitate the creation of two models, including: 1) a model using the full wavelength range(38-2400 cm-1 ), and 2) a model including a truncated wavelength
IP T
region (845-900 cm-1 ) that contained the APAP peak.
Prediction errors for the test set at the different humidity levels are provided in Table 3. The global model
SC R
demonstrated a lower prediction error compared to models created at unique humidity conditions (Figure 6A). Also of significance, the model created using the full variable range provided similar
U
prediction performance compared to the truncated variable range.
A
N
Figure 6B presents Hotelling’s 𝑇 2 and 𝑄 – residual plots for global modeling compared to Figure 5(B). The
M
reconstruction error for all the test sets was lower for global modeling, indicating that the global
ED
calibration model was more robust against moisture variation, which ultimately improved prediction
PT
performance.
Global models are practical when the number of interfering elements is limited. Their ability to handle
CC E
large ranges of variability in components not related to the parameter of interest will be seen through improved method robustness, although sometimes at the cost of reduced standard error of prediction.
A
5. Summary and Conclusion: It is difficult to preemptively identify all sources of variation that will be encountered during the pharmaceutical life-cycle. The experimental approach in this study provides an example of how a simple
phenomenon such as moisture variability, arising from such phenomena as inconstant environmental conditions, can, through fluorescence, significantly affect spectral baseline and model performance.
In this study, the presence of components containing 1) fluorophores and 2) moisture, negatively affected
IP T
model performance. Pharmaceutical raw materials often contain fluorescent components. In this study, lignin in the microcrystalline cellulose caused laser induced fluorescence [43, 44]. Moreover, different
SC R
lots of excipients may exhibit variable amounts of lignin, thus impacting the fluorescent background.
Variable moisture content also introduced spectral variance. The water content of tablets will often vary
U
between and within different tablet batches, depending on the relative humidity conditions during
A
N
production, packaging, storage and analysis. Water may be sorbed by samples very quickly depending on
M
the relative humidity of the environment and material attributes [34]. While it is likely that the relative
ED
humidity of the laboratory may be controlled, it is very unlikely that the same environmental conditions will be observed in the packaging facilities, warehouses and manufacturing plants. This is particularly true
PT
in situations where spectroscopic method development takes place at one geographic and climatic
CC E
location and routine use of the method occurs elsewhere[34].
It is typically beyond the capacity of an analyst to eliminate sample components or strictly control the
A
sources of variability. Understanding the effect of variabilities such as moisture and fluorescence during multivariate calibration model development will facilitate the development of calibration models which are robust against these variations. In the present study, the effect of fluorescence background was
reduced by preprocessing the spectra. However, when calibration and test sets contained different moisture content, prediction performance was degraded.
An understanding of potential confounding variables is key to the development of robust models.
IP T
Without that knowledge, a tendency to generate less robust models with better performance statistics will supersede the long-term interests of analytical methods that are robust over the lifecycle of a
SC R
product. Global calibration methods were tested in this work to demonstrate opportunities for the
M
A
N
U
creation of robust models for laboratory and production scale tablets.
PT
ED
Acknowledgement
CC E
The authors are grateful to Agilent Technologies, Inc (previously Cobalt Light Systems Ltd.) for the use of the TRS 100 transmission Raman optical engine and the technical support received during the research
A
period.
References
[1] Food, D. Administration, Guidance for industry: PAT—A framework for innovative pharmaceutical development, manufacturing, and quality assurance, DHHS, Rockville, MD, (2004). [2] C.J. Strachan, T. Rades, K.C. Gordon, J. Rantanen, Raman spectroscopy for quantitative analysis of
IP T
pharmaceutical solids, Journal of pharmacy and pharmacology, 59 (2007) 179-192. [3] A. Hédoux, Y. Guinet, M. Descamps, The contribution of Raman spectroscopy to the analysis of phase
SC R
transformations in pharmaceutical compounds, International journal of pharmaceutics, 417 (2011) 17-31. [4] T. De Beer, A. Burggraeve, M. Fonteyne, L. Saerens, J.P. Remon, C. Vervaet, Near infrared and Raman
U
spectroscopy for the in-process monitoring of pharmaceutical production processes, International Journal
N
of Pharmaceutics, 417 (2011) 32-47.
M
A
[5] H. Wikström, P.J. Marsac, L.S. Taylor, In‐line monitoring of hydrate formation during wet granulation using Raman spectroscopy, Journal of pharmaceutical sciences, 94 (2005) 209-219.
ED
[6] J. Müller, K. Knop, M. Wirges, P. Kleinebudde, Validation of Raman spectroscopic procedures in
PT
agreement with ICH guideline Q2 with considering the transfer to real time monitoring of an active
CC E
coating process, Journal of pharmaceutical and biomedical analysis, 53 (2010) 884-894. [7] H. Trnka, J. Rantanen, H. Grohganz, Well-plate freeze-drying: a high throughput platform for screening of physical properties of freeze-dried formulations, Pharmaceutical development and technology, 20
A
(2015) 65-73.
[8] D.M. Haaland, E.V. Thomas, Partial least-squares methods for spectral analyses. 1. Relation to other quantitative calibration methods and the extraction of qualitative information, Analytical Chemistry, 60 (1988) 1193-1202.
[9] S. Wold, M. Sjöström, L. Eriksson, PLS-regression: a basic tool of chemometrics, Chemometrics and intelligent laboratory systems, 58 (2001) 109-130. [10] T. Naes, T. Isaksson, T. Fearn, T. Davies, A user friendly guide to multivariate calibration and
IP T
classification, NIR publications 2002. [11] P.C. Williams, Implementation of near-infrared technology, Near-infrared technology in the
SC R
agricultural and food industries, 2 (2001) 143-167.
[12] B. Igne, Z. Shi, J.K. Drennen, C.A. Anderson, Effects and Detection of Raw Material Variability on the
U
Performance of Near‐Infrared Calibration Models for Pharmaceutical Products, Journal of pharmaceutical
N
sciences, 103 (2014) 545-556.
M
A
[13] V.S. Dave, S.D. Saoji, N.A. Raut, R.V. Haware, Excipient variability and its impact on dosage form functionality, Journal of pharmaceutical sciences, 104 (2015) 906-915.
ED
[14] M.V. Pellow-Jarman, P.J. Hendra, R.J. Lehnert, The dependence of Raman signal intensity on particle
PT
size for crystal powders, Vibrational Spectroscopy, 12 (1996) 257-261.
CC E
[15] H. Wang, C.K. Mann, T.J. Vickers, Effect of powder properties on the intensity of Raman scattering by crystalline solids, Applied spectroscopy, 56 (2002) 1538-1544. [16] T. De Beer, W. Baeyens, Y. Vander Heyden, J.P. Remon, C. Vervaet, F. Verpoort, Influence of particle
A
size on the quantitative determination of salicylic acid in a pharmaceutical ointment using FT-Raman spectroscopy, European journal of pharmaceutical sciences, 30 (2007) 229-235.
[17] N. Townshend, A. Nordon, D. Littlejohn, M. Myrick, J. Andrews, P. Dallin, Comparison of the determination of a low-concentration active ingredient in pharmaceutical tablets by backscatter and transmission Raman spectrometry, Analytical chemistry, 84 (2012) 4671-4676.
IP T
[18] N. Townshend, A. Nordon, D. Littlejohn, J. Andrews, P. Dallin, Effect of particle properties of powders on the generation and transmission of Raman scattering, Analytical chemistry, 84 (2012) 4665-4670.
SC R
[19] Y. Hu, H. Wikström, S.R. Byrn, L.S. Taylor, Analysis of the effect of particle size on polymorphic quantitation by Raman spectroscopy, Applied spectroscopy, 60 (2006) 977-984.
U
[20] A. Sparén, M. Hartman, M. Fransson, J. Johansson, O. Svensson, Matrix effects in quantitative
N
assessment of pharmaceutical tablets using transmission Raman and near-infrared (NIR) spectroscopy,
M
A
Applied spectroscopy, 69 (2015) 580-589.
[21] D. Oelkrug, E. Ostertag, R.W. Kessler, Quantitative Raman spectroscopy in turbid matter: reflection or
ED
transmission mode?, Analytical and bioanalytical chemistry, 405 (2013) 3367-3379.
PT
[22] H. Wikström, S. Romero-Torres, S. Wongweragiat, J.A.S. Williams, E.R. Grant, L.S. Taylor, On-line
CC E
content uniformity determination of tablets using low-resolution Raman spectroscopy, Applied spectroscopy, 60 (2006) 672-681. [23] B. Igne, J.K. Drennen, C.A. Anderson, Improving Near-Infrared Prediction Model Robustness with
A
Support Vector Machine Regression: A Pharmaceutical Tablet Assay Example, Applied spectroscopy, 68 (2014) 1348-1356. [24] H. Swierenga, A. De Weijer, R. Van Wijk, L. Buydens, Strategy for constructing robust multivariate calibration models, Chemometrics and Intelligent Laboratory Systems, 49 (1999) 1-17.
[25] J. Johansson, A. Sparén, O. Svensson, S. Folestad, M. Claybourn, Quantitative transmission Raman spectroscopy of pharmaceutical tablets and capsules, Applied spectroscopy, 61 (2007) 1211-1218. [26] R.L. McCreery, Raman spectroscopy for chemical analysis, John Wiley & Sons 2005.
IP T
[27] W.R. Browne, J.J. McGarvey, The Raman effect and its application to electronic spectroscopies in metal-centered species: Techniques and investigations in ground and excited states, Coordination
SC R
Chemistry Reviews, 251 (2007) 454-473.
[28] J.M. Chalmers, H.G. Edwards, M.D. Hargreaves, Infrared and Raman spectroscopy in forensic science,
U
John Wiley & Sons 2012.
N
[29] J.F. Kauffman, M. Dellibovi, C.R. Cunningham, Raman spectroscopy of coated pharmaceutical tablets
biomedical analysis, 43 (2007) 39-48.
M
A
and physical models for multivariate calibration to tablet coating thickness, Journal of pharmaceutical and
ED
[30] Y. Li, B. Igne, J.K. Drennen, C.A. Anderson, Method development and validation for pharmaceutical
CC E
(2016) 318-325.
PT
tablets analysis using transmission Raman spectroscopy, International journal of pharmaceutics, 498
[31] R.N. Palermo, S.M. Short, C.A. Anderson, H. Tian, J.K. Drennen III, Determination of figures of merit for near-infrared, Raman and powder X-ray diffraction by net analyte signal analysis for a compacted
A
amorphous dispersion with spiked crystallinity, Journal of Pharmaceutical Innovation, 7 (2012) 56-68. [32] S. Romero-Torres, J.D. Pérez-Ramos, K.R. Morris, E.R. Grant, Raman spectroscopy for tablet coating thickness quantification and coating characterization in the presence of strong fluorescent interference, Journal of pharmaceutical and biomedical analysis, 41 (2006) 811-819.
[33] I. Barman, C.-R. Kong, G.P. Singh, R.R. Dasari, Effect of photobleaching on calibration model development in biological Raman spectroscopy, Journal of biomedical optics, 16 (2011) 011004. [34] B. Igne, N. Hossain, J.K. Drennen III, C.A. Anderson, Robustness considerations and effects of moisture
IP T
variations on near infrared method performance for solid dosage form assay, Journal of Near Infrared Spectroscopy, 22 (2014) 179-188.
SC R
[35] S. Pieters, W. Saeys, T. Van den Kerkhof, M. Goodarzi, M. Hellings, T. De Beer, Y. Vander Heyden, Robust calibrations on reduced sample sets for API content prediction in tablets: Definition of a cost-
U
effective NIR model development strategy, Analytica chimica acta, 761 (2013) 62-70.
N
[36] T. Van den Kerkhof, R. De Maesschalck, K. Vanhoutte, M. Coene, Augmentation of near infrared
M
A
diffuse reflectance and transmittance spectral data for the development of robust PLSBC models for classifying double blind clinical trial tablets, Journal of pharmaceutical and biomedical analysis, 42 (2006)
ED
517-522.
PT
[37] D.M. Parikh, Handbook of pharmaceutical granulation technology, CRC Press 2016.
CC E
[38] M. Hajjou, Y. Qin, S. Bradby, D. Bempong, P. Lukulay, Assessment of the performance of a handheld Raman device for potential use as a screening tool in evaluating medicines quality, Journal of pharmaceutical and biomedical analysis, 74 (2013) 47-55.
A
[39] M.D. Hargreaves, N.A. Macleod, M.R. Smith, D. Andrews, S.V. Hammond, P. Matousek, Characterisation of transmission Raman spectroscopy for rapid quantitative analysis of intact multicomponent pharmaceutical capsules, Journal of pharmaceutical and biomedical analysis, 54 (2011) 463468.
[40] S. Tobita, K. Ida, S. Shiobara, Water-induced fluorescence quenching of aniline and its derivatives in aqueous solution, Research on Chemical Intermediates, 27 (2001) 205-218. [41] Acetaminophen Tablets. United States Pharmacopeial Monograph 29, USP29-NF24.
IP T
[42] R.P. Cogdill, C.A. Anderson, J.K. Drennen, Process analytical technology case study, part III: calibration monitoring and transfer, AAPS PharmSciTech, 6 (2005) E284-E297.
SC R
[43] U.P. Agarwal, An overview of Raman spectroscopy as applied to lignocellulosic materials, Advances in lignocellulosics characterization, (1999) 201-225.
U
[44] G. Thoorens, F. Krier, B. Leclercq, B. Carlin, B. Evrard, Microcrystalline cellulose, a direct compression
N
binder in a quality by design environment—A review, International journal of pharmaceutics, 473 (2014)
A
CC E
PT
ED
M
A
64-72.
Figure Captions
Figure 1: Raw spectra of calibration tablets at different moisture conditions: 11% RH (red), 32% RH (blue),
IP T
52% RH (green), and 75% RH (magenta).
SC R
Figure 2: Raw spectra of a tablet (target formulation) exposed to repeat scans over eight weeks to test the photobleaching effect.
U
Figure 3: Pure component raw spectra of Acetaminophen, Hypromellose, Lactose, Magnesium Stearate,
M
A
75% RH in red, blue, green and magenta respectively)
N
and Microcrystalline Cellulose stored in four different humidity conditions (11% RH, 32% RH, 52% RH and
Figure 4: Raman spectra collected in the presence of molecular gas A) MCC in presence of O2, B) MCC in
ED
presence of N2, C) Acetaminophen in presence of air, D) Acetaminophen in the presence of O2 (all are raw
PT
spectra).
CC E
Figure 5: (A) Concentration correlation plot of predicted vs. reference APAP concentrations corresponding to prediction samples from different humidity conditions: 11% RH (red), 32% RH (blue), 52% RH (green)
A
and 75% RH (magenta). The calibration model was displayed with Black dots. (B) Hotelling’s 𝑇 2 vs. 𝑄residuals plot was generated from the same tablets, represented with similar colors.
Figure 6: (A) Concentration correlation plots of predicted versus reference APAP concentrations corresponding to prediction samples from 11% RH (red), 32% RH (blue), 52% RH (green) and 75% RH (magenta). The global calibration model performance is displayed with black dots. (B) The 𝑇 2 vs 𝑄-
A
CC E
PT
ED
M
A
N
U
SC R
IP T
residuals plot was generated from similar tablets, represented by similar colors.
A ED
PT
CC E
IP T
SC R
U
N
A
M
Fig 1
A ED
PT
CC E
IP T
SC R
U
N
A
M
Fig 2
A ED
PT
CC E
IP T
SC R
U
N
A
M
Fig 3
A ED
PT
CC E
IP T
SC R
U
N
A
M
Fig 4
A ED
PT
CC E
IP T
SC R
U
N
A
M
Fig 5
A ED
PT
CC E
IP T
SC R
U
N
A
M
Fig 6
Table 1: Experimental Design (%W/W).
Acetaminophen
Hypromellose
Intragranular
Extragranular
lactose
Microcrystalline cellulose
Magnesium Stearate
Excipient ratio
lactose
19. 11
2. 73
5. 46
51. 77
20. 43
0. 50
2. 0
2
23. 21
3. 32
6. 63
48. 65
17. 70
0. 50
2. 0
3
27. 30
3. 90
7. 80
45. 53
14. 97
0. 50
2. 0
4
31. 40
4. 49
8. 97
42. 41
12. 24
0. 50
2. 0
5
35. 49
5. 07
10. 14
39. 29
9. 51
0. 50
2. 0
6
19. 11
2. 73
5. 46
38. 83
33. 37
0. 50
1. 0
7
23. 21
3. 32
6. 63
36. 49
29. 86
0. 50
1. 0
8
27. 30
3. 90
7. 80
34. 15
26. 35
0. 50
1. 0
9
31. 40
4. 49
8. 97
31. 81
22. 84
0. 50
1. 0
10
35. 49
5. 07
10. 14
29. 47
19. 33
0. 50
1. 0
11
19. 11
2. 73
5. 46
46. 31
0. 50
0. 5
12
23. 21
3. 32
6. 63
24. 33
42. 02
0. 50
0. 5
13
27. 30
3. 90
7. 80
22. 77
37. 73
0. 50
0. 5
14
31. 40
4. 49
8. 97
21. 21
33. 44
0. 50
0. 5
15
35. 49
10. 14
19. 65
29. 15
0. 50
0. 5
A
CC E
PT
5. 07
M
SC R
U
N
A
25. 89
IP T
1
ED
Design No.
Table 2: Effect of Fluorescent containing material and Moisture on Raman Spectra MCC: Lactose Ratio
RS=276 cm-1
RS=750 cm-1
RS= 1404 cm-1
Slope (Calculated from least square)
Intensity count
Intensity count
Intensity count
A. U.
11
85896
47659
14711
35593
32
59688
33290
10039
24824
52
34189
19205
5590
14300
75
22719
12889
3585
11
81946
45242
13930
32
56868
31702
52
33534
18615
75
24557
13951
11
61985
34349
32
42609
23625
52
27139
75
22087
%RH
SC R 5531
14001
3982
10287
11061
25462
7491
17559
15339
4803
11168
12424
3659
9214
U
N
A M
ED PT CC E
34008 23565
0. 5
A
9567
9737
1
RS= Raman Shift.
IP T
2
Table 3: Effect of sample moisture variability on test set predictions: Lab Scale Tablet
Calibration Sets
Global
Global
Modeling
11%
32%
52%
75% (38-2400 cm-1)
RMSEP (%, w/w)
3
3
3
3
3
11% RH
1. 81
2. 12
3. 87
3. 96
1. 6
1. 54
32% RH
2. 93
1. 87
3. 03
2. 66
1. 66
1. 48
52% RH
4. 62
2. 48
1. 37
75% RH
6. 54
4. 72
3. 28
RMSEP – Root Mean Square Error of Prediction
2. 11
1. 92
1. 94
1. 84
1. 75
1. 66
U
RH – Relative humidity
IP T
Test Set Prediction errors,
3
SC R
# Latent var.
APAP(845-900 cm-1)
N
Grey cells correspond to instances when a model developed with tablets stored at a particular relative humidity was
A
CC E
PT
ED
M
A
used to predict a test set stored in the same environmental conditions.
S0731-7085(18)31529-2 https://doi.org/10.1016/j.jpba.2018.10.022 PBA 12271
To appear in:
Journal of Pharmaceutical and Biomedical Analysis
Received date: Revised date: Accepted date:
27-6-2018 12-10-2018 12-10-2018
Please cite this article as: Hossain MN, Igne B, Anderson CA, Drennen JK, Influence of Moisture Variation on the Performance of Raman Spectroscopy in Quantitative Pharmaceutical Analyses, Journal of Pharmaceutical and Biomedical Analysis (2018), https://doi.org/10.1016/j.jpba.2018.10.022 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Influence of Moisture Variation on the Performance of Raman Spectroscopy in Quantitative Pharmaceutical
U
SC R
Md Nayeem Hossaina, Benoît Igneb, Carl A. Andersona,b and James K. Drennen IIIa,b,*
IP T
Analyses
a
A
N
Duquesne University Graduate School for Pharmaceutical Sciences, Pittsburgh, Pennsylvania 15282
b
Corresponding author at: Duquesne University Center for Pharmaceutical Technology, Duquesne
PT
*
ED
M
Duquesne University Center for Pharmaceutical Technology, Pittsburgh, Pennsylvania 15282
CC E
University, Pittsburgh, Pennsylvania 15282
A
Tel.: 1-412-396-1172, Fax: 1-412-396-4660, E-mail: [email protected]
Graphical abstract
IP T SC R U N A M
Raman spectral variations of the pharmaceutical materials were observed due to the
PT
ED
Highlight
presence of water.
Unwanted spectral baseline change due to moisture variation caused spectral
CC E
A
inconsistency across calibration and test samples which degraded quantitative model
performance.
Including moisture variations into the calibration set significantly improved Raman spectroscopic model performance.
Abstract
To develop a robust quantitative calibration model for spectroscopy, different sources of variability that
IP T
are not directly related to the components of interest should be included in the calibration samples; this variability should be similar to that which is anticipated during validation and routine operation. Moisture
SC R
content of pharmaceutical samples can vary as a function of supplier, storage conditions, geographic
origin or seasonal variation. Additionally, some pharmaceutical operations (e. g., wet granulation) cause
N
U
exposure of excipients and API to water. Although water is a weak Raman scatterer, moisture variability
A
has an indirect effect on analytical model performance. Because many pharmaceutical components have
M
intrinsic fluorescent characteristics (with broad spectral features), moisture variability may cause spectral
ED
artifacts in the form of baseline variation associated with fluorescence quenching. This work investigates the deleterious effects of water quenching on quantitative prediction accuracy of a multivariate
PT
calibration algorithm for Raman spectroscopy. To demonstrate this, a formulation composed of
CC E
acetaminophen, lactose, microcrystalline cellulose, HPMC and magnesium stearate was used. Tablets were manufactured using laboratory scale equipment. A full-factorial design was used to vary
A
acetaminophen (5 levels), and excipient ratios (3 levels) to generate tablets for calibration and testing. Tablet moisture variation was introduced by placing samples in different humidity chambers. Significant spectral effects arising from fluorescence were identified in the Raman spectra and due to moisture variation: the fluorescence related spectral variability caused substantial degradation of the prediction
performance for API. The work demonstrated that accounting for moisture variation during method
IP T
development reduced the prediction error of the multivariate prediction model.
SC R
Keywords: Raman Spectroscopy; Pharmaceutical; Quantitative analysis; Robustness; Moisture;
M
A
N
U
Fluorescence; PLS modelling.
PT
ED
1. Introduction
CC E
In the nearly seventeen years since the start of the PAT (Process Analytical Technology) initiative, spectroscopic techniques have become more widely used for designing, monitoring and controlling pharmaceutical unit operations. Such PAT methods have become an important component of a
A
comprehensive quality management system[1]. Among these, Raman spectroscopy has emerged as a promising process analytical tool due to its sharp peak features offering inherently good chemical specificity, its non-destructive nature, and the flexibility in sampling interfaces offering various opportunities for in-line, on-line or off-line analysis of solid oral solid dosage forms. Raman spectroscopy
has proven valuable for quantitative prediction of critical components of pharmaceutical products[2], managing critical process parameters[3-5], and for assuring that products meet real-time release criteria[6, 7].
IP T
Before implementing a quantitative spectroscopic prediction model, calibration model development is required. While developing Raman methods, it is necessary to ensure that models meet the required
SC R
performance standards; in particular, the accuracy, precision, and robustness of the model must meet predetermined validation criteria. A variety of multivariate calibration techniques have been used to
U
create models that were considered fit-for-purpose [8-10]. Though these methods allow the analyst to
A
N
extract relevant chemical and physical information about samples, numerous sources of spectral
M
variability that occur during the method life-cycle may confound measurements. To ensure long-term
ED
model performance, model robustness may be enhanced by including such sources of spectral variability in the calibration model. These sources of variability may be identified during a risk assessment.
PT
Understanding the impact of such variability on the spectra and the effect of these spectral changes on
CC E
model performance is a critical part of a sound model development process [10, 11].
Many sources of spectral variability arise during the pharmaceutical manufacturing process, including
A
sampling, analytical instrument instability, sample properties and environmental fluctuation. Sample variability may be introduced through raw material or process variability or as a result of sampling procedure: the time, location and method of sampling is often at fault. Typically, lot-to-lot differences of incoming raw materials introduce new physical and chemical variations [12, 13]. The consequence of
these variations on spectra may depend on the sampling configuration, spectral collection modalities, etc. [14-16]. As an example, the effect of particle size variation in tablets has been compared for backscatter and transmission Raman spectroscopy[17]. Because backscatter Raman generates most of the signal
IP T
from very near the tablet surface [18], particle size had a reduced impact relative to the transmission mode. In the case of transmission measurements, photon propagation characteristics were more
SC R
influenced by particle size variation due to the effect of light scattering on the effective optical path
length. When particle size variability was not included in the calibration set, model accuracy suffered.
U
Spectral processing methods such as derivative, normalization, etc., were applied to remove the baseline
N
effect of particle size variation, thereby improving model performance. Also, by matching particle size
M
A
variability for both API and excipients in the calibration, adequate predictions were obtained for independent test sets [17, 19-21]. A change of size, shape or density of tablets due to changes in the
ED
manufacturing process can also introduce spectral variability [22, 23]. By adjusting laser power and
PT
accumulation time, or enhancing density range in the calibration model, robustness can be enhanced
CC E
against tablet size and density variations [20, 22].
While the application of Raman in pharmaceutical applications has increased dramatically over the past
A
decade due to the improvement of instrumentation [24-26], one continuing challenge, and a major source of spectral variability causing interference with the Raman spectral signal, is fluorescence. Produced by emission of photons from low-lying excited electronic states, fluorescence often masks the vibrational shift of Raman spectra. This effect occurs in a sample whenever the frequency of the excitation laser coincides with transition energy from the ground to an electronically excited state.
Different experimental strategies and instrument modifications have been developed to overcome this problem[27]. By using excitation sources in the near-infrared region coupled with multichannel detectors, modern dispersive Raman spectrometers decrease the fluorescence background and provide adequate
IP T
Raman intensity[28]. Multivariate regression and spectral preprocessing methods have been used to reduce the impact of fluorescence signal on prediction results [29-32]. Moreover, model performance
background without introducing other spectral disturbances[33].
SC R
should be adequate as long as both calibration and validation sets contain similar fluorescence
U
One of the challenging sources of variability that analysts often encounter during spectroscopic model
A
N
development and deployment is environmental variability; for example, relative humidity will often vary
M
from the warehouses where raw materials are stored to the facilities where products are manufactured.
ED
This can cause significant variability in the amounts of water sorbed by pharmaceutical samples [34-36]. Moreover, different pharmaceutical unit operations, for example high shear granulation or film-coating
PT
processes, introduce significant amounts of water which can create batch to batch moisture variability
CC E
[23, 37]. Most often, Raman analytical methods ignore the effect of moisture because Raman presents relatively weak water scattering. However, because pharmaceutical samples exhibit fluorescence from
A
Active Pharmaceutical Ingredient (API) [38] , excipient [30, 31], dyes [32], capsule shells [39], etc., the effect of moisture as a fluorescence quencher needs to be considered [40]. This can introduce unwanted spectral inconsistency across calibration and test samples.
The objective of this paper is to investigate the effect of moisture and fluorescence on Raman spectroscopic analysis for pharmaceutical solid oral dosage forms. The effect of these changes on Raman spectra and multivariate quantitative model performance was assessed, with the aim to provide strategies
IP T
for the development of robust calibration models.
2. Material and Method
SC R
2.1. Samples and Environment conditions
Granules of acetaminophen (APAP; Mallinckrodt Inc., Raleigh, NC, USA), hydroxypropyl methylcellulose
N
U
(HPMC; Pharmacoat 606, Shin-Etsu Chemical Co. LTD, Tokyo, Japan) and lactose (modified spray-dried;
A
Foremost Farms USA, Rothschild, WI, USA) were manufactured using a fluid bed granulator (model WSG
M
5, Glatt, Binzen, Germany). The active containing granules were mixed with extra-granular MCC (MCC;
ED
Avicel PH 200, FMC Biopolymer, Mechanicsburg, PA, USA), lactose and magnesium stearate (Fisher Scientific, Waltham, MA, USA) in a bin blender. Tablets comprised of the aforementioned components
PT
were manufactured at laboratory scale using a Carver Automatic Tablet Press (Model 3887. 1SD0A00,
CC E
Wabash, IN, USA) and 13 mm die and flat-faced punches. Target tablet weight was 700 mg. HPLC was performed on the tablets to ensure uniformity of blend and tablets.
A
A five by three level, two factor [active content (19. 11%, 23. 21%, 27. 30%, 31. 40% and 35. 49% w/w; and excipient ratio levels 2, 1, 0. 5)] full factorial design was created. The use of a full-factorial experimental design provided orthogonality between the active ingredient and the excipient ratios. While it would have been possible to have all other components vary independently from the active ingredient,
the use of excipient ratio helped to reduce the number of design points. Table 1 summarizes the design. Three replicate tablets per level were created from the same blend for a total of 45 samples. From each design point, one tablet was randomly chosen to create a test set. Therefore, a total of 30 tablets were
IP T
used for calibration and 15 tablets were included in the test set.
After allowing the tablets to undergo viscoelastic relaxation (two weeks), they were placed in
SC R
environmental chambers at room temperature. The four chambers equilibrated with a relative humidity (RH) of 11% (saturated lithium chloride), 32% (saturated magnesium chloride), 52% (saturated
U
magnesium nitrate) and 75% (saturated sodium chloride). Tablets were left to equilibrate in each
A
N
chamber, beginning with the lowest moisture level. Once the mass of tablets had equilibrated in the 11%
M
RH chamber, they were analyzed and moved to the 32% RH chamber. The equilibration and analysis was
ED
repeated for each chamber. Tablets were moved from low to high relative humidity to simplify the experimental design and reduce the experimental period.
PT
2.2. Fluorescence Quenching Effect
CC E
To understand the nature of spectral change due to fluorescence quenching, spectra of pure components were collected in the presence of water and molecular gas, oxygen (O2) and nitrogen (N2). To investigate
A
the impact of water, pure component compacts were created by compressing approximately 700 mg of each material in the Carver press. Then, these pure component compacts were stored in the four aforementioned humidity conditions to observe the effect of moisture on Raman spectra.
However, spectra were collected in the presence of O2 and N2 using custom gas chambers rather than a humidity chamber.
2.3. Raman Data Collection
IP T
Calibration and test sets of Raman spectra were collected on a TRS 100 (Cobalt Light Systems Ltd., Abington, Oxfordshire, UK) for analysis of compacts. The transmission spectrometer is equipped with a
SC R
diode laser operating at 830 nm as the radiation source, a long pass spectral filter and a thermoelectrically cooled CCD camera for the detector. All compacts were scanned over the wavenumber range of 38-2400
N
U
cm-1. The measurements were conducted at room temperature (approximately 24oC). Parameters were
A
tuned according to the tablet scale: exposure time for lab scale and manufacturing scale samples was 0. 5
M
sec and 1.3 seconds, respectively. Five accumulations were collected per spectrum with a laser power of
ED
0. 6W.
PT
The effect of O2 and N2 gasses on the spectra of powdered samples was evaluated with a RamanRxn2TM backscattering spectrometer (Kaiser Optical Systems, Inc, Ann Arbor, MI, USA) and iCRaman software
CC E
(version 4. 1, Mettler Toledo, Columbus, OH, USA). A fiber-coupled PhAT probe (HoloGRAMS version 4. 0, Kaiser Optical Systems, Inc, Ann Arbor, MI, USA) was equipped with a 400 mW laser at 785 nm. The
A
diameter of laser spot size was 6 mm. Accumulation time was 3 seconds, and the spectral range was 382400 cm-1.
2.4. Reference Testing
Acetaminophen reference values for all compacts were determined using high-pressure liquid chromatography (HPLC; Waters Alliance 2790, Milford, MA, USA), followed by ultraviolet detection (Waters 2487). A method, modified from USP29-NF24 (‘‘Acetaminophen Tablets’’), was employed [41].
IP T
The mobile phase was water: methanol: acetic acid (80: 17: 3), and the stationary phase was a 15 cm by 4. 6 mm C18 column with 3 µm packing. The detection wavelength was 243 nm. The error of the reference
SC R
measurement was estimated at 0. 83%. Reference testing was performed after all laboratory tablets had been equilibrated and scanned at all four RH conditions.
N
U
2.5. Model Evaluation
A
In order to develop a calibration model for the prediction of API concentration and to determine the
M
potential impact of moisture variation, partial least-squares regression (PLS) was used with the nonlinear
ED
iterative partial least squares (NIPALS) algorithm [9]. The capability of a model developed at one moisture level to predict the active ingredient concentration of compacts stored at other relative humidity levels
PT
was tested. The prediction error was used to evaluate the effect of moisture variability on model
CC E
performance. Various diagnostic tests were employed to understand the underlying causes of model performance.
A
2.6. Performance measurement by Prediction error
The root-mean-square error (RMSE) was used as a performance criterion. The choice of latent variables was determined by comparing the evaluation of the root mean-square error of calibration (RMSEC) and of cross-validation (RMSECV). The ability of a model developed at one moisture level to predict accurately
an active ingredient concentration of tablets stored at other relative humidity conditions was tested by prediction error (RMSEP):
(1)
IP T
∑𝑛 (𝑦̂𝑖 − 𝑦𝑖 )2 𝑅𝑀𝑆𝐸𝑃 = √ 𝑖=1 𝑛
SC R
where 𝑦̂ and 𝑦 represent vectors of predicted and reference values, respectively, and 𝑛 the number of samples.
N
U
2.7. Diagnostic Tests
A
Statistics such as the Hotelling’s 𝑇 2 and the 𝑄-residuals (𝑄𝑟 ) can be used to ensure the suitability of a
M
model for predicting any sample. Hotelling’s 𝑇 2 tests the membership of an observation to the
ED
population forming the calibration set. 𝑄𝑟 determines whether a sample spectrum is different due to an
failure, etc. ) [42].
PT
un-modelled source of variance, including a possible instrumental fault ( e. g., sampling error, component
𝑇𝑖2 = 𝑡𝑖 (
(𝑇 𝑇 𝑇) −1 ) 𝑡𝑖 𝑛
(2)
A
CC E
Hotelling’s 𝑇 2 was defined as follows:
where 𝑡𝑖 is the latent variable scores for 𝑡𝑖 th sample spectrum and 𝑇 is the matrix of calibration sample model scores.
𝑄𝑟 was calculated as:
𝑄𝑟 = 𝑠𝑢𝑚(𝑥𝑖 − 𝑡𝑖 𝑃𝑇 )2
IP T
(3)
where 𝑄𝑟 represents sum of squared reconstruction error, 𝑥𝑖 is the sample spectrum, 𝑡𝑖 is the latent
SC R
variable scores for the 𝑖 𝑡ℎ sample spectrum, and P is the model loading. More information about these statistics and their use for process monitoring can be found in the literature[42].
N
U
4. Results
A
4.1. Effect of moisture variability on Raman spectroscopy
M
Figure 1 shows spectral baseline changes when tablets were exposed to four different moisture levels
ED
(11% (red), 32% (blue), 52% (green) and 75% (magenta) RH). It can be observed that increasing the water
PT
content in the tablet leads to a decrease in the fluorescent background. Mechanistically, the decrease in fluorescence intensity was caused by the deactivation of the excited-state fluorophores in contact with
CC E
certain molecules (e.g., water in this study) in the sample. These molecules are referred to as quenchers [39]. This spectral change, due to the presence of water, was much more significant than the effect of
A
photobleaching (due to successive exposure of the sample to the laser). The effect of photobleaching is illustrated in Figure 2, where a tablet was stored at a relative humidity 52% and scanned repeatedly over eight weeks. With an increasing number of scans and storage time, there was a small decrease in the baseline due to photobleaching.
To understand the source of spectral variation in the presence of moisture, a variety of compacts were prepared using pure components; the compacts were then exposed to different moisture levels. Figure 3 reveals that MCC and HPMC compacts demonstrated a significant decrease in the baseline with increasing
IP T
water adsorption, which may be due to fluorescence quenching (Figure 3). However, APAP, Lactose and Magnesium Stearate spectra demonstrated very small changes in the different conditions. This is likely
SC R
due to the absence of fluorescence with these three pure components. The results indicate that the
primary source of spectral variation in the prepared samples was from MCC and HPMC, with water acting
U
as a quencher.
A
N
In Figure 4, fluorescence quenching for MCC and HPMC was demonstrated in the presence of oxygen, a
M
known quenching agent in cellulosic materials [42]. Previous studies demonstrated that, due to the
ED
presence of lignin in the fiber component, a large number of pulp and paper samples exhibit laser induced fluorescence (LIF). However, in the presence of molecular oxygen, lignin caused LIF contributions to
PT
Raman spectra were significantly decreased[43]. Figure 4 shows that due to the presence of O2, the
CC E
spectral baseline of MCC changed over time. However, when spectra were collected in the presence of N2 or air, quenching was reduced. This result suggests that fluorescence decreases dramatically in the
A
presence of the quenching agent O2. A small spectral change was observed with MCC in the presence of air due to the continuous laser exposure. This provides another demonstration that the effect of photobleaching was less than that of quenching associated with moisture. Acetaminophen and Lactose spectra demonstrated limited spectral change due to the lack of fluorophores.
The effect of fluorescence quenching on spectral baseline has been shown to vary with the content of both fluorophores and moisture. To characterize the impact of these two factors, spectral slope and intensity was observed for raw spectra. The slope of the spectra was calculated from a simple least-
IP T
squares regression between Raman shift and Raman intensity. Three Raman shifts (276, 750, 1404 cm -1) were chosen from the spectra to calculate this slope. The slope values are displayed in Table 2, where
SC R
results from four moisture levels and three excipient ratios (2:1, 1:1 and 0. 5:1 of MCC: Lactose) are presented.
U
First, the amount of fluorescent components affected the baseline intensity, as anticipated. Fluorescence
A
N
intensity decreased with a decreasing amount of MCC (for a given moisture level). This change was
M
reflected in the slope. Second, the moisture content had a significant impact on the baseline: the Raman
ED
intensity decreased with increasing water content (for a given compact composition), causing a dramatically reduced slope with increasing moisture content (Table 2). These results emphasize the
PT
quenching effect of water on Raman spectra.
CC E
4.2. Effect of sample moisture variability on quantitative model performance
The observed spectral variations due to the presence of water had a significant impact on quantitative
A
model performance. Performance was evaluated by developing four models (one per relative humidity condition). These models were then used to predict four test sets corresponding to the four relative humidity conditions.
A number of preprocessing methods such as baseline weighted least squares, normalization, mean centering, derivative, and detrending were evaluated to remove the baseline changes of the full spectral range described previously. Random block cross-validation (five blocks) was used to determine the most
IP T
appropriate spectral pre-treatments and model complexity (number of latent variables). An optimization was independently conducted based on the minimization of cross-validation error. Savitzky-Golay first
SC R
derivative (window size-31, polynomial order-2) followed by normalization to unit area and mean-
centering was chosen for all models. Derivative methods were helpful to remove baseline variations and
U
normalization decreased the intensity of variations introduced due to the fluorescence quenching or any
N
physical differences between calibration and test sets. The responses were auto-scaled (zero mean, unit
M
A
variance).
ED
As shown in Table 3, models offered acceptable error when predicting API content for samples from the same moisture levels, where the preprocessing method was successful at removing fluorescent baseline
PT
variability. However, poor results were achieved for prediction of samples from unique moisture levels;
CC E
error statistics increased with increasing range between the moisture levels of calibration and test sets. The prediction error was largely due to bias, as seen in Figure 5A. These results may be interpreted as a
A
lack of robustness of the calibration model to moisture variability.
4.3. Robustness enhancement methods
Different strategies are applied to develop a model that is robust against different sources of variations. In the present study, a global modeling approach was used as a means to improve model accuracy.
Global models include all of the expected variability in the calibration set. In this case, calibration sets from four relative humidity levels were combined to facilitate the creation of two models, including: 1) a model using the full wavelength range(38-2400 cm-1 ), and 2) a model including a truncated wavelength
IP T
region (845-900 cm-1 ) that contained the APAP peak.
Prediction errors for the test set at the different humidity levels are provided in Table 3. The global model
SC R
demonstrated a lower prediction error compared to models created at unique humidity conditions (Figure 6A). Also of significance, the model created using the full variable range provided similar
U
prediction performance compared to the truncated variable range.
A
N
Figure 6B presents Hotelling’s 𝑇 2 and 𝑄 – residual plots for global modeling compared to Figure 5(B). The
M
reconstruction error for all the test sets was lower for global modeling, indicating that the global
ED
calibration model was more robust against moisture variation, which ultimately improved prediction
PT
performance.
Global models are practical when the number of interfering elements is limited. Their ability to handle
CC E
large ranges of variability in components not related to the parameter of interest will be seen through improved method robustness, although sometimes at the cost of reduced standard error of prediction.
A
5. Summary and Conclusion: It is difficult to preemptively identify all sources of variation that will be encountered during the pharmaceutical life-cycle. The experimental approach in this study provides an example of how a simple
phenomenon such as moisture variability, arising from such phenomena as inconstant environmental conditions, can, through fluorescence, significantly affect spectral baseline and model performance.
In this study, the presence of components containing 1) fluorophores and 2) moisture, negatively affected
IP T
model performance. Pharmaceutical raw materials often contain fluorescent components. In this study, lignin in the microcrystalline cellulose caused laser induced fluorescence [43, 44]. Moreover, different
SC R
lots of excipients may exhibit variable amounts of lignin, thus impacting the fluorescent background.
Variable moisture content also introduced spectral variance. The water content of tablets will often vary
U
between and within different tablet batches, depending on the relative humidity conditions during
A
N
production, packaging, storage and analysis. Water may be sorbed by samples very quickly depending on
M
the relative humidity of the environment and material attributes [34]. While it is likely that the relative
ED
humidity of the laboratory may be controlled, it is very unlikely that the same environmental conditions will be observed in the packaging facilities, warehouses and manufacturing plants. This is particularly true
PT
in situations where spectroscopic method development takes place at one geographic and climatic
CC E
location and routine use of the method occurs elsewhere[34].
It is typically beyond the capacity of an analyst to eliminate sample components or strictly control the
A
sources of variability. Understanding the effect of variabilities such as moisture and fluorescence during multivariate calibration model development will facilitate the development of calibration models which are robust against these variations. In the present study, the effect of fluorescence background was
reduced by preprocessing the spectra. However, when calibration and test sets contained different moisture content, prediction performance was degraded.
An understanding of potential confounding variables is key to the development of robust models.
IP T
Without that knowledge, a tendency to generate less robust models with better performance statistics will supersede the long-term interests of analytical methods that are robust over the lifecycle of a
SC R
product. Global calibration methods were tested in this work to demonstrate opportunities for the
M
A
N
U
creation of robust models for laboratory and production scale tablets.
PT
ED
Acknowledgement
CC E
The authors are grateful to Agilent Technologies, Inc (previously Cobalt Light Systems Ltd.) for the use of the TRS 100 transmission Raman optical engine and the technical support received during the research
A
period.
References
[1] Food, D. Administration, Guidance for industry: PAT—A framework for innovative pharmaceutical development, manufacturing, and quality assurance, DHHS, Rockville, MD, (2004). [2] C.J. Strachan, T. Rades, K.C. Gordon, J. Rantanen, Raman spectroscopy for quantitative analysis of
IP T
pharmaceutical solids, Journal of pharmacy and pharmacology, 59 (2007) 179-192. [3] A. Hédoux, Y. Guinet, M. Descamps, The contribution of Raman spectroscopy to the analysis of phase
SC R
transformations in pharmaceutical compounds, International journal of pharmaceutics, 417 (2011) 17-31. [4] T. De Beer, A. Burggraeve, M. Fonteyne, L. Saerens, J.P. Remon, C. Vervaet, Near infrared and Raman
U
spectroscopy for the in-process monitoring of pharmaceutical production processes, International Journal
N
of Pharmaceutics, 417 (2011) 32-47.
M
A
[5] H. Wikström, P.J. Marsac, L.S. Taylor, In‐line monitoring of hydrate formation during wet granulation using Raman spectroscopy, Journal of pharmaceutical sciences, 94 (2005) 209-219.
ED
[6] J. Müller, K. Knop, M. Wirges, P. Kleinebudde, Validation of Raman spectroscopic procedures in
PT
agreement with ICH guideline Q2 with considering the transfer to real time monitoring of an active
CC E
coating process, Journal of pharmaceutical and biomedical analysis, 53 (2010) 884-894. [7] H. Trnka, J. Rantanen, H. Grohganz, Well-plate freeze-drying: a high throughput platform for screening of physical properties of freeze-dried formulations, Pharmaceutical development and technology, 20
A
(2015) 65-73.
[8] D.M. Haaland, E.V. Thomas, Partial least-squares methods for spectral analyses. 1. Relation to other quantitative calibration methods and the extraction of qualitative information, Analytical Chemistry, 60 (1988) 1193-1202.
[9] S. Wold, M. Sjöström, L. Eriksson, PLS-regression: a basic tool of chemometrics, Chemometrics and intelligent laboratory systems, 58 (2001) 109-130. [10] T. Naes, T. Isaksson, T. Fearn, T. Davies, A user friendly guide to multivariate calibration and
IP T
classification, NIR publications 2002. [11] P.C. Williams, Implementation of near-infrared technology, Near-infrared technology in the
SC R
agricultural and food industries, 2 (2001) 143-167.
[12] B. Igne, Z. Shi, J.K. Drennen, C.A. Anderson, Effects and Detection of Raw Material Variability on the
U
Performance of Near‐Infrared Calibration Models for Pharmaceutical Products, Journal of pharmaceutical
N
sciences, 103 (2014) 545-556.
M
A
[13] V.S. Dave, S.D. Saoji, N.A. Raut, R.V. Haware, Excipient variability and its impact on dosage form functionality, Journal of pharmaceutical sciences, 104 (2015) 906-915.
ED
[14] M.V. Pellow-Jarman, P.J. Hendra, R.J. Lehnert, The dependence of Raman signal intensity on particle
PT
size for crystal powders, Vibrational Spectroscopy, 12 (1996) 257-261.
CC E
[15] H. Wang, C.K. Mann, T.J. Vickers, Effect of powder properties on the intensity of Raman scattering by crystalline solids, Applied spectroscopy, 56 (2002) 1538-1544. [16] T. De Beer, W. Baeyens, Y. Vander Heyden, J.P. Remon, C. Vervaet, F. Verpoort, Influence of particle
A
size on the quantitative determination of salicylic acid in a pharmaceutical ointment using FT-Raman spectroscopy, European journal of pharmaceutical sciences, 30 (2007) 229-235.
[17] N. Townshend, A. Nordon, D. Littlejohn, M. Myrick, J. Andrews, P. Dallin, Comparison of the determination of a low-concentration active ingredient in pharmaceutical tablets by backscatter and transmission Raman spectrometry, Analytical chemistry, 84 (2012) 4671-4676.
IP T
[18] N. Townshend, A. Nordon, D. Littlejohn, J. Andrews, P. Dallin, Effect of particle properties of powders on the generation and transmission of Raman scattering, Analytical chemistry, 84 (2012) 4665-4670.
SC R
[19] Y. Hu, H. Wikström, S.R. Byrn, L.S. Taylor, Analysis of the effect of particle size on polymorphic quantitation by Raman spectroscopy, Applied spectroscopy, 60 (2006) 977-984.
U
[20] A. Sparén, M. Hartman, M. Fransson, J. Johansson, O. Svensson, Matrix effects in quantitative
N
assessment of pharmaceutical tablets using transmission Raman and near-infrared (NIR) spectroscopy,
M
A
Applied spectroscopy, 69 (2015) 580-589.
[21] D. Oelkrug, E. Ostertag, R.W. Kessler, Quantitative Raman spectroscopy in turbid matter: reflection or
ED
transmission mode?, Analytical and bioanalytical chemistry, 405 (2013) 3367-3379.
PT
[22] H. Wikström, S. Romero-Torres, S. Wongweragiat, J.A.S. Williams, E.R. Grant, L.S. Taylor, On-line
CC E
content uniformity determination of tablets using low-resolution Raman spectroscopy, Applied spectroscopy, 60 (2006) 672-681. [23] B. Igne, J.K. Drennen, C.A. Anderson, Improving Near-Infrared Prediction Model Robustness with
A
Support Vector Machine Regression: A Pharmaceutical Tablet Assay Example, Applied spectroscopy, 68 (2014) 1348-1356. [24] H. Swierenga, A. De Weijer, R. Van Wijk, L. Buydens, Strategy for constructing robust multivariate calibration models, Chemometrics and Intelligent Laboratory Systems, 49 (1999) 1-17.
[25] J. Johansson, A. Sparén, O. Svensson, S. Folestad, M. Claybourn, Quantitative transmission Raman spectroscopy of pharmaceutical tablets and capsules, Applied spectroscopy, 61 (2007) 1211-1218. [26] R.L. McCreery, Raman spectroscopy for chemical analysis, John Wiley & Sons 2005.
IP T
[27] W.R. Browne, J.J. McGarvey, The Raman effect and its application to electronic spectroscopies in metal-centered species: Techniques and investigations in ground and excited states, Coordination
SC R
Chemistry Reviews, 251 (2007) 454-473.
[28] J.M. Chalmers, H.G. Edwards, M.D. Hargreaves, Infrared and Raman spectroscopy in forensic science,
U
John Wiley & Sons 2012.
N
[29] J.F. Kauffman, M. Dellibovi, C.R. Cunningham, Raman spectroscopy of coated pharmaceutical tablets
biomedical analysis, 43 (2007) 39-48.
M
A
and physical models for multivariate calibration to tablet coating thickness, Journal of pharmaceutical and
ED
[30] Y. Li, B. Igne, J.K. Drennen, C.A. Anderson, Method development and validation for pharmaceutical
CC E
(2016) 318-325.
PT
tablets analysis using transmission Raman spectroscopy, International journal of pharmaceutics, 498
[31] R.N. Palermo, S.M. Short, C.A. Anderson, H. Tian, J.K. Drennen III, Determination of figures of merit for near-infrared, Raman and powder X-ray diffraction by net analyte signal analysis for a compacted
A
amorphous dispersion with spiked crystallinity, Journal of Pharmaceutical Innovation, 7 (2012) 56-68. [32] S. Romero-Torres, J.D. Pérez-Ramos, K.R. Morris, E.R. Grant, Raman spectroscopy for tablet coating thickness quantification and coating characterization in the presence of strong fluorescent interference, Journal of pharmaceutical and biomedical analysis, 41 (2006) 811-819.
[33] I. Barman, C.-R. Kong, G.P. Singh, R.R. Dasari, Effect of photobleaching on calibration model development in biological Raman spectroscopy, Journal of biomedical optics, 16 (2011) 011004. [34] B. Igne, N. Hossain, J.K. Drennen III, C.A. Anderson, Robustness considerations and effects of moisture
IP T
variations on near infrared method performance for solid dosage form assay, Journal of Near Infrared Spectroscopy, 22 (2014) 179-188.
SC R
[35] S. Pieters, W. Saeys, T. Van den Kerkhof, M. Goodarzi, M. Hellings, T. De Beer, Y. Vander Heyden, Robust calibrations on reduced sample sets for API content prediction in tablets: Definition of a cost-
U
effective NIR model development strategy, Analytica chimica acta, 761 (2013) 62-70.
N
[36] T. Van den Kerkhof, R. De Maesschalck, K. Vanhoutte, M. Coene, Augmentation of near infrared
M
A
diffuse reflectance and transmittance spectral data for the development of robust PLSBC models for classifying double blind clinical trial tablets, Journal of pharmaceutical and biomedical analysis, 42 (2006)
ED
517-522.
PT
[37] D.M. Parikh, Handbook of pharmaceutical granulation technology, CRC Press 2016.
CC E
[38] M. Hajjou, Y. Qin, S. Bradby, D. Bempong, P. Lukulay, Assessment of the performance of a handheld Raman device for potential use as a screening tool in evaluating medicines quality, Journal of pharmaceutical and biomedical analysis, 74 (2013) 47-55.
A
[39] M.D. Hargreaves, N.A. Macleod, M.R. Smith, D. Andrews, S.V. Hammond, P. Matousek, Characterisation of transmission Raman spectroscopy for rapid quantitative analysis of intact multicomponent pharmaceutical capsules, Journal of pharmaceutical and biomedical analysis, 54 (2011) 463468.
[40] S. Tobita, K. Ida, S. Shiobara, Water-induced fluorescence quenching of aniline and its derivatives in aqueous solution, Research on Chemical Intermediates, 27 (2001) 205-218. [41] Acetaminophen Tablets. United States Pharmacopeial Monograph 29, USP29-NF24.
IP T
[42] R.P. Cogdill, C.A. Anderson, J.K. Drennen, Process analytical technology case study, part III: calibration monitoring and transfer, AAPS PharmSciTech, 6 (2005) E284-E297.
SC R
[43] U.P. Agarwal, An overview of Raman spectroscopy as applied to lignocellulosic materials, Advances in lignocellulosics characterization, (1999) 201-225.
U
[44] G. Thoorens, F. Krier, B. Leclercq, B. Carlin, B. Evrard, Microcrystalline cellulose, a direct compression
N
binder in a quality by design environment—A review, International journal of pharmaceutics, 473 (2014)
A
CC E
PT
ED
M
A
64-72.
Figure Captions
Figure 1: Raw spectra of calibration tablets at different moisture conditions: 11% RH (red), 32% RH (blue),
IP T
52% RH (green), and 75% RH (magenta).
SC R
Figure 2: Raw spectra of a tablet (target formulation) exposed to repeat scans over eight weeks to test the photobleaching effect.
U
Figure 3: Pure component raw spectra of Acetaminophen, Hypromellose, Lactose, Magnesium Stearate,
M
A
75% RH in red, blue, green and magenta respectively)
N
and Microcrystalline Cellulose stored in four different humidity conditions (11% RH, 32% RH, 52% RH and
Figure 4: Raman spectra collected in the presence of molecular gas A) MCC in presence of O2, B) MCC in
ED
presence of N2, C) Acetaminophen in presence of air, D) Acetaminophen in the presence of O2 (all are raw
PT
spectra).
CC E
Figure 5: (A) Concentration correlation plot of predicted vs. reference APAP concentrations corresponding to prediction samples from different humidity conditions: 11% RH (red), 32% RH (blue), 52% RH (green)
A
and 75% RH (magenta). The calibration model was displayed with Black dots. (B) Hotelling’s 𝑇 2 vs. 𝑄residuals plot was generated from the same tablets, represented with similar colors.
Figure 6: (A) Concentration correlation plots of predicted versus reference APAP concentrations corresponding to prediction samples from 11% RH (red), 32% RH (blue), 52% RH (green) and 75% RH (magenta). The global calibration model performance is displayed with black dots. (B) The 𝑇 2 vs 𝑄-
A
CC E
PT
ED
M
A
N
U
SC R
IP T
residuals plot was generated from similar tablets, represented by similar colors.
A ED
PT
CC E
IP T
SC R
U
N
A
M
Fig 1
A ED
PT
CC E
IP T
SC R
U
N
A
M
Fig 2
A ED
PT
CC E
IP T
SC R
U
N
A
M
Fig 3
A ED
PT
CC E
IP T
SC R
U
N
A
M
Fig 4
A ED
PT
CC E
IP T
SC R
U
N
A
M
Fig 5
A ED
PT
CC E
IP T
SC R
U
N
A
M
Fig 6
Table 1: Experimental Design (%W/W).
Acetaminophen
Hypromellose
Intragranular
Extragranular
lactose
Microcrystalline cellulose
Magnesium Stearate
Excipient ratio
lactose
19. 11
2. 73
5. 46
51. 77
20. 43
0. 50
2. 0
2
23. 21
3. 32
6. 63
48. 65
17. 70
0. 50
2. 0
3
27. 30
3. 90
7. 80
45. 53
14. 97
0. 50
2. 0
4
31. 40
4. 49
8. 97
42. 41
12. 24
0. 50
2. 0
5
35. 49
5. 07
10. 14
39. 29
9. 51
0. 50
2. 0
6
19. 11
2. 73
5. 46
38. 83
33. 37
0. 50
1. 0
7
23. 21
3. 32
6. 63
36. 49
29. 86
0. 50
1. 0
8
27. 30
3. 90
7. 80
34. 15
26. 35
0. 50
1. 0
9
31. 40
4. 49
8. 97
31. 81
22. 84
0. 50
1. 0
10
35. 49
5. 07
10. 14
29. 47
19. 33
0. 50
1. 0
11
19. 11
2. 73
5. 46
46. 31
0. 50
0. 5
12
23. 21
3. 32
6. 63
24. 33
42. 02
0. 50
0. 5
13
27. 30
3. 90
7. 80
22. 77
37. 73
0. 50
0. 5
14
31. 40
4. 49
8. 97
21. 21
33. 44
0. 50
0. 5
15
35. 49
10. 14
19. 65
29. 15
0. 50
0. 5
A
CC E
PT
5. 07
M
SC R
U
N
A
25. 89
IP T
1
ED
Design No.
Table 2: Effect of Fluorescent containing material and Moisture on Raman Spectra MCC: Lactose Ratio
RS=276 cm-1
RS=750 cm-1
RS= 1404 cm-1
Slope (Calculated from least square)
Intensity count
Intensity count
Intensity count
A. U.
11
85896
47659
14711
35593
32
59688
33290
10039
24824
52
34189
19205
5590
14300
75
22719
12889
3585
11
81946
45242
13930
32
56868
31702
52
33534
18615
75
24557
13951
11
61985
34349
32
42609
23625
52
27139
75
22087
%RH
SC R 5531
14001
3982
10287
11061
25462
7491
17559
15339
4803
11168
12424
3659
9214
U
N
A M
ED PT CC E
34008 23565
0. 5
A
9567
9737
1
RS= Raman Shift.
IP T
2
Table 3: Effect of sample moisture variability on test set predictions: Lab Scale Tablet
Calibration Sets
Global
Global
Modeling
11%
32%
52%
75% (38-2400 cm-1)
RMSEP (%, w/w)
3
3
3
3
3
11% RH
1. 81
2. 12
3. 87
3. 96
1. 6
1. 54
32% RH
2. 93
1. 87
3. 03
2. 66
1. 66
1. 48
52% RH
4. 62
2. 48
1. 37
75% RH
6. 54
4. 72
3. 28
RMSEP – Root Mean Square Error of Prediction
2. 11
1. 92
1. 94
1. 84
1. 75
1. 66
U
RH – Relative humidity
IP T
Test Set Prediction errors,
3
SC R
# Latent var.
APAP(845-900 cm-1)
N
Grey cells correspond to instances when a model developed with tablets stored at a particular relative humidity was
A
CC E
PT
ED
M
A
used to predict a test set stored in the same environmental conditions.