Food Chemistry 158 (2014) 296–301
Contents lists available at ScienceDirect
Food Chemistry journal homepage: www.elsevier.com/locate/foodchem
Analytical Methods
Rapid detection of sildenafil analogue in Eurycoma longifolia products using a new two-tier procedure of the near infrared (NIR) spectra database Mazlina Mohd Said ⇑, Simon Gibbons, Anthony C. Moffat, Mire Zloh 1 UCL School of Pharmacy, 29-39 Brunswick Square, London WC1N 1AX, UK
a r t i c l e
i n f o
Article history: Received 18 January 2013 Received in revised form 28 September 2013 Accepted 23 February 2014 Available online 6 March 2014 Keywords: NIRs Spectral database Eurycoma longifolia Sildenafil analogue Principal component analysis SIMCA
a b s t r a c t A simple and cost-effective two-tier drug screening procedure comprises a ‘dedicated’ NIR spectral database of common medicines and a ‘unified’ database was developed to detect the sildenafil analogue in Eurycoma longifolia products. Diffuse reflectance spectra of ten commercial herbal products containing E. longifolia were obtained over the wavelength range of 1100–2500 nm. The spectral search of two products purchased via the internet against a dedicated database of reputable E. longifolia products have resulted in the similarity index of more than 0.1 which indicated significantly different spectra. Further searches against the unified database showed a close match to the spectra of drug containing sildenafil citrate suggesting the presence of a sildenafil analogue. This finding was supported by clustering of these spectra in the PCA score plot within 5% significance level. This approach has alleviated the use of reference product or standard active for direct comparison and has a potential to be used for adulterated food and drugs detection. Ó 2014 Elsevier Ltd. All rights reserved.
1. Introduction Eurycoma longifolia (Family Simaroubaceae) has a long therapeutic history and has become one of the well-known health food products across the world. It is traditionally used as a general tonic to treat high blood pressure, tuberculosis, fever, diarrhoea, jaundice and dysentery (Zhari, Norhayati, & Jaafar, 1999). However, the popularity of E. longifolia today is based on its aphrodisiac effect, which is due to its ability to stimulate the production or action of androgen hormones, especially testosterone (Bhat & Karim, 2010). Most of the quality issues of commercial E. longifolia preparations are associated with contamination with heavy metals (Ang, Lee, & Cheang, 2004; Ang, Lee, & Matsumoto, 2003) and intentional adulteration with drugs or their analogues (Zhang, Wider, Shang, Li, & Ernst, 2012). Sildenafil and its related compounds are the most common adulterants found in herbal preparation used as sexual enhancer or man’s virility products (Champagne & Emmel, 2011). Previous
⇑ Corresponding author. Current address: Drug and Herbal Research Center, Universiti Kebangsaan Malaysia, Jalan Raja Muda Abdul Aziz, 50300 Kuala Lumpur, Malaysia. Tel.: +60 392897972. E-mail address: [email protected] (M.M. Said). 1 Current address: Department of Pharmacy, University of Hertfordshire, College Lane, Hatfield AL10 9AB, UK. http://dx.doi.org/10.1016/j.foodchem.2014.02.125 0308-8146/Ó 2014 Elsevier Ltd. All rights reserved.
works have shown the identification of sildenafil and its analogue in E. longifolia preparations available in the market were carried out using FTIR (Champagne & Emmel, 2011), ESI-MS (Li, Low, Aliwarga, Teo, Ge, Zeng et al., 2009), HPLC–MS (Fleshner, Harvey, Adomat, Wood, Eberding, Hersey & Guns, 2005). Identification using all these methods was made based on the comparison of suspected sample with a reference standard or reference spectra. The combination of NIRS and chemometric analyses such as PCA, PLS and SIMCA, has proven to be simple, fast and robust drug analyses procedures for counterfeit drug detection. Analysis can be made through the comparison of suspected tablets with the spectra of authentic tablets (Moffat, Assi, & Watt, 2010; O’Neil, Jee, Lee, Charvill & Moffat, 2008) and determination of the active pharmaceutical ingredients (APIs) (Khan, Jee, Watt, & Moffat, 1997) or the sources of the tablets (Said, Gibbons, & Zloh, 2009). However, application of these methods mostly depends on the availability of reference compounds or products for comparison with the test spectra. Here, an experimental strategy was designed in order to demonstrate the use of two-tier methods as a potential rapid screening tool for identification and classification of herbal products. A customized spectral database of different brand names of E. longifolia products sold in the Malaysian market was created and validation of the proposed strategy was carried out using three sets of test samples of E. longifolia products purchased via the internet. This
297
M.M. Said et al. / Food Chemistry 158 (2014) 296–301
method has shown the possibility of direct identification of the adulterated samples without relying on standard which is almost impossible using the NIRS techniques mentioned earlier. 2. Experimental
by taking the original averaged NIR spectra, exported in ASCII format and then converted to JCAM-DX files using an in-house programme. The spectral files were then imported into The Unscrambler v9.7 software, (CAMO Software, Oslo, Norway). The spectral database was created using SPECTRAL ID v9.0 database from Gram Suite Software (Thermo Scientific, Waltham, USA).
2.1. Samples 2.4. Database creation and search Seven commercial herbal products stated as containing extracts of E. longifolia were purchased from pharmacies in Malaysia using the convenient sampling. The registration number of each sample was verified with the regulatory agency to ensure the authenticity of these products. Three sets of E. longifolia samples; E. longifolia Complex (manufacturer not stated), Malaysian Ginseng Complex (manufacturer not stated) and Tongkat Ali 100:1 (manufacturer not stated) were purchased via the internet from United Kingdom. Table 1 lists the ingredients of these products as stated on the products’ packaging vendors. Apart from these samples, 20 capsules from each of the 7 brands of E. longifolia were included in a customized spectral database of E. longifolia products. 2.2. NIR analysis The NIR diffuse reflectance analyses were carried out using a NIRSystems 6500 spectrophotometer, equipped with a Rapid Content Analyser (FOSS NIRSystems, Silver Springs, USA). Capsules were emptied into 4 ml clear glass vials with snap caps 45 15 mm (height diameter) (Waters Ltd., Herts, UK). The glass vial was placed in the middle of the sample stage. Each sample was analysed four times, shaking the vials in between acquisition time to avoid the compaction of the powder materials. Eight accumulations of 32 scans of reflectance spectra were recorded for each sample over the wavelength range 1100–2500 nm.
2.4.1. Database development The spectra of reference sample were directly imported into the E. longifolia database (dedicated database) after baseline correction using the auto-baseline correction algorithm to reduce the scattering effects that can highly contaminate NIR spectra (GRAMS Suite Help, 2009). A combination of different types of conventional medicines database that were developed throughout this research, known as the unified database, was also used in this work. In the database, the X-axes of the spectra represented the spectral wavelength in nanometres between the range 1100 and 2500 and the Y-axes were the spectral absorbance values. For each brand of samples, at least 20 averaged spectral profiles with 700 individual points in each were combined as a multi-file to enhance the accuracy of the spectral search. The spectra were saved in 32-bit resolution. Other information recorded together with the spectra included the brand and proprietary names, batch numbers, expiry dates, manufacturer names and addresses, sample origins, other excipients (where available) and a description of the samples. 2.4.2. Database search The spectra of the samples were searched against the dedicated E. longifolia database followed by the universal database using the correlation algorithm based on the whole spectrum (GRAMS Suite Help, 2009):
2.3. Software
HQIc ¼ 1 The NIR instrument was controlled by Vision Spectral Analysis Software for Windows v2.11 (FOSS NIRSystems, Silver Springs, USA) for data acquisition. Chemometric analysis was conducted
ðLm Am Þ2 ðLm Lm ÞðAm Am Þ
where Lm and Am are vectors calculated using all the points in the spectra as defined by equations:
Table 1 List of seven brands of E. longifolia products used as the reference spectra and three test samples purchased via the internet with their names, labels and main active ingredients as stated on the packaging. No.
Samples
Label
Ingredients as stated on the products’ packaging
Reference samples 1 2 3 4 5 6 7
Tongkat Ali Plus Tonex Tongkat Ali Gold Box Tongkat Ali Puteri Rembulan Tongkat Ali Tongkat Ali Capsule LKH Tongkat Ali Nu-Prep100 Tongkat Ali
TA TA TA TA TA TA TA
Mixed preparation with 50 mg radix extract Radix extract 300 mg Radix extract 300 mg Radix extract 500 mg Mix preparations with 354 mg radix extract Radix extract 350 mg Radix extract 100 mg
Internet samples 8
E. longifolia Complex
ELC
Malaysian Ginseng Complex
MGC
Lewtress Tongkat Ali 100:1
LTA
E. longifolia 154 mg Floscarthami 24.5 mg Rhizoma cucurmaelongae 49 mg Gingko biloba extract 49 mg Epimedii 24.5 mg Cistanches 24.5 mg Astragalus membranaceus 24.5 mg Malaysian ginseng complex 157.5 mg Fructus tribulusterestris 52.5 mg Herba epimedium 52.5 mg Radix Smilax myosotiflora 28 mg Radix Panax ginseng 24.5 mg Fructus coriandrumsativum 17.5 mg Trigonella foenumgraecum 17.5 mg 500 mg Pure 100:1 Tongkat ali root extract
9
10.
298
M.M. Said et al. / Food Chemistry 158 (2014) 296–301
Pn Lm ¼ L
i¼1 Li
n
Pn Am ¼ A
i¼1 Ai
n
Ai is the absorbance values of the ith point in the unknown spectrum, Li is the corresponding point of the reference spectrum being searched in the database being compared to, and n is the number of wavelengths in each vector. Using this algorithm, the least squares dot products of the unknown spectra were compared with the spectra in the database after being centralized to their respective means using the hit quality index (HQI) calculated. A low HQI value indicated a good match between the unknown spectra and their matching spectra in the database. HQI equals to zero indicating 100% similarity between two spectra. Based on HQI, a hit list was generated after each search, listing the highest to lowest match of the spectrum tested to each spectrum in the database. 2.5. Chemometric analysis Principal Component Analysis (PCA) and Soft Independent Modelling of Class Analogy (SIMCA) was conducted using The Unscrambler v9.7 software (CAMO Software, Oslo, Norway). 2.5.1. PCA PCA was used to reduce the number of variables in a multivariate data table and to present it in a low dimensional space. PC models were constructed individually for each batch of the samples. After the first run, the presence of outliers, groups, clusters and trends were determined based on the observation of the score plots. At this stage, the outliers detected belonged to the same population but they were poorly described by the model. The optimum number of PCs was determined based on the total explained variance plot. PCA was conducted on the data with leverage correction as the validation method and the scaling factor was set as 1. 2.5.2. SIMCA Classification using the SIMCA method was carried out on the suspected samples by projecting them on all of the PC models with the significance level set at 95%. The orthogonal distances from the new objects to two different classes (models) were presented in Cooman’s plots. In this plot, the orthogonal distances from the test samples to two different classes (models) were compared at the same time in pair wise plots. 3. Results and discussion Spectroscopic fingerprints obtained by NIR provided a distinct representation of individual herbal formulations despite their similar chemical entities since it reflected not only the active constituents but also all of the other chemical compositions/excipients and it also indicated the variability in the production processes used. The average NIR spectra of each of the E. longifolia samples after SNV transformation are shown in Fig. 1A. The test set consisted of three products purchased via the internet (ELC, MGC, and LTA). Their NIR spectra are shown in red in three different line shapes. The spectra for ELC and LTA can be distinguished from the rest of the set while the spectrum for sample MGC overlaps with other spectra. The PCA of all spectra for the E. longifolia products was carried out and the PCA score plot indicated that each brand of the herbal product was clustered according to the manufacturer (Fig. 1B). Large variations of herbal products from different productions were mainly due to the different extraction methods and use of excipients in some of the samples. Two sets of the commercial preparation (TA and TA) with the three sets of test samples contained other constituents of plant origin in addition to
E. longifolia. However, based on the product labelling (Table 1), these constituents present in a small amount compared to the main active component. Therefore, The PCA score plot of all spectra were grouped in the 95% confidence limit indicating that the purchased samples were similar to other products which did not suggest that there were any quality issues with the samples. 3.1. Two-tier method analysis The use of the NIR spectral database as a basis for the two-tier method has been demonstrated in our previous work on the qualitative analysis of different brands of paracetamol tablets purchased in pharmacies and supermarkets in Malaysia (Said, Gibbons, Moffat, & Zloh, 2011; Said, Gibbons, & Zloh, 2010). In this study, the application of dedicated and unified databases for the quick identification of unknown samples and similarity assessment of products are described. A ‘dedicated’ database is a single product database that contains spectra from a specific type of product or class of medicines. Here, the dedicated database consisted of herbal preparations with E. longifolia as the main (if not only) active component. The adequate use of this type of database can only be used when the main components of the sample are known. This procedure was used to verify and to confirm if a sample belongs to a particular class of products and can also be used to further identify the sample according to the manufacturer or batches available in the database. The use of cut-off points were required to perform classification (Kramer & Ebel, 2000). The ‘unified’ database is the combination of all the dedicated product databases and it contained all the spectra of the products purchased up to date. In a larger database, that includes different types of conventional medicines, it might be possible to detect the presence of other compounds or adulterants in a product. The use of cut-off points was no longer relevant for this purpose. However, similarity between spectra could provide some information about sample composition. 3.2. Spectra search using the dedicated database The similarity of the spectra was first assessed using the dedicated database containing only the spectra of E. longifolia products. The average NIR spectra of the three test samples (samples ELC, LTA, and MGC) bought over the internet were imported into the single product database; consisting of different brands of E. longifolia products. A full spectrum search was conducted on these samples against the database to confirm their similarity using a correlation algorithm for the whole spectrum. A hit-list of the top matches was provided together with hit quality index (HQI) values after each search. The cut-off values determined in our previous work were used for samples classification (Said et al., 2011). These classifications were divided into four types; Type 1 (same batch of sample will generate a HQI of less than 0.0001); Type 2 (different batch of products of the same brand name will generate HQI of less than 0.01; Type 3 (different brand names of samples from the same class of drugs or herbals will generate HQI of less than 0.1; and Type 4 (completely different type of samples will generate HQI of more than 0.2). According to the pre-determined cut-off values, spectra search of sample LTA (HQI = 0.0464) fell within the same classification with the other spectra of E. longifolia in the database (Type 3). However, the search outcomes on the other two samples ELC and MGC have generated higher HQI values 0.1133 and 0.1297, respectively (Table 2). This has led to an inconclusive finding of the similarity between these sets of samples to the database and indicated the possibility of these samples having a different composition.
299
M.M. Said et al. / Food Chemistry 158 (2014) 296–301
Fig. 1. (A) Average SNV spectra from the different commercial products of E. longifolia in the wavelength range 1400–2400 nm. (B) PCA score plots of the NIR spectra of commercial E. longifolia herbal formulations purchased randomly in the Malaysian market. The test samples; ELC, MGC and LTA (pointed) were clustered within the 95% confidence limit of other samples.
The database search outcome and the observation of the PCA score plot (Fig. 1B) did not give agreeable results in this analysis. Use of the database indicated that there was some dissimilarity of the spectra based on the higher HQI while PCA indicated that
the test samples were similar to the reference spectra as both sets clustered within the 95% confidence ellipse. The contradicting outcome was possible as the database and PCA used different methods in assessing the similarity between the spectra.
Table 2 The hit-list showing the database search outcomes for samples (A) ELC (B) LTA and (C) MGC against the E. longifolia dedicated database. The HQI and sample in bold showed slightly higher values that the cut-off threshold pre-determined. (A) ELC
(B) LTA
(C) MGC
Hit #
HQI
Sample ID
Hit#
HQI
Sample ID
Hit #
HQI
Sample ID
1 2 3 4 5 6 7 8 9 10
0.1133 0.1707 0.1881 0.1995 0.2049 0.2247 0.2355 0.2500 0.2504 0.3760
TA TA TA TA TA TA TA TA TA TA
1 2 3 4 5 6 7 8 9 10
0.0464 0.0571 0.1170 0.1564 0.2434 0.3421 0.3725 0.3853 0.3878 0.5127
TA TA TA TA TA TA TA TA TA TA
1 2 3 4 5 6 7 8 9 10
0.1297 0.1564 0.1810 0.2036 0.2212 0.2498 0.2595 0.2745 0.2754 0.4046
TA TA TA TA TA TA TA TA TA TA
300
M.M. Said et al. / Food Chemistry 158 (2014) 296–301
Table 3 The hit-list showing the database search outcomes for samples (A) ELC and (B) MGC against the general customized database with the top-ten closest match. Both samples ELC and MGC showed highest similarity with Viagra tablets. Bold form is to highlight the similarity of the samples tested with viagra tablets. (A) ELC
(B) MGC
Hit #
HQI
Sample ID
Hit #
HQI
Sample ID
1 2 3 4 5 6 7 8 9 10
0.2179 0.2476 0.3308 0.3339 0.3621 0.3794 0.3814 0.4415 0.4626 0.4685
viagra01 vor05 vor07 vor06 mac28 mac29 mac27 sim39 AmoxP Amox E
1 2 3 4 5 6 7 8 9 10
0.2337 0.2505 0.3412 0.3437 0.3745 0.3918 0.3935 0.4308 0.4444 0.4448
viagra01 vor05 vor07 vor06 mac28 mac29 mac27 sim39 Amox E AmoxP
The database performed a direct (physical) comparison of the test spectra with the reference spectra from the sub-library using a specific algorithm, which was a correlation coefficient in this work. On the other hand, PCA was conducted using latent variables which were the principal components (PC), derived from the maximum variability of the spectra set. The PCA score plot was observed based on the variability presented by PC1 and PC2 only and not the whole spectrum. In this case, PC1 and PC2 would probably represent the active components of E. longifolia. Hence, it could be deduced that all of the test samples contained E. longifolia extract as one of the main components. However, two of the test samples with an HQI above the threshold, ELC and MGC, may also contain other compounds that are significantly different from the rest of the samples. To confirm this, another search was conducted against the unified database that consisted of over 3000 NIR spectra of common medicines acquired
Fig. 2. The 3-dimensional PCA score plot showing the distribution of the test samples purchased as E. Longifolia compared with the spectra of conventional medicine samples. Samples ELC and MGC were clustered close to the cluster of Viagra tablets. (The axis is labelled as X = PC1, Y = PC2, and Z = PC3).
Fig. 3. Flowchart of the two-tier screening protocol summarising the search outcomes for the analysis of E. longifolia samples using (A) dedicated product database; and (B) unified database. The combined used of these two databases indicated the presence of a sildenafil-like compound in ELC and MGC samples.
M.M. Said et al. / Food Chemistry 158 (2014) 296–301
in our laboratory for fifteen types of Malaysian pharmaceutical products. 3.3. Spectra search using the unified database A spectral search of the test samples against the unified database indicated some degree of similarity of ELC and MGC with a set of Viagra tablets (Table 3). In this type of analysis, cut-off point was no longer be used but spectra search giving out the lowest HQI among the 3000 spectra in the database indicated the most similar spectra. This finding was supported by PCA analysis on test and reference spectra from the database whereby close clustering of samples ELC and MGC with the similar set of Viagra tablets was observed (Fig. 2). Considering these outcomes, it can be hypothesised that both ELC and MGC were contaminated with the active substance of Viagra (sildenafil citrate) or a compound chemically related to it. These two samples were analysed using NMR and LC–MS and the presence of sildenafil citrate analogue (hydroxythiohomosildenafil) was confirmed (Supplementary A). The flowchart indicating the search outcomes of the database searches conducted at each stage of data analysis is presented in Fig. 3. Further classification analysis was conducted using SIMCA.
301
the database approach was the lack of graphical representations which was well compensated by the use of PCA. The identification of unknown samples depends on the database content, the identification successes can be improved by including the spectra of diverse pharmaceutical and herbal products. This strategy can be used to build incremental spectral databases of products on different markets as repositories of data for further monitoring and chemometric analysis. These could be used for product identification, drug quality surveillance and as potential methods for counterfeit and adulterated drug screening, particularly in cases where reference samples are difficult to obtain. Acknowledgements The authors thank Professor Peter Hylands (King’s College London) for his valuable comments and inputs in this work. Special thanks to Professor Ibrahim Jantan (Universiti Kebangsaan Malaysia) for his expert assistance in the preparation of the manuscript. This research was sponsored by the Higher Education Ministry of Malaysia and Universiti Kebangsaan Malaysia. Appendix A. Supplementary data
3.4. SIMCA Classification by Cooman’s plot in SIMCA analysis indicated that an object belongs to a model if it distributed within the membership limit, which is to the left of the vertical lines. And if the distribution is near the origin, sample is classified as belonging to both model (Esbensen, Guyot, Westad, & Houmollar, 2005). SIMCA analysis of samples MGC and ELC showed the distribution of the spectra overlapping each other near the origin in the Cooman’s plot (figure included in Supplementary B). Therefore, it can be deduced that these samples were actually the same product despite bearing different brand names and different ingredients on the packaging of both products). The primary packaging (blister packs) of both products and the distributor shown on the secondary packaging looked the same and both products were purchased from the same vendor. Based on the product declarations, the MGC product was 20% more potent than the ELC product and, consequently, the MGC product was sold at a higher price. However, our findings indicate that these two products actually had a very similar composition. 4. Conclusions The application of the two-tier screening method using the NIR spectral database in analysing E. longifolia herbal preparations purchased via the internet was presented in this work. Double screening analyses were conducted using firstly, the dedicated product database (tier 1) which indicated the dissimilarity between two samples (ELC and MGC) followed by analysis using unified database (tier 2), which shown the presence of a sildenafil-like compound in these samples, in addition to E. longifolia compound as claimed. The use of a dedicated product database was justified because the composition of the test samples was known. This method has allowed quick screening on the test samples to verify their content as labelled despite not having the spectra of those products in the database. This is one of the advantages of this procedure whereby analysis can be conducted without depending on the reference product or standard active compounds in the database. Other than that, the method was able to screen the presence of undeclared compound(s) in the products. A limitation of
Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.foodchem.2014. 02.125. References Ang, H. H., Lee, E. L., & Cheang, H. S. (2004). Determination of mercury by cold vapour atomic absorption spectrophotometer in Tongkat Ali preparations obtained in Malaysia. International Journal of Toxicology, 23, 65–71. Ang, H. H., Lee, E. L., & Matsumoto, K. (2003). Analysis of lead content in herbal preparations in Malaysia. Human & Experimental Toxicology, 22, 445–451. Bhat, R., & Karim, A. A. (2010). Tongkat Ali (Eurycoma longifolia Jack): A review on its ethnobotany and pharmacological importance. Fitoterapia, 81(7), 669–679. Champagne, A. B., & Emmel, K. V. (2011). Rapid screening test for adulteration in raw materials of dietary supplements. Vibrational Spectroscopy, 55(2), 216–223. Esbensen, K. H., Guyot, D., Westad, F., & Houmollar, L. P. (2005). Multivariate data analysis. In practice: An introduction to multivariate data analysis and experimental design (5th ed.). Norway: Camo Software AS (Chapter 15). Fleshner, N., Harvey, M., Adomat, H., Wood, C., Eberding, A., & Hersey, K. G. (2005). Evidence for contamination of herbal erectile dysfunction products with phosphodiesterase type 5 inhibitors. The Journal of Urology, 174, 636–641. Khan, P. R., Jee, R. D., Watt, R. A., & Moffat, A. C. (1997). The identification of active drugs in tablets using near infrared spectroscopy. Journal of Pharmaceutical Science, 3, 447–453. Kramer, K., & Ebel, S. (2000). Application of NIR reflectance spectrometry for the identification of pharmaceutical excipients. Analytical Chimica Acta, 420, 155–161. Li, L., Low, M.-Y., Aliwarga, F., Teo, J., Ge, X.-W., Zeng, Y., et al. (2009). Isolation and identification of hydroxythiohomosildenafil in herbal dietary supplements sold as sexual performance enhancement products. Food Additives and Contaminants, 26, 145–151. Moffat, A. C., Assi, S., & Watt, R. A. (2010). Identifying counterfeit medicines using near infrared spectroscopy. Journal of Near Infrared Spectroscopy, 18, 1–15. O’Neil, A. J., Jee, R. D., Lee, G., Charvill, A., & Moffat, A. C. (2008). Use of a portable near infrared spectrometer for the authentication of tablets and the detection of counterfeit versions. Journal of Near Infrared Spectroscopy, 16, 327–333. Said, M. M., Gibbons, S., Moffat, A. C., & Zloh, M. (2011). Near-infrared spectroscopy (NIRS) and chemometric analysis of Malaysian and UK paracetamol tablets: A spectral database study. International Journal of Pharmaceutics, 415, 102–109. Said, M. M., Gibbons, S., & Zloh, M. (2009). Qualitative comparison of Malaysian and UK paracetamol tablets by Near-Infra Red. Journal of Pharmacy and Pharmacology, 61, A112–A113. Said, M. M., Gibbons, S., & Zloh, M. (2010). Chemometrics analysis of solid herbal products and development of spectral databases. Journal of Pharmacy and Pharmacology, 62, 1203–1204. Zhang, J., Wider, B., Shang, H., Li, X., & Ernst, E. (2012). Quality of herbal medicines: Challenges and solutions. Complementary Therapies in Medicine, 20, 100–106. Zhari, I., Norhayati, I., & Jaafar, L. (1999). Malaysian herbal Monograph (Vol. 1). Kuala Lumpur: Malaysian Monograph Committee.
Contents lists available at ScienceDirect
Food Chemistry journal homepage: www.elsevier.com/locate/foodchem
Analytical Methods
Rapid detection of sildenafil analogue in Eurycoma longifolia products using a new two-tier procedure of the near infrared (NIR) spectra database Mazlina Mohd Said ⇑, Simon Gibbons, Anthony C. Moffat, Mire Zloh 1 UCL School of Pharmacy, 29-39 Brunswick Square, London WC1N 1AX, UK
a r t i c l e
i n f o
Article history: Received 18 January 2013 Received in revised form 28 September 2013 Accepted 23 February 2014 Available online 6 March 2014 Keywords: NIRs Spectral database Eurycoma longifolia Sildenafil analogue Principal component analysis SIMCA
a b s t r a c t A simple and cost-effective two-tier drug screening procedure comprises a ‘dedicated’ NIR spectral database of common medicines and a ‘unified’ database was developed to detect the sildenafil analogue in Eurycoma longifolia products. Diffuse reflectance spectra of ten commercial herbal products containing E. longifolia were obtained over the wavelength range of 1100–2500 nm. The spectral search of two products purchased via the internet against a dedicated database of reputable E. longifolia products have resulted in the similarity index of more than 0.1 which indicated significantly different spectra. Further searches against the unified database showed a close match to the spectra of drug containing sildenafil citrate suggesting the presence of a sildenafil analogue. This finding was supported by clustering of these spectra in the PCA score plot within 5% significance level. This approach has alleviated the use of reference product or standard active for direct comparison and has a potential to be used for adulterated food and drugs detection. Ó 2014 Elsevier Ltd. All rights reserved.
1. Introduction Eurycoma longifolia (Family Simaroubaceae) has a long therapeutic history and has become one of the well-known health food products across the world. It is traditionally used as a general tonic to treat high blood pressure, tuberculosis, fever, diarrhoea, jaundice and dysentery (Zhari, Norhayati, & Jaafar, 1999). However, the popularity of E. longifolia today is based on its aphrodisiac effect, which is due to its ability to stimulate the production or action of androgen hormones, especially testosterone (Bhat & Karim, 2010). Most of the quality issues of commercial E. longifolia preparations are associated with contamination with heavy metals (Ang, Lee, & Cheang, 2004; Ang, Lee, & Matsumoto, 2003) and intentional adulteration with drugs or their analogues (Zhang, Wider, Shang, Li, & Ernst, 2012). Sildenafil and its related compounds are the most common adulterants found in herbal preparation used as sexual enhancer or man’s virility products (Champagne & Emmel, 2011). Previous
⇑ Corresponding author. Current address: Drug and Herbal Research Center, Universiti Kebangsaan Malaysia, Jalan Raja Muda Abdul Aziz, 50300 Kuala Lumpur, Malaysia. Tel.: +60 392897972. E-mail address: [email protected] (M.M. Said). 1 Current address: Department of Pharmacy, University of Hertfordshire, College Lane, Hatfield AL10 9AB, UK. http://dx.doi.org/10.1016/j.foodchem.2014.02.125 0308-8146/Ó 2014 Elsevier Ltd. All rights reserved.
works have shown the identification of sildenafil and its analogue in E. longifolia preparations available in the market were carried out using FTIR (Champagne & Emmel, 2011), ESI-MS (Li, Low, Aliwarga, Teo, Ge, Zeng et al., 2009), HPLC–MS (Fleshner, Harvey, Adomat, Wood, Eberding, Hersey & Guns, 2005). Identification using all these methods was made based on the comparison of suspected sample with a reference standard or reference spectra. The combination of NIRS and chemometric analyses such as PCA, PLS and SIMCA, has proven to be simple, fast and robust drug analyses procedures for counterfeit drug detection. Analysis can be made through the comparison of suspected tablets with the spectra of authentic tablets (Moffat, Assi, & Watt, 2010; O’Neil, Jee, Lee, Charvill & Moffat, 2008) and determination of the active pharmaceutical ingredients (APIs) (Khan, Jee, Watt, & Moffat, 1997) or the sources of the tablets (Said, Gibbons, & Zloh, 2009). However, application of these methods mostly depends on the availability of reference compounds or products for comparison with the test spectra. Here, an experimental strategy was designed in order to demonstrate the use of two-tier methods as a potential rapid screening tool for identification and classification of herbal products. A customized spectral database of different brand names of E. longifolia products sold in the Malaysian market was created and validation of the proposed strategy was carried out using three sets of test samples of E. longifolia products purchased via the internet. This
297
M.M. Said et al. / Food Chemistry 158 (2014) 296–301
method has shown the possibility of direct identification of the adulterated samples without relying on standard which is almost impossible using the NIRS techniques mentioned earlier. 2. Experimental
by taking the original averaged NIR spectra, exported in ASCII format and then converted to JCAM-DX files using an in-house programme. The spectral files were then imported into The Unscrambler v9.7 software, (CAMO Software, Oslo, Norway). The spectral database was created using SPECTRAL ID v9.0 database from Gram Suite Software (Thermo Scientific, Waltham, USA).
2.1. Samples 2.4. Database creation and search Seven commercial herbal products stated as containing extracts of E. longifolia were purchased from pharmacies in Malaysia using the convenient sampling. The registration number of each sample was verified with the regulatory agency to ensure the authenticity of these products. Three sets of E. longifolia samples; E. longifolia Complex (manufacturer not stated), Malaysian Ginseng Complex (manufacturer not stated) and Tongkat Ali 100:1 (manufacturer not stated) were purchased via the internet from United Kingdom. Table 1 lists the ingredients of these products as stated on the products’ packaging vendors. Apart from these samples, 20 capsules from each of the 7 brands of E. longifolia were included in a customized spectral database of E. longifolia products. 2.2. NIR analysis The NIR diffuse reflectance analyses were carried out using a NIRSystems 6500 spectrophotometer, equipped with a Rapid Content Analyser (FOSS NIRSystems, Silver Springs, USA). Capsules were emptied into 4 ml clear glass vials with snap caps 45 15 mm (height diameter) (Waters Ltd., Herts, UK). The glass vial was placed in the middle of the sample stage. Each sample was analysed four times, shaking the vials in between acquisition time to avoid the compaction of the powder materials. Eight accumulations of 32 scans of reflectance spectra were recorded for each sample over the wavelength range 1100–2500 nm.
2.4.1. Database development The spectra of reference sample were directly imported into the E. longifolia database (dedicated database) after baseline correction using the auto-baseline correction algorithm to reduce the scattering effects that can highly contaminate NIR spectra (GRAMS Suite Help, 2009). A combination of different types of conventional medicines database that were developed throughout this research, known as the unified database, was also used in this work. In the database, the X-axes of the spectra represented the spectral wavelength in nanometres between the range 1100 and 2500 and the Y-axes were the spectral absorbance values. For each brand of samples, at least 20 averaged spectral profiles with 700 individual points in each were combined as a multi-file to enhance the accuracy of the spectral search. The spectra were saved in 32-bit resolution. Other information recorded together with the spectra included the brand and proprietary names, batch numbers, expiry dates, manufacturer names and addresses, sample origins, other excipients (where available) and a description of the samples. 2.4.2. Database search The spectra of the samples were searched against the dedicated E. longifolia database followed by the universal database using the correlation algorithm based on the whole spectrum (GRAMS Suite Help, 2009):
2.3. Software
HQIc ¼ 1 The NIR instrument was controlled by Vision Spectral Analysis Software for Windows v2.11 (FOSS NIRSystems, Silver Springs, USA) for data acquisition. Chemometric analysis was conducted
ðLm Am Þ2 ðLm Lm ÞðAm Am Þ
where Lm and Am are vectors calculated using all the points in the spectra as defined by equations:
Table 1 List of seven brands of E. longifolia products used as the reference spectra and three test samples purchased via the internet with their names, labels and main active ingredients as stated on the packaging. No.
Samples
Label
Ingredients as stated on the products’ packaging
Reference samples 1 2 3 4 5 6 7
Tongkat Ali Plus Tonex Tongkat Ali Gold Box Tongkat Ali Puteri Rembulan Tongkat Ali Tongkat Ali Capsule LKH Tongkat Ali Nu-Prep100 Tongkat Ali
TA TA TA TA TA TA TA
Mixed preparation with 50 mg radix extract Radix extract 300 mg Radix extract 300 mg Radix extract 500 mg Mix preparations with 354 mg radix extract Radix extract 350 mg Radix extract 100 mg
Internet samples 8
E. longifolia Complex
ELC
Malaysian Ginseng Complex
MGC
Lewtress Tongkat Ali 100:1
LTA
E. longifolia 154 mg Floscarthami 24.5 mg Rhizoma cucurmaelongae 49 mg Gingko biloba extract 49 mg Epimedii 24.5 mg Cistanches 24.5 mg Astragalus membranaceus 24.5 mg Malaysian ginseng complex 157.5 mg Fructus tribulusterestris 52.5 mg Herba epimedium 52.5 mg Radix Smilax myosotiflora 28 mg Radix Panax ginseng 24.5 mg Fructus coriandrumsativum 17.5 mg Trigonella foenumgraecum 17.5 mg 500 mg Pure 100:1 Tongkat ali root extract
9
10.
298
M.M. Said et al. / Food Chemistry 158 (2014) 296–301
Pn Lm ¼ L
i¼1 Li
n
Pn Am ¼ A
i¼1 Ai
n
Ai is the absorbance values of the ith point in the unknown spectrum, Li is the corresponding point of the reference spectrum being searched in the database being compared to, and n is the number of wavelengths in each vector. Using this algorithm, the least squares dot products of the unknown spectra were compared with the spectra in the database after being centralized to their respective means using the hit quality index (HQI) calculated. A low HQI value indicated a good match between the unknown spectra and their matching spectra in the database. HQI equals to zero indicating 100% similarity between two spectra. Based on HQI, a hit list was generated after each search, listing the highest to lowest match of the spectrum tested to each spectrum in the database. 2.5. Chemometric analysis Principal Component Analysis (PCA) and Soft Independent Modelling of Class Analogy (SIMCA) was conducted using The Unscrambler v9.7 software (CAMO Software, Oslo, Norway). 2.5.1. PCA PCA was used to reduce the number of variables in a multivariate data table and to present it in a low dimensional space. PC models were constructed individually for each batch of the samples. After the first run, the presence of outliers, groups, clusters and trends were determined based on the observation of the score plots. At this stage, the outliers detected belonged to the same population but they were poorly described by the model. The optimum number of PCs was determined based on the total explained variance plot. PCA was conducted on the data with leverage correction as the validation method and the scaling factor was set as 1. 2.5.2. SIMCA Classification using the SIMCA method was carried out on the suspected samples by projecting them on all of the PC models with the significance level set at 95%. The orthogonal distances from the new objects to two different classes (models) were presented in Cooman’s plots. In this plot, the orthogonal distances from the test samples to two different classes (models) were compared at the same time in pair wise plots. 3. Results and discussion Spectroscopic fingerprints obtained by NIR provided a distinct representation of individual herbal formulations despite their similar chemical entities since it reflected not only the active constituents but also all of the other chemical compositions/excipients and it also indicated the variability in the production processes used. The average NIR spectra of each of the E. longifolia samples after SNV transformation are shown in Fig. 1A. The test set consisted of three products purchased via the internet (ELC, MGC, and LTA). Their NIR spectra are shown in red in three different line shapes. The spectra for ELC and LTA can be distinguished from the rest of the set while the spectrum for sample MGC overlaps with other spectra. The PCA of all spectra for the E. longifolia products was carried out and the PCA score plot indicated that each brand of the herbal product was clustered according to the manufacturer (Fig. 1B). Large variations of herbal products from different productions were mainly due to the different extraction methods and use of excipients in some of the samples. Two sets of the commercial preparation (TA and TA) with the three sets of test samples contained other constituents of plant origin in addition to
E. longifolia. However, based on the product labelling (Table 1), these constituents present in a small amount compared to the main active component. Therefore, The PCA score plot of all spectra were grouped in the 95% confidence limit indicating that the purchased samples were similar to other products which did not suggest that there were any quality issues with the samples. 3.1. Two-tier method analysis The use of the NIR spectral database as a basis for the two-tier method has been demonstrated in our previous work on the qualitative analysis of different brands of paracetamol tablets purchased in pharmacies and supermarkets in Malaysia (Said, Gibbons, Moffat, & Zloh, 2011; Said, Gibbons, & Zloh, 2010). In this study, the application of dedicated and unified databases for the quick identification of unknown samples and similarity assessment of products are described. A ‘dedicated’ database is a single product database that contains spectra from a specific type of product or class of medicines. Here, the dedicated database consisted of herbal preparations with E. longifolia as the main (if not only) active component. The adequate use of this type of database can only be used when the main components of the sample are known. This procedure was used to verify and to confirm if a sample belongs to a particular class of products and can also be used to further identify the sample according to the manufacturer or batches available in the database. The use of cut-off points were required to perform classification (Kramer & Ebel, 2000). The ‘unified’ database is the combination of all the dedicated product databases and it contained all the spectra of the products purchased up to date. In a larger database, that includes different types of conventional medicines, it might be possible to detect the presence of other compounds or adulterants in a product. The use of cut-off points was no longer relevant for this purpose. However, similarity between spectra could provide some information about sample composition. 3.2. Spectra search using the dedicated database The similarity of the spectra was first assessed using the dedicated database containing only the spectra of E. longifolia products. The average NIR spectra of the three test samples (samples ELC, LTA, and MGC) bought over the internet were imported into the single product database; consisting of different brands of E. longifolia products. A full spectrum search was conducted on these samples against the database to confirm their similarity using a correlation algorithm for the whole spectrum. A hit-list of the top matches was provided together with hit quality index (HQI) values after each search. The cut-off values determined in our previous work were used for samples classification (Said et al., 2011). These classifications were divided into four types; Type 1 (same batch of sample will generate a HQI of less than 0.0001); Type 2 (different batch of products of the same brand name will generate HQI of less than 0.01; Type 3 (different brand names of samples from the same class of drugs or herbals will generate HQI of less than 0.1; and Type 4 (completely different type of samples will generate HQI of more than 0.2). According to the pre-determined cut-off values, spectra search of sample LTA (HQI = 0.0464) fell within the same classification with the other spectra of E. longifolia in the database (Type 3). However, the search outcomes on the other two samples ELC and MGC have generated higher HQI values 0.1133 and 0.1297, respectively (Table 2). This has led to an inconclusive finding of the similarity between these sets of samples to the database and indicated the possibility of these samples having a different composition.
299
M.M. Said et al. / Food Chemistry 158 (2014) 296–301
Fig. 1. (A) Average SNV spectra from the different commercial products of E. longifolia in the wavelength range 1400–2400 nm. (B) PCA score plots of the NIR spectra of commercial E. longifolia herbal formulations purchased randomly in the Malaysian market. The test samples; ELC, MGC and LTA (pointed) were clustered within the 95% confidence limit of other samples.
The database search outcome and the observation of the PCA score plot (Fig. 1B) did not give agreeable results in this analysis. Use of the database indicated that there was some dissimilarity of the spectra based on the higher HQI while PCA indicated that
the test samples were similar to the reference spectra as both sets clustered within the 95% confidence ellipse. The contradicting outcome was possible as the database and PCA used different methods in assessing the similarity between the spectra.
Table 2 The hit-list showing the database search outcomes for samples (A) ELC (B) LTA and (C) MGC against the E. longifolia dedicated database. The HQI and sample in bold showed slightly higher values that the cut-off threshold pre-determined. (A) ELC
(B) LTA
(C) MGC
Hit #
HQI
Sample ID
Hit#
HQI
Sample ID
Hit #
HQI
Sample ID
1 2 3 4 5 6 7 8 9 10
0.1133 0.1707 0.1881 0.1995 0.2049 0.2247 0.2355 0.2500 0.2504 0.3760
TA TA TA TA TA TA TA TA TA TA
1 2 3 4 5 6 7 8 9 10
0.0464 0.0571 0.1170 0.1564 0.2434 0.3421 0.3725 0.3853 0.3878 0.5127
TA TA TA TA TA TA TA TA TA TA
1 2 3 4 5 6 7 8 9 10
0.1297 0.1564 0.1810 0.2036 0.2212 0.2498 0.2595 0.2745 0.2754 0.4046
TA TA TA TA TA TA TA TA TA TA
300
M.M. Said et al. / Food Chemistry 158 (2014) 296–301
Table 3 The hit-list showing the database search outcomes for samples (A) ELC and (B) MGC against the general customized database with the top-ten closest match. Both samples ELC and MGC showed highest similarity with Viagra tablets. Bold form is to highlight the similarity of the samples tested with viagra tablets. (A) ELC
(B) MGC
Hit #
HQI
Sample ID
Hit #
HQI
Sample ID
1 2 3 4 5 6 7 8 9 10
0.2179 0.2476 0.3308 0.3339 0.3621 0.3794 0.3814 0.4415 0.4626 0.4685
viagra01 vor05 vor07 vor06 mac28 mac29 mac27 sim39 AmoxP Amox E
1 2 3 4 5 6 7 8 9 10
0.2337 0.2505 0.3412 0.3437 0.3745 0.3918 0.3935 0.4308 0.4444 0.4448
viagra01 vor05 vor07 vor06 mac28 mac29 mac27 sim39 Amox E AmoxP
The database performed a direct (physical) comparison of the test spectra with the reference spectra from the sub-library using a specific algorithm, which was a correlation coefficient in this work. On the other hand, PCA was conducted using latent variables which were the principal components (PC), derived from the maximum variability of the spectra set. The PCA score plot was observed based on the variability presented by PC1 and PC2 only and not the whole spectrum. In this case, PC1 and PC2 would probably represent the active components of E. longifolia. Hence, it could be deduced that all of the test samples contained E. longifolia extract as one of the main components. However, two of the test samples with an HQI above the threshold, ELC and MGC, may also contain other compounds that are significantly different from the rest of the samples. To confirm this, another search was conducted against the unified database that consisted of over 3000 NIR spectra of common medicines acquired
Fig. 2. The 3-dimensional PCA score plot showing the distribution of the test samples purchased as E. Longifolia compared with the spectra of conventional medicine samples. Samples ELC and MGC were clustered close to the cluster of Viagra tablets. (The axis is labelled as X = PC1, Y = PC2, and Z = PC3).
Fig. 3. Flowchart of the two-tier screening protocol summarising the search outcomes for the analysis of E. longifolia samples using (A) dedicated product database; and (B) unified database. The combined used of these two databases indicated the presence of a sildenafil-like compound in ELC and MGC samples.
M.M. Said et al. / Food Chemistry 158 (2014) 296–301
in our laboratory for fifteen types of Malaysian pharmaceutical products. 3.3. Spectra search using the unified database A spectral search of the test samples against the unified database indicated some degree of similarity of ELC and MGC with a set of Viagra tablets (Table 3). In this type of analysis, cut-off point was no longer be used but spectra search giving out the lowest HQI among the 3000 spectra in the database indicated the most similar spectra. This finding was supported by PCA analysis on test and reference spectra from the database whereby close clustering of samples ELC and MGC with the similar set of Viagra tablets was observed (Fig. 2). Considering these outcomes, it can be hypothesised that both ELC and MGC were contaminated with the active substance of Viagra (sildenafil citrate) or a compound chemically related to it. These two samples were analysed using NMR and LC–MS and the presence of sildenafil citrate analogue (hydroxythiohomosildenafil) was confirmed (Supplementary A). The flowchart indicating the search outcomes of the database searches conducted at each stage of data analysis is presented in Fig. 3. Further classification analysis was conducted using SIMCA.
301
the database approach was the lack of graphical representations which was well compensated by the use of PCA. The identification of unknown samples depends on the database content, the identification successes can be improved by including the spectra of diverse pharmaceutical and herbal products. This strategy can be used to build incremental spectral databases of products on different markets as repositories of data for further monitoring and chemometric analysis. These could be used for product identification, drug quality surveillance and as potential methods for counterfeit and adulterated drug screening, particularly in cases where reference samples are difficult to obtain. Acknowledgements The authors thank Professor Peter Hylands (King’s College London) for his valuable comments and inputs in this work. Special thanks to Professor Ibrahim Jantan (Universiti Kebangsaan Malaysia) for his expert assistance in the preparation of the manuscript. This research was sponsored by the Higher Education Ministry of Malaysia and Universiti Kebangsaan Malaysia. Appendix A. Supplementary data
3.4. SIMCA Classification by Cooman’s plot in SIMCA analysis indicated that an object belongs to a model if it distributed within the membership limit, which is to the left of the vertical lines. And if the distribution is near the origin, sample is classified as belonging to both model (Esbensen, Guyot, Westad, & Houmollar, 2005). SIMCA analysis of samples MGC and ELC showed the distribution of the spectra overlapping each other near the origin in the Cooman’s plot (figure included in Supplementary B). Therefore, it can be deduced that these samples were actually the same product despite bearing different brand names and different ingredients on the packaging of both products). The primary packaging (blister packs) of both products and the distributor shown on the secondary packaging looked the same and both products were purchased from the same vendor. Based on the product declarations, the MGC product was 20% more potent than the ELC product and, consequently, the MGC product was sold at a higher price. However, our findings indicate that these two products actually had a very similar composition. 4. Conclusions The application of the two-tier screening method using the NIR spectral database in analysing E. longifolia herbal preparations purchased via the internet was presented in this work. Double screening analyses were conducted using firstly, the dedicated product database (tier 1) which indicated the dissimilarity between two samples (ELC and MGC) followed by analysis using unified database (tier 2), which shown the presence of a sildenafil-like compound in these samples, in addition to E. longifolia compound as claimed. The use of a dedicated product database was justified because the composition of the test samples was known. This method has allowed quick screening on the test samples to verify their content as labelled despite not having the spectra of those products in the database. This is one of the advantages of this procedure whereby analysis can be conducted without depending on the reference product or standard active compounds in the database. Other than that, the method was able to screen the presence of undeclared compound(s) in the products. A limitation of
Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.foodchem.2014. 02.125. References Ang, H. H., Lee, E. L., & Cheang, H. S. (2004). Determination of mercury by cold vapour atomic absorption spectrophotometer in Tongkat Ali preparations obtained in Malaysia. International Journal of Toxicology, 23, 65–71. Ang, H. H., Lee, E. L., & Matsumoto, K. (2003). Analysis of lead content in herbal preparations in Malaysia. Human & Experimental Toxicology, 22, 445–451. Bhat, R., & Karim, A. A. (2010). Tongkat Ali (Eurycoma longifolia Jack): A review on its ethnobotany and pharmacological importance. Fitoterapia, 81(7), 669–679. Champagne, A. B., & Emmel, K. V. (2011). Rapid screening test for adulteration in raw materials of dietary supplements. Vibrational Spectroscopy, 55(2), 216–223. Esbensen, K. H., Guyot, D., Westad, F., & Houmollar, L. P. (2005). Multivariate data analysis. In practice: An introduction to multivariate data analysis and experimental design (5th ed.). Norway: Camo Software AS (Chapter 15). Fleshner, N., Harvey, M., Adomat, H., Wood, C., Eberding, A., & Hersey, K. G. (2005). Evidence for contamination of herbal erectile dysfunction products with phosphodiesterase type 5 inhibitors. The Journal of Urology, 174, 636–641. Khan, P. R., Jee, R. D., Watt, R. A., & Moffat, A. C. (1997). The identification of active drugs in tablets using near infrared spectroscopy. Journal of Pharmaceutical Science, 3, 447–453. Kramer, K., & Ebel, S. (2000). Application of NIR reflectance spectrometry for the identification of pharmaceutical excipients. Analytical Chimica Acta, 420, 155–161. Li, L., Low, M.-Y., Aliwarga, F., Teo, J., Ge, X.-W., Zeng, Y., et al. (2009). Isolation and identification of hydroxythiohomosildenafil in herbal dietary supplements sold as sexual performance enhancement products. Food Additives and Contaminants, 26, 145–151. Moffat, A. C., Assi, S., & Watt, R. A. (2010). Identifying counterfeit medicines using near infrared spectroscopy. Journal of Near Infrared Spectroscopy, 18, 1–15. O’Neil, A. J., Jee, R. D., Lee, G., Charvill, A., & Moffat, A. C. (2008). Use of a portable near infrared spectrometer for the authentication of tablets and the detection of counterfeit versions. Journal of Near Infrared Spectroscopy, 16, 327–333. Said, M. M., Gibbons, S., Moffat, A. C., & Zloh, M. (2011). Near-infrared spectroscopy (NIRS) and chemometric analysis of Malaysian and UK paracetamol tablets: A spectral database study. International Journal of Pharmaceutics, 415, 102–109. Said, M. M., Gibbons, S., & Zloh, M. (2009). Qualitative comparison of Malaysian and UK paracetamol tablets by Near-Infra Red. Journal of Pharmacy and Pharmacology, 61, A112–A113. Said, M. M., Gibbons, S., & Zloh, M. (2010). Chemometrics analysis of solid herbal products and development of spectral databases. Journal of Pharmacy and Pharmacology, 62, 1203–1204. Zhang, J., Wider, B., Shang, H., Li, X., & Ernst, E. (2012). Quality of herbal medicines: Challenges and solutions. Complementary Therapies in Medicine, 20, 100–106. Zhari, I., Norhayati, I., & Jaafar, L. (1999). Malaysian herbal Monograph (Vol. 1). Kuala Lumpur: Malaysian Monograph Committee.
