Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
Contents lists available at ScienceDirect
Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy journal homepage: www.elsevier.com/locate/saa
Rapid identification of illegal synthetic adulterants in herbal anti-diabetic medicines using near infrared spectroscopy Feng Yanchun a, Lei Deqing b, Hu Changqin a,⇑ a b
National Institutes for Food and Drug Control, Beijing 100050, China Shaoyang Institute for Drug Control of Hunan, Shaoyang 422000, China
h i g h l i g h t s
g r a p h i c a l a b s t r a c t
A NIR procedure to detect illicit
A NIR procedure for the rapid, nondestructive identification of herbal medicines containing illicit synthetic drugs was constructed during this study.
adulterants in herbal medicines was constructed. This rapid, nondestructive procedure can reach an accuracy level over 80%. The strategy for RCCM threshold determination has been improved.
a r t i c l e
i n f o
Article history: Received 1 February 2013 Received in revised form 23 January 2014 Accepted 26 January 2014 Available online 8 February 2014 Keywords: Near infrared spectroscopy Reverse correlation coefficient method Synthetic drugs Adulterants Herbal medicines
a b s t r a c t We created a rapid detection procedure for identifying herbal medicines illegally adulterated with synthetic drugs using near infrared spectroscopy. This procedure includes a reverse correlation coefficient method (RCCM) and comparison of characteristic peaks. Moreover, we made improvements to the RCCM based on new strategies for threshold settings. Any tested herbal medicine must meet two criteria to be identified with our procedure as adulterated. First, the correlation coefficient between the tested sample and the reference must be greater than the RCCM threshold. Next, the NIR spectrum of the tested sample must contain the same characteristic peaks as the reference. In this study, four pure synthetic anti-diabetic drugs (i.e., metformin, gliclazide, glibenclamide and glimepiride), 174 batches of laboratory samples and 127 batches of herbal anti-diabetic medicines were used to construct and validate the procedure. The accuracy of this procedure was greater than 80%. Our data suggest that this protocol is a rapid screening tool to identify synthetic drug adulterants in herbal medicines on the market. Ó 2014 Elsevier B.V. All rights reserved.
Introduction
⇑ Corresponding author. Address: National Institutes for Food and Drug Control, Antibiotic Division, #2 Tiantan Xili, Beijing 100050, China. Tel.: +86 10 67095308; fax: +86 10 65115148. E-mail addresses: [email protected], [email protected] (C. Hu). http://dx.doi.org/10.1016/j.saa.2014.01.117 1386-1425/Ó 2014 Elsevier B.V. All rights reserved.
Herbal medicines are popular worldwide for their natural origins and healing properties. In contrast to conventional synthetic drugs, herbal medicines are generally perceived as safe, harmless and without deleterious side effects. However, nowadays,
364
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
herbal medicines have not remained trustworthy as lots of evidences are coming into literature about adulteration of these products with synthetic drugs, in order to enhance the claims stated on the label [1]. Over the past several years, there have been various reports from Asia [2,3], Europe [4], Africa [5], USA [6] and Brazil [7], regarding this unethical trend. Such counterfeit drugs or adulterated products are commonly analyzed using high-performance liquid chromatography (HPLC) [8], mass spectrometry (MS) [9] or nuclear magnetic resonance (NMR) [10]. Recently voltammetry of microparticles [11] and high-performance thin layer chromatography (HPTLC) densitometry [12] have been used to detect these illicit drugs. Each of these techniques requires substantial time and complicated sample preprocessing, preventing opportunities for identifying and seizing such products. Near infrared (NIR) is a powerful tool for counterfeit drug identification because it is a quick, non-destructive, and reagent-free technique [13]. Since 2002, the National Institutes for Food and Drug Control (NIFDC) has conducted a large scale study on NIR rapid drug screening and more than 300 mobile laboratory vehicles in China have equipped with NIR prescreening systems. These vehicles are used to quickly evaluate drug quality on site, in open market as well as in the distribution channel. The NIR prescreening system, utilizing a universal model, can successfully detect illegal common synthetic drugs [14,15]. Recently, we have used NIR to identify herbal medicines adulterated with synthetic drugs. Because herbal medicines are complex, NIR is more problematic for identifying adulterants. We did observe that in the NIR region of 6200–5500 cm1, the herbal medicine adulterated with sildenafil citrate had considerable similarity to the sildenafil citrate reference substance. Thus the NIR reverse correlation coefficient method (RCCM) was developed and it can detect sildenafil citrate illegally adulterated in herbal aphrodisiacs medicines accurately [16]. Moreover, Lu and co-workers [17] demonstrated infrared absorption differences between synthetic drugs which had sharp, abrupt peaks and herbal medicines that had broad, smooth peaks. Therefore, we hypothesized that NIR profiles of herbal medicines adulterated with a synthetic drug would have characteristic peaks of the corresponding synthetic adulterant. Upon validation of this hypothesis, construction of a robust NIR procedure, capable of detecting the most common synthetic adulterants in herbal medicines, is a worthwhile endeavor. Herbal anti-diabetic medicines containing undeclared pharmaceuticals are a significant problem because patients taking these illicit products may experience potentially fatal adverse effects. Such illicit herbal medicines usually contain undeclared, registered and/ or banned, oral anti-diabetic agents, including metformin, gliclazide, glibenclamide and glimepiride etc. [18]. Here herbal antidiabetic medicines dosed with synthetic adulterants were tested to investigate the usefulness of RCCM in this application. Based on these findings, a general detection procedure for common synthetic adulterants found in herbal medicines was developed using the RCCM in tandem with NIR characteristic peaks comparisons.
Materials and methods
medicines referred to herbal medicines adulterated with synthetic drugs. Reference substances of metformin, gliclazide, glibenclamide and glimepiride were obtained from the NIFDC. To determine if synthetic anti-diabetic drugs had characteristic regions when adulterated into herbal medicines, five kinds of antidiabetic herbal medicines, named Qi Lu Wen Shen Jiao Nang, Gan Lu Xiao Ke Jiao Nang, Pu Nuo Ning Si Jiao Nang, Zhong Yan Wan Tong Jiao Nang and Sang Ge Jiao Nang, were selected as blanks. These products were removed from capsules and milled into small granules before using as blanks. Laboratory samples were made by mixing given amounts of pure synthetic drugs into blank samples. See Table 1 for the composition of the laboratory samples. Instrumentation and data acquisition FT-NIR spectrometers, MATRIX-F, from Bruker Optik GmbH (Ettlingen, Germany) were used for all experiments. All spectrometers are equipped with a 1.5 m fiber-optic diffuse reflectance probe and an extended TE-cooled indium gallium arsenide (InGaAs) detector. Data were collected and processed using OPUS software (version 6.5). Diffuse reflectance spectra were recorded using a 1.5 m fiber optic probe at 8 cm1 resolution with 32 co-added scans over the spectral range of 4000–12,000 cm1. For capsule samples, contents were transferred into glass bottles and a fiber-optic probe was inserted to record spectra. For tablet samples, spectra were recorded via a randomly selected tablet surface scan. Spectral pretreatment Data pretreatment was performed to reduce the influence of physical parameters. First, the average spectrum was calculated using six original spectra from each sample. Then a second derivation was performed on the averaged spectra with 17-point smoothing followed by vector normalization, to enhance spectral information and reduce baseline drift. Only the preprocessed, averaged spectra were used in the construction and validation of these methods. Reverse correlation coefficient method RCCM is a new detection method for illicit synthetic adulterants in herbal medicines using NIR spectroscopy [16]. RCCM and the traditional correlation coefficient method use the same formula to calculate the correlation coefficient (r) (Eq. 1). However, significant differences exist between the two methods: RCCM uses a pure adulterant spectrum as the reference spectrum, and can detect herbal medicines from different manufacturers simultaneously, whereas the traditional correlation coefficient method requires a spectrum of a genuine sample from one manufacturer as the reference spectrum (Fig. 1). Because different manufacturers produce various products, one traditional correlation coefficient method can be only used for detecting the corresponding manufacturer’s product.
Cov ðy1 ðkÞ; y2 ðkÞÞ ry1 ry2
Sample preparation
r¼
Both genuine herbal medicines and counterfeit ones adulterated with synthetic drugs were obtained from the Chinese market by the NIFDC, the Beijing Institute for Drug Control, and the Yulin Institute for Drug Control in Guangxi Province. Most of the herbal medicines used here were capsules and the others were tablets. All samples were identified according to official laboratory methods. Here genuine herbal medicines were the samples passed the examination of its official laboratory method and counterfeit herbal
y(k): all Y data points, when a spectrum is presented as a data point table (X, Y), Cov(y1 (k), y2 (k)): the covariance of y1(k) and y2(k) and ry: the standard deviation of y(k). The key parameter of RCCM is the threshold. By comparing the r and a preset corresponding threshold, we could identify whether the tested herbal medicine is adulterated with synthetic drugs. The minimum effective concentration (MEC) of an adulterant was used to set the RCCM threshold (Eq. 2) as we previously reported
ð1Þ
365
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374 Table 1 Adulteration of herbal laboratory samples. Sample No.
1 2 3 4 5 6 7 8 9 10 11 12 13
Metformin
Gliclazide
Glibenclamide
Blank (mg)
Metformin (mg)
Blank (mg)
Gliclazide (mg)
Blank (mg)
Glibenclamide (mg)
Blank (mg)
Glimepiride (mg)
400 400 400 400 400 400 400 400 400 400 400 400 400
200 160 120 60 40 20 12 8 4 2 1.2 0.6 0.4
400 400 400 400 400 400 400 400 – – – – –
48 40 32 20 10 8 6 4
400 400 400 400 400 400 400 – – – – – –
40 20 10 5 4 2 1.2
400 400 400 400 400 400 400 400 400 – – – –
20 16 10 8 4 2 1.2 0.8 0.4
Fig. 1. Comparison between traditional correlation coefficient method and the reverse correlation coefficient method ⁄ ‘‘Counterfeit drugs’’ refer to herbal medicines adulterated with synthetic drugs.
[16]. The maximum weight of a single tablet or capsule in Eq. (2) is set at 500 mg. A laboratory sample with a content equal to the MEC was prepared, and the r between the reference sample and the corresponding laboratory sample was set as the RCCM threshold.
MEC ¼
Glimepiride
The minimum specification of the pure adulterant The maximum weight of a single tablet or capsule
characteristic spectral regions of metformin, gliclazide, glimepiride and glibenclamide as herbal adulterants, spectra of laboratory samples, counterfeit anti-diabetic herbal medicines with adulterants as well as pure synthetic drug were investigated. Fig. 2 depicts original spectra for reference substances of metformin, gliclazide, glimepiride and glibenclamide. Data show that these compounds can be distinguished from one another. Next, we compared the spectra of laboratory samples and their corresponding pure adulterants. Spectra of metformin and the laboratory samples dosed with various concentrations of metformin are shown in Fig. 3. These spectra closely matched in regions 6701–6515 cm1, 6030–5570 cm1, 5070–4700 cm1, and 4480–4220 cm1. Moreover, their spectral intensities decreased as metformin concentration declined. Multiple counterfeit anti-diabetic herbal medicines adulterated with metformin were compared to determine if the characteristic regions selected using laboratory samples could be used for real counterfeit samples from different manufacturers which had different excipients and formulations. Typical spectra of the counterfeit medicines are depicted in Fig. 4. These counterfeit herbal medicines were similar to pure metformin, with similar spectral peaks in our selected regions. There were, however, significant differences among their spectral whole shape. This technique was then used to search characteristic regions of herbal medicines adulterated with gliclazide, glimepiride and glibenclamide. Data presented in Table 2 indicate that all four synthetic drugs have unique NIR peaks, even when incorporated into herbal medicines. In addition, a common characteristic spectral region (6200–5500 cm1) appeared in all spectra of herbal medicines adulterated with these four synthetic drugs. Therefore, the 6200– 5500 cm1 region may be an important region for detecting other synthetic drugs in herbal medicines. Next, NIR spectra of the reference substances of metformin, gliclazide, glimepiride and glibenclamide were used as reference spectra to construct RCCMs, respectively, using their corresponding characteristic regions shown in Table 2.
ð2Þ
Results and discussion Characteristic region selection Herbal medicines adulterated with sildenafil citrate are similar to pure sildenafil citrate in the spectral region of 6200–5500 cm1 [16]. Previous studies suggest that most APIs (active pharmaceutical ingredients) also have obvious characteristic peaks within the NIR regions of 6200–5500 cm1 and 4100–4900 cm1 [19]. To find
Setting threshold for RCCM A critical step in RCCM model construction is determination of the threshold. If the r between the tested spectrum and the reference spectrum is greater than the threshold, the tested herbal medicine will be identified as containing the examined synthetic adulterant. Our previous study used the MEC of illicit adulterants to set the threshold for the RCCM. However, when setting the RCCM thresholds for metformin, gliclazide, glimepiride and glibenclamide, we found that the MECs of the compounds were not always suitable for this purpose. Based on the following study,
366
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
Fig. 2. NIR spectra of pure synthetic drugs. (1. metformin 2. gliclazide 3. glimepiride 4. glibenclamide).
the minimum detectable concentration (MDC) of the adulterants in herbal medicines was best for setting RCCM thresholds. In this study, any MDC of the adulterants in herbal medicines must meet two criteria: First, when the adulterant concentration in an herbal medicine is higher than its MDC, the r between the herbal medicine and the pure adulterant increases with the adulterant concentration increasing; Second, when the adulterant concentration in an herbal medicine is lower than its MDC, the r between the herbal medicine and the pure adulterant has no obvious correlation with the adulterant concentration, and/or the RCCM method cannot distinguish the herbal medicine with the adulterant from genuine herbal medicines. To set the RCCM threshold for metformin detection in herbal medicines, the r between the reference spectrum (namely the spectrum of the pure metformin) and laboratory samples (Qi Lu Wen Shen Jiao Nang was as a blank) was calculated in the selected regions. The r between the metformin reference spectrum and 12 batches of genuine anti-diabetic herbal products was also calculated. All r values are presented in Fig. 5A and B. Data indicate that at lower concentrations of metformin, the correlation coefficient decreased. When the concentration of metformin decreased below 2%, the r between the metformin reference and laboratory samples no longer had correlation with their corresponding adulterant concentrations; besides, the r between the reference spectrum and laboratory samples was similar to the r between the reference spectrum and the genuine herbal anti-diabetic products. Thus, when metformin in Qi Lu Wen Shen Jiao Nang was less than 2%, it became difficult to correctly identify whether the herbal medicine was adulterated with metformin by RCCM. Therefore, the MDC of metformin in Qi Lu Wen Shen Jiao Nang was established at 2%, and the corresponding r for this metformin concentration was 29.63% (Fig. 5B). The r between the other three pure synthetic drugs (gliclazide, glimepiride, and glibenclamide), their corresponding laboratory samples (Qi Lu Wen Shen Jiao Nang was as a blank), and the genuine products are shown in Fig. 5C–E. For gliclazide (Fig. 5C), when the concentration was below 2.5%, the r between pure gliclazide and laboratory samples was zero, a value smaller than the values
between pure gliclazide and the genuine products. Thus, the MDC for gliclazide in Qi Lu Wen Shen Jiao Nang was 2.5%. Likewise, the MDC of both glimepiride and glibenclamide in Qi Lu Wen Shen Jiao Nang was 1.0% (Fig. 5D and E). According to the Chinese National Formulary [20], the minimum specifications (MS) of metformin, gliclazide, glimepiride and glibenclamide are given in Table 3. The MEC for each compound was calculated from Eq. (2) (Table 3). When comparing these data, MDCs of glimepiride and glibenclamide were higher than their MECs. Under this condition, the r calculated from the pure adulterant and its laboratory sample at MEC is incredible, which may be confused with the r between the pure adulterant and genuine herbal medicines. If the RCCM threshold for glimepiride or glibenclamide is set based on its MEC, the RCCM may not correctly identify part of genuine herbal medicines. So MDC values must be used to set the RCCM threshold for glimepiride and glibenclamide. In contrast, MDC values for metformin and gliclazide were lower than their MECs. So, if their MECs are used to set RCCM thresholds, herbal adulterants less than the MEC but higher than the MDC would produce a false negative result. Collectively, using MDC to set RCCM thresholds is more reliable than using the MEC. First, for the adulterants with higher MS, their MEC in herbal products exceeds their MDC. In this case, if the MDC is used to set the threshold, the RCCM detectable range is significantly increased. Conversely, for the synthetic adulterants with lower MS, the calculated MEC may be smaller than their MDC. Then, the RCCM threshold calculated from MEC is incredible because the RCCM could not correctly identify whether an herbal medicine is adulterated with a synthetic drug when the concentration of the synthetic drug in herbal medicines is lower than its MDC. Moreover, an MEC is an empirical value, equal to the MS of the pure adulterant divided by the maximum weight of a single tablet or a single capsule (i.e. 500 mg). The maximum weight is a statistical value, and the real weight of a single tablet or capsule may be higher or lower than 500 mg, which increases the uncertainty in the accuracy of thresholds calculated using MEC values. In contrast, not only is the MDC an exact value derived from experimentation, but it is also the minimum concentration which can be
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
367
Fig. 3. NIR spectra of blank, pure metformin, and laboratory samples containing metformin. (A) Qi Lu Wen Shen Jiao Nang was selected as blank. (B) Gan Lu Xiao Ke Jiao Nang was selected as blank. (C) Pu Nuo Ning Si Jiao Nang was selected as blank. (D) Zhong Yan Wan Tong Jiao Nang was selected as blank. (E) Sang Ge Jiao Nang was selected as blank. Numbers 1–7 represent different metformin concentrations in the blank.
identified by its specific RCCM within specific ranges. Therefore, in this context, the MDC values were used to set RCCM thresholds. The initial RCCM thresholds for the four adulterants calculated from their MDCs are shown in Table 3. To validate the initial thresholds, set using Qi Lu Wen Shen Jiao Nang as a blank, is universal to other blanks, the r between the laboratory samples made using other four kinds of genuine herbal medicines as blanks (Gan Lu Xiao Ke Jiao Nang, Pu Nuo Ning Si Jiao Nang, Zhong Yan Wan Tong Jiao Nang and Sang Ge Jiao Nang) and their corresponding reference were calculated, and then were
plotted with adulterant concentrations (Fig. 6). Most of the r between the laboratory samples and their corresponding pure adulterants increased with the increase of their adulterant concentrations in the whole concentration range. Therefore, the major determinant of MDC was the r between the reference spectrum and the genuine herbal anti-diabetic products. Fig. 6A indicated as long as the RCCM threshold was set to about 30%, the metformin RCCM can detect metformin adulterated in herbal medicines correctly. So the MDC of metformin was the concentration corresponding to 30% in Fig. 6A. It was found that although the MDCs
368
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
Fig. 3 (continued)
of metformin in different blanks were different, the r values corresponding to the MDCs in different blanks were identical. In other words, the thresholds of the metformin RCCM set by these four kinds of blanks respectively were consistent with that set using Qi Lu Wen Shen Jiao Nang. Likewise, the RCCM thresholds of glimepiride and glibenclamide set using Qi Lu Wen Shen Jiao Nang as a blank, were similar to that set using other four kinds of blanks (Fig. 6C and D). Only for gliclazide RCCM, because there were no enough laboratory samples made by Qi Lu Wen Shen Jiao Nang, which was result in high threshold. Based on the data presented in Fig. 6B, the threshold for gliclazide was adjusted to 4%. In addition, it was also found that RCCM thresholds related to the representativeness of the genuine herbal products we selected to set
MDC. Therefore with the spectra of genuine herbal products we can obtain increasing during RCCM use, RCCM thresholds will be adjusted to more reasonable values. Characteristic peak comparisons used in tandem with RCCM One problem showed up when the RCCM model construction was finished. Using correlation coefficient to evaluate similarities between the two spectra may be limited: the spectral correlation coefficient between the reference and the herbal medicines, which did not contain the examined synthetic drug, might exceed the threshold. For example, one batch of anti-diabetic herbal capsules adulterated with rosiglitazone and pioglitazone, the RCCM
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
369
Fig. 3 (continued)
Fig. 4. NIR spectra of counterfeit herbal medicines adulterated with metformin. (1. Yi Dao Zai Sheng Jiang Tang Jiao Nang 2. Hong Qi Yi Hao 3. Huo Yi Xue Tang Ping Jiao Nang 4. Sheng Jin Zhi Ke Jiao Nang 5. Jiang Tang Ning Jiao Nang).
Table 2 Characteristic regions of synthetic anti-diabetic drugs. No.
Synthetic anti-diabetic adulterants
Characteristic regions (cm1)
1
Metformin
2 3 4
Gliclazide Glimepiride Glibenclamide
6701–6515, 6030–5570, 5070–4700, 4480–4220 6100–5600, 4333–4220, 6000–5500, 5000–4490 6200–5500, 5000–4500
incorrectly identified the adulterant as metformin. Its correlation coefficient with pure metformin was 68.76%; while the threshold of metformin RCCM was 30%. However, no characteristic metformin peaks were found in the spectrum of the tested herbal capsule (Fig. 7). Comparing the spectra of the laboratory samples adulterated with different concentrations of metformin and the misidentified sample revealed that the characteristic peaks of the metformin adulterated samples were obvious when their second derivative spectra were observed, even with low concentrations
370
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
Fig. 5. Plots of correlation coefficients vs. the concentration of the synthetic adulterants in herbal medicines. (A) Metformin ranged from 5% to 40%. (B) Metformin ranged from 0% to 5%. (C) gliclazide, (D) glimepiride and (E) glibenclamide.
Table 3 Parameters for RCCM threshold settings.
MS MEC MDC r at MDC Initial threshold
Metformin
Gliclazide
Glimepiride
Glibenclamide
250 mg 50% 2% 29.63% 30%
30 mg 6% 2.5% 40.15% 40%
2.5 mg 0.5% 1.0% 21.74% 22%
1.0 mg 0.2% 1.0% 21.54% 21%
of metformin. As shown in Fig. 8, four characteristic peaks appeared at 6627 cm1, 5917 cm1, 5006 cm1, and 4418 cm1. Characteristic peaks can also be observed in the second derivative spectra of herbal medicines adulterated with gliclazide, glibenclamide and glimepiride (See Table 4 for corresponding characteristic peaks). To improve the accuracy of identification, a characteristic peak comparison in the selected ranges was used in tandem with RCCM, as an identification system (Fig. 9). When an herbal medicine is tested, the r between its spectrum and the reference is calculated within the selected ranges. If r is less than the threshold, the product will be identified as an unadulterated drug. If r is greater than the threshold, a comparison is made between its NIR spectrum and the corresponding reference spectrum of the RCCM. Only after the
characteristic peaks are identified in the tested spectrum is the tested herbal medicine considered to be adulterated. If no characteristic peaks appear in the selected ranges, the product is thought to be suspicious and further investigated using its official laboratory method. External validation To validate the performance of the RCCM in tandem with the characteristic peaks comparison method, some independent genuine and counterfeit products were analyzed using the RCCM tandem protocol and traditional official laboratory protocols (Table 5). A total of 111 batches of independent products were used for external validation. The accuracy of NIR identification procedures are 94% (metformin), 80% (glibenclamide), 100% (glimepiride) and 99% (gliclazide). Because so few challenging counterfeit herbal medicines to the RCCMs of glimepiride and gliclazide were available, the accuracy for these two drugs maybe higher than their actual conditions. Two types of error may occur during external validation: false positives (an herbal medicine is labeled as adulterated when it is not) and false negatives (the sample is adulterated but not identified as such). Fourteen batches of genuine herbal anti-diabetic medicines, from eleven different manufacturers, were tested using
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
371
Fig. 6. Plots of correlation coefficients vs. the concentration of the synthetic adulterants in different herbal medicines. (A) Metformin, (B) gliclazide, (C) glimepiride and (D) glibenclamide.
Fig. 7. NIR spectra for the reference substance of metformin (1) and the herbal anti-diabetic medicine containing rosiglitazone, pioglitazone (2).
the four RCCMs constructed in this study. Here, the r between the genuine medicines and the reference spectra were less than the corresponding thresholds. Two of the 111 batches yielded false
positive results for metformin according to its RCCM and the RCCM also gave false positive data for glibenclamide in two batches and glimepiride in seven other batches. False positive results for these
372
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
Fig. 8. Characteristic peaks of herbal medicines adulterated with metformin and those without metformin (1. blank 2. herbal anti-diabetic medicine containing rosiglitazone, pioglitazone 3. laboratory sample adulterated with 5% metformin 4. laboratory sample adulterated with 10% metformin 5. laboratory sample adulterated with 15% metformin 6. laboratory sample adulterated with 30% metformin 7. laboratory sample adulterated with 40% metformin 8. pure metformin).
Table 4 Characteristic peaks of synthetic anti-diabetic drugs. No.
Synthetic anti-diabetic adulterants
Characteristic peaks
1 2 3 4
Metformin Gliclazide Glimepiride Glibenclamide
6627, 5971, 5974, 6009,
5917, 5740, 5716, 5968,
5006, 5800, 4909, 5712,
a
(cm1)
4418 4274 4504 4895
a Second derivative after 17 points smoothing followed by vector normalization were used to preprocess spectra.
11 batches were verified by characteristic peaks comparison and this technique excluded these false positive errors. Eight batches of counterfeit herbal anti-diabetic medicines containing metformin were correctly identified with the NIR system. The r between the herbal medicines containing metformin and the reference spectrum exceeded the threshold, whereas, the r between these products and the reference spectra of gliclazide, glimepiride and glibenclamide were below their corresponding thresholds. For fifty-six batches of counterfeit herbal medicines with glibenclamide, four batches were yielded false negative results. For these four false negative batches, the glibenclamide concentration of two batches (0.72% and 0.55%) was lower than its MDC (1.0%) in Qi Lu Wen Shen Jiao Nang. The glibenclamide concentrations of the other two batches were 1.62% and 1.11% and their corresponding r to the reference spectrum of glibenclamide were 14% and 0%, respectively. To identify the source of the false negative results, three capsules of each batch were randomly selected to be analyzed by official laboratory methods again and the data showed that the content uniformity of the capsules was not optimal. Glibenclamide within the three capsules of one batch was 1.15%, 0.79% and 1.35% respectively, likely contributing to misleading RCCM results. In the twenty-six batches of counterfeit herbal medicines containing both glibenclamide and metformin, in eighteen batches, we did not identify glibenclamide and in seven batches we did not identify metformin (Table 6). In the eighteen batches
Fig. 9. Procedure of RCCM in tandem with characteristic peaks comparison method to detect herbal medicines adulterated with synthetic drugs.
with false negative results for glibenclamide, fifteen batches had glibenclamide that was less than its MDC. For seven batches of false negative results for metformin, the concentrations of glibenclamide and metformin were all higher than their corresponding MDC and the RCCM of glibenclamide gave correct results. Thus, false negative results for metformin may have arisen from interference from glibenclamide. When an herbal anti-diabetic
373
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374 Table 5 Identification of independent herbal anti-diabetic medicines. No.
Legally official laboratory results (adulterants)
Number of batches were evaluated
1 Genuine herbal medicines 2 Metformin 3 Glibenclamide 4 Glimepiride 5 Repaglinide 6 Glibenclamide, Metformin 7 Glibenclamide and Phenformin 8 Glibenclamide and Rosiglitazone 9 Metformin and Rosiglitazone 10 Rosiglitazone and Pioglitazone 11 Gliclazide, Phenformin and Rosiglitazone Total Accuracy
14 8 56 1 1 26 1 1 1 1 1 111
Table 6 Identification of counterfeit herbal anti-diabetic medicines adulterated with glibenclamide and metformin. Sample No.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
Adulterant concentration determined by legally official laboratory method
r determined by RCCM
Glibenclamide (%)
Metformin (%)
Glibenclamide (%)
Metformin (%)
0.29 0.49 0.65 0.59 0.69 0.52 0.90 0.82 0.60 0.95 0.76 0.80 1.33 0.97 1.32 0.56 0.79 1.21 1.08 0.94 1.37 1.68 0.98 1.47 1.02 7.83
48.70 22.76 30.41 16.56 41.13 7.20 5.77 2.63 8.22 12.74 9.70 8.46 10.38 16.13 9.70 4.84 3.90 7.82 14.47 19.89 6.68 6.25 7.73 7.03 14.07 73.99
0a 0 0 0 0 0 0 5 0 0 0 0 4 14 2 7 13 11 30 47 27 21 31 26 33 42
80 77 72 72 68 65 65 64 64 64 63 58 53 44 43 35 35 33 33 24 15 3 2 0 0 0
a The bold figures represent that their values were smaller than their corresponding RCCM thresholds.
medicine is analyzed for adulterants, the four NIR procedures for detecting metformin, gliclazide, glimepiride and glibenclamide are used as a database, so any adulterant will flag the tested sample as adulterated. This, for the herbal anti-diabetic medicines only adulterated with glibenclamide, the false negative rate was about 7%, a data point likely arising from low concentrations of the adulterant. For the herbal anti-diabetic medicines adulterated with glibenclamide and metformin, at least one of adulterants would be detected if the concentration exceeded the MDC. Conclusions A NIR procedure to detect synthetic drugs adulterated in herbal medicines was developed to offer a direct, rapid analysis in the
Number of batches were incorrectly identified by RCCM in tandem with characteristic peaks comparison Metformin method
Glibenclamide method
Glimepiride method
Gliclazide method
0 0 0 0 0 7 0 0 0 0 0 7 94%
0 0 4 0 0 18 0 0 0 0 0 22 80%
0 0 0 0 0 0 0 0 0 0 0 0 100%
0 0 0 0 0 0 0 0 0 0 1 1 99%
field. The analytical process involves identification of unknown samples by RCCM, followed by confirmation of RCCM results using a characteristic peaks comparison. A positive result from both protocols suggests an adulterated product. The RCCM strategy for threshold setting was improved during this research also. A threshold based on the MDC of an adulterant in herbal medicines proved to be more reasonable than one based on its MEC. Finally, the NIR procedure was validated using real herbal medicines, both genuine and counterfeit, obtained from the Chinese market. We observed that NIR detection is a valid alternative to existing methods of synthetic drug identification in herbal medicines, a finding that is relevant in China where LC–MS equipment availability is limited. However, mobile laboratories equipped with NIR spectrometers are portable and accessible, so this NIR procedure can be used in the open market and throughout the herbal medicines distribution channel. Acknowledgements This work was supported by grants from the National Department Public Benefit Research Foundation (Grant No. 2012104008) and the National Natural Science Foundation for Less Developed Regions of China (Grant No. 21365008). References [1] S.A. Jordan, D.G. Cunningham, R.J. Marles, Toxicol. Appl. Pharm. 243 (2010) 198–216. [2] A.A. Savaliya, R.P. Shah, B. Prasad, S. Singh, J. Pharm. Biomed. Anal. 52 (2010) 406–409. [3] D.N. Patel, W.L. Low, L.L. Tan, M.M.B. Tan, Q. Zhang, M.Y. Low, C.L. Chan, H.L. Koh, Clin. Toxicol. 50 (2012) 481–489. [4] S. Balayssac, V. Gilard, C. Zedde, R. Martino, M. Malet-Martino, J. Pharm. Biomed. Anal. 63 (2012) 135–150. [5] T. Snyman, M.J. Stewart, A. Grove, V. Steenkamp, Ther. Drug Monit. 27 (2005) 86–89. [6] G.M. Miller, R. Stripp, Legal Med. 9 (2007) 258–264. [7] L.M. de Carvalho, M. Martini, A.P.L. Moreira, A.P.S. de Lima, D. Correia, T. Falcão, S.C. Garcia, A.V. de Bairros, P.C. do Nascimento, D. Bohrer, Forensic Sci. Int. 204 (2011) 6–12. [8] U. Wollein, W. Eisenreich, N. Schramek, J. Pharm. Biomed. Anal. 56 (2011) 705– 712. [9] J.R. Kesting, J. Huang, D. Sørensen, J. Pharm. Biomed. Anal. 51 (2010) 705–711. [10] N. Klinsunthorn, A. Petsom, T. Nhujak, J. Pharm. Biomed. Anal. 55 (2011) 1175– 1178. [11] A. Doménech-Carbó, M. Martini, L.M. de Carvalho, C. Viana, M.T. DoménechCarbó, M. Silva, J. Pharm. Biomed. Anal. 74 (2013) 194–204. [12] E. Ariburnu, M.F. Uludag, H. Yalcinkaya, E. Yesilada, J. Pharm. Biomed. Anal. 64–65 (2012) 77–81. [13] A.C. Moffat, S. Assi, R.A. Watt, J. Near Infrared Spectrom. 18 (2010) 1–15. [14] Y.C. Feng, C.Q. Hu, J. Pharm. Biomed. Anal. 41 (2006) 373–384. [15] Y.C. Feng, X.B. Zhang, C.Q. Hu, J. Pharm. Biomed. Anal. 51 (2010) 12–17. [16] X.L. Wang, Y.C. Feng, C.Q. Hu, Chin. J. Anal. Chem. 37 (2009) 1825–1828.
374
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
[17] F. Lu, S. Li, J. Le, G. Chen, Y. Cao, Y. Qi, Y. Chai, Y. Wu, Anal. Chim. Acta 589 (2007) 200–207. [18] C.K. Ching, Y.H. Lam, A.Y.W. Chan, T.W.L. Mak, Br. J. Clin. Pharmacol. (2011), http://dx.doi.org/10.1111/1365-2125.2011.04135.x.
[19] D.Q. Lei, Y.C. Feng, C.Q. Hu, Chin. Pharm. J. 45 (2010) 1097–1104. [20] Editorial Committee of Chinese National Formulary, Chinese National Formulary, Chemicals and Biological Products, People’s Military Medical Press, Beijing, 2010.
Contents lists available at ScienceDirect
Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy journal homepage: www.elsevier.com/locate/saa
Rapid identification of illegal synthetic adulterants in herbal anti-diabetic medicines using near infrared spectroscopy Feng Yanchun a, Lei Deqing b, Hu Changqin a,⇑ a b
National Institutes for Food and Drug Control, Beijing 100050, China Shaoyang Institute for Drug Control of Hunan, Shaoyang 422000, China
h i g h l i g h t s
g r a p h i c a l a b s t r a c t
A NIR procedure to detect illicit
A NIR procedure for the rapid, nondestructive identification of herbal medicines containing illicit synthetic drugs was constructed during this study.
adulterants in herbal medicines was constructed. This rapid, nondestructive procedure can reach an accuracy level over 80%. The strategy for RCCM threshold determination has been improved.
a r t i c l e
i n f o
Article history: Received 1 February 2013 Received in revised form 23 January 2014 Accepted 26 January 2014 Available online 8 February 2014 Keywords: Near infrared spectroscopy Reverse correlation coefficient method Synthetic drugs Adulterants Herbal medicines
a b s t r a c t We created a rapid detection procedure for identifying herbal medicines illegally adulterated with synthetic drugs using near infrared spectroscopy. This procedure includes a reverse correlation coefficient method (RCCM) and comparison of characteristic peaks. Moreover, we made improvements to the RCCM based on new strategies for threshold settings. Any tested herbal medicine must meet two criteria to be identified with our procedure as adulterated. First, the correlation coefficient between the tested sample and the reference must be greater than the RCCM threshold. Next, the NIR spectrum of the tested sample must contain the same characteristic peaks as the reference. In this study, four pure synthetic anti-diabetic drugs (i.e., metformin, gliclazide, glibenclamide and glimepiride), 174 batches of laboratory samples and 127 batches of herbal anti-diabetic medicines were used to construct and validate the procedure. The accuracy of this procedure was greater than 80%. Our data suggest that this protocol is a rapid screening tool to identify synthetic drug adulterants in herbal medicines on the market. Ó 2014 Elsevier B.V. All rights reserved.
Introduction
⇑ Corresponding author. Address: National Institutes for Food and Drug Control, Antibiotic Division, #2 Tiantan Xili, Beijing 100050, China. Tel.: +86 10 67095308; fax: +86 10 65115148. E-mail addresses: [email protected], [email protected] (C. Hu). http://dx.doi.org/10.1016/j.saa.2014.01.117 1386-1425/Ó 2014 Elsevier B.V. All rights reserved.
Herbal medicines are popular worldwide for their natural origins and healing properties. In contrast to conventional synthetic drugs, herbal medicines are generally perceived as safe, harmless and without deleterious side effects. However, nowadays,
364
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
herbal medicines have not remained trustworthy as lots of evidences are coming into literature about adulteration of these products with synthetic drugs, in order to enhance the claims stated on the label [1]. Over the past several years, there have been various reports from Asia [2,3], Europe [4], Africa [5], USA [6] and Brazil [7], regarding this unethical trend. Such counterfeit drugs or adulterated products are commonly analyzed using high-performance liquid chromatography (HPLC) [8], mass spectrometry (MS) [9] or nuclear magnetic resonance (NMR) [10]. Recently voltammetry of microparticles [11] and high-performance thin layer chromatography (HPTLC) densitometry [12] have been used to detect these illicit drugs. Each of these techniques requires substantial time and complicated sample preprocessing, preventing opportunities for identifying and seizing such products. Near infrared (NIR) is a powerful tool for counterfeit drug identification because it is a quick, non-destructive, and reagent-free technique [13]. Since 2002, the National Institutes for Food and Drug Control (NIFDC) has conducted a large scale study on NIR rapid drug screening and more than 300 mobile laboratory vehicles in China have equipped with NIR prescreening systems. These vehicles are used to quickly evaluate drug quality on site, in open market as well as in the distribution channel. The NIR prescreening system, utilizing a universal model, can successfully detect illegal common synthetic drugs [14,15]. Recently, we have used NIR to identify herbal medicines adulterated with synthetic drugs. Because herbal medicines are complex, NIR is more problematic for identifying adulterants. We did observe that in the NIR region of 6200–5500 cm1, the herbal medicine adulterated with sildenafil citrate had considerable similarity to the sildenafil citrate reference substance. Thus the NIR reverse correlation coefficient method (RCCM) was developed and it can detect sildenafil citrate illegally adulterated in herbal aphrodisiacs medicines accurately [16]. Moreover, Lu and co-workers [17] demonstrated infrared absorption differences between synthetic drugs which had sharp, abrupt peaks and herbal medicines that had broad, smooth peaks. Therefore, we hypothesized that NIR profiles of herbal medicines adulterated with a synthetic drug would have characteristic peaks of the corresponding synthetic adulterant. Upon validation of this hypothesis, construction of a robust NIR procedure, capable of detecting the most common synthetic adulterants in herbal medicines, is a worthwhile endeavor. Herbal anti-diabetic medicines containing undeclared pharmaceuticals are a significant problem because patients taking these illicit products may experience potentially fatal adverse effects. Such illicit herbal medicines usually contain undeclared, registered and/ or banned, oral anti-diabetic agents, including metformin, gliclazide, glibenclamide and glimepiride etc. [18]. Here herbal antidiabetic medicines dosed with synthetic adulterants were tested to investigate the usefulness of RCCM in this application. Based on these findings, a general detection procedure for common synthetic adulterants found in herbal medicines was developed using the RCCM in tandem with NIR characteristic peaks comparisons.
Materials and methods
medicines referred to herbal medicines adulterated with synthetic drugs. Reference substances of metformin, gliclazide, glibenclamide and glimepiride were obtained from the NIFDC. To determine if synthetic anti-diabetic drugs had characteristic regions when adulterated into herbal medicines, five kinds of antidiabetic herbal medicines, named Qi Lu Wen Shen Jiao Nang, Gan Lu Xiao Ke Jiao Nang, Pu Nuo Ning Si Jiao Nang, Zhong Yan Wan Tong Jiao Nang and Sang Ge Jiao Nang, were selected as blanks. These products were removed from capsules and milled into small granules before using as blanks. Laboratory samples were made by mixing given amounts of pure synthetic drugs into blank samples. See Table 1 for the composition of the laboratory samples. Instrumentation and data acquisition FT-NIR spectrometers, MATRIX-F, from Bruker Optik GmbH (Ettlingen, Germany) were used for all experiments. All spectrometers are equipped with a 1.5 m fiber-optic diffuse reflectance probe and an extended TE-cooled indium gallium arsenide (InGaAs) detector. Data were collected and processed using OPUS software (version 6.5). Diffuse reflectance spectra were recorded using a 1.5 m fiber optic probe at 8 cm1 resolution with 32 co-added scans over the spectral range of 4000–12,000 cm1. For capsule samples, contents were transferred into glass bottles and a fiber-optic probe was inserted to record spectra. For tablet samples, spectra were recorded via a randomly selected tablet surface scan. Spectral pretreatment Data pretreatment was performed to reduce the influence of physical parameters. First, the average spectrum was calculated using six original spectra from each sample. Then a second derivation was performed on the averaged spectra with 17-point smoothing followed by vector normalization, to enhance spectral information and reduce baseline drift. Only the preprocessed, averaged spectra were used in the construction and validation of these methods. Reverse correlation coefficient method RCCM is a new detection method for illicit synthetic adulterants in herbal medicines using NIR spectroscopy [16]. RCCM and the traditional correlation coefficient method use the same formula to calculate the correlation coefficient (r) (Eq. 1). However, significant differences exist between the two methods: RCCM uses a pure adulterant spectrum as the reference spectrum, and can detect herbal medicines from different manufacturers simultaneously, whereas the traditional correlation coefficient method requires a spectrum of a genuine sample from one manufacturer as the reference spectrum (Fig. 1). Because different manufacturers produce various products, one traditional correlation coefficient method can be only used for detecting the corresponding manufacturer’s product.
Cov ðy1 ðkÞ; y2 ðkÞÞ ry1 ry2
Sample preparation
r¼
Both genuine herbal medicines and counterfeit ones adulterated with synthetic drugs were obtained from the Chinese market by the NIFDC, the Beijing Institute for Drug Control, and the Yulin Institute for Drug Control in Guangxi Province. Most of the herbal medicines used here were capsules and the others were tablets. All samples were identified according to official laboratory methods. Here genuine herbal medicines were the samples passed the examination of its official laboratory method and counterfeit herbal
y(k): all Y data points, when a spectrum is presented as a data point table (X, Y), Cov(y1 (k), y2 (k)): the covariance of y1(k) and y2(k) and ry: the standard deviation of y(k). The key parameter of RCCM is the threshold. By comparing the r and a preset corresponding threshold, we could identify whether the tested herbal medicine is adulterated with synthetic drugs. The minimum effective concentration (MEC) of an adulterant was used to set the RCCM threshold (Eq. 2) as we previously reported
ð1Þ
365
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374 Table 1 Adulteration of herbal laboratory samples. Sample No.
1 2 3 4 5 6 7 8 9 10 11 12 13
Metformin
Gliclazide
Glibenclamide
Blank (mg)
Metformin (mg)
Blank (mg)
Gliclazide (mg)
Blank (mg)
Glibenclamide (mg)
Blank (mg)
Glimepiride (mg)
400 400 400 400 400 400 400 400 400 400 400 400 400
200 160 120 60 40 20 12 8 4 2 1.2 0.6 0.4
400 400 400 400 400 400 400 400 – – – – –
48 40 32 20 10 8 6 4
400 400 400 400 400 400 400 – – – – – –
40 20 10 5 4 2 1.2
400 400 400 400 400 400 400 400 400 – – – –
20 16 10 8 4 2 1.2 0.8 0.4
Fig. 1. Comparison between traditional correlation coefficient method and the reverse correlation coefficient method ⁄ ‘‘Counterfeit drugs’’ refer to herbal medicines adulterated with synthetic drugs.
[16]. The maximum weight of a single tablet or capsule in Eq. (2) is set at 500 mg. A laboratory sample with a content equal to the MEC was prepared, and the r between the reference sample and the corresponding laboratory sample was set as the RCCM threshold.
MEC ¼
Glimepiride
The minimum specification of the pure adulterant The maximum weight of a single tablet or capsule
characteristic spectral regions of metformin, gliclazide, glimepiride and glibenclamide as herbal adulterants, spectra of laboratory samples, counterfeit anti-diabetic herbal medicines with adulterants as well as pure synthetic drug were investigated. Fig. 2 depicts original spectra for reference substances of metformin, gliclazide, glimepiride and glibenclamide. Data show that these compounds can be distinguished from one another. Next, we compared the spectra of laboratory samples and their corresponding pure adulterants. Spectra of metformin and the laboratory samples dosed with various concentrations of metformin are shown in Fig. 3. These spectra closely matched in regions 6701–6515 cm1, 6030–5570 cm1, 5070–4700 cm1, and 4480–4220 cm1. Moreover, their spectral intensities decreased as metformin concentration declined. Multiple counterfeit anti-diabetic herbal medicines adulterated with metformin were compared to determine if the characteristic regions selected using laboratory samples could be used for real counterfeit samples from different manufacturers which had different excipients and formulations. Typical spectra of the counterfeit medicines are depicted in Fig. 4. These counterfeit herbal medicines were similar to pure metformin, with similar spectral peaks in our selected regions. There were, however, significant differences among their spectral whole shape. This technique was then used to search characteristic regions of herbal medicines adulterated with gliclazide, glimepiride and glibenclamide. Data presented in Table 2 indicate that all four synthetic drugs have unique NIR peaks, even when incorporated into herbal medicines. In addition, a common characteristic spectral region (6200–5500 cm1) appeared in all spectra of herbal medicines adulterated with these four synthetic drugs. Therefore, the 6200– 5500 cm1 region may be an important region for detecting other synthetic drugs in herbal medicines. Next, NIR spectra of the reference substances of metformin, gliclazide, glimepiride and glibenclamide were used as reference spectra to construct RCCMs, respectively, using their corresponding characteristic regions shown in Table 2.
ð2Þ
Results and discussion Characteristic region selection Herbal medicines adulterated with sildenafil citrate are similar to pure sildenafil citrate in the spectral region of 6200–5500 cm1 [16]. Previous studies suggest that most APIs (active pharmaceutical ingredients) also have obvious characteristic peaks within the NIR regions of 6200–5500 cm1 and 4100–4900 cm1 [19]. To find
Setting threshold for RCCM A critical step in RCCM model construction is determination of the threshold. If the r between the tested spectrum and the reference spectrum is greater than the threshold, the tested herbal medicine will be identified as containing the examined synthetic adulterant. Our previous study used the MEC of illicit adulterants to set the threshold for the RCCM. However, when setting the RCCM thresholds for metformin, gliclazide, glimepiride and glibenclamide, we found that the MECs of the compounds were not always suitable for this purpose. Based on the following study,
366
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
Fig. 2. NIR spectra of pure synthetic drugs. (1. metformin 2. gliclazide 3. glimepiride 4. glibenclamide).
the minimum detectable concentration (MDC) of the adulterants in herbal medicines was best for setting RCCM thresholds. In this study, any MDC of the adulterants in herbal medicines must meet two criteria: First, when the adulterant concentration in an herbal medicine is higher than its MDC, the r between the herbal medicine and the pure adulterant increases with the adulterant concentration increasing; Second, when the adulterant concentration in an herbal medicine is lower than its MDC, the r between the herbal medicine and the pure adulterant has no obvious correlation with the adulterant concentration, and/or the RCCM method cannot distinguish the herbal medicine with the adulterant from genuine herbal medicines. To set the RCCM threshold for metformin detection in herbal medicines, the r between the reference spectrum (namely the spectrum of the pure metformin) and laboratory samples (Qi Lu Wen Shen Jiao Nang was as a blank) was calculated in the selected regions. The r between the metformin reference spectrum and 12 batches of genuine anti-diabetic herbal products was also calculated. All r values are presented in Fig. 5A and B. Data indicate that at lower concentrations of metformin, the correlation coefficient decreased. When the concentration of metformin decreased below 2%, the r between the metformin reference and laboratory samples no longer had correlation with their corresponding adulterant concentrations; besides, the r between the reference spectrum and laboratory samples was similar to the r between the reference spectrum and the genuine herbal anti-diabetic products. Thus, when metformin in Qi Lu Wen Shen Jiao Nang was less than 2%, it became difficult to correctly identify whether the herbal medicine was adulterated with metformin by RCCM. Therefore, the MDC of metformin in Qi Lu Wen Shen Jiao Nang was established at 2%, and the corresponding r for this metformin concentration was 29.63% (Fig. 5B). The r between the other three pure synthetic drugs (gliclazide, glimepiride, and glibenclamide), their corresponding laboratory samples (Qi Lu Wen Shen Jiao Nang was as a blank), and the genuine products are shown in Fig. 5C–E. For gliclazide (Fig. 5C), when the concentration was below 2.5%, the r between pure gliclazide and laboratory samples was zero, a value smaller than the values
between pure gliclazide and the genuine products. Thus, the MDC for gliclazide in Qi Lu Wen Shen Jiao Nang was 2.5%. Likewise, the MDC of both glimepiride and glibenclamide in Qi Lu Wen Shen Jiao Nang was 1.0% (Fig. 5D and E). According to the Chinese National Formulary [20], the minimum specifications (MS) of metformin, gliclazide, glimepiride and glibenclamide are given in Table 3. The MEC for each compound was calculated from Eq. (2) (Table 3). When comparing these data, MDCs of glimepiride and glibenclamide were higher than their MECs. Under this condition, the r calculated from the pure adulterant and its laboratory sample at MEC is incredible, which may be confused with the r between the pure adulterant and genuine herbal medicines. If the RCCM threshold for glimepiride or glibenclamide is set based on its MEC, the RCCM may not correctly identify part of genuine herbal medicines. So MDC values must be used to set the RCCM threshold for glimepiride and glibenclamide. In contrast, MDC values for metformin and gliclazide were lower than their MECs. So, if their MECs are used to set RCCM thresholds, herbal adulterants less than the MEC but higher than the MDC would produce a false negative result. Collectively, using MDC to set RCCM thresholds is more reliable than using the MEC. First, for the adulterants with higher MS, their MEC in herbal products exceeds their MDC. In this case, if the MDC is used to set the threshold, the RCCM detectable range is significantly increased. Conversely, for the synthetic adulterants with lower MS, the calculated MEC may be smaller than their MDC. Then, the RCCM threshold calculated from MEC is incredible because the RCCM could not correctly identify whether an herbal medicine is adulterated with a synthetic drug when the concentration of the synthetic drug in herbal medicines is lower than its MDC. Moreover, an MEC is an empirical value, equal to the MS of the pure adulterant divided by the maximum weight of a single tablet or a single capsule (i.e. 500 mg). The maximum weight is a statistical value, and the real weight of a single tablet or capsule may be higher or lower than 500 mg, which increases the uncertainty in the accuracy of thresholds calculated using MEC values. In contrast, not only is the MDC an exact value derived from experimentation, but it is also the minimum concentration which can be
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
367
Fig. 3. NIR spectra of blank, pure metformin, and laboratory samples containing metformin. (A) Qi Lu Wen Shen Jiao Nang was selected as blank. (B) Gan Lu Xiao Ke Jiao Nang was selected as blank. (C) Pu Nuo Ning Si Jiao Nang was selected as blank. (D) Zhong Yan Wan Tong Jiao Nang was selected as blank. (E) Sang Ge Jiao Nang was selected as blank. Numbers 1–7 represent different metformin concentrations in the blank.
identified by its specific RCCM within specific ranges. Therefore, in this context, the MDC values were used to set RCCM thresholds. The initial RCCM thresholds for the four adulterants calculated from their MDCs are shown in Table 3. To validate the initial thresholds, set using Qi Lu Wen Shen Jiao Nang as a blank, is universal to other blanks, the r between the laboratory samples made using other four kinds of genuine herbal medicines as blanks (Gan Lu Xiao Ke Jiao Nang, Pu Nuo Ning Si Jiao Nang, Zhong Yan Wan Tong Jiao Nang and Sang Ge Jiao Nang) and their corresponding reference were calculated, and then were
plotted with adulterant concentrations (Fig. 6). Most of the r between the laboratory samples and their corresponding pure adulterants increased with the increase of their adulterant concentrations in the whole concentration range. Therefore, the major determinant of MDC was the r between the reference spectrum and the genuine herbal anti-diabetic products. Fig. 6A indicated as long as the RCCM threshold was set to about 30%, the metformin RCCM can detect metformin adulterated in herbal medicines correctly. So the MDC of metformin was the concentration corresponding to 30% in Fig. 6A. It was found that although the MDCs
368
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
Fig. 3 (continued)
of metformin in different blanks were different, the r values corresponding to the MDCs in different blanks were identical. In other words, the thresholds of the metformin RCCM set by these four kinds of blanks respectively were consistent with that set using Qi Lu Wen Shen Jiao Nang. Likewise, the RCCM thresholds of glimepiride and glibenclamide set using Qi Lu Wen Shen Jiao Nang as a blank, were similar to that set using other four kinds of blanks (Fig. 6C and D). Only for gliclazide RCCM, because there were no enough laboratory samples made by Qi Lu Wen Shen Jiao Nang, which was result in high threshold. Based on the data presented in Fig. 6B, the threshold for gliclazide was adjusted to 4%. In addition, it was also found that RCCM thresholds related to the representativeness of the genuine herbal products we selected to set
MDC. Therefore with the spectra of genuine herbal products we can obtain increasing during RCCM use, RCCM thresholds will be adjusted to more reasonable values. Characteristic peak comparisons used in tandem with RCCM One problem showed up when the RCCM model construction was finished. Using correlation coefficient to evaluate similarities between the two spectra may be limited: the spectral correlation coefficient between the reference and the herbal medicines, which did not contain the examined synthetic drug, might exceed the threshold. For example, one batch of anti-diabetic herbal capsules adulterated with rosiglitazone and pioglitazone, the RCCM
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
369
Fig. 3 (continued)
Fig. 4. NIR spectra of counterfeit herbal medicines adulterated with metformin. (1. Yi Dao Zai Sheng Jiang Tang Jiao Nang 2. Hong Qi Yi Hao 3. Huo Yi Xue Tang Ping Jiao Nang 4. Sheng Jin Zhi Ke Jiao Nang 5. Jiang Tang Ning Jiao Nang).
Table 2 Characteristic regions of synthetic anti-diabetic drugs. No.
Synthetic anti-diabetic adulterants
Characteristic regions (cm1)
1
Metformin
2 3 4
Gliclazide Glimepiride Glibenclamide
6701–6515, 6030–5570, 5070–4700, 4480–4220 6100–5600, 4333–4220, 6000–5500, 5000–4490 6200–5500, 5000–4500
incorrectly identified the adulterant as metformin. Its correlation coefficient with pure metformin was 68.76%; while the threshold of metformin RCCM was 30%. However, no characteristic metformin peaks were found in the spectrum of the tested herbal capsule (Fig. 7). Comparing the spectra of the laboratory samples adulterated with different concentrations of metformin and the misidentified sample revealed that the characteristic peaks of the metformin adulterated samples were obvious when their second derivative spectra were observed, even with low concentrations
370
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
Fig. 5. Plots of correlation coefficients vs. the concentration of the synthetic adulterants in herbal medicines. (A) Metformin ranged from 5% to 40%. (B) Metformin ranged from 0% to 5%. (C) gliclazide, (D) glimepiride and (E) glibenclamide.
Table 3 Parameters for RCCM threshold settings.
MS MEC MDC r at MDC Initial threshold
Metformin
Gliclazide
Glimepiride
Glibenclamide
250 mg 50% 2% 29.63% 30%
30 mg 6% 2.5% 40.15% 40%
2.5 mg 0.5% 1.0% 21.74% 22%
1.0 mg 0.2% 1.0% 21.54% 21%
of metformin. As shown in Fig. 8, four characteristic peaks appeared at 6627 cm1, 5917 cm1, 5006 cm1, and 4418 cm1. Characteristic peaks can also be observed in the second derivative spectra of herbal medicines adulterated with gliclazide, glibenclamide and glimepiride (See Table 4 for corresponding characteristic peaks). To improve the accuracy of identification, a characteristic peak comparison in the selected ranges was used in tandem with RCCM, as an identification system (Fig. 9). When an herbal medicine is tested, the r between its spectrum and the reference is calculated within the selected ranges. If r is less than the threshold, the product will be identified as an unadulterated drug. If r is greater than the threshold, a comparison is made between its NIR spectrum and the corresponding reference spectrum of the RCCM. Only after the
characteristic peaks are identified in the tested spectrum is the tested herbal medicine considered to be adulterated. If no characteristic peaks appear in the selected ranges, the product is thought to be suspicious and further investigated using its official laboratory method. External validation To validate the performance of the RCCM in tandem with the characteristic peaks comparison method, some independent genuine and counterfeit products were analyzed using the RCCM tandem protocol and traditional official laboratory protocols (Table 5). A total of 111 batches of independent products were used for external validation. The accuracy of NIR identification procedures are 94% (metformin), 80% (glibenclamide), 100% (glimepiride) and 99% (gliclazide). Because so few challenging counterfeit herbal medicines to the RCCMs of glimepiride and gliclazide were available, the accuracy for these two drugs maybe higher than their actual conditions. Two types of error may occur during external validation: false positives (an herbal medicine is labeled as adulterated when it is not) and false negatives (the sample is adulterated but not identified as such). Fourteen batches of genuine herbal anti-diabetic medicines, from eleven different manufacturers, were tested using
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
371
Fig. 6. Plots of correlation coefficients vs. the concentration of the synthetic adulterants in different herbal medicines. (A) Metformin, (B) gliclazide, (C) glimepiride and (D) glibenclamide.
Fig. 7. NIR spectra for the reference substance of metformin (1) and the herbal anti-diabetic medicine containing rosiglitazone, pioglitazone (2).
the four RCCMs constructed in this study. Here, the r between the genuine medicines and the reference spectra were less than the corresponding thresholds. Two of the 111 batches yielded false
positive results for metformin according to its RCCM and the RCCM also gave false positive data for glibenclamide in two batches and glimepiride in seven other batches. False positive results for these
372
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
Fig. 8. Characteristic peaks of herbal medicines adulterated with metformin and those without metformin (1. blank 2. herbal anti-diabetic medicine containing rosiglitazone, pioglitazone 3. laboratory sample adulterated with 5% metformin 4. laboratory sample adulterated with 10% metformin 5. laboratory sample adulterated with 15% metformin 6. laboratory sample adulterated with 30% metformin 7. laboratory sample adulterated with 40% metformin 8. pure metformin).
Table 4 Characteristic peaks of synthetic anti-diabetic drugs. No.
Synthetic anti-diabetic adulterants
Characteristic peaks
1 2 3 4
Metformin Gliclazide Glimepiride Glibenclamide
6627, 5971, 5974, 6009,
5917, 5740, 5716, 5968,
5006, 5800, 4909, 5712,
a
(cm1)
4418 4274 4504 4895
a Second derivative after 17 points smoothing followed by vector normalization were used to preprocess spectra.
11 batches were verified by characteristic peaks comparison and this technique excluded these false positive errors. Eight batches of counterfeit herbal anti-diabetic medicines containing metformin were correctly identified with the NIR system. The r between the herbal medicines containing metformin and the reference spectrum exceeded the threshold, whereas, the r between these products and the reference spectra of gliclazide, glimepiride and glibenclamide were below their corresponding thresholds. For fifty-six batches of counterfeit herbal medicines with glibenclamide, four batches were yielded false negative results. For these four false negative batches, the glibenclamide concentration of two batches (0.72% and 0.55%) was lower than its MDC (1.0%) in Qi Lu Wen Shen Jiao Nang. The glibenclamide concentrations of the other two batches were 1.62% and 1.11% and their corresponding r to the reference spectrum of glibenclamide were 14% and 0%, respectively. To identify the source of the false negative results, three capsules of each batch were randomly selected to be analyzed by official laboratory methods again and the data showed that the content uniformity of the capsules was not optimal. Glibenclamide within the three capsules of one batch was 1.15%, 0.79% and 1.35% respectively, likely contributing to misleading RCCM results. In the twenty-six batches of counterfeit herbal medicines containing both glibenclamide and metformin, in eighteen batches, we did not identify glibenclamide and in seven batches we did not identify metformin (Table 6). In the eighteen batches
Fig. 9. Procedure of RCCM in tandem with characteristic peaks comparison method to detect herbal medicines adulterated with synthetic drugs.
with false negative results for glibenclamide, fifteen batches had glibenclamide that was less than its MDC. For seven batches of false negative results for metformin, the concentrations of glibenclamide and metformin were all higher than their corresponding MDC and the RCCM of glibenclamide gave correct results. Thus, false negative results for metformin may have arisen from interference from glibenclamide. When an herbal anti-diabetic
373
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374 Table 5 Identification of independent herbal anti-diabetic medicines. No.
Legally official laboratory results (adulterants)
Number of batches were evaluated
1 Genuine herbal medicines 2 Metformin 3 Glibenclamide 4 Glimepiride 5 Repaglinide 6 Glibenclamide, Metformin 7 Glibenclamide and Phenformin 8 Glibenclamide and Rosiglitazone 9 Metformin and Rosiglitazone 10 Rosiglitazone and Pioglitazone 11 Gliclazide, Phenformin and Rosiglitazone Total Accuracy
14 8 56 1 1 26 1 1 1 1 1 111
Table 6 Identification of counterfeit herbal anti-diabetic medicines adulterated with glibenclamide and metformin. Sample No.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
Adulterant concentration determined by legally official laboratory method
r determined by RCCM
Glibenclamide (%)
Metformin (%)
Glibenclamide (%)
Metformin (%)
0.29 0.49 0.65 0.59 0.69 0.52 0.90 0.82 0.60 0.95 0.76 0.80 1.33 0.97 1.32 0.56 0.79 1.21 1.08 0.94 1.37 1.68 0.98 1.47 1.02 7.83
48.70 22.76 30.41 16.56 41.13 7.20 5.77 2.63 8.22 12.74 9.70 8.46 10.38 16.13 9.70 4.84 3.90 7.82 14.47 19.89 6.68 6.25 7.73 7.03 14.07 73.99
0a 0 0 0 0 0 0 5 0 0 0 0 4 14 2 7 13 11 30 47 27 21 31 26 33 42
80 77 72 72 68 65 65 64 64 64 63 58 53 44 43 35 35 33 33 24 15 3 2 0 0 0
a The bold figures represent that their values were smaller than their corresponding RCCM thresholds.
medicine is analyzed for adulterants, the four NIR procedures for detecting metformin, gliclazide, glimepiride and glibenclamide are used as a database, so any adulterant will flag the tested sample as adulterated. This, for the herbal anti-diabetic medicines only adulterated with glibenclamide, the false negative rate was about 7%, a data point likely arising from low concentrations of the adulterant. For the herbal anti-diabetic medicines adulterated with glibenclamide and metformin, at least one of adulterants would be detected if the concentration exceeded the MDC. Conclusions A NIR procedure to detect synthetic drugs adulterated in herbal medicines was developed to offer a direct, rapid analysis in the
Number of batches were incorrectly identified by RCCM in tandem with characteristic peaks comparison Metformin method
Glibenclamide method
Glimepiride method
Gliclazide method
0 0 0 0 0 7 0 0 0 0 0 7 94%
0 0 4 0 0 18 0 0 0 0 0 22 80%
0 0 0 0 0 0 0 0 0 0 0 0 100%
0 0 0 0 0 0 0 0 0 0 1 1 99%
field. The analytical process involves identification of unknown samples by RCCM, followed by confirmation of RCCM results using a characteristic peaks comparison. A positive result from both protocols suggests an adulterated product. The RCCM strategy for threshold setting was improved during this research also. A threshold based on the MDC of an adulterant in herbal medicines proved to be more reasonable than one based on its MEC. Finally, the NIR procedure was validated using real herbal medicines, both genuine and counterfeit, obtained from the Chinese market. We observed that NIR detection is a valid alternative to existing methods of synthetic drug identification in herbal medicines, a finding that is relevant in China where LC–MS equipment availability is limited. However, mobile laboratories equipped with NIR spectrometers are portable and accessible, so this NIR procedure can be used in the open market and throughout the herbal medicines distribution channel. Acknowledgements This work was supported by grants from the National Department Public Benefit Research Foundation (Grant No. 2012104008) and the National Natural Science Foundation for Less Developed Regions of China (Grant No. 21365008). References [1] S.A. Jordan, D.G. Cunningham, R.J. Marles, Toxicol. Appl. Pharm. 243 (2010) 198–216. [2] A.A. Savaliya, R.P. Shah, B. Prasad, S. Singh, J. Pharm. Biomed. Anal. 52 (2010) 406–409. [3] D.N. Patel, W.L. Low, L.L. Tan, M.M.B. Tan, Q. Zhang, M.Y. Low, C.L. Chan, H.L. Koh, Clin. Toxicol. 50 (2012) 481–489. [4] S. Balayssac, V. Gilard, C. Zedde, R. Martino, M. Malet-Martino, J. Pharm. Biomed. Anal. 63 (2012) 135–150. [5] T. Snyman, M.J. Stewart, A. Grove, V. Steenkamp, Ther. Drug Monit. 27 (2005) 86–89. [6] G.M. Miller, R. Stripp, Legal Med. 9 (2007) 258–264. [7] L.M. de Carvalho, M. Martini, A.P.L. Moreira, A.P.S. de Lima, D. Correia, T. Falcão, S.C. Garcia, A.V. de Bairros, P.C. do Nascimento, D. Bohrer, Forensic Sci. Int. 204 (2011) 6–12. [8] U. Wollein, W. Eisenreich, N. Schramek, J. Pharm. Biomed. Anal. 56 (2011) 705– 712. [9] J.R. Kesting, J. Huang, D. Sørensen, J. Pharm. Biomed. Anal. 51 (2010) 705–711. [10] N. Klinsunthorn, A. Petsom, T. Nhujak, J. Pharm. Biomed. Anal. 55 (2011) 1175– 1178. [11] A. Doménech-Carbó, M. Martini, L.M. de Carvalho, C. Viana, M.T. DoménechCarbó, M. Silva, J. Pharm. Biomed. Anal. 74 (2013) 194–204. [12] E. Ariburnu, M.F. Uludag, H. Yalcinkaya, E. Yesilada, J. Pharm. Biomed. Anal. 64–65 (2012) 77–81. [13] A.C. Moffat, S. Assi, R.A. Watt, J. Near Infrared Spectrom. 18 (2010) 1–15. [14] Y.C. Feng, C.Q. Hu, J. Pharm. Biomed. Anal. 41 (2006) 373–384. [15] Y.C. Feng, X.B. Zhang, C.Q. Hu, J. Pharm. Biomed. Anal. 51 (2010) 12–17. [16] X.L. Wang, Y.C. Feng, C.Q. Hu, Chin. J. Anal. Chem. 37 (2009) 1825–1828.
374
Y. Feng et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 125 (2014) 363–374
[17] F. Lu, S. Li, J. Le, G. Chen, Y. Cao, Y. Qi, Y. Chai, Y. Wu, Anal. Chim. Acta 589 (2007) 200–207. [18] C.K. Ching, Y.H. Lam, A.Y.W. Chan, T.W.L. Mak, Br. J. Clin. Pharmacol. (2011), http://dx.doi.org/10.1111/1365-2125.2011.04135.x.
[19] D.Q. Lei, Y.C. Feng, C.Q. Hu, Chin. Pharm. J. 45 (2010) 1097–1104. [20] Editorial Committee of Chinese National Formulary, Chinese National Formulary, Chemicals and Biological Products, People’s Military Medical Press, Beijing, 2010.
