Chemical Barcoding: A Nuclear-Magnetic-Resonance-Based

Jun 26, 2019 - Read OnlinePDF (3 MB) ... experiment, reproducibility studies, and NMR spectral files (PDF). pdf. jf9b01066_si_001.pdf (9.59 MB). back...
0 downloads 0 Views 3MB Size
Article Cite This: J. Agric. Food Chem. 2019, 67, 7765−7774

pubs.acs.org/JAFC

Chemical Barcoding: A Nuclear-Magnetic-Resonance-Based Approach To Ensure the Quality and Safety of Natural Ingredients Camilo F. Martinez-Farina,† Stephen Driscoll,‡,§ Chelsi Wicks,‡,§ Ian Burton,† Peter D. Wentzell,§ and Fabrice Berrue*́ ,† †

Downloaded via BUFFALO STATE on July 29, 2019 at 18:01:20 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

Aquatic and Crop Resource Development, National Research Council of Canada, 1411 Oxford Street, Halifax, Nova Scotia B3H 3Z1 Canada § Trace Analysis Research Centre, Department of Chemistry, Dalhousie University, Post Office Box 15000, Halifax, Nova Scotia B3H 4R2 Canada S Supporting Information *

ABSTRACT: One of the greatest challenges facing the functional food and natural health product (NHP) industries is sourcing high-quality, functional, natural ingredients for their finished products. Unfortunately, the lack of ingredient standards, modernized analytical methodologies, and industry oversight creates the potential for low quality and, in some cases, deliberate adulteration of ingredients. By exploring a diverse library of NHPs provided by the independent certification organization ISURA, we demonstrated that nuclear magnetic resonance (NMR) spectroscopy provides an innovative solution to authenticate botanicals and warrant the quality and safety of processed foods and manufactured functional ingredients. Two-dimensional NMR experiments were shown to be a robust and reproducible approach to capture the content of complex chemical mixtures, while a binary normalization step allows for emphasizing the chemical diversity in each sample, and unsupervised statistical methodologies provide key advantages to classify, authenticate, and highlight the potential presence of additives and adulterants. KEYWORDS: nuclear magnetic resonance, chemical barcoding, natural health products, food quality, multivariate statistical analysis



INTRODUCTION Much of the worldwide population still relies on traditional medicine as a primary source of health care, while the use of natural remedies has increased in developed countries over the past 3 decades.1−5 In the context of a global economy, the use of natural products and functional ingredients has been steadily increasing, with application in cosmetic, nutraceutical, animal care, and human health and wellness.6−8 With herbal remedies becoming readily available in drug stores, food stores, and supermarkets, policy-makers, health professionals, and consumers have raised important concerns regarding the safety, quality, and effectiveness of natural health products (NHPs) and their natural ingredients in the marketplace.9−13 In Canada, NHPs are considered a class of health products that include vitamins, mineral supplements, herbal preparations, traditional and homeopathic medicines, probiotics, and enzymes.14,15 Similarly, the United States of America has defined these products as dietary supplements under the Dietary Supplement Health and Education Act of 1994.16,17 Both NHPs and dietary supplements are therefore neither food products nor drugs and must obey specific good manufacturing practices and regulatory framework to ensure their quality and safety to the consumer. This regulatory framework was called into question in 2015 when the New York State Attorney General conducted an investigation into NHPs sampled among major retail stores in the U.S. for the presence of fraudulent and adulterated products.18,19 The investigation was based on a newly developed technique based on DNA barcoding,20−25 and although later disputed, four cease-and-desist letters were Published 2019 by the American Chemical Society

published and subsequent extensive media coverage had a severe negative impact on both consumer confidence and the reputation of the industry. It is indisputable that the NHP industry faces great challenges in sourcing high-quality plants and functional ingredients to ensure the quality and safety of their finished products in the context of a rapidly growing economic sector and global trade.1,10,26−29 It is therefore critical to develop new scientific technology as well as promote the application of adequate and rigorous methodologies suitable to the evaluation of the identity, stability, reproducibility, and bioavailability of NHPs.7,30−32 Although DNA barcoding is effective in the authentication of raw natural ingredients (in their original form), its effectiveness quickly diminishes when trying to authenticate processed ingredients, extracts, or blends.33−36 Indeed, DNA barcoding is a targeted approach looking at genetic biomarkers and cannot distinguish between different plant parts (root, flower, leaf, and entire aerial part) because they share the same genetic information. Moreover, gene-based methods do not provide any information on the presence of exogenous chemicals, including excipients, fillers, or pharmaceuticals. The ability to distinguish between different plant parts is important because their chemical composition varies greatly, as highlighted in the nuclear magnetic resonance (NMR) study of Echinacea purpurea (Figure S4 of the Received: Revised: Accepted: Published: 7765

February 18, 2019 May 26, 2019 June 3, 2019 June 26, 2019 DOI: 10.1021/acs.jafc.9b01066 J. Agric. Food Chem. 2019, 67, 7765−7774

Journal of Agricultural and Food Chemistry

Article

based methodology will be discussed as well as the benefit of binary normalization to enhance chemical diversity and create a unique chemical barcode for each natural ingredient that is easy to interpret.

Supporting Information). To capture the differences in chemical composition and differentiate these samples, a chemical-based method is therefore recommended. In that context, liquid chromatography−mass spectrometry (LC−MS) has been an analytical method of choice for many years as being efficient and sensitive in separating and detecting metabolites, respectively.37−40 LC−MS is the most widely used analytical technique to qualitatively and quantitatively analyze targeted compounds and adulterants in complex mixtures.41−44 However, it often requires sample preparation and extensive method development, while carry-over, injection cross-contamination, and mass spectrometer interferences are some of the many challenges inherent in the analysis of the data.45 Consequently, the development of robust and reproducible unsupervised methods remains challenging for metabolomics studies, including the alignment of MS data in both the time and m/z domains.46 The development of analytical methods suitable for the analysis of NHPs with respect to quality and safety is made even more difficult because natural ingredients often constitute hundreds of natural products, whose identities and abundances may greatly vary depending upon the plant part used, growing conditions, storage conditions, transportation conditions, adulteration, contamination, manufacturing process, and shelf stability. Therefore, proper investigation of these complex samples requires the use of analysis systems that are multivariate in nature.32,47−50 Modern analytical platforms allow for this type of analysis by recording data comprised of responses over thousands of variables. To manage the chemical complexity and variability within these analytical data sets, data analysis techniques, such as principal component analysis (PCA),51,52 hierarchical cluster analysis (HCA),53,54 and projection pursuit analysis (PPA),55−57 should be employed to visualize subspaces of the data that may be of interest to researchers. For example, these methods can be used to determine relationships among samples, which can be informative to researchers in terms of classification or measurement sensitivity and can reveal many other important characteristics about the data. Herein, we propose the use of two-dimensional (2D) NMR spectroscopy in conjunction with multivariate statistical analysis to chemically fingerprint complex mixtures toward authenticating and ensuring the quality of NHPs, botanicals, and functional ingredients. NMR is a non-destructive and quantitative technique based on an equimolar response among all organic compounds composing an extract or a NMR sample.58−62 The association of multivariate statistical analysis and 1H NMR spectroscopy has shown to be a relevant method for the chemical investigation of botanicals and the detection, identification, and quantitation of adulterants.63−67 Because high-resolution and 2D NMR spectroscopy is generating megavariate data sets, the development of new untargeted and exploratory methods are required to exploit NMR quantitative capabilities and dimension reduction techniques must be used to access the latent chemical information in the data.68−70 In this study, heteronuclear single-quantum coherence−total correlation spectroscopy (HSQC−TOCSY) NMR experiments provided detailed and unique fingerprints by resolving correlations over a wide (1H−13C) heteronuclear chemical space.71,72 By surveying a chemically and biologically diverse library of herbal materials and extracts (more than 200 samples), the key advantages and limitations of the proposed HSQC−TOCSY-



EXPERIMENTAL SECTION

Sample Library. The sample library was assembled over the past few years with standards acquired from Alkemist Laboratories, ChromaDex, National Institute of Standards and Technology, National Research Council of Canada−Measurement Science and Standards, and United States Pharmacopeial Convention, as indicated in Table S2 of the Supporting Information. A sizable contribution to the library was made through a collaboration with ISURA, a NHP and food product verification and certification organization. The NHP and natural ingredient library was composed of a large, diverse list of raw botanicals and extracts used in the preparation of NHPs as well as formulated products that are commonly found in the marketplace. Sample Preparation. Raw natural ingredients were preprocessed by cutting them into smaller pieces or crushed using mortar and pestle to allow for better extraction of the chemical components. The NMR samples were then prepared by dissolving approximately 50 mg of sample in their natural state (liquid or solid) (Tables S1 and S2 of the Supporting Information) within 1.2 mL of deuterated methanol (CD3OD) with sonication (30 min, 25 °C). After sonication, the samples were centrifuged (5 min, 3000 rpm) and 600 μL was removed for NMR analysis. CD3OD was chosen as the solvent because it offered greater chemical diversity when using a diverse test set of samples compared to deuterated dimethyl sulfoxide (DMSOd6) and an ethyl acetate−water extraction run in CD3OD (Figures S1−S3 of the Supporting Information). All NMR samples were prepared identically to reliably evaluate the proposed one-size-fits-all analytical method. NMR Acquisition and Processing. NMR spectra used to determine the appropriate solvent for this study were recorded on a Bruker AVANCE III 500 MHz spectrometer equipped with a TXI 5 mm room-temperature probe. All other NMR spectra were recorded on a Bruker AVANCE III 700 MHz spectrometer equipped with a 5 mm TCI cryoprobe. All samples were tuned and matched to 50 Ω resistive impedance. 1H NMR was recorded for qualitative purposes only to obtain a spectral reference for the HSQC−TOCSY NMR by calibrating these to the residual methanol peak (3.31 ppm). These spectra were not used beyond this application within this study. HSQC−TOCSY NMR was recorded into 1024 complex points in the 1 H dimension and 256 complex points in the 13C dimension with spectral widths of 14.0025 and 165.6580 ppm, respectively. Delays in the HSQC were optimized for one-bond JCH of 145 Hz, and the TOCSY mixing time was 120 ms. Shaped adiabatic pulses were used for the 13C 180° refocusing pulses. Each transient was averaged over 16 scans with 16 preparation scans at the beginning of the experiment with a relaxation delay of 1.5 s. The spectra were processed using a 2D Fourier transform into 1024 real points in both dimensions after apodization with a squared sine function shifted, so that the maximum was at T = 0 (squared cosine). Prior to Fourier transform, the time domain in T1 was expanded to 1024 real points using the linear prediction function in TopSpin with 128 coefficients rather than simple zero filling. The resulting spectra were then phased to pure absorption mode and baseline-corrected in F2 using a fourth-order polynomial. These spectra were calibrated by copying the spectral reference from the 1H NMR spectrum for the 1H dimension, while calibration in the 13C dimension was performed using the residual methanol peak (49.0 ppm). Bucketing of NMR Spectra. Bucketing was conducted on all NMR spectra to improve data analysis. AMIX Statistics 3.9.12 was used to apply fixed grid buckets for the 2D NMR spectra using simple rectangular buckets with positive intensities and without scaling. The chemical ranges used for bucketing were between −1 and 12 ppm in the proton dimension and between −1 and 150 ppm in the carbon dimension. Various bucket sizes were evaluated with the HSQC− TOCSY spectra, including 1H of 0.01 ppm and 13C of 1 ppm, 1H of 7766

DOI: 10.1021/acs.jafc.9b01066 J. Agric. Food Chem. 2019, 67, 7765−7774

Journal of Agricultural and Food Chemistry

Article

0.5 ppm and 13C of 1 ppm, 1H of 0.5 ppm and 13C of 3 ppm, and 1H of 0.5 ppm and 13C of 5 ppm. The best results were obtained using the bucket size of 1H of 0.01 ppm with 13C of 1 ppm, resulting in a 151 × 1300 matrix for each sample prior to vectorization. At the same time, the residual solvent signals (water and methanol) were “removed” from these matrices. The residual water solvent signal was “removed” within the AMIX bucketing process by removing all 1 H signals and their associated 13C signals between 4.75 and 5.06 ppm, while the residual methanol solvent signal was “removed” at the same time by removing all 1H signals between 3.16 and 3.45 ppm and all 13C signals between 47.0 and 51.0 ppm. The matrices that remain of size 151 × 1300 from “removing” the solvent signal merely fill those cells with zeroes. All spectra were then vectorized in the 1H direction. This linear transformation converts the 151 × 1300 spectrum into a 1 × 196 300 row vector by concatenating the rows of the original matrix beside one another. This methodology was further investigated by creating a second data set containing only berry samples. For this data set, it was found that improved results could be obtained by applying further dimension reduction on each vectorized 1H of 0.01 ppm and 13C of 1 ppm spectrum (1 × 196 300). Specifically, each spectrum was bucketed with a bin size of 650, resulting in a 1 × 302 vector for each sample. After bucketing, all spectra were baseline-corrected using an iterative asymmetric weighted least squares algorithm available in PLS_Toolbox73 and normalized to unit length before statistical analysis. Binary Normalization. In addition to the processing methodology described in the previous section, binary normalization of the spectra was performed. The binary bucket tables were produced in MATLAB74 by first defining a threshold within each spectrum. For the vectorized and preprocessed 1H of 0.01 ppm and 13C of 1 ppm spectra (1 × 302), the threshold was arbitrarily defined as the mean of each spectrum plus 0.25 times its standard deviation. This threshold was then used to find peaks in the spectrum (using the findpeaks.m MATLAB function) according to their prominence. Any peak above the threshold was then given a value of 1, while all other bins were given a value of 0. Although there are obvious drawbacks to this thresholding method, such as penalizing spectra with a large range of response values, it was found to capture most of the informative peaks over all spectra in the data set. Chemical barcodes were then produced from this binary data set by coloring the bins; those assigned a 1 were colored black, and those containing a 0 were colored white. The barcode was then simplified by removing all of the variables that contained only zeroes throughout the entire sample set. This reduces the size of the data set while keeping informationcontaining variables. Multivariate Analysis. All data analysis was performed in MATLAB using software written “in-house”, unless otherwise stated. PCA was performed using singular value decomposition. The “quasipower” algorithm was used to perform exploratory PPA based on the optimization of kurtosis.75 The code for PPA can be found in the following public domain: http://groupwentzell.chemistry.dal.ca/ software.html. HCA was performed using the MATLAB functions linkage.m and dendrogram.m using the squared Euclidian distance metric and unweighted average distance for determining the distance between clusters. The Hamming distance metric was used for producing the dendrogram plots for the binary data.76

Table 1. Subset of 20 Representative Samples code

sample name

state

NRC-0012 NRC-0015 NRC-0023 NRC-0024 NRC-0030 NRC-0033 NRC-0040 NRC-0052 NRC-0055 NRC-0066 NRC-0068 NRC-0070 NRC-0072 NRC-0075 NRC-0080 NRC-0082 NRC-0088 NRC-0092 NRC-0094 NRC-0095

white grape juice concentrate Korean ginseng boysenberry liquid certified organic chia rich duck liver powder papaya powder whole Echinacea angustifolia root pineapple fruit powder blueberry juice concentrate 65 °Brix milled organic cherry rich cherry powder 36:1 spray-dried aloe vera 1:1 xollagen type II hydrolyzed stinging nettle cat’s claw powder papaya fruit powder dandelion root raw Gymnema sylvestre beef flavor noni juice concentrate

liquid powder liquid seeds powder powder powder powder liquid powder powder powder powder powder powder powder whole powder powder liquid

and maintaining good resolution between signals compared to the other solvents tested. Another study was also conducted to explore the error characteristics associated with NMR (see the Supporting Information). These preliminary results showed that the dominant error source in these studies is associated with the solvents (water and MeOH) and re-emphasized that NMR spectroscopy and the proposed experimental protocol are suitable for determining the chemical information in the data sets. Following these initial assessments, the versatility and reproducibility of the method was tested by extracting 20 representative samples (Table 1) in triplicate and recording HSQC−TOCSY NMR experiments over several months, an extensive period of time (see Table S1 of the Supporting Information). The 2D (1H−13C) NMR was also chosen over typical one-dimensional (1D) 1H NMR data to spread out the data over a larger chemical space and to provide information with higher resolution. For this reason, 1D techniques are not further discussed in this paper. The (1H−13C) HSQC− TOCSY experiment was specifically selected to take advantage of two distinct types of correlations. While HSQC correlates protons (δ1H of 0−12 ppm) to their adjacent carbons, allowing the data to be further resolved over the carbon dimension (δ13C of 0−150 ppm), TOCSY detects correlations between proton signals within a given spin system and, therefore, builds in redundancy into the data sets, allowing for greater discrimination between samples. The use of 2D NMR also avoids the use of deconvolution algorithms that are often a source of variance in the data set. Therefore, HSQC−TOCSY NMR (Figure 1A), coupled with multivariate statistical analysis (PPA, PCA, and HCA), were shown to be an adequate approach for assessing the chemical composition of a large and chemically diverse library of samples. Following the acquisition of the HSQC−TOCSY data, the three-dimensional (3D) data set (δ13C, δ1H, and signal intensity) was reduced to two dimensions [chemical shift coordinate/bucket (δ13C and δ1H) and intensity] by binning the spectra using fixed-width rectangular regions, which effectively vectorizes each spectrum (Figure 1B). Four



RESULTS AND DISCUSSION To ensure that the collected data were representative of the chemical composition of our samples, several deuterated solvents were evaluated for extraction (see the Supporting Information). This study was conducted using a sample set to represent the wide chemical and biological variety observed within the NHP library provided by ISURA (Table 1). The results showed that CD3OD offered the greatest chemical diversity while extracting the highest number of components 7767

DOI: 10.1021/acs.jafc.9b01066 J. Agric. Food Chem. 2019, 67, 7765−7774

Journal of Agricultural and Food Chemistry

Article

Figure 2. Investigating the reproducibility of the method using 20 representative samples (in triplicate) with (A) PCA scores plot of the first three principal components (PCs). (B) HCA performed on the PCA scores obtained from the first six PCs. (C) HCA performed on the multivariate PPA scores obtained after PCA compression to six components. (D) PPA scores plot of the first set classes (SCs).

Figure 1. (A) Example of a HSQC−TOCSY NMR spectrum recorded on a 700 MHz NMR spectrometer. (B) Overlay of the vectorized HSQC−TOCSY data of 20 spectra, after binning.

different bin sizes were explored (see the Experimental Section) to evaluate the influence of the matrix resolution on the statistical analysis outputs. These bin sizes were applied to the 20 samples, and then a series of multivariate statistical tools were investigated to rapidly assess the grouping patterns between each triplicate experiment while measuring the discrimination between the different types of samples. Overall, good separation between samples was observed in the PCA− HCA dendrograms using all four bucket sizes, and no significant variations were observed in the clustering patterns among replicates when the smallest bin size [1 and 0.01 ppm (13C and 1H)] is applied. Remarkably, the latter observations further emphasize the reproducibility of NMR as an analytical technique and its ability to generate high-resolution and content-rich data sets. As represented in Figure 2A, the PCA plot demonstrates good grouping and reproducibility among replicates using a bin size of 1 ppm (13C dimension) × 0.01 ppm (1H dimension). Moreover, the clustering patterns in the HCA−PCA dendrogram (Figure 2B) also demonstrate good discrimination among the 20 representative samples when performing HCA based on the PCA scores. As an aside, sample NRC-0012-1 was determined to not group with NRC-0012-2 and NRC-0012-3. This can be explained by examining the spectra for these samples in the Supporting Information. This result shows that the method can be used for easy identification of samples that are not following their expected groupings. The chosen approach does not only demonstrate the reproducibility of the method by grouping triplicate experi-

ments but also demonstrates that similar natural ingredients are also grouped together (Figure 2). This is most notably observed where fruit juice concentrates (NRC-0012, NRC0023, and NRC-055) and papaya powder samples (NRC-0033 and NRC-0082) form a clade, which is mostly explained by the abundant presence of carbohydrates in these samples. On the other hand, samples NRC-0024 (certified organic chia rich), NRC-0030 (duck liver powder), NRC-0066 (milled organic cherry), and NRC-0092 (Gymnema sylvestre) were shown to have unique NMR chemical fingerprints by forming distinct clusters in both PPA and PCA plots. Although PPA usually performs well only with a small and even number of balanced classes, it was found to be a pertinent approach to process the data set prior to performing HCA by conserving the relevant variables (rich in information) and reducing the multidimensional space (panels C and D of Figure 2). Unlike PCA, PP finds “interesting” projections of the data. Where PCA finds these directions based on the variance in the data, the PP algorithm used in this study finds these projections using the kurtosis of the data. By comparison of the projection plots and dendrograms for both multivariate projection statistical techniques (PCA and PPA), all of the triplicate extracts, except for NRC-0012, were closely clustered in both cases. However, it also became apparent that the type of methodology employed to process the data set has an impact on the relative distances observed among the classes of products in the dendrogram. As examples, 7768

DOI: 10.1021/acs.jafc.9b01066 J. Agric. Food Chem. 2019, 67, 7765−7774

Journal of Agricultural and Food Chemistry

Article

Figure 3. (A) Chemical barcodes and dendrogram (binary HCA), with triplicate samples being denoted following their respective sample codes with a 1, 2, or 3. Colors are added to aid the reader in identifying clusters but do not imply relationship between samples of different parent codes. (B) Average number (with standard deviations as error bars) of positive bins in each barcode for the sample set investigating method reproducibility.

the relative amounts of major components. Although these major metabolites are often good chemical markers to identify botanicals and functional ingredients, they also became chemicals of choice for the spiking and adulteration of NHPs. As a result, Figure 3A represents the chemical barcodes of the 20 representative NHPs and natural ingredients (in triplicate) with their corresponding location on the HCA dendrogram calculated from the binary normalized data using the Hamming distance. For reference to the HSQC−TOCSY spectra, the positive variables in the chemical barcode were sorted by increasing carbon chemical shifts in which the proton dimension is continuously cycling through each carbon unit (δ13C of 1 ppm). Remarkably, each class of sample exhibited a distinct and characteristic chemical barcode, despite reducing the data

NRC-0066 (milled organic cherry) was clustering with the two papaya samples (NRC-0033 and NRC-0082) in the reduced multidimensional space induced by PPA, while a better separation is obtained between the non-related samples NRC-0015 (Korean ginseng) and NRC-0075 (stinging nettle) in contrast to the PCA data. In addition to selecting an adequate statistical technique, normalization of the bucket table also plays an important role in emphasizing the variables and trends of interest. Herein, we are proposing binary normalization to support the authentication of NHPs and their natural ingredients. Binary normalization consists of assigning a value of 1 to all variables above a set threshold and 0 for the rest of the data set.77 The subsequent coloring of these bins results in the chemical barcode, which emphasizes the chemical diversity rather than 7769

DOI: 10.1021/acs.jafc.9b01066 J. Agric. Food Chem. 2019, 67, 7765−7774

Journal of Agricultural and Food Chemistry

Article

bins with a value close to the binary threshold and within the range of intrinsic experimental variances. The quality and reproducibility are directly correlated to the chemical contents extracted by the chosen solvent system as well as the processing parameters, including the bin size and the binary normalization threshold. In the proposed one-size-fits-all analytical method, it is critical to evaluate the information content of each barcode before performing multivariate statistical analyses and drawing any conclusions. For example, in the representative sample set, the barcode obtained for NRC-0095 (noni fruit concentrate) exhibited fewer bin counts in comparison to other samples and, consequently, could inherently cluster with adjacent samples as a result of the lack of information rather than similarities in their chemical profiles. Finally, it is also worthwhile to notice that the clustering pattern and distances between clades in a dendrogram are highly dependent upon the chemical diversity in the data set or the biodiversity in the sample set. To complete the illustration of the proposed NMR methodology and better appreciate its application for authenticating NHPs and natural ingredients based on their chemical profiles, a subset of berry samples, including certified reference materials (dried berries) as well as closely related manufactured ingredients, was investigated (see Table 2). The resulting data set was built by extracting each sample in duplicate within several months between replicates. Following the NMR data acquisition and processing of the binary normalized data (vectorized, binned, and preprocessed 1H of

matrix to a simplistic binary data set. These results highlight the ability of the methodology to efficiently reduce complex chemical compositions into binary and content-rich information suitable for the development of mobile traceability systems [e.g., quick response (QR) code] for the food and NHP industry.78 Moreover, all triplicate samples showed high similarity in their grouping patterns and chemical code compositions, as supported by the reproducible numbers of positive variables for most of the samples (Figure 3B). Interestingly and in a similar manner as previously observed, the samples are still grouping according to similarities in their chemical composition expected from the different natural ingredients, such as juice, fruit, berry, plant materials, etc. Indeed, the bottom part of the dendrogram is composed by two clades: the first formed by the cherry samples NRC-0068 and NRC-0066, while the adjacent clade is composed of a group of fruit-derived samples, including the three juice concentrates (NRC-0012, NRC-0023, and NRC-0055) and the two papaya samples (NRC-0033 and NRC-0082). Moreover, the chemical barcoding representation allows for a rapid visualization of the main similarities and differences in the chemical composition for these samples. For example, by the observation of upfield 13C NMR signals (left section of the chemical barcode), it can be seen that there are points in the cherry samples that are clearly absent in both adjacent clades of papaya and juice concentrates. These signals were assigned to NMR signals attributable to the presence of lipids or fatty-acidderived metabolites. To complement these results, Figure 3B displays the average number of positive buckets captured in each chemical barcode and its associated standard deviation calculated from the triplicate analysis. These values provide critical information to evaluate the relevance of a chemical barcode as a representative way of capturing the chemical diversity of a given sample. Indeed, solvent extraction methods, NMR acquisition parameters, and data processing parameters (e.g., bin size and binary threshold) are directly correlated to the quality and content of a barcode, and it is therefore important to evaluate whether a chemical barcode contains sufficient and reproducible information to confidently authenticate a specific natural ingredient. For example, the chemical barcode of an unknown sample exhibiting high similarities to a certified reference barcode with a high number of positive variables would therefore be identified and authenticated with a high degree of confidence. As indicated in Figure 3B, most chemical barcodes extrapolated in this small study encompassed significant numbers of positive bins (more than 100), with the largest standard variations observed for NRC-0070 (spray-dried aloe vera 1:1) and NRC-0072 (collagen type II hydrolyzed). Despite the large number of positive bins, NRC-0070 and NRC-0072 were also identified as outliers using a binary normalization approach as a result of the large Hamming distances within replicates. These observations and the large standard deviations are explained by the presence of few abundant signals coupled with noisy baselines that are leading to discrepancies among the replicate experiments. Indeed, the binning threshold is defined by the mean of each spectrum and led to the selection of signals belonging to the noise as a result of the absence of significant NMR signals, as indicated in the HSQC−TOCSY spectra for NRC-0070 and NRC-0072 (Figures S79 and S81 of the Supporting Information). Indeed, the presence or absence strategy during normalization may emphasize variabilities in the chemical barcodes caused by any

Table 2. Berry Sample Subset code NRC-0007 NRC-0008 NRC-0009 NRC-0010 NRC-0018 NRC-0020 NRC-0023 NRC-0027 NRC-0028 NRC-0043 NRC-0045 NRC-0046 NRC-0047 NRC-0055 NRC-0059 NRC-0060 NRC-0062 NRC-0083 NRC-0085 NRC-0089 NRC-0090 NRC-0091 NRC-0102 NRC-0191 NRC-0192 NRC-0195 NRC-0196 NRC-0197 NRC-0198 NRC-0214 7770

sample name ́ açai 4:1 chokeberry rich 36:1 black raspberry extract milled cranberry elderberry rich blueberry milled boysenberry liquid cranberry powder raspberry rich cranberry 20:1 blueberry powder basic mulberry leaf powder strawberry rich blueberry juice concentrate 65 °Brix cranberry fruit raw cranberry powder black currant rich Hawthorn berry basic powder Schisandra berry powder Saskatoon berry 36:1 organic cranberry rich açai ́ rich powder organic blueberry rich Vaccinium oxycoccus (cranberry) Vaccinium myrtillus (bilberry) Rubus idaeus (raspberry) Vaccinium macrocarpon (cranberry) Vaccinium myrtillus (bilberry) Vaccinium corymbosum (blueberry) Ribes nigrum (black currant)

state powder powder powder powder powder powder liquid powder powder powder powder powder powder liquid powder powder powder powder powder powder powder powder powder fruit (blended) whole fruit whole fruit whole fruit whole fruit whole fruit whole fruit

DOI: 10.1021/acs.jafc.9b01066 J. Agric. Food Chem. 2019, 67, 7765−7774

Journal of Agricultural and Food Chemistry

Article

Figure 4. (A) Chemical barcodes and dendrogram (binary HCA), with replicate samples being denoted following their respective sample codes with a 1, 2, or 3. (B) Number of positive bins in each barcode for the berry sample set. Standards are denoted as std.

0.01 ppm and 13C of 1 ppm spectra, where each spectrum has dimensions of 1 × 302), chemical barcodes were generated for each berry-based product (Figure 4). Similar to Figure 3, the robustness and reproducibility of the methodology were highlighted in this study because the duplicate experiments were found to cluster together, which can be seen in the dendrogram (Figure 4). This is rather noticeable because the closely related berry-derived products are expected to display higher similarity in their chemical compositions (i.e., chemical barcodes) in comparison to the previous representative data set that displayed a wider biological and chemical diversity. As described in Table 2,

samples NRC-0191−NRC-0214 are dried fruit reference materials that have been authenticated and certified by a commercial analytical laboratory. Interestingly, these samples formed one clade on the dendrogram with the addition of the two juice concentrates NRC-0023 and NRC-0055. Moreover, the two cranberry reference materials (NRC-0191 and NRC0196) are located on two separate subclades that are distinct from the remaining certified berries and, therefore, suggest that cranberry fruits can be distinguished among other berries and potentially further classified between the two cranberry species Vaccinium oxycoccus and Vaccinium macrocarpon. By visualization of the chemical barcodes, two trends in the chemical 7771

DOI: 10.1021/acs.jafc.9b01066 J. Agric. Food Chem. 2019, 67, 7765−7774

Journal of Agricultural and Food Chemistry

Article

fingerprints were observed in the data set and resulted in the formation of the two major clades I and II. From the interpretation of the chemical shifts attributed to the positive variables in the barcodes, clade II appears to represent berrybased products that have been processed using excipients or fillers in their formulations. Indeed, characteristic chemical signatures are unique to the clade II samples and are not found in any of the berry reference materials. As a result, the cranberry powders NRC-0010, NRC-0043, and NRC-0060 displayed antipodal chemical barcodes when compared to both reference samples NRC-0191 and NRC-0196. Consequently, the labeling of these three dietary supplements can be misleading, especially for sample NRC-0010, because it was described as milled cranberry. On the other hand, NRC-0090 (organic cranberry rich) was located in the subclade Ib and showed high similarity in its chemical profile to the two cranberry reference materials (NRC-0191 and NRC-0196). It is also worth noting that the dietary supplement NRC0083 (Hawthorn berry basic powder) was closely clustering with the reference berry materials, while the three blueberryderived products NRC-0020 (blueberry milled), NRC-0045 (blueberry powder), and NRC-0102 (organic blueberry rich) were grouping together within clade Ib. On the other hand, NRC-007 (açai 4:1), NRC-0046 (basic mulberry leaf powder), and NRC-0085 (Schisandra berry powder) produced unique clades with a significant Hamming distance with adjacent samples, which are in agreement with the unique chemical profiles expected for those samples. Although the total number of samples is limited, it is conceivable to introduce an unknown berry-derived product and propose its identity by locating its position on the dendrogram. With the addition of more reference materials in replicates, the association of HCA and NMR-based chemical barcoding would provide a new and relevant analytical method for the authentication of raw materials and formulated products with high confidence. Moreover, the proposed methodology is untargeted and groups botanicals and formulated products according to their chemical composition rather than a DNA sequence. Indeed, the cranberry samples are spread over the entire dendrogram by displaying different chemical compositions, despite sharing the same genetic information. This observation also highlights the importance of creating diverse libraries of authentic plant materials and known processed and formulated products to allow for direct comparison between similar types of samples. Because the nutritional value and health benefits are directly correlated to chemical composition, the development of a chemical-based methodology is therefore highly relevant to characterize the quality of food and functional ingredients. The proposed methodology leveraged the advantage of NMR as a reproducible and quantitative analytical technique to accurately capture the composition of chemically diverse samples. Moreover, the 2D HSQC−TOCSY experiment greatly improved the dispersion of the NMR data on a wide chemical space and provided data sets that are highly suitable for multivariate statistical methods. Herein, both PCA and HCA were explored to highlight key features and trends in the data sets to cluster NHPs according to their chemical composition. This study also introduced binary normalization and its benefits for reducing the dimensionality of complex data sets as well as creating chemical barcodes as unique chemical signatures for a given sample. The chemical barcoding approach also provides the advantage of emphasizing chemical

diversity rather than metabolite abundance while maintaining enough information to discriminate among classes of samples. Therefore, NMR chemical barcoding provides a complementary analytical method to DNA barcoding to qualitatively highlight the similarities and differences among NHPs. By building libraries of reference chemical barcodes recorded from authenticated raw materials and formulated products, the proposed NMR methodology enables the development of prediction models for the identification of botanicals, ensuring the quality of formulated products and highlighting the presence of potential adulterants. To validate this hypothesis, additional studies are underway and require access to broader sample sets with detailed metadata on the vouchers by developing collaborations with botanical producers and NHP manufacturers.



ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jafc.9b01066. Complete sample list, solvent determination experiment, reproducibility studies, and NMR spectral files (PDF)



AUTHOR INFORMATION

Corresponding Author

*Telephone: 1-902-426-0636. E-mail: [email protected]. ORCID

Stephen Driscoll: 0000-0002-4943-2571 Ian Burton: 0000-0002-9387-0595 Fabrice Berrué: 0000-0003-2389-4388 Author Contributions ‡

Stephen Driscoll and Chelsi Wicks contributed equally to this work. Funding

The authors thank the Natural Sciences and Engineering Research Council of Canada (NSERC) for funding. Notes

The authors declare no competing financial interest.

■ ■

ACKNOWLEDGMENTS The authors acknowledge the non-profit organization ISURA for providing raw materials and formulated products. REFERENCES

(1) Ekor, M. The growing use of herbal medicines: Issues relating to adverse reactions and challenges in monitoring safety. Front. Pharmacol. 2014, 4, 1−10. (2) Cheung, F. TCM: Made in China. Nature 2011, 480, S82. (3) Kroes, B. H. The legal framework governing the quality of (traditional) herbal medicinal products in the European Union. J. Ethnopharmacol. 2014, 158, 449−453. (4) WHO Expert Committee on Specifications for Pharmaceutical Preparations WHO Guidelines for Selecting Marker Substances of Herbal Origin for Quality Control of Herbal Medicines (Annex 1); World Health Organization (WHO): Geneva, Switzerland, 2017; WHO Technical Report Series 1003, http://apps.who.int/medicinedocs/ en/d/Js23240en/ (accessed Oct 2018). (5) Job, K. M.; Kiang, T. K. L.; Constance, J. E.; Sherwin, C. M. T.; Enioutina, E. Y. Herbal medicines: Challenges in the modern world. Part 4. Canada and United States. Expert Rev. Clin. Pharmacol. 2016, 9, 1597−1609. 7772

DOI: 10.1021/acs.jafc.9b01066 J. Agric. Food Chem. 2019, 67, 7765−7774

Journal of Agricultural and Food Chemistry

Article

(6) Mosihuzzaman, M. Herbal medicine in healthcareAn overview. Nat. Prod. Commun. 2012, 7, 807−812. (7) Schilter, B.; Andersson, C.; Anton, R.; Constable, A.; Kleiner, J.; O’Brien, J.; Renwick, A. G.; Korver, O.; Smit, F.; Walker, R. Guidance for the safety assessment of botanicals and botanical preparations for use in food and food supplements. Food Chem. Toxicol. 2003, 41, 1625−1649. (8) Khan, R. A. Natural products chemistry: The emerging trends and prospective goals. Saudi Pharm. J. 2018, 26, 739−753. (9) Avigan, M. I.; Mozersky, R. P.; Seeff, L. B. Scientific and Regulatory Perspectives in Herbal and Dietary Supplement Associated Hepatotoxicity in the United States. Int. J. Mol. Sci. 2016, 17, 331− 361. (10) Calahan, J.; Howard, D.; Almalki, A. J.; Gupta, M. P.; Calderon, A. I. Chemical Adulterants in Herbal Medicinal Products: A Review. Planta Med. 2016, 82, 505−515. (11) Raynor, D. K.; Dickinson, R.; Knapp, P.; Long, A. F.; Nicolson, D. J. Buyer beware? Does the information provided with herbal products available over the counter enable safe use? BMC Med. 2011, 9, 94. (12) Ogbogu, U.; Necyk, C. Community Pharmacists’ Views and Practices Regarding Natural Health Products Sold in Community Pharmacies. PLoS One 2016, 11, e0163450. (13) Heinrich, M. Quality and safety of herbal medical products: Regulation and the need for quality assurance along the value chains. Br. J. Clin. Pharmacol. 2015, 80, 62−66. (14) Health Canada. The Approach to Natural Health Products; Health Canada: Ottawa, Ontario, Canada, 2018; https://www.canada. ca/content/dam/hc-sc/migration/hc-sc/dhp-mps/alt_formats/pdf/ prodnatur/nhp-psn-eng.pdf (accessed Nov 24, 2018). (15) Smith, A.; Jogalekar, S.; Gibson, A. Regulation of natural health products in Canada. J. Ethnopharmacol. 2014, 158, 507−510. (16) U.S. Food and Drug Administration (FDA) Dietary Supplements; FDA: Silver Spring, MD, 2018; https://www.fda.gov/Food/ DietarySupplements/. (17) National Institutes of Health (NIH) Dietary Supplement Health and Education Act of 1994; NIH: Bethesda, MD, 1994. (18) Marcus, D. M. Dietary supplements: What’s in a name? What’s in the bottle? Drug Test. Anal. 2016, 8, 410−412. (19) O’Connor, A. New York Attorney General Targets Supplements at Major Retailers. The New York Times; The New York Times Company: New York, 2015. (20) Newmaster, S. G.; Grguric, M.; Shanmughanandhan, D.; Ramalingam, S.; Ragupathy, S. DNA barcoding detects contamination and substitution in North American herbal products. BMC Med. 2013, 11, 222. (21) Tnah, L. H.; Lee, S. L.; Tan, A. L.; Lee, C. T.; Ng, K. K. S.; Ng, C. H.; Nurul Farhanah, Z. DNA barcode database of common herbal plants in the tropics: A resource for herbal product authentication. Food Control 2019, 95, 318−326. (22) Gao, Z.; Liu, Y.; Wang, X.; Song, J.; Chen, S.; Ragupathy, S.; Han, J.; Newmaster, S. G. Derivative Technology of DNA Barcoding (Nucleotide Signature and SNP Double Peak Methods) Detects Adulterants and Substitution in Chinese Patent Medicines. Sci. Rep. 2017, 7, 5858. (23) de Boer, H. J.; Ichim, M. C.; Newmaster, S. G. DNA Barcoding and Pharmacovigilance of Herbal Medicines. Drug Saf. 2015, 38, 611−620. (24) Palhares, R. M.; Gonçalves Drummond, M.; dos Santos Alves Figueiredo Brasil, B.; Pereira Cosenza, G.; das Graças Lins Brandão, M.; Oliveira, G. Medicinal plants recommended by the world health organization: DNA barcode identification associated with chemical analyses guarantees their quality. PLoS One 2015, 10, e0127866. (25) Xin, T.; Xu, Z.; Jia, J.; Leon, C.; Hu, S.; Lin, Y.; Ragupathy, S.; Song, J.; Newmaster, S. G. Biomonitoring for traditional herbal medicinal products using DNA metabarcoding and single molecule, real-time sequencing. Acta Pharm. Sin. B 2018, 8, 488−497.

(26) Han, J.; Pang, X.; Liao, B.; Yao, H.; Song, J.; Chen, S. An authenticity survey of herbal medicines from markets in China using DNA barcoding. Sci. Rep. 2016, 6, 18723. (27) Srirama, R.; Santhosh Kumar, J. U.; Seethapathy, G. S.; Newmaster, S. G.; Ragupathy, S.; Ganeshaiah, K. N.; Uma Shaanker, R.; Ravikanth, G. Species Adulteration in the Herbal Trade: Causes, Consequences and Mitigation. Drug Saf. 2017, 40, 651−661. (28) Zhang, J.; Wider, B.; Shang, H.; Li, X.; Ernst, E. Quality of herbal medicines: Challenges and solutions. Complementary therapies in medicine 2012, 20, 100−106. (29) Zovko Koncic, M. Getting More Than You Paid For: Unauthorized “Natural” Substances in Herbal Food Supplements on EU Market. Planta Med. 2018, 84, 394−406. (30) Kuszak, A. J.; Hopp, D. C.; Williamson, J. S.; Betz, J. M.; Sorkin, B. C. Approaches by the US National Institutes of Health to support rigorous scientific research on dietary supplements and natural products. Drug Test. Anal. 2016, 8, 413−417. (31) Stadler, R. H.; Tran, L. A.; Cavin, C.; Zbinden, P.; Konings, E. J. Analytical Approaches to Verify Food Integrity: Needs and Challenges. J. AOAC Int. 2016, 99, 1135−1144. (32) Pinu, F. R. MetabolomicsThe new frontier in food safety and quality research. Food Res. Int. 2015, 72, 80−81. (33) Raclariu, A. C.; Heinrich, M.; Ichim, M. C.; de Boer, H. Benefits and Limitations of DNA Barcoding and Metabarcoding in Herbal Product Authentication. Phytochem. Anal. 2018, 29, 123−128. (34) Pawar, R. S.; Handy, S. M.; Cheng, R.; Shyong, N.; Grundel, E. Assessment of the Authenticity of Herbal Dietary Supplements: Comparison of Chemical and DNA Barcoding Methods. Planta Med. 2017, 83, 921−936. (35) Simmler, C.; Chen, S.-N.; Anderson, J.; Lankin, D. C.; Phansalkar, R.; Krause, E.; Dietz, B.; Bolton, J. L.; Nikolic, D.; van Breemen, R. B.; Pauli, G. F. Botanical Integrity: The Importance of the Integration of Chemical, Biological, and Botanical Analyses, and the Role of DNA Barcoding. HerbalGram 2015, 106, 58−60. (36) Parveen, I.; Gafner, S.; Techen, N.; Murch, S. J.; Khan, I. A. DNA Barcoding for the Identification of Botanicals in Herbal Medicine and Dietary Supplements: Strengths and Limitations. Planta Med. 2016, 82, 1225−1235. (37) Goodarzi, M.; Russell, P. J.; Vander Heyden, Y. Similarity analyses of chromatographic herbal fingerprints: A review. Anal. Chim. Acta 2013, 804, 16−28. (38) Glish, G. L.; Vachet, R. W. The basics of mass spectrometry in the twenty-first century. Nat. Rev. Drug Discovery 2003, 2, 140−150. (39) Bouslimani, A.; Sanchez, L. M.; Garg, N.; Dorrestein, P. C. Mass spectrometry of natural products: Current, emerging and future technologies. Nat. Prod. Rep. 2014, 31, 718−729. (40) Bianchi, F.; Riboni, N.; Termopoli, V.; Mendez, L.; Medina, I.; Ilag, L.; Cappiello, A.; Careri, M. MS-Based Analytical Techniques: Advances in Spray-Based Methods and EI-LC-MS Applications. J. Anal. Methods Chem. 2018, 2018, 1−24. (41) Al Lawati, H. A.; Al Busaidi, I.; Kadavilpparampu, A. M.; Suliman, F. O. Determination of Common Adulterants in Herbal Medicine and Food Samples using Core-shell Column Coupled to Tandem Mass Spectrometry. J. Chromatogr. Sci. 2017, 55, 232−242. (42) Hemdan, A.; Tawakol, S. M. HPLC−UV Chromatographic Methods for Detection and Quantification of Undeclared Withdrawn Synthetic Medications in Counterfeit Herbal Medicines with Confirmation by HPLC−PDA and Mass Spectrometry. Chromatographia 2018, 81, 777−783. (43) Vaclavik, L.; Krynitsky, A. J.; Rader, J. I. Mass spectrometric analysis of pharmaceutical adulterants in products labeled as botanical dietary supplements or herbal remedies: A review. Anal. Bioanal. Chem. 2014, 406, 6767−6790. (44) Wang, L.; Liu, L. F.; Wang, J. Y.; Shi, Z. Q.; Chang, W. Q.; Chen, M. L.; Yin, Y. H.; Jiang, Y.; Li, H. J.; Li, P.; Yao, Z. P.; Xin, G. Z. A strategy to identify and quantify closely related adulterant herbal materials by mass spectrometry-based partial least squares regression. Anal. Chim. Acta 2017, 977, 28−35. 7773

DOI: 10.1021/acs.jafc.9b01066 J. Agric. Food Chem. 2019, 67, 7765−7774

Journal of Agricultural and Food Chemistry

Article

supplements marketed for weight loss. J. Pharm. Biomed. Anal. 2016, 124, 34−47. (65) Kumar, D. Nuclear Magnetic Resonance (NMR) Spectroscopy For Metabolic Profiling of Medicinal Plants and Their Products. Crit. Rev. Anal. Chem. 2016, 46, 400−412. (66) Monakhova, Y. B.; Holzgrabe, U.; Diehl, B. W. K. Current role and future perspectives of multivariate (chemometric) methods in NMR spectroscopic analysis of pharmaceutical products. J. Pharm. Biomed. Anal. 2018, 147, 580−589. (67) Nguyen, H. T.; Lee, D. K.; Choi, Y. G.; Min, J. E.; Yoon, S. J.; Yu, Y. H.; Lim, J.; Lee, J.; Kwon, S. W.; Park, J. H. A 1H NMR-based metabolomics approach to evaluate the geographical authenticity of herbal medicine and its application in building a model effectively assessing the mixing proportion of intentional admixtures: A case study of Panax ginseng: Metabolomics for the authenticity of herbal medicine. J. Pharm. Biomed. Anal. 2016, 124, 120−128. (68) Engelsen, S. B.; Savorani, F.; Rasmussen, M. A. Chemometric Exploration of Quantitative NMR Data. In eMagRes; Harris, R. K., Wasylishen, R. L., Eds.; John Wiley & Sons, Ltd.: Chichester, U.K., 2013; Vol. 2, pp 267−278, . (69) Mahrous, E. A.; Farag, M. A. Two dimensional NMR spectroscopic approaches for exploring plant metabolome: A review. J. Adv. Res. 2015, 6, 3−15. (70) Smolinska, A.; Blanchet, L.; Buydens, L. M.; Wijmenga, S. S. NMR and pattern recognition methods in metabolomics: From data acquisition to biomarker discovery: A review. Anal. Chim. Acta 2012, 750, 82−97. (71) Bingol, K.; Bruschweiler-Li, L.; Li, D.-W.; Brüschweiler, R. Customized Metabolomics Database for the Analysis of NMR 1H− 1H TOCSY and 13C−1H HSQC-TOCSY Spectra of Complex Mixtures. Anal. Chem. 2014, 86, 5494−5501. (72) Buedenbender, L.; Habener, L. J.; Grkovic, T.; Kurtböke, D. I.;̇ Duffy, S.; Avery, V. M.; Carroll, A. R. HSQC−TOCSY Fingerprinting for Prioritization of Polyketide- and Peptide-Producing Microbial Isolates. J. Nat. Prod. 2018, 81, 957−965. (73) Eigenvector Research, Inc. PLS_Toolbox 8.6; Eigenvector Research, Inc.: Manson, WA, 2018. (74) The MathWorks, Inc. MATLAB, Release 2018a; The MathWorks, Inc.: Natick, MA, 2018. (75) Hou, S.; Wentzell, P. D. Fast and simple methods for the optimization of kurtosis used as a projection pursuit index. Anal. Chim. Acta 2011, 704, 1−15. (76) Hamming, R. W. Error detecting and error correcting codes. Bell Syst. Tech. J. 1950, 29, 147−160. (77) Forner, D.; Berrué, F.; Correa, H.; Duncan, K.; Kerr, R. G. Chemical dereplication of marine actinomycetes by liquid chromatography-high resolution mass spectrometry profiling and statistical analysis. Anal. Chim. Acta 2013, 805, 70−79. (78) Cai, Y.; Li, X.; Wang, R.; Yang, Q.; Li, P.; Hu, H. Quality Traceability System of Traditional Chinese Medicine Based on Two Dimensional Barcode Using Mobile Intelligent Technology. PLoS One 2016, 11, e0165263.

(45) Scalbert, A.; Brennan, L.; Fiehn, O.; Hankemeier, T.; Kristal, B. S.; van Ommen, B.; Pujos-Guillot, E.; Verheij, E.; Wishart, D.; Wopereis, S. Mass-spectrometry-based metabolomics: Limitations and recommendations for future progress with particular focus on nutrition research. Metabolomics 2009, 5, 435−458. (46) Berg, M.; Vanaerschot, M.; Jankevics, A.; Cuypers, B.; Breitling, R.; Dujardin, J.-C. LC-MS metabolomics from study design to dataanalysisUsing a versatile pathogen as a test case. Comput. Struct. Biotechnol. J. 2013, 4, e201301002. (47) Zhu, J.; Fan, X.; Cheng, Y.; Agarwal, R.; Moore, C. M. V.; Chen, S. T.; Tong, W. Chemometric Analysis for Identification of Botanical Raw Materials for Pharmaceutical Use: A Case Study Using Panax notoginseng. PLoS One 2014, 9, e87462. (48) Bansal, A.; Chhabra, V.; Rawal, R. K.; Sharma, S. Chemometrics: A new scenario in herbal drug standardization. J. Pharm. Anal. 2014, 4, 223−233. (49) Gad, H. A.; El-Ahmady, S. H.; Abou-Shoer, M. I.; Al-Azizi, M. M. Application of Chemometrics in Authentication of Herbal Medicines: A Review. Phytochem. Anal. 2013, 24, 1−24. (50) Fujimura, Y.; Kawano, C.; Maeda-Murayama, A.; Nakamura, A.; Koike-Miki, A.; Yukihira, D.; Hayakawa, E.; Ishii, T.; Tachibana, H.; Wariishi, H.; Miura, D. A Chemometrics-driven Strategy for the Bioactivity Evaluation of Complex Multicomponent Systems and the Effective Selection of Bioactivity-predictive Chemical Combinations. Sci. Rep. 2017, 7, 2257. (51) Bro, R.; Smilde, A. K. Principal component analysis. Anal. Methods 2014, 6, 2812−2831. (52) Lever, J.; Krzywinski, M.; Altman, N. Principal component analysis. Nat. Methods 2017, 14, 641−642. (53) Zhang, Z.; Murtagh, F.; Van Poucke, S.; Lin, S.; Lan, P. Hierarchical cluster analysis in clinical research with heterogeneous study population: Highlighting its visualization with R. Ann. Transl. Med. 2017, 5, 75−75. (54) Gil-Garcia, R. J.; Badia-Contelles, J. M.; Pons-Porrata, A. A General Framework for Agglomerative Hierarchical Clustering Algorithms. Proceedings of the 18th International Conference on Pattern Recognition (ICPR’06); Hong Kong, Aug 20−24, 2006; pp 569−572. (55) Friedman, J. H.; Tukey, J. W. A Projection Pursuit Algorithm for Exploratory Data Analysis. IEEE Trans. Comput. 1974, C-23, 881− 890. (56) Jiang, T.; Jia, H.; Yuan, H.; Zhou, N.; Li, F. Projection Pursuit: A General Methodology of Wide-Area Coherency Detection in Bulk Power Grid. IEEE Trans. Power Syst. 2016, 31, 2776−2786. (57) Posse, C. Projection pursuit exploratory data analysis. Comput. Stat. Data Anal. 1995, 20, 669−687. (58) Burton, I. W.; Quilliam, M. A.; Walter, J. A. Quantitative 1H NMR with External Standards: Use in Preparation of Calibration Solutions for Algal Toxins and Other Natural Products. Anal. Chem. 2005, 77, 3123−3131. (59) Pauli, G. F. qNMRA versatile concept for the validation of natural product reference compounds. Phytochem. Anal. 2001, 12, 28−42. (60) Pauli, G. F.; Gödecke, T.; Jaki, B. U.; Lankin, D. C. Quantitative 1H NMR. Development and potential of an analytical method: An update. J. Nat. Prod. 2012, 75, 834−851. (61) Simmler, C.; Napolitano, J. G.; McAlpine, J. B.; Chen, S.-N.; Pauli, G. F. Universal quantitative NMR analysis of complex natural samples. Curr. Opin. Biotechnol. 2014, 25, 51−59. (62) Barding, G. A., Jr.; Salditos, R.; Larive, C. K. Quantitative NMR for bioanalysis and metabolomics. Anal. Bioanal. Chem. 2012, 404, 1165−79. (63) Gilard, V.; Balayssac, S.; Tinaugus, A.; Martins, N.; Martino, R.; Malet-Martino, M. Detection, identification and quantification by 1H NMR of adulterants in 150 herbal dietary supplements marketed for improving sexual performance. J. Pharm. Biomed. Anal. 2015, 102, 476−493. (64) Hachem, R.; Assemat, G.; Martins, N.; Balayssac, S.; Gilard, V.; Martino, R.; Malet-Martino, M. Proton NMR for detection, identification and quantification of adulterants in 160 herbal food 7774

DOI: 10.1021/acs.jafc.9b01066 J. Agric. Food Chem. 2019, 67, 7765−7774