Non-targeted Chemical Analysis of Consumer Botanical Products Labelled as Blue Cohosh (Caulophyllum thalictroides), Goldenseal (Hydrastis Canadensis), or Yohimbe Bark (Pausinystalia johimbe) by NMR and MS

Abstract

Consumers have unprecedented access to botanical dietary supplements through on-line retailers, making it difficult to ensure product quality and authenticity. Therefore, methods to survey and compare chemical compositions across botanical products are needed. Nuclear magnetic resonance (NMR) spectroscopy and non-targeted mass spectrometry (MS) were used to chemically analyze commercial products labelled as containing one of three botanicals: blue cohosh, goldenseal, and yohimbe bark. Aqueous and organic phase extracts were prepared and analyzed in tandem with NMR followed by MS. We processed the non-targeted data using multivariate statistics to analyze the composition similarity across extracts. In each case there were several product outliers that were identified using principal component analysis (PCA). Evaluation of select known constituents proved useful to contextualize PCA subgroups, which in some cases supported or excluded product authenticity. The NMR and MS data reached similar conclusions independently but were also complementary.

Graphical Abstract

Introduction

Botanical dietary supplements are becoming increasingly popular throughout the U.S. because of their uses as ‘natural’ remedies and in functional foods (i.e., foods claiming to have added health benefits) (1). However, botanical supplements are composed of complex mixtures of constituents that can vary depending on multiple factors (e.g., growth conditions, plant part, processing), making them difficult to characterize or to define “standard” samples. Additionally, the dietary supplement industry has limited regulation, which can lead to adulterated products being sold in the market (2). Adulteration may be motivated by increasing the efficacy of these products by adding active ingredients or pharmaceuticals, or adulteration may be incentivized by cost reduction (3). The natural complexity, deficient regulatory constraints, and prevalence of adulteration all contribute to variability in quality and authenticity of botanical products in the marketplace.

Numerous techniques have been used to determine product quality and authenticity ranging from product origin tracking, visual investigation of packing, and chemical analysis (2). Of the many techniques used for chemical analysis, nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry (MS) remain primary approaches.(4, 5) NMR measures the resonance frequencies of nuclei within a magnetic field, indicative of the structure of individual chemicals as well as chemicals within mixtures. MS measures the mass-to-charge of ionized chemicals, providing molecular formulae, as well as structural information via fragmentation of the ionized chemical. Often, chemical separation (e.g., liquid chromatography) is performed prior to MS analysis which serves to improve differentiation of chemical constituents in complex samples. Combining NMR and MS has been applied to studying the authenticity of Ginkgo biloba botanical products, as well as other food metabolite research projects. (6, 7)

As an initial step in creating a methodological workflow for the chemical analysis of consumer botanical products, we devised a strategy for measuring extracts of botanical products by NMR, non-destructively, and subsequently using ultra-high performance liquid chromatography coupled to high resolution tandem mass spectrometry (UHPLC-HRMS/MS) to strengthen characterization information with the same samples. This approach was applied to botanical products, consisting of powders, tablets, and capsules, extracted using two different methods (aqueous and organic) to cover a broad range of chemistry. The chemical profiles were investigated using principal component analysis (PCA) and supported by assessment of individual chemical components for each method of analysis (i.e., NMR, MS). The chemical profiles of each botanical product were compared, without bias to the intra-type (powders, tablets, capsules) differences within each. Mid-level data fusion (8) was used to generate a singular overall chemical profile that was interpreted for similarities and differences.

Herein, we demonstrate the utility of our approach on multiple products containing one of three different botanicals: blue cohosh (Caulophyllum thalictroides), goldenseal (Hydrastis Canadensis), and yohimbe bark (Pausinystalia yohimbe). These three botanical products have been proposed to have various health effects due to specific chemical constituents, a few of which we will summarize here. This paragraph is not intended to be comprehensive regarding health effects or important chemical constituents, but rather to introduce a few compounds that will be discussed later with respect to product authenticity. First, blue cohosh has been commonly used to ease pain for menstrual cramps and childbirth and to induce labor (9). Magnoflorine is a key constituent of blue cohosh and thought to possess a wide variety of the pharmacological properties (10). Blue cohosh also contains cauloside C, which has been reported to exhibit concentration and time-dependent mitochondrial toxicity at certain exposure levels (11). Second, goldenseal is a common ingredient in supplements used to treat respiratory illnesses and promote immune function (12). Berberine is one of the main active constituents of goldenseal and purportedly possesses anti-inflammatory and antioxidant properties (12). However, berberine has been shown to possess a high potential for adverse effects including inhibition of CYP enzymes (13) and inhibition of topoisomerase II (14). Goldenseal also contains hydrastine, which is structurally related to berberine and has been suggested to possess various pharmacological properties (6). Similarly, canadine is also a known constituent of goldenseal and produced antioxidant properties with a low cytotoxic effect (15). Third, yohimbe bark has been used for sexual dysfunction in men, performance enhancement, and weight loss (16). Yohimbine is the main active alkaloid found in yohimbe bark; however, it is known to possess adverse effects including elevated blood pressure and heart rate, anxiety, and increased urinary output (17). In this study, the primary read-out was non-targeted chemical analysis; however, it will be shown that analyzing a few of these key chemical constituents also provided valuable information to characterize differences among products.

Material and Methods

Botanical Products

Botanical products of blue cohosh, goldenseal, and yohimbe were purchased from different commercial online vendors including Etsy, Amazon, and Walmart. The botanical products obtained included powders, capsules, and tablets in a variety of packaging (e.g., bottles, zip-top bags, and foil bags), see SI-Table 1. The variety of products was intended to reflect the variety of options and marketing strategies presented to consumers (Figure 1). We noted various health claims on the packaging including: “…promote healthy mucous membranes.”, “…support healthy digestive and immune system function.”, “…provide menstrual support for women.”, and “… aid in warding off evil, …”.

Figure 1.

Photographs of consumer botanical products purchased labeled as: A) blue cohosh, B) goldenseal, and C) yohimbe.

Chemical Standards

Authentic chemical standards of reported, primary chemical constituents for each botanical were purchased from commercial vendors, namely berberine chloride (Sigma-Aldrich), hydrastine and N-methylcytisine (the nature network), and yohimbine hydrochloride (US Pharmacopeia). The chemical standards were analyzed by NMR and UHPLC-HRMS/MS.

For NMR, the standards were dissolved in 594 μL of [²H]-DMSO with 6 μL of 50 mM sodium trimethylsilylpropanesulfonate (DSS) as the proton chemical shift reference. This solution was vortexed until it became a consistent solution, and 600 μL of that solution was then added to an NMR tube for analysis.

Standards were prepared for UHPLC-HRMS/MS by removing samples from the NMR tubes via glass Pasteur pipette and aliquoted into 1.5 mL microcentrifuge tubes. Twenty (20) μL of each sample were then added to a new 1.5 mL microcentrifuge tube and diluted 1:10 v/v with HPLC grade water. Then 10 μL of the 1:10 diluted sample was added to a new 1.5 mL microcentrifuge tube and diluted 1:60 v/v with 590 μL of HPLC grade water. The 1:600 v/v diluted samples were stored at room temperature in the dark until analysis. Immediately prior to analysis, 50 μL of the 1:600 diluted sample was transferred to a 2 mL amber screw top autosampler vial with a deactivated glass insert and PTFE/silicone septa (Agilent).

Extractions

One gram of each product was weighed in a 15 mL falcon tube. For the different formulations: powdered product was directly weighted, tablet products were crushed, or capsule products were opened, and the exterior discarded. Ten (10) mL of deionized H₂O was added to create the suspension. The suspension was mixed by vortexing until it was a consistent solution, then rocked for 2-3 hours. After extraction, the solution was centrifuged at 5x g for 15 minutes to remove particulate matter. The supernatant was filtered with a 0.45 μm filter and the filtered solution was dried in a speed-vac for 3-4 hours. If the solution was not completely dry, it was frozen at −80°C, and freeze dried in the lyophilizer overnight. Samples were rehydrated with 694 μL of [²H]-PBS (pH~7.4) and 6 μL of 50 mM DSS. The solution was sonicated for 5 minutes then centrifuged for 15 minutes at 20K rpm. The supernatant was then added to an NMR tube. Each botanical product was extracted in triplicate. For the organic extraction, the same methods were used with ethanol as the solvent instead of diH₂O, and the samples were rehydrated with [²H]-DMSO instead of [²H]-PBS.

Chemical Analysis by Nuclear magnetic resonance spectroscopy

NMR spectra were acquired on a 600 MHz Agilent DD2 console using a room-temperature or cryogenically cooled probe. A 1-dimensional NOESY sequence with 4 s acquisition time, 1 s recycle delay, and 100 ms mixing time was used. Chenomx software (Alberta, Canada) was used for processing and spectral analysis. Chenomx compressed the NMR intensity data into 0.04 ppm bins for non-targeted analysis. Metaboanalyst (https://www.metaboanalyst.ca/)was used for statistical comparisons (18, 19).

Chemical Analysis by Mass Spectrometry

Pre-analytical Steps

Following NMR data acquisition, samples (dissolved in [²H]-DMSO or [²H]-PBS) were removed from the NMR tubes via glass Pasteur pipette and added to 1.5 mL microcentrifuge tubes. The samples were centrifuged at 14K rpm for 10 minutes. Twenty (20) μL of each sample were added to a new 1.5 mL microcentrifuge tube and diluted 1:10 v/v with HPLC grade water. Then, 10 μL of the 1:10 diluted sample was added to a new 1.5 mL microcentrifuge tube and diluted 1:60 v/v with 590 μL of HPLC grade water. The 1:600 v/v diluted samples were stored at room temperature in the dark until analysis. Immediately prior to analysis, 50 μL of the 1:600 diluted sample was transferred to a 2 mL amber screw top autosampler vial with a deactivated glass insert and PTFE/silicone septa (Agilent). In addition to the individual samples, a pooled quality control sample was generated by transferring an aliquot of 10 μL of each sample (by botanical) into a 1.5 mL microcentrifuge tube and vortexing to produce a homogenous solution.

Data Acquisition

Samples were analyzed using an UHPLC (Vanquish, Thermo Scientific) coupled to a HRMS (Orbitrap Fusion Tribrid, Thermo Scientific). LC-MS and LC-MS/MS data were acquired. LC-MS data were collected from individual samples (n = 1 injection), system blanks (injection of solvent used to resolubilize samples), and a pooled quality control sample. The pooled quality control (QC) was injected multiple times at different volumes and used in data processing. LC-MS/MS data were collected using AcquireX (Thermo Scientific) deep scan methodology in which pooled QC was injected multiple times (n = 7).

Chromatographic separation was carried out on a F5 analytical column (2.1 x 100 mm, 100 Å, 2.6 μm, Phenomenex) with corresponding guard cartridge. The column was maintained at 30°C during separation with solvent pre-heater. Gradient elution was performed after an initial period of isocratic elution using water with 0.1% acetic acid v/v (A) and acetonitrile with 0.1% acetic acid v/v (B). Separation was performed as follows: 0% B from 0 - 2.0 min, 0% to 100% B from 2.0 to 10.5 min, 100% B from 10.5 to 12.0 min, 100% to 0% B from 12.0 to 13.0 min, 0% B from 13.0 to 20.0 min. The flow rate was 0.5 mL min⁻¹.

Ionization was performed via heated electrospray ionization (NG Ion Max, Thermo Scientific). The source parameters in positive ionization mode were as follows: spray voltage of +4,000 V, sheath gas of 50 arbitrary units (arb), auxiliary gas of 10 arb, sweep gas of 1 arb, ion transfer tube at 325°C, vaporizer at 350 °C. The source parameters used in negative ionization mode were identical except for the spray voltage of −3,000 V. Data collected in positive or negative ionization mode are labeled as MS+ or MS−, respectively.

MS and MS/MS data were collected with an anticipated LC peak width of 8 s and a default charge of 1. EASY-ICT^M (Thermo Scientific) was installed and used during data collection; a lock-mass was measured, concurrently to experimental measurement, and used for instrument mass calibration. MS data were acquired at 120,000 resolutions from m/z 100-1000 with an RF lens of 60% and maximum injection time of 50 ms. MS/MS data were acquired at 30,000 resolutions using an isolation width of 1.5 (m/z), stepped assisted HCD (energy steps were 20, 35, and 60, and a maximum injection time of 54 ms. The inclusion list was generated and updated via AcquireX with a low and high mass tolerance of 5 ppm. An intensity filter was applied via the “Intensity” node in the workflow with an intensity threshold of 2.0 x 10⁴. The “Dynamic Exclusion” node was used with the following parameters: exclude after n = 3 times, if occurs within 15 s, exclusion duration of 6 s, a low mass tolerance of 5 ppm, a high mass tolerance of 5 ppm, and excluding isotopes.

Data Processing and Analysis

Raw data were processed using Compound Discoverer 3.3.550 (Thermo Scientific), which was used to identify chemical features (unique set of m/z and retention time value) in the dataset. The resulting feature table was exported from Compound Discoverer and was processed using in-house R code to reformat the data, remove features of poor measurement (based on pooled quality controls), and output processed tables used in subsequent statistics and interpretation. All raw data are publicly available in the Chemical Effects in Biological Systems (CEBS) database at https://cebs-ext.niehs.nih.gov/cebs/paper/15729/private/MTExODdkMWJlZjMyN2QyNjRlMGY3YTAzZmM0OGNmOGYK.

Results

Identical methods of extraction were applied to the blue cohosh, goldenseal, and yohimbe samples as described in the methods. The similarities and differences among the botanical products will be presented first using a principal component analysis (PCA). For the NMR data, the input vector was the NMR spectral intensities, compressed to 0.04 ppm bins. For the MS positive and negative data, the input vector was the log₁₀ transformed feature table. The MS data and NMR data were combined using a mid-level data fusion approach (8). This approach analyzed the top five principal components (i.e., latent variables) of the NMR and MS analyses in order to combine the disparate datasets. This fusion analysis can lead to further differentiation of the botanical products if the two methods capture different chemical constituents which distinguish the products. After visual analysis of product clustering using the first 2 PCA scores and PERMANOVA comparisons to identify outliers, this was followed by an analysis of the chemical features that led to differentiating the products. Either the organic or aqueous phase extract data are presented in the main document and the other recorded in supplemental, depending on which provided more differentiation.

Blue cohosh

The PCA of blue cohosh sample extracts (Figure 2) revealed that total variance among samples appeared to be due largely to the type of product: clustering among subgroups was observed (BC-2, BC-4, BC-5, BC-7, and BC-8), but separation between subgroups was also observed (BC-1 and BC-3, BC-6, and BC-9). Variance was small between replicate extractions of the same product, and the differences observed were between products. The permuted ANOVA (PERMANOVA) statistic confirmed that there were significant outlier groups in the data (Table 1). The tablet products (BC-1 and BC-3) were distinct from the powder and capsule products being observed at the far right of Figure 2A. A representative example of spectra from the main group appears in Figure 3A. The corresponding loading plot (SI-Figure 1A) indicates that these observed differences are primarily driven by chemical shifts 3.54, and 3.7 ppm. Figure 3B shows a representative spectrum from this group and the data indicate an abundance of sucrose and maltose. Figure 3I confirms that the MS data revealed high concentrations of a disaccharide in these two products, see below.

Figure 2.

PCA score plots for (A) NMR, (B) MS+, (C) MS−, and (D) mid-level fused data for Organic Blue Cohosh. Points are colored according to product identifier and shape corresponds to formulation.

Table 1.

The permuted ANOVA (PERMANOVA) statistic

Botanical	Preparation	Data Type	p-value	F value
Blue Cohosh	Aqueous	MS+	0.001	12.1
Blue Cohosh	Aqueous	MS−	0.001	10.2
Blue Cohosh	Aqueous	NMR	0.002	8.1
Blue Cohosh	Aqueous	Fused	0.001	18.0
Blue Cohosh	Organic	MS+	0.001	45.0
Blue Cohosh	Organic	MS−	0.001	20.7
Blue Cohosh	Organic	NMR	0.001	10.4
Blue Cohosh	Organic	Fused	0.001	67.0
Goldenseal	Aqueous	MS+	0.001	46.1
Goldenseal	Aqueous	MS−	0.001	34.1
Goldenseal	Aqueous	NMR	0.005	11.6
Goldenseal	Aqueous	Fused	0.001	495.9
Goldenseal	Organic	MS+	0.001	37.0
Goldenseal	Organic	MS−	0.001	29.7
Goldenseal	Organic	NMR	0.005	9.6
Goldenseal	Organic	Fused	0.001	370.8
Yohimbe	Aqueous	MS+	0.001	30.9
Yohimbe	Aqueous	MS−	0.001	20.1
Yohimbe	Aqueous	NMR	0.014	4.3
Yohimbe	Aqueous	Fused	0.001	153.6
Yohimbe	Organic	MS+	0.001	7.3
Yohimbe	Organic	MS−	0.001	13.0
Yohimbe	Organic	NMR	0.051	2.8
Yohimbe	Organic	Fused	0.001	16.8

Figure 3. NMR Spectra of Organic Blue Cohosh and Detection of Reported Phytochemicals Present in Blue Cohosh.

(A-C) NMR spectra from three different patterns observed upon visual inspection. Red annotation for Sucrose and Blue for Maltose. (D) MS+ feature with m/z 343.16975 at 4.573 min, annotated as Magnoflorine. Note, the presence of an outlier sample in BC-6. (E) MS+ feature with m/z 205.13349 at 3.579 min, annotated as N-methylcytisine. (F) MS− feature with m/z 767.45809 at 5.47 min, annotated as Cauloside C. (G) MS− feature with m/z 307.11746 at 4.873 min, annotated as Cimifugin. (H) MS− feature with m/z 118.0862 at 0.514 min, annotated as Betaine. (I) MS− feature with m/z 343.12334 at 0.532 min, annotated as Lactose. Box indicates first and third quartiles and whiskers indicate 1.5 times the inter-quartile range. Median is displayed. Individual data points are indicated.

In the MS+ data (Figure 2B), the PCA score plot indicates that the same brands cluster together (BC-2, BC-4, BC-5, BC-7, and BC-8), while three subgroups appear chemically different (BC-1 and BC-3, BC-6, and BC-9) along PC1. The PCA loading plot (SI-Figure 1B) indicated several important features that contribute to the observed grouping in the PC score plot along PC1, namely features annotated as magnoflorine, N-methylcytisine, and betaine. Qualitative comparisons of representative compounds are shown in Figure 3. Greater levels of magnoflorine were observed in BC-2, BC-4, BC-5, BC-7, and BC-8 products (Figure 3D), in agreement with the grouping of samples in Figure 2B. An alpha/beta – disaccharide was found to be highly prevalent in BC-1 and BC-3 of which both are tablet products, in agreement with the subgrouping observed in PCA. While annotated as lactose, the feature is more appropriately annotated as a disaccharide given that many disaccharides have the same mass.

In Figure 2C, PCA of blue cohosh MS− data revealed a main group with a few smaller groups (BC-1 and BC-3, BC-6, and BC-9). These same PCA observations were seen in both NMR and MS+ PCAs. The PCA loading plot indicated the features important in distinguishing the products from one another and includes alpha/beta – disaccharide and betaine as observed in SI-Figure 1B, Figure 3H. Betaine contributes to the observed separation of samples along the PC2 axis with the highest levels observed in BC-6. Betaine is found in many plants including what, beets, and spinach (20). As observed in the MS+ data, disaccharide signals (e.g., sucrose and lactose) differentiated the products BC-1 and BC-3.

The MS+, MS−, and NMR data were combined via mid-level data fusion resulting in the PCA plot in Figure 2D. The totality of chemical information obtained for each sample is represented by a point in the score plot and the source of the chemical information indicated in the loading plot (SI-Figure 1). As in each data set individually, we observed a main group of products and subgroups which contain BC-1, BC-3, BC-6, and BC-9.

A distinct pattern was noted in the NMR and MS data for BC-9. In the NMR spectrum of BC-9 (Figure 3C), the signals between 7 – 9 ppm were absent in comparison to the representative spectrum of the majority of samples (Figure 3A). The MS data supported the presence of cauloside C (Figure 3F) and N-methylcytisine (Figure 3E), indicative of blue cohosh. However, there were additional phytochemical constituents annotated, such as cimifugin (Figure 3G) and ferulic acid (not shown), that were uniquely present in BC-9. Cimifugin and ferulic acid are known constituents of black cohosh (Actaea racemosa, formerly known as Cimicifuga racemosa) (21).

Aqueous phase extracts produced less distinct grouping of blue cohosh products in the MS and fused data compared to the organic phase extractions; compare Figure 2 with SI-Figure 2 and Figure 3 with SI-Figure 3.

Goldenseal

PCA of the NMR data (Figure 4A) differentiated the products into three distinct subgroups. Products GS-2, GS-5, GS-6, GS-7, and GS-8 were observed in one group, products GS-1, GS-3, and GS-4 were observed in another group separated along PC1, and the third group comprised of only product GS-9 was distinguished from all others along PC1. This observation is due to the quantitative measurement of berberine along PC1. Figure 5 panel A shows an example of the GS-1, GS-3, and GS-4 product subgroup, which has a high amount of berberine (red line), but hydrastine was not detected (green line) as can be seen in panel B. Note that the y-axis scale of panel B is 10 times larger than panel A, which shows that the berberine levels are much lower in the outlier products. Product GS-9 is an outlier because very few chemicals could be extracted, suggesting the composition was primarily filler.

Figure 4.

PCA score plots for (A) NMR, (B) MS+, (C) MS−, and (D) mid-level fused data for Organic Goldenseal. Points are colored according to product identifier and shape corresponds to formulation.

Figure 5. NMR Spectra of Organic Goldenseal and Detection of Reported Phytochemicals Present in Goldenseal.

(A-B) NMR spectra from three different patterns observed upon visual inspection. Red annotation for berberine and green for hydrastine. (C) MS+ feature with m/z 384.1438 at 5.066 min, annotated as Hydrastine. Note, the presence of an outlier sample in BC-6. (D) MS+ feature with m/z 338.13868 at 5.463 min, annotated as Berberine. (E) MS− feature with m/z 354.1698 at 5.763 min, annotated as Palmatine. (F) MS− feature with m/z 340.15408 at 5.707 min, annotated as Canadine. Box indicates first and third quartiles and whiskers indicate 1.5 times the inter-quartile range. Median is displayed. Individual data points are indicated.

PCA of goldenseal MS+ data (Figure 4B) largely confirmed the NMR data with the same products organized into three distinct subgroups. However, intra-product variance was relatively small with tight groupings of replicate samples. The PCA loading plot (SI-Figure 4B) indicated several important features that contribute to the observed grouping in the PC score plot along PC1, namely features annotated as hydrastine and canadine. The negative mode data acquired from the organic extract of goldenseal products mirrored that of the positive mode data. The same subgrouping of products is noted in the PCA score plot, Figure 4C. Additionally, we note that all of the outliers were capsule products, but the main group included both capsules and powders.

Upon examining the PCA plot of the data fusion of the NMR and MS data, displayed in Figure 4D, products were evident in 4 subgroups, driven by the combination of distinguishing data from MS and NMR. The notable difference is the clear distinction of GS-4 as a subgroup based on NMR data indicating quantitatively more berberine than GS-1 and GS-3 (SI-Figure 5), illustrative of how the combination of multiple data sources is beneficial to understand the chemical composition of complex botanical products more comprehensively. Indeed, the PERMANOVA F value increases significantly for the data fusion approach compared to the individual measurements demonstrating better recognition of outliers in the data (Table 1).

Looking at the specific chemical constituents observed by MS, greater levels of hydrastine were observed in products GS-2, GS-5, GS-6, GS-7, and GS-8 (Figure 5C) and were contrastingly low in product GS-1, GS-3, and GS-4. Berberine was observed in lower levels in products GS-1, GS-3, GS-4, and GS-9. Product GS-9 was lower in berberine signal, but contained hydrastine, palmatine, and canadine at a slightly lower level compared to products GS-2, GS-5, GS-6, GS-7, and GS-8 (Figure 5C-F). Products GS-1, GS-3, and GS-4 have disproportionately less hydrastine and canadine, than berberine and palmatine which suggests that these products may contain filler spiked with berberine.

Figures SI-6 and SI-7 of the aqueous extracts mirror the results and conclusions above for the organic extracts.

Yohimbe

When analyzing the yohimbe products, there were subtle differences between the organic and aqueous phase extracts. PCA of NMR data, Figure 6A, revealed aqueous yohimbe products Y-1, Y10, and Y-12 were differentiated from the other products along PC1, which Y-9 and Y-11 were differentiated from other products along PC2 (Figure 6A). Of note, product Y-10 showed high variability and we noticed that it had a very high viscosity. From the NMR, we posit that the aqueous extraction for Y-10 contained very high levels of maltose; the signal of which overwhelmed other peaks.

Figure 6.

PCA score plots for (A) NMR, (B) MS+, (C) MS−, and (D) mid-level fused data for Aqueous Yohimbe. Points are colored according to product identifier and shape corresponds to formulation.

Furthermore, the MS+/− PCA plots revealed similar outliers along PC1 that were observed in the NMR data; however, a different grouping pattern was observed (Figure 6B,6C). The subgroupings appear to be clustered more tightly along PC2 compared to those from NMR data (Figure 6A). Contrastingly, the outliers were separated by much larger principal component values compared to the NMR PCA (Figure 6). Additionally, a third subgroup, Y-5, appeared to drift from the main group. According to the loading plot (SI-Figure 8) corynathine, which is a known constituent of yohimbe and an isomer of yohimbine, was a primary driver of the outliers Y-1 and Y-12. The aqueous extract of yohimbe products negative mode data mirrored that of the positive mode data. The same subgrouping of products (Y-1, Y-5, and Y-12) is noted in the PCA score plot, Figure 6C. Although, it was not the same differentiating factors as the positive mode. According to the loading plot (SI-Figure 8C), stachyose (more appropriately described as a tetrasaccharide) was annotated to be the driving factor in the negative mode.

The comparison of the variance in the data fusion approach suggested that 9 of the products belong in one primary group, while products Y-1 and Y-12 are significant outliers, which was only partially apparent looking at NMR and MS separately (Figure 6D). According to the loading plot (SI-Figure 8D), distinct principal components from the data fusion were the differentiating factors for Y-1 and Y-12, respectively. This is a good example of where the different chemical analyses in the data fusion approach contributed in different degrees to the principal components. This helped identify which products are outliers. Table 1 confirms the visual interpretation, via the PERMANOVA, that there are significant outliers and the largest F value is found for the data fusion of the aqueous extracts. The organic phase extracts were not as useful for clustering as can be seen in the smaller F-values. This is visually apparent in SI-Figure 9 and 10. Similar to goldenseal, we observed that all of the outliers were capsule products; however, the main group included both capsules and powders.

Additionally, in order to understand possible chemical drivers of Y-1 and Y-12 differences, we first looked at the organic phase extract for yohimbine and yohimbine was clearly found in all organic extracts (data not shown). All aqueous products clearly contain yohimbine in various proportions and it is notably higher in products Y-1 and Y-12 (Figure 7C). Corynanthine is also at a higher level in product Y-12 compared to the rest of the products (Figure 7D). Products Y-1 and Y-12 also contain a surprising quantity of maltose, which is disproportionately high, compared to all the other products purchased (Figure 7E). However, in looking at the NMR spectra of the aqueous extracts some clear differences were apparent. For example, products Y-2 through Y-12 contain quinic acid, which is a known component of bark, which is the typical source of yohimbe products (Figure 7A). In product Y-1 this ingredient was not detected in the NMR spectra because the signal is swamped by the high levels of sugars; note again the y-axis scale in Figures 7A and B. However, it was putatively detected by MS in all yohimbe products (Figure 7F). In the case of the yohimbe analysis, it was useful to utilize both NMR and MS to better characterize the chemical constituents.

Figure 7. NMR Spectra of Aqueous Yohimbe and Detection of Reported Phytochemicals Present in Yohimbe.

(A-B) NMR spectra from three different patterns observed upon visual inspection. Blue annotation for Quinic Acid and orange for Maltose. (C) MS+ feature with m/z 355.20133 at 4.991 min, annotated as Yohimbine. Note, the presence of an outlier sample in BC-6. (D) MS+ feature with m/z 355.20136 at 5.608 min, annotated as Corynanthine. (E) MS− feature with m/z 325.11309 at 3.43 min, annotated as Maltose. (F) MS− feature with m/z 193.07069 at 0.61 min, annotated as Quinic Acid. Box indicates first and third quartiles and whiskers indicate 1.5 times the inter-quartile range. Median is displayed. Individual data points are indicated.

Discussion

Sales of botanical dietary supplements in the United States totaled approximately 12.35 billion in 2021 (1). According to the NIH Dietary Supplement Label Database, there are currently over 25,000 products on the market containing botanical ingredients (22). The current regulatory structure for botanicals and other dietary supplements in the United States, as detailed in the Dietary Supplement Health and Education Act of 1994, does not require pre-market safety and quality documentation (23). Instead, the burden of proof for adulteration or lack of safety falls on the Food and Drug Administration. Considering this regulatory landscape that favors access to dietary supplements over stringent quality and safety requirements, it is not surprising that there have been numerous reports of adulterated botanical products in the marketplace (24, 25).

In the current work, we observed chemical differences between botanical products with identical labels via the non-targeted chemical analysis workflow that utilized both NMR and MS. Non-targeted chemical analysis is particularly useful for complex mixtures that have a large unidentified fraction (3), such as botanical ingredients in dietary supplements. The non-targeted analysis workflow was informative without complete annotation or identification of all features. Several known constituents from each botanical product were used to confirm the identification of purported bioactive markers and compare their relative quantities across samples. A further benefit of the presented workflow is the ability to determine the similarity and differences, chemically, between products for which there is no standard reference material available. Outliers could be identified visually via the PCA analysis and confirmed statistically with the PERMANOVA.

Both NMR and MS were able to identify constituent differences between products, particularly those that did not resemble the majority of purchased botanicals. In the case of products labeled as blue cohosh, there were 3 subgroups of outliers (BC-1 and BC-3, BC-6, and BC-9). BC-1 and BC-3 were tablet formulations that contained primarily sugar. The presence of sugar molecules likely reflects the homeopathic tablet formulation of these two products. Homeopathy posits that miniscule quantities of an ingredient that would cause toxicity at higher doses can mitigate existing disease, and sugars including lactose and saccharose, observed in the current evaluation, are common homeopathic diluents (26). This finding highlights the importance of formulation in comparing across samples available in the marketplace. While BC-6 had low levels of black cohosh constituents (magnoflorine, N-methylcytisine, and cauloside C), it was the only sample to contain betaine, which could indicate the presence of substitute plant material since betaine is found in high quantities in other plants (20). The BC-9 sample clearly contained compounds not known to be present in blue cohosh, but which are known to be in black cohosh (i.e., cimifugin and ferulic acid), a completely unrelated plant. While multiple cases of black cohosh adulteration with other Asian cohosh species have been reported (25), to our knowledge, this is the first report finding blue cohosh adulteration with black cohosh.

Interestingly, the goldenseal products displayed a different pattern of distinguishing features. In the case of goldenseal, several of the products had lower signals for berberine, hydrastine, and canadine (relative comparisons), which are benzylisoquinoline alkaloids that are believed to be responsible for bioactivity of the botanical (27). Products with low levels of these marker constituents could reflect harvesting conditions (27), or adulteration with additional filler material, which has been observed in another survey of goldenseal quality (28). In the case of the yohimbe products the NMR data were overwhelmed by the levels of sugar filler, but the MS data helped identify that none of the products appear inauthentic. The various levels of filler and yohimbine related compounds provided strong differentiation of the outliers.

In terms of utilizing aqueous versus organic extracts for clustering, very similar conclusions were found for goldenseal. This was despite the fact that for goldenseal, the important ingredients hydrastine and berberine were found primarily in the organic phase. In contrast, it was insightful to look at both extraction methods in analyzing the blue cohosh and yohimbe products. The differentiation of the products in the organic phase was relatively weaker, looking at the size of the principal components. However, it was the identification of sugar and quinic acid in the aqueous phase that clearly distinguished 2 products as significant outliers. Hence, when dealing with other botanicals in the future, it is strongly encouraged to look at both phases.

When comparing the effectiveness of NMR versus MS, there is a vast literature on the choice for metabolomics analysis. NMR is orders of magnitude less sensitive than MS, and therefore identifies fewer molecules. In the case of goldenseal, the hydrastine levels were undetected in several products by NMR and in the case of yohimbe, the filler concentrations swamped identification of quinic acid, a key constituent of bark. However, both techniques showed very similar patterns in their PCA profiles with some subtleties in differentiating outliers. Certain key points were easier to examine quickly with one technique or the other. For example, it was facile with NMR to identify that the tablets contained primarily sucrose or maltose, whereas non-targeted MS analysis does not differentiate similar mass structures like these disaccharides. It was also easy to identify quinic acid by NMR which is a fairly common metabolite. In contrast MS performed better when the sample contained a plant product not advertised on the label. Here MS was able to identify cimifugin, usually found in black cohosh, in the blue cohosh extracts. Therefore, we view the combination of NMR and MS methods as advantageous in that they can confirm findings and complement one another.

One limitation of the current study is that the product selection contained only dry products, which were easy to handle and used a nearly identical extraction procedure. The botanical market also frequently includes tinctures. In some preliminary experiments (data not shown) we examined a few tinctures and found various solvents and high concentrations of excipients like glycerol. The high concentration of these substances made the samples difficult to analyze. Similarly, Yohimbe product 10 was difficult to analyze due to the high concentrations of maltose. Future studies will need to develop methods for addressing these challenges. It is possible that other -omics technologies may be useful in this regard including genomics as well as other statistical techniques (29).

In conclusion, the NMR and MS non-targeted chemical analysis workflow using aqueous and organic extracts was successful in measuring chemical similarity and differences in botanical products. Of the products tested, one blue cohosh product is suspected to be inauthentic (i.e., contaminated with black cohosh). In general, products formulated as capsules had the greatest variability among products with the same labeling. We do not believe this to be due to the capsule formulation, but rather due to variation in the botanical source material that went into the capsules. In contrast, the powdered products appeared to be the most consistent. Our results demonstrate, at the proof-of-concept level, that our workflow is useful in characterizing and comparing across botanical products.

Supplementary Material

EBE1F82774A45558D0965D0BE17382AC

SI-Figure 6. PCA score plots for (A) NMR, (B) MS+, (C) MS−, and (D) mid-level fused data for aqueous goldenseal. Points are colored according to product identifier and shape corresponds to formulation.

395B28F3AFE14A9AA21457337426611D

SI-Figure 4. PCA loading plots for (A) NMR, (B) MS+, (C) MS−, and (D) mid-level fused data for organic goldenseal.

713BAE781153249EBDF410ED65985B10

SI-Figure 9. PCA score plots for (A) NMR, (B) MS+, (C) MS−, and (D) mid-level fused data for organic yohimbe. Points are colored according to product identifier and shape corresponds to formulation.

0679D922C3C275DD3045AB5C118EB990

Figure 8. Loading plots for (A) NMR, (B) MS+, (C) MS−, and (D) mid-level fused data for aqueous yohimbe.

50BA1166C06D2E40B65201F9BF7F8D48

SI-Figure 7. PCA loading plots for (A) NMR, (B) MS+, (C) MS−, and (D) mid-level fused data for aqueous goldenseal.

8FBF5EBC74405519CA68FB9E865C1ABB

SI-Figure 10. PCA loading plots for (A) NMR, (B) MS+, (C) MS−, and (D) mid-level fused data for organic yohimbe.

34F9528D47500A327C73969B57330D2C

SI-Figure 5. Bar graph denoting Berberine and Hydrastine content levels in organic goldenseal products.

5F322E1FBB70BD0CC4CC5D37B0A46B3D

SI-Figure 3. PCA loading plots for (A) NMR, (B) MS+, (C) MS−, and (D) mid-level fused data for aqueous blue cohosh.

0D3BD82147005867B358462C2233C05B

SI-Figure 2. PCA score plots for (A) NMR, (B) MS+, (C) MS−, and (D) mid-level fused data for aqueous blue cohosh. Points are colored according to product identifier and shape corresponds to formulation.

BA8160CCEC9765C67C330874DC93618A

SI-Figure 1. PCA loading plots for (A) NMR, (B) MS+, (C) MS−, and (D) mid-level fused data for organic blue cohosh.

50C3B9608CAD474AB987B3E81C8D3134

SI-Table 1. Table of consumer botanical products purchased

Abbreviations:

NMR

Nuclear Magnetic Resonance

MS

Mass Spectrometry

UHPLC-HRMS/MS

Ultra-High Performance Liquid Chromatography-High Resolution MS/MS

PCA

Principal Component Analysis

DSS

sodium trimethylsilylpropanesulfonate

DMSO

Di-Methyl Sulfoxide

PTFE

Polytetrafluoroethylene

PBS

Phosphate-Buffered Saline

NOESY

Nuclear Overhauser Effect SpectroscopY

LC-MS/MS

Liquid Chromatography-MS/MS

MS/MS

Tandem mass spectrometry

HCD

Higher-energy collision-induced dissociation

QC

Quality Control

MS+

Mass Spectrometry Positive Mode

MS−

Mass Spectrometry Negative Mode

BC

Blue Cohosh

GS

Goldenseal

Y

Yohimbe

CYP

Genes encoding cytochrome P450 enzymes

m/z

Mass-to-charge ratio

rpm

Revolutions per minute