Principal Component Analysis Followed by Mahalanobis Distance Classification for Authentication of Vaccinium Berry Ingredients and Supplements via LC-MS/MS-based Anthocyanin Fingerprinting

Abstract

The rising popularity of Vaccinium berry dietary supplements for their antioxidant and anti-inflammatory benefits has raised concerns about mislabeling and adulteration. Traditional authentication methods often compare anthocyanin profiles to a certified reference material, often overlooking variances from different cultivars, environments, and cultivation practices. Thus, a more comprehensive approach is imperative. This study developed a chemometric approach using anthocyanin profiles to distinguish bilberry (Vaccinium myrtillus L.), blueberry (Vaccinium corymbosum L.), and cranberry (Vaccinium macrocarpon Aiton) from one another and from potential adulterants. Anthocyanin fingerprints from 48 Vaccinium and non-Vaccinium samples were generated via LC-MS/MS due to its ability to rapidly quantify low-level analytes across diverse sample matrices. Principal component analysis (PCA) was applied to the relative abundance ratios of 18 selected anthocyanins, followed by a Mahalanobis Distance Classification model for classifying unknown samples with a decision boundary approach. By using voucher information and high-performance-thin layer chromatography (HPTLC) test results, the model successfully classified 25 authentic Vaccinium ingredients, non-Vaccinium ingredients, and Vaccinium-containing supplements with 100% accuracy in a verification study. Among the four dietary supplements tested, three were correctly labeled, while one V. myrtillus product was determined to be adulterated, as confirmed by HPTLC analysis. While the number of reference samples was constrained by marketplace availability, the model demonstrates strong initial potential for authentication, providing a foundation for future validation using larger data sets. To the best of our knowledge, this is the first study to authenticate three high-value Vaccinium species in dietary ingredients and supplements using anthocyanin fingerprints and chemometric methods. This approach complies with FDA cGMP 21 CFR Part 111 and targets to meet AOAC (Association of Analytical Collaboration International) 2014.007 Standard Method Performance Requirements, showing promising potential to be adopted as a consensus method for Vaccinium authentication.

Introduction

Vaccinium is a genus of terrestrial shrubs widely distributed in the Northern Hemisphere, with more than 400 species. Within the genus, bilberry (Vaccinium myrtillus L.), blueberry (Vaccinium corymbosum L.), and cranberry (Vaccinium macrocarpon Aiton) hold high economic value. Vaccinium berries are generally high in polyphenolic compounds, especially anthocyanins, leading to their high antioxidant capacity. As a result, Vaccinium fruits have historic use in folk medicine and are widely consumed as food. Modern clinical studies support many therapeutic claims of Vaccinium species, such as prevention or treatment of cardiovascular diseases, diabetes, obesity, cancer, urinary tract infections, and aging diseases. −

A recent spike in demand for dietary ingredients and dietary supplements containing Vaccinium species has resulted in quality concerns. In many cases, blueberry products are intentionally and unintentionally marketed as bilberry. Cranberry juice or extracts are fully or partially replaced by cheaper anthocyanin sources such as grape seed, black rice and mulberry extracts. In a 2016 study, Lee et al. reported that over 30% of investigated Vaccinium products differed from the claimed ingredient based on anthocyanin profiles. Gasper et al. (2021) reported that 45% of bilberry-containing dietary supplements did not pass the authentication screen. Multiple researchers have reported that up to 24% of extracts and 70% of Vaccinium supplements were mislabeled in terms of species composition or active ingredient dosage. , Although most adulteration or mislabeling poses limited major risk for consumers’ health, some may cause severe health hazards due to toxic or allergic substances and can result in a loss of intended therapeutic effects. Allergic reactions are a common concern regarding berry adulteration, since consumers may have reactions to some, but not all berries.

In several studies, adulteration or mislabeling of Vaccinium products has been detected by comparing the sample’s anthocyanin profile to that of authentic materials. Anthocyanins are naturally occurring water-soluble phenolic compounds responsible for the red, blue, and violet colors in various fruits, vegetables, and grains. The chemical structure of anthocyanins varies depending on the type of aglycone (anthocyanidin) and the type, number, and position of glycosylation and acylation. To date, more than 700 distinct anthocyanins have been identified.

The variations in the anthocyanin composition among Vaccinium and non-Vaccinium species can be primarily attributed to inherent genetic differences. For example, bilberries contain a diverse range of anthocyanins, including delphinidin, cyanidin, petunidin, peonidin, and malvidin derivatives. Conversely, cranberries have a distinct anthocyanin profile primarily comprising peonidin, cyanidin, and malvidin derivatives. Non-Vaccinium species, such as elderberry (), blackberry (Rubus subg. rubus), and chokeberry (Aronia), also contain high levels of various anthocyanins. Besides interspecies differences, intraspecies variations have also been observed, influenced by factors such as cultivar, maturity, and geographic region. For instance, bilberry anthocyanin accumulation increased with altitude, particularly for delphinidin and malvidin glucosides. Wild blueberry can have 1.7 times higher total anthocyanin content per berry mass than cultivated ones. Even adverse environmental conditions affect the anthocyanin profiles. For example, levels of delphinidin, cyanidin, and petunidin-derivatives in blueberry increased with mechanical injury or pest infestation, while the levels of malvidin-derivatives decreased. Furthermore, various manufacturing processes, like extraction and purification, may result in hydrolysis and degradation of anthocyanins. In this context, relying solely on baseline anthocyanin profiles in a certified reference material may not be sufficient for reliable identification of Vaccinium products, and considering seasonal, temporal, and geographic variation is essential for a robust authentication study.

One strategy to address this challenge is using a combination of analytical techniques, or an orthogonal approach, to obtain a more comprehensive understanding of the chemical composition of targeted samples. This synergistic approach can provide robust insight into the chemical relationships within species. One such orthogonal approach begins with generating an in-silico database containing anthocyanin profiles from various authentic Vaccinium samples, encompassing different cultivars, growing regions, and environmental conditions. New samples can then be compared to the reference library rather than a single reference sample, to determine their similarities to known profiles. Principal Component Analysis (PCA) is an unsupervised chemometric method that aids in identifying potentially mislabeled samples by analyzing inter- and intraspecies chemical variations following dimensionality reduction.

Instrument selection is a major factor when developing authentication schemes, for example our group has found that combining HPTLC and HPLC-PDA (Photo Diode Array) is valuable for elderberry authentication, while an AMS (Accelerator mass spectrometry) and HPLC-PDA combination is preferable for turmeric’s natural origin determination. HPLC-PDA is a powerful technique that is frequently used for industrial consensus methods. , However, in the context of this study, LC-MS/MS provides numerous benefits over HPLC-PDA because of its high sensitivity and selectivity, enabling the detection of specific anthocyanins even in complex mixtures or at very low concentrations. This is particularly crucial when dealing with dietary supplement samples. Second, LC-MS/MS can accurately distinguish closely related compounds with similar UV spectra, thereby contributing to a more comprehensive understanding of the anthocyanin profile. Integrating LC-MS/MS as an analytical tool contributes to the reliability and robustness of the chemometric approach, and optionally verifying results by HPTLC provides an additional layer of false-negative prevention.

Up to now, only a limited number of studies have focused on detecting adulteration of Vaccinium containing dietary supplements, and a general analytical approach for authentication of different Vaccinium species is yet to be published. Recent efforts have been made to enhance the quality of dietary supplements in the United States. AOAC (Association of Analytical Collaboration International) has established the Standard Method Performance Requirements (SMPR 2014.007) that require analytical methods that effectively authenticate Vaccinium ingredients in dietary supplements. Our goal was to develop a chemometric approach to differentiate anthocyanin profile of selected (target) Vaccinium species from excluded (nontarget) species and evaluate anthocyanin-containing dietary ingredients and supplements. In combination with a post hoc classification step using Mahalanobis distance decision boundaries, we applied PCA to a preselected group of anthocyanins for each authentic sample and known adulterants, eliminating the need to identify additional marker compounds for each potential adulterant. The present study builds a primary method on LC-MS/MS, utilizes known samples as a training data set to define and calibrate decision boundaries for in-silico classification, tests unknown samples for distinguishing authentic samples from potential adulterants, and optionally verifies sample classification with HPTLC as an orthogonal approach for adulteration confirmation. Future applications of this method hold the potential to be adopted as a universal quality control method, meeting the botanical identification requirements of the United States 21 CFR Part 111 FDA cGMP (Current Good Manufacturing Practices) regulation for dietary supplements.

Materials and Methods

Chemicals and Reagents

Cyanidin-3-glucoside chloride and other anthocyanin reference standards were donated by ChromaDex, Inc. (Irvine, CA). LC/MS grade methanol, acetonitrile, and Optima water were purchased from Fisher Scientific (St. Louis, USA). LC/MS grade formic acid and HPLC grade trifluoroacetic acid (TFA) were purchased from Thermo Scientific (St. Louis, USA) and Sigma-Aldrich (St. Louis, USA), respectively.

Sample Collection

Considering AOAC SMPR 2014.007 and material availability, a total of 50 samples were included in this study. A full sample list is shown in Supplemental Tables S1. Samples used at the method development and model training stages were botanical reference materials (BRMs) purchased from or donated by certified vendors such as US Pharmacopeia (USP) (Rockville, MD, USA), Extrasynthese (Grasse, France), ChromaDex (Irvine, CA, USA), American Herbal Pharmacopoeia (AHP) (Scotts Valley, CA, USA), Layn (Guangxi, USA), Starwest (Sacramento, CA, USA), and Jiaherb (Shaanxi, China). The identification of in-house samples were verified via a selective HPTLC method comparing the sample to reference material of the claimed species. Samples employed during the method verification stage were commercially available dietary supplements or ingredients obtained from a local market. The dietary supplements selection adhered to the requirements outlined in AOAC SMPR 2014.007, ensuring they contained a single Vaccinium species in diverse forms, such as concentrate, extract, syrup, and juice. Supplement brand information was withheld to prevent any potential bias during the analysis.

Among all 50 samples, the authenticity panel (target group) was consisted of Northern highbush blueberry (V. corymbosum L.) (n = 5), American cranberry (V. macrocarpon Aiton) (n = 13), and bilberry (V. myrtillus L.) (n = 11). The nontargeted (exclusion) species group included black rice (Oryza sativa L.) (n = 2), aronia (Aronia melanocarpa (Michx.) Elliott) (n = 3), European elderberry ( L.) (n = 3), acai (Euterpe oleracea Mart.) (n = 5), black soybean ( L. Merr.) (n = 1), pomegranate (Punica granatum L.) (n = 1), blackberry (Rubus spp.) (n = 1), grape ( L.) (n = 1), which were the most common adulterants of Vaccinium products reported in the literature. Four dietary supplements (n = 4) containing single Vaccinium species or a nontarget species were purchased online or in local market.

For verification study, authentic target ingredient sample information can be found in Table , nontarget ingredient sample information can be found in Table , and dietary supplement sample information can be found in Table . Samples not listed in these three tables were used as training data set to establish the classification model.

1. List of Target Verification Samples and Their Botanical Identification Results by the Classification Model.

Claimed Species	Sample Name	Sample Code	Mahalanobis Distance Classification Result	Correctly labeled by manufacturer?	Correctly identified by the classification method?
V. corymbosum extract powder	Blueberry BRM 3	B5	V. corymbosum	Yes	Yes
V. macrocarpon extract powder	Cranberry BRM 4	C4	V. macrocarpon	Yes	Yes
V. myrtillus extract powder	Bilberry BRM 6	A9	V. myrtillus	Yes	Yes

2. List of Nontarget Verification Samples and Their Botanical Identification Results by the Classification Model.

Claimed Species	Sample Name	Sample Code	Mahalanobis Distance Classification Result	Correctly labeled by manufacturer?	Correctly identified by the classification method?
O. sativa extract powder	Black rice 1	N1	Nontarget group	Yes	Yes
O. sativa extract powder	Black rice 2	N2	Nontarget group	Yes	Yes
A. melanocarpa dry powder	Chokeberry 1	N3	Nontarget group	Yes	Yes
A. melanocarpa dry powder	Chokeberry 2	N4	Nontarget group	Yes	Yes
V. vinifera extract powder	Grape seed and skin	N5	Nontarget group	Yes	Yes
S. nigra extract powder	Elderberry 1	N6	Nontarget group	Yes	Yes
S. nigra extract powder	Elderberry 2	N7	Nontarget group	Yes	Yes
S. nigra extract powder	Elderberry 3	N8	Nontarget group	Yes	Yes
E. oleracea extract powder	Acai 1	N10	Nontarget group	Yes	Yes
E. oleracea extract powder	Acai 2	N11	Nontarget group	Yes	Yes
E. oleracea extract powder	Acai 3	N12	Nontarget group	Yes	Yes
E. oleracea extract powder	Acai 4	N13	Nontarget group	Yes	Yes
E. oleracea extract powder	Acai 5	N14	Nontarget group	Yes	Yes
R. subg. Rubus dry powder	Blackberry	N15	Nontarget group	Yes	Yes
P. granatum dry powder	Pomegranate	N16	Nontarget group	Yes	Yes
A. melanocarpa dry powder	Chokeberry 3	N17	Nontarget group	Yes	Yes
G. max dry powder	Black Soybean	N18	Nontarget group	Yes	Yes

3. List of Consumer Available Products and Their Botanical Identification Results by the Classification Model.

Claimed Species	Sample Name	Sample code	Mahalanobis Distance Classification Result	HPTLC result for secondary confirmation	Correctly labeled by manufacturer?	Correctly identified by the classification method?
V. corymbosum liquid concentrate	Blueberry Supplement	B6	V. corymbosum	Not tested	Yes	Yes
V. macrocarpon juice	Cranberry Supplement	C14	V. macrocarpon	Not tested	Yes	Yes
V. myrtillus capsule	Bilberry Supplement	A12	Nontarget group	Not match with V. myrtillus BRM	No	Yes
S. nigra syrup	Elderberry supplement	N9	Nontarget group	Match with S. nigra BRM	Yes	Yes

Sample Preparation

Fresh samples were blended and homogenized by a coffee grinder. Depending on the visual color density, an appropriate amount of sample was weighed into a volumetric flask. Anthocyanins were extracted by acidified methanol (5% formic acid (v/v)) with sonication for 10 min. The extract was brought to 20 mL with acidified water (5% formic acid (v/v)) in a volumetric flask. After mixing, an aliquot of sample was pipetted into centrifuge tube and centrifuged for 4 min at 10,000 rpm. The supernatant was diluted by acidified water (5% formic acid (v/v)), filtered through a 0.22 um PTFE filter, and transferred to HPLC vial for analytical analysis.

Marker Selection and Analytical Method Development

Anthocyanins of 6 predominant anthocyanin aglycones (cyanidin (Cy), delphinidin (Dp), petunidin (Pt), peonidin (Pn), pelargonidin (Pg), and malvidin (Mv)) with glycosylation of galactose (-gal), glucose (-glu), and arabinose (-arab) were selected as marker compounds. Since it would be quite challenging to separate all 18 compounds based on their retention time in a chromatographic column, a LC-MS/MS method was developed to identify and quantify the selected anthocyanins by their mass-to-charge ratio (m/z) and fragmentation patterns.

Sample extracts were analyzed by Waters Acquity Xevo TQ MS (Waters Corp., Milford, MD, US) equipped with an Acquity UPLC CSH Phenyl-hexyl column (2.1 × 150 mm, 1.7 μm) (Waters Corp., Milford, MD, US). The UPLC system included a binary pump, a cooled autosampler maintained at 20 °C, a 5 μL sample loop, and a Photo Diode Array (PDA) detector. The oven temperature was set at 45 °C. The flow rate was 0.25 mL/min for a binary gradient. The mobile phase A (MPA) was consisted of water, acetonitrile, formic acid and TFA at a ratio of 19:1:0.002:0.01; the mobile phase B (MPB) was consisted of methanol, acetonitrile, water, formic acid and TFA at a ratio of 7:8:5:0.002:0.01. 50% Methanol was used as needle wash. The gradient settings was as follows: initial at 2% MPB and increased to 4% MPB in 5 min, from 4 to 20% MPB from 5 to 8 min, from 20 to 90% MPB from 8 to 8.5 min, 9.5 min at 10% MPA, 10 min at 96% MPA, and 13 min at 96% MPA. This gradient ensured anthocyanins with the same mass-to-charge ratio (m/z) and fragmentation patterns could be separated by their retention time.

Mass spectrometry was operated under positive ion mode. The capillary voltage was set at 2.0 kV with a cone gas flow of 150 L/h and source temperature of 150 °C. The desolvation flow was set at 1000 L/h with the desolvation temperature of 600°C, and the nebulizer at 7 bar. For each compound, the MRM transition with the highest signal-to-noise ratio was selected as the quantifier. Anthocyanins’ identifications were confirmed by reference standard comparison. Masslynx software was used to control the instruments and acquire the data. TargetLynx software was used for data processing.

A 50 ppm of Cyanidin-3-glucoside (C3g) reference standard was injected in triplicate at the beginning of each run and periodically throughout the batch to check system suitability. RSD of peak area and retention time were calculated by TargetLynx. For inclusion in the analysis, only batches with RSDs ≤ 2% for both peak area and retention time were considered. This criterion ensured the evaluation of adequate system suitability and instrument performance.

High-Performance-Thin Layer Chromatography (HPTLC) for adulteration confirmation

HPTLC was performed on a single plate with the following samples, which were all prepared at 5 mg/mL in methanol, V. myrtillus leaf BRM, V. myrtillus berry BRM, V. myrtillus supplement, S. nigra fruit BRM, S. nigra supplement. Samples were applied as 10 mm wide bands to a silica 60 gel HPTLC plate (Merck, fluorescence indicator F254) via a CAMAG Automatic TLC Sampler. Bands were eluted in a CAMAG Automatic Development Chamber with butanol: formic acid: H₂O (8:2:3) as the solvent. Plates were visualized with CAMAG Visualizer 2.

Data Processing and Chemometrics

Peak area of each anthocyanin generated by MassLynx were transferred to Microsoft Excel for additional processing. The area of individual anthocyanins was divided by the total peak area of all 18 anthocyanins and expressed as a peak area percentage. Peak area percentage was used as opposed to absolute peak area to allow a more direct comparison between samples with minimum and maximum total anthocyanin content. Peak area percentage was exported as a CSV file for analysis in RStudio (Posit, Version 2024.09.0 Build 375 using R 4.3.2.). Samples were labeled as their respective target species (cranberry, blueberry, or bilberry), nontarget species, or unknown species in a separate CSV file.

Principal component analysis (PCA) was conducted in RStudio for the data analysis. Data were first log transformed, all “0” cells were replaced by a small value (0.0001), and autoscaled with the prep.autoscale function in the mdatools package. Singular Value Decomposition-based PCA was conducted using the prcomp function in the stats package, and the scores plot was generated with ggplot2. Hotelling’s 95% confidence intervals were calculated and represented with ellipses for each species (bilberry, blueberry, cranberry, or nontarget). A scree plot (not shown) confirmed that PC1 and PC2 contained the highest variance and were appropriate for a sample comparison.

PCA was initially performed using the reference materials of target species to determine species groupings and to establish a confidence region for each group using Hotelling’s 95% confidence intervals. For sample classification, each unknown sample was assessed in the context of these groupings by incorporating it into the data set prior to PCA computation to ensure transformation consistency with the training data. The unknown sample was then labeled as “unknown” in the newly generated principal component space. Mahalanobis distance was subsequently calculated to measure how far this unknown sample was from the centroids of each target species cluster, based solely on the first two principal components. If the unknown sample fell within the Hotelling’s 95% confidence interval of a certain centroid based on this distance, it was classified as a member of that species. This approach effectively created a Mahalanobis Distance Classification model using confidence region boundaries as decision rules. If a sample did not fall within any of the defined ellipses, it was classified as a nontarget species.

Results

Anthocyanin Profiling of Main Vaccinium Species

Variations between the anthocyanin profiles of the three target species, V. myrtillus, V. macrocarpon, and V. croybosum, were apparent (Figure ). Figure provides an overview of the average anthocyanin content of the reference materials used in the study belonging to the three target species. There were differences in the major anthocyanins in each species; for example Pn-3-gal has an average of over 40% total peak area in V. macrocarpon samples but is only present in trace amounts in the other two species (Figure ). The standard deviation of each anthocyanin, represented by black bars, highlights the anthocyanin variation within reference materials of the same species.

1.

Average % peak area of 18 selected anthocyanins in the three target species: V. mytillus, V. corybosum, and V. macrocarpon. Averages are across all BRMs used in the study. Bars represent the standard deviation.

Taking V. myrtillus USP Reference Standard (Sample Code A1) as an example, anthocyanins were identified as Dp-3-gal, Dp-3-glu, Dp-3-arab, Cy-3-gal, Cy-3-glu, Pt-3-gal, Pt-3-glu, Pg-3-gal, Cy-3-arab, Pg-3-glu, Pt-3-arab, Pn-3-gal, Mv-3-gal, Pg-3-arab, Pn-3-glu, Mv-3-glu, Pn-3-arab, Mv-3-arab, following the elution order in LC chromatogram. The most abundant anthocyanin in V. myrtillus USP Reference Standard was Mv-3-glu, accounting for 13.4% of the total anthocyanin relative abundances, followed by Cy-3-arab at 11.8%. Other major anthocyanins included Dp-3-gal, Dp-3-glu, Dp-3-arab, Cy-3-gal, Cy-3-glu, Pn-3-glu, Pt-3-glu, and Mv-3-gal, with relative abundances ranging between 5-10% (Supplemental Tables S2A; Sample A1)

Similar anthocyanin profiles were observed in other V. myrtillus reference materials, although there were slight variations in the proportions of individual anthocyanins. For instance, the BRM from Starwest (Starwest Botanicals, Sacramento, CA, USA) exhibited a higher proportion of Dp-3-glu (19.7%) and a lower abundance of Cy-3-arab (6.1%) compared to the USP reference standard. Despite these minor differences, the overall anthocyanin profiles within V. myrtillus species remained consistent, with standard deviations ranging from 0.02% to 1.60% for each individual anthocyanin. In general, V. myrtillus fruit extracts contained significant amount of Cy-, Dp- and Mv-derivatives, while Pg-derivatives were present in negligible amounts, indicating a high similarity in anthocyanin profiles (Supplemental Tables S2A, Figure ).

The anthocyanin profiles of V. macrocarpon fruits were much simpler compared to V. myrtillus. The main anthocyanins identified in V. macrocarpon extracts were Pn-3-gal, Pn-3-arab, Cy-3-gal and Cy-3-arab. These four anthocyanins accounted for approximately 95% of the total anthocyanin content (Supplemental Tables S2C, Figure ). Pn-3-glu, Mv-3-gal, Pg-3-arab, and Mv-3-arab were detected at low levels in only a few V. macrocarpon fruits. Similar to that of V. myrtillus, the anthocyanin profile within the V. macrocarpon species exhibited a high degree of consistency, with the standard deviation (SD) of the main anthocyanins ranging from 2.2% to 4.9%.

However, the anthocyanin profile exhibited significant variation within the V. corybosum samples. As an example, the extract of NIST BRM (Sample Code B3) was characterized by high Mv-derivatives, with Mv-3-gal (29.6%), Mv-3-glu (22.9%), and Mv-3-arab (11.9%) collectively accounting for approximately 65% of the total peak area (Supplemental Tables S2B, Figure ). Other anthocyanins detected in the NIST sample included Pt-3-gal, Dp-3-gal, Pn-3-glu, Pn-3-gal, Pt-3-glu, Cy-3-gal, Cy-3-arab, Dp-3-arab, Pn-3-arab, Dp-3-glu, Cy-3-glu, and Pt-3-arab, listed in descending order of relative abundances. In contrast, the extracts from the ChromaDex (Sample Code B1) and AHP BRMs (Sample Code B2) exhibited negligible amounts of Mv-3-glu, while showing significantly higher proportions of Dp-3-gal, Dp-3-arab, and Mv-3-arab compared to the NIST SRM. As seen in Figure , V. corybosum and V. myrtillus have similar anthocyanin patterns, however, V. corybosum consistently exhibited a higher proportion of Mv- derivatives compared to V. myrtillus.

PCA Highlights Differences between Target Species

Following anthocyanin profiling via LC-MS/MS, Principal Component Analysis (PCA) was used to visualize species specific anthocyanin patterns. PCA is an unsupervised multivariate statistics technique that reduces multiple variables to smaller principal components (PC) which describe how samples vary and correlate based on the overall chemical profile. The resulting scores plot groups samples with more similar profiles together and those with more distinct profiles farther apart (Figure ). In this case, the model finds relationships between the % peak area of each anthocyanin, resulting in separation of samples based on the similarity in their overall anthocyanin profiles.

2.

Principal component analysis (PCA) result for distinguishing between the three Vaccinium species: V. myrtillus (blue squares), V. macroparpon (green circle), and V. corybosum (brown plus). The PCA score plot is shown for a model built with (A) 18 anthocyanins and (B) 18 anthocyanins and the ratio of cyanidin-to-malvidin derivatives.

Figure A demonstrates that PC1 and PC2, which explain 56% and 28% of the total variance, respectively, were able to separate V. macrocarpon from the other two target species using the 18 targeted anthocyanins. Each ellipse represents the 95% confidence interval for the specified group. However, there was an overlap between the V. myrtillus and V. corybosum ellipses, meaning that the two species cannot be fully distinguished using only the selected anthocyanins. The top contributing anthocyanins to PC1 were Pn-3-arab, Pn-3-gal, Dp-3-gal, DP-3-arab, and Pt-3-gal. The top contributing anthocyanins to PC2 were Pn-3-glu, Cy-3-glu, Dp-3-glu, Mv-3-arab, and Pt-3-glu.

As previously mentioned, Figure highlights that V. myrtillus and V. corybosum differ in their overall levels of cyanidin and malvidin. To improve separation, the ratio of cyanidin (Cy) and malvidin (Mv) was calculated as an additional variable, which allowed full separation of the three species (Figure B). Notably, adding the Cy/Mv ratio as an additional variable did not change the variance explained by each PC or the order of contributing variables along each PC, indicating that the overall chemical variability explained by the principal components was still preserved. Therefore, this model was selected as basis for the following steps.

Verification Study for Classification Model

To test the classification model performance, a method verification study was conducted to examine if the model can accurately classify target and nontarget Vaccinium species. Although PCA is unsupervised chemometrics, which does not inherently provide predictive classification, we utilized it in conjunction with Mahalanobis distance to analyze new samples within a model constructed by reference materials from the three target species. This approach enabled us to determine whether new samples aligned with the expected clusters on the scores plot, facilitating sample classifications.

We conducted three tests in the verification study. First, we added new botanical reference materials (BRMs) for each target species (excluded from initial model construction) to the data set individually. Each new sample consistently clustered within the 95% confidence interval of its respective species, showing 100% model accuacy (Figure , Table ).

3.

Potential of the presented model to correctly classify new samples was confirmed by adding a new BRM (red triangle) from each target species, one at a time. (A) V. myrtillus verification, (B) V. corybosum verification, and (C) V. macroparpon verification.

We further evaluated the model’s ability to distinguish nontarget samples (i.e., any sample not belonging to V. myrtillus, V. corymbosum, or V. macrocarpon) from the target clusters. We introduced 17 nontarget BRMs (Table ) individually one at a time, and all samples fell outside the clusters of the three target species (Figure , Table ). This demonstrated that the model could successfully identify when a sample did not belong to blueberry, cranberry, or bilberry, using profiles of 18 anthocyanins and the Cy/Mv ratio without requiring additional information. Mahalanobis distance calculations confirmed that these nontarget samples were outside the confidence intervals of the target clusters.

4.

Example of a new nontarget sample classification outside of the three target groups. Red triangles show that a new elderberry BRM is chemically distinct from the target groups.

To assess the model applicability to commercially available products, we tested four dietary supplements, each labeled as containing one of the target species or, in one instance, (elderberry). Table summarizes the test results for these products. Each product was extracted and profiled using the same protocol as the BRMs, and their anthocyanin profiles were added to the classification model. The V. macrocarpon, V. corymbosum, and S. nigra supplements clustered in accordance with their label claims (Figure A, B, D, and Table ). Among them, the V. macrocarpon product exhibited a high Pn-3-gla concentration (46%), consistent with V. macrocarpon BRM profiles. The V. corymbosum product showed the highest Mv-3-glu concentration (22%), aligning with the V. corymbosum BRM profiles (Supplemental Tables S2). The S. nigra product, with 97% Cy-3-glu and trace amounts of other anthocyanins, clustered distinctly outside the target species when it was added to the PCA model (Figure D). This classification as non-Vaccinium was confirmed by HPTLC analysis (Figure F), where the supplement’s banding pattern matched the in-house S. nigra BRM. The V. myrtillus supplement had a high relative abundance of Cy-3-glu (19%), Pn-3-glu (15%), and Mv-3-glu (14%) which looks similar to a mixture of the V. myrtillus and V. corybosum BRMs’ profiles. When added to the classification model, the supplement fell outside any cluster, indicating that it was incorrectly labeled by the manufacturer (Figure C). HPTLC confirmed that this supplement does not match the banding patterns of an V. myrtillus BRM (Figure E), indicating a possible incidence of adulteration.

5.

Use of PCA to classify consumer available products based on 18 anthocyanins and the ratio of Cy/Mv derivatives. (A) V. macroparpon supplement, (B) V. corybosum supplement, (C) V. myrtillus supplement, (D) S. nigra supplement. Model classifications were confirmed via HPTLC for the (E) V. myrtillus supplement (arrows point to missing bands in the supplement) and (F) S. nigra supplement.

Discussion

Among the Vaccinium species investigated in our study, V. corybosum (blueberry) contains various cultivars specifically bred for traits, such as size, firmness, and disease resistance, resulting in diverse anthocyanin profiles. Our study highlighted a notable difference in the anthocyanin profiles between the available BRMs (Supplemental Tables S2). The ChromaDex and AHP BRMs displayed similar anthocyanin profiles, characterized by higher proportions of Cy-, Dp-, and Pt-derivatives, and a limited presence of anthocyanins glycosylated with glucose. In contrast, the NIST sample exhibited higher abundances of Mv- and Pn-derivatives, with approximately 33% of the anthocyanins being glycosylated with glucose (Supplemental Tables S2). The observation aligned with the findings of Li et al. (2016), who identified and quantified anthocyanins in 19 blueberry samples and concluded that while the types of anthocyanins were similar across cultivars, the proportion of each individual anthocyanin was cultivar-dependent.

In contrast to V. corybosum, V. myrtillus is typically wild-harvested and less extensively cultivated. In our study, we analyzed 8 BRMs of V. myrtillus from different vendors, and all identified anthocyanins demonstrated a standard derivation less than 2.0%, confirming the consistent anthocyanin profile observed in previous studies (Supplemental Tables S2).

The difference between BRM anthocyanin profiles of the same species is critical to consider when developing adulteration detection methods. In many cases, Vaccinium adulteration is detected by comparing a sample’s anthocyanin profile to that of a single reference material. However, as previously discussed, substantial differences were present within V. corybosum species regarding the proportion of individual anthocyanins; therefore, a single reference material is not representative of the species. Furthermore, V. corybosum and V. myrtillus shared a similar pattern of anthocyanin composition. Both species contained anthocyanin from 5 different aglycones (Cy-, Dp-, Mv-, Pn-, and Pt-), with each of three sugar moieties (galactose, glucose, and arabinose), making it more challenging to distinguish the two species from the anthocyanin profile without chemometrics.

In our study, individual anthocyanins were semiquantified by % peak area. The relative, rather than absolute, content of selected marker anthocyanins allowed for a comprehensive evaluation of the entire anthocyanin profile, allowing us to assess the overall relative abundance of various anthocyanins within the sample with minimal normalization and chromatography data processing. Moreover, since botanical ingredients appear in various dietary supplements and raw materials at different concentrations, using relative abundance enables us to disregard their absolute concentrations and treat all samples uniformly. In summary, our semiquantification approach provided a holistic perspective on the distribution and relative proportions of marker compounds, enabling a more comprehensive pattern recognition analysis.

PCA is an accessible and simple chemometric tool to use the comprehensive chemical profiles generated via the semiquantitative approach to address the challenges associated with interspecies variation. By reducing the dimensionality of a data set, PCA uncovers patterns and visualizes relationships among different samples. In our case, PCA highlights the variations and similarities among different species based on the anthocyanin profiles. Additionally, PCA using a standardized set of anthocyanins eliminates the need for information about potential marker compounds from other nontarget samples, simplifying data collection. To further streamline our analysis, we focused on the 18 most common anthocyanins found in Vaccinium species.

Initial PCA analysis using 18 selected anthocyanin markers resulted in a significant overlap between the V. corybosum and V. myrtillus clusters due to the high anthocyanin variations within the V. corybosum species (Figure A). Figure highlights that although V. corybosum exhibited significant variations in individual anthocyanin accumulation among different samples, the percentage of malvidin derivatives remained consistently high (51.0% ± 11.8%). In contrast, cyanidin derivatives constituted only a small portion toward total anthocyanin content (7.5% ± 1.7%). On the other hand, V. myrtillus extracts displayed a more balanced distribution of cyanidin (24.0% ± 5.9%) and malvidin derivatives (23.9% ± 4.9%), with both accounting for approximately 25% of the total anthocyanin content (Table , Figure ). Thus, we hypothesized that a ratio between cyanidin and malvidin derivatives captures species-specific differences, offering an additional predictor for distinguishing between the two Vaccinium species in chemometric analysis.

By incorporation of the ratio of cyanidin and malvidin as a 19th marker in the PCA model, all three Vaccinium clusters were completely separated visually (Figure B). This addition of the cyanidin-to-malvidin ratio as a marker proved to be effective in achieving better differentiation between the species without changing the amount of variation explained by each principal component or the key variables contributing to each component. The sufficiency of the marker selection was further confirmed during model verification, in which new target BRMs were added to the model and successfully clustered with their respective species each time (Figure ).

To ensure classification consistency, PCA was always performed using the three target species training data sets along with one unknown sample. This approach ensures that the unknown undergoes the same transformation as the training data, preventing inconsistencies that could arise if verification samples were projected onto a precomputed PCA model. While incorporating an unknown sample into PCA computation slightly alters the clustering structure each time, these effects are minimal as the training data set from three target species remains dominant (as shown in the ellipse comparisons between Figure B vs Figure A–C, , and A–D). This method ensures that verification samples are transformed consistently with the training data set before Mahalanobis classification, preserving classification accuracy.

Our classification approach was based on the Mahalanobis distance, a statistical method used in chemometric applications. A similar technique was employed in an earlier study, where Mahalanobis distance and residual variance analysis were combined for spectral classification of near-infrared (NIR) reflectance data. While their study focused on pattern recognition in spectroscopy, our study is the first to leverage Mahalanobis distance for targeted LC-MS/MS-based anthocyanin fingerprinting. By integrating PCA with Mahalanobis distance classification, we enabled accurate species authentication in botanical ingredients and dietary supplements, directly addressing real-world cGMP regulatory requirements in the dietary supplement industry. Both studies demonstrated the Mahalanobis distance to be a powerful classification tool across different analytical techniques and data types.

In addition to demonstrating that this classification model successfully distinguishes the three Vaccinium species, it was also important to differentiate authentic Vaccinium extracts from nontarget species. AOAC SMPR provided a list including 14 fruit and 9 nonfruit sources to separate from Vaccinium species (AOAC SMPR 2014.07, 2014). Considering the availability of reference materials and previous reported adulterants, the nontargeted group in this study included extracts from acai (Euterpe oleracea), black soybean (), chokeberry (Aronia melanocarpa), pomegranate (Punica granatum), blackberry (Rubus spp.), elderberry (), grape (), and black rice (Oryza sativa L.). ,, When each of the nontarget BRMs were added to the model one at a time they all landed outside of the target clusters (Figure , Table ), indicating that their anthocyanin profile is distinct from the three target species.

Using multiple BRMs to build and verify the model is a unique strength of this study. Selecting BRMs and verified in-house samples allowed us to include relatively broad variability in the sample source. In other words, we used a diverse library of samples to expand the reference anthocyanin profiles. As a result, samples that may not match a single BRM’s anthocyanin profile but are truly the correct species were well represented. It should be noted that an ideal classification model would incorporate a broader range of authenticated samples for each target species to fully represent the natural intraspecies variability. However, due to the limited commercial availability of verified botanical reference materials, this study relied on all accessible, well-characterized samples. While this may result in slightly narrower 95% confidence intervals and potentially optimistic classification accuracy, the current work serves as a proof-of-concept for applying LC-MS/MS-based anthocyanin fingerprinting combined with chemometric classification to regulatory identity testing. These preliminary findings provide a practical and scalable framework for future validation studies using expanded data sets.

Under the U.S. FDA 21 CFR 111.75 cGMP regulations, dietary supplement manufacturers are required to conduct at least one test to verify the identity of each botanical ingredient. Additionally, according to section 402­(g)(1) of the U.S. Code (21 U.S.C. § 342­(g)(1)), a dietary supplement is considered adulterated if it is prepared, packed, or held in conditions that do not comply with cGMP standards. , To assess whether this classification model could serve as a quality control method to meet cGMP requirements by identifying adulterated dietary supplements, four commercially available supplements were analyzed for 18 anthocyanins (Table ). Based on their location within the scores plot, a V. corybosum and V. macrocarpon supplement were correctly labeled by the manufacturer (Figure A,B). Similarly, a nontarget supplement, (elderberry) was distinguished from the three target groups (Figure D), which was confirmed by comparing the supplement’s HPTLC patterns to that of an elderberry BRM (Figure F). However, the V. myrtillus supplement did not fall within the target cluster, suggesting that it is a mislabeled product. We confirmed that this sample was likely not V. myrtillus with HPTLC where the supplement’s densitogram did not contain signature banding patterns of V. myrtillus BRM (Figure E).

In a regulatory testing environment, multispecies comparative results provide a strong starting point for various testing options. In the case where a sample clusters with the expected group (the ingredient on the label), like in Figure A and B, we acknowledge it as authentic. On the other hand, if the supplement does not group with the expected species (Figure E), there is cause to suggest the sample differs from the claimed species. Since the developed LC-MS/MS method for anthocyanin detection is quite simple to perform and analyze, this method would save time over traditional analytical approaches by comparing more than one species at a time while including geographic, environmental, processing, and other potential varying factors. Single or small-group anthocyanin investigations typically performed using HPTLC use only one or two reference materials, so the full extent of sample variation is not accounted for. Though the initial cost of LC-MS/MS analysis is high, it is justified through the improved speed and reliability of the proposed approach for industrial applications. For example, a reputable company or organization can store the LC-MS/MS training set in silico database for authentic samples and use the classification model to analyze unknown samples. Or the database can be made publicly available for other laboratories to use it directly. Notably, our method is particularly suited for addressing instances of adulteration within Vaccinium species such as the substitution of bilberries with blueberries or wild lingonberries with American cranberries. With our approach, not only can we detect the presence of adulterants, but we may also discern the specific adulterant involved, underscoring its indispensable industrial applications. To the best of our knowledge, this is the first comprehensive analytical approach for authenticating different Vaccinium species in dietary supplement.

Further steps can be considered to implement this study in quality control testing. The development of a standard solution including all 18 anthocyanins would give more validity in retention time in the case of instrument issues or retention time shifting. Additional PCA models comparing only two target species would allow a closer look at samples that fall between potential groups. Despite our best efforts to collect samples that could be considered authentic, it is generally recommended that one include a much larger number of authentic samples for both model construction and method verification. Given the limited availability of authenticated reference materials from commercial sources in the United States, future studies may benefit from collaborations with farmers and producers worldwide to obtain verified botanical samples in a special initiative. Future work may also involve exploring the applicability of this methodology to other botanical groups, enhancing its utility for industry applications.

Conclusion

In conclusion, this study highlights the effectiveness of using a semitargeted anthocyanin profile to differentiate between three primary Vaccinium species and identify nontarget adulterants. Through a combination of PCA for visualization and Mahalanobis distance for classification, we demonstrated that this method can distinguish target Vaccinium samples from potential nontarget or adulterated samples. This approach is a preliminary but promising tool for identifying mislabeled products and verifying botanical identity in compliance with regulatory standards.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.analchem.4c06037.

• Study sample summary and Anthocyanin profile screening results for (A) Vaccinium myrtillus, (B) V. corybosum, and (C) V. macrocarpon reference materials (PDF)

‡.

X.W. and H.Y. authors contributed equally. The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. X.W.: Formal analysis, Writing - original draft. Conceptualization, Funding acquisition; H.Y.: Writing - original draft, Methodology, Resources, Conceptualization, Project administration. Y.Z.: Writing - original draft, Investigation, Methodology; E.A.: Writing - original draft, Investigation, Methodology; J.C.: Validation; S.C.: Project administration, Resources; Q.X.: Conceptualization, Funding acquisition.

The authors declare no competing financial interest.