Analytical Developments for Identification and Authentication of Botanicals James Harnly Food Composition and Methods Development Lab Beltsville Human Nutrition Research Center Agricultural Research Service U.S. Department of Agriculture Beltsville, MD, USA National Capital Area Chapter Society of Toxicology Annual Meeting, March 23, 2017
Authentication of Botanicals Terminology: authentication, identification, similarity, phyto-equivalence, taxonomic exactness, or adulteration. Perspective: these are all the same problem and best addressed with the same approach: non-targeted metabolite fingerprinting with chemometric analysis. Non-targeted methods: information is limited to the composition of the reference materials. No target compounds, no target adulterants. As comprehensive as possible. Usually a form of metabolite fingerprinting. Every data point is used. None are arbitrarily discarded. Chemometric analysis: use one-class modeling. Model is based only on reference samples. No other info needed.
Authentication Requires reference samples. Verification of the similarity/dissimilarity of the test sample and the reference samples. Authentication is like a Sesame Street Question Peter Scholl, FDA Which one is different? Or, rephrased: Is this one,, the same as the others?
Conceptually, Authentication is Simple 1. Build a model with reference samples. 2. Set statistical limits (how much deviation from the model will be tolerated). 3. Then, compare the test sample to the model. The Hard Part: 1. Specifying and collecting reference samples for the model that will encompass as many naturally occurring sources of variance as possible. 2. Specifying a method that will allow you to build a model that will encompass as many of the specific botanical features/properties as possible.
1. Collecting Reference Samples Most critical aspect of the analysis and usually the most difficult. Samples may be vouchered, self collected, historical. It depends on the purpose. Samples must be representative of the material of interest & the expected variance. Differences may arise from many factors (G x E x M): Genetics: species & sub-species Environment: geography, weather Management: conventional/organic farming, postharvest, processing Must compare apples to apples!
2. Selection of Methods Many possible measures of botanical properties. Modern methods: Genetic: Full/partial sequencing, DNA barcoding, mini-barcoding, next generation sequencing. Chemical: Metabolomics (targeted, markers) Metabolic fingerprinting (non-targeted, patterns require multivariate analysis). Prefer quantitative methods that produce values that can be treated statistically to assure objectivity. The method must be appropriate for the question being asked.
Non-Targeted Analysis - Metabolite Fingerprinting High throughput qualitative screening of the metabolic composition of an organism or tissue with the primary aim of sample comparison and discrimination analysis. Generally no attempt is initially made to identify the metabolites present. All steps from sample preparation, separation, and detection should be rapid and as simple as is feasible.* * Hall, RD. New Phytologist 169:453-468 (2006).
Non-Targeted Analysis Chromatographic and Spectral Fingerprints Fingerprints come from: 1) direct analysis of solids, 2) direct analysis of extracts (no separation), or 3) chromatograms of extracts. Fingerprints are complex chromatograms or spectra. Fingerprints require statistical or chemometric analysis to extract information. Any chromatographic or spectroscopic method can be used. Extract Separation Chromatograms Solid Sample Direct Spectra Fingerprints Solid
Basic Question: Do These Patterns Match? HPLC-Any Detector UV NIR NMR MS
How Do We Determine if the Patterns Match? Chemometric Methods! Approach a Method Supervision Model Exploratory PCA b None All data Class Modeling SIMCA ID classes 1 or more (soft modeling) (multi-pca) classes One-Class Modeling PCA ID 1 class 1 class Classification PLS-DA ID classes All Classes (hard modeling) a Richard G. Brerton, Chemometric for Pattern Recognition, John Wiley & Sons, West Sussex, UK, 2009, ISBN 978-0-470-98725-4 b PCA - Principal Component Analysis, SIMCA - Soft independent modeling of class analogy, PLS-DA Partial Least Squares-Discriminant Analysis.
Linear, 1D model (1 PC) fit to bivariate data: ( * ) authentic samples (mean-centered) One-Class Modeling Hotelling T 2 and Q Statistics Hotelling T 2 statistic Multivariate analog to Student s t value in univariate statistics. Characterizes the variance within the model.
Linear model (1 PC) fit to bivariate data: ( ) authentic samples * ( ) test samples * One-Class Modeling Hotelling T 2 and Q Statistics Hotelling T 2 statistic Multivariate analog to Student s t value in univariate statistics. Characterizes the variance within the model. Q Statistic Characterizes the variance outside the model. No equivalent in univariate statistics.
Simplest Approach to Authentication Step Collect authentic materials Obtain fingerprints Build a model Establish statistical limits Compare test material Comments Desired reference materials (fit for purpose) Chromatographic or spectral Fit a PCA model to authentic fingerprints Use Q statistic Does test material lie outside the 95% confidence limits?
Example #1: Authentication of Gingko biloba Samples: 18 commercial samples from local stores. 2 NIST Ginkgo biloba standard reference materials: SRM 3247 Powdered Extract SRM 3248 Oral Dosage Product (tablet) Analysis: HPLC-DAD (absorbance 220-400 nm) Data processing (3 approaches): Computed areas for 22 peaks (normalized to 100%) Chromatograms as images UV absorbance profiles as images (no separation) Model: One-class SIMCA
Chromatograms with UV Detection 18 Commercial Samples Retention Times Aligned (areas not normalized)
Peak Area Approach #1 - Measuring Individual Peaks Relative Peak Areas for 22 Flavonol glycosides Samples
PCA: Relative Peak Areas (numbers correspond to samples on previous slide) 17 11 2,10 1 5,8 3247,3248, 3,4,6,7,9,12, 13,14,15,16,18
95% Confidence Limit SIMCA: Authentic Samples Modeled Q Values vs Hotelling T 2 Values 95% Confidence Limit
PCA: Relative Peak Areas PC1, PC2, and PC3 11 2,10 6 1 5,8 17 SRM3247,SRM3248, 3,4,7,9,12,13, 14,15,16,&18
Peak Area Relative Peak Areas Additional unusual sample detected (outlined) Samples
Approach #2: Chromatograms as Images Retention Times Aligned (areas not normalized)
PCA: Chromatograms as Images Retention Times Aligned (sample spectra normalized) 2,10,11,&17 1,5,&8 SRM3247,SRM3248, 3,4,6,7,9,12,13, 14,15,16,&18
95% Confidence Limit SIMCA: Chromatograms as Images Q Statistic vs Hotelling T 2 Values 6 6 95% Confidence Limit
Absorbance Approach #3: UV Absorbance Spectra as Images Typical UV spectra Wavelength (nm)
PCA: UV Spectra (no chromatographic separation) 1,2,5,8,10,11,&17 18 SRM3247,SRM3248, 3,4,7,9,12,13,14,15, 16,&18 6
SIMCA: Authentic Samples Modeled Q Statistic vs Hotelling T 2 Values
PCA: Ginkgo Leaves and Commercial Supplements Apples vs Oranges 3246 Leaves Commercial Supplements 3247 Extract 3248 Tablets Leaves
1.E+06 Example 2: American Ginseng Flow Injection MS Rb2/Rb3/Rc 1.E+06 Rb1 8.E+05 Rd/Re 6.E+05 4.E+05 Rf/Rg1 2.E+05 0.E+00 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500
PCA: FIMS, Panax Species P ginseng P notoginseng P quinquefolius AHP K1 SC80 SC25 K3 K5
95% CL PCA & SIMCA: NIR, Panax Species PCA P quinquefolius P notoginseng SIMCA P ginseng 95% CL
95% CL PCA & SIMCA: NIR, Panax Quinquefolius Wisconsin grown PCA SIMCA Canadian grown 95% CL
Example #3: Authentication of Black Cohosh (Actaea racemosa) Obtained authentic A. racemosa and other species from American Herbal Pharmacopoiea (AHP), Strategic Sourcing (SS), North Carolina Arboretum (NCA), & NIST. Obtained commercial supplements and whole roots from the internet and stores in the US and China. Obtained DNA barcodes from AuthenTechnologies. Obtained spectral fingerprints using nuclear magnetic resonance spectrometry (NMR) and flow injection mass spectrometry (MS). Analyzed data using PCA and SIMCA.
PCA & SIMCA: Black Cohosh FIMS for AHP Samples 95% CL PCA A. racemosa A. dahurica A. pachypoda A. podocarpa A. rubra SIMCA 95% CL
PCA & SIMCA: Black Cohosh 1 H-NMR for AHP Samples 95% CL PCA A. racemosa A. dahurica A. pachypoda A. podocarpa A. rubra SIMCA 95% CL
PCA: Authentic Black Cohosh MS American Herbal Pharmacopoeia N Carolina Arboretum Germplasm Repository Strategic Sourcing NIST SRM 3295 NMR NMR
PCA: MS - A. racemosa from 22 Sites (Source - North Carolina Arboretum; Method - FIMS)
SIMCA: A. racemosa from 2 Sites as Examples (Source - North Carolina Arboretum; Method - NMR) Location #1 Location #22
A. racemosa cultivated from Different Sites NCA samples were collected from 22 sites along the eastern mountain range. They cluster together compared to the other species. Within the cluster, samples from the same location form sub-clusters. Variance within the A. racemosa samples may be due to: Isolated genetic differences Local climate and soil conditions Different types and levels of endophytic fungi Work on this project continues.
PCA: All Authentic Black Cohosh and Commercial Roots & Supplements SRM 3297, 3298 Rhizome extract & tablet Authentic Black Cohosh Commercial Root Samples SRM 3295 Rhizome Commercial Supplements
SIMCA: Model Based on All Authentic Black Cohosh vs Commercial Roots & Supplements SRM 3297, 3298 Rhizome extract & tablet Commercial Root Samples Commercial Supplements Authentic Black Cohosh SRM 3295 Rhizome
Example 4: Perspective on Chemical Identification PCA: Flow Injection MS Spectral Fingerprints
PCA: Echinacea 2 Species - 2 Plant Parts EAR E. angustifolia root EPA E. purpurea aerial EPR E. purpurea root EPR
PCA: Echinacea 2 Species - 2 Plant Parts - Supplements EPA E. purpurea aerial EPR E. purpurea root EAR E. angustifolia root S - Supplement EAR S S S EPA EPR
SIMCA: E. purpurea aerial Aerial Ingredient and Solid and Liquid Supplements EPA E. purpurea aerial EPAS E. purpurea aerial solid supplement EPAL E. purpurea aerial liquid supplement EPAL EPAS EPA
Q Statistic Q Statistic: Echinacea Fingerprints PCA based on E. Purpurea Aerial (1) 1 E. purpurea aerial 4 E. purpurea root 6 E. angustifolia root 2,5,7 solid single ingredient supplements 3,8 liquid single ingredient supplements 9-15 mixed ingredient supplements 6 8 3 4 5 2 9 13 +2 1 Mean -2 7 10 11 12 15 14 Sample
Summary Basic concept: build a model, set statistical limits, and compare test samples. PCA, SIMCA are essential mathematical tools. Must compare apples to apples. Must select appropriate method. Comparison of raw botanical materials and processed commercial supplements is difficult.
Acknowledgements USDA Agricultural Research Service, Office of Dietary Supplements at the National Institutes of Health, Kim Colson, Jimmy Yuk, Bruker BioSpin, Bellerica, MA, USA Joe-Ann McCoy, North Carolina Arboretum, Bent Creek Germplasm Repository, Asheville, NC, USA Danica Harbaugh Reynaud, AuthenTechnologies LLC, Richmond,CA, USA