Multi-Platform Chemical Fingerprinting for Authentication and Characterisation of New Zealand Monofloral Honeys
Loading...
Files
Size: 8.38 MB, File format: Adobe PDF
Date
Authors
Dissanayake Mudiyanselage, Rushan Lakshitha Bonuwan Diyanilla
Supervisor
Hamid, Nazimah
Le, Thao T.
Item type
Degree name
Journal Title
Journal ISSN
Volume Title
Publisher
Auckland University of Technology
Abstract
This research developed and evaluated an evidence-chain framework for authenticating New Zealand monofloral honeys (thyme, mānuka, kānuka and clover) while linking compositional fingerprints to functional bioactivity. As melissopalynology and single-marker testing can be inconclusive, especially for the closely related Myrtaceae honeys (mānuka/kānuka), this research tested whether combining orthogonal analytical platforms with interpretable modelling could produce defensible, class-consistent drivers suitable for routine screening. An intentionally unbalanced design (thyme n=13; mānuka/kānuka/clover n=3 each), reflecting thyme’s commercial relevance and strong preliminary bioactivity, was used to stress-test robustness under realistic sampling constraints. Twenty-two raw and commercial samples were characterised using (i) HS-SPME-GC-MS volatile profiling, (ii) a dual-window ¹H NMR strategy (direct honey profiling and C18-SPE enrichment to mitigate sugar-matrix suppression), (iii) targeted compositional panels (sugars, minerals, phenolics and selected bioactive constituents), and (iv) bioactivity assays (FRAP, CUPRAC, DPPH and anti-tyrosinase). Across chapters, PLS-DA, multinomial elastic net, Random Forest with Shapley Additive Explanations (RF-SHAP) and multiblock integration (DIABLO and sPLS regression) were implemented with sample-level stratified cross-validation and repeated resampling. Candidate marker sets were further constrained by requiring dominant target-class abundance (highest median), a minimum mean positive SHAP threshold and class-consistency support (positive SHAP across all target-class samples) to reduce outlier-driven findings.
HS-SPME-GC-MS delivered rapid discrimination with near-ceiling performance (5-fold micro-average ROC-AUC 0.995) and chemically interpretable drivers consistent with floral origin: thyme was characterised by oxygenated terpenoids and related aroma chemistry, clover by distinctive aldehydic/benzenoid features, and the Myrtaceae honeys by overlapping but separable plant-derived signatures. Direct-honey ¹H NMR, although dominated by the carbohydrate manifold, still achieved strong out-of-fold separability (micro-AUC ≈0.98) and highlighted regions consistent with established mānuka chemistry (including low-ppm and mid-field features aligned with MGO/DHA-associated resonances), while thyme was supported by aromatic windows around 7.42–7.38 ppm. C18-SPE NMR increased dynamic range (micro-AUC ≈0.99) and revealed enriched aromatic/phenolic regions that strengthened interpretability and cross-platform convergence.
Integrating targeted sugars, minerals, and phenolics translated complex chemistry into audit-ready decision rules: PLS-DA, elastic net and RF-SHAP achieved strong discrimination (ROC-AUC ≈0.95-0.99) and resolved botanically coherent drivers. Lumichrome emerged as a robust kānuka-associated discriminant, mānuka aligned with a Leptospermum-type profile including DL-3-phenyllactic acid, and thyme showed a coordinated antioxidant-leaning signature spanning bioactive constituents and composition (vitamin C/ascorbic acid, syringic acid and sucrose).
Finally, multiblock integration linking composition to function demonstrated a coherent chemico-biological gradient: Welch ANOVA confirmed honey-type effects across all bioactivity endpoints (p ≤ 0.004), with thyme highest and clover lowest, while mānuka/kānuka were intermediate and partially overlapping. DIABLO and sPLS regression indicated that the high-activity phenotype reflected a convergent multi-domain signature involving vitamin C, sucrose, phosphorus and phenolic acids, whereas lumichrome contributed mainly to genus-level discrimination within the Myrtaceae profile. Overall, the thesis shows that authenticity is best supported by convergent multi-platform signatures rather than single biomarkers, and provides a reproducible, interpretable workflow for New Zealand monofloral honey authentication and bioactivity interpretation.