A Novel Extraction Method for Spices and Non-destructive Evaluation of Spice Extracts

概要

Appended Form No. 3(Doctor)

Abstract of Thesis
Affiliation

Doctoral Program in Agricultural Sciences
Degree Programs in Life and Earth Sciences
Graduate School of Science and Technology

Student ID Number

202030249

Name

Bui Thi Bao Chau

Thesis Title

A Novel Extraction Method for Spices and Non-destructive
Evaluation of Spice Extracts
（スパイスの新規抽出法および抽出液の非破壊評価法の開発）

Abstract of Thesis

Spices, the edible, non-leaf, flavory parts of the plants like seeds or barks are well-known as
the reservoirs of many beneficial phytochemicals. However, spice extraction is very
challenging because many dry spices are very hard to crush. Intense pre-processing is often
separately applied, increasing the risk of losing trace or volatile compounds in multistage
extraction protocols. On the other hand, evaluation of spice extracts is a compulsory step to
directly evaluate the quality of spice extracts for usage or to indirectly assess of the spice
material and extraction method. Nonetheless, current evaluation methods are very laborious,
time-consuming with lots of toxic chemical waste. Altogether, these issues have raised the
concern and demands for simple but efficient methods to extract the hard-to-crush spices and
evaluate the quality of spice extracts. To answer these demands, in this study, a novel extraction
method named Simultaneous stone-milling and extraction (SME) method and a new extract
quality evaluation method based on Fluorescence Fingerprint (FF) in conjunction with machine
learning (ML) were established using four hard spices including three spice seeds (anise seeds,
dill seeds, fennel seeds) and one spice fruit (star anise) as the representative materials.
First, the four spice materials were shown to be significantly different in the criteria of
morphology, moisture content, solvent absorptivity and especially size, hardness, thus were
deemed suitable as representatives.
Using the characterized representative spices, SME was established to extract spices in
virtually one step with minimal preparation. Standard SME was set up as a process of milling
the whole spice seeds or 5 s pre-ground star anise twice (1 loop) in ethanol 70% at a materialto-solvent ratio of 1:20 (w/v) and mill rotation speed of 50 rpm. Standard SME achieved up to
6.5-fold higher total polyphenol content (TPC) and up to 4-fold more flavonoids (Al-F),
scavenging antioxidants (ScA) and reducing antioxidants (ReA) compared to the ordinary
maceration extraction. Next, SME was shown to be as efficient on whole, dry seeds without
pre-soaking and on dry star anise with the minimum pre-grinding time of 5 s as on pre-soaked

seeds and dry star anise pre-ground for longer time. SME was also proven most efficient with
ethanol 70% compared to ethanol 30% and 99.5%, with extraction efficiency staying the same
for 1 or 3 milling loops. Finally, SME under its most promising setting (standard setting) was
compared to common extraction methods including exhaust maceration (EM), hot extraction
(HE), ultrasonic extraction using ultrasonic bath (UB) and ultrasonic crusher (UC) on whole or
ground seeds and 5 s pre-ground star anise. SME had comparable extraction ability to the toptier methods EM and UC with higher consistency for all chemical attributes investigated. In
addition, star anise extracts prepared by SME showed higher safety with neutral effects on two
common cell lines HeLa and U-2 OS while exhibiting high inhibitory effects against the cancerrelated retinaldehyde dehydrogenase type 3 that were comparable to EM and UC. Collectively,
SME has been proven to be the promising new alternative extraction method for spices and for
hard plant materials in general.
Spice seed extracts prepared by SME and other extraction methods that had been chemically
evaluated were used as the materials for developing the novel evaluation method using nondestructive FF with ML to predict four chemical attributes (TPC, Al-F, ScA and ReA). FFs of
extracts were acquired at different dilution levels (DLs) including non-dilution (1×), and 2-, 5and 10-times dilution (2×, 5×, 10×). Not only peak intensity but also peak number and
appearance of the characteristic FFs of extract changed significantly with different DLs, seed
type and extraction methods, suggesting that more accurate prediction may be achieved using
data of all DLs. Partial Least Squares (PLS) models were built using data of individual DLs,
combined data of all DLs or stacked data of all DLs as benchmarks. More complex ML
algorithms including Support Vector Machine (SVM), Artificial Neural Network (ANN),
Random Forest (RF), Heterogenous Averaging Ensemble (HAE) and Heterogenous Stacking
Ensemble (HSE) were optimized to predict the four chemical attributes using combined data of
all DLs with or without feature selection (FS). While ScA could be adequately predicted with
the simple linear PLS models, TPC, Al-F and ReA required more complex models, especially
HAE and HSE on combined data of all DLs. On the other hand, FS only showed positive effects
on prediction of ReA. Altogether, it is confirmed that the use of combined data was more
appropriate to predict the chemical attributes of extracts using ML algorithms. In summary, the
best prediction models were suggested to be HAE without FS for TPC (test accuracy 0.73),
HAE without FS for Al-F (test accuracy 0.62), PLS with 1× FF data for ScA (test accuracy
0.81), and HSE with FS for ReA (test accuracy 0.76).
In conclusion, this study has laid the groundwork for efficient production of high-quality
spice extracts, from the extraction of spices by the one-step SME method to quick and nondestructive evaluation of chemical quality of the extracts by FF in conjunction with ML models.
This study is expected to encourage the use of beneficial plant materials like spices and to make
significant impact on natural sciences and ultimately contribute to human well-being. ...