Infrared (IR) spectroscopy combined with chemometrics has been developed for simple analysis of flavonoid in the medicinal plant extract. Flavonoid was extracted from medicinal plant leaves by ultrasonication and maceration. IR spectra of selected medicinal plant extract were correlated with flavonoid content using chemometrics. The chemometric method used for calibration analysis was Partial Last Square (PLS) and the methods used for classification analysis were Linear Discriminant Analysis (LDA), Soft Independent Modelling of Class Analogies (SIMCA), and Support Vector Machines (SVM). In this study, the calibration of NIR model that showed best calibration with
Indonesia shows an amazing diversity of plants species that have been associated with the human health from time immemorial. Many of them were reported to have various desirable activities; however, only 20–22% were cultivated [
Studies have shown that many plants have chemical components and biological activities. The most important of these bioactive constituents of plant are alkaloids, flavonoids, terpenoids, steroids, tannins, and saponins [
Several analytical techniques have been developed for determining total flavonoids concentration such as gas chromatographic (GC) [
Infrared spectroscopy is a technique based on the vibrations of the atoms of a molecule. The advantage of the infrared technique is that it can be nondestructive, requires a relatively small amount of sample, is fast, and is accurate [
Multivariate statistical methods are very useful for processing of IR spectra. The big advantage of multivariate statistical methods is their capability to extract the information of IR spectra and explore this spectral information for qualitative or quantitative applications. The most frequently used of multivariate statistical methods (often called chemometric methods) are Linear Discriminant Analysis (LDA) and Partial Least Squares (PLS) regression [
The objective of this research is to develop a simple, rapid, and validated model of IR spectra for the determination of the flavonoid content. Furthermore, IR spectroscopy and chemometric methods were applied for determining flavonoid content in commercial samples.
In this study, samples used were leaves samples collected from Materia Medica Botanical Garden, Malang, Indonesia (Table
Identity code of samples.
Number | Code | Leaves samples |
---|---|---|
(1) | A | |
(2) | B | |
(3) | C | |
(4) | D | |
(5) | E | |
(6) | F | |
(7) | G | |
(8) | H | |
(9) | I | |
(10) | J | |
(11) | K | |
(12) | L | |
(13) | M | |
(14) | N | |
(15) | O | |
(16) | P | |
(17) | Q | |
(18) | R | |
(19) | S | |
(20) | T | |
Dry leaves samples were mixed and finely powdered. 80.0 g of powdered sample was extracted with 800 mL of methanol in an ultrasonicator for an hour and continued being extracted by maceration for 24 hours. The extract was filtered through Whatman filter paper and then the solvent was evaporated using a rotavapour at 60°C. Extract was dried using Aerosil to yield dry extract.
Samples were scanned with a Brimrose, Luminar 3070 (Brimrose Corp, Baltimore, MD), with an integrating sphere. Before samples were measured, the instrument was warmed up for 30 minutes. The monochromator entrance slit was set on 500 pm, the amplifier was set on 200. the response time is smooth (1 ms), and light intensity was set on 14 volts. The wavelength range of spectra is from 8500–2000 nm and the data were measured in 5 nm intervals, which resulted in 120 points reflection.
FTIR spectrometer (Alpha FTIR Spectrometer from Bruker optic), equipped with a deuterated triglycine sulphate (DTGS) as a detector and a germanium as beam splitter, interfaced to computer operating under Windows-based system, and connected to software of OPUS operating system (Version 7.0 Bruker optic), was used during FTIR spectra acquisition. A few drops of each sample were positioned in contact with attenuated total reflectance (ATR) plate.
FTIR spectra were collected at frequency regions of 4000–650 cm−1 by coadding 32 scans and at resolution of 4 cm−1. All spectra were substracted against a background of air spectra. After every scan, a new reference of air background spectra was taken. The ATR plate was carefully cleaned by scrubbing with isopropyl 70% twice followed by drying with soft tissue before being filled in with the next sample, making it possible to dry the ATR plate. These spectra were recorded as absorbance values at each data point in replicate two times.
The flavonoids content was determined by aluminum chloride method using quercetin as a reference compound [
Chemometric analysis was performed using The Unscrambler software package (Version 10.2. CAMO ASA, Norway). Calibration and classification model for determination of flavonoids content were formed by training set samples that consists of fifteen medicinal plant extracts, quercetin, Aquadest, and Aerosil. The training set samples of medicinal plant extracts have been traced having varied flavonoid content which is expected to represent variations of flavonoid content of all plants. The Linear Discriminant Analysis (LDA), Soft Independent Modelling of Class Analogies (SIMCA), and Support Vector Machines (SVM) were used to develop classification model. These models were using two kinds of category, matrix and flavonoid. Matrix category was sample without flavonoids content (Aquadest and Aerosil) and flavonoid category was sample with flavonoid content (leaves extracts and quercetin). Partial Least Square (PLS) was used to develop calibration model for total flavonoids content. The PLS model was then validated with leave-one-out cross-validation (LOOCV) and 2-fold cross-validation (five test set samples). The training set and test set samples were shown in Table
Training set and test set samples.
Number | Samples code | Identity of group |
---|---|---|
(1) | A | Training set |
(2) | B | Training set |
(3) | C | Training set |
(4) | D | Training set |
(5) | E | Training set |
(6) | F | Training set |
(7) | G | Training set |
(8) | H | Training set |
(9) | I | Training set |
(10) | J | Training set |
(11) | K | Training set |
(12) | L | Training set |
(13) | M | Training set |
(14) | N | Training set |
(15) | O | Training set |
(16) | P | Test set |
(17) | Q | Test set |
(18) | R | Test set |
(19) | S | Test set |
(20) | T | Test set |
The results for total flavonoids content in samples are presented in Table
Total flavonoids content in samples.
Number | Samples code | mg QE/g extract ± SD |
---|---|---|
(1) | A | 9.87 ± 0.25 |
(2) | B | 11.23 ± 0.39 |
(3) | C | 51.49 ± 0.21 |
(4) | D | 31.53 ± 0.02 |
(5) | E | 24.17 ± 0.10 |
(6) | F | 35.61 ± 0.01 |
(7) | G | 9.74 ± 0.48 |
(8) | H | 32.74 ± 0.87 |
(9) | I | 27.87 ± 0.02 |
(10) | J | 32.55 ± 0.07 |
(11) | K | 15.06 ± 0.13 |
(12) | L | 39.22 ± 0.11 |
(13) | M | 26.41 ± 0.17 |
(14) | N | 36.46 ± 0.04 |
(15) | O | 40.25 ± 0.25 |
(16) | P | 14.39 ± 0.09 |
(17) | Q | 20.25 ± 0.72 |
(18) | R | 46.07 ± 0.28 |
(19) | S | 26.23 ± 0.78 |
(20) | T | 4.03 ± 0.07 |
Figure
NIR spectra of quercetin (a), dry extract (b), Aquadest (c), and Aerosil (d).
The correlation data of PLS (NIR model).
In order to validate the developed model, leave-one-out cross-validation (LOOCV) and 2-fold cross-validation were used. LOOCV was performed as follows: one sample was left out from the calibration set, a model was built with the remaining samples in the calibration set, then the left-out sample was predicted by this model, and the procedure was repeated by leaving out each sample in the calibration set.
The leave-one-out cross-validation (LOOCV) of PLS (NIR model).
Twofold cross-validation was used to validate the developed model using independent samples (test set). Five medicinal plant extracts were used as test set.
2-fold cross-validation of PLS (NIR model).
The ability of NIR model (LDA, SIMCA, and SVM) to classify samples in flavonoid and matrix category can be seen through the accuracy of classification models. Table
The accuracy of classification of NIR model (LDA, SIMCA, and SVM).
Model | Accuracy |
---|---|
LDA | 100% |
SIMCA | 100% |
SVM | 100% |
Figure
The calibration of FTIR model.
Wavelength number | | | RMSEC | RMSECV |
---|---|---|---|---|
4000 | 0.8558883 | 0.5403671 | 9.2037029 | 16.860432 |
3500 | 0.8527114 | 0.5758341 | 11.093782 | 18.985434 |
1300 | 0.8234164 | 0.7321395 | 10.187981 | 12.915498 |
1650 | 0.8653689 | 0.8201284 | 8.8958149 | 10.315225 |
FTIR spectra of quercetin (a), dry extract (b), Aquadest (c), and Aerosil (d).
The ability of FTIR model (LDA, SIMCA, and SVM) was less than 100%, which means that the model could not classify fifteen training set samples in a correct category (Table
The accuracy of classification of FTIR model (LDA, SIMCA, and SVM).
Model | Accuracy |
---|---|
LDA | 86.0% |
SIMCA | 91.2% |
SVM | 77.3% |
PLS and LDA developed models of NIR spectra were further used to predict flavonoid in commercial samples. The results of flavonoids content in samples measured by NIR and UV-Vis spectrophotometry method are presented in Table
Analysis of flavonoid content with NIR and UV-Vis spectrophotometry.
Commercial sample | Flavonoid content (mg QE/g extract) with NIR | Flavonoid content (mg QE/g extract) with UV-Vis spectrophotometry |
---|---|---|
Stimuno | 36.30 ± 2.78 | 35.94 ± 0.14 |
Daun Salam | 17.18 ± 0.06 | 15.12 ± 0.02 |
The NIR spectroscopy combined with multivariate calibrations methods can be used to determine flavonoid in medicinal plant extract. The suggested method is simple, selective, validated, and ecofriendly.
The authors declare that they have no competing interests.
The authors are grateful to the financial support of Kemenristek Dikti, Indonesia, for funding this fundamental research project.