FTIR-based fingerprinting and discriminant analysis of Apium graveolens from different locations

Materials

Apium graveolens var. secalinum Alef. leaves were determined by the Center for Information and Development of Traditional Medicine, University of Surabaya (certificate number: 1403/D.T/VII/2019), and by B2P2TOOT Tawangmangu (certificate number: YK.01.03/2/290/2020). Plant materials were collected in June 2019 from several provinces on Java Island, i.e., six samples from Jawa Timur, one sample from Jawa Barat, three samples from Jawa Tengah, and three samples of commercially available A. graveolens crude drugs (Table 1). The chemicals used were potassium bromide (KBr, spectroscopy grade, from Merck, Darmstadt, Germany). As for the research equipment, it included a Jasco FT/IR-4200 spectrophotometer (Japan) with a deuterated triglycine sulfate detector, moisture content balance (Moisture Analyzer HB43 Mettler-Toledo GmbH, Laboratory & Weighing Technologies, Switzerland), and analytical balances (Ohaus).

Table 1. The geographical origins of A. graveolens. [Click here to view]

Crude drugs preparation

The aerial parts of A. graveolens (Fig. 1) were cut, and then the leaves were separated from the petioles. Afterward, these leaves were washed with tap water, drained, and air-dried at room temperature. The crude drugs (Fig. 2) were then ground and sieved using a 45 mesh.

FTIR spectrum

The FTIR spectrum of A. graveolens was recorded using an FTIR spectrophotometer. A total of 5 mg of crude drug powder was finely crushed with 95 mg of KBr in an agate mortar and then pressed with a manual hydraulic press to form a pellet. Afterward, the pellet was put into a sample holder, and then the spectrum was read at the wavenumbers of 4,000–400 cm⁻¹ with a 4 cm⁻¹ resolution and processed in Spectra Manager version 2 software.

Chemometrics analysis

The FTIR spectra of A. graveolens from various places were then analyzed by chemometrics, i.e., PCA, CA, and DA, in Minitab v.16 software (Minitab Inc., State College, PA). PCA works by translating the spectrum from an n-dimensional variable (spectra) into a PC, in which one score describes each spectrum in the data set in a new, much lower dimensional space. The PCs are sorted based on their eigenvalues that quantify the amount of variance captured by the PC. Then, the PCs that capture the most significant variation in the FTIR spectra (PCs with the greatest eigenvalues) are retained, while the PCs that only contain noises are discarded (PCs that have the smallest eigenvalues). PCA score plots were used to classify samples from their measured properties objectively. The distribution of samples on the graph may reveal a pattern that might be correlated to the general characteristics of the said samples (Yang et al., 2007).

Figure 1. The aerial part of A. graveolens. [Click here to view]

Figure 2. Apium graveolens crude drugs collected from Bondowoso (a), Magetan (b), Ponorogo (c), Bogor (d), Mojokerto (e), Batu (f), Karanganyar (g), Pasuruan (h), Banyumas (i), and Tawangmangu (j). [Click here to view]

CA was carried out to classify samples based on similarities in their chemical properties. For this reason, the samples were grouped in a high dimensional space and thereby formed a dendrogram. In the first step, each sample forms a cluster, and then two nearest objects are grouped. In the next step, either a third sample joins the first two or two other samples join a different cluster. Each of these steps results in one cluster less than the previous step until, eventually, all samples are in one cluster (Yang et al., 2007). In this study, single linkage was used and the Euclidean distances produced were calculated.

DA is one of the supervised pattern recognition methods. In general, supervised techniques make use of calibration or training sets with a priori information to build a classification model. The model is then tested using an independent sample set with a priori information to validate the predictive properties of the model before using it on unknown samples. DA assumes that the data are normally distributed, and the variance–covariance matrices of all classes are equal. It focuses on finding the optimal boundaries between object classes. It is a feature reduction method and designed to achieve maximum separation between the different classes. The basis of DA is to carry out linear combinations of the selected descriptors called canonical variates (CV) or discriminant functions that have high-class discriminating power (Gad et al., 2013).

Physical characteristics of crude drugs

Apium graveolens crude drugs used in this study were collected from 10 regions on Java Island, Indonesia. Three samples of commercially available crude drugs were also analyzed. Their clusters were then evaluated by FTIR (fingerprint data) and DA. The physical characteristics and moisture contents of each sample are shown in Figure 2 and Table 2, respectively.

Table 2. The moisture contents of A. graveolens crude drugs. [Click here to view]

Apium graveolens samples from Bondowoso and Pasuruan represent the population in low-lying areas (0–100 m a.s.l.), while the ones from Magetan, Bogor, Batu, Karanganyar, and Tawangmangu typify A. graveolens from highlands (>700 m a.s.l.). Regions at moderately elevated altitudes are represented by samples from Ponorogo, Mojokerto, and Banyuwangi. An organoleptic analysis was conducted as the first step in quality evaluation. All samples (Fig. 2) showed significant similarity, with merely slight differences in color. Because the standard harvesting and postharvesting processes were applied to all samples, such differences in organoleptic features could be attributed to their origin. Furthermore, the moisture contents of all samples were not higher than 10%. This parameter is of vital importance because water and/or moisture content in a crude drug is an excellent medium for the growth of microbes, fungi, or insects, and it can encourage hydrolysis and, subsequently, deterioration (WHO, 1998). The Indonesian Herbal Pharmacopoeia sets the loss-on-drying of A. graveolens at no higher than 10% (Depkes, 2008).

The fingerprint regions of FTIR spectra of A. graveolens

The FTIR profile of one sample of A. graveolens [Tawangmangu, Jawa Tengah (JTG)] is shown in Figure 3. It exhibited some typical bands. A fairly wide band at the wavenumber of 3,368 cm⁻¹ indicates the stretching vibrations of the O-H group. At the same time, the stretching vibrations of C-H and C=O appeared at 2,926, 1,734, and 1,651 cm⁻¹, respectively, marking the presence of phenolic and flavonoid compounds. The FTIR profile of the other samples is shown in Figure 4. All spectra shared similarities with merely a slight difference in the transmittance value of each major or minor band. This finding indicates that their chemical compounds are identical.

Fingerprint analysis by PCA

The general principle of fingerprint analysis is to display and analyze, as much as possible, the compounds or characters of samples, so that a valid conclusion can be drawn. The FTIR fingerprints of A. graveolens were analyzed by tabulating the FTIR spectrum data (data not shown). Rows represent the origin of the samples, while the columns show the transmittance value of each peak at each wavenumber. The spectra at the wavenumbers of 1,800–400 cm⁻¹ were tabulated to form 10 × 21 matrix data. These data were then analyzed by PCA (Fig. 5).

Figure 3. The FTIR profile of the A. graveolens crude drug from Tawangmangu. [Click here to view]

Figure 4. The FTIR spectra of A. graveolens crude drugs from 10 places of origin and three commercially available A. graveolens crude drugs. [Click here to view]

Figure 5. The PCA score plot. [Click here to view]

Figure 6. The dendrogram resulted from the linkage CA. 1–10 represents A. graveolens from Tawangmangu (1), Karaganyar (2), Mojokerto (3), Batu (4), Ponorogo (5), Magetan (6), Bogor (7), Pasuruan (8), Banyumas (9), and Bondowoso (10). [Click here to view]

The overlaid spectra of 10 A. graveolens samples (Fig. 4) were then analyzed by PCA. The first and second PCs described 94.4% and 3.4% of the variability in the original observations, respectively, and both PCs accounted for 97.8% of the total variance. Therefore, the first two PCs concentrated the multidimensional information into a 2D data set to classify the samples. Figure 5 shows that A. graveolens samples from 10 origins were classified into two groups: the samples from Banyumas, Batu, Bogor, Bondowoso, Karanganyar, Magetan, Ponorogo, and Tawangmangu belonged to group 1, while samples from Mojokerto and Pasuruan formed group 2. Group 1 consisted of samples from low-lying areas, regions at moderately elevated altitudes, and highlands, whereas group 2 was composed of A. graveolens from low-lying areas and regions at moderately elevated heights. This grouping apparently is not related to the height of the growing location. Other factors, like the type of soil and cultivation process, also play an essential role in shaping the quality of medicinal plants. The results of this study are consistent with Royani et al. (2014), which found that the andrographolide content in Andrographis paniculata varies according to provenance and that there is no correlation between the height of location and the andrographolide concentration (Royani et al., 2014).

Figure 7. The FTIR profiles of commercially available A. graveolens crude drugs 1 (a), 2 (b), and 3 (c). [Click here to view]

Table 3. The DA of commercially available A. graveolens crude drugs. [Click here to view]

Fingerprint analysis by CA

CA was carried out to derive groups from the autoscaled data set and compare them with the previous ones yielded by PCA. Two main components were selected as the basis for classification. The clusters formed (Fig. 6) illustrated different characteristics in A. graveolens. Similar to PCA, CA also found that A. graveolens can be divided into two clusters. The first cluster consisted of crude drugs originating in Tawangmangu, Ponorogo, Bogor, Magetan, Karanganyar, Banyumas, Bondowoso, and Batu, while the second one included samples from Mojokerto and Pasuruan. The high similarity observed between the results of PCA and CA further confirms that the fingerprint and PCA analysis discussed in this article are reliable for assessing the quality of A. graveolens.

Fingerprint analysis by DA

In this study, the FTIR profiles of three commercial samples (Fig. 7) with unknown origins were also analyzed. DA was applied to predict which samples belonged to which group. This method is used to identify new samples so that they can be grouped into the existing ones. If the PC score 1 of the new sample is higher than the PC score 2, the new sample can be classified into group 1. However, if the PC score 2 of the new sample is greater than the PC score 1, then the new sample belongs to group 2. The grouping of the commercial samples is shown in Table 3. Based on the DA results, the three commercial samples can be classified into group 1 or 2. Commercial samples falling into group 1 indicates that their quality is similar to those of group 1 members, and the same case applies to those in group 2.