Analysis of sunflower oil in ternary mixture with grapeseed oil and candlenut oil in the ternary mixture system using FTIR spectroscopy and chemometrics

Candlenut oil (CDNO) has a high price in the market so that it can be adulterated with one or more oils, therefore, the authentication of CDNO is very important. This study was aimed to develop a Fourier transform infrared (FTIR) spectroscopy combined with chemometrics to analyze sunflower oil (SFO) in ternary mixture with grapeseed oil (GPO) and CDNO. SFO, GPO, CDNO and its ternary mixture were prepared and scanned randomly using FTIR spectrophotometer in the mid-infrared region of 4000-650 cm, 8 cm resolution and thirty-two scans. SFO quantification was performed using two multivariate calibrations of principle component regression (PCR) and partial least square (PLS). PLS using the second derivative FTIR–ATR spectra at wavenumbers regions of 3100-2750 cm was used for quantitative analysis of SFO in ternary mixtures with GPO and CDNO, with correlation coefficient (R) values obtained for the relationship between actual values and FTIR predicted values of SFO of 0.9999 and 0.8744 in calibration and validation models, respectively. The errors in calibration and prediction models, expressed by the root mean square error of calibration (RMSEC) and root mean square error of prediction (RMSEP), were low, i.e. 0.233% and 7.90%, respectively. The use of FTIR spectroscopy method combined with multivariate calibration techniques is an appropriate method for detecting the adulteration of candlenut oil from GPO and SFO.


Introduction
Candlenut (Aleurites mollucana (L.) willd.) is a plant often found in eastern Indonesia (Sulistyo et al., 2009). Linoleic, oleic acid is a polyunsaturated oil contained in candlenut oil (Imdadul et al., 2017). Due to the unsaturated fatty acid content, candlenut oil (CDNO) is more expensive than other vegetable oil, especially in Indonesian fats industry (Yuliani et al., 2017). Candlenut originates from the Indo-Malaysia region and the productivity of the oil yield obtained from its seed is approximately 3200 kg/ha annually (Elevitch and Manner, 2006;Riyanta et al., 2020). Candlenut leave's traditionally used to treat coughs, diarrhea, headaches, ulcers, fever, hernia, and gonorrhoea (Quintãoa et al., 2019), Unfortunately, the seeds of the plant are considered as toxic for oral use (González-Stuart and Rivera, 2003). In addition, candlenut oil extracted from candlenut fruit was applied in the cosmetics industry, as good material for varnish, soap, and other oil-based cosmetics products (Subroto et al., 2017). Grapeseed oil (GSO) is considered as one of the possible sources of special lipids (Fernandes et al., 2013). GSO has a high content of polyunsaturated fatty acids (PUFA) at a range of 85-90% and among them, linoleic acid is dominant. Linoleic acid is associated with the promotion of cardiovascular health by down-regulating low-density lipoprotein cholesterol (LDL-c). GSO also contains some phytochemical components such as phytosterols, tocopherols, tocotrienols, flavonoids, phenolic acids and carotenoids which are important to human health (Crews et al., 2006).
The analytical methods applied for analysis for oil content are mainly based on chromatographic techniques (Maffei et al., 1992) such as gas chromatography for the analysis of fatty acid composition (Al-Ismail et al., 2011) and high performance liquid chromatography for the analysis of triacylglycerol composition . However, this method is time-consuming and needs skillful analyst, therefore, a more rapid and simple analytical method must be developed for quality control of high price oils. One of the emerging methods for the analysis of oils is Fourier transform infrared (FTIR) spectroscopy, especially combined with multivariate data analysis (chemometrics). Chemometrics is the application of mathematics and statistics to treat the chemical data including FTIR spectra to be extracted in more information which is easy to understand and is applied widely in the authentication of fats and oils (Rohman and Che Man, 2012).
The combination FTIR spectroscopy with chemometrics has been used for authentication studies of candlenut oil from soybean oil (Yuliani et al., 2017), olive oil from other vegetable oils (Lizhi et al., 2010), and analysis of plant oil in binary mixtures with pyridyl, abamectin, spinosad and malathion (Abd El-Razik and Zayed, 2014). Besides, this technique was also used for the analysis of pumpkin seed oil (Irnawati et al., 2020), analysis of walnut oil in binary mixture of sunflower oil (Ge et al., 2014) and analysis of essential oil (Khudzaifi et al., 2020). The reported publications are mostly on the authentication of high price oil with the lower one in the binary mixture and there are limited studies regarding the use of FTIR spectroscopy for analysis of sunflower in ternary mixtures with other oils. Therefore, this study highlighted the development of FTIR spectroscopy for analysis of sunflower oil in the ternary mixture with grapeseed oil and candlenut oil.

Preparation of calibration and validation samples
A total of twenty-nine samples consisting of SFO, GPO and CDNO in ternary mixtures at concentration ranges of 0-100.0% v/v were prepared. The composition of calibration samples were prepared randomly Excel (Microsoft Inc., USA), as shown in Table 1. All samples were analyzed using FTIR spectrophotometer.

Discriminant analysis
Discriminant analysis (DA) was used for discriminating between pure SFO and SFO in ternary mixture with GPO and CDNO. For training set, the authentic CDNO and CDNO added with SFO and GSO in ternary system were prepared at concentration of 0.5-50.0% (v/v) of SFO were prepared. Both groups (pure SFO and SFO-GPO-CDNO) were classified and discriminated using DA on the basis of Cooman's plot using FTIR spectral absorbance at specific wavenumbers as variables.

FTIR spectra acquisition
The scanning of FTIR spectra of oil samples were done using the FTIR spectrophotometer (Thermo Scientific Nicolet iS10, Madison, WI), equipped with the Omnic software for spectral processing. Horizontal attenuated total reflectance (HATR) at wavenumbers of 4000-650 cm -1 using thirty-two scans with resolution of 8 cm -1 was applied. Air used as background for all FTIR spectra were corrected against FTIR spectrum. A new reference air background spectrum was taken after every scan. These spectra were recorded as absorbance values at each data point in triplicate.

Chemometrics analysis
Chemometrics analysis including discriminant analysis and multivariate calibrations of Principle component regression (PCR) and partial least square (PLS) was performed using TQ Analyst software version 9 (Thermo Fisher Scientific, Inc.). Statistical parameters used for the evaluation of multivariate calibrations performance were coefficient of determination (R 2 ) and root mean square error of calibration (RMSEC) as well as root mean square error of prediction (RMSEP). Figure 1 shows that the three edible oils, sunflower oil (SFO), grapeseed oil (GPO), and candlenut oil (CDNO) as FTIR-ATR spectra were scanned at midinfrared regions corresponding to wavenumbers of 4000-650 cm −1 . The characteristic peaks in FTIR spectra represented the presence of several functional groups of the compounds. The interpretation of each peak and shoulder in relation to functional groups in FTIR spectra was presented in Table 2. Among SFO, GPO and CDNO spectra are very similar based on all peaks in FTIR-ATR. It's because the main components composing edible fats and oils were triglyceride (95-98%), glycerol esterified with three fatty acids. However, FTIR-ATR spectra of SFO, GPO and CDNO reveal slight differences upon detailed investigation, in terms of band intensities (absorbances) and the exact wavenumbers in each peak and shoulders.

Results and discussion
Two multivariate calibrations performed to quantitative analysis of SFO in ternary mixtures with GPO and CDNO, namely principle component regression (PCR) and partial least square (PLS). PCR is one of inverse calibration in which the concentration (xaxis) is used as a predictor, while response such as absorbance at several wavelengths is located in the y-axis. PCR performs multiple inverse calibrations of predictor variables against the scores (knowns as principal components) rather than original variables. PCR uses regression to converts scores of principal components (PCs) into concentration. PCs are obtained during principal component analysis (PCA). PLS is the most reported multivariate calibration for quantitative analysis to solve problems involving high collinearity. The main difference between PCR and PLS is that the first PC or factor in PCR represents the highest variations in the responses, whereas the first PC1 in PLS represents the most relevant variations showing the best correlation with the target property values (Rohman and Putri, 2019). Table 3 presents the performance of PCR and PLS for prediction of SFO in ternary mixtures with GPO and CDNO along with wavenumbers regions and FTIR spectral modes used (normal and derivatives). Based on the variation existed, the wavenumbers regions were selected among SFO, GPO, and CDNO. In addition, derivatization was intended to resolve any extensive overlapping peaks. The selection of FTIR spectroscopy conditions (multivariate calibrations, spectral modes, and FULL PAPER Figure 1. FTIR spectra using attenuated total reflectance mode of sunflower, grapeseed oil, and pumpkin candlenut oil acquired at mid-infrared region (4000-650 cm -1 )  wavenumbers regions) was based on the highest coefficient of correlation (R) and the lowest errors in calibration and prediction, expressed with root mean square error of calibration (RMSEC) and root mean square error of prediction (RMSEP). Finally, the second derivative FTIR spectra at the wavenumbers regions of 3100-2750 cm -1 aided with PLS regression offered the best prediction models for quantitative analysis of SFO in ternary mixtures with GPO and CDNO. The R values obtained for the correlation between actual values and FTIR predicted values were 0.9999 and 0.874 in calibration and validation models, while RMSEC and RMSEP values were 0.233% and 7.90%, respectively ( Figure 2). The low errors either in calibration and prediction models were confirmed by residual analysis in which no systematic errors were observed. The high R values and lowest values of RMSEC and RMSEP indicated that FTIR spectroscopy aided with PLS offered a reliable technique for analysis of SFO. This results could be extended for analysis of SFO as adulterants in high valuable oils such as GPO and CDNO either in binary or ternary models.
Discriminant analysis (DA), one of the supervised pattern recognition techniques, was used for discrimination between pure SFO and SFO in ternary mixtures with GPO and CDNO. Both classes (pure and the mixture) were clearly separated indicating that SFO was different from GPO and CDNO (Figure 3) Table 3. The performance of principle component regression (PCR) and partial least square (PLS) for prediction of sunflower in ternary mixtures with grapeseed oil and candlenut oil along with wavenumbers regions and FTIR spectral modes used (normal and derivatives).
Bold indicates selected variables. R = correlation coefficient; RMSEC = root mean square error of prediction; RMSEP = root mean square error of prediction.
groups are very similar. This result could be extended that DA based on Mahalanobis distance of variables used (absorbances at wavenumbers of 3100-2750 cm -1 ) was an effective means for authentication of GPO and CDNO from the possible adulteration with low-value edible oils such as sunflower oil.

Conclusion
Fourier transform infrared (FTIR) spectroscopy combined with PLS using second derivative FTIR-ATR spectra at the wavenumber's regions of 3100-2750 cm -1 is capable of quantifying sunflower oil in ternary mixtures with grapeseed oil and candlenut oil. In addition, discriminant analysis is also successfully used for classifying SFO and SFO in ternary mixture with grapeseed oil and candlenut oil. The developed method is reliable and is suitable to be used for authentication analysis of high price oils (candlenut oil and grapeseed oil from potential adulterant).
Khudzaifi, M., Retno, S.S. and . The employment of FTIR spectroscopy and chemometrics for authentication of essential oil of FULL PAPER Figure 2. The correlation between actual values (x-axis) and FTIR predicted values (y-axis) of sun flower oil in ternary mixtures with grapeseed oil (GPO) and candlenut oil (CDNO) with coefficient correlation of 0.9999 and 0.8744 in calibration and validation models. RMSEC = root mean square error of calibration; RMSEP = root mean square error of prediction. Figure 3. The discriminant analysis expressed by Cooman's plot for the discrimination of sunflower and sunflower oils in ternary mixtures with grapeseed oil and candlenut oil.