Fourier transform infrared spectroscopy (FTIR) combined with multivariate calibration of partial least square (PLS) was developed and optimized for the analysis of
Adulteration of foodstuffs including oil is a serious problem because of the dangerous effects that may arise from additional ingredients that mixed into foods, such as the emergence of an allergic reaction [
FTIR spectroscopy combined with multivariate calibration is a rapid and reliable technique for quantitative analysis of oils in mixture. Multivariate calibration is an analysis that uses several variables (absorbances in many wavenumbers) and is often used for the analysis of complex mixture [
In authentication study, FTIR spectroscopy in combination with multivariate calibration has been used for authentication of extra virgin olive oil from palm oil [
In order to assure the authenticity of the used oils (NSO, CO, and SO), their fatty acid (FA) compositions were determined using a gas chromatograph (Shimadzu GC-2010, Shimadzu Corp., Tokyo, Japan), equipped with flame ionization. Before being analyzed, the samples of oils were derivatized using sodium methoxide to form FAMEs according to the method described by the American Oil Chemists Society (AOCS) [
The calibration samples composed of NSO in binary and ternary mixtures with CO and SO in the concentration range 0–100% (v/v) was prepared. Furthermore, a series of independent samples was also prepared as validation samples in order to evaluate the predictive ability of the developed calibration model.
FTIR spectra of all evaluated samples were acquired using FTIR spectrophotometer ABB MB3000 (Canada) equipped with ZnSe crystal, with sample handling technique of attenuated total reflectance (ATR), using detector of deuterated triglycine sulfate (DTGS), and connected to Horizon MB software. Samples are placed on the ATR crystal with restrained temperature (20°C). The collection of FTIR spectra was carried out at 32 scans with resolution of 8 cm−1 in the frequency regions of 4000–650 cm−1. After every scan, a new reference air background spectrum was taken. The ATR plate was carefully cleaned using hexane twice followed by acetone and dried with a soft tissue before filling with the next samples. These spectra were recorded as absorbance values, and replication was done 2 times.
The software Horizon MB (Canada) was used during performing multivariate calibration. Spectral regions that showed FTIR spectra difference between NSO, CO, and SO were selected to make PLS model. Worksheet Excel 2010 was used to correlate between actual concentration and predicted concentration. The performance of the obtained calibration model was evaluated using the values of coefficient of determination (
Figure
FTIR spectra of
If one examines the spectra closely, they reveal some differences which can be observed in the region around 1750–1700 cm−1 (peak d). NSO has two peaks at frequency region of 1750–1700 cm−1, meanwhile CO and SO revealed one peak. These peaks were attributed to carbonyl C=O stretching vibration from the ester linkage of triacylglycerol. Furthermore, at frequency region of 1128–1084 cm−1 (peak j and k), NSO also has two peaks; meanwhile, CO and SO appear with one peak. The magnified FTIR spectra at frequency region of 1750–1700 cm−1 and at 1128–1084 cm−1 were shown in Figure
Functional groups and mode of vibration from FTIR spectra of the evaluated oils [
Marker | Peak position of FTIR spectra (cm−1) | Assignment of bands | Mode of vibration |
---|---|---|---|
(a) | 3009 | C=CH ( |
Stretching |
(b) and (c) | 2922 and 2852 | –CH (CH3) | Stretching asymmetric |
(d) | 1742 | –C=O (ester) | Stretching |
(e) | 1658 | –C=C ( |
Stretching |
(f) | 1461 | –C–H (CH2) | Bending (scissoring) |
(g) | 1378 | –C–H (CH3) | Bending symmetric |
(h) and (i) | 1235 and 1161 | C–O (ester) | Stretching |
(j) and (k) | 1118 and 1098 | C–O | Stretching |
(l) | 964 |
|
Bending out of plane |
(m) | 914 |
|
Bending out of plane |
(n) | 871 |
|
Bending out of plane |
(o) | 844 |
|
Bending out of plane |
(p) | 721 |
|
Bending out of plane |
(q) | 1715 | –C=O | Stretching |
The magnified FTIR spectra frequency region of 1750–1700 cm−1 (a) and at 1128–1084 cm−1 (b).
FTIR spectra of NSO, CO, and SO are very similar since the main components of oils are triglycerides with certain fatty acids (Table
The composition of fatty acids in NSO, CO, and SO.
Fatty acids | Concentration (%) | ||||
---|---|---|---|---|---|
NSO | CO | CO (Codex) | SO | SO (Codex) | |
Caproic acid (C6:0) | 0.18 | ND | ND | 0.01 | ND |
Caprylic acid (C8:0) | 0.04 | 0.00 | ND | 0.01 | ND |
Capric acid (C10:0) | 0.06 | 0.00 | ND | 0.04 | ND |
Lauric acid (C12:0) | 0.07 | 0.14 | ND–0.30 | 0.02 | ND–0.10 |
Myristic acid (C14:0) | 0.02 | 0.00 | ND–0.30 | 0.06 | ND–0.20 |
Myristoleic acid (C14:1) | 0.01 | ND | ND | ND | ND |
Palmitic acid (C16:0) | 19.42 | 0.08 | 8.60–16.50 | 11.29 | 8–13.50 |
Palmitoleic acid (C16:1) | 0.29 | 0.001 | ND–0.50 | 0.11 | ND–0.20 |
Oleic acid (C18:1) | 45.22 | 15.32 | ND–3.30 | 0.02 | 17–30 |
Linoleic acid (C18:2) | 30.18 | 64.19 | 34.00–65.60 | 87.05 | 48.00–59.00 |
Linolenic acid (C18:3) | 2.07 | 16.76 | ND–2.00 | 0.64 | 4.50–11.00 |
Behenic acid (C22:0) | 0.23 | ND | ND–0.50 | ND | ND–0.70 |
Eicosatrienoic acid (C20:3) | 1.62 | 0.43 | ND | ND | ND |
Arachidonic acid (C20:4) | 0.59 | ND | ND | ND | ND |
ND: not detected; the level is below 0.001%.
During developing PLS regression, the samples were divided into the calibration and the validation sets. In the PLS model, evaluation of the method linearity was performed to show the proportional relationship between responses (absorbance) versus analyte concentrations of NSO in the mixtures over the working range [
Based on the highest value of
PLS performances at some frequency regions for the determination of NSO in binary mixtures with CO.
Frequencies region (cm−1) | Spectra | Calibration | Validation | ||
---|---|---|---|---|---|
|
RMSEC (% v/v) |
|
RMSEP (% v/v) | ||
4000–650 | Normal | 0.9818 | 3.50 | 0.9542 | 3.06 |
Derivative 1 | 0.9904 | 2.12 | 0.9975 | 1.32 | |
Derivative 2 | 0.9968 | 1.61 | 0.9975 | 1.15 | |
| |||||
2977–3028, |
Normal | 0.9929 | 2.23 | 0.9849 | 2.80 |
Derivative 1 | 0.9933 | 1.85 | 0.9970 | 1.41 | |
|
|
|
|
|
|
| |||||
1666–1739 | Normal | 0.9926 | 2.48 | 0.9929 | 1.83 |
Derivative 1 | 0.9966 | 1.55 | 0.9950 | 1.88 | |
Derivative 2 | 0.9915 | 2.08 | 0.9978 | 1.21 | |
| |||||
740–1446 | Normal | 0.9983 | 1.11 | 0.9965 | 1.34 |
Derivative 1 | 0.9983 | 1.19 | 0.9983 | 0.87 | |
Derivative 2 | 0.9921 | 2.20 | 0.9952 | 1.82 |
*The frequency region and FTIR spectral treatment selected for quantification were marked with italics.
The calibration performance of PLS for quantification of NSO in binary mixture with SO is shown in Table
PLS performances at some frequency regions for the determination of NSO in binary mixtures with SO.
Frequencies region (cm−1) | Spectra | Calibration | Validation | ||
---|---|---|---|---|---|
|
RMSEC (% v/v) |
|
RMSEP (% v/v) | ||
4000–650 | Normal | 0.9918 | 2.11 | 0.9927 | 2.44 |
Derivative 1 | 0.9992 | 0.73 | 0.9988 | 0.59 | |
Derivative 2 | 0.9995 | 0.67 | 0.9991 | 0.53 | |
| |||||
2977–3028, |
Normal | 0.9995 | 0.71 | 0.9991 | 0.58 |
Derivative 1 | 0.9994 | 0.75 | 0.9984 | 0.66 | |
|
|
|
|
|
|
| |||||
2985–3024, |
Normal | 0.9993 | 0.76 | 0.9992 | 0.66 |
Derivative 1 | 0.9997 | 0.47 | 0.9988 | 0.63 | |
Derivative 2 | 0.9993 | 0.66 | 0.9994 | 0.57 | |
| |||||
752–1755 | Normal | 0.9992 | 0.90 | 0.9996 | 0.46 |
Derivative 1 | 0.9993 | 0.74 | 0.9997 | 0.44 | |
Derivative 2 | 0.9996 | 0.59 | 0.9997 | 0.64 |
*The frequency region and FTIR spectral treatment selected for quantification were marked with italics.
The PLS calibration model was further used to analyze the validation samples. From the results obtained, it can be shown that PLS using first derivative spectra at frequency region of 2985–3024 and 752–1755 cm−1 is the better model, in terms of the highest value of
PLS model has been done at certain frequencies, including those selected for analysis of NSO in binary mixture with CO or SO. The calibration performance of PLS for quantification of NSO in ternary mixture with CO and SO is shown in Table
PLS performances at some frequency regions for the determination of NSO in ternary mixtures with CO and SO.
Frequencies region (cm−1) | Spectra | Calibration | Validation | ||
---|---|---|---|---|---|
|
RMSEC (% v/v) |
|
RMSEP (% v/v) | ||
4000–650 | Normal | 0.9975 | 1.40 | 0.9986 | 1.04 |
Derivative 1 | 0.9989 | 0.99 | 0.9987 | 0.76 | |
Derivative 2 | 0.9986 | 0.97 | 0.9990 | 0.69 | |
| |||||
2985–3024; |
Normal | 0.9985 | 1.07 | 0.9989 | 0.873 |
Derivative 1 | 0.9993 | 0.73 | 0.9993 | 0.70 | |
|
|
|
|
|
|
| |||||
2977–3028; |
Normal | 0.9987 | 1.07 | 0.9984 | 0.97 |
Derivative 1 | 0.9993 | 0.80 | 0.9986 | 0.78 | |
Derivative 2 | 0.9993 | 0.86 | 0.9995 | 0.58 | |
| |||||
1666–1739 | Normal | 0.9969 | 1.56 | 0.9973 | 1.27 |
Derivative 1 | 0.9978 | 1.37 | 0.9964 | 1.33 | |
Derivative 2 | 0.9971 | 1.53 | 0.9970 | 1.13 | |
| |||||
740–1446 | Normal | 0.9704 | 4.39 | 0.9777 | 3.57 |
Derivative 1 | 0.9990 | 0.88 | 0.9986 | 0.89 | |
Derivative 2 | 0.9954 | 1.55 | 0.9991 | 0.90 |
*The frequency region and FTIR spectral treatment selected for quantification were marked with italics.
The PLS calibration model was further used to analyze the validation samples. From the results obtained, PLS using second derivative spectra at frequency region of 2977–3028, 1666–1739, and 740–1446 cm−1 can predict well the validation sample due to the highest value of
In conclusion, FTIR spectroscopy combined with chemometrics of PLS regression is a powerful technique for the quantitative analysis of NSO in binary and ternary mixtures with CO and SO. The developed method is fast, does not need excessive sample preparation, and does not involve the use of reagents and chemicals.