Geographical origin classification of peanuts and processed fractions using stable isotopes

Highlights

• •Stable isotopes of peanuts and their different fractions are investigated.
• •Stable C, N, O and H isotopes of peanuts are used to assign production origin.
• •Peanuts are leguminous plants and fix nitrogen from the atmosphere.
• •Peanut δ¹⁵N is unaffected by processing and indicates soil nitrification processes.
• •LDA model achieved higher classification rates than k-NN and SVM models.

Abstract

This study investigates the use of stable isotopes (C, N, H, and O) to characterize the geographical origin of peanuts along with different peanut fractions including whole peanut kernel, peanut shell, delipidized peanuts and peanut oil. Peanut samples were procured in 2017 from three distinctive growing regions (Shandong, Jilin, and Jiangsu) in China. Peanut processing significantly influenced the δ¹³C, δ²H, and δ¹⁸O values of different peanut fractions, whereas δ¹⁵N values were consistent across all fractions and unaffected by peanut processing. Geographical differences of peanut kernels and associated peanut fractions showed a maximum variance for δ¹⁵N and δ¹⁸O values which indicated their strong potential to discriminate origin. Different geographical classification models (SVM, LDA, and k-NN) were tested for peanut kernels and associated peanut fractions. LDA achieved the highest classification percentage, both on the training and validation sets. Delipidized peanuts had the best classification rate compared to the other fractions.

Introduction

Peanut (Arachis hypogaea L.) is one of the major oilseed crops and is grown globally, mainly as a cash crop. In China, peanuts are widely cultivated, with China ranked among the top peanut producers and exporters in the world (Zhao, Wang, & Yang, 2020). Total peanut production in China reached 17.33 million tons in 2019 (National Bureau of Statistics of China. (2019), 2019). Its kernel is rich in edible oil (45 to 55 %) and provides a valuable source of macro and micronutrients (Shokunbi, Fayomi, Sonuga, & Tayo, 2012). However, its quality and safety is strongly associated with its growing origin. Peanuts are highly enriched in oil and are widely used as a raw material for oil extraction. Its oil content is strongly associated with the grown origin (Liang, 2021, Wang et al., 2021). However the potential contamination of peanuts from aflatoxins and heavy metals uptaken through irrigation water etc. is a serious food safety issue in peanut-producing areas (Liu et al., 2022, Lu et al., 2013). In China, peanuts produced in some regions including Heishan, Luanxian and Tongxu, etc. have been declared as Protected Geographical Indication (PGI) products and sell at a high premium than other regions. Mislabeling and adulteration of these PGI products for financial gain is a growing concern that now requires an effective analytical technique for authentication and origin identification of PGI peanuts and their products.

Recently, a few studies have explored the potential of different analytical techniques to determine the geographical authenticity of peanuts. In this context, only trace element analysis and NIR fingerprints have been applied to discriminate the geographical origin of peanuts (Wang et al., 2021, Zhao et al., 2020). Multivariate analysis revealed infrared spectral absorbance at different wavenumbers including 2923 cm⁻¹, 2851 cm⁻¹, 1742 cm⁻¹, 1162 cm⁻¹, and 1051 cm⁻¹ were associated with sub origin classification (Wang et al., 2021). Isotope ratio mass spectrometry (IRMS) and the analysis of carbon, nitrogen, hydrogen, oxygen, and sulfur stable isotopes have also been widely used in food traceability research (Krauß et al., 2020, Liu et al., 2019, Wadood et al., 2018) and depends on a number of abiotic factors. Isotope ratio measurement of a bulk sample or a particular fraction can provide useful information regarding its provenance. Carbon (δ¹³C) isotopes of plants mainly depends on their photosynthetic pathways such as C₃ (Calvin cycle) pathway, C₄ (Hatch-Slack) pathway, and CAM photosynthesis which provide clear carbon isotopic distinctions (Roy, Hall, Mix, & Bonnichsen, 2005). Peanuts are leguminous C₃ plants and follow the Calvin cycle for carbon assimilation. Nitrogen (δ¹⁵N) isotope ratios in the plants depend on various factors such as farming or fertilizer practices, soil and climatic conditions, and plant type (nitrogen-fixing or non-nitrogen fixing (Bateman, Kelly, & Woolfe, 2007). Peanuts are nitrogen-fixing, leguminous plants which remove nitrogen from the air and fix it in the soil via nitrogen-fixing bradyrhizobium nodules located on peanut roots (Wang et al., 2019). Oxygen (δ¹⁸O) and hydrogen (δ²H) isotopes fractionate during condensation and evaporation processes of the water cycle and the rate of fractionation depends on rainwater or groundwater input, local temperature, atmospheric oxygen, and CO₂ pressure. In addition, plant physiology also plays an important isotopic fractionation role within the plants (de Rijke et al., 2016) Moreover, the variation of these stable isotope ratios in agricultural products was also reported during different industrialized food processing steps. In this context, isotope effects from wheat processing including wheat milling, noodle fabrication, and cooking was investigated and found that wheat processing had no effect on δ¹³C or δ¹⁵N values or only slightly influenced δ²H and δ¹⁸O values by different processing techniques (Wadood et al., 2018, Wadood et al., 2019).

There is a growing demand by Chinese consumers for high-quality regional products. Peanuts are a very popular snack food and widely used for cooking. However peanuts are especially vulnerable to aflatoxin contamination, so consumers are more concerned about peanut traceability and geographical origin to ensure higher standards of food safety. To date, there are no reported studies on the use of stable isotopes to geographically trace the origin of peanuts. In this study, stable isotopes (C, N, H, and O) were measured in different peanut fractions (whole peanut kernel, shell, delipidized peanuts and oil) procured from three different geographical areas in China. The aim of this study was to measure the isotopic variations among different peanut fractions and identify geographical origin characteristics of peanuts using stable isotopes combined with multivariate data analysis.

Materials and methods

Sample collection and site description

Around 100 g of peanut samples were harvested and collected in 2017 from 76 farms in three main peanut-producing regions of China, including Shandong Province (n = 30), Jilin Province (n = 30), and Jiangsu Province (n = 16) as shown in Fig. 1.

Fig. 1.

Geographical location of peanuts sampled from different producing regions in China.

Processing of peanut samples

Each sample consisted of different peanut fractions, including peanut shell, whole peanut kernel, delipidized kernel, and peanut oil. About 10 g of peanut pods were selected for each sample analysis. The separated and peeled fractions as whole peanut kernel and shell were freeze-dried and grounded into a fine powder. Delipidized peanuts (kernel) and peanut oil were obtained following an oil extraction procedure: 0.5 g of powder peanut kernel samples were delipidized using Soxhlet solvent extraction (SER 148, VELP Scientifica) using 50 mL of solvent (chloroform: petroleum ether, 70/30 v:v) for 6 h. The lipid extracts were stored in a freezer at −20 °C and the dried, powdered samples were stored in a desiccator at room temperature until analysis. All samples were analyzed within 1 month after preparation.

Stable isotope analysis

A quantity of 4.5 mg of dried powdered samples of different peanut fractions (peanut kernel, peanut shell, and delipidized kernel) and 0.2 mg of peanut oil were weighed in duplicate and packed into tin capsules (3 × 5 mm). For C and N isotopes, samples were analyzed in an elemental analyzer (Vario Pyro Cube, Elementar, Germany) coupled with an isotope ratio mass spectrometer (EA-IRMS) (IsoPrime100, England). The samples were converted into CO₂ and NO_x at high temperature (920 ^ͦC) in the combustion furnace under a flow of oxygen. Subsequently, an inert carrier gas (helium) with a flow rate of 230 mL/min transferred the combusted sample into a reduction chamber (600 ^ͦC) where NO_x is reduced into nitrogen (N₂) gas. The gases were finally transferred to the IRMS for isotope determination.

For H and O isotopes, 0.9 to 1.2 mg of powdered kernel, shell and delipidized kernel were weighed and folded into isotope grade silver capsules (6 × 4 mm). The samples were equilibrated in a desiccator for seventy-two hours. Subsequently, samples were introduced into an elemental analyzer via an autosampler. High temperature combustion (1450 ^ͦC) was achieved using pyrolysis to convert organic H and O to gaseous H₂ and CO, respectively and finally the gases transferred into the IRMS for the isotope determination. The carrier gas (helium) was maintained at the flow rate of 120 mL/min and the pressure of the reference gas was 4 bar. The isotope values are expressed in delta notation and calculated using the following equation:

$$\delta \mathrm{E}=({\mathrm{R}}_{\mathrm{sample}}/{\mathrm{R}}_{\mathrm{standard}}-1)$$

Where δE represents δ¹³C, δ¹⁵ N, δ²H, and δ¹⁸O whereas R_sample and R_standard represent the ¹³C/¹²C, ¹⁵N/¹⁴N, ²H/¹H or ¹⁸O/¹⁶O ratio in samples and references, respectively. Peanut kernel, shell and delipidized kernel samples were also calibrated against IAEA-CH6 (Sucrose, δ¹³C = -10.4 ‰), USGS 64(Glycine, δ¹³C = -40.8 ‰), IAEA-N-2 (Ammonium Sulfate, δ¹⁵ N = 20.3 ‰), USGS 40 (l-glutamic acid, δ¹⁵ N = -4.5 ‰). The peanut oil was also calibrated using B2172 (olive oil, δ¹³C = -29.3 ‰) which was obtained from Elemental Microanalysis (United Kingdom). Reference standard materials USGS 54 (δ²H = -150.4 ‰, δ¹⁸O = 17.8 ‰) and USGS 56 (δ²H = -44.0 ‰, δ¹⁸O = 27.3 ‰) were used for H and O isotope calibration of kernels and shells. Reference standard materials IAEA-CH-7 (polyethylene, δ²H = -100.3 ‰) and IAEA-601 (Benzoic Acid, δ¹⁸O = 23.3 ‰) supplied by IAEA were calibrated for H and O isotopes of peanut oil samples. The method precision was lower than 0.1 ‰ for δ¹³C, 0.2 ‰ for δ¹⁵N, 2.0 ‰ for δ²H, and 0.5 ‰ for δ¹⁸O, respectively.

Statistical analysis

Different statistical analyses including one-way analysis of variance (one way ANOVA), multi-way ANOVA, box and whisker plots, k-nearest neighbor (k-NN), support vector machines (SVM) and Linear discriminant analysis (LDA) were applied. All analysis except k-NN and SVM models were conducted using SPSS for Windows 26.0 (SPSS Inc., Chicago, USA). k-NN and SVM models were analyzed using XLSTAT. One-way analysis of variance combined with Duncan’s test was applied to test the statistical significance between different origins of peanut and the different peanut fractions. Box plots combined with Duncan’s test were used to check the isotope variations between different peanut fractions. Multi-way ANOVA was applied to check the influence of different factors (regions, fractions, and their interactions) by calculating the contribution rate following mean square values. Finally, LDA, k-NN and SVM were applied to classify the geographical origin of the peanut samples. Segmented cross-validation was applied to all samples to validate the model.

Results and discussion

Stable isotopes (δ¹³C, δ¹⁵N, δ²H, δ¹⁸O) of peanut samples among different regions

Carbon, nitrogen, hydrogen, and oxygen isotopes (δ¹³C, δ¹⁵N, δ²H and δ¹⁸O) and carbon and nitrogen contents of peanut kernel and different peanut fractions including peanut shell, delipidized kernel and oil were determined and are summarized in Table 1.

Table 1.

Mean %C, %N, δ¹³C, δ¹⁵N, δ²H, and δ¹⁸O values of different peanut fractions from different geographical regions.

		Fractions
		Peanut Kernels	Peanut Shells	Defatted Kernel	Oil
Isotopes	Regions	Mean ± SD	Mean ± SD	Mean ± SD	Mean ± SD
C%	Shandong	59.7 ± 3.1a	46.1 ± 0.7 a	N/A	N/A
	Jilin	58 ± 1.7b	46.1 ± 0.7 a	N/A	N/A
	Jiangsu	59.9 ± 1.1a	45.6 ± 1.4 a	N/A	N/A
N%	Shandong	4.4 ± 0.4 a	0.8 ± 0.3 a	N/A	N/A
	Jilin	4.5 ± 0.5 a	0.8 ± 0.1 a	N/A	N/A
	Jiangsu	4.5 ± 0.4 a	0.8 ± 0.1 a	N/A	N/A
δ13C	Shandong	−28.4 ± 0.4b	−27.3 ± 0.7b	−25.5 ± 0.5b	−29.3 ± 1.3a
	Jilin	−27.8 ± 0.6 a	−26.6 ± 0.6a	−24.7 ± 0.4a	−29.5 ± 0.4a
	Jiangsu	−29.3 ± 0.7c	−28.1 ± 0.6c	−26.6 ± 0.6c	−30.8 ± 0.6b
δ15N	Shandong	−0.3 ± 1.2b	1.8 ± 1.0a	−0.5 ± 1.1b	NA
	Jilin	−0.1 ± 1.3b	1.4 ± 1.9a	0.1 ± 1.3b	NA
	Jiangsu	1.9 ± 1.9a	2.1 ± 1.6a	2.2 ± 1.9a	NA
δ2H	Shandong	−192.4 ± 11.9b	−95.3 ± 6.3b	−93.0 ± 7.4a	−265.3 ± 18.6c
	Jilin	−174.4 ± 4.9a	−93.8 ± 4.1ab	−90.4 ± 5.9a	−256.3 ± 7.5b
	Jiangsu	−195.8 ± 10.6b	−90.9 ± 4.7a	−90.9 ± 6.2a	−246.4 ± 13.5a
δ18O	Shandong	14.2 ± 1.6c	16.3 ± 1.3b	16.0 ± 1.7b	15.7 ± 1.6b
	Jilin	18.7 ± 1.4a	18.9 ± 0.9a	20.4 ± 0.8a	17.6 ± 0.9a
	Jiangsu	15.2 ± 0.9b	16.4 ± 1.2b	16.2 ± 0.5b	15.2 ± 0.9b

The δ¹³C values of peanut kernels ranged between −29.3 to −27.8 ‰. Post-hoc tests showed significant δ¹³C differences among different regions (p < 0.05). The highest δ¹³C values were observed in Jilin samples followed by Shandong and the lowest δ¹³C values were found in Jiangsu. Overall, δ¹³C values exhibited a significant increase when going from east (Shandong province, warmer temperatures) to north (Jilin province, cooler, drier temperatures). Plants in cooler climates may experience a higher degree of stomatal closure, less water availability and higher water use efficiency than in warmer climates (Bontempo, Camin, Paolini, Micheloni, & Laursen, 2016). Jiangsu, Shandong, and Jilin are located at different altitudes 16.7 m, 49.31 m, and 196 m, respectively and it has been reported that several climatic (e.g., rain, atmospheric pressure, sunshine) and edaphic (e.g., nutrient content, soil water holding capacity) factors that covary with altitude are responsible to impart significant variations in δ¹³C values. Most importantly, δ¹³C fractionation during photosynthesis in C₃ plants decreases with altitude because of greater carboxylation effect (Graves, Romanek, & Rodriguez Navarro, 2002). Other factors including plant physiology, solar radiation are also responsible for variations in δ¹³C values (Anderson and Smith, 2006, Wadood et al., 2018).

Peanut kernel δ¹⁵N values ranged from −0.34 to 1.89 ‰ with slightly higher values found in Jiangsu samples and the lower values in Shandong. Peanuts are nitrogen-fixing plants which typically have low δ¹⁵N values similar to the δ¹⁵N values in air (0 ‰). Slight variations between the three regions are more likely attributed to different climatic conditions which can change the physiology of plants and/or soil conditions allowing more or less nitrogen stockage in soils and uptake in plants. Slightly higher δ¹⁵N values in Jiangsu region peanuts compared to Jilin and Shandong peanuts reflect the uptake of stored atmospheric nitrogen from the soil, while lower δ¹⁵N values in the Shandong region reveal direct plant uptake of atmospheric nitrogen (ammonia mineralization) from nitrogen-fixing (Charles & Garten, 1993). Furthermore, soil characteristics and agricultural practices also impart variations in the δ¹⁵N values (Yuan et al., 2018). Soil characteristics among these regions have great diversity and N-stocking ability which influences δ¹⁵N values. Jilin region soils are mostly chernozem, aeolian sandy, and meadow soils, those in Jiangsu are sandy, brown, and yellow–brown soils; while Shandong has brown and cinnamon soils (Zhao et al., 2020).

δ²H peanut values also showed variations among different regions with the highest mean value (-174.39 ‰) found in Jilin while the lowest value (-195.76 ‰) was observed in the Jiangsu region. The main source of plant hydrogen is water taken up by the roots which is subsequently transpired through leaf stomata (Ziegler, Osmind, Stichler, & Trimborn, 1976). Therefore, more positive δ²H values are found in cooler, drier regions, where the evapotranspiration rate from leaf stomata is higher due to lower humidity. It has been reported that δ²H values become depleted with increasing inland distance from the coast (Dansgaard, 1964, Gat, 1996, Krauß et al., 2020). No significant difference was observed between peanut samples from Shandong and Jiangsu regions probably because of similar climatic conditions.

In the case of peanut oxygen (δ¹⁸O) isotopes, significant differences were observed for all three regions. A decreasing trend was observed for mean δ¹⁸O peanut values for the three regions with Jilin > Jiangsu > Shandong, respectively. δ¹⁸O values of local meteoric water are mainly associated with altitude, latitude, distance from the ocean, and the rate of evapotranspiration (Kern et al., 2020). Climatic differences among these regions are distinct and are the main contributor to peanut isotopic fractionation. Jilin province experiences a temperate, continental monsoon climate; Jiangsu province is a temperate to subtropical zone, whereas Shandong province experiences a warmer temperate, monsoon climate. Isotopic variations are reported between climatic conditions and δ¹⁸O values (Brescia et al., 2002). Mean peanut carbon content (%C) showed significant differences between Jilin/Shandong and Jilin/Jiangsu, while %N of peanuts did not exhibit any significant differences among regions.

Variation of stable isotopes (δ¹³C, δ¹⁵N, δ²H, δ¹⁸O) among different peanut fractions

The δ¹³C, δ¹⁵N, δ²H and δ¹⁸O values of the different peanut fractions including peanut kernel, peanut shell, delipidized kernel, and oil from each region were calculated and shown in Fig. 2 in the form of univariate box and whisker plots. In the case of δ¹³C, the most positive mean δ¹³C value (-25.4 ‰) was found in the delipidized kernel and the most negative mean δ¹³C value (-29.7 ‰) in peanut oil. For the different peanut fractions, a decreasing δ¹³C trend of delipidized kernel > peanut shell > peanut kernel > peanut oil, respectively. Post-hoc tests (DMR) showed significant differences among all fractions (p < 0.05). Peanuts contain high-fat contents (>50 %) we compare our bulk and delipidized peanut δ¹³C values with those from peanut oil. Our findings were consistent with previous research (Guo et al., 2010, Steele et al., 2010). Lipids are the most depleted naturally occurring organic material, therefore the removing oil from peanuts (delipidization) will have an impact on its δ¹³C values, especially as peanuts have a high fat content (>40 %) (DeNiro and Epstein, 1977, Krauß et al., 2020, Post et al., 2007). For this reason, the delipidized sample is usually considered to more accurately reflect its carbon isotopic value (Tulli et al., 2020).

Fig. 2.

Box and whisker diagram of carbon (δ¹³C), nitrogen (δ¹⁵N), oxygen (δ¹⁸O), and hydrogen (δ²H) of peanuts and different peanut fractions from three production regions. The centre line is the median value and the box represents the 25 to 75 percentile. The whiskers represent the minimum and the maximum non outlier values and small circles “o” represent the outliers. ^a-d letters represent significant differences.

Peanut shells showed the highest δ¹⁵N values (+1.7 ‰) followed by the delipidized kernels (0.3 ‰), and the lowest values were observed in peanut kernels (0.2 ‰), respectively. No significant differences were observed between whole peanuts and delipidized kernels which is consistent with previous findings (Liu et al., 2018). In terms of δ²H values, more positive values were observed in delipidized kernels (-91.5 ‰) followed by peanut shells (-93.8 ‰), peanut kernels (-186.0 ‰), and the more negative values were observed in peanut oil (-257.8 ‰), respectively. Duncan multiple range (DMR) tests depicted significant differences for different fractions (p < 0.05). A decreasing δ¹⁸O trend was observed delipidized kernels > peanut shells > peanut oil > peanut kernels, respectively. All fractions showed individual ability to resolve the geographical origin of peanut samples. The effect of lipid extraction on the δ²H and δ¹⁸O values of delipidized peanuts remains unclear due to a limited number of studies.

Multivariate analysis

Analysis of variance across the three regions was applied to determine the influence of factors such as region, fraction and region × fraction on δ¹³C, δ¹⁵N, δ²H and δ¹⁸O values (Table 2). All factors showed a significant influence on the stable isotope values. The relative contribution of each factor was evaluated using mean square values. In the case of δ¹³C, the different fractions explained the maximum variance (76.6 %) followed by different regions (23.2 %), and their interaction (0.1 %), respectively. For δ¹⁵N, different regions explained the maximum variance (55.2 %) followed by fractions (34.6 %), and the least well explained was their interaction (8.1 %). In the case of δ²H, the maximum variance was explained by different fractions followed by different regions and their interaction. The relative contribution rate in terms of δ¹⁸O was by different regions (87.0 %), fractions (10.3 %), and regions × fractions (2.2 %), respectively. These results are consistent with previous findings where the processing of wheat explained the maximum variance for δ¹³C and δ²H and geographical origin explained the maximum δ¹⁵N variance (Liu et al., 2018).

Table 2.

Combined analysis of variance for carbon (δ¹³C), nitrogen (δ¹⁵N), oxygen (δ¹⁸O), and hydrogen (δ²H) values of peanuts.

		δ13C (‰)		δ15N (‰)		δ18O (‰)		δ2H (‰)
Source of variation	df	MS	F	MS	F	MS	F	MS	F
Region (R)	2	43.85**	150.26	54.52**	26.29	369.54**	238.005	1343.462**	25.38
Fraction (F)	2	145.078**	497.107	34.127**	16.456	43.990**	28.331	210123.5**	3970.15
R × F	4	0.222**	0.762	8.008**	3.862	9.238**	5.949	1097.947**	20.745
Error	219	0.292		2.074		1.553		52.92

Geographical origin classification of peanut by LDA, k-NN and SVM

Classification methods based on LDA, k-NN, and SVM were developed and compared for their predictive ability to discriminate the geographical origin of peanut samples using stable isotopes. LDA is broadly applied for the classification of data due to its simplicity and empirical success in finding hidden trends in data. Similarly, SVM and k-NN are the widely used techniques that are developed through supervised learning. SVM can perform well with relatively small data set as it has a regularization parameter which work against over fitting and can effectively deal non-linearly separable data through the kernel trick. k-NN is easy, simple, and low cost algorithm that also performs well with small data sets and multiclass problems (Maione & Barbosa, 2019). To develop most robust model, discrimination capabilities of all these methods were compared. Table 3 summarizes the classification percentages. All fractions were individually used to check their discrimination power and finally-two matrices including peanut kernels and delipidized peanut samples were chosen because of their higher discrimination potential. Initially, linear discriminant analysis (LDA) was performed on the peanut kernel using 76 samples. Thirty samples were sourced from Shandong, 30 from Jilin, and the remaining 16 from Jiangsu. The LDA model correctly classified the origin of 88.2 % of samples and 86.8 % were correctly identified in cross-validation (Table 3). Subsequently, the second matrix (delipidized peanut samples) was used to develop a further LDA model. Two discriminant functions were obtained which were highly statistically significant (Wilk’s ƛ < 0.6). Function 1 explained the maximum variance (81.6 %) of the total variance. The developed model achieved a very high classification rate both in training and cross-validation. Samples were correctly classified with an accuracy rate of 94.7 % which was slightly reduced (92.1 %) in cross-validation. A few samples from Shandong and Jiangsu were misclassified, although Shandong had an accuracy rate of 93.3 % and 6.6 % of samples were misclassified as originating from Jilin and Jiangsu. Similarly, 87.5 % samples from Jiangsu were correctly classified, and 12.5 % samples were misclassified as Shandong, while 100 % samples were correctly classified as originating from Jilin. Delipidized peanut samples had a higher overall discrimination result which accounts for the removal of lipids (peanut oil) which adds a variable isotopic contribution depending on oil content. The geographical origin separation and classification is shown by plotting both discriminant functions (Fig. 3a).

Table 3.

Classification results of LDA, SVM, and k-NN of Peanut geographical origins by using the training and validation set.

a. LDA
	Origins	Peanut Kernel				Defatted Sample
		Predicted				Predicted
		Shandong	Jilin	Jiangsu	Total	Shandong	Jilin	Jiangsu	Total
Training	Shandong	27	3	0	30	28	1	1	30
	Jilin	3	26	1	30	0	30	0	30
	Jiangsu	2	0	14	16	2	0	14	16
	% correct	90	86.7	87.5	88.2	93.3	100	87.5	94.7
Validation	Shandong	2	4	0	30	26	3	1	30
	Jilin	3	26	1	30	0	30	0	30
	Jiangsu	2	0	14	16	2	0	14	16
	% correct	86.7	86.7	87.5	86.8	86.7	100	87.5	92.1
b. k-NN
Training	Shandong	21	2	0	23	23	0	1	24
	Jilin	1	22	0	23	3	16	0	19
	Jiangsu	4	0	8	12	6	0	4	10
	% correct	91.3	95.7	66.7	87.9	95.8	84.2	40	81.1
Validation	Shandong	7	0	0	7	5	0	1	6
	Jilin	2	5	0	7	2	9	0	11
	Jiangsu	1	0	3	4	2	0	4	6
	% correct	100	71.4	75	83.3	83.3	81.8	66.7	78.3
c. SVM
Training	Shandong	22	2	0	24	23	1	1	25
	Jilin	2	22	0	24	2	25	0	27
	Jiangsu	2	0	10	12	0	0	8	8
	% correct	91.66	91.66	83.33	90	92	92.5	100	93.3
Validation	Shandong	6	0	0	6	5	0	0	5
	Jilin	1	5	0	6	0	3		3
	Jiangsu	4	0	0	4	2	0	6	8
	% correct	100	83.33	100	93.7	100	100	75	87.5

Fig. 3.

(a) Cross plot of the first two discriminant functions obtained from the linear discriminant analysis of delipidized peanut matrices for different regions. (b) k-NN analysis of peanut kernel using the peanut kernel matrix.

Subsequently, a k-NN model was applied and prior to performing the analysis, the data set was divided into a training set (76 %) to develop a model and a test set (24 %) to calculate its classification performance. The k-NN model achieved the best performance when k = 3. The classification percentages for both the training and validation sets are summarized in Table 3c. The k-NN model achieved a total accuracy of 87.9 % in training and 83.3 % in the validation set, respectively, for the peanut kernels. The total classification percentages for delipidized peanut samples were 81.1 % for the training and 78.3 % for the validation sets, respectively. Fig. 3b depicts the geographical separation of peanut samples generated by the k-NN model. Finally, a SVM model was applied to classify samples according to different geographical origins. A SVM model was run with four types of kernel functions, including linear, power, RBF and sigmoid. The best performance was achieved with the power function. The SVM model achieved a higher classification rate both in the training and test sets. The total classification rates of the peanut kernel samples were 90 % in the training and 93 % in the validation sets, respectively. Similarly, the model achieved a total classification rate for the delipidized peanuts of 93.3 % and 87.5 % in the training and test sets, respectively (Table 3b). This comparison between the different statistical models showed that all the tested models were suitable for discriminating the geographical origin of peanut samples and each model had its own advantages. However, LDA proved to be a more effective and robust method since its classification rates were higher than the other two methods.

Conclusion

In this study, the geographical classification of peanuts and different peanut fractions from three provinces in China was successfully demonstrated using stable isotopes (C, N, H, and O). Peanut processing influenced the δ¹³C, δ²H, and δ¹⁸O values due to diploidization, while the δ¹⁵N values were mostly unaffected during processing. Climatic differences explained key isotopic variances between geographical regions, and similar trends were observed across peanut kernel and its fractions for each region. Three different geographical classification models were investigated. LDA had the highest classification ability for the delipidized kernels and SVM had the best model accuracy for the peanut kernel for both the training and validation sets. In general, the delipidized samples resulted in slightly improved classification results. The outcome of this study provides a useful method to verify peanut and peanut product origin in China and improve food safety through increase traceability using stable isotope and chemometric methods.

Author contributions

Syed Abdul Wadood and Yuwei Yuan conceived the idea, Yongzhi Zhang, Syed Abdul Wadood, Jing Nie drafted the manuscript, Chunlin Li analyzed data, Karyne M. Rogers and Yuwei Yuan edited the manuscript. The final version was approved by all authors.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.