A review of conventional and rapid analytical techniques coupled with multivariate analysis for origin traceability of soybean

Abstract

Soybean has developed a reputation as a superfood due to its nutrient profile, health benefits, and versatility. Since 1960, its demand has increased dramatically, going from a mere 17 MMT to almost 358 MMT in the production year 2021/22. These extremely high production rates have led to lower-than-expected product quality, adulteration, illegal trade, deforestation, and other concerns. This necessitates the development of an effective technology to confirm soybean’s provenance. This is the first review that investigates current analytical techniques coupled with multivariate analysis for origin traceability of soybeans. The fundamentals of several analytical techniques are presented, assessed, compared, and discussed in terms of their operating specifics, advantages, and shortcomings. Additionally, significance of multivariate analysis in analyzing complex data has also been discussed.

Keywords:

• Soy
• analytical techniques
• multivariate analysis
• geographical origin
• spectroscopy
• mass spectrometry

Introduction

Soy, popularly known as soyabean/soybean (Glycine max (L.) Merr.), is believed to have originated in east Asia and was domesticated in China around 1100 BC using its wild variety Glycine soja (Hirst 2019; Lee et al. 2011). Soybean is known as a “nutritional powerhouse” as it possesses a high protein content of 36% with all the nine essential amino acids (Michelfelder 2009), which accounts for approximately 60% of the world’s vegetable protein supply (Wijewardana, Reddy, and Bellaloui 2019). In addition, soybean comprises 35% carbohydrates, 19% lipids, and approximately 5% ash. Micro and macro-minerals (K, P, Mg, S, Cl, Na, Fe, Zn, Cu), vitamins (tocopherols, thiamin, riboflavin), phytins, and bioactive compounds (Isoflavones and anthocyanins) are among the other minor components (Liu 1997). Besides the nutritional value, soy has culinary qualities for humans and animals, health benefits, construction, and biofuel properties, and is inexpensive (Singh 2010). Soy is consumed in several forms, including soy sauce, tofu, edamame, soy milk, tempeh, miso, and edible oil. Approximately 77% of soy is processed into soybean meal as feed for animals farmed for human consumption (Charbeneau 2021; Ritchie and Roser 2021).

As the world’s population grows, so does the demand for soy protein-based foods. China has been the world’s top soybean importer since 2000, accounting for 65% of global soybean imports (Hasan, Suryani, and Hendrawan 2015; Chiang and Ikeda 2017). To meet these astronomically high demands, the global soybean production has increased dramatically over the years (Figure 1), reaching to 358 million metric tons (MMT) in 2021/2022 as compared to mere 17 MMT in 1960 (Hartman, West, and Herman 2011), with Brazil being the largest producer (129.5¹) and exporter (79), followed by the USA (121.5), Argentina (44), China (16.4) and India (12) (USDA 2023b). According to the USDA’s (United States department of Agriculture) most recent projections, Brazil’s soybean production will rise to 153 MMT, and global soybean production will reach approximately 391 MMT in 2022/2023 (USDA 2023c, 2023a). Due to these increased demands and production rate, the supply chains have been pushed, resulting in reduced product quality by utilizing questionable fillers and substitutes, food fraud and tax fraud by mislabelling, and deforestation to make more room for soy farming to meet the rising demands. Mato Grosso is Brazil’s largest soy producer and has some of the highest deforestation rates, accounting for 16% of Cerrado and 31% of Amazon deforestation over the past two decades (Vasconcelos et al. 2020). Countries including China, South Korea, and Japan are battling the problem of counterfeit soybean as the suppliers often mix authentic soybean with relatively inexpensive and low-quality foreign-acquired soybean (Lee et al. 2018; Oh et al. 2021; Otaka, Hokura, and Nakai 2014). Food falsification hurts the economy and threatens public health, as sometimes a non-compliant or unknown substitute additive can cause allergic reactions (Gerbig et al. 2017). Consumers, producers, and regulatory agencies have all emphasized the importance of food product authenticity since the 1980s (Rodríguez-Bermúdez et al. 2018). Food authentication is now more crucial than ever as consumers’ awareness of food safety grows, especially in the wake of the COVID-19 outbreak. Furthermore, lifestyle choices, religious beliefs, and health complications can restrict people from consuming certain foods, e.g., vegans avoid animal-based products; some communities don’t consume pork; people with specific allergies avoid certain foods such as gluten, nuts, dairy, and others. Conventional labeling is not entirely reliable and can be forged, thereby misleading consumers. This is another reason why consumers are compelled to know if the product’s content complies with the information provided on the packaging about the product composition, geographical origin, processing methods, and allergen alerts. Determination of geographical origin is an essential step in the food authentication process (Esteki et al. 2018). Deceptive labelling, health risks, tax fraud, and illegal deforestation are among the many reasons why establishing the geographical origin of soybeans is so important today.

Figure 1. Demonstrates the pattern of soybean production from 1960 to 2022 (Ritchie and Roser 2021; USDA 2023c).

All the components in soybean are substantially influenced by variety, cultivation practices (e.g., fertilizers), soil characteristics, and environmental factors (total rainfall and average annual temperature). This inherent benefit aids in distinguishing the geographical origin of the soybean, and a variety of analytical techniques can be employed to measure these variations in soybean (Dhungana et al. 2021; Ibáñez et al. 2020; Karr-Lilienthal et al. 2004; Lee and Choung 2011). The analytical techniques for verifying origin are transitioning from conventional geochemical methods to more rapid, sensitive, and inexpensive methods as new techniques demonstrate their reliability in determining the provenance of food products. Additionally, as chemometrics has developed, data extraction from these methodologies has improved in complexity, robustness, and accuracy. For example, spectroscopy techniques now provide complex chemical information and are no longer merely limited to providing structural information. Chemometrics methods enable spectral pre-processing, variable selection, pattern recognition, data dimension reduction, quantitative calibration, and multispectral data fusion (Wang et al. 2022). Several conventional and new advanced analytical techniques coupled with chemometrics have been reported for the origin determination of soybean, including the mass spectrometric techniques such as IRMS (Oh et al. 2021; Wu et al. 2015), ICP-MS (Sheng et al. 2020; Cui et al. 2021), HPLC-MS (Lee and Choung 2011; Wu et al. 2017), GC-TOF-MS and LC-orbitrap-MS (Lee et al. 2019); separation techniques such as HPLC (Azam et al. 2021; Desta et al. 2022), UPLC (Azam et al. 2021) and GC-FID (Desta et al. 2022; Lee and Choung 2011); spectroscopy techniques such as NMR (Zhou et al. 2021), NIR (Ahn and Kim 2012), FT-IR (Lee et al. 2018), terahertz spectroscopy (Wei, Zhu, et al. 2020) and, ED-XRF spectroscopy (Lai et al. 2020; Otaka, Hokura, and Nakai 2014). Furthermore, instrumental neutron activation analysis (NAA) has also been reported for the origin determination of soybean (Okada, Suzuki, and Hirai 1987).

Other potential DNA-based and spectroscopy techniques such as Raman (Zhu et al. 2018) and hyperspectral imaging spectroscopy (Yin et al. 2017) can effectively determine the origin but haven’t been reported previously for soybean. To the best of our knowledge, no review article has summarized the various analytical techniques to determine the geographical origin of soybean. In this review, several analytical techniques (Figure 2) combined with chemometrics for determining the geographical origin of soybeans are presented, assessed, compared, and discussed in terms of their operating specifics, benefits, and drawbacks.

Figure 2. Outline of analytical techniques discussed in this review.

Chemometrics in food traceability and authentication

Chemometrics is a multidisciplinary approach that employs multivariate statistics to extract relevant information from intricate chemical systems. Over the years, chemometrics has developed to the point that it can now be utilized to extract minute differences in measurements between the target products. Both qualitative and quantitative analyses are now possible in food authentication thanks to the advancements in multivariate data analysis. Qualitative methods can further be categorized into unsupervised and supervised machine learning methods. Both methods are relevant for feature selection and classification (Berrueta, Alonso-Salces, and Héberger 2007; Berry, Mohamed, and Yap 2020). The primary distinction between unsupervised and supervised learning methods is that the former categorizes the data and presents a pattern without any prior information about the groups. In contrast, the supervised chemometrics method latter uses a training set of data based on prior information assignment of the entities to a known group to construct a prediction model (Esteki et al. 2018).

Principle component analysis (PCA) and cluster analysis (CA) are unsupervised pattern recognition methods most employed. Linear discriminant analyses (LDA) and partial least square discriminant analyses (PLS-DA) are the most commonly used supervised techniques (Berrueta, Alonso-Salces, and Héberger 2007). The supervised technique by itself can be biased as the samples are always classified into one of the given categories, even if they do not belong to any of them, which is why the most common approach in food science research is to initiate with an unsupervised learning method followed a supervised learning technique (Berrueta, Alonso-Salces, and Héberger 2007; Berry, Mohamed, and Yap 2020).

Quantitative models are based on linear and non-linear regression supervised models. Both methods help establish the relationship between the independent and dependent variables and fit the model depending on the type of equation. Multiple linear regression (MLR), principal component regression (PCR), partial least squares (PLS) regression, and orthogonal partial least square (OPLS) regression are examples of linear regression. In contrast, artificial neural networks (ANN), support vector machine (SVM), and wavelet neural networks (WNN) are most commonly used to explain the non-linear relationship between the variables (Esteki et al. 2018; Guillén-Casla et al. 2011).

Separation techniques

Chromatographic methods, one of the most used separation techniques, permit separation, identification, and purification of the components of a mixture for qualitative and quantitative analysis. The separation of molecules is based on their chemical affinities, liquid-solid partitioning, liquid-solid adsorption, and variations in molecular weights. Column chromatography, thin-layer chromatography, gas chromatography, ion exchange chromatography, gel permeation chromatography, and high-pressure liquid chromatography are among the most used chromatographic techniques. The fundamental building blocks of chromatography techniques are the stationary phase, mobile phase, separated molecules, and a detector (Coskun 2016). Numerous chromatography methods that have been employed to establish the geographical origin of soybean are summarized in Table 1 below.

Table 1. The summary overview of all the chromatographic techniques used to determine soybean’s geographical origin.

Analytical techniques	Geographical origin and sample size	Traceability traits	Data processing techniques	Key results	Ref.
HPLC/PDA	India and Bulgaria	Isoflavones	ANOVA and LSD^2	The isoflavone content of soybeans from either origin did not differ significantly.	(Sakthivelu et al. 2008)
HPLC/DAD	The USA, Russia, Serbia, and China(n = 5 each)	Isoflavones	ANOVA and PCA	Origin had no influence on isoflavone composition. Isoflavone content of different soybean cultivars varied substantially.	(Tepavčević et al. 2010)
HPLC/PDA, GC-FID	The USA (n = 27), China (68), and South Korea (n = 109)	Isoflavones, phenolics, and tocopherol	GLM^3 and LSD	Soybean from these origins could be differentiated based on the given traceability traits.	(Kim et al. 2012)
UPLC-RID, GC, HPLC-UV	The USA (n = 115) and China (n = 218)	Soluble sugars, fatty acids, and total isoflavone	ANOVA, correlation analysis and PCA	Fatty acid, sucrose, and total isoflavone contents were significantly different in soybean from the USA and China.	(Azam et al. 2021)
HPLC-RID, GC-FID	South Korea, The USA, Japan, Taiwan, Nepal, and China	Isoflavone, fatty acids and free sugar	Duncan’s multiple-range test	Soybeans from Nepal could be differentiated from Chinese soybeans based on the isoflavone content.	(Lee and Choung 2011)
GC-FID, HPLC-UV-Vis	The USA (n = 11), China (n = 20), Japan (n = 6) and South Korea (n = 17)	Protein, oil, Fatty acids, and Isoflavones	ANOVA and HCA	Korean soybean had the highest average total isoflavone content.	(Desta et al. 2022)
GC-FID	Four Chinese regions: Beian, Harbin, Jian Sanjiang and Qiqihar	Palmitic acid, heptadecanoic acid, Stearic acid, cis-13-octadecenoic acid and arachidic acid	-NA-	Origin identification rate: 98%. Origin prediction rate: 93%.	(Xi-Chun 2018)
GC-FID	Northern China: Heilongjiang (n = 30), the Inner Mongolia Autonomous Region (n = 20), Jilin (n = 20), Liaoning (n = 21)	Oleic acid and Stearic acid	PCA, HCA and PLS-DA	Origin prediction precision: 90%.	(Sheng et al. 2020)

Liquid Chromatography (LC)

LC efficiently separates and identifies all types of organic compounds irrespective of their polarity and volatility, including proteins, amino acids, carbohydrates, lipids, nucleic acids, phenolic compounds, and other biologically active molecules (Lehotay and Hajšlová 2002). The more advanced high-performance (HP) and ultra-performance (UP) LC have mostly supplanted LC. Apart from the fundamental building blocks, a typical HPLC instrument consists of a high-pressure pump to ensure better separation. Numerous detectors, such as an ultraviolet-visible (UV-Vis) light absorbance detector, refractive index detector (RID), fluorescence detector, and mass spectrometer, are widely employed with HPLC. The basic working principle of an LC system is that components in the sample mixture move through the mobile and stationary phases at different rates as a result of their different chemical affinities, and depending on the stationary phase in the column, they can be separated based on their charge, molecular mass, polarity, and other characteristics. Another enhanced variation of LC that emerged from HPLC is UPLC. It operates at an even higher pressure than HPLC and supports particles smaller than 2 µm for packing material, which enables better separation in less time (Gumustas et al. 2013).

Owing to these advancements, the use of HPLC in food analysis has increased exponentially since 1969 (Lehotay and Hajšlová 2002). In addition, both HPLC and UPLC have been widely used in the field of food traceability (Di Cagno et al. 2003; Guo et al. 2013, 2017; Liu, Chen, et al. 2018; Rodrıguez-Delgado et al. 2002; Stefanoudaki, Kotsifaki, and Koutsaftakis 1997; Tomás-Barberán et al. 1993). Similarly, several studies have been conducted in the past decade that utilized HPLC and UPLC to examine several nutritional characteristics (e.g., isoflavones, soluble sugar) in multiple soybean cultivars from various regions to determine their association with the geographical origin (Azam et al. 2021; Zhang and Li 2009; Desta et al. 2022; Kim et al. 2012; Lee and Choung 2011).

Azam et al. (2021) combined UPLC-Refractive Index Detector (UPLC-RID) and HPLC-UV with chemometrics to analyze soluble sugar and isoflavones in soybean. The results demonstrated that USA soybean cultivars had significantly higher total soluble sugar and total isoflavone contents than Chinese samples. Similar patterns in the total isoflavones levels and soluble sugar were reported in previous studies (Desta et al. 2022; Lee and Choung 2011). Desta et al. (2022) utilized HPLC-UV to distinguish the soybean cultivars from the USA, China, South Korea, and Japan based on their total isoflavone contents (except gycitin). They demonstrated that South Korean samples had significantly higher total isoflavone content than other countries. This is in line with a previous study done by Kim et al. (2012), where they utilized an HPLC-photodiode array detector (HPLC/PDA) to analyze isoflavones and phenolic compounds in soybean germplasms from the USA, South Korea, and China. They were able to distinguish the samples from different origins based on these metabolites. In contrast to these investigations, other studies that used similar techniques found no significant variations in the isoflavones levels in soybean cultivars from India and Bulgaria (Sakthivelu et al. 2008), the USA, Serbia, Russia and China (Tepavčević et al. 2010). However, they could establish the differences in the multiple cultivars based on isoflavones.

The cost of an HPLC apparatus is reasonable, and the analysis is simple, robust, and reproducible. HPLC can detect materials from picograms to milligrams (McMaster 2007). Depending on the compounds to be analyzed, the specific detectors can be selected, and the instrument elements (e.g., stationary, mobile phase) can be easily customized (Luykx and van Ruth 2008). Additionally, the enhanced permeability and efficiency of the new and improved HPLC columns (e.g., silica-based monolithic) reduce the overall analytical time (Cabrera 2004). Despite the benefits, HPLC still has several disadvantages, such as the need for additional extraction and purification steps. In addition, it doesn’t have a universal detector (Wadood et al. 2020).

Gas chromatography (GC)

GC has been used to help determine food composition and is a great tool for food traceability and authentication. Unlike HPLC, GC can only be used for non-polar, semi-polar, volatiles, and semi-volatiles (e.g., sterols, oils, low-chain fatty acids, aroma components, and pesticides) without chemical derivatization. However, most researchers prefer to use GC to measure these compounds because of the higher separation efficiency (Lehotay and Hajšlová 2002; McMaster 2007). The basic operation of GC starts with injecting the sample mixture into the GC, where it is vaporized in a heated inlet. An inert gas then carries this gaseous mixture through the GC column’s stationary phase, where its components separate as they interact with the stationary phase. The abundance of components can then be measured by a range of detectors, including flame ionization (FID), atomic emission, thermal conductivity, photo-ionization detectors, and mass spectrometers (Luykx and van Ruth 2008).

Several studies have combined GC with chemometrics for the geographical origin traceability of a variety of food products including wine (Etièvant et al. 1989), milk (Collomb et al. 2002), edible oils (Ollivier et al. 2003), cocoa (Oliveira et al. 2016), and cereals (Tres, Heenan, and van Ruth 2014). Similarly, GC-FID with multivariate analysis has been widely applied to establish the geographical origin of soybean cultivars based on their fatty acid profiles (Azam et al. 2021; Desta et al. 2022; Sheng et al. 2020; Lee and Choung 2011; Xi-Chun 2018). Most of these studies analyzed five essential fatty acids, including palmitic acid (PA), stearic acid (SA), oleic acid (OA), linoleic acid (LA), and linolenic acid (LLA). Azam et al. (2021) found that four of the five measured fatty acids (PA, SA, OA, LA) positively correlated with geographical origin. In addition, Chinese cultivars showed higher total average contents of PA, SA, and OA, whereas the USA samples were rich in LA and LLA contents. Similar patterns (except SA) were observed by Desta et al. (2022). Desta et al. (2022) were also able to differentiate the soybean germplasms from the USA, China, and South Korea based on individual fatty acid contents. Interestingly, Lee and Choung (2011) showed contradicting patterns in the five fatty acids of soybean seeds from China and the USA. They concluded that the total average contents of SA and LLA correlated with the geographical origin. Xi-Chun (2018) also employed fatty acid fingerprint technology to detect and measure five fatty acids (PA, SA, heptadecanoic acid, cis-13-octadecenoic acid, arachidic acid) that could be used as traceability indicators for the four regions in Heilongjiang province (Beian, Harbin, Jian Sanjiang and Qiqihar). With respect to identification and prediction, they attained high rates of about 98% and 93%, respectively.

Based on the profiles of twelve trace elements and two fatty acids (OA and SA), Sheng et al. (2020) built an OPLS-DA model and differentiated the soybean samples from four producing areas in northern China. It must be noted that these studies analyzed different genotypes from multiple cultivation years, which could have resulted in the variations observed in the fatty acids. In addition, average annual temperature plays a significant role in the abundance of saturated and unsaturated fatty acids in different parts of plants (Sheng et al. 2020). Soybean germplasms from the USA, China, and South Korea have also been discriminated based on tocopherol concentrations using GC-FID (Kim et al. 2012).

Qualities like versatility, reproducibility, and, most importantly, higher sensitivity have made GC one of the most widely used analytical separation techniques. While GC equipment is now reasonably priced, its operational expenses continue to be a drawback. In addition, it requires a specialist to perform and interpret the data (Wadood et al. 2020). Furthermore, other than by pyrolysis, the nonvolatile substances, such as inorganic salts, proteins, polysaccharides, nucleic acids, and other large molecular weight organics, are no amenable to GC (Lehotay and Hajšlová 2002).

Mass spectrometric techniques

Mass spectrometry (MS) is a valuable analytical tool for identifying and quantifying the constituents in inorganic and organic compounds. MS’s fundamental premise is to generate ions from the sample, separate the ions based on their mass and measure their mass-to-charge ratio (m/z) to determine a sample’s composition. MS is often coupled with other techniques, such as geochemical and chromatographic techniques (Luykx and van Ruth 2008). Table 2 summarizes the mass spectrometry techniques that have been employed to determine the geographical origin of soybean.

Table 2. The summary overview of all the mass spectrometry techniques used to determine soybean’s geographical origin.

Analytical techniques	Geographical origin and sample size	Traceability traits	Data processing techniques	Key results	Ref.
IRMS	South Korea (n = 13) and China (n = 10)	δ^15N, δ^13C, δ^18O values	ANOVA and Duncan’s multiple range test	δ^15N values could not be used for origin provenance. No significant difference in the δ^18O values of soybean from South Korea and China.	(Oh et al. 2021)
IRMS	Heilongjiang, Shandong, and Jiangsu provinces of China(n = 18)	δ^15N, δ^13C, values	ANOVA and Duncan’s least significant test	The δ^13C values of soybeans from the three origins varied significantly.	(Wu et al. 2015)
ICP-MS	Vietnam (n = 15), Canada (n = 8), the USA (n = 9) and Brazil (n = 6)	33 trace elements	PCA and HCA	PCA could differentiate soybean from the four origins.	(Nguyen-Quang, Bui-Quang, and Truong-Ngoc 2021)
ICP-MS	Northern China: Heilongjiang (n = 30), the Inner Mongolia Autonomous Region (n = 20), Jilin (n = 20) and Liaoning (n = 21)	12 trace elements	PCA, HCA and PLS-DA	Origin prediction precision: 90%.	(Sheng et al. 2020)
ICP-MS	Chinese Heilongjiang Province: Daqing, Suihua, Heihe and Jiamusi(n = 116)	17 trace elements	OPLS-DA	Origin prediction precision in cross-validation: 93%.	(Cui et al. 2021)
ICP-MS	Argentina provinces: Chaco, Cordoba and San Luis(n = 120)	20 trace elements	SVM-DA	Origin prediction precision in cross-validation: 99%.	(Hidalgo et al. 2020)
HPLC-MS	Southwest China: Sichuan, Chongqing, Guizhou and Yunnan(n = 16)	12 Isoflavones and 8 anthocyanins	Correlation analysis and OPLS-DA	Origin prediction precision: 71%.	(Wu et al. 2017)
HPLC-DAD-ESI/MS	South Korea, USA, Japan, Taiwan, Nepal, and China	Isoflavones	Duncan’s multiple-range test	The isoflavone content in soybean from various sources varied significantly.	(Lee and Choung 2011)
GC-TOF-MS and UPLC-Orbitrap MS	7 Provinces of South Korea	Five metabolites (tryptophan, malonylgenistin, malonyldaidzin, N-acetylornithine, and allysine)	OPLS-DA	Origin prediction rate: 78%.	(Lee et al. 2019)
GC-MS	South Korea (n = 17), China (n = 13) and North America (n = 6)	52 Volatile compounds	PCA, HCA and PLS-DA	There were significant discrepancies between the volatile components of soybean from China and North America.	(Kim et al. 2020)
GC-MS	Argentina, the USA, Indonesia, Australia, Indian, Pakistan, Egypt, China, and Syria(n = 24)	88 Volatile compounds	PCA	-NA-	(Alghamdi et al. 2018)

Isotope ratio mass spectrometry (IRMS)

IRMS is a conventional and well-established geochemical method that uses the isotope content of compounds to identify their geographical, chemical, and biological origin. It is widely used in forensic science, biology, geology, medicine, and archaeology. In a typical EA-IRMS (elemental analyzer), samples are combusted at 1020 °C in the presence of oxygen to transform them into gaseous form. The resulting gases are then reduced and separated before being introduced into the ion source of the MS, where beams of ions are separated according to their m/z and finally detected (Rodrigues et al. 2013). IRMS precisely measures the variations in the abundance of isotopic ratios of elements such as ¹³C/¹²C, ¹⁵N/¹⁴N, ¹⁸O/¹⁶O, ²H/¹H, and ³⁴S/³²S. The isotope ratios are assessed relative to an isotopic standard to eliminate bias or systematic error in the measurements. Some internationally recognized standards include Vienna Pee Dee Belemnite (VPDB) for carbon, Vienna Canyon Diablo Troilite meteorite (V-CDT) for sulfur, Vienna Standard Mean Ocean Water (VSMOW) for oxygen and hydrogen, and laboratory air for nitrogen. Various natural abundant standards are provided by the International Atomic Energy Agency (IAEA; Vienna, Austria) and the National Institute of Standards and Technology (NIST; Washington, DC, USA). Isotope ratios are measured relative to the universal standards and are reported in the delta notation (δ), δ = (R_Sample/R_Standard − 1), where R_Sample is the abundance ratio of heavier and lighter isotopes (e.g., ²H/¹H) and standards chosen to establish the R_Standard values represent a stable and homogenous material (Muccio and Jackson 2009). δ -values are small and so are multiplied by 1000 and reported as per mille (‰).

Fertilizers, climatic variations, farming methods, and geological elements (such as soil characteristics and altitude) all affect a plant’s isotopic composition (proteins, carbohydrates, fats, minerals) and can serve as indicators of geographical origin (Li et al. 2016; Luykx and van Ruth 2008). Some of these factors affect specific stable isotopes, e.g., compared to plants produced using chemical fertilizers, plant materials fertilized with organic fertilizer or animal manure had greater δ¹⁵N values as these fertilizers are enriched in ¹⁵N (Watzka, Buchgraber, and Wanek 2006); δ¹⁸O values depends broadly on latitude the altitude of the production site (Luykx and van Ruth 2008).

A variety of plant and animal products have been authenticated and tracked using IRMS, including some essential commodities such as oils (Portarena et al. 2014), rice (Chung et al. 2021; Liu et al. 2022), coffee (Brodie, Kracht, and Hilkert 2019; Rodrigues et al. 2009), fruit juices (Perez, Smith, and Anderson 2006), meat (Piasentier et al. 2003; Schmidt et al. 2005), dairy products (Ehtesham et al. 2013; Kalpage et al. 2022), honey (McComb and Frew 2013; Zhou et al. 2018), wines (Ogrinc et al. 2001), and the world’s most expensive spice; saffron (Wakefield et al. 2019). To the best of our knowledge, merely two studies have been published on stable isotope analysis of soybeans to determine their geographical origin. Oh et al. (2021) measured δ¹³C, δ¹⁵N, and δ¹⁸O values for origin authentication of red pepper powders, chickpea, and soybean originating from South Korea, China, Vietnam, and Egypt. Soybeans were purchased commercially from South Korea and China without any information about the growing environment of the samples. Wu et al. (2015) conducted a similar study to measure δ¹³C and δ¹⁵N values of five kinds of cereal grains, including rice, soybean (imported from three Chinese regions: Jiangsu, Shandong, and Heilongjiang), millet, wheat, and corn, from different origins to investigate their geographical origin traceability. Both of these studies used analysis of variance (ANOVA) followed by Duncan’s least significant test to determine the significance (p < 0.05) of the results.

The results from both studies demonstrated that soybean from China had similar average δ¹⁵N values of 1.40 ± 0.32 ‰ (Wu et al. 2015) and 1.39 ± 0.82 ‰ (Oh et al. 2021). In addition, the δ¹⁵N values weren’t significantly different between the regions (p > 0.05). They concluded that δ¹⁵N could not be used for origin provenance because δ¹⁵N values are highly influenced by the soil nutrition and type of fertilizers and are not merely depend on the origin (Li et al. 2016; Wu et al. 2015). The low nitrogen isotopic composition of soybean indicates that the soybeans were grown using chemical fertilizers. However, it couldn’t be verified as the soil and fertilizer information was not accessible. Furthermore, due to the similar latitudes of South Korea and China, Oh et al. (2021) discovered no discernible variation between the δ¹⁸O values of soybean from these two countries.

Wu et at. (2015) showed a significant difference (p < 0.05) in the δ¹³C values of the soybean obtained from three different Chinese provinces. In contrast, Oh et al. (2021) found no statistically significant difference between the δ¹³C values of South Korean and Chinese soybean samples. In addition, the average δ¹³C values of Chinese samples from both studies were also different. It is difficult to pinpoint the precise cause of this discrepancy because the sample metadata was inaccessible. However, the variation in the ¹³C/¹²C could be explained based on a plants’ photosynthesis that affect the carbon fixation process. Wind stress and water usage efficiency can impact CO₂ diffusion in the leaves, potentially resulting in isotope fractionation in the crop and varying δ¹³C values (Farquhar et al. 1989; Kaler et al. 2018).

IRMS leverages certain environmental parameters (such as climate and altitude) to establish a fingerprint specific to a given product. However, similar climate conditions can occur in different locations making it difficult to distinguish based on their isotopic compositions. Other drawbacks of employing stable isotope analysis include the high cost of the equipment, its maintenance and sample preparation, and the time-consuming aspect of sample preparation and processing (Luykx and van Ruth 2008).

Inductively couples’ plasma mass spectrometry (ICP-MS)

ICP-MS is believed to be the fastest-growing analytical trace element technique due to its unique capabilities, such as the range of detection limit (from ultra-trace; ppq to higher quantities; ppm), rapid multi-element analysis, isotopic ability, and sensitivity. ICP-MS works on a principle that the samples (solid or liquid) are first converted to fine aerosols using a nebulizer, followed by ionization in a high-temperature argon plasma torch. The targeted ions are separated by a mass spectrometer (m/z) and finally sent to the detector. Numerous sectors, including environmental, biomedical, geological, semiconductor, and nuclear applications, have demonstrated the technique’s utility (Thomas 2008). ICP-MS, coupled with chemometrics, has also been extensively utilized in food authentication and traceability since it can analyze several inorganic elements simultaneously to produce a chemical fingerprint. This approach is predicated on the notion that an agricultural product’s chemical composition is influenced by the soil’s structural makeup, which is crucial to distinguish the geographical origin as no two soils would have the same structure (Nguyen-Quang, Bui-Quang, and Truong-Ngoc 2021).

ICP-MS has been successfully applied to screen the geographical origin of numerous food products, including vegetables (Ariyama et al. 2007; Laursen et al. 2011), rice (Kelly et al. 2002), wines (Gao et al. 2022; Geana et al. 2013), honey (Chen et al. 2014; Wu et al. 2021), coffee (Santato et al. 2012; Valentin and Watling 2013), tea (Liu, Zeng, et al. 2020), nuts (von Wuthenau et al. 2022), and cacao and cocoa beans (Bertoldi et al. 2016). The concentration of mineral elements in the soil, which is ultimately transferred to the crop, is influenced by the soil type and the capacity of certain elements to solubilize in the soil. For example, the chernozem soil is rich in Mg, Ca, and K, whereas black soil is rich in elements such as Fe, Zn, Mn, Cu, and Mo (Sheng et al. 2020). Alkali metals like rubidium and cesium are labile in soil, so they are readily transported to the plant. Such factors serve as excellent indicators of geographical origin (Kelly et al. 2002). Cui et al. (2021) demonstrated a significant correlation between the element composition of soybeans and that of the corresponding soil. Furthermore, factors like temperature and precipitation impact the soil’s organic matter and humus, impacting the crop’s elemental composition (Chung et al. 2021; Sheng et al. 2020). The concentrations of the soil’s mineral elements could also be affected by several other factors, such as environmental pollution, agricultural practices (e.g., fertilizers, irrigation), crop variety, and climate change. Still, these variations would be less pronounced than those seen between production sites if suitable elements were selected (Luykx and van Ruth 2008). Recent advancements have been made in the authentication of soybeans’ origin using ICP-MS. We have identified various studies conducted over the last two years using ICP-MS and multivariate analysis to establish the geographical specificity of soybeans (Sheng et al. 2020; Hidalgo et al. 2020; Nguyen-Quang, Bui-Quang, and Truong-Ngoc 2021).

Nguyen-Quang, Bui-Quang, and Truong-Ngoc (2021) used ICP-MS with PCA and hierarchical clustering analysis (HCA) to identify 33 out of 40 examined elements in soybean from Vietnam, Canada, the USA, and Brazil as essential markers for soybean origin detection. The HCA successfully differentiated the samples from four regions with a slight overlap between the Canadian and the USA samples. PCA modeling clearly clustered samples from four countries and was also able to determine discriminant elements as some of the elements were dramatically higher in specific locations, e.g., Ti, Mn, Zn, Mo, Dy, and Tl concentrations were significantly higher in Brazilian samples as compared to the other regions. Similarly, contents of Se, Sr, Nb, and Ba were highest in Vietnamese samples, Canadian samples typically possessed the highest levels of Lu, and the USA samples stood out due to their highest concentrations of Ti and Hf.

Sheng et al. (2020) combined gas chromatography and ICP-MS to measure five fatty acids and 17 trace elements in soybean from four production sites in northern China (Heilongjiang, Inner Mongolia Autonomous, Liaoning, and Jilin). HCA revealed the key elements unique to each production region. Mg, Ca and Pd contents were found to be highest in soybeans from Heilongjiang, whereas the soybean from Liaoning possessed the highest concentration of Zn, Mn, Se, and C18:1. Similarly, the concentrations of Mo and Cr were measured in the soybean from Inner Mongolia Autonomous region and Jilin regions respectively. Employing 14 classification indexes (Mn, Se, Ni, Sr, Cd, Pd, Mg, Fe, Ca, Zn, Co, Rb, C18:1, and C18:0), the PLS-DA model was able to predict 89.9% of the sample’s origin accurately. This study used yearly temperature, precipitation, and soil organic matter to explain the fluctuation of various elements in various producing regions. The results, however, may have been skewed due to not considering the impact of different soybean varieties in different geographical areas.

Another study by the same author, Cui et al. (2021), examined the elemental composition of soybean and soil samples from four production regions within the Heilongjiang province (Daqing, Suihua, Heihe, and Jiamusi) of China. They established a geographical traceability system using the difference in the element content between each production area, correlation analysis of all variables in soybean and soil, cluster heat map, and the orthogonal partial least squares discriminant analyses (OPLS-DA). The cluster heat map utilizing 17 (Mg, Al, K, Mn, Mo, O, Cu, Cr, Rb, Ni, Ca, Fe, Se, Pd, Zn, Ga, and Pd) of the 25 analyzed trace elements contained sufficient discriminating ability, and OPLS-DA model showed a high prediction percentage of 93.33. Furthermore, they demonstrated that, as a result of bidirectional absorption aid and unidirectional absorption inhibition, respectively, there is a positive and negative correlation between the elemental composition of the soybean and the soil. They also successfully reduced the sampling area from approximately 1.97 million Km² (Sheng et al. 2020) to 143,985 Km², i.e. approximately 13 times, and presented a more effective and sensitive model to discriminate the samples from the production area within a single province.

Hidalgo et al. (2020) collected soybeans from Argentina’s three provinces (Córdoba, San Luis, and Chaco) and measured a panel of 20 trace elements using ICP-MS. They compared many supervised and unsupervised classification and discriminant techniques such as PCA, support vector machine (SVM) discriminant model (SVM-DA), soft independent modeling of class analogy (SIMCA), k-nearest number (k-NN), random forest (RF), potential functions (PF), one-class support vector machine (OC-SVM). The results demonstrated significant variations in Cr, Li, Mn, and Zn concentrations (p < 0.01) between the three provinces. PCA discriminated the samples from Chaco and showed a slight overlap between the other two origins. The supervised technique, SVM-DA, could even differentiate these two neighboring regions (San Luis and Córdoba) with an accuracy of 91.7%. They concluded that SVM-DA and SIMCA outperformed all the other tested models in terms of prediction ability in cross-validation with average values of 99.3% for SVM-DA and a median value of balanced accuracy of 96% for Chaco, 92% for Córdoba, 88% for San Luis using SIMCA.

Homura et al. (2006) used LDA to distinguish between Japanese and Chinese soybean samples based on six chosen elements; Baoxin and Dongjie (2017) examined 52 mineral elements in whole soybean samples, produced a PCA-DA model based on four elements, and successfully discriminated the samples from Qiqihar and Beian regions of China’s Heilongjiang province; Hua Jao et al. (2017) were able to determine the correct production region of the vegetable soybean samples (both kernel and pods) collected from Taiwan, China and Japan based on eight trace elements.

ICP-atomic emission spectroscopy (ICP-AES), Atomic absorption, x-ray fluorescence spectroscopy (discussed in the later section), and neutron activation analysis (NAA) are additional methods for trace elemental analysis. ICP-AES was used by Wan et al. (2010) to distinguish soybean from the Chinese provinces of Jiangxi, Heilongjiang, Jilin, and Anhui. They concluded that the production region significantly impacts the concentrations of Al, Ca, Fe, Na, Ma, and Zn in soybeans (p < 0.05). Okada, Suzuki, and Hirai (1987) nuclear study using instrumental NAA demonstrated that the concentration of selenium in soybean from China, the USA, and Japan differ significantly and can be used as an indicator of geographical origin. These techniques haven’t been used widely due to their drawbacks, such as high detection limit, large sample size, high cost of equipment, inaccessibility of a nuclear reactor, and radioactive hazards (Kučera, Pp, and In 2009; Misof, Roschger, and Fratzl 2011; Wilschefski and Baxter 2019). ICP-MS has largely overcome these limitations; however, it still has shortcomings, much like IRMS. The main drawback is the expensive equipment, operating, and sample preparation costs. Additionally, it is susceptible to both spectroscopic and non-spectroscopic interference. Thus, great caution must be used to obtain good data (Wilschefski and Baxter 2019).

Liquid Chromatography Mass spectrometry (LC-MS)

LC coupled with an MS provides high sensitivity, selectivity, and resolution to measure both non-targeted/targeted metabolites in a wide range of compounds (Dong and Llanas 2013; Luykx and van Ruth 2008). HPLC-MS has been used widely in the field of metabolomics for drug bioanalysis and to discover and develop new medicines (Wilson et al. 2005).

The quality characteristics of soybeans are primarily affected by genotype, origin, maturity groups, climatic conditions (temperature, precipitation, rainfall), soil properties, and water (Azam et al. 2021; Desta et al. 2022; Lee et al. 2019; Zhang et al. 2014) and these could act as unique and identifiable markers for the origin. Based on the examination of nonvolatile metabolite concentrations including proteins, nucleotides and secondary metabolites such as isoflavones, and anthocyanin, HPLC-MS has demonstrated its dependability in the field of traceability, e.g., for origin authentication of coffee (Choi et al. 2010), tea (Fraser et al. 2013), rice, oils (Nescatelli et al. 2014), wine (Sagratini et al. 2012), and juices (Jandrić et al. 2017).

Similarly, multiple studies have combined LC with MS to examine the secondary metabolites (e.g., isoflavones, anthocyanins), in multiple soybean cultivars from various regions to determine their association with the geographical origin (Lee et al. 2019; Lee and Choung 2011; Wu et al. 2017). Lee and Choung (2011) used HPLC with diode array detection and electrospray ionization/mass spectrometry (HPLC-DAD-ESI/MS) to screen and differentiate soybean varieties from the USA, China, South Korea, Japan, Taiwan, and Nepal. They found no significant correlation between the isoflavone content and the country of origin. However, they discovered a positive correlation between the isoflavone concentration and soybean varieties from each country. Other studies also demonstrated a correlation between the isoflavones and geographical origin (Lee et al. 2019; Wu et al. 2017). Lee et al. (2019) coupled UPLC-orbitrap-MS with chemometrics to examine the high geographical specificity of metabolomic traits among the Korean samples collected from 7 provinces (Paju, Inje, Boeun, Gunsan, Gangjin, Daegu, and Hamyang). PCA analysis on flavonoids and isoflavonoid contents was able to distinguish between the Gunsan, Inje, and Daegu regions, and HCA was able to establish a correlation between the regions (Paju and Boeun; Hamyang and Gangjin; Inje and Daegu) considering the geographical characteristics (rainfall, average temperature, and soil texture) and soybean varieties. Tryptophan, malonylgenistin and malonyldaidzin, N-acetylornithine, and allysine were the five discriminating metabolites prioritized using the OPLS-DA model, which had a fitness and predictability of 95% and 78%, respectively.

Wu et al. (2017) differentiated the black soybean seeds collected from four different Chinese regions: Sichuan, Yunnan, Guizhou, and Chongqing, based on isoflavone and anthocyanin content. They utilized HPLC-MS and profiled twelve isoflavones and eight anthocyanins. Using the combined data from the geographical analysis and OPLS-DA model, they concluded that longitude and latitude significantly influence the accumulation of these secondary metabolites. Samples grown at higher latitudes showed a higher accumulation of isoflavones and anthocyanins (e.g., Sichuan and Chongqing). Similar observations were made by Zhang and Li (2009). Wu et al. (2017) also observed a biosynthesis tradeoff between the isoflavones and anthocyanins, which may be affected by the longitude. Samples from Guizhou and Chongqing, grown in a high-longitudinal environment, were found to have higher levels of isoflavones. In contrast, higher anthocyanins content was observed in the sample grown in the low-longitudinal areas of Sichuan.

LC-MS has been used in all of these investigations because it is versatile, accurate, and highly reproducible for profiling and quantitative analysis; nevertheless, there are some drawbacks, such as extensive extraction procedures, complex data analysis, compound identification challenge costly equipment and maintenance expenses, and lower separation efficiency than gas chromatography (Kamal and Karoui 2015; Timchenko 2021).

Gas Chromatography Mass Spectrometry (GC-MS)

Like LC-MS, GC combined with an MS provides comprehensive information about the identity of components in a sample mixture. GC-MS best quantifies the metabolites that are volatile and thermally stable (Luykx and van Ruth 2008). Due to its specificity, GC-MS has established its importance in metabolomics-based research (Beale et al. 2018). The cultivation and post-harvest conditions, including temperature, precipitation, light, water, and soil characteristics, fertilizer treatments have a significant impact on the volatile organic compounds of soybeans; therefore, they can be used to distinguish the soybean from different growing regions (Kim et al. 2020; Maestri et al. 1998). Thus, GC-MS has been widely applied in the field of authentication and traceability of oils (Gonzalez-Fernandez et al. 2019), fruits (Giannetti et al. 2017), tea (Lin et al. 2013), Saffron (Liu, Chen, et al. 2018), dairy products (Gaspardo et al. 2010; Palladino et al. 2010), and other animal and agricultural products (Kamal and Karoui 2015).

Numerous studies have effectively identified the region of soybean production by analyzing and profiling volatile compounds such as aldehydes, terpenes, ketones, alcohols, phenolic compounds, and fatty acids (Alghamdi et al. 2018; Kim et al. 2020; Lee et al. 2019). By combining GC-MS and multivariate analysis (PLS-DA), Kim et al. (2020) were able to identify 52 discriminating volatile compounds in soybean from South Korea (15), China (25), and North America (12) and could establish the association of soybean to their origin. Similarly, Alghamdi et al. (2018) identified 88 volatile compounds categorized into 13 classes, including aldehydes, ketones, alcohols, carboxylic acids, esters, alkanes, heterocyclic compounds, phenolic compounds, sugar moiety, ether, amide, alkene and fatty acid ester in soybean genotypes from nine countries (the USA, Egypt, Syria, China, India, Pakistan, Indonesia, Argentina and Australia). They found that Indonesian and Australian samples had the highest amounts of phenolic and ester compounds, respectively. In a subsequent study, Lee et al. (2019) used an OPLS-DA model to identify five distinguishing metabolites in seven Korean provinces.

GC-MS is considered a versatile analytical technique due to its ability to separate the components with high selectivity, sensitivity, and reproducibility. The well-established libraries, which facilitate compound identification, are the most compelling advantage of GC-MS. Among its other advantages, it is easy to use in terms of analysis time and running costs. In addition, GC-MS minimizes the problems common to LC-MS, such as matrix effects, and achieves greater chromatographic resolution (Beale et al. 2018). However, similar to LC-MS, the instrumentation is expensive, and analysis is time-consuming. The sample preparation for GC-MS is extremely particular because the MS detector is designed to analyze clean materials (Luykx and van Ruth 2008). Furthermore, GC-MS can only be used to separate and identify low molecular weight (ca. 50–600 Da) and volatile compounds (Beale et al. 2018). Moreover, chemical derivatization is needed to measure polar and nonvolatile compounds using GC (Raynie 2018).

Spectroscopy techniques

Spectroscopy techniques are based on the premise that molecules and atoms can interact with electromagnetic radiation. By using the wavelength or frequency detected in the emitted or absorbed energy spectrum, they reveal intricate structural and physicochemical details of the compounds in the sample (Ibañez and Cifuentes 2001). The available spectroscopy techniques include nuclear magnetic resonance (NMR), energy dispersive X-Ray fluorescence spectroscopy (ED-XRF), and a wide array of vibrational spectroscopy techniques, such as near-infrared (NIR), mid-infrared, terahertz spectroscopy and others. The primary benefits are that spectroscopic techniques are nondestructive, noninvasive, affordable, rapid, and can be used for qualitative and quantitative analysis. Combined with chemometrics, these techniques have proven effective for assessing authenticity and detecting adulteration (Wadood et al. 2020). Table 3 summarizes the spectroscopy techniques that have been employed to determine the geographical origin of soybean.

Table 3. The summary overview of the spectroscopy techniques to determine the geographical origin of soybean.

Analytical techniques	Geographical origin and sample size	Traceability traits	Data processing techniques	Key results	Ref.
^1H NMR	North-eastern, middle, and southern regions of China (n = 9) and South Korea	9 metabolites, 11 metabolites within 3 regions of China, Kernel size	OPLS-DA	Origin prediction rate: 94% Origin classification accuracy: 96%	(Zhou et al. 2021)
Low-field NMR (10 mhz)	China (n = 24) and South Korea (n = 25)	Crude protein and relaxation times	Canonical discriminant analyses	Origin classification accuracy: 96%	(Kim, Rho, and Lee 2009)
Low-field NMR (20 mhz)	South Korea (n = 22) and other countries^4 (n = 34)	Crude fat, Moisture, crude protein, and relaxation times	Long-term discriminant analyses	Origin classification accuracy: 70-92%	(Rho et al. 2003)
FTIR	South Korea (Upper side, Right side, Left side) (n = 8) and China (South-eastern, North-eastern, and Eastern regions) (n = 13)	Three wavenumber regions	PLS-DA and HCA	Predictability: China Vs South Korea: 94%, within China: 89%, and within South Korea: 97%.	(Lee et al. 2018)
NIR	South Korea, China (n = 22), the USA (n = 6), Taiwan (n = 6), Australia (n = 3), Russia (n = 2), Canada (n = 2), Brazil (n = 1), and Ukraine (n = 1)	-NA-	PCA and MPLS	Origin prediction rate: 98%	(Ahn and Kim 2012)
FT-NIR	South Korea (n = 234) and other countries^1 (n = 237)	Eight wavenumber regions	PLS	Origin prediction rate: 96%	(Lee et al. 2022)
NIR and ED-XRF	South Korea (n = 37) and other countries^1 (n = 75)	Al, P, S, Cl, Ca, Fe, Ni, Cu, Zn, Br and Rb	PCA, MPLS, and canonical DA	Correct classification rate: 99%	(Kim et al. 2022)
ED-XRF	South Korea (n = 100) and imported (USA, China, Vietnam, and Thailand; n = 97)	Five trace elements (P, Ni, Br, Zn and Mn)	Canonical DA	Accuracy in discriminating south Korean soybean from other countries: 100%. Accuracy in discriminating soybean within south Korea: 93%	(Lee et al. 2020)
ED-XRF	Northern China: Henan, Inner Mongolia, Xinjiang, Heilongjiang, and Liaoning(n = 296)	Nine trace elements (Mg, K, Ca, Mn, Fe, Ni, Cu, Zn, and Rb)	PCA and MLP	Origin classification accuracy: 96%	(Lai et al. 2020)
ED-XRF with 3 D polarization optics	Japan (n = 23), the USA (n = 8), Canada (n = 7), and China (n = 7)	Eight trace elements	PCA and LDA	Origin prediction precision: 91%	(Otaka, Hokura, and Nakai 2014)
Terahertz spectroscopy	Argentina (n = 25), the USA (n = 9), and China (n = 41)	-NA-	iPLS, ELM, GA-SVM, ABC-SVM	Origin prediction rate: 95%	(Wei, Zhu, et al. 2020)

Nuclear magnetic resonance (NMR)

NMR is one of the most used techniques in metabolomic fingerprinting research because of its numerous advantages. NMR is noninvasive, environmentally friendly, and requires relatively simple sample preparation. It is based on the measurement of radiofrequency radiation absorbed by the nuclei of the atoms present in a strong magnetic field. The absorption by atomic nuclei is influenced by the atoms in their immediate environment, and this effect is utilized to acquire structural data about the compound under study. The most employed isotopes in NMR are ¹H and ¹³C, with ¹⁵N, ¹⁷O, ¹⁹F, and ³¹P, among others (Wadood et al. 2020). Depending on the magnetic field strengths, low and high field NMR-based metabolomics are widely used for food authentication and traceability. Factors such as pedoclimatic conditions of production areas, sample variety, year of production, and seasonality affect the agro-products’ nutritional characteristics, which can also be measured and quantified using NMR (Bellaloui et al. 2009; Masetti, Sorbo, and Nisini 2021). The geographical origin of various plant and animal products has been identified by profiling metabolites using NMR (Aursand et al. 2009; Brescia et al. 2003; Consonni, Cagliani, and Cogliati 2012; Huo et al. 2017; Mannina et al. 2001). Similarly, three studies applied coupled NMR and chemometric techniques to differentiate soybean from South Korea and China (Kim, Rho, and Lee 2009; Rho et al. 2003; Zhou et al. 2021).

Zhou et al. (2021) analyzed twenty-five metabolites, including ten amino acids, three carbohydrates, nine organic acids, one fatty acid, and others, and developed an OPLS-DA model with a sensitivity, specificity, and accuracy of 96.9%, 94.4%, and 95.6% respectively and discriminated the soybean samples from South Korea and China. They also distinguished soybean from the regions within China (north-eastern, middle, and southern regions) using a PLS-DA model that showed more than 95% accuracy. However, the samples within Korean provinces could not be differentiated, possibly because the region of origin had similar climatic, soil, and rainfall conditions. In addition, no information regarding the soybean cultivars was available to the author and wasn’t considered in this study. Using low-field NMR (10 MHz) and performing the canonical discriminant analysis (96% accuracy), Kim, Rho, and Lee (2009) found that Korean soybean had higher relaxation times and crude protein content (p < 0.05) as compared to the Chinese soybean. In a similar study, soybean samples were discriminated based on relaxation times measured using general components and 20 MHz pulsed NMR. In particular, yellow beans showed a significant difference in the crude fat contents whereas black beans showed a considerable variation in the moisture and protein contents between the countries (Rho et al. 2003).

The most appealing feature of NMR spectroscopy is the ability to analyze and quantify several chemical compounds in a single run using the same sample. However, it is still expensive compared to other spectroscopy techniques and requires an expert operator (Wadood et al. 2020). NMR’s main limitation compared to MS is its low sensitivity. MS has detection limits in the nanomolar range. MS is also applicable for targeted and untargeted analyses, whereas NMR is often employed for nonselective assays (Emwas 2015).

Infrared spectroscopy (IR)

When exposed to a specific wavelength of light that creates a vibration in the bonds, each functional group or structural feature of a sample absorbs infrared at different intensities that can be detected by IR spectroscopy. Since the vibration frequencies of the various bonds in the functional groups differ, if we consider them all, they provide a fingerprint of the sample that can reveal the potential compositional and/or structural identity of the sample (Luykx and van Ruth 2008). Two principal zones of the IR spectrum are widely used for analytical purposes, the mid-infrared (mid-IR) region from 4000 to 400 cm⁻¹ and the near-infrared (NIR) from 14,000 to 4000 cm⁻¹ (Ibañez and Cifuentes 2001). Absorption in the NIR region mainly appears as combination and overtone bands of molecular vibrations of -CH, -OH, and -NH functional groups typical of oil, carbohydrates, and protein components. Fourier transform infrared (FT-IR) spectroscopy, on the other hand, relies on the mid-infrared region of the electromagnetic spectrum. It provides the fundamental vibrational information and associated rotational-vibrational structure of the sample instead of overtones and combination modes. The mid-infrared region can be split into two broad regions: functional (4000–1500 cm⁻¹) and fingerprint (1500–400 cm⁻¹) regions. The former provides extensive information about the sample, including the major functional groups present in the sample. The latter contains various overlapped bands generated by different compounds in this region making it harder to identify individual functional groups without any uncertainty. However, this region can still be used as a fingerprint of a food product (Karoui, Downey, and Blecker 2010; Wadood et al. 2020).

Multivariate statistical analysis must be utilized to interpret the spectroscopy data since the NIR and FT-IR spectra of soybeans from different origins are superficially similar but contain a substantial amount of hidden and complex information about the sample. Furthermore, multivariate analysis is necessary for data pre-processing for any sample background, noise, light scattering, device response, and so forth (Rinnan, Berg, and Engelsen 2009; Wei, Zhu, et al. 2020). Both FT-IR and NIR combined with chemometrics have been employed to determine the geographical origin of various commodities, including rice (Kim et al. 2003), olive oils (Tapp, Defernez, and Kemsley 2003), dairy products (Bassbasi et al. 2014; Karoui et al. 2004), coffee (Medina et al. 2017), wines (Hu et al. 2019), tea (Esteki, Memarbashi, and Simal-Gandara 2022), alcoholic beverages (Uríčková and Sádecká 2015), saffron (Liu, Chen, et al. 2018), and honey (Ruoff et al. 2006).

By integrating NIR spectroscopy and chemometrics, Kim et al. (2022) distinguished Korean soybeans from imported soybeans with a correct classification rate of 100%. They also experimented with various spectral pre-processing approaches to create a model using DA and modified partial least squares OPLS(MPLS) regression techniques and discovered that the classification rate could range from 90% to 100%. Ahn and Kim (2012) conducted a similar analysis and successfully distinguished between Korean and imported soybeans with a discriminating rate of 98%. Using a PLS algorithm based on eight wavenumber regions, Lee et al. (2022) found that FT-NIR had about 96% prediction accuracy in separating Korean and imported soybeans. Similar prediction accuracy was reported by Kang et al. (2016). They successfully discriminated Korean soybeans in cheonggukjang (a traditional Korean dish made using fermented soybean) from imported soybeans (China and the USA). Tangendjaja (2020) successfully discriminated soybean meal (SBM) from Argentina, Brazil, and the USA. They measured proximate amino acids, metabolizable energy, and carbohydrate composition of SBM using near-infrared reflectance spectroscopy and applied a stepwise regression technique to select the distinct variables. They concluded that Brazilian SBM possessed higher protein (2%) and amino acid content than the USA and the Argentine SBM. Azam et al. (2021) also used NIR and concluded that protein and oil contents were higher in China and the USA soybeans, respectively. Previous studies made similar conclusions (Grieshop and Fahey 2001; Lee and Choung 2011).

One published study used FT-IR and multivariate statistical analysis to distinguish between Chinese and Korean soybeans (Lee et al. 2018). This study also revealed the regions within China and South Korea where the soybeans were cultivated by analyzing the regional characteristics without considering the variety. Based on three wavenumber regions, they developed a PLSR model and experimented with different scaling and normalization techniques, and VIP-cutoff values. They found that when a VIP cutoff was not used, discriminating against the country of origin (China vs South Korea) was more precise (93.5%). In contrast, a VIP cutoff value of 1.5 was optimal for discriminating the origin of soybean within a country.

FT-IR and NIR spectroscopy provides a cheap, rapid, and noninvasive way of analyzing samples to find their region of origin. In addition, they can detect numerous components in a single scan. As compared to FT-IR spectroscopy, the sample preparation for NIR spectroscopy is simpler. If a shorter wavelength area is used, sample homogenization is not even necessary. However, if a qualitative analysis is the objective, the samples’ moisture content needs to be regulated (Kawano 2001). Both of these techniques have a few limitations too. For example, NIR has low sensitivity associated with a high signal-to-noise ratio plus low selectivity of NIR spectra due to the superposition of various overtones and combination bands in the NIR region (Wadood et al. 2020). As the particle size decreases, the intensity and area of IR bands increases, therefore the sample preparation for FT-IR needs to be carefully done and can be time-consuming (Udvardi et al. 2017; Yang and Fateley 1986). Water is a significant absorber in the MIR spectral region and can interfere with determining other components present in food systems (Karoui, Downey, and Blecker 2010).

Energy dispersive X-Ray fluorescence spectroscopy (ED-XRF)

ED-XRF is a spectroscopy-based technique for trace element analysis and a great alternative to ICP-MS. The basic operation of an ED-XRF spectrometer begins with irradiating a sample using an X-ray source which ionizes the sample and produces X-ray fluorescence unique to each inorganic element. A semiconductor detector detects this fluorescence, and the various energies are separated and processed at a multi-channel analyzer. Finally, the analogue signal is converted to a digital signal, and both qualitative and quantitative analysis occur (Yao, Wang, and Zhao 2015). ED-XRF was initially used mainly for metal authentication. It has recently been proven effective for screening purposes in organic food control, geographical origin authentication, adulteration, and detecting the impact of manufacturing procedures on the food product (Ghidotti, Papoci, et al. 2021). Numerous food products have been geographically authenticated using ED-XRF, including olive oil (Lia, Zammit Mangion, and Farrugia 2020), honey (Ghidotti, Fiamegos, et al. 2021), tea (Rajapaksha et al. 2017), milk (Potocnik et al. 2016), cocoa powder (Herreros-Chavez, Cervera, and Morales-Rubio 2019), and sesame seeds (Choi et al. 2016).

Multiple studies analyzed and quantified several trace elements for the provenance analysis of soybeans using ED-XRF (Kim et al. 2022; Lai et al. 2020; Lee et al. 2020; Otaka, Hokura, and Nakai 2014). Kim et al. (2022) demonstrated the highest correct prediction rate of 99% while employing the canonical DA based on 11 elements. They revealed that Al, P, S, Cl, Ca, Fe, Ni, Cu, Zn, Br, and Rb concentrations were significantly different (p < 0.05) between the Korean and imported soybeans. Similarly, Lee et al. (2020) were also able to differentiate between Korean and imported soybeans by using just five elements (P, Ni, Br, Zn, and Mn) with a prediction efficiency of 93.4%. With a high prediction efficiency of 95%, Kang et al. (2016) could even distinguish the fermented Korean soybeans from the imported soybeans (from China and the USA) used in a traditional Korean dish called cheonggukjang.

Otaka, Hokura, and Nakai (2014) found a significant difference in 8 elements (Mg, P, Cl, K, Mn, Cu, Br, and Ba) in soybean sourced from Japan and other countries (the USA, Canada, and China). Using LDA, they could discriminate the geographical origins with a prediction accuracy of 91%. According to Lai et al. (2020), soybean from the five provinces of China (Henan, Inner Mongolia, Heilongjiang, Xinjiang, and Liaoning) could be differentiated by performing a multilayer perception (MLP) analysis (96.2% prediction efficiency) based on 8 trace elements (K, Ca, Mn, Fe, Ni, Cu, Zn, and Rb).

The minimal sample preparation is a significant factor in the growing use of ED-XRF spectroscopy and gives it a significant advantage over methods like ICP-AES/MS. Before analysis, the sample must be adequately homogenized, which is the sole substantial sample treatment requirement. Another advantage of ED-XRF spectroscopy over ICP-MS is that it is simple and rapid. In addition, it is noninvasive, the results are highly reproducible, and an operator specialist isn’t necessarily required, making it easier to use at daily foodstuff markets. Compared to ICP-MS methods, the sensitivity of ED-XRF spectroscopy is still a limitation. However, Otaka, Hokura, and Nakai (2014) demonstrated that, except for light components Mg, Al, and P, the findings measured by XRF had standard deviations of less than 10%, and the results acquired via XRF and ICP-MS were in good agreement. The most common way to decrease the detection limit is to increase the irradiation time at the expense of the sample throughput. ED-XRF spectroscopy can be used to analyze a wide range of matrices. Once calibration curves are created, they can easily be corrected for the matrix effects using mathematical algorithms such as fundamental parameters (Ghidotti, Papoci, et al. 2021). The ED-XRF method can even be validated without needing matrix-matched calibration curves (Fiamegos and Guntiñas 2018).

Terahertz spectroscopy

Terahertz (THz) region refers to a very small gap between the microwave (MW) and IR regions of the electromagnetic spectrum with a frequency range of 0.1 to 10 THz (3.3 cm⁻¹ to 333.6 cm⁻¹). The energy of photons in the THz range makes it possible to analyze the vibrational activity of molecules using THz spectroscopy. THz also has distinctive qualities in addition to reflecting some IR and MW features (Afsah-Hejri et al. 2019). For example, THz radiation is non-ionizing and can also travel in the field of vision, and like MW, it can penetrate a broad range of non-conductive materials. The spectral responses in the THz frequency range comprise evidence of intramolecular and inter-molecular motions, which are beneficial to reflect a single-molecular structure and molecular arrangements. Similar to other spectroscopy techniques, the radiation source (electron beam), lenses or polarizers for manipulating the radiation, and the detector are the three fundamental components of the THz spectrometer (Patil et al. 2022).

In 1960, THz radiation was employed for the first time to generate astronomical images. It wasn’t until 1995 that THz time-domain spectroscopy (THz-TDS) depicted its use in food inspection (Sethy et al. 2015). Over the last two decades, THz spectroscopy has been applied in a wide range of fields, including security screening, astronomical studies, communication, medical imaging, pharmaceuticals, agriculture, environmental monitoring (Afsah-Hejri et al. 2019), adulteration detection (Liu, Zhang, et al. 2020), and to differentiate cultivars (Chen et al. 2016). Furthermore, it has recently been used to classify the geographical origin of food products such as wheat (Shen et al. 2022), olive oils (Liu, Liu, et al. 2018), coffee beans (Yang et al. 2021), tea (Li, Dai, et al. 2017), and honey (Liu, Zhang, et al. 2018).

Multiple studies have identified soybean cultivars, including transgenic soybeans using THz spectroscopy (Liu et al. 2016; Luo et al. 2019; Tu et al. 2021; Wei, Zheng, et al. 2020). However, to the best of our knowledge, only one study has coupled THz-TDS with chemometrics to successfully establish the geographic origin of soybean (Wei, Zhu, et al. 2020). They employed iPLS to optimize THz-TDS spectra and developed three identification models; extreme learning machine (ELM), genetic algorithm support vector machine (GA-SVM), and artificial bee colony algorithm support vector machine (ABC-SVM) combined with 8 pre-processing techniques. They concluded that ABC-SVM with auto-scaling was the best model to differentiate the geographical origin with an identification accuracy of approximately 95%.

Like infrared spectroscopy, THz spectroscopy is green, noninvasive, rapid, and doesn’t require extensive sample preparation (Liu et al. 2016). However, it contrasts with infrared approaches and hyperspectral imaging in that it can provide information on both low-frequency intramolecular and intermolecular modes of molecules (Liu, Chen, et al. 2018). Furthermore, THz waves can penetrate through various materials and have low photon energy, which lowers the risk of possible sample damage. Moreover, the waves are coherent, which offers them an edge over FTIR spectroscopy (Patil et al. 2022). Compared to FTIR and Raman spectroscopy, the coherent character of this approach amplifies its sensitivity and dynamic range (GuoZhong et al. 2012). The non-ionizing nature of THz radiation makes it a great and safer alternative to X-rays in generating high-resolution images from the inner parts of a solid object. One of the main drawbacks of the THz technique is that the THz radiation can be entirely absorbed by water, limiting its use in the food sector (Mathanker et al. 2013). The scattering effect, limited sensitivity and penetration, low limit of detection, and lack of a well-established THz database library are among the other limitations of THz spectroscopy (Afsah-Hejri et al. 2019).

Research gap

As discussed in the sections above, various approaches have been used for the provenance analysis of soybean. However, there are additional potential methods that could be applied, albeit they haven’t yet been investigated to determine the geographical origin of soybean. To our knowledge, techniques such as Raman spectroscopy, hyperspectral imaging spectroscopy, and DNA-based techniques haven’t been reported for soybean provenance. These techniques have been combined with multivariate analysis in several studies for other food product traceability and fingerprinting. For example, Zhu et al. (2018) utilized Raman spectroscopy with multivariate data analysis to determine the geographical origin of rice with a recognition rate of approximately 97%. Similarly, provenance studies of honey (Corvucci et al. 2015) and wine (Mandrile et al. 2016) have demonstrated the effectiveness of Raman spectroscopy in the traceability field.

As the name suggests, hyperspectral imaging spectroscopy integrates imaging and spectroscopic techniques into one system (Li et al. 2017). This technique is most popularly used to discriminate the crop varieties, including soybean (Tan et al. 2014; Zhu et al. 2020), fruits (Liu et al. 2014; Zhao et al. 2018), edible oils (Xie, Wang, and He 2014), meat and beverages (Feng et al. 2021). The use of hyperspectral imaging in traceability is still novel, so it hasn’t been applied widely. However, numerous provenance investigations have demonstrated that it has a promising scope. For instance, Hai-Yan, Zhan-Qi, and Zhen-Hong (2019) used machine learning algorithms to differentiate millets from three regions of China with an average prediction rate of 90%; Kim, Lim, et al. (2020) employed PLS-DA and achieved a high discrimination accuracy of 99% in discriminating rice from South Korea and China; Choi, Kim, and Moon (2021) applied the same approach and achieved a discrimination prediction accuracy of 91% for chia seeds from Argentina, Paraguay, and Bolivia. Similarly, it has been used for traceability analysis of Chinese wolfberries (Yin et al. 2017), peaches (Sun et al. 2021), tea (Hong and He 2020), and saffron (Lu et al. 2020).

Numerous DNA-based techniques have been used to identify countries of origin, including microsatellites, species-specific PCR amplification, single nucleotide polymorphisms (SNPs), DNA sequencing, and cleaved amplified polymorphisms (Wilkes, Nixon, and Burns 2016). DNA fingerprinting has been employed to trace the geographical origin of several products, including beef (Vázquez et al. 2004), timber (Hung, Lin, and Ju 2017), honey (Khansaritoreh et al. 2020), olive oil (Woolfe and Primrose 2004), and seafood (Takashima et al. 2014). Hung, Lin, and Ju (2017) determined and genotyped 16 microsatellites to discriminate between the legal and illegal logged timber from Taiwan. The provenance of soybeans may benefit substantially from such a strategy.

Conclusion

Consumer interest in high-quality food and its supply chain transparency has increased rapidly over the past decade, favouring the development of several analytical techniques for food traceability purposes, such as geochemical, chromatographic, and spectroscopic techniques. New advancements in machine learning methods have also offered us an innovative approach to looking at the data differently and better comprehending it. Combining chemometrics with analytical techniques allows for analyzing specific product attributes influenced by geographic origin. In this review, we have discussed ten techniques, broadly categorized into three groups that have been utilized to determine the geographical origin of soybean. However, each of these methods has advantages and disadvantages of its own. For example, the major drawback of geochemical techniques is their instrumental and operational costs, which can be mitigated by spectroscopic methods at the expense of sensitivity. Lastly, we have discovered research gaps in analytical techniques, including DNA techniques, Raman, and hyperspectral spectroscopies that have the potential to determine the provenance of soybeans.

Acknowledgements

We would like to acknowledge the University of Otago for the Doctoral Scholarship.

Disclosure statement

The authors declare that they have no known conflict of interests.

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.

Notes

1 All units are in million metric tons (MMT).

2 Least significant difference

3 General linear model

4 Exact origins weren’t specified.