The Ability of Near-Infrared Spectroscopy to Discriminate Plant Protein Mixtures: A Preliminary Study

Abstract

The aim of this paper was to evaluate the effect of two different matrices (e.g., starch base flour vs. protein base flour) on the ability of near-infrared (NIR) spectroscopy to classify binary mixtures of chickpea (protein), corn and tapioca (starch) flours. Binary mixtures were made by mixing different proportions of chickpea plus corn, chickpea plus tapioca, and corn plus tapioca flour. Spectra were collected using NIR spectroscopy and the data analyzed using techniques such as principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA). The results showed an effect of the matrix on the PLS-DA classification results, in both classification rates and PLS loadings. The different combinations of flours/mixtures showed changes in absorbance values around 4752 cm⁻¹ that are associated with starch and protein. Nevertheless, the use of NIR spectroscopic might provide a valuable initial screening and identification of the potential contamination of flours along the supply and value chains, enabling more costly methods to be used more productively on suspect samples.

1. Introduction

The realm of plant-based and animal-free protein foods is a highly dynamic and a rapidly expanding domain within the contemporary and sustainable food industry [1]. These protein sources are regarded as compelling alternatives to conventional protein sources (e.g., animal proteins) [1,2]. Nonetheless, despite their surging popularity, the development and production of such products encounter substantial challenges, which may elicit concerns among consumers [2,3]. Similar to other food ingredients, plant-based proteins are subject to inherent variability stemming from natural and induced factors [1,2,3]. Consequently, it has become imperative for both the food manufacturing industry and consumers to conduct thorough testing to ensure the safety, chemical properties, and nutritional composition of these ingredients [1,2,3]. By doing so, potential risks associated with ingredient variability can be mitigated, ensuring consistent quality and maintaining consumer confidence [3].

As the plant-based protein industry continues to evolve, a pressing need arises for dependable and efficient methods to assess the safety and quality of these alternative protein sources [1,2,3]. The development and implementation of robust testing protocols are indispensable in guaranteeing the production of high-quality plant-based protein foods that meet regulatory standards and consumer expectations. Such methods should encompass comprehensive evaluations encompassing safety considerations, chemical characteristics, and nutritional profiles [1,2,3]. To meet these demands, collaborative efforts among researchers, industry professionals, and regulatory bodies are essential. Through the establishment of standardized testing procedures and the advancement of analytical techniques, the food manufacturing industry can bolster consumer trust in plant-based and animal-free protein products [1,2,3]. Consequently, these advancements can further propel the growth and sustainability of the plant-based protein sector while addressing the prevailing challenges in product development and manufacturing [2,3].

The composition and nutritional quality of raw ingredients utilized in the production of plant-based proteins exhibit considerable variability, which can be attributed to a multitude of factors [2]. These factors include the genetic diversity of plant varieties, environmental conditions such as temperature, rainfall, and soil fertility, as well as potential contamination risks during transportation and storage, such as mold formation [1,2,3]. Additionally, concerns related to fraud and artificial adulteration further contribute to the variability observed [2,4,5]. The substantial variability in composition and nutritional quality of these raw ingredients carries significant practical and economic implications for the food manufacturing industry. Ensuring consistency in both the composition and quality of the final product is imperative to meet the expectations and requirements of consumers [2,4,6]. Achieving such consistency poses a challenge for manufacturers, as they must navigate the inherent variability in the raw materials and implement measures to mitigate its impact. This necessitates the development and implementation of robust quality control systems and testing protocols to monitor and regulate the input materials throughout the manufacturing process.

By adopting these measures, the food manufacturing industry can strive to deliver plant-based protein products that meet predetermined standards and specifications consistently. Moreover, such efforts enhance consumer confidence by providing them with products that are reliable and of consistent quality [4,5,6]. The economic viability of the plant-based protein sector also depends on the ability to manage ingredient variability effectively, ensuring efficient and cost-effective production processes while meeting consumer demands [2,4,6]. Thus, understanding and addressing the sources of variability in raw ingredients is crucial for the sustainable growth and success of the plant-based protein industry.

A diverse array of methods can be employed to characterize and analyze the safety and chemical composition of plant-based protein ingredients and products throughout the supply and value chain [2,4,6]. These methods encompass a spectrum of approaches, ranging from traditional techniques such as proximate analysis (e.g., quantifying protein, fat, and carbohydrate content) to more advanced and sophisticated methodologies, including high performance liquid chromatography (HPLC), liquid chromatography mass spectrometry (LC-MS), and vibrational spectroscopy [7]. Vibrational spectroscopy techniques, such as mid-infrared (MIR) and near-infrared (NIR) spectroscopy, have long been utilized for the routine analysis of food ingredients and products at various stages of the supply and value chain [2,4,6]. These techniques not only enable the analysis of the proximate composition, but also serve as valuable tools for evaluating functional properties of food and monitoring issues associated with food safety and security, including fraud detection, traceability, authenticity, and adulteration [2,4,6,8].

By employing vibrational spectroscopy, researchers and industry professionals can gain insights into the chemical composition and quality of plant-based protein ingredients and products in a rapid and non-destructive manner. Infrared techniques offer advantages such as speed, cost-effectiveness, and minimal sample preparation, making them suitable for large-scale analysis and quality control purposes throughout the supply chain. Both MIR and NIR spectroscopy have become increasingly important as analytical methods in the analysis of plant-based protein ingredients and products. These techniques offer valuable insights into the composition and functional properties of these ingredients, which in turn enables various applications in the food industry. One of the primary benefits of using vibrational spectroscopy techniques is their ability to provide a comprehensive analysis of the proximate composition of plant-based proteins [2]. MIR and NIR spectroscopy can identify and quantify important components such as proteins, carbohydrates, lipids, and other constituents present in the sample in a rapid and non-destructive manner [4]. This information is crucial for determining the nutritional profile of plant-based protein ingredients and products and for comparing them to desired standards and regulatory requirements [2,6].

In addition to proximate composition analysis, vibrational spectroscopy techniques also offer insights into the functional properties of the sample [8]. These properties include solubility, emulsifying capacity, foaming ability, gelling properties, and overall texture [8]. By characterizing these functional properties, manufacturers can optimize formulation processes and improve the overall quality of plant-based protein products [6,8]. This is especially important as the demand for plant-based alternatives to animal-based proteins continues to grow, and consumers expect these products to mimic the sensory and functional attributes of traditional animal-based products. Another significant advantage of vibrational spectroscopy techniques is their ability to address food safety concerns [9]. These techniques can detect and quantify contaminants, such as allergens, mycotoxins, pesticides, and heavy metals, in plant-based protein ingredients and products [4,6,9]. Ensuring the safety of plant-based protein foods is crucial to preventing adverse health effects and maintaining consumer confidence [4,6,7]. Therefore, MIR and NIR spectroscopy provides a reliable and efficient means of monitoring and controlling potential contaminants throughout the supply chain, contributing to maintaining the integrity of the plant-based protein supply chain [2,4,6,7,8,9]. For example, by establishing spectral fingerprints or chemical signatures, these techniques can be used for authentication and identification purposes. They can provide information used to verify the origin and authenticity of plant-based protein ingredients, preventing adulteration, and ensuring that products meet the claimed standards and labeling requirements [2,8]. This is particularly important as the plant-based protein industry continues to expand, and there is a need to maintain transparency and trust in the market [2,4,6,7,8,9].

Numerous investigations have provided evidence of the efficacy of NIR spectroscopy in assessing and tracking the presence of adulterants and contaminants in various binary mixtures within diverse food matrices. These studies encompass a range of applications, such as the analysis of the adulteration of infant formula [9], the identification of fiber addition to semolina samples [10], the utilization of NIR combined with hyperspectral imaging to detect contamination in soybean meal [11] and metanil adulteration in chickpea meal [12], the detection of soy additives in rice beverages [13], and the determination of adulteration in purple sweet potato with white sweet potato [14]. Although NIR spectroscopy has demonstrated its potential in assessing adulteration and contamination in different binary mixtures across various food matrices, there remains a dearth of comprehensive investigations examining the influence of the flour type on NIR spectra and classification outcomes.

This study aimed to assess the efficacy of near-infrared (NIR) spectroscopy in discriminating binary mixtures of plant proteins and starchy flours such as chickpea, corn, and tapioca flours to detect instances of food adulteration and contamination.

2. Materials and Methods

Three batches of commercial chickpea (CP), corn flour (CF), and tapioca flour (TP) samples were purchased from a local supermarket in Brisbane (Queensland, Australia). Each of the flour types were randomly split into three replicates, where binary mixtures were made using the following ratios; 100:0% (pure flour with no addition of other flour), 95:5% w/w, 90:10% w/w, 85:15% w/w, 80:20% w/w, 75:25% w/w, 70:30% w/w, 65:35% w/w, 60:40% w/w, and 50:50% w/w. For example, the 100% CP and CF mixtures were made using 10 g w/w CP and 0 g w/w CF, while the 95% mixture was made using 9.5 g w/w CP and 0.5 g w/w CF, and so forth. In addition, three replicates for each combination were also created. In this study, only the high range of adulteration was considered, as it will be the one that has the higher economic effect. Thus, the total number of samples obtained for each of the binary mixtures was 30 (10 mixtures × 3 biological replicates = 30), and the overall number of 90 samples.

The FT-NIR spectra of both pure flour and mixture samples were collected randomly using a Bruker Tango-R spectrophotometer (Bruker Optics GmbH, Ettlingen, Germany) with a gold-coated integrating sphere (diffuse reflection). The samples were placed in a borosilicate-glass cuvette with a 10 mm diameter (Bruker Optics GmbH, Ettlingen, Germany), and the reflectance spectra of the sample were recorded using OPUS software (version 8.5, Bruker Optics GmbH, Ettlingen, Germany) (64 interferograms at a resolution of 4 cm⁻¹, wavenumber range of 11,550 to 3950 cm⁻¹). To avoid cross-contamination, the cuvettes were cleaned with a 70% w/w ethanol/water solution and dried with paper wipes between samples.

The NIR spectra were exported from the instrument using the OPUS software into the Vektor Direktor™ (Version 1.1; KAX Group, Sydney, NSW, Australia) for chemometric analysis. Before data analysis and classification, the NIR spectra were pre-processed using the Savitzky–Golay second derivative (second polynomial order and 21 smoothing points) [15]. Principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) were used to classify the flour mixture samples according to the level of adulteration. The PCA and PLS-DA models were validated using cross validation [16,17,18]. The PLS-DA models were evaluated using the coefficient of determination in cross validation (R²_cv) and the standard error in cross validation (SECV). Samples from replicates 1 and 3 were used to develop the cross validation models, while samples from replicate 2 were used as a validation set. Thus, the models were evaluated using the R²_val coefficient of determination in prediction or validation, and the standard error in prediction (SEP). In addition, the number of samples correctly classified (%CC) was also calculated.

3. Results and Discussion

The second derivative NIR spectra of TP vs. CP (Panel A) and CF vs. CP (Panel B) flour mixtures are displayed in Figure 1. Changes in the NIR spectra of the mixtures can be observed around 5808 cm⁻¹, 5632 cm⁻¹, and 4752 cm⁻¹. The absorption band around 5808 cm⁻¹ is associated with C-H₂, corresponding with the high concentration of CP flour in the mixtures (identified with a circle). This band also decreases as the concentration of either CF or TP (starch) flour that is added to the mixture increases. Changes in the absorbance around 5632 cm⁻¹ can be associated with variations in the C-H₂ as well as with aromatic groups. An increase in the absorbance values at this wavenumber was observed as consequence of an increase in the addition of CF and TP flour to the mixture samples [11,12,19]. The absorption band at 4752 cm⁻¹ is associated with CHO, C=O and O-H, and C-H corresponding with plant polymers such as starch [10,11,12,19,20]. The absorbance values at this wavenumber appear to remain constant along the three flour matrices and mixtures analyzed. Absorbance values corresponding with this band also increase as the amount of CF and TP flour added to the mixtures increases. Other authors have also reported that the deformation of O–H and stretching of C–O related to starch can be also observed in the range between 4800 to 4200 cm⁻¹; where stretching and deformation of C–H can be observed around 4316 cm⁻¹ and 4015 cm⁻¹ associated with the stretching of C–H and C–C related to starch [10,11,12,19,20]. These changes can also explain the observed differences in the NIR spectra of the samples and the classification results obtained. Figure 2 plots the absorbances (second derivative) at 4752 cm⁻¹ at the different levels from each of the three mixture samples, analyzed in triplicate. It was observed that the absorbance values from the two mixtures containing CP flour decreased until the 80:20% w/w mixture, where after this point the absorbance increases. On the other hand, the absorbance values at the same wavenumbers for the CF vs. TP flour mixture samples remain almost constant independently of the level of the mixture analyzed. This can be explained by the occurrence of starch as the main carbohydrate present in the set of flour samples analyzed. These results also agreed with those reported by other studies, where different mixtures or combinations of different flours were analyzed (e.g., soy and semolina) using NIR spectroscopy [10]. Similar results were also reported by other authors when hyperspectral imaging was combined with wavelengths in the NIR region to detect contamination in chickpea and soybean meal samples [11,12].

The PCA score plot of all binary mixtures analyzed using NIR spectroscopy (second derivative) is shown in Figure 3 (Panel A scores and Panel B loadings). The first and second principal components (PC) accounted for 90% and 8% of the variance in the NIR spectra of the flour mixture samples analyzed. It can be observed that the three pure flours are clearly separated from the mixtures, with the arrow indicating the increase in the amount of CP flour added into the mixtures. The highest loadings derived from the PCA analysis were observed around 5808 cm⁻¹, 5632 cm⁻¹, and 4752 cm⁻¹. These wavenumbers are associated to the same chemical components described above. Figure 4 shows the PCA score plot (A and C) and loadings (B and D) for the individual binary mixtures comparing CP and CF (A and B), and CF and TP (C and D), respectively. In all cases, samples tend to be separated according to the inherent differences between the pure flour samples along the PC1 (>80%), while PC2 (>10%) explains the separation between the mixtures. Overall, the PCA score plots are very similar between the different flour mixture analyses. However, the main differences were observed in the PCA loadings. The highest loadings derived from the PCA analysis of the CP and CF mixtures were observed around 5808 cm⁻¹ (C-H), 5632 cm⁻¹ (C-H and C-H₂), and 4752 cm⁻¹ (C=O and C-H₂) [10,19,20]. These wavenumbers are mainly associated with the protein and starch present in the mixtures. In the case of the PCA analysis of the flour mixtures between TP and CF, the highest loading was observed at around 5072 cm⁻¹ (N-H bonds) in both PC1 and PC2, respectively [10,19,20]. It is important to note that this wavenumber is mainly related to the starch content in the samples [10,19,20].

The PLS-DA regression statistics used to predict the level of addition or adulteration in the binary mixtures are summarized in

Table 1. Cross validation and validation statistics for the prediction of the level of adulteration in the binary mixture samples analyzed using near-infrared reflectance spectroscopy.

	CP vs. TP	TP vs. CF
Calibration
R2cv	0.85	0.73
SECV (%)	1.50	1.53
Bias	0.29	−0.10
slope	0.75	0.63
LV	3	3
Validation
SEP	1.65	1.25
R2val	0.80	0.79

R²_cv: coefficient of determination in cross validation; R²_val: coefficient of determination in validation; SECV: standard error in cross validation; SEP; standard error in prediction; LV: latent variables; CP: chickpea flour; CF: corn flour; TP: tapioca flour.

, along with their cross-validation statistics results. The results showed that the PLS-DA regression cross-validation models explained 80% (R²_cv: 0.80) and 79% (R²_cv: 0.79) of the variability in the binary mixtures between CP with the starchy flour and the CF and TP mixture, respectively. The standard error in prediction (SEP) was 1.65% and 1.25 for the CP+TP and TP+CF mixtures, respectively. Overall, 100% of the samples were correctly classified. Similar to what was observed for the PCA loadings, differences in the PLS loadings used to develop the models were also observed. It is important to note that the models used the same number of latent variables (LV), with three in each case. The highest PLS loadings used for the discrimination between CP samples and the starchy flour mixtures were observed around 5264 cm⁻¹ (O-H), 4832 cm⁻¹ (N-H deformations), and 4320 cm⁻¹ (C-H) associated with water or moisture content, which is the presence of amide groups and protein content [10,19,20]. Meanwhile, the highest PLS loadings for the adulteration between CF and TP were observed around 5072 cm⁻¹ (N-H bonds) and 4512 cm⁻¹ (C=O and C-H) associated with protein and starch content in the samples analyzed, as reported by other authors [10,19,20,21,22,23]. Moreover, we propose implementing stratified sampling for the test set. This will ensure that both the calibration and test sets accurately represent all adulteration ratios and flour types. One viable approach is to categorize the samples into strata based on the adulteration ratio. Within each stratum, a random selection of samples can then be chosen for your calibration and test sets.

NIR spectroscopy is widely recognized as a valuable tool in the food industry due to its ability to provide an initial level of screening. This advantage can be utilized in the food chain and enable more costly methods to be used more productively on suspect specimens and can be easily implemented by the food manufacturing industry [24,25,26,27,28]. While it is important to measure and analyze the chemical composition of raw materials to optimize the processing of plant proteins, qualitative screening by means of NIR spectroscopy should also be incorporated as a routine analysis tool for food ingredients. Incorporating NIR spectroscopy in this manner can help ensure the quality and safety of food products while also increasing efficiency and productivity in the food industry.

4. Conclusions

The results of this study showed an effect of the matrix on the PLS-DA classification results, in both the classification results and the PLS loadings used by the models. The different combinations of flours/mixtures showed differences in the absorbance values around 4752 cm⁻¹ associated with changes in starch and protein. Nevertheless, NIR spectroscopy could be utilized as an initial screening tool for detecting potential flour contamination along the supply and value chain, allowing more expensive methods to be used more efficiently on suspect samples. In addition, the use of NIR spectroscopy for the initial screening of potential contamination in food ingredients can significantly improve the efficiency and effectiveness of quality control measures in the food industry. This can ultimately reduce the risk of food fraud and ensure the safety and authenticity of food products. However, this is a preliminary study (proof-of-concept) due to the number of samples and mixtures analyzed and requires further developments before the proposed method can be fully implemented by the food industry. This also includes the optimization of the analytical method to be used as a reference, the establishment of robust and reliable calibration models by including a wide range of flour and mixture samples, and the validation of the method using an independent set of samples. Continued research in this field will be valuable in advancing the application of NIR spectroscopy for food analysis and quality control.