Application of infrared spectroscopy for the prediction of nutritional content and quality assessment of faba bean (Vicia faba L.)

With growing consumer interest and demand for health‐benefiting functional foods such as faba beans, particularly evident in developed countries, commercial production of this crop is increasing. In concert with increased production levels comes an equally great need for the inexpensive rapid measurement of nutritional parameters for quality determining purposes. As an analytical tool, near‐infrared spectroscopy has been well explored for the quantification of proximate nutritional composition, such as protein, starch and oil contents in faba bean and faba bean‐derived products. Near‐infrared spectroscopy has also been demonstrated to have potential for the noninvasive prediction of low‐level micronutrients such as the total polyphenol content in faba bean and faba bean‐derived products, although further exploration in this area is required to provide a more acceptable model. In some instances, the authors may be inadvertently measuring micronutrient concentrations through a secondary correlation with certain macronutrients. It is particularly difficult to determine if this is the case if exacerbated by the lack of an independent validation test set in the paper in question. The associated technique of mid‐infrared spectroscopy shows particular promise for the rapid, noninvasive characterisation of structural components of faba bean, such as carbohydrates and proteins. Complementary applications of these two technologies are likely to yield a wealth of potential applications.

has been well explored for the quantification of proximate nutritional composition, such as protein, starch and oil contents in faba bean and faba bean-derived products.
Near-infrared spectroscopy has also been demonstrated to have potential for the noninvasive prediction of low-level micronutrients such as the total polyphenol content in faba bean and faba bean-derived products, although further exploration in this area is required to provide a more acceptable model. In some instances, the authors may be inadvertently measuring micronutrient concentrations through a secondary correlation with certain macronutrients. It is particularly difficult to determine if this is the case if exacerbated by the lack of an independent validation test set in the paper in question. The associated technique of mid-infrared spectroscopy shows particular promise for the rapid, noninvasive characterisation of structural components of faba bean, such as carbohydrates and proteins. Complementary applications of these two technologies are likely to yield a wealth of potential applications.

K E Y W O R D S
Fourier transformed infrared (FTIR) spectroscopy, attenuated total reflectance mid-infrared (ATR-MIR) spectroscopy, near-infrared spectroscopy (NIRS)

| INTRODUCTION
The modern consumer is more connected and informed than ever before, and one of the results of this is an increased consumer demand for functional foods. Functional foods are food sources that have the potential to provide health-benefiting effects in addition to those expected from traditional staple crops such as wheat. Faba bean (Vicia faba L.), also known as broad bean or fava bean, is one of the oldest cultivated crops (Singh & Bharati, 2013) and is an example of such a species that has benefited from this shift in consumer attitudes towards accepting functional foods. In particular, the high levels of antioxidant and phenolic compounds present in faba bean seeds have been linked to their health benefiting effects, which include protection against radical species, antihypertensive and anticancer activity (Siah, Konczak, Agboola, Wood, & Blanchard, 2012;Turco, Ferretti, & Bacchetti, 2016). The seed material from faba bean may be consumed as is or utilized as an adjunct additive for the creation of novel valueadded foodstuffs (López-Barrios, Gutiérrez-Uribe, & Serna-Saldívar,-2014;Vioque, Alaiz, & Girón-Calle, 2012). Traditionally consumed in countries in the Middle East and South East Asia (AEGIC, 2017), commercial production of this leguminous crop is steadily increasing in developed countries such as Australia, particularly over the last few decades (Siddique et al., 2000). Australia is now regarded as one of the top five producers and the largest exporter of faba bean in the world (Australian Export Grains Innovation Centre [AEGIC], 2017).
With both increased commercial production and greater consumer awareness and acceptance, there is an increasing need to assess and predict the overall nutritional quality of this crop in terms of protein and carbohydrate composition, as well as other bioactive/functional compounds of commercial crops for quality assurance purposes. However, most analytical procedures are either too expensive, require extensive sample preparation, or take long periods to perform. Hence, the rapid assessment and prediction of the overall nutritional quality of faba bean could be extensively applied amongst producers, buyers and manufacturers alike. Development of such nondestructive analytical tools would reduce the overall analytical costs and time, allowing for more representative sampling and subsequent analysis of the crops.

| INFRARED SPECTROSCOPY
Infrared spectroscopy is a noninvasive, nondestructive and rapid spectrophotometric analytical tool that is gradually being applied across many disciplines and to a range of matrices worldwide (Cozzolino, 2014;Johnson, 2020;Roggo et al., 2007). Infrared spectroscopy works on the principles of absorption due to molecular vibrations, using electromagnetic wavelengths with a lower frequency than light. The infrared spectrometer emits the full spectrum of infrared wavelengths, which penetrate the sample, with certain wavelengths absorbed by specific chemical bonds present within the sample. The amount of light energy absorption is directly proportional to the concentration or quantity of bonds present in the sample. From the reflected or transmitted wavelengths, the identity and quantity of the compounds present in the sample may be deduced.
There are two principal types of infrared spectroscopy used in food analysis, namely, near-infrared spectroscopy (NIRS) and midinfrared (MIR) spectroscopy (MIRS) (Figure 1). MIRS is generally defined as the range of wavenumbers between 4,000 and 400 cm −1 (or wavelengths of 2,500 to 25,000 nm), while NIRS incorporates the spectrum from wavelengths of 750 to 2,500 nm (Pasquini, 2003).
Although the wavelengths below 1,000 nm mainly result from overlapped, relatively weak third overtones of chemical bonds (Dowell, Throne, & Baker, 1998), the longer pathlengths and opportunity to work with intact samples have made this technology invaluable for food researchers. For example, shortwave NIRS may penetrate centimetres into a sample. Transmission spectroscopy is often used with shortwave NIRS for the analysis of whole grains, as allowed for a more representative spectra to be collected from the sample. On the other hand, longwave NIRS penetrates only millimetres into the sample and hence is often used when analysing surface layers or homogenous materials. Reflectance or interactance spectroscopy is used in the latter instances, as the spectral information is collected from only the sample surface.
The majority of modern MIRS systems use a Fourier transform in order to simultaneously measure all wavelengths across the MIR spectrum, thus the term FTIR (Fourier transform infrared) spectroscopy will be used throughout this review to refer to mid-infrared spectroscopy. To provide greater signal amplitude and clarity, attenuated total reflectance (ATR) sampling is often used with FTIR spectroscopy. This necessitates direct contact between the sample and the ATR platform.
As the signal amplitude depends to a large degree on the pressure applied, it can be quite difficult to obtain quantitative results from ATR-FTIR. The particle/grain size can have an impact on the spectra obtained, particularly for ATR-FTIR spectroscopy (Lee, Liong, & F I G U R E 1 Three commercial infrared spectrophotometers. (a) Nicolet Antaris FT-NIR (b) Bruker Alpha II FTIR with ATR platform (c) FOSS NIRSystems Model 6500 Jemain, 2017;Udvardi et al., 2017). However, grain size also impacts upon the spectra obtained from NIRS (Rinnan, Berg, & Engelsen, 2009;Wiley, Tanner, Chandler, & Anderssen, 2009). Data preprocessing methods performed on the spectra, such as the use of derivatives or multiplicative scatter correction, are often used to overcome this (Cozzolino, 2014;Lee et al., 2017). Nevertheless, there do not appear to be any systematic, controlled studies investigating the specific effects of particle size on the resultant spectra in cereal/pulse matrices.
Historically, NIRS has been the dominant form of infrared spectroscopy used for food analysis, largely due to the low instrumentation cost, high signal to noise ratio of the detector and the greater penetration of the infrared wavelengths into the sample matrix, due to the longer wavelengths used. Contemporary applications are increasing in their use of portable and in-line instrumentation. On the other hand, the MIR spectrum contains a larger array of more specific and characteristic absorption peaks for a range of functional group chemical bonds present (Cozzolino, 2014;Johnson, Collins, Skylas, & Naiker, 2019), hence has the potential to provide a more detailed fingerprint of the sample being analysed ( Figure 2). The application of FTIR for the analysis of food crops has been steadily increasing, particularly over the past decade. For instance, over the past year, studies have reported using FTIR in the analysis of grain crops such as wheat , mungbean (Johnson, Collins, Power, Chandra, Portman, Blanchard et al., 2020) and faba bean (Johnson, Collins, Skylas, Quail, Blanchard, & Naiker, 2020). This review focuses on faba bean and the historical and emerging application of both forms of infrared spectroscopy for the determination of its overall nutritional quality. This area was reviewed very briefly by Rodriguez Espinosa, Guevara-Oquendo, Sun, ; however, these authors focused only on selected applications of FTIR and did not include NIRS in their review.

| PROXIMATE NUTRITIONAL COMPOSITION
Proximate nutritional composition refers to the broad classes of macronutrients that make up the majority of foodstuffs and is usually determined through relatively basic analytical procedures. Aspects of the proximate composition include moisture, crude protein content, ash, crude fat and crude fibre content. NIRS has been widely used for the determination of numerous aspects of proximate nutritional composition in most crops, including faba bean (Font, del Río-Celestino, & de Haro-Bailón, 2006;Williams, Stevenson, Starkey, & Hawtin, 1978).

| Protein
As has been the case with other grain crops such as wheat, the quantification of protein content was one of the first applications of NIRS in faba bean. By the late 1970s, Williams et al. (1978) had reported the development of a reflectance NIRS model (utilising the ratio of absorption at 2,180 nm to that at 2,100 nm) that was used to predict the level of protein in ground faba bean samples ( Table 1). The calibrated protein prediction model had an R 2 of 0.96, indicating that nearly all points fell on a straight line. The root mean square error (RMSE), a measure of the differences between the predicted and true values, was 0.56. The coefficient of variability was around 1.5%, comparing favourably to 1.2% for the Kjeldahl method. Analysis speed F I G U R E 2 Locations of selected functional groups in FTIR spectra collected from milled faba bean flour. Note the spectral variation between the three commercial Australian faba bean varieties illustrated here was reported as >200 samples per instrument per day, although with use of in-line apparatuses this could easily exceed thousands of samples per day. Nowadays, commercial instruments including in-built calibration models are available for the determination of protein, fibre, fat, ash and moisture (BUCHI, 2019).
El-Sherbeeny and Robertson (1992) used NIRS (wavelength range not stated) to measure the protein content of 840 faba bean lines.
The calibration sample size was quite small, at only 50 samples, with validation performed using the Kjeldahl method for total protein determination for every tenth sample. In addition, the authors of this work did not report whether whole or powdered faba bean samples were used. Protein concentrations for the samples included in the calibration curve ranged from 18.0% to 31.1%. The results obtained from these researchers showed NIRS to be quite accurate, with the standard deviation between the values obtained from NIR and Kjeldahl methods at 0.28%, while the coefficient of variability was 1.13%.
Lepse, Dane, Zeipiņa, Domínguez-Perles and Rosa (2017) also determined protein contents via NIRS in their attempt to determine optimise crop combinations of faba bean and a range of vegetables (e.g., onion, cabbage and carrot) for the greatest protein yield per hectare, but did not report any measures of the error of prediction. (2014) reported a NIRS model for the determination of protein in ground faba bean seed powder with an R 2 of 0.94 and root mean square error of cross-validation (RMSECV) of 0.34% (leave-one-out cross-validation). They also demonstrated that protein content could also be predicted from the intact seed, but with lower accuracy (R 2 = 0.76, RMSECV = 0.60%). This can be attributed to the greater heterogeneity of the intact seed, as only the outer layer of the seed volume is sampled, which in turn reduces the reproducibility of the infrared spectra and decreases the prediction accuracy of the model created.

Wang, Liu and Ren
Another aspect that deserves consideration in terms of protein content is the structural and conformational composition, which may influence the availability and digestibility of the protein when consumed, and hence the bioavailability of potential nutritional health benefits of the faba bean crop. Rodriguez Espinosa (2018) applied FTIR spectroscopy to quantify the amounts of various protein structures in ground faba bean seeds (Table 2). Significant differences in most measures of protein molecular structure were found between low tannin and normal tannin containing faba bean genotypes, particularly in the ratios of Amide I: Amide II bonds and the amount of β-structures, highlighting the potential interaction between tannins and nutritional quality. The Amide I bond results mainly from C O stretch and its level of absorption is modulated by the protein secondary structure (Byler & Susi, 1986). Amide II is produced by C 3.2 | Moisture Williams et al. (1978) reported an R 2 of 0.93 and RMSE of 0.30% for the reflectance NIRS model they developed (ratio of wavelengths at  using a slightly different wavelength ratio to quantify crystalline and amorphous starch (1,048 to 1,016 cm −1 ). Variation between faba bean genotypes was found in both the initial orderliness of the starch present (with ratios ranging from 0.772 to 0.889) and the level of decreased orderliness in response to heat and moisture treatment (11.3% to 13.2% decrease).

| Starch and other carbohydrates
Other researchers have also used ATR-FTIR to assess changes in starch granular architecture following various hydrothermal treatments, in addition to their molecular interactions with the proteins present (Chávez-Murillo, Veyna-Torres, Cavazos-Tamez, de la Rosa-

| QUANTIFICATION OF TANNINS AND POLYPHENOLS
One of the major groups of phytochemicals present in faba bean is the polyphenols. While most polyphenols are considered beneficial due to their positive cardiovascular effects, certain polyphenols such as condensed tannins are considered antinutritive as they can decrease the efficiency of nutrient uptake and metabolism (Chung, Wong, Wei, Huang, & Lin, 1998).

| Total polyphenols
Polyphenols are phytochemical micronutrients, most of which are well known for their health-benefiting effects (Shahidi & Ambigaipalan, 2015). As a traditional measure of polyphenol content, the total polyphenols in a sample are usually quantified using the Folin-Ciocalteu assay, which can be quite time consuming . Wang et al. (2014) used NIRS to predict the total polyphenol content in ground faba bean, with an R 2 of 0.79, RMSECV of 0.40 and RPD of 2.20 for leave-one-out cross-validation. In contrast to proximate components such as moisture, protein and carbohydrates, no commercial instruments appear to be marketed for the purpose of polyphenol determination.

| Tannins
Tannins are a group of complex phenolic polymers derived from flavonol. As an important antinutritive component of faba bean seeds, the condensed tannin concentration, which can range from negligible to up to 7% w/w, is thus one factor that must be taken into consideration when developing new faba bean varieties (Helsper, Hoogendijk, van Norel, & Burger-Meyer, 1993;Marquardt, Ward, & Evans, 1978).
As such, development of a reliable, rapid method for quantification of the tannin contents in faba bean would be greatly beneficial to the industry, particularly for the plant breeders. In order to assist in this area, De Haro, López-Medina, Cabrera and Martín (1988) demonstrated that NIRS can noninvasively be used to measure the tannin content in whole faba bean seeds. Tannins are largely located in the seed coat, ensuring that NIRS is measuring nearly all of the tannins present. Across the range of 0.01-7% w/fw tannin in the faba bean seeds measured, the best cross-validation results obtained had an R 2 value of 0.93 and standard error of prediction of 0.54%. The authors used a calibration set of 60 samples, with a similar-size validation set, but did not report the cross-validation method used. It should be noted that in this study, no calibration samples were included with tannin concentrations in the range of 1%-3.5%.

| Vicine and convicine
Two other important antinutritional factors present in faba bean are vicine and convicine, both of which are alkaloid glycosides (Burbano, Cuadrado, Muzquiz, & Cubero, 1995). If consumed in high levels by individuals suffering from a genetic mutation in the red blood cell enzyme glucose-6-phosphate (estimated to occur in around 5% of the world population), this can lead to a form of haemolytic anaemia known as favism (Burbano et al., 1995). Vicine and convicine are currently quantified using a lengthy HPLC procedure , so a rapid analytical method would save considerable time. The major challenge results from the typically low concentrations of these compounds (around 0.6%-0.9% by weight). Nevertheless, Puspitasari (2017) did report the prediction of these compounds in faba bean flour using NIRS. The best performing prediction model gave an R 2 of 0.968 and standard error of cross-validation of 0.094%.
Subsets of the calibration samples were used for validation purposes.
The RPD, which should ideally be at least 3, ranged from 2.67-3.14 for the five validation subsets, leading the authors to suggest that NIRS would be suitable for screening purposes.

| DISCRIMINATION BETWEEN VARIETIES AND GROWING LOCATIONS
Determining the variety or growing location of a faba bean sample can be an important aspect of quality assurance or authentication. In contrast to the previously mentioned studies, which report either quantification or identification of a compound using infrared spectroscopy, determination of variety/growing location is a qualitative Also working on Chinese faba bean, Xu et al. (2015) reported the discrimination of white and green varieties using a combination of their polysaccharide and protein absorption bands, measured using FTIR, and their mineral contents, measured using ICP-MS (inductively coupled plasma mass spectrometry).

| Leaves and stems
Whole faba bean plants were included amongst other plant species investigated by Bruun et al. (2005)   Isolated studies on faba bean leaves have been conducted through the light measurement tool spectroradiometry, typically using visible and shortwave NIR wavelengths (Malthus & Madeira, 1993).
This technique has been used to detect the infestations of Botrytis fabae (Malthus & Madeira, 1993) and determine the radiation usage by the crop (the fraction of absorbed photosynthetically active radiation, commonly abbreviated fAPAR) (Ridao, Conde, & Mínguez, M. I., 1998;Ridao, Oliveira, Conde, & Minguez, 1996). This latter application could be used to remotely assess water stress and crop biomass, as shortwave NIR wavelengths can be obtained as satellite imagery.

| Roots
FTIR and Fourier transform near-infrared spectroscopy (FT-NIR) have been used to investigate the effective responses of faba bean saplings following exposure to high levels of inorganic arsenic (Boccia, Meconi, Mecozzi, & Sturchio, 2013;Sturchio, Napolitano, Beni, & Mecozzi, 2012). Spectral data were obtained from freeze-dried, lyophilized root meristems of the seedlings. Changes in the content of polysaccharides, lipids, aliphatic compounds, nucleic acids and proteins were determined, as well as altered levels of hydrogen bonding in these samples. Given the level of changes observed, both quantitative and structural changes in these compounds were likely to be occurring. In particular, it was believed that inorganic arsenic would replace the phosphate groups within DNA molecules, resulting in the proliferation of DNA damage being observed through genotoxicity assays.
Similar work on faba bean roots, stems and leaves used FTIR to investigate the potential mechanisms of aluminium toxicity on faba bean and determine varietal differences in aluminium tolerance (Wang et al., 2011). Through FTIR spectral features, a faba bean variety with predicted high susceptibility to aluminium toxicity was identified. This result was confirmed through the use of traditional heavy metal toxicity assays (root elongation experiment and chrome azurol dyeing).
With growing concerns surrounding pesticide resistance, essential oils have been proposed as a potential bioherbicidal agent. However, their components may have negative physiological effects on the crops as well as the target pest species. Following on from the previously reported work on arsenic and aluminium toxicity, Mecozzi and Sturchio (2015) explored the effects of essential oil treatments upon the protein structure in freeze-dried, lyophilized faba bean roots using diffuse reflectance FTIR. In addition to transitions in the secondary structure of the proteins (α-helices, β-turns and β-sheets), development of random coil structures and oxidation of proteins was observed. Subsequent work incorporating FT-NIR demonstrated further alterations in bonds relating to DNA and other nucleotides, carbohydrate backbones, proteins and lipids (Mecozzi et al., 2017).
One way to increase crop harvest while limiting resource consumption is through mixed cropping of multiple species. However, the underground interactions between the roots of different species are still poorly understood, in part due to the complexity of identifying the species of a root isolated from a soil core. With this in mind, Streit, Meinen, Nelson, Siebrecht-Schöll and Rauber (2019) applied ATR-FTIR to identify ground, dried roots collected from various soil cores.
The faba bean roots were readily discriminated from wheat roots, with no apparent effect of genotype or growing year. The technique was subsequently used to investigate the overyielding effect of different faba bean genotypes grown in combination with winter wheat and thus identify the optimum cropping combination.

| DEVELOPMENTS IN INFRARED SPECTROPHOTOMETERS
One major development that has occurred and continues to occur in this field is the reduction in the cost of spectrophotometers. These almost exclusively use NIR wavelengths, as the apparatus is much cheaper than that for MIRS. The lower prices have made this technology more accessible to small organisations or research groups, increasing the range of potential applications (Kosmowski & Worku, 2018). In concert with the decreasing price have come smaller, more portable instruments. As before, this increases the range of possible applications, even allowing for in-field analysis of crops.
Another aspect of this is in-line spectrophotometers, which allow for much greater analysis speeds. Mainly using transmission spectrophotometry, these may be suitable for the analysis of freshly harvested whole crops. Interest is also growing in single-seed/single-kernel analysis using NIRS (Agelet, 2011). This allows for more accurate sampling of the crops, including the determination of interkernel variation in nutritional parameters, given that each kernel is characterised separately. This technique is likely to play an increasing role in future research (Wrigley, Matakovsky, Melnik, Pascual, & Romanov, 2019).

| CONCLUSION
Both MIRS and NIRS have been used in a range of applications for the prediction of quality assessment of faba bean and/or or faba beanderived products. While NIRS is a well-established analytical tool for the determination of proximate nutritional parameters such as protein, oil and starch content, MIRS is more proficient for establishing the structural composition of constituents such as starch and protein. In the latter instance, MIRS can discriminate between secondary structures such as unordered aggregates, α-helices, β-pleated sheets, β-turns and anti-parallel β-sheets. Additionally, the greater sensitivity of MIRS is likely to make it a better candidate for the determination of micronutrients. However, this aspect remains largely unexplored. In terms of the faba bean crop, there remains much scope for future studies using these technologies, particularly in the area of functional foods.