Rapid estimation of non-starch polysaccharides and lignin in monogastric feed ingredients using near-infrared spectroscopy

The fiber composition in protein meals and by-products is highly variable and is influenced by plant genetics, environmental conditions, and processing. Therefore, enabling the rapid analysis of fiber in various ingredients can support the inclusion of fiber-rich ingredients in monogastric diets. This study aimed to develop and validate near-infrared reflectance spectroscopy (NIRS) calibrations for predicting non-starch polysaccharides (NSP), lignin, and related components in feed ingredients. In this study, 628 samples of diverse feed ingredients and by-products were collected globally between 2015 and 2022. Chemical analyses were conducted to determine the NSP content, including total and insoluble NSP, as well as the monosaccharide sugars. Separate assays were also conducted to determine the cellulose and lignin contents. The same samples were screened using NIRS to predict the NSP, mono-component sugar (e.g., xylose, glucose, and mannose), and cellulose and lignin contents. The diversity in raw materials enabled the development of a global NIRS calibration for the NSP fractions, cellulose, lignin, and key monosaccharides. Global calibrations achieved strong predictive performance for total NSP [R² = 0.96, ratio of standard error of performance to standard deviation (RPD) = 4.6], cellulose (R² = 0.97, RPD = 5.1), and lignin (R² = 0.90, RPD = 2.8), while the monosaccharides were predicted with robust accuracy (R² typically ≥0.90, RPD ≥ 3.0). Ingredient-specific models for soybean meal showed good performance for the key fiber components, with insoluble NSP achieving R² = 0.85 (RPD = 4.5), cellulose with R² = 0.94 (RPD = 4.1), and the xylose and glucose calibrations demonstrating strong accuracy (R² ≥ 0.9). The sunflower meal models also performed well, with total NSP achieving R² = 0.92 (RPD = 3.3) and lignin with R² = 0.90 (RPD = 3.2), while the total xylose and total glucose predictions showed strong accuracy (R² ≥ 0.87, RPD ≥ 3.0). The rapeseed meal models showed strong performance for insoluble NSP (R² = 0.93, RPD = 4.0) and insoluble arabinose (R² = 0.94, RPD = 3.9), while the lignin and cellulose calibrations achieved moderate accuracy (R² = 0.81 and 0.61, respectively). These findings demonstrate the potential of NIRS as a reliable, rapid method for fiber characterization in monogastric feed evaluation.

1 Introduction

Non-starch polysaccharides (NSP) and lignin constitute the fiber fraction that cannot be broken down by the endogenous enzymes of pigs and poultry. This fiber fraction represents the majority of undigested dietary components in commercial diets (Kim et al., 2022). However, their chemical composition is highly diverse across feedstuff sources, processing methods, and environmental conditions (Holtekjølen et al., 2006; Serena and Knudsen, 2007; Knudsen, 2014). Moreover, the biological effects of fiber remain inconsistent. While the inclusion of fiber-rich ingredients, such as sugar beet pulp, has been shown to improve specific markers of gut health, including enhanced short-chain fatty acid production and intestinal barrier function (Feng et al., 2024), the effects on animal performance remain uncertain and inconsistent. Some studies indicate that certain types of fiber, particularly in soluble form, may significantly depress growth performance and nutrient digestibility (Jaworski et al., 2015; Lee et al., 2022). To further complicate the picture, the physiological responses to fiber appear to be influenced not only by its chemical composition but also by its physicochemical (e.g., solubility, and viscosity) and functional (e.g., fermentability) properties (Nguyen et al., 2021; Lee et al., 2022). Furthermore, the level of fiber in the diet dictates the magnitude of these effects (Nguyen et al., 2021; Lee et al., 2022), highlighting the need for accurate and rapid quantification of the fiber fractions.

Fiber has been largely overlooked in feed formulations due to its vague definition and limited analytical resolution. Nutritionists must rely on outdated metrics such as crude fiber (CF), which underestimate the total fiber and fail to account for the significant fractions in ingredients such as soybean meal (Choct, 2015). Similarly, acid detergent fiber (ADF) and neutral detergent fiber (NDF) represent nonspecific fractions and exclude soluble hemicelluloses and pectins (Fahey et al., 2019). These limitations highlight the need for more precise, chemically defined metrics such as NSP and lignin to improve formulation accuracy.

The term “dietary fiber,” which is widely used in human nutrition, is understood within the context of monogastric animal nutrition as the sum of NSP and lignin (Choct, 2015). This classification is based on their shared chemical composition and resistance to enzymatic digestion in the small intestine. Their quantification generally involves enzymatic removal of starch, followed by hydrolysis and gravimetric determination of lignin using standardized analytical procedures (method 991.43; AOAC, 2012).

These methods are reliable, but are often time-consuming and costly, limiting their use for routine analysis. Therefore, to increase the number of samples analyzed and to reduce the turnaround time, quicker and fit-for-purpose methods are needed. In this sense, near-infrared reflectance spectroscopy (NIRS) technology is available for obtaining rapid, non-destructive, and accurate estimation of the chemical composition of feedstuffs (Fontaine et al., 2001; Blanco and Villarroya, 2002). Furthermore, numerous compositional determinations can be made simultaneously, providing an overall characterization of a sample and allowing the nutritional quality to be evaluated quickly. However, very few studies have developed NIRS calibrations for fiber characterization in monogastric feed ingredients. Existing models are typically restricted to specific ingredients, often cereals, and are based on limited sample sizes or region-specific datasets (Owens et al., 2009; Ferreira et al., 2015; Albanell et al., 2021). These limitations reduce the applicability of the current calibrations for feed formulations and highlight the need for robust models built from large, diverse datasets covering multiple fiber fractions across a wide range of ingredients collected worldwide.

Characterizing the fiber fraction in the diverse group of feed ingredients used in monogastric nutrition is a crucial first step toward understanding and managing its potential negative impacts. However, practical application is constrained by the absence of robust, global NIRS calibrations for all NSP fractions and lignin across various feed ingredients. This knowledge gap limits the ability to formulate diets that accurately reflect the actual fiber levels.

Given the growing use of exogenous carbohydrases, such as xylanase, β-glucanase, and β-mannanase, in poultry and swine nutrition, precise quantification of the fiber constituents, particularly the NSP fractions and lignin, has become essential to interpret enzyme responses and to formulate diets with greater confidence.

Building on our previous work, which developed NIR calibrations to predict NSP, monosaccharide sugars, and in vitro starch digestibility in cereals and cereal by-products commonly used in monogastric feed formulations (Nieto-Ortega et al., 2022), the present study extends this approach to a globally sourced, diverse set of feed ingredients and by-products, including protein-rich meals, by developing global and ingredient-specific NIRS calibrations for the rapid prediction of NSP, monosaccharide sugars, and lignin.

2 Materials and methods

2.1 Non-starch polysaccharide, lignin, and cellulose contents of feed ingredients and by-products

A total of 628 feed ingredient samples were collected globally, including soybean meal (140), sunflower meal (87), rice bran (69), rapeseed meal (56), corn DDGS (dried distiller’s grains with solubles) (47), palm kernel meal (36), sugar beet pulp (34), corn gluten feed (32), full-fat soybean (30), corn gluten meal (28), tapioca or cassava (10), pea (9), corn germ meal (8), soy hull (7), alfalfa meal (7), apple fiber (5), grape fruit pulp (5), sunflower seed (5), lupin (3), ground nut expeller meal (3), straw (2), cotton seed (1), oat flour (1), potato (1), wheat gluten (1), and mango by-product (1). The samples were analyzed for NSP (total and insoluble), including individual constituent sugars (e.g., arabinose, xylose, glucose, mannose, galactose, rhamnose, and galacturonic acid), using the gas chromatography version of the assay developed by Englyst et al. (1994). In addition, cellulose was determined as the NSP glucose fraction requiring dispersion with 12 M sulfuric acid treatment. Klason lignin was determined gravimetrically as the acid-hydrolyzed residue using a modified method from Englyst et al. (1992). The modification included an effective enzymatic removal of protein and the soluble carbohydrates to minimize artifacts. The remaining insoluble fraction was treated with 12 M sulfuric acid using the previously published methods of Englyst et al. (1992). Analyses of the NSP, lignin, cellulose, and sugar fractions were performed at Englyst Carbohydrate Ltd. (Southampton, UK) on samples ground using a 0.5-mm sieve.

2.2 Near-infrared reflectance spectroscopy of feed ingredients and by-products

The ingredient samples were ground using a 0.75–1 mm sieve with an Ultra Centrifugal Mill (ZM 200, Retsch, Germany) before their analysis by NIRS. The ISO FDIS 12099 standard guidelines were followed. Each sample was scanned using two spectrometers: model DS2500 scanning monochromator device (FOSS, Hillerød, Denmark) and a Bruker Tango FT-NIR machine (Bruker, Billerica, MA, USA). Hereafter, only the results from the FOSS machine are discussed with similar statistics obtained from the Bruker device. The recorded spectra range was from 400 to 2,500 nm, every 0.5 nm. The spectra were exported to the WinISI software version 4.0, and the spectral data were combined with the chemical reference data. The manipulation of the spectra for the Bruker device was done with OPUS 7.2 software (Bruker-Optics, Billerica, MA, USA).

Global calibration models were developed to characterize the studied parameters of the diverse feed ingredients, while individual calibrations were established for three main protein-rich ingredients (i.e., soybean meal, sunflower meal, and rapeseed meal). This means that there is one NIRS calibration equation for each of the wet chemistry parameters evaluated for the mixture of diverse ingredients, as well as one calibration equation for each parameter for the sunflower meal, soybean meal, and rapeseed meal samples. The NIRS calibrations included NSP (total and insoluble), monosaccharide sugars (total and insoluble), lignin, and cellulose. The modified partial least squares regression (mPLS) method (Shenk and Westerhaus, 1991a) was employed to establish correlations between the reference wet chemistry analysis and the collected NIRS spectra. In the mPLS methodology, similarly to other methodologies, the spectral data are reduced to a few independent factors holding the decisive spectral information. For the mathematical treatment, the standard “SNV and de-trend” method for scatter correction was applied to reduce the particle size effect (Barnes et al., 1989). The mathematical derivative treatment applied was 1.4.4.1 following the nomenclature of Shenk and Westerhaus (1991b). The first digit is the number of the derivative, the second is the gap over which the derivative is calculated in nanometers, the third is the number of data points in smoothing in nanometers, and the last one is a second smooth of 1 nm. The accuracy of the chosen mathematical method was evaluated through the following coefficients: coefficient of determination of calibration (R²), standard error of calibration (SEC), standard error of cross-validation (SECV), and the coefficient of determination of cross-validation (1 − VR). The standard error of prediction (SEP) and the ratio of the standard deviation (SD) of the reference data to the SECV (ratio of performance to deviation, RPD) was used to evaluate the calibration and equation performance.

3 Results

3.1 Chemical analyses of feed ingredients and by-products

The wet chemistry results (i.e., the mean, range, and SD) for the total and insoluble NSP and the constituent sugars (arabinose, xylose, glucose, rhamnose, galacturonic acid, galactose, and mannose) in ingredients and by-products are shown in Tables 1–3 on an “as is” basis. The arabinose-to-xylose ratio was calculated from the total concentrations (Table 3). Soluble NSP and sugars were derived by subtracting the insoluble from the total value for each ingredient.

Table 1

Table 1. Total, insoluble, and soluble arabinose, xylose and non-starch polysaccharide (NSP) content of diverse feed ingredients, % as received.

Table 2

Table 2. Total, insoluble, and soluble glucose, galacturonic acid, and rhamnose contents of diverse feed ingredients (percent as received).

Table 3

Table 3. Calculated arabinose-to-xylose ratio and the total mannose and galactose contents of diverse feed ingredients (percent as received).

The chemical analysis confirmed a high degree of compositional variability across the diverse feed ingredients, which is a key challenge in monogastric feed formulation. Total NSP, constituent sugars, lignin, and cellulose varied widely among ingredients. For instance, total NSP ranged from a low of 1.91% in corn gluten meal to a high of 49.66% in sugar beet pulp.

The data highlight significant differences in the types of fiber across ingredients, which has distinct physiological implications for monogastric animals. Sugar beet pulp, with the highest total NSP (49.66%), was primarily characterized by high levels of arabinose (14.43%), glucose (16.45%), and galacturonic acid (10.81%). In contrast, palm kernel meal (44.41% total NSP) was dominated by mannose (30.2%) and glucose (7.73%). This contrast in the monosaccharide profile, despite similar total NSP contents, demonstrates the necessity for monomeric sugar characterization. Soluble NSP was six times higher in sugar beet pulp than in palm kernel meal (19.86% vs. 2.98%), while palm kernel meal had 1.4 times more insoluble NSP (41.43% vs. 29.78%). Corn gluten meal had the lowest total (1.91%), insoluble (1.51%), and soluble NSP (0.41%).

The major protein-rich ingredients, i.e., soybean meal, rapeseed meal, and sunflower meal, contained 16.24%, 20.47%, and 27.01% total NSP, respectively. In soybean meal, glucose (4.49%) and galactose (3.91%) predominated; in rapeseed meal, glucose (6.8%), arabinose (4.55%), and galacturonic acid (4.06%); and in sunflower meal, glucose (11.46%) and xylose (5.96%). Insoluble NSP accounted for >60% of the total NSP in all three meals, except for sunflower meal where soluble galacturonic acid comprised ~63% of the total galacturonic acid. The mannose content was low (0.12%–1.3%) in all ingredients except for palm kernel meal (30.2%), while rhamnose remained below 1.1% across all feed ingredients. Galactose was highest in lupins (11.3%) and ranged from 0.15% to 4.24% in other feedstuffs.

The cellulose and lignin contents (Table 4) also varied across ingredients, confirming the high heterogeneity of the fiber matrix. Sugar beet pulp had the highest cellulose (16.18%), followed by peas (14.35%), lupins (11.4%), and sunflower meal (10.82%). Moderate levels were found in rice bran (8.05%), palm kernel meal (7.1%), corn DDGS (6.55%), rapeseed meal (6.04%), corn gluten feed (5.86%), and soybean meal (4.26%). The lignin content in palm kernel meal (9.98%), sunflower meal (9.12%), rapeseed meal (8.18%), and rice bran (7.34%) was approximately four to five times higher than that in other ingredients (≤2.1%).

Table 4

Table 4. Determined cellulose and lignin contents of diverse feed ingredients (percent as received).

3.2 NIRS predictions using individual calibrations: soybean meal, sunflower meal, and rapeseed meal

Individual calibration equations for the main protein-rich ingredients (soybean meal, sunflower meal, and rapeseed meal) were developed. The number of samples, mean, minimum, maximum, and SD of the calibration files (set) used for NIRS prediction of the NSP content (insoluble and total), along with the corresponding calibrations and cross-validation statistics, are depicted in Tables 5–7.

Table 5

Table 5. Predicted nutrient content of soybean meal and near-infrared reflectance spectroscopy (NIRS) performance indices of predictability (individual calibration).

Table 6

Table 6. Predicted nutrient content of sunflower meal and near-infrared reflectance spectroscopy (NIRS) performance indices of predictability (individual calibration).

Table 7

Table 7. Predicted nutrient content of rapeseed meal and near-infrared reflectance spectroscopy (NIRS) performance indices of predictability (individual calibration).

The individual calibrations for soybean meal resulted in a robust correlation (R² > 0.85) between the NIRS prediction and the wet chemistry for the NSP, cellulose, xylose, and glucose contents (Table 5). The 1 − VR values were also >0.80 for the same parameters. For the mannose, galactose, and galacturonic acid contents, the correlation was lower, with R² values of 0.75, 0.73, and 0.78, respectively. The estimate of the total NSP content based on the calibration equations was within the limits of 13%–22% (Figure 1A). For the total NSP and insoluble NSP calibrations, the SEP values were 0.87% and 0.84% and the RPD values 4.53 and 2.66, respectively. The estimate of cellulose content based on the calibration equations was within the limits of 2.5%–7%, with an R² of 0.94. The SEP was close to the SECV and SEC, with values between 0.37% and 0.39% and an RPD value of 4.05. The calibrations for the total and insoluble xylose and glucose fractions showed similar behavior to those for total and insoluble NSP. The estimated xylose content was within the limits of 0.5%–2.1% for the insoluble fraction and 0.9%–2.1% for the total fraction. The R² values were 0.90 and 0.94, respectively, and the 1 − VR values were 0.92 and 0.93 for the insoluble and total fractions, respectively. The SEP values were very low at 0.12% and 0.09% for the insoluble and total xylose contents, respectively, indicating a good correlation. The glucose calibration showed higher SEP values, with 0.42% for the insoluble fraction and 0.38% for the total fraction. This SEP is closer or identical to the SECV and SEC values in both cases, and the calibration had an RPD > 3.0. The estimated mannose content was within the limits of 0.6%–1.3%, with an R² of 0.75. The mean value was 0.93%, and SEP, SECV, and SEC had the same value of 0.14%. The estimated galactose content was within 3.2%–4.8%, with an R² of 0.73. In contrast, the galacturonic acid values were more limited, 1%–2.1% for the insoluble fraction and 1.7%–3.2% for the total fraction. The SEP, SECV, and SEC parameters were 0.22% and 0.23% for the insoluble and total fractions, respectively. The lignin, arabinose, and rhamnose calibrations were also evaluated; however, their low concentrations and limited variability in soybean meal limited the development of robust individual NIRS calibrations.

Figure 1

Figure 1. Near infrared reflectance spectroscopy (NIRS) calibrations for (A) total non-starch polysaccharide (NSP) in soybean meal (n = 135, R²=0.85, SEP = 0.87%); (B) cellulose in sunflower meal (n=63, R²=0.86, SEP = 1.05%); and (C) galacturonic acid in rapeseed meal (n = 65, R² = 0.79, SEP 0.2%).

Individual NSP and constituent sugar, cellulose, and lignin NIRS calibrations for sunflower meal were also developed (Table 6). The number of samples of this raw material was slightly lower (n = 87); however, the variability of the samples was sufficient and allowed for the development of robust individual NIRS calibrations. The validation of the NSP calibrations was robust, with an R² and 1 − VR greater than 0.85 for both total and insoluble NSP. The SEP was higher (1.54%) for the insoluble NSP compared with the total NSP (1.24%). In addition, the R² and 1 − VR parameters were 0.86, which is an indication of a robust calibration. The estimate of the cellulose concentration according to the calibration equation ranged from 8% to 14% (Figure 1B). The SEP, SEC, and SECV were approximately 1.06%, with an RPD of 2.67. For lignin, the NIRS calibration equation showed SEP, SEC, and SECV values below 1%. The RPD value was 3.20, which represents reliable predictive performance. The contents of rhamnose, mannose, and galactose were too low in sunflower meal to obtain suitable NIRS calibrations for this ingredient. The xylose, arabinose, glucose, and galacturonic acid concentrations and variations in the samples were adequate for the development of robust individual NIRS calibrations. The R² and 1 − VR for insoluble xylose and total xylose were above 0.86. The measured errors (SEP, SEC, and SECV) were between 0.54% and 0.67%. Surprisingly, we measured a small calibration range for arabinose, but with the R² and 1 − VR above 0.85, meaning that the NIRS could detect the small variation in the arabinose content in sunflower meal. The corresponding errors (SEP, SEC, and SECV) were around 0.12%, with RPD values of 2.82 and 1.76 for the insoluble and total fractions, respectively. Although the R² values for galacturonic acid were relatively low (0.66 for the insoluble fraction and 0.80 for the total content), the calibration curves still demonstrated acceptable linearity for quantification within individual ingredients. The R² and 1 − VR values were close to 0.90, with an RPD higher than 3.0 in the insoluble glucose and total glucose fractions. The calculated errors (SEP, SEC, and SECV) were around 0.9%, which is an indication of a robust calibration for the glucose measurements in sunflower meal.

For rapeseed meal, the RPD values were higher or near an RPD of 3 for the majority of the parameters analyzed. The 1 − VR values for the insoluble and total NSP were slightly lower than the R² values (1 − VR values of 0.67 and 0.70 and R² values of 0.93 and 0.88 for the insoluble and total NSP, respectively). The error values (SEP, SEC, and SECV) were around 0.41%. In the case of rapeseed meal, the cellulose calibration equation did not show the robustness observed in soybean meal and sunflower meal. An individual calibration equation was also built for the lignin content in rapeseed meal, with an R² of 0.81 and an RPD of 2.29%. The calculated SEP, SEC, and SECV errors were around 0.6%. The low concentrations of rhamnose, galactose, and mannose from the wet chemistry analyses (Table 1) prevented the development of robust NIRS calibrations for these three sugar fractions (data not shown). The insoluble arabinose fraction had a greater range; therefore, the R² and 1 − VR values indicated a slightly better prediction compared with the total fraction (R² of 0.94 for the insoluble fraction vs. 0.83 for the total fraction). The SEP, SEC, and SECV values were around 0.12% for both the insoluble and total arabinose contents. The R² for both total and insoluble xylose were ≥0.90, with the 1 − VR around 0.72, an indication of good linearity between the NIRS-predicted parameter and the wet chemistry reference method. The insoluble glucose calibration equation showed less linearity with the wet chemistry reference method (R² = 0.49) when compared with the total glucose calibration for rapeseed meal (R² = 0.82). The SEP, SEC, and SECV values were around 0.2%–0.3% for both insoluble and total glucose. The galacturonic acid content ranged from 1.8% to 3.3% for the insoluble fraction and from 3% to 4.8% for the total fraction (Figure 1C). The R² was approximately 0.80 for both parameters, while the 1 − VR was lower at 0.55 for the insoluble fraction and was 0.69 for total galacturonic acid. The calculated errors (SEP, SEC, and SECV) were between 0.20 and 0.25.

3.3 NIRS predictions using a global calibration containing diverse feed ingredients and by-products

In addition to the individual calibrations from the three protein-rich ingredients, global calibrations for each of the total and insoluble NSP fractions, monosaccharide sugars, lignin, and cellulose were developed, including the complete sample set of approximately 628 samples. The number of samples and the mean, minimum, maximum, and SD used for the NIRS predictions of the NSP content (insoluble and total) and the cellulose, lignin, and sugar fractions, along with the corresponding calibrations and cross-validation statistics, are depicted in Table 8.

Table 8

Table 8. Predicted nutrient content in the global calibration (diverse feed ingredients) and near-infrared reflectance spectroscopy (NIRS) performance indices of predictability.

The global calibration achieved excellent predictive performance for the most relevant fiber fractions, supporting the application of NIRS for routine screening. The highest predictability was observed for cellulose (R² = 0.97, RPD = 5.1) and total NSP (R² = 0.96, RPD = 4.6). Lignin also showed strong predictive accuracy (R² = 0.90, RPD = 2.8). The ability to accurately predict individual sugar components is a significant finding as it allows nutritionists to move beyond simple total NSP or total fiber values and formulate diets based on the specific type of fiber (e.g., mannan, arabinoxylan, or pectin) present in the ingredient.

The correlation between the NIRS and the reference data for the validation of the total NSP content was robust (R² = 0.96) (Table 8), with a 1 − VR of 0.96, confirming the capability of the global NIRS calibrations to predict the total NSP content in various feed ingredients. The estimated content of total NSP, as determined using the calibration equations, ranged from 0.1% to 60% (Figure 2A). The insoluble NSP calibration ranged from 0.1% to 51%, with an R² of 0.96. For both calibrations, the SEP was close to the SECV and SEC, with values around 2.2% and 2.32% for both fractions. The RPD values were between 5.26 and 4.59, respectively.

Figure 2

Figure 2. Near infrared reflectance spectroscopy (NIRS) global calibrations for (A) total non-starch polysaccharide (NSP) (n=624, R² = 0.97, SEP = 2.32 g/100); (B) cellulose (n = 435, R²= 0.97, SEP 7 = 0.94%); (C) lignin (n = 525, R²=0.90, SEP = 1.63%); and (D) galactose (n = 595, R² = 0.93, SEP = 8 0.36%.

The cellulose calibration exhibited an R² of 0.97 and a 1 − VR of 0.98 (Figure 2B). The calculated errors were <1%. The RPD value was 5.12. For lignin, the linearity was also strong, with an R² of 0.90 and a 1 − VR of 0.91 (Figure 2C). The calculated errors (SEP, SECV, and SEC) were ~1.63%, and the RPD value was 2.81.

For the monosaccharide sugar fractions, the development of a global calibration equation was possible for all parameters, except for rhamnose. For both total and insoluble arabinose, the R² and the 1 − VR presented values >0.90 (Table 8). The calculated errors were around 0.5% for both the insoluble and total arabinose fractions, and the RPD values were 2.94 and 5.12 for the insoluble and total arabinose, respectively. The xylose fraction in the global calibration performed in a similar manner to the arabinose calibration. The R² and the 1 − VR showed values >0.90 as well, with RPDs of 3.27 and 3.31 for the total fraction and the insoluble fraction, respectively. The calculated errors for xylose were higher than those for arabinose. The SEC, SEP, and SECV were around 0.8%, which is higher than the calculated errors from the individual calibration, in particular for soybean meal (see Tables 1, 8). The total and insoluble glucose calibration equations showed an R² and 1 − VR above 0.96, whereas the RPD values were >4.60. For total mannose, the calculated errors (SEC, SEP, and SECV) were around 0.78% and the RPD was 6.86. The galactose calibration exhibited an R² and 1 − VR of 0.93 and 0.96, respectively, with an RPD of 3.67 (Figure 2D). For galactose, the SEC, SEP, and SECV were 0.36% and the RPD value was 3.67. In the case of glucuronic acid, the R² and the 1 − VR were above 0.80, and the RPD was 2.17 for the insoluble fraction and was 3.58 for the total glucuronic acid. The calculated errors were 0.4% for the insoluble fraction and 0.63% for the total fraction.

4 Discussion

This study revealed considerable variability in NSP and its constituent sugar, lignin, and cellulose contents both across and within feed ingredients, emphasizing the need for rapid and accurate quantification of the fiber components. The results of the wet chemistry analyses of NSP, cellulose, and lignin in the protein-rich ingredients (i.e., soybean meal, rapeseed meal, and sunflower meal) are generally consistent with previously reported values (Lannuzel et al., 2022). Similarly, the fiber composition of corn DDGS, peas, lupins, and rice bran aligned with earlier findings by Sinha et al. (2011) and Casas and Stein (2016). Taken together, these results indicate that the analytical concentrations of the NSP, monosaccharides, cellulose, and lignin used for the NIRS calibrations are robust and consistent with previously published wet chemistry data. This is especially important as some of the ingredient variability reported in the literature may be the result of the different enzymatic assays used to determine the NSP content (Agyekum and Nyachoti, 2017; Ibáñez et al., 2020; Lannuzel et al., 2022). In the present work, this was mitigated by using the same assays and laboratory to conduct all wet chemistry analyses.

The successful development of the NIRS calibrations represents a significant advance in feed analysis, providing a validated, rapid, and cost-effective alternative to the conventional wet chemistry for fiber characterization. While the reference method (Englyst et al., 1994) remains the gold standard for chemical accuracy, it is labor-intensive, requires specialized laboratory equipment, uses hazardous reagents, and typically takes several days to complete. This lengthy turnaround time is a significant constraint for routine quality control and real-time feed formulation adjustments, especially given the high compositional variability observed in feed ingredients.

The compositional variability in the NSP content of soybean meal, rapeseed meal, and sunflower meal can be attributed to the differences in the oil extraction methods and the extent of hull reintroduction to the meal. Similarly, the variability in corn DDGS, palm kernel meal, and corn gluten meal can reflect differences in the processing conditions for starch, syrup, or oil extraction (Giannenas et al., 2017; Caldas et al., 2020).

The samples were sourced from several countries over approximately 6 years, and the exact production process applied to each sample is unknown. Additional sources of compositional variability within each ingredient include growing conditions, harvest and storage practices, and genetic variety (Lannuzel et al., 2022). Maharjan et al. (2019) reported substantial variability in the fiber composition among different soybean genetic lines, with the insoluble uronic acid content ranging from 2.2% to 5.3%, depending on the genotype. Furthermore, the total dietary fiber content has been shown to be negatively correlated with the energy digestibility in pigs, as observed by Lopez et al. (2020).

NIRS is a widely used tool for rapid assessment of the chemical composition of feed ingredients and diets. Traditionally used to predict chemical parameters such as ash, moisture, CF, and crude protein (Williams, 2006), its application has expanded to include predictions of the NSP content, starch digestibility, the cellulose and lignin in cereal grains (Nieto-Ortega et al., 2022), and anti-nutritional factors such as phytate-phosphorus (Aureli et al., 2017). In the current work, the NIRS calibrations developed to predict the fiber composition demonstrated strong performance and closely matched the corresponding wet chemistry data. The coefficients of determination (R²) were generally high for the total and insoluble fractions across all calibrations, except for mannose, galacturonic acid, and galactose. In these cases, lower concentrations, particularly in the individual calibrations, limited the predictive accuracy, resulting in fair model performance. Overall, the RPD values for the global calibrations were above 2.5 for all parameters, except for the insoluble galacturonic acid, indicating excellent predictive performance. In general, the total fractions were predicted with slightly better accuracy than their insoluble counterparts. For instance, total xylose and glucose showed higher R² values (0.94 and 0.92, respectively) compared with their insoluble forms (both at 0.90). For soybean meal and rapeseed meal, majority of the RPD values exceeded 2.5, which is generally considered good, while values ≥3 are indicative of excellent prediction. The prediction errors (SEC, SECV, and SEP) were lower in the individual calibrations (soybean meal, sunflower meal, and rapeseed meal) compared with the global calibration, likely due to the reduced analytical variability within single ingredients relative to the broader variability observed across diverse ingredients in the global set. The SEP for the total NSP content in the global calibration was 2.32%, while lower values were observed in the ingredient-specific calibrations. Soybean meal showed an SEP of 0.87%, sunflower meal of 1.24%, and rapeseed meal of 0.43%. The lower prediction errors observed in the ingredient-specific calibrations likely reflect the reduced compositional variability within single ingredients compared with the broader heterogeneity captured in the global calibration. This observation is consistent with previous findings, emphasizing that the robustness of NIRS calibrations depends more on the diversity of the calibration set than on its size (Nieto-Ortega et al., 2022). A representative dataset must cover the full range of values for the target parameters, such as the NSP content, to ensure reliable predictions. Achieving this level of variability is particularly challenging for single ingredients, which highlights the importance of globally sourced and heterogeneous datasets for the development of applicable models.

In the global calibration, reliable prediction of the NSP fractions such as mannose, arabinose, galactose, and glucuronic acid was enabled by the broader concentration range and greater variability in the sample set, resulting in an improved linearity compared with the individual calibrations. For example, the global calibration achieved an R² of 0.97 for arabinose, whereas the sunflower-specific calibration reached 0.89. However, this increased variability also led to higher SEP values, with total xylose showing an SEP of 0.84% in the global calibration compared with 0.53% in the sunflower calibration. Despite this, all fractions except rhamnose were predicted with good or very good agreement with the wet chemistry data. The lignin content was also determined across the three ingredients and in the global calibration. While soybean meal exhibited low lignin levels (0.1%–0.65%), limiting the calibration performance, rapeseed and sunflower meals showed higher and more variable lignin concentrations, enabling robust individual calibrations. The global calibration, incorporating all samples, demonstrated good predictive performance relative to the wet chemistry values.

The purpose of the NIRS technology has changed over the last years from predicting the gross chemical content of feed ingredients to a broad range of applications including the use of NIRS to predict the carbohydrate composition (Pérez-Marı́n et al., 2004; Nieto-Ortega et al., 2022). Majority of the literature, however, is based on the analysis of one ingredient, such as barley (Albanell et al., 2021), soybean meal (Ferreira et al., 2015), or wheat (Owens et al., 2009), or a mix of cereal grains (Blakeney and Flinn, 2005). The results obtained in the literature are comparable to the results described in this work and show that NIRS can predict the ingredient NSP, monosaccharide sugar, cellulose, and lignin contents in a fast, reliable, and accurate manner, thereby replacing or mitigating the need for time-consuming chemical methodology.

Currently, the term CF is commonly used in monogastric feed formulation to estimate the dietary fiber levels. However, relying solely on CF can be misleading as it does not capture the complexity of the fiber composition or differentiate between the soluble and insoluble NSP fractions, which can have markedly different effects on the nutrient digestibility, microbial fermentation, and gut health (Agyekum and Nyachoti, 2017). In this study, for example, sugar beet pulp showed the highest total NSP level (49.66%), primarily composed of arabinose (14.43%), glucose (16.45%), and galacturonic acid (10.81%). Palm kernel meal followed with 44.41% total NSP, with mannose (30.2%) and glucose (7.73%) as the main constituents. These contrasting sugar profiles highlight the importance of quantifying both the total NSP and the individual sugar fractions as their composition can significantly influence the functional and nutritional properties of feed ingredients. For example, the high mannose content in palm kernel meal (30.2%) suggests a greater proportion of β-mannans, which are known to be highly viscous and can negatively impact the nutrient digestibility in monogastric animals (Azizi et al., 2021). Conversely, the high galacturonic acid content in sugar beet pulp indicates a high proportion of pectin, which is highly soluble and fermentable (Feng et al., 2024).

A distinctive feature of this study is the dual calibration strategy, which integrates both global and ingredient-specific modeling approaches. While previous research has typically focused either on broad-spectrum calibrations across diverse feedstuffs or on narrowly defined ingredient groups, the present work combines both by developing robust NIR calibrations using a wide range of feed ingredients and, uniquely, by establishing dedicated models for key protein-rich sources such as soybean meal, rapeseed meal, and sunflower meal. To the best of our knowledge, this is the first study to report NIR calibrations for such an extensive set of carbohydrate residues while simultaneously addressing ingredient-specific variability in protein-rich feedstuffs. Enabling real-time quantification of specific soluble and insoluble NSP fractions will be essential for improving the sustainable and efficient use of protein meals and by-products in monogastric production. The application of NIRS to estimate the chemical composition of feed ingredients, account for NSP fraction variability, and support precision feed formulation, including the strategic use of exogenous fiber-degrading enzymes, has the potential to support poultry and swine production, reduce feed costs, and improve overall sustainability.

5 Conclusion

The rapid and reliable estimation of NSP, mono-component sugars, cellulose, and lignin using NIRS offers animal nutritionists a powerful tool to optimize the inclusion and utilization of feed ingredients in monogastric diets. Such rapid and precise analysis is not only essential for optimizing the feed formulation and ingredient selection but also for guiding the targeted use of exogenous carbohydrases, ensuring that the right type and dosage are applied to effectively modulate fiber utilization and enhance nutrient availability. The NIRS calibrations developed for these components demonstrated strong robustness, with high correlations between the reference analyses and the predicted values. Global calibrations, encompassing a broader range of sample diversity, enhanced applicability across various feed ingredients. When sufficient sample size and variability are available, as observed for soybean meal, rapeseed meal, and sunflower meal, individual calibrations resulted in lower prediction errors than the global models. These findings confirm the robustness and reliability of NIRS for quantifying the total and insoluble NSP, cellulose, lignin, and mono-component sugars such as arabinose, xylose, glucose, galacturonic acid, mannose, and galactose across a wide array of feed ingredients used in monogastric nutrition. This study represents a significant step forward in unraveling the structural and compositional complexity of dietary fiber in monogastric feed ingredients. While fiber is known to affect gut health and nutrient digestibility, its diverse nature remains insufficiently captured in current nutritional models. By enabling rapid quantification of distinct fiber constituents, NIRS provides a foundation for more precise characterization and functional interpretation of fiber in feed formulations.

Data availability statement

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Author contributions

GA: Conceptualization, Writing – original draft, Writing – review & editing. BN: Writing – original draft, Validation, Formal Analysis. JA: Writing – review & editing, Validation, Methodology. NC: Writing – review & editing, Resources. KS: Conceptualization, Writing – review & editing. AS: Conceptualization, Project administration, Writing – review & editing.

Funding

The author(s) declared that financial support was not received for this work and/or its publication.

Conflict of interest

Authors GP, BN-O and KS were employed by company DSM-Firmenich.

Authors JJ-A and NC were employed by company DSM Nutritional Products Iberia SA.

Author AS was employed by company DSM Nutritional Products Ltd.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

Agyekum A. K. and Nyachoti C. M. (2017). Nutritional and metabolic consequences of feeding high-fiber diets to swine: A review. Engineering 3, 716–725. doi: 10.1016/J.ENG.2017.03.010

Crossref Full Text | Google Scholar

Albanell E., Martínez M., De Marchi M., and Manuelian C. L. (2021). Prediction of bioactive compounds in barley by near-infrared reflectance spectroscopy (NIRS). J. Food Compos. Anal. 97, 103763. doi: 10.1016/j.jfca.2020.103763

Crossref Full Text | Google Scholar

AOAC (2012). Official Methods of Analysis. 19th ed (Arlington, VA, USA: AOAC International).

Google Scholar

Aureli R., Ueberschlag Q., Klein F., Noël C., and Guggenbuhl P. (2017). Use of near infrared reflectance spectroscopy to predict phytate phosphorus, total phosphorus, and crude protein of common poultry feed ingredients. Poult. Sci. 96, 160–168. doi: 10.3382/ps/pew214

PubMed Abstract | Crossref Full Text | Google Scholar

Azizi M. N., Loh T. C., Foo H. L., and Teik Chung E. L. (2021). Is palm kernel cake a suitable alternative feed ingredient for poultry? Animals 11, 338. doi: 10.3390/ani11020338

PubMed Abstract | Crossref Full Text | Google Scholar

Barnes R. J., Dhanoa M. S., and Lister S. J. (1989). Standard normal variate transformation and de-trending of near-infrared diffuse reflectance spectra. Appl. Spectrosc. 43, 772–777. doi: 10.1366/0003702894202201

Crossref Full Text | Google Scholar

Blakeney A. B. and Flinn P. C. (2005). Determination of non-starch polysaccharides in cereal grains with near-infared reflectance spectroscopy. Mol. Nutr. Food Res. 49, 546–550. doi: 10.1002/mnfr.200500038

PubMed Abstract | Crossref Full Text | Google Scholar

Blanco M. and Villarroya I. (2002). NIR spectroscopy: a rapid-response analytical tool. TrAC Trends Anal. Chem. 21, 240–250. doi: 10.1016/S0165-9936(02)00404-1

Crossref Full Text | Google Scholar

Caldas J. V., Hilton K., Mullenix G., Xuemei D., England J. A., and Coon C. N. (2020). Corn distillers dried grains with solubles: nutrient analysis, metabolizable energy, and amino acid digestibility in broilers. J. Appl. Poult. Res. 29, 1068–1083. doi: 10.1016/j.japr.2020.09.015

Crossref Full Text | Google Scholar

Casas G. A. and Stein H. H. (2016). Effects of microbial xylanase on digestibility of dry matter, organic matter, neutral detergent fiber, and energy and the concentrations of digestible and metabolizable energy in rice coproducts fed to weanling pigs1. J. Anim. Sci. 94, 1933–1939. doi: 10.2527/jas.2015-0064

PubMed Abstract | Crossref Full Text | Google Scholar

Choct M. (2015). Feed non-starch polysaccharides for monogastric animals: classification and function. Anim. Prod. Sci. 55, 1360–1366. doi: 10.1071/AN15276

Crossref Full Text | Google Scholar

Englyst H. N., Quigley M. E., and Hudson G. J. (1994). Determination of dietary fibre as non-starch polysaccharides with gas–liquid chromatographic, high-performance liquid chromatographic or spectrophotometric measurement of constituent sugars. Analyst 119, 1497–1509. doi: 10.1039/AN9941901497

PubMed Abstract | Crossref Full Text | Google Scholar

Englyst H. N., Quigley M. E., Hudson G. J., and Cummings J. H. (1992). Determination of dietary fibre as non-starch polysaccharides by gas–liquid chromatography. Analyst 117, 1707–1714. doi: 10.1039/AN9921701707

PubMed Abstract | Crossref Full Text | Google Scholar

Fahey G. C., Novotny L., Layton B., and Mertens D. R. (2019). Critical factors in determining fiber content of feeds and foods and their ingredients. J. AOAC Int. 102, 52–62. doi: 10.5740/jaoacint.18-0067

PubMed Abstract | Crossref Full Text | Google Scholar

Feng L., Luo Z., Wang J., Wu K., Wang W., Liu Z., et al. (2024). Effects of different ratios of soluble to insoluble dietary fiber on growth performance and intestinal health of piglets. Anim. Nutr. 18, 257–271. doi: 10.1016/j.aninu.2024.05.005

PubMed Abstract | Crossref Full Text | Google Scholar

Ferreira D. S., Poppi R. J., and Lima Pallone J. A. (2015). Evaluation of dietary fiber of Brazilian soybean (Glycine max) using near-infrared spectroscopy and chemometrics. J. Cereal Sci. 64, 43–47. doi: 10.1016/j.jcs.2015.04.004

Crossref Full Text | Google Scholar

Fontaine J., Hörr J., and Schirmer B. (2001). Near-infrared reflectance spectroscopy enables the fast and accurate prediction of the essential amino acid contents in soy, rapeseed meal, sunflower meal, peas, fishmeal, meat meal products, and poultry meal. J. Agric. Food Chem. 49, 57–66. doi: 10.1021/jf000946s

PubMed Abstract | Crossref Full Text | Google Scholar

Giannenas I., Bonos E., Anestis V., Filioussis G., Papanastasiou D. K., Bartzanas T., et al. (2017). Effects of protease addition and replacement of soybean meal by corn gluten meal on the growth of broilers and on the environmental performances of a broiler production system in Greece. PloS One 12, e0169511. doi: 10.1371/journal.pone.0169511

PubMed Abstract | Crossref Full Text | Google Scholar

Holtekjølen A. K., Uhlen A. K., Bråthen E., Sahlstrøm S., and Knutsen S. H. (2006). Contents of starch and non-starch polysaccharides in barley varieties of different origin. Food Chem. 94, 348–358. doi: 10.1016/j.foodchem.2004.11.022

Crossref Full Text | Google Scholar

Ibáñez M. A., de Blas C., Cámara L., and Mateos G. G. (2020). Chemical composition, protein quality and nutritive value of commercial soybean meals produced from beans from different countries: A meta-analytical study. Anim. Feed Sci. Technol. 267, 114531. doi: 10.1016/j.anifeedsci.2020.114531

Crossref Full Text | Google Scholar

Jaworski N. W., Lærke H. N., and Stein H. H. (2015). Carbohydrate composition and in vitro digestibility of dry matter and nonstarch polysaccharides in corn, sorghum, and wheat and coproducts from these grains1. J. Anim. Sci. 93, 1103–1113. doi: 10.2527/jas.2014-8147

PubMed Abstract | Crossref Full Text | Google Scholar

Kim E., Morgan N. K., Moss A. F., Li L., Ader P., and Choct M. (2022). Characterisation of undigested components throughout the gastrointestinal tract of broiler chickens fed either a wheat- or maize-based diet. Anim. Nutr. 8, 153–159. doi: 10.1016/j.aninu.2021.09.011

PubMed Abstract | Crossref Full Text | Google Scholar

Knudsen K. E. B. (2014). Fiber and nonstarch polysaccharide content and variation in common crops used in broiler diets1. Poult. Sci. 93, 2380–2393. doi: 10.3382/ps.2014-03902

PubMed Abstract | Crossref Full Text | Google Scholar

Lannuzel C., Smith A., Mary A. L., Della Pia E. A., Kabel M. A., and de Vries S. (2022). Improving fiber utilization from rapeseed and sunflower seed meals to substitute soybean meal in pig and chicken diets: A review. Anim. Feed Sci. Technol. 285, 115213. doi: 10.1016/j.anifeedsci.2022.115213

Crossref Full Text | Google Scholar

Lee G. I., Hedemann M. S., Jørgensen H., and Bach Knudsen K. E. (2022). Influence of dietary fibre on nutrient digestibility and energy utilisation in growing pigs fed diets varying in soluble and insoluble fibres from co-products. Animal 16, 100511. doi: 10.1016/j.animal.2022.100511

PubMed Abstract | Crossref Full Text | Google Scholar

Lopez D. A., Lagos L. V., and Stein H. H. (2020). Digestible and metabolizable energy in soybean meal sourced from different countries and fed to pigs. Anim. Feed Sci. Technol. 268, 114600. doi: 10.1016/j.anifeedsci.2020.114600

Crossref Full Text | Google Scholar

Maharjan P., Mayorga M., Hilton K., Weil J., Beitia A., Caldas J., et al. (2019). Non-cellulosic polysaccharide content in feed ingredients and ileal and total tract non-cellulosic polysaccharide digestibility in 21- and 42-day-old broilers fed diets with and without added composite enzymes. Poult. Sci. 98, 4048–4057. doi: 10.3382/ps/pez079

PubMed Abstract | Crossref Full Text | Google Scholar

Nguyen H. T., Bedford M. R., Wu S.-B., and Morgan N. K. (2021). Soluble non-starch polysaccharide modulates broiler gastrointestinal tract environment. Poult. Sci. 100, 101183. doi: 10.1016/j.psj.2021.101183

PubMed Abstract | Crossref Full Text | Google Scholar

Nieto-Ortega B., Arroyo J.-J., Walk C., Castañares N., Canet E., and Smith A. (2022). Near infrared reflectance spectroscopy as a tool to predict non-starch polysaccharide composition and starch digestibility profiles in common monogastric cereal feed ingredients. Anim. Feed Sci. Technol. 285, 115214. doi: 10.1016/j.anifeedsci.2022.115214

Crossref Full Text | Google Scholar

Owens B., McCann M. E. E., McCracken K. J., and Park R. S. (2009). Prediction of wheat chemical and physical characteristics and nutritive value by near-infrared reflectance spectroscopy. Br. Poult. Sci. 50, 103–122. doi: 10.1080/00071660802635347

PubMed Abstract | Crossref Full Text | Google Scholar

Pérez-Marı́n D. C., Garrido-Varo A., Guerrero-Ginel J. E., and Gómez-Cabrera A. (2004). Near-infrared reflectance spectroscopy (NIRS) for the mandatory labelling of compound feedingstuffs: chemical composition and open-declaration. Anim. Feed Sci. Technol. 116, 333–349. doi: 10.1016/j.anifeedsci.2004.05.002

Crossref Full Text | Google Scholar

Serena A. and Knudsen K. E. B. (2007). Chemical and physicochemical characterisation of co-products from the vegetable food and agro industries. Anim. Feed Sci. Technol. 139, 109–124. doi: 10.1016/j.anifeedsci.2006.12.003

Crossref Full Text | Google Scholar

Shenk J. S. and Westerhaus M. O. (1991a). Population definition, sample selection, and calibration procedures for near infrared reflectance spectroscopy. Crop Sci. 31, cropsci1991.0011183X003100020049x. doi: 10.2135/cropsci1991.0011183X003100020049x

Crossref Full Text | Google Scholar

Shenk J. S. and Westerhaus M. O. (1991b). Populations structuring of near infrared spectra and modified partial least squares regression. Crop Sci. 31, cropsci1991.0011183X003100060034x. doi: 10.2135/cropsci1991.0011183X003100060034x

Crossref Full Text | Google Scholar

Sinha A. K., Kumar V., Makkar H. P. S., De Boeck G., and Becker K. (2011). Non-starch polysaccharides and their role in fish nutrition – A review. Food Chem. 127, 1409–1426. doi: 10.1016/j.foodchem.2011.02.042

Crossref Full Text | Google Scholar

Williams P. (2006). “Near-infrared spectroscopy of cereals,” in Handbook of Vibrational Spectroscopy (USA: John Wiley & Sons, Ltd). doi: 10.1002/0470027320.s6601

Crossref Full Text | Google Scholar