Near-Infrared Spectroscopy for Predicting Fumonisin and Deoxynivalenol in Maize: Development of Preliminary Chemometric Models ^†

Abstract

Fumonisins and deoxynivalenol (DON) are toxic secondary metabolites, produced by Fusarium species frequently contaminating maize, representing a critical challenge to food safety and human health. Conventional analytical methods, such as HPLC and ELISA, are accurate but time-consuming and require complex sample preparation. In contrast, near-infrared spectroscopy (NIR) has emerged as a rapid, non-destructive, and cost-effective alternative for mycotoxin screening. This study investigates the potential of NIR spectroscopy combined with chemometric techniques to detect and quantify fumonisins (primarily FB1 and FB2) and DON in maize. A total of 60 maize samples were analyzed with mean concentrations of 534 µg/kg for FB1, 208 µg/kg for FB2, and 130 µg/kg for DON. The highest cumulative contamination of FB1 + FB2 reached 3420 µg/kg. While 30% of the samples showed no detectable fumonisin contamination, DON was absent in 17% of the samples. The best performing predictive models were developed using second-derivative pre-processing of the NIR spectra. The NIR calibration model yielded coefficients of determination (R²) of 0.91 for FB1, 0.88 for FB2, and 0.92 for DON, with corresponding root mean square errors (RMSE) of 683, 282, and 115 µg/kg, respectively. These results demonstrate that NIR spectroscopy, particularly when integrated with multivariate analysis, is a promising tool for distinguishing contaminated from uncontaminated maize and estimating mycotoxin levels with reasonable accuracy. These findings support the application of NIR as a practical tool for routine screening and quality control in the maize supply chain.

1. Introduction

Maize (Zea mays L.) is a vital crop for both food and feed, ranking as the most widely produced cereal worldwide, with 1.2 million tonnes of production in 2022. In Portugal, maize is the fourth most-produced food crop, with approximately 720,000 tonnes [1].

Maize varieties have their own distinct applications in feed and various food products. In the food industry, maize is widely utilized to produce flour, breakfast cereals, bread, snacks, porridge, and alcoholic and non-alcoholic beverages [2].

However, maize is highly susceptible to contamination by mycotoxins—toxins produced by molds in contaminated food—which represent significant risks to human and animal health, reaching from gastrointestinal and kidney disorders to immune deficiency and cancer [3]. Common mycotoxins in maize include aflatoxins, fumonisins, deoxynivalenol (DON), zearalenone, and ochratoxin A, produced by fungi such as Aspergillus, Fusarium, and Penicillium species. As reported by Rapid Alert System for Food and Feed (RASFF), Iran, China and the USA reported the highest levels of fumonisin contamination in maize or maize-based foods, ranging 50–170 mg/kg [4]. The occurrence of mycotoxins may occur at field level, in farms after harvesting and during the storage process [5]. Several studies have been focused on strategies of mycotoxin mitigation and fungal activity, including agricultural practices, bio-control strategies, optimization of storage conditions [6], and grain processing, such as roasting, baking, and nixtamalization [7].

For assurance of food safety and protection of public health, it is crucial to prioritize research, advanced analytical techniques for detection and quantification, adoption of good agricultural practices, and robust regulatory frameworks [8].

Mycotoxin analysis is typically expensive and time-consuming, meaning that it is imperative to apply advanced tools with robust developed models for detection and quantification of mycotoxins [4,9]. Near-infrared spectroscopy is a fast and non-destructive analytical technique, where the observed absorptions in the spectrum correspond to combination modes or overtones of fundamental vibrational modes [10]. This technique has been widely utilized in maize for quantification of mycotoxins, such as aflatoxins [11]. DON occurrence has been detected using UV-Vis-NIR diffuse reflection spectroscopy [9] and near-infrared hyperspectral imaging [12] in maize samples. Fumonisins B1 + B2 in maize samples have already been detected and quantified using near-infrared spectroscopy [13,14], by UV-Vis-NIR [15]. However, the development of predictive models using NIR spectroscopy for individual fumonisin B1, fumonisin B2 and DON remains limited, possibly due to the lower accuracy of predictive models when dealing with smaller quantities [13], as observed for fumonisin B2 levels. To ensure the food safety it is essential to adopt reliable and accurate methods for monitoring food quality. In this way, the objective of this preliminary study is the development of the predictive models for detection and quantification of fumonisin B1, fumonisin B2 and deoxynivalenol in maize samples using near-infrared spectroscopy.

2. Materials and Methods

2.1. Maize Samples

Sixty maize varieties were collected from three regions in the Tagus Valley, central Portugal, during 2018 and 2019. The maize grains were dried in an oven (Memmert UFB 400, Schwabach, Germany) until their moisture content reached 11–13%. The grains were then thoroughly mixed and ground using a Retsch rotor mill (SK300, Retsch GmbH, Haan, Germany) equipped with a 1.00 mm sieve. Each maize flour variety was stored in a plastic container at −20 °C until analysis.

2.2. Mycotoxin Analysis

Maize samples were extracted using 80% acetonitrile. For fumonisins, the extract was diluted with ultra-pure water, whereas for other mycotoxins, the extract was evaporated and reconstituted in 40% acetonitrile, following the method described by Silva et al. [16]. Quantification of fumonisins, aflatoxins, T-2 toxin, and zearalenone was carried out using UHPLC-TOF-MS analysis under chromatographic conditions validated by Silva et al. [16]. Deoxynivalenol (DON) quantification was performed in negative mode, as detailed by Freitas et al. [17]. LOD and LOQ were calculated as three and ten times the blank response, 62.5 and 120 µg/kg for fumonisins; and 12.6 and 42 µg/kg for DON.

2.3. Near-Infrared Spectroscopy Analysis

NIR spectra were acquired in the range of 4000–13,000 cm⁻¹, with a spectral resolution of 16 cm⁻¹ and 16 scans, using NIR transflection MPA apparatus (Bruker Optics, Ettlingen, Germany). For developing reliable quantitative models, multiplicative scatter correction (MSC), 1st and 2nd derivatives, was applied, enhancing the signal of the raw NIR spectra. Evaluation of partial least square (PLS) was established based on the root mean square (RMSE) and the correlation coefficient (R) [18]. The preprocessing of spectral data and the PLS models was performed using the toolbox developed for MATLAB software (R2023a) [19]. The samples obtained were grouped randomly into two groups: the calibration range corresponds to 65% of the total number of samples and the test range corresponds to 35% of the total number of samples [20].

2.4. Statistical Analysis

The levels of fumonisin B1, fumonisin B2 and DON in maize samples were determined in triplicate. The explanation of the UHPLC-ToF-MS results achieved for fumonisins was performed through PeakView™ 2.2 and MultiQuant™ 3.0.3 (SCIEX, Foster City, CA, USA) software. Statistical analysis of the fumonisin B1 and B2 levels was performed using SPSS Statistics 21.0 software (SPSS Inc., Chicago, IL, USA).

3. Results and Discussion

Mycotoxins

The levels of fumonisins B1, B2 and deoxynivalenol in maize samples harvested in 2020 are summarized in

Table 1. Minimum, maximum and mean levels of fumonisins detected in maize samples analyzed.

	Positive Samples (n)	Min. (µg/kg)	Max. (µg/kg)	Mean (µg/kg)	Standard Deviation
Fumonisin B1	42	<LOD	2582.1	534.5	496.6
Fumonisin B2	22	<LOD	837.9	208.2	149.8
Fumonisin B1 + B2	48	<LOD	3420.0	758.6	646.1
Deoxynivalenol	50	<LOD	484.1	130.0	71.0

.

In the sixty maize samples analyzed, only fumonisin B1, B2 and DON were detected. The mean level of fumonisin B1 was 534 µg/kg, that of fumonisin B2 was 208 µg/kg and that of DON was 130.0 µg/kg. The highest fumonisin B1 and B2 concentration reached 3420 µg/kg. However, all detected levels were below the maximum limits established by the European Union [21]. In all maize samples analyzed, fumonisin B1 levels consistently exceeded those of B2. Approximately 87% of the samples were contaminated with fumonisins B1 + B2, slightly higher than reported by Tarazona et al. [22]. In contrast, Price et al. [23] reported no detection of fumonisins in 83 fungal strains analyzed. About 30% of maize samples tested showed no fumonisin B1 or fumonisin B2; 60% of the samples were free of fumonisin B2 contamination, while DON was absent in 17% of the samples.

The spectra pre-processed using multiplicate scatter correction (MSC), first derivative and second derivative, for fumonisin B1 and fumonisin B2 are shown in Figure 1.

In the present study, three pre-processing techniques were applied to the samples: multiplicative scatter corrections (MSC), first derivative and second derivative. The main goal was to improve the precision of the spectral information, highlight key differences between samples, and minimize noise in each spectrum [10]. Spectral information above 9000 cm⁻¹ was deemed irrelevant across all three pre-processing techniques. The first- and second-derivative techniques emphasized the most critical spectral regions, particularly 4000–6500 cm⁻¹. Despite the pre-processing approaches used, no significant differences were observed between fumonisins B1 and B2, as both exhibited similar spectral profiles.

The NIR spectra of a maize sample with mycotoxin contamination (orange color) and without mycotoxin contamination (blue color) are represented in Figure 2. The main spectral differences are observed between 6400 and 7000 cm⁻¹ in the first overtone region and in 7700–6000 cm⁻¹, corresponding to O-H stretching vibrations, represented at grey circle. Additional differences are noted around 4100–4300 cm⁻¹ and in the 5500–4000 cm⁻¹ region, associated with combination bands of O-H and C-H stretching vibrations of fumonisins, represented at grey color. The most prominent peak in the spectrum at 5200 cm⁻¹ is attributed to the combination and overtones of C-H, O-H and N-H stretching vibrations of the chemical composition of maize [13]. The spectrum of the uncontaminated sample exhibits higher absorption compared to the contaminated sample. This trend aligns with observations in maize samples contaminated by aflatoxins [24] and deoxynivalenol [9]. However, an opposite tendency has been reported for maize contaminated with fumonisins [24]. These spectral differences between contaminated and uncontaminated samples are associated with fungal attacks, which disrupt tissues and cells, modifying nutritional properties and subsequently affecting their spectral signature [25].

Figure 3 displays the results of the correlation between determined and predicted fumonisin levels using the calibration models developed after pre-processing with the second derivative. PLS models developed for quantification of DON were used with MSC and second-derivative preprocessing (Figure 4).

The preliminary models for predicting fumonisin B1 and B2 occurrence in maize using near-infrared spectroscopy show promising results. The partial least squares (PLS) regression yielded correlation coefficients and RMSE of 0.91 and 0.88, and 683 and 282 for fumonisin B1 and B2, respectively. The slightly lower correlation coefficient for fumonisin B2 may be attributed to the presence of potential outlier samples. However, as a preliminary study, the data supporting the predictive models remain robust. Promising PLS models were developed for quantification of DON using MSC pre-processing with R² of 0.84, and R² of 0.99 with second-derivative pre-processing. These results are consistent with findings of Ghilardelli et al. [25], who reported an R² of calibration 0.899 for sum fumonisins, and Tyska et al. [14], who achieved an R² of 0.899 and RMSE of 659 µg/kg for sum of fumonisin B1 and B2.

4. Conclusions

The application of NIR spectroscopy offers significant advantages over traditional analytical methods, particularly its speed, non-destructive nature, and cost-effectiveness. This technology enables the simultaneous estimation of fumonisin B1, B2 and deoxynivalenol levels. In the present study, preliminary robust calibration models were developed for fumonisin B1, fumonisin B2 and deoxynivalenol, yielding promising results with second-derivative pre-processing.

In conclusion, NIR spectroscopy demonstrates its potential as a powerful tool for the rapid and non-destructive detection of fumonisins B1, B2 and deoxynivalenol in maize. This method provides an efficient approach to ensuring food safety and quality. Continued advancements in calibration techniques and chemometric analyses are anticipated to further enhance the precision and reliability of NIR spectroscopy, primarily by increasing the number of samples tested and expanding the range of fumonisin levels. This will reinforce the role of NIR spectroscopy as an essential technology in food safety monitoring.