Parallel validation of a green-solvent extraction method and quantitative estimation of multi-mycotoxins in staple cereals using LC-MS/MS

In this study, 15 different mycotoxins were estimated in three staple cereals from selected agro-ecological regions in Nigeria using a ‘novel’ green extraction method, pressurized hot water extraction (PHWE) in comparison to a conventional solvent extraction method. Discrimination of the results of PHWE and solvent extraction using principal component analysis (PCA) and orthogonal projection to latent structures discriminate analysis (OPLS-DA) did not yield any differential clustering patterns. All maize samples (n = 16), 32% (n = 38) of sorghum and 35% (n = 37) of millet samples were positive for at least one of the 15 tested mycotoxins. Contamination levels for the cereals were higher in the warm humid rain forest region and gradually decreased towards the hot and arid region in the north of the country. The results demonstrate the applicability of PHWE as a possible alternative extraction method to conventional methods of extraction, which are solvent based.

in outer dimension and 0.18 mm inner dimension), and the pressure maintained at 1000 ± 200 psi by means of a back-pressure valve (Swagelok, Johannesburg, South Africa). The extract was made to pass through a cooling coil and collected into a 50 mL centrifuge tube up to the 50 mL mark. Two (2) mL of the extracts were filtered through a 0.22 µm PVDF syringe filter into a 2 mL HPLC vial for subsequent analysis on HPLC-MS/MS. Solvent extraction. Solvent extraction was performed using the method of Majeed et al. 26 . Briefly, 5 g of samples were spiked with internal standards, DOM (1 µg/kg) and ZAN (1 µg/kg), prior to extraction. For the extraction, 20 mL of extraction solvent, acetonitrile/water/acetic acid (79/20/1, v/v/v), added to the spiked samples, agitated for 1 h on an overhead shaker (AG 6 A, Exacta, Mery sour Oise, France) and centrifuged for 15 min at 3,300 g using an IEC Central (type MP4) centrifuge (VWR International, Leuven Belgium). The supernatant was passed through a pre-conditioned (10 mL of the extraction solvent) octadecyl (C18) solid phase extraction (SPE) column (Grace octadecyl C18, Lokeren, Belgium) under gravity. A second extraction was performed on similar samples by adding 5 mL of extraction solvent, agitating, centrifuging and passing through the SPE column as described above. The total volume of the eluate was adjusted to 25 mL and defatted with 10 mL n-hexane. The defatted extract was split into two equal parts for further clean-up using different approaches. The first portion (10 mL defatted extract) was diluted with 20 mL acetonitrile/acetic acid (99/1, v/v) and was subjected to clean-up by a Multisep226, Afla-ZON + multifunctional column, under gravity, followed by further column washing using 5 mL of acetonitrile/acetic acid (99/1, v/v). The second portion (10 mL defatted extract) was filtered using a Whatman glass microfilter (VWR International, Zaventem, Belgium). All of the first extract was combined with 2 mL of the second extract, evaporated to dryness under a stream of nitrogen gas at 40 °C. The residue was reconstituted in 150 µL of mobile phase, methanol/water/acetic acid (57.20/41.80/1, v/v/v) and 5 mM ammonium acetate, and filtered through a 0.22 µm Ultrafree-MC centrifugal filter (Bedford, MA, USA) at 14,000 g for 5 min.
Liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS). Chromatographic separation and HPLC-MS/MS for PHWE. Chromatographic separation, detection and quantitation of mycotoxin levels was achieved using Shimadzu HPLC-MS/MS 8030 equipment (Shimadzu Corporation, Tokyo, Japan). The system consisted of a chromatograph (LC-30AD Nexera) linked to an autosampler (SIL-30 AC Nexera). Two µL of sample were injected and pumped through a Raptor ARC-18 column (2.7 µm, 2.1 mm × 100 mm) (Restek Corporation, Pennsylvania USA) maintained at 40 °C in a column oven (CTO-20 AC Prominence). A binary pump (LC-20AD) connected to the system was used to pump the mobile phases A (aqueous phase) and B (organic phase) through the column at a constant flow rate of 0.2 mL/min. Mobile phase A consisted of 0.1% formic acid (FA) in deionized water, and mobile phase B consisted of 0.1% FA in methanol/acetonitrile (50/50 v/v). The elution gradient program started with pumping 10% B for 0.1 min, which was steadily increased to 95% B within 8.4 min, held constant at 95% B for 3 min, initial gradient condition of 10% B re-established within 1 min, and then the column was allowed to re-equilibrate at this condition for 4.5 min prior to the next injection, making a total run time of 17 min.
Separated analytes were delivered to a Shimadzu triple-quadrupole MS model 8030 (Shimadzu Corporation, Kyoto, Japan) instrument equipped with an electron spray ionization operated in positive mode (ESI + ). A time-scheduled ultrafast multiple reaction monitoring (MRM) MS method was used for the identification and quantification of the analytes of interest. To improve specificity and confidence in analytical detection, two MRM-transitions were monitored per analyte. The desolvation line (DL) temperature was 250 °C, heat block temperature was 400 °C, drying gas flow rate was 15 L/min, and interface nebulizing gas flow rate was 3 L/min. The Shimadzu LabSolutions software was used for subsequent data visualization and analysis. The optimized chromatographic and MS method parameters of the 15 mycotoxins under investigation are presented in Table 2.
Chromatographic separation and LC-MS/MS for solvent extraction. A Waters Acquity UPLC system (Waters, Milford, MA, USA) linked to a Waters Micromass Quattro Micro triple-quadrupole mass spectrometer (Waters, Milford, MA, USA) was used for chromatographic separation, detection and quantification of the extracts obtained from the solvent extraction. The above described LC-MS/MS system was equipped with a Waters www.nature.com/scientificreports www.nature.com/scientificreports/ Symmetry C18 analytical column (5 µm, 2.1 × 150 mm) and a Waters Sentry guard column (3.5 µm 2.1 × 10 mm) purchased from the same vendor. The column oven was kept at room temperature (25 °C), and 20 µL of sample was injected into the column. The aqueous portion of the mobile phase (i.e., mobile phase A) consisted of water/ methanol/acetic acid (94/5/1, v/v/v) and 5 mM ammonium acetate, while the organic portion of the mobile phase (i.e., mobile phase B) consisted of methanol/water/acetic acid (97/2/1, v/v/v) and 5 mM, ammonium acetate. The gradient elution program and MS parameters are as described by Monbaliu et al. 27 .
Validation of the modified PHWE method for multi-mycotoxin extraction. The effects of the cereal-matrix components on the analytical signals of the different mycotoxins in the MS were determined using the signal suppression/enhancement method described by Arroyo-Manzanares et al. 28 and Sulyok et al. 29 . Calibration curves were plotted for standards prepared in analyte-free sample extracts, as well as for standards prepared in neat organic solvents (100% methanol). Matrix-effect (ME) was determined as the percentage ratio of the difference between the slope of matrix-matched calibration curve and neat standards calibration curve divided by the slope of neat standards calibration curve (Eq. 1).
where ME is the matrix effect, Slope m is the slope of calibration curve of standards prepared in sample extracts and Slope n is the slope of standards prepared in neat solvent. The limits of detection (LOD) and limits of quantification (LOQ) (Eq. 2) of the mycotoxins were determined using the signal-to-noise ratio of the matrix-matched standards as described by Kim et al. 30  www.nature.com/scientificreports/ was determined by least-square regression of a 6-point matrix-matched calibration curve within the ranges of 9 to 5,000 µg/kg depending on the mycotoxin. Recovery efficiency of the method was determined by spiking analyte-free samples with known concentrations of mycotoxins, extracting the analytes the same day using PHWE as described above. The percentage ratio of post-extraction concentration (recovered concentration) to that of pre-extraction concentration (initial concentration) was taken as the recovery value (Eq. 3) 28 .
where Limit DQ is the LOD or LOQ depending on the value of the multiplication factor F, which is 3.33 for LOD and 10 for LOQ. C is the concentration, while S is the signal at concentration C, and N is the noise level at similar concentration. E r is the recovered concentration after spiking, and E i is the spiked concentration.
Multivariate discriminant analysis. The pre-processed data set was subjected to multivariate discriminant analysis in order to scrutinize for discriminatory patterns between the two extraction methods. Using the SIMCA-P + 14.0 chemometrics software (Umetrics, MKS Instruments Inc., Sweden), the data was mean-centered, Pareto-scaled 31 and subjected to PCA and OPLSD-DA analysis in order to extract maximum information from the data set. The adopted models, PCA and OPLS-DA, are advanced dimensionality reduction tools which highlights contrasts or similarities between data groups via construction of few interpretable linearly uncorrelated variables called latent variable or principal components from the dataset 32 . PCA does not supervise the construction of latent variables from the dataset while the OPLS-DA supervises the construction of latent variables from the dataset 31,32 .
For the OPLS-DA model, the data variables were classified into two major groups depending on the adopted method of analysis, either PHWE or solvent extraction. This was critical because, OPLS-DA is a supervised model, as such, information regarding variable class member is a prerequisite for location of the principal components. Usually, OPLS-DA is best applicable when there are only two classification groups in the data set, such as a control group and a dependent group. As such, classification of the data set into two groups permitted the extraction of a between-class variation referred to as the Y-predictive block, and a within-class variation referred to as the Y-orthogonal block or uncorrelated variation 33 . By doing so, OPLS-DA maximizes the discrimination of the two groups of variables and provides an improved model interpretability without modifying its predictive power 34 .
For the evaluation of model performances, the quantitative goodness-of-fit parameters i.e., R2X(cum) and Q2X(cum) values, and the goodness-of-prediction parameters i.e., the Q2(cum) values, were calculated. The R2X(cum) and Q2X(cum) values for the PCA model were used to measure the degree to which the latent structures (i.e., principal components) describe the variations and patterns in the data set 32 . Whereas for the OPLS-DA model, the R2X(cum) was used to estimate the cumulative fraction of the variation of the X variables explained by the model, R2Y(cum) was used to measure the cumulative ratio of the variation of the Y variables explained by the model, and the Q2(cum) estimated the cumulative predictive capacity of the full model 31,35 . For internal validation of the OPLS-DA models to assess the statistical significance of the model, the goodness-of-fit and goodness-of-prediction of the OPLS-DA model was compared with those of 100 random Y-permutated models, which generates a distribution of Q2 values that are suitable for testing the null hypothesis for a model's Q2 32,36 .

Results and discussion
Pressurized hot water extraction, a 'novel' green extraction technique was adopted for the analysis of multi-mycotoxin in 91 samples of maize, sorghum and millet intended for human consumption obtained from different agroecological zones in Nigeria as described in Section 2.2.1. In order to authenticate the PHWE method, a method validation was performed, as well as, a comparison with a solvent-based extraction method.

Method validation and comparative evaluation of pHWe and solvent extraction. The results
of PHWE method validation in comparison with solvent extraction is presented in Appendix A and Table 3. In order to compensate for matrix effects, matrix-matched calibrations were adopted for quantification of the mycotoxin concentrations in the samples for the two methods.
Validation of PHWE and solvent extraction. The performances of both methods showed good consistency with EC, AOAC, and ICH guidelines 28,[37][38][39][40] . The linearity correlation (R 2 ) of PHWE ranged from 0.98 to 1.00 for the 15 mycotoxins in all 3 sample matrices within the linear ranges of 60 to 2,000 µg/kg for FB 1 , 16 to 500 µg/kg for FB 2 , 10 to 300 µg/kg for AFB 1 and 30 to 1,000 µg/kg for the other analytes (Appendix A). The benchmark for the acceptance of linearity of R 2 equal or higher than 0.95 by the International Conference on Harmonization (ICH) was fulfilled for all understudied analytes 39 , though there existed a significant difference (p ≤ 0.05) in the linearities of the two methods. The sensitivity of the methods was determined by assessing the LOD's and LOQ's of the method for each of the mycotoxins. For PHWE, the LOD's and LOQ's ranged respectively from 0.06 to 41 µg/kg and 0.32 to 123 µg/kg for maize, 0.07 to 98 µg/kg and 0.21 to 295 µg/kg for sorghum, and 0 to 26 µg/kg and 0.01 to 79 µg/kg for millet (Table 3). These values were sufficiently low for detection and quantitation of small amounts of www.nature.com/scientificreports www.nature.com/scientificreports/ the analytes under investigation in cereal grains 12,30 , as such, trace amounts of the analytes in the sample extracts can be quantitatively reported with a high degree of confidence. There was no statistically significant difference (p ≤ 0.05) between the LOD's and LOQ's of PHWE and the solvent extraction method.
The recovery rates of the 15 mycotoxins varied from 74 to 126% in maize, 73 to 115% in sorghum and 67 to 133% in millet for PHWE, which is not far from the AOAC recommendations of between 60% to 125% for foodstuff contaminated with 10 μg/kg of mycotoxins, and the EC recommendations of 60 and 130% 28,40 . The recovery rates for the solvent extraction method varied from 99 to 100% for maize, 100 to 101% in sorghum, and 100 to 102% for millet. Though the recoveries of the solvent extraction method were more consistent and closer to 100%, when compared to those of PHWE using the Independent Sample's T-test at a 95% probability, there existed no statistically significant differences in the mean recoveries of the two methods. The RSDr values for all the analytes in maize, sorghum and millet matrices ranged from 2 to 19% for PHWE and satisfied the guideline criterion of <25% by CODEX and the EC 28,38,40 . In comparison with the RSD values for the solvent extraction, those of maize ranged from 0 to 5% and 5 to 20% for sorghum. These values were significantly different (p ≤ 0.05) from the corresponding RSD values for PHWE while those of millet (3 to 15%) were not significantly different (p > 0.05).
In general, there was less variability in the validation parameters (i.e., recovery, linearity and RSD) of the solvent extraction method as compared to those of PHWE. This could be due to the fact that the solvent extraction  www.nature.com/scientificreports/ method involved multiple clean-up steps (defatting and two SPE purification procedures using different SPE cartridges). This was in addition to the use of two internal standards to correct for loss of analytes during the sample preparation or injection steps in the analytical process. On the other hand, PHWE was designed with the aim of reduction of cost and the amount of harmful organic solvents used during extraction and increased speed of the analytical process. Forfeiture of a clean-up step facilitated the achievement of these objectives. This however contributed to the higher variations observed in the PHWE analytical results, which was not unexpected. Such variations are negligible provided they were within the acceptable limits stipulated by regulatory bodies 28,38,[40][41][42][43] .
As would be discussed in the succeeding sections of this chapter, the overall variations between the two analytical procedures have been shown to be statistically insignificant.
Comparative evaluation of PHWE and solvent extraction. Principal component analysis (PCA) and orthogonal projection to latent structures discriminate analysis (OPLS-DA) approaches were adopted to scrutinize the entire data set (i.e., combined data from method validation and samples analysis) for inherent global discriminatory patterns such as multiple pairwise correlations and/or co-variances between the data obtained by the two extraction methods, which may not be readily observed by using conventional statistical analysis. Discriminatory analysis of PHWE and solvent extraction methods using PCA and OPLS-DA. The results of the PCA and OPLS-DA analysis of the data and their corresponding model-fit quality parameters are shown in Table 4 and Fig. 1. PCA was used for an initial screening of the overall structure of the data set for discriminatory patterns and detection of outliers. The results revealed a single latent variable, which indicated there were no differential patterns between the variables pertaining to PWHE and corresponding variables pertaining to the solvent extraction. Notwithstanding, the PCA model accounted for 97% of the variations in the data set [i.e., R2X(cum) = 0.97], with a predictive ability of 77% [i.e., Q2X(cum) = 0.77]. This observation was in agreement with the results of the Independent samples t-test performed on each of the variables from the two extraction methods (Appendix A), where it was observed that the majority of the variables were not significantly different (p > 0.05).
Since PCA does not take into account the classification label associated with the data set, discrimination of the data groups is not maximized. This problem is solved by the OPLS-DA model which supervises the construction of the latent variables, hence, yielding a more class-specific discrimination of the data. Results of the OPLS-DA analysis showed the construction of 2 explanatory and 1 predictive principal component with a 99% total explained variation in X [i.e., R2X(cum) = 0.99]. However, the model could not account for the cumulative variations in Y [R2Y(cum) = −0.11], and the total amount of predicted variability in the full model was low [i.e., Q2(cum) = 0.05]. As such, despite the OPLS-DA models making reference to the pre-defined sample class membership in order to maximize separation of the data, there was no clear separation of the data groups. A pattern which can be more clearly visualized on the OPLS-DA scores plot (Fig. 1). On this score plot, the green circular dots represent variables of the PHWE class, while the red triangular dots represent variables of the solvent extraction (i.e., SOLV-EXT) class. Variables from the two data groups can be seen clustered together towards the middle of the plot and while some arbitrarily distributed across the graph. This is an indication of intrinsic similarities and randomness in the patterns within and between the groups. This observation is in strong agreement to the results of PCA analysis as well as the Independent Sample's T-test. Overall, PHWE compared well with the solvent-based extraction method, and other validated methods for multi-mycotoxin extraction reported in literature [28][29][30]44 . Hence, considering its good validation performance, it was deemed adequate for investigating the natural occurrence of mycotoxins in food commodities.
Mycotoxin contamination of staple cereals. The above validated PHWE method together with the solvent extraction method was used for the screening of 15 different mycotoxin in samples of maize (n = 16), sorghum (n = 38) and millet (n = 37) obtained from six agroecological zones from Nigeria (Table 5). Only samples contaminated with mycotoxins at levels above the respective LODs were considered positive. The minor differences between the results from the two extraction methods, are possibly due to the differences in the LOD's and LOQ's, as well as the recovery rates of the two methods. For the sake of discussing the results, in the remaining sections of this paper, reference is made only to the results of the PHWE.
Mycotoxin levels and incidence rate in maize, sorghum and millet from Nigeria. The incidence rates and levels of mycotoxin contamination in maize, sorghum and millet samples are presented in Table 5 www.nature.com/scientificreports www.nature.com/scientificreports/ ranging from 17 to 7,947 µg/kg for FB 1 and 30 to 14,603 µg/kg for ΣFB 1 &FB 2 . Out of the 16 analyzed maize samples, 7 samples (i.e., 44%) contained ΣFB 1 &FB 2 at levels above the maximum levels of 4,000 µg/kg in unprocessed maize stipulated by the European Commission (EC) and CODEX 28,45,46 . Previous studies have equally reported high incidence rates and levels of FBs contamination in maize from Nigeria 20,47 . Bankole and Mabekoje 48 found that FB 1 was the predominant mycotoxin, occurring in 79% of samples of maize obtained from a similar region from Nigeria as we sampled (Southern Nigeria).
Such high incidence rates and even higher levels of FBs contamination have likewise been reported in other West African countries. Ngoko et al. 49 found FB 1 in 16 out of 18 maize samples from Cameroon at levels within the range 300 to 26,000 µg/kg. Fumonisins occurred in sorghum samples at relatively lower levels, with a mean value of 17 µg/kg and maximum value of 248 µg/kg for ΣFB 1 &FB 2 , while, no FBs contamination was recorded in millet samples. Chilaka et al. 12 reported a mean value of 83 µg/kg and a maximum value of 180 µg/kg for ΣFB 1 &FB 2 in sorghum. The International Agency for Research on Cancer (IARC) has classified FB 1 as a group 2B carcinogen (possibly carcinogenic to humans) 50 . Consumption of foods contaminated with FBs have been directly linked with upper gastro-intestinal tract cancer 51 . Moreover, FBs are also nephrotoxic, hepatotoxic, immunosuppressive, atherogenic and embryotoxic in experimental animal systems 52 .
AFs contamination was also relatively high in the cereal samples. In maize, levels for AFB 1 ranged from 6.5 to 315 µg/kg, with a mean of 54 µg/kg, while ΣAFs ranged from 16 to 323 µg/kg. Sixty-nine percent (n = 16) of the maize samples were contaminated above the maximum level of 2 and 4 µg/kg for AFB 1 and ΣAFs, respectively, stipulated by the European Commission (EC) 28,53 . These levels are similar to those reported by Bandyopadhyay et al. 18 , who reported ΣAFs contamination ranging from 1.1 to 480 µg/kg and a mean of 36 µg/kg in freshly harvested maize in Nigeria. In sorghum, 8% of samples were above the EC limits for AFB 1 (2 µg/kg) and ΣAFs (4 µg/ kg), respectively, whereas, 5% of millet samples exceeded similar limits for AFB 1 and ΣAFs. Aflatoxin B 2 , AFG 1 and AFG 2 were not detected in any of the millet samples, however, the observed levels for AFB 1 (5.5 to 50 µg/kg) were in agreement to those reported by Apeh et al. 13 (1.1 to 15 µg/kg) in millet grain from Nigeria. Observed levels for ΣAFs in sorghum (15 to 116 µg/kg) were less than those reported by Makun et al. 54 (<LOQ to 1,164 μg/kg) in stored sorghum samples. This could be due to variations in fungi colonization of crops over different years which could be stimulated by annual variations in temperature and rainfall, in addition to other climatic conditions 55,56 . Generally, AFs were more prevalent in maize, followed by sorghum and then millet. A similar trend was observed      www.nature.com/scientificreports www.nature.com/scientificreports/ by Bandyopadhyay et al. 18 , in their study on the relative severity of AFs contamination of cereal crops in West Africa. In fact, they observed that Nigerians consume 138 kg cereals annually, and if the main cereal is sorghum instead of maize, associated AF problems will be diminished 4-fold, whereas, if it is millet, then the AF-related risks will be reduced at least 8-fold 18 . Diversification of diets, instead of diets that are heavily dependent on maize could greatly reduce exposure to AFs and their consequent health-related problems. Aflatoxins are highly carcinogenic and are equally recognized as being immunosuppressive. Among the AFs group, AFB 1 is considered the most toxic, and has been identified as the most potent naturally occurring carcinogen known to man 50,57 .
Ochratoxin A was present in 31% (n = 16) of the maize samples at levels ranging from 6.5 to 54 µg/kg. All positive maize samples for OTA were above the maximum level of 5 µg/kg in unprocessed cereals stipulated by the EC 53 . In sorghum, the levels of OTs varied from 7.5 to 15 µg/kg, while those for millet ranged from 0.75 to 20 µg/ kg. Sangare-Tigori et al. 58 reported higher OTA levels (17 to 204 μg/kg) in millet from the West African country of Côte d'Ivoire sampled between 1998 to 2002, which may be due to yearly variations in mycotoxin contamination patterns across the continent 55,56,59 . Exposure to OTA has been linked with nephropathy 60 , urinary tract tumors 61 and oxidative DNA damage leading to mutagenesis and eventually cancer 62 . Based on its carcinogenicity in animal studies, OTA has also been classified as a group 2B possible human carcinogen by the International Agency for Research on Cancer (IARC) 63 .
The incidence rate and contamination range of ZEN in maize and sorghum and millet was 19% (7.4 to 33 µg/ kg) and 11% (4.8 to 22 µg/kg), respectively. These levels are negligible when compared with the EC maximum limit of 350 μg/kg for unprocessed maize and 100 µg/kg for unprocessed cereals other than maize, respectively 45 . On the other hand, 5 out of 37 millet samples were positive for ZEN within the range 80 to 94 µg/kg, none of which exceeded the EC limits of 100 µg/kg in unprocessed cereals other than maize 45 . The ZEN analogues, α-ZEL and β-ZEL, were also detected in at least one of maize and sorghum samples each, while none of the analogues was detected in millet samples Chilaka et al. (2016) and Adetunji et al. (2014). reported maize and millet contamination with α-ZEL and β-ZEL 12,20 .
Cereal contamination by ZENs could be a major health concern as this toxin is known to be chemically stable both during various food processing operations such as cooking, heating, fermentation, milling etc., and has been quantified in a number of processed cereal-based products from Africa [64][65][66] . Zearalenone has been implicated in the manifestations of gynecomastia with testicular atrophy in rural males in Southern Africa 67 . Among human populations, children are the most vulnerable to ZEN exposure and the toxin has been implicated in several incidents of precocious pubertal changes 68 and other fertility problems 69 . The potency of ZEN's estrogenic activity is reportedly greater than that of many naturally occurring non-steroidal estrogens 70 . Exposure to high concentrations of ZEN in cattle feed has been linked with enlargement of the mammary gland, infertility, reduced milk production, vaginal secretions and vaginitis particularly in young dairy heifers 71 . Whereas in swine, effects of ZEN include enlargement of the uterus, vaginal prolapse, swelling of the vulva, infertility, reduced litre size and embryonic death 72 .
The Alternaria toxin, AME, and the trichothecene toxin, T-2, were not detected in any of the maize samples. Bankole et al. 73 also reported the absence of T-2 contamination in maize from Nigeria. In sorghum (n = 38) and millet (n = 37) samples, T-2 occurred in 3% each, whereas, the average AME contamination was 0.99 and 21 µg/ kg for sorghum and millet, respectively. Sterigmatocystin occurred in 13% of both maize (n = 16) and sorghum (n = 38) samples within the ranges of 6.5 to 7.3 µg/kg and 1.9 to 330 µg/kg, respectively, whereas 9 out of 37 millet samples were positive for STEG at concentrations ranging from 5 to 208 µg/kg. Elsewhere, STEG was reported as a contaminant of Nigerian maize 20 . While the toxicity of T-2 has been established in literature 74,75 , the toxic effects of STEG and AME to humans have remained largely limited. Nonetheless, it is known that STEG is a precursor for the biosynthesis of AFB 1 and both have similar structural configurations, as such, STEG is considered as a potent mutagen, carcinogen, and teratogen 76,77 . The IARC classifies STEG as a group 2B carcinogen 77,78 .
A number of studies have demonstrated the stability of AME during extreme food processing conditions such as during wet baking of bread 66,79 , as such, there is a risk of secondary exposure to AME through processed cereal-based foods. Adekoya et al. 65 reported AME contamination of gruels derived directly from maize and sorghum in Nigeria. Some studies have demonstrated the possible carcinogenicity and mutagenicity of AME [80][81][82] . For example, NIH/3T3 cells mutated by AME caused subcutaneous tumors in mice 83 . It has also been shown to induce DNA strand breaks in cell cultures 84 . In general, mycotoxin contamination was higher in maize, followed by sorghum and then millet (Table 5). A similar trend was observed by Makun et al. 85 and Gwary et al. 86 . It has been shown that tannin-rich varieties of sorghum and millet are less susceptible to fungal colonization 13,87 , indicating that phytochemicals present in these cereals could exert antimycological properties as such resulting in less mycotoxin contamination.
Mycotoxin distribution patterns in maize, sorghum and millet from different agroecological zones of Nigeria. Geoclimatic conditions under which crops are cultivated are critical for fungi proliferation and attendant mycotoxin production 13,88,89 . Table 6 presents the distribution patterns of mycotoxins in maize, sorghum and millet across 5 agroecological zones in Nigeria. Maize samples generally had higher incidence rate and mycotoxin contamination levels. This could be because all the maize samples were obtained from the HRF/ DRS agroecological zones. Also, as stated earlier in Section 3.2.1, maize is generally more susceptible to contamination by mycotoxigenic fungal species and consequent mycotoxin production as compared to sorghum and millet 12,90 . Likewise, in millet samples contamination levels increased from Southern to Northern Nigeria, with samples from Northern having the lowest levels of contamination and incidence rates. A similar trend was observed by Chilaka et al. 12 , who reported contamination/incidence rate of multi-mycotoxin contamination in DRS > NGS > SHS.
The observed high incidence rate and contamination levels of mycotoxins in maize, sorghum and millet samples from the HRF/DRS agroecological zones could be partly due to the climatic conditions in these regions www.nature.com/scientificreports Scientific RepoRtS | (2020) 10:10334 | https://doi.org/10.1038/s41598-020-66787-z www.nature.com/scientificreports/ that favor the proliferation of mycotoxigenic fungi species and subsequent mycotoxin production 22 . As already described in Section 2.2.1, the HRS is characterized by abundant rainfall (1,500 to 2,000 mm/yr), high humidity (78 to 100%) and average temperatures between 25 to 28 °C 91,92 , while the DRS vegetation represents a transition between the humid rainforest and guinea savannah zones. Annual rainfall in this zone ranges from 1,200 mm to 1,700 mm, while average humidity and temperature are 66 to 78% and 26 to 27 °C, respectively 93 . Adejumo and Adejoro 93 , noted that the important agroecological zones in terms of mycotoxin research in Nigeria are the HRF, DRS and SGS/NGS zones.
Interestingly, this HRF zone spans across many of the West African countries, as such, similar patterns of mycotoxin contamination have been recorded in some of these countries 49,94 . Hell et al. 94 , Udoh et al. 95 and Gong et al. 96 observed significant AFs contamination of maize in lowland areas in Cameroon, Nigeria, Benin and Togo, respectively. A similar trend was reported for FBs (10 to 16,040 μg/kg) in maize from Burkina Faso 97 and FB 1 (300 to 26,000 μg/kg) in maize from Cameroon 49 . On the other hand, the lower levels of mycotoxin contamination and incidence rates observed in the Middle-Belt and Northern parts of Nigeria i.e., the SGS/NGS and SHS, respectively, could be due to dryer and more arid climatic conditions in these regions. Particularly the SHS has a much lower annual rain fall and humidity as compared to DRS and HRF. Temperatures can reach as high as 40 °C 22,98 . Such conditions may not favor the proliferation of mycotoxigenic fungi species.
Simultaneous occurrence of multiple mycotoxins in maize, sorghum and millet from Nigeria as analyzed using PHWE followed by HPLC-MS/MS. Based on the results of this study, exposure of humans and animals to multiple mycotoxins is highly likely in the selected regions in Nigeria, as we observed co-occurrence of different groups of mycotoxins in many of the analyzed samples. This observation is in line with previous literature reports 12,99,100 . Mycotoxin co-contamination of crops is a complex phenomenon, and a number of interrelating factors, such as fungi species, crop genotype, and climatic conditions may be responsible for the co-occurrence of mycotoxins in foods. It is known that a single fungal species may be able to produce more than one mycotoxin 101,102 , while one mycotoxin can be produced by different fungi species 103,104 . Table 7 and Appendix B describe all the possible co-occurrence patterns of the five groups of regulated mycotoxins detected in the samples, which are AFs, FBs, OTs, T-2 and ZENs. This was achieved using the Venn diagram web application (Bioinformatics and Evolutionary Genomics, Ghent University, Belgium) 105 . The 5 groups of mycotoxins subjected to this analysis yielded 9 different unique intersections as summarized in Table 7. The highest co-occurrence of mycotoxins that appeared in the maize samples was the co-contamination of AFs and  www.nature.com/scientificreports www.nature.com/scientificreports/ FBs (i.e. AFs + FBs) which occurred in 6 (n = 16) of the maize samples, while 3 of the samples simultaneously contained AFs, FBs and OTs. A similar pattern was observed in 5,000 samples submitted to the GEMS/Food contaminants database between 2011 and 2016, where reiterations of FBs + AFs combinations occurred in approximately 6% of the maize samples, 1% of sorghum 106 . A combination of AFs + FBs + OTs + ZENs also occurred in one (n = 16) of the maize samples, whereas, a combination of AFs + OTs + ZENs occurred in one (n = 38) of the sorghum samples. The co-occurrence of OTs + ZENs was observed in a single millet sample.
Such co-contamination patterns have previously been reported in cereals from Nigeria 12,20,48,107 . Bankole and Mabekoje 48 reported that 15 samples (n = 103) of pre-harvest maize from Southern Nigeria were contaminated with both FBs and AFs simultaneously. A prevalence rate of 10% and mean contamination level of 111 µg/kg have been reported for OTA, concurrently with OTB 7% (7.5 µg/kg), STEG 37% (3 µg/kg), ZEN 17% (174 µg/ kg), α-ZEL 1% (17 µg/kg), and β-ZEL 1% (13 µg/kg) in stored maize from five different agroecological zones in Nigeria 20 . A review of over a hundred papers between 1987 to 2016, revealed 127 mycotoxin combinations, of which AFs + FBs, AFs + OTA, DON + ZEN, and FBs + ZEN were amongst the most frequently co-occurring combinations in cereal crops 108 . In Tanzania, co-exposure to FBs + AFs has been confirmed by means of plasma or urinary biomarkers of AF 1 and FB 1 106 . Co-occurrence can be caused by substrate colonization with a single fungus that produces more than one mycotoxin, or due to colonization by different fungi species that produce different mycotoxins. It has been reported that ZEN usually co-occurs with one or more of the trichothecenes (THs), because of the ability of its producing fungi to synthesize more than one mycotoxin 109 .
Since the individual toxins that make these combinations are all amongst the most potent mycotoxins, their co-existence must not be neglected 108 . The combined effects of different mycotoxins have been extensively reviewed in literature 106,108,110 , and could manifest as additive, synergistic or antagonistic 106 . For example, exposure of F344 rats to FB 1 + AFB 1 combinations increased liver preneoplastic changes suggestive of a synergistic interaction 111 , whereas, health concerns in humans include possible childhood stunting 112 . All cytotoxic effects of the binary combinations of OTA, FB 1 and AFB 1 in low concentrations at their EU regulatory limits to MDBK cell lines were additive, and in the order OTA + FB 1 > AFB 1 + FB 1 > AFB 1 + OTA 113 , while FB 1 + α-ZEL combination significantly diminished interferon γ mRNA expression as compared to α-ZEL alone 114 .
Significance of mycotoxin contamination of Nigerian staple cereals. We reiterate that mycotoxin contamination of maize, sorghum and millet crops in Nigeria represents a major food safety concern because these crops are staples. Atanda et al. 14 asserted that AF contamination in maize and peanuts in Nigeria contributed to at least 7,761 cases of liver cancer, resulting in a total burden of 100,965 Disability-Adjusted Life Years (DALYs) 14 . Derived food products from these cereals such as masa, tuwo, ogi, ogi-baba, kunu, burukutu and others are also at risk of mycotoxin contamination 12,65,93,[115][116][117][118] . Some of these processed food products (e.g. ogi and ogi-baba) are used as weaning foods for children. Sadly, a number of studies have reported mycotoxin contamination in these weaning foods 65,119,120 . In fact, a study 120 , reported AFB 1 contamination levels up to 4,806 µg/ kg in home-made weaning food made from maize and soybeans 120 . Moreover, a posthumous autopsy of infants who suffered from kwashiorkor showed a significant level of AFs in their brains, because of consumption of contaminated maize-derived gruels 15 .
Aside the health effects on humans and animals, mycotoxin prevalence in Nigeria and Africa as a whole has other significant socio-economic impacts, ranging from food security, decreased market value of crops, regulatory rejections of goods mainly at ports of exit, damage to the African agricultural export brand amongst others 5 . For example, in 2010, the monetized burden of AFs contamination in Nigeria was estimated to be between 112 million U$D and 942 million U$D, which accounts for roughly 0.5% of the nation's Gross Domestic Product (GDP) 14 . These enormous impacts of mycotoxin could significantly jeopardize prospects of attaining the UN's sustainable development goal number 2, of achieving food security, improved nutrition, and a healthy agroeconomic growth by 2030 5 .
Based on the results obtained in this study and considering the economic importance of these cereals in Nigeria, it is thus imperative to prioritize the adoption of functional mitigation strategies that are both  www.nature.com/scientificreports/ cost-effective, crop-specific and locally adapted to the climatic conditions and agronomic practices of the region, in order to adequately combat the prevalence of these toxins in Nigeria. Further to this, more research should be done to determine the toxicological effects of different mycotoxin combinations in order to better understand their associated public health risks. If this is done, then future establishment of regulatory limits or as the case may be, reassessment of already existing limits could take into account the effects of co-existing mycotoxins. The successful adoption of PHWE in this study further represents a step closer to sustainability in green-solvent extraction in the field of mycotoxicology.