Impaginato 223 Adv. Hort. Sci., 2020 34(2): 223­232 DOI: 10.13128/ahsc­8125 Feasibility of Vis/NIR spectroscopy to detect and estimate fungicide residues on intact lettuces A.J. Steidle Neto (*), D.C. Lopes, W.A. Silva Federal University of São João del‐Rei, Campus Sete Lagoas, Sete Lagoas, Minas Gerais, Brazil. Key words: dithiocarbamate, Lactuca sativa L., multivariate data analysis, spec­ tral reflectance. Abstract: Pesticides are applied repeatedly to grain, fruit, and vegetable crops for protection against pathogens, pests, and weeds, in short periods of time before harvest. The effective and fast monitoring of chemical residues in agri­ cultural products is important for the assurance of healthy food. This study was accomplished to evaluate the feasibility to detect and estimate the concentra­ tion of dithiocarbamate fungicide (mancozeb) residues on intact lettuce leaves based on Vis/NIR spectral reflectance measurements and multivariate data analysis. In the pre­harvest interval, a high initial rate of decline of dithiocarba­ mate residues was observed between one and seven days after pesticide spray­ ing (decrease of 90.3%), while a slower decline was verified from seventh to fourteenth day (decrease of 8.7%). The usefulness of this spectrometric method has been evidenced by determination of dithiocarbamate residues at concen­ trations between 0.23 and 10.3 mg CS2 kg­1, with detection and quantitation limits of 0.49 and 1.41 mg CS2 kg­1, respectively. Vis/NIR spectral reflectance combined to the partial least square analysis have potential to be applied for estimating dithiocarbamate concentrations on intact lettuce leaves, presenting advantages such as real­time measurements and the possibility to be built into the industrial processing lines. 1. Introduction Consumers demand grain, fruits, and vegetables with high sensorial and nutritional qualities, but without pesticides. Although the correct use of fungicides does not cause problems of public concern in health and environmental areas, undesirable residues can remain on agricultural products after harvest if inappropriate or abusive treatments are applied without respecting safety recommendations indicated by the specialized agencies and manufacturers (López­Fernández et al., 2013). In modern agriculture, pesticides are applied repeatedly to grain, fruit, and vegetable crops for protection against pathogens, pests, and weeds, in short periods of time before harvest (Jankowska et al., 2019). The main exposure to pesticides for humans is via food, especially by consumption of agricultural products. (*) Corresponding author: antonio@ufsj.edu.br Citation: STEIDLE NETO A.J., LOPES D.C., SILVA W.A., 2020 ­ Feasibility of Vis/NIR spectroscopy to detect and estimate fungicide residues on intact lettuces ‐ Adv. Hort. Sci., 34(2): 223­232. Copyright: © 2020 Steidle Neto A.J., Lopes D.C., Silva W.A. This is an open access, peer reviewed article published by Firenze University Press (http://www.fupress.net/index.php/ahs/) and distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Competing Interests: The authors declare no competing interests. Received for publication 20 February 2020 Accepted for publication 13 May 2020 AHS Advances in Horticultural Science http://creativecommons.org/licenses/by/4.0/ http://creativecommons.org/licenses/by/4.0/ http://creativecommons.org/licenses/by/4.0/ Adv. Hort. Sci., 2020 34(2): 223­232 224 Introduced between 1940 and 1980, dithiocarba­ mate fungicides still represent an important class of pesticides widely used in agriculture. They are char­ acterized by a broad spectrum of activity against vari­ ous plant pathogens, low acute mammal toxicity, and low production costs (Crnogorac and Schwack, 2009). However, laboratory studies showed that dithiocar­ bamates can result in neuropathology, thyroid toxici­ ty, and developmental toxicity to the central nervous system (IPCS, 1993; Caldas et al., 2006). Mancozeb is an ethylene­bis­dithiocarbamate used as fungicide to protect fruit and vegetable crops from a range of fungal diseases (Pereira et al., 2014) and was considered a multipotent carcinogenic agent in a long­term study (Belpoggi, 2002). This fungicide is registered to 38 food crops, occupy the third place in the ranking of the most commercialized pesticides in Brazil in terms of tones of active ingredient (ANVISA, 2018; IBAMA, 2018). Caldas et al. (2006) reported data obtained from the Program on Pesticide Residue Analysis in Food (PARA), coordinated by the Brazilian National Sanitary Surveillance Agency (ANVISA), about dithio­ carbamate residues on 34% of lettuce samples in a total of 297 analyzed, based on vegetable samples collected from 2001 to 2004. In another Brazilian study, accomplished from 2005 to 2015, Jardim et al. (2018) showed that 14.6% of 1483 lettuces presented dithiocarbamate residues. López­Fernández et al. (2013) verified the presence of mancozeb and other dithiocarbamate residues on 72% of lettuce samples collected at Spain, with some samples containing residues three times higher than the maximum limit (5 mg kg­1). Various methods have been improved for the determination of dithiocarbamate residues, including gas and/or liquid chromatography, often in conjunc­ tion with mass spectrometry (Crnogorac and Schwack, 2009). These current methods requiring sample preparation (destructive), time consuming, and laboratory labor demanding, well­trained per­ sonnel and relatively expensive process chemistry. They also produce chemical and sample waste, which adversely affects product traceability by preventing real­time decision­making (Salguero­Chaparro et al., 2013; Steidle Neto et al., 2017). Although more sensi­ tive, these methods are suitable for spot checks. The reasons for Vis/NIR spectroscopy great suc­ cess as one of the most important and versatile tech­ niques in analytical chemistry include its speediness and easiness to handle and provide molecular specif­ ic information for different types of samples in any physical state, with little or no previous chemical treatment (Gonzálvez et al., 2011). However, one possible drawback is the range of concentration of the target analyte. Pesticide residues tend to have very small concentrations in foods. Despite this, pre­ vious studies proved the feasibility of spectroscopy to detect and quantify low concentrations of analytes in fruits and vegetables (Saranwong and Kawano, 2005; Gonzálvez et al. , 2011; Acharya et al. , 2012). Nevertheless, very few published studies have addressed the use of Vis/NIR spectroscopy for pre­ dicting pesticides residues in harvested intact sam­ ples. The scientific researches of Sánchez et al. (2010), for peppers, and Jamshidi et al. (2016), for cucumbers, reported estimates of pesticides without pre­treatment or preparation of samples. This advan­ tage allows that spectroscopy technique can be built into the processing line, enabling large­scale individu­ al analysis and real­time decision­making. Another recent technique for trace level detection of pesti­ cides is the Surface Enhanced Raman Spectroscopy (SERS), that use noble metal nanostructures (e.g. gold) to increase the weak signals from analytes. But, SERS requires spectrometer, laser source, probe, sample holder, and substrates that are more expen­ sive when compared with spectral reflectance equip­ ment. The present scientific research was carried out to evaluate the feasibility to detect and estimate the concentration of dithiocarbamate fungicide residues on intact lettuce leaves based on Vis/NIR spectral reflectance measurements and multivariate data analysis. 2. Materials and Methods Lettuce cultivation Lettuce (Lactuca sativa L. cv. Regina) with plain green leaves, was cultivated under organic conditions in a certified farm located at the Capim Branco city, Minas Gerais State, Brazil (19° 34’ S latitude, 44° 10’ W longitude, and 816 m a.s.l.). According to Köppen classification, the region climate is Cwa (warm tem­ perate ­ mesothermal), with dry winter and rainy summer. The lettuce seeds were planted in plastic trays containing organic substrate. Seedling production occurred under a low­density polyethylene cover (Suncover Av Blue 120 μm, Ginegar Polysack, São Paulo, Brazil), which allowed a better irrigation con­ trol and was internally coated with a photoselective Steidle Neto et al. ‐ Spectroscopy to estimate fungicide on lettuces 225 shading mesh (ChromatiNet Raschel Red 35%, Ginegar Polysack, São Paulo, Brazil), capable of reducing the percentage of beam solar radiation on the plants, also increasing the fresh mass production. The vigorous and healthy seedlings were individu­ ally transplanted to plastic pots containing a thin gravel layer overlapped by soil mixed with organic compound (cattle manure and vegetable biomass). The crop irrigation was performed by a drip system, controlled by a digital timer. The experiment consist­ ed of 500 lettuce plants. However, 355 plants were effectively used as experimental units and 145 plants were cultivated for boundary effects. Some days before plants reach the physiological maturation, the lettuces were transported to a greenhouse with the objective of performing the fungicide spraying. The greenhouse is located at the Campus Sete Lagoas of the Federal University of São João del­Rei, which is distant from the organic farm about 22 km. This greenhouse also was covered and coated with the same polyethylene film and photo selective shading mesh described above. Fungicide spraying A non­systemic fungicide (mancozeb) was used, which is classified in the alkylenebis group (dithiocar­ bamate). The active ingredient of this pesticide is a Carbon Disulphide (CS2) precursor and is registered in the Brazilian Ministry of Agriculture, Livestock, and Food Supplies (MAPA) for application on some crops. However, the dithiocarbamate (mancozeb) is not authorized by the Brazil ian National Sanitary Surveillance Agency (ANVISA) for lettuce crop. This fungicide was chosen based on reports of the Brazilian Monitoring Program for Pesticides Residues in Food, developed by ANVISA, that mentioned the indiscriminate use of this pesticide on lettuces and other crops by some Brazilian farmers. Contrarily, this fungicide is authorized for lettuce crop by other agencies and committees, such as the European Food Safety Authority and the FAO Codex Alimentarius. The solubilization of mancozeb in water was done to provide a sufficient volume for uniform and homo­ geneous application on lettuces. The dosage pre­ scribed in the Brazilian package leaflet for other green leafy vegetables, such as cabbage, was adopt­ ed (2­3 kg of fungicide ha­1). This dose is recommend­ ed to control the mildew (Peronospora parasítica) and the pod spot (Alternaria brassicae). Excepting for five plants (control units), which were randomly selected in the greenhouse, the application of man­ cozeb to lettuces was performed using an electric sprayer (droplets with average diameter of 29 µm). The application time and distance from the sprayer nozzle to the plants were standardized. Similar to the dosage, the pre­harvest interval for consumption of green leafy vegetables informed in the Brazilian package leaflet (14 days) was consid­ ered in this study. Samplings for the spectral reflectance measurements and laboratory analyzes were carried out on alternate days during the pre­ harvest interval, totaling 7 days and starting one day after the fungicide spraying. Daily 10 samples were collected, each one weight­ ing more than 500 g and corresponding to five let­ tuce plants randomly selected. The fresh mass was determined using a precision balance and discarding the roots. Therefore, 50 plants were used per day, totaling 350 plants during the sampling period. The statistical design was entirely randomized with 7 treatments (alternate days after spraying) and 10 repetitions (samples). Reflectance measurements Spectral reflectance was measured by a miniature and hand­held spectrometer (JAZ­EL350, Ocean Optics, Dunedin, USA), coupled to a tungsten­halo­ gen light source, and preconfigured to acquire and store reflectance data from 350 to 1000 nm, with spectral resolution of 1.3 nm. A specific clip probe (SpectroClip­R, Ocean Optics, Dunedin, USA) was used to collect reflected light from the lettuce leaves (Fig. 1). This probe contains an integrating sphere that captures diffuse reflected light more efficiently Fig. 1 ­ Hand­held spectrometer, clip probe, and diffuse reflec­ tance standard used to collect reflected light from the lettuce leaves. Adv. Hort. Sci., 2020 34(2): 223­232 226 than lens­based collection optics. The active illumi­ nated leaf area in the clip probe is 5 mm diameter. Two premium fibers Vis/NIR (600 μm) interconnect­ ed the spectrometer and the light source to the clip probe. A diffuse reflectance standard with Spectralon™ was used as a reference to measure spectral reflectance. After the warm up time of the light source, the reference standard measurements were made before the spectral reflectance measurements on let­ tuce leaves. Data acquisitions were performed in a temperature­controlled environment with the pur­ pose of avoiding the overheating of the light source and the spectrometer detector due to the extended using time. The reflectance values were calibrated by means of the software (OceanView, Ocean Optics, Dunedin, USA) and expressed as a relative percent­ age of the reference standard (Xing et al., 2006): Rλ cal = [(Rλ leaf ‐ Rλ dark)/(Rλ ref ‐ Rλ dark)] x 100 (1) where Rl cal is the calibrated spectral reflectance from the leaves (%), Rl leaf is the spectral reflectance from the leaves (dimensionless), Rl dark is the spectral reflectance considering light absence (dimension­ less), and Rl ref is the spectral reflectance from the dif­ fuse reflectance standard (dimensionless). Three leaves of each plant were randomly select­ ed from the external, middle, and central parts of the lettuce head. Three separate measurements on stan­ dardized and equidistant locations of the adaxial sur­ face were performed in each leaf, avoiding its central vein and boundaries. Thus, 3195 spectral signatures were obtained during the sampling period, consider­ ing the 350 sprayed plants and the five control let­ tuces (without dithiocarbamate). The electronic files containing the spectral signa­ tures were stored in a memory card and later trans­ ferred to a notebook for analyzes performed with electronic spreadsheets. During the analyzes, spec­ tral signature averages were obtained for each let­ tuce sample. After this, data was stored in an exter­ nal hard disk. Dithiocarbamate analytical determination Samples were quartered, milled in an electric grinder, placed in hermetic packages, and frozen at ­ 30°C in an ultrafreezer for minimizing the degrada­ tion and metabolization of the dithiocarbamate. The analytical determination of the concentration of dithiocarbamate was performed based on the method proposed by Cullen (1964) and improved by Keppel (1971). Mancozeb residues were measured by the spectrophotometric determination of the cupric complex formed with the CS2 evolved from the acid decomposition of dithiocarbamate in the presence of stannous chloride as a reducing agent (Caldas et al., 2004). The solution of the complex formed from the reaction between CS2 and copper (II) acetate mono­ hydrate was measured at 435 nm in UV/Vis spec­ trophotometer (Cary 50, Varian, Agilent Technologies Inc., USA). At the end of the laboratory analyzes, 70 reference measurements were obtained, corre­ sponding to 10 values for each treatment. Data analysis The Partial Least Squares (PLS) multivariate analy­ sis was applied with the purpose of developing a mathematical model capable of predicting the dithio­ carbamate concentrations based on lettuce pre­ treated spectral signatures. Thus, a response matrix, composed by the dithiocarbamate concentrations obtained by laboratory analytical measurements, was correlated with a spectral matrix, containing the average reflectance measurements. An orthogonal basis of latent variables was constructed one by one in such a way that they were oriented along the directions of maximal covariance between the two original spaces (response and spectral matrices), try­ ing to achieve an optimal prediction for new data (Wold et al., 2001; Anderson, 2009; Cozzolino et al., 2011). The latent variables were calculated by iterative methods as linear combinations of the original inde­ pendent variables (spectral reflectances) and the dependent ones (dithiocarbamate concentrations). New variables were found, representing estimates of the latent variables or their rotations. These new variables were called X­scores and were predictors of the response ones. A weight matrix was also calculat­ ed so that each of their elements maximized the covariance between response variables and the cor­ responding latent variable scores. The unexplained part of the predictor variables was represented by the deviations between the measured and predicted responses, which were also calculated and called Y­ residuals (Wold, 2001; Lopes and Steidle Neto, 2018). The detrending pre­treatment was applied to the spectral signatures prior to the model calibration and external validation. The detrending algorithm cor­ rected scatter and simple deformations of the spec­ tra baseline as vertical shift and slope (Barnes et al., 1989; Steidle Neto et al., 2016). According to Moura et al. (2016), this was the most effective pre­treat­ ment for removing irrelevant information which could not be handled by the regression technique and principal component analysis. Steidle Neto et al. ‐ Spectroscopy to estimate fungicide on lettuces 227 As recommended by Huang et al. (2008), two thirds of the data for each response variable were used as the calibration with cross­validation set, whereas one third of the data as the external valida­ tion set. The calibration set was used for developing the model and the external validation set was employed for assessing the calibrated model predic­ tion performance. This procedure was applied for predicting independent data, not related with those included in the calibration with cross­validation set (Agelet and Hurburgh, 2014). During calibration the leave­group­out cross­validation technique was applied to data. This sampling plan also agrees with the suggested by Kramer (1998), who affirmed that the number of data in the calibration set should be more than 10 times the number of variable compo­ nents in the experiment. In this study the dithiocar­ bamate residues represented the independent source of significant variation in the data. The software SPECTOX was specifically developed for this study with the purpose of assisting in the spectral reflectance data processing and the multi­ variate analysis (calibration and validation), also allowing the performance evaluation of the predic­ tion model for dithiocarbamate concentrations. The SPECTOX was written in Java language, by using the free NetBeans IDE (Apache Software Foundation, Maryland, USA). Additional algorithms of the pre­ treatments and multivariate methods were included in the SCILAB software (Scilab Enterprises, Versailles, France). The optimal number of latent variables was deter­ mined as recommended by Jha (2010), considering the minimum value for the root mean square error to avoid over fitting (Eq. 2). RMSE = ∑ (Yo ­ Yp)² (2) n Where RMSE is the root mean square error (mg kg­1), Yo are the values measured by the UV/Vis spec­ trophotometer (mg kg­1), Yp are the values predicted by the model (mg kg­1), and n is the number of sam­ ples (dimensionless). The calibration and cross­validation processes were evaluated by the root mean square errors for calibra­ tion (RMSEC) and cross­validation (RMSECV) sets, respectively. Additionally, the predictive capacity of the adjusted model regarding external validation was evaluated by the statistical parameters mean absolute error (MAE), mean bias error (BIAS), coefficient of determination (R2), and index of agreement (d). Willmott and Matsuura (2005) pointed out that MAE (Eq. 3) is unambiguous and the most natural measure of the mean error magnitude. These authors considered that MAE should be used as the basis for all dimensioned evaluations and inter­com­ parisons of model performance. The BIAS (Eq. 4) rep­ resents the average difference between measured and predicted data. Thus, values close to zero indi­ cate low systematic error between the measured and predicted values (high accuracy of the model). Also, negative BIAS values indicate underfitting, while posi­ tive BIAS values reveal overfitted predictions (Steidle Neto et al., 2016). The coefficient of determination (Eq. 5) represents the proportion of explained vari­ ance of the response variable in the validation set, with results varying from 0 to 1, and the maximum value reflecting a perfect agreement between mea­ sured and predicted data (Steidle Neto et al., 2017). Finally, index of agreement (Eq. 6) varies from 0 to 1, where the maximum value reflects a perfect agree­ ment between measured and predicted data. As affirmed by Willmott (1981), this is an important index since it is not a measure of correlation or asso­ ciation in the formal sense but rather a measure of the degree to which the model’s predictions are error free. MAE = ∑│Yp ­ Yo│ (3) n BIAS = ∑(Yo ­ Yp) (4) n R² = [∑(Yp ­ Ȳp)(Yo ­ Ȳo)]² (5) ∑(Yp ­ Ȳp)²∑(Yo ­ Ȳo)² d = 1­ ∑(Yp ­ Yo)² (6) ∑(│Yp ­ Ȳo│+│Yo ­ Ȳo│)² Where MAE is mean absolute error (mg kg­1), BIAS is the mean bias error (mg kg­1), R2 is the coefficient of determination (dimensionless), and d is index of agreement (dimensionless). The limit of detection (LOD) and limit of quantita­ tion (LOQ) are frequently used to describe the small­ est concentrations of a sample that can be reliably measured by an analytical procedure. The LOD corre­ sponds to the lowest analyte concentration at which detection is feasible, while the LOQ is the lowest con­ centration at which the analyte can be effectively quantified. Thus, LOQ tends to be equivalent or high­ er than LOD (Armbruster and Pry, 2008). In this study, the statistical LOD and LOQ determinations were applied (CLSI, 2004; Jeon et al., 2007). That is, 228 Adv. Hort. Sci., 2020 34(2): 223­232 the spectra of five representative blank lettuce sam­ ples (containing no dithiocarbamate residues) were measured, following the same procedures used dur­ ing the calibration and external validation of the PLS model. The mean and standard deviation of the pre­ dicted CS2 concentrations were calculated based on these spectra, which were used as input for the developed model. The LOD and LOQ were equal to the mean of the predicted CS2 concentrations plus three times or ten times the standard deviation of the mean, respectively. According to CLSI (2004) and Armbruster and Pry (2008), although the blank sam­ ples are devoid of analyte, they can produce an ana­ lytical signal that might be consistent with a low con­ centration of dithiocarbamate. 3. Results Figure 2 presents the degradation curve of dithio­ carbamate residues on lettuce leaves during the pre­ harvest interval, obtained from average values of lab­ oratory analytical measurements. The estimated half­ life of the dithiocarbamate residues was 5­7 days. The rates of decline of dithiocarbamate concentration in two­day intervals were quite variable from first to seventh day after spraying (4.7, 3.2, and 1.4 mg CS2 kg­1), corresponding to a decrease of 90.3%. After the seventh day, the rates of decline were 0.3 mg CS2 kg­1, indicating a decrease of 8.7%. Although very small concentration residues persisted at the final of pre­ harvest interval, the results indicate that after this period the pesticide metabolization is advanced, assuring reliability for lettuce consumption. The PLS model for dithiocarbamate concentration on lettuces was more precise and accurate when detrending pre­treatment was applied to spectral sig­ natures, compared with predictions obtained from spectra without pre­treatment or treated with other methods (centering, standardization, first and second derivatives). Despite NIR bands were more sensible, all wavelengths (350­1000 nm) presented potential to explain the dithiocarbamate concentration residues on lettuce leaves from spectral reflectance, contributing to the good performance of the predic­ tion PLS model. Table 1 shows the statistical results for the dithio­ carbamate concentration model, considering the cali­ bration with cross­validation and external validation datasets. The data processing showed that the use of more than four latent variables for dithiocarbamate concentrations resulted in an over­fitting, character­ ized by a slight divergence in the downward trend of the RMSE, which continued to decrease for calibra­ tion (RMSEC), but almost established for cross­valida­ tion (RMSECV). The proposed PLS model was satisfactory since presented low RMSEC, RMSECV, RMSE, MAE, and BIAS when compared to the range values (Table 1). Additionally, the performance of the model for the external validation presented high index of agree­ ment (0.94), reflecting a good accuracy for indepen­ dent predictions of the dithiocarbamate concentra­ Fig. 2 ­ Dithiocarbamate concentration decay (mg CS2 kg­1) on lettuce leaves at seven intervals after mancozeb spraying. Vertical bars represent the standard error of the average values. Table 1 ­ Statistical parameters of calibration with cross­validation and external validation processes for the estimation model of dithio­ carbamate concentration on lettuces Calibration External validation Number of latent variables 4 R2 (dimensionless) 0.87 RMSEC (mg CS2 kg­1) 1.86 RMSE (mg CS2 kg­1) 1.41 RMSECV (mg CS2 kg­1) 2.74 MAE (mg CS2 kg­1) 1.24 Range (mg CS2 kg­1) 0.23­10.3 BIAS (mg CS2 kg­1) ­0.37 d (dimensionless) 0.94 Steidle Neto et al. ‐ Spectroscopy to estimate fungicide on lettuces 229 tions on lettuces. The negative BIAS value indicated a majority tendency of underfitting predictions by the model, mainly from 2.3 mg CS2 kg­1 (Fig. 3). Also, low BIAS value represented small systematic error (Table 1). Regarding the differences among statistics pre­ sented in Table 1, the spectral measurements and predictions of external validation may deviate from the calibration as they originate from different sam­ ple sets. As mentioned by Liu and Ying (2005), in this way, the ability of the calibration model to withstand unknown variability is assessed. The correlations between values determined by the reference analytical method and those predicted by the external validation of the PLS model are pre­ sented in figure 3. Prediction performance resulted in good agreement between reference and estimated values, with R2 of 0.87 (Table 1), indicating that 87% of the measured values were accurately represented by the model. The comparison of parameters (LOD, LOQ, and range) associated to different methods of determina­ tion of dithiocarbamates on lettuces are showed in Table 2. The LOD and LOQ values verified in this study were higher than those obtained with the gas or liquid chromatography, as well as the spectropho­ tometric method. The usefulness of the methodology presented in this study has been evidenced by the determination of dithiocarbamate residues on lettuce samples at concentrations between 0.23 and 10.3 mg CS2 kg­1. This range can be considered appropriated when the results of previous studies are used as reference (Table 2). Further, values between these lower and upper limits are sufficient to measure a wide range of dithiocarbamate concentrations on lettuces. 4. Discussion and Conclusions Past studies showed dithiocarbamate degradation profiles for lettuces similar to those found in this study. Yip et al. (1971) found that the dithiocarba­ mate concentration (maneb) from a single spray on lettuces declined from 45 mg kg­1 initially to about 5 mg kg­1 after 15 days (decrease of 89%). On the other hand, Hughes and Tate (1982) monitored mancozeb residues on lettuces and verified a reduction of 90 mg kg­1 after a 14­day interval (decrease of 72%). These authors also reported a high initial rate of decline in dithiocarbamate concentration during the first seven days after spraying, confirming the high degradation of analyte. Despite of the dithiocarbamates with active ingre­ dient based on mancozeb are not authorized by ANVISA for lettuces in Brazil, the European Food Safety Authority (EFSA) allows the use of this pesti­ cide on lettuce crop in the countries that integrate the European Union, considering a maximum residue limit of 5 mg kg­1 (EFSA, 2013). In addition to the dietary risk, the edafoclimatic differences and dietary patterns of the countries justify the distinct positions between ANVISA and EFSA. Based on the results, let­ tuces presented dithiocarbamate concentrations Fig. 3 ­ External validation of the PLS model for estimating dithiocarbamate concentrations (mg CS2 kg­1) on lettuce leaves. Table 2 ­ Limit of detection, limit of quantitation, and range (expressed in mg CS2 kg­1) for different methods of determination of dithio­ carbamates on lettuces Method Limit of detection Limit of quantitation Range Reference Gas chromatography 0.004 0.013 0.04­5.0 Česnik and Gregorčič (2006) Liquid chromatography 0.04 0.11 0.50­9.3 López­Fernández et al. (2012) Gas chromatography 0.02 0.05 0.04­1.0 Pizzutti et al. (2017) Spectrophotometric 0.28 0.40 0.20­4.5 Pizzutti et al. (2017) Spectrometric 0.49 1.41 0.23­10.3 Present study Adv. Hort. Sci., 2020 34(2): 223­232 230 mancozeb has 2 NH and 2 CH2, and that the NH bond position is unhindered within the chemical structure, there is a high likelihood it will produce a sharp and strong absorption band, which should improve detec­ tion of mancozeb. The mean spectral signatures of intact lettuce leaves with absence and presence of fungicide pre­ sented differences mainly in NIR region, with fungi­ cide­contaminated samples resulting in greater absorbance than that in the fungicide­free lettuce leaves. According to that reported by Sánchez et al. (2010) and Jamshidi et al. (2016), increase of the absorbance in NIR region after 900 nm could be due to the C­H absorption. Based on the results, it can be said that the Vis/NIR spectroscopy combined to the multivariate data analysis have potential to be applied as an alter­ native method to estimate dithiocarbamate concen­ trations on lettuce leaves. However, the success and widespread adoption of this method also depends of suitable measurement practices. It is important that measurements are performed using a spectrometer with high spectral resolution, after the time required to warm­up the light source, and after adequate spectrometer calibration (proper sampling of refer­ ence standard). Additionally, important factors to achieve good results include the standardization of the measurement points in the samples, the homo­ geneity of the target area, and the positioning of samples on a black non­reflective panel with the pur­ pose of prevent the light reflection going through the leaf. Another essential aspect is related to the inci­ dence angle of the light bunch, which is emitted by the light source over the sample and directly influ­ ences the light reflection by sample (Steidle Neto et al., 2017). Although the effects of the abovemen­ tioned individual error sources may appear small, their combined presence may result in measure­ ments with significant errors, influencing the spectral reflectance independently of dithiocarbamate con­ centrations, as well as, the model predictions. Vis/NIR spectral reflectance measurements com­ bined to the partial least square analysis showed to be feasible and effective as a promising method for estimating concentration of dithiocarbamate fungi­ cide residues on intact lettuce leaves. The developed PLS model allowed the detection and quantification of the dithiocarbamate without any preparation and/or processing of lettuce samples. This method offers advantages over traditional laboratory meth­ ods, such as real­time measurements and the possi­ bility to be built into the industrial processing lines, lower than the maximum residue limit established by EFSA from the fourth day after spraying. According to Lopes and Steidle Neto (2018) detrending, which was the most appropriated spec­ tral pre­treatment in this study, is usually used to remove specific data offsets that are not related to the chemical or physical properties of interest for the chemometric modeling. Sánchez et al. (2010) also developed PLS models for predicting pesticide residues on intact peppers using near­infrared reflectance spectroscopy, applying detrending method for scatter correction in data, and obtaining good results. Steidle Neto et al. (2016) affirmed that detrending was the best spectra pre­treatment when predicting chlorophyll content in lettuces, helping to remove non­linear trends in spectroscopic data, and consequently correcting scatter. The number of latent variables considered ade­ quate for predicting dithiocarbamate concentrations on lettuce leaves in this study agree with Cozzolino et al. (2011), who affirmed that if more than optimum number of latent variables is used, the solution can become over­fitted and the model will be very dependent on the dataset, giving poor predictions. Otherwise, using less than the optimum number of latent variables will cause under­fitting and the model will not be accurate enough to capture the variability in the data. This result also agrees with other researchers who applied spectroscopy and PLS models to non­destructively predict of pesticide con­ centrations. Jamshidi et al. (2016) showed that 5­7 latent variables were required for PLS models when predicting diazinon residues on cucumbers. Although the proposed model tended to underes­ timate the pesticide concentrations, slight overesti­ mates were verified for low dithiocarbamate concen­ trations, evidenced in this study until the fifth day after spraying. The method based on spectral reflectance provid­ ed sufficient sensibility for detecting and quantifying concentrations lower than the maximum residue limit allowed by the European Food Safety Authority for mancozeb­based dithiocarbamates on lettuces (5 mg kg­1). Makino et al. (2009) used spectral reflectance for detection and quantification of chlor­ pyrifos residues on apple surface, also demonstrating the feasibility of spectroscopy for estimating low con­ centrations of pesticide residues in fruits. According to Acharya et al. (2012), the spectral features can be assigned to overtone and combina­ tion bands of various C­H and N­H bonds within these molecules. Considering that the chemical structure of Steidle Neto et al. ‐ Spectroscopy to estimate fungicide on lettuces 231 CALDAS E.D., TRESSOU J., BOON P.E., 2006 ­ Dietary expo‐ sure of Brazilian consumers to dithiocarbamate pesti‐ cides ‐ A probabilistic approach. ­ Food. Chem. Tox., 44: 1562­1571. ČESNIK H.B., GREGORČIČ A., 2006 ­ Validation of the method for the determination of dithiocarbamates and thiuram disulphide on apple, lettuce, potato, strawber‐ ry and tomato matrix. ­ Acta Chim. Slov., 53: 100­104. CLSI, 2004 ­ Protocols for determination of limits of detec‐ tion and limits of quantitation (guideline). Document EP17. ­ Clinical and Laboratory Standards Institute (CLSI), Wayne, USA. COZZOLINO D., CYNKAR W.U., SHAH N., SMITH P., 2011 ­ Multivariate data analysis applied to spectroscopy: potential application to juice and fruit quality. ­ Food Res. Int., 44: 1888­1896. CRNOGORAC G., SCHWACK W., 2009 ­ Residue analysis of dithiocarbamate fungicides. ­ Trends Anal. Chem., 28: 40­50. CULLEN T.E., 1964 ­ Spectrophotometric determination of dithiocarbamates residues on food crops. ­ Anal. Chem., 36: 221­224. EFSA, 2013 ­ The 2010 European Union report on pesticide residues in food. ­ European Food Safety Authority, Parma, Italy, pp. 511. GONZÁLVEZ A., GARRIGUES S., ARMENTA S., DE LA GUARDIA M., 2011 ­ Determination at low ppm levels of dithiocarbamate residues in foodstuff by vapour phase‐liquid phase microextraction‐infrared spec‐ troscopy. ­ Anal. Chim. Acta, 688: 191­196. HUANG H., YU H., XU H., YING Y., 2008 ­ Near infrared spectroscopy for on/in‐line monitoring of quality in foods and beverages: a review. ­ J. Food Eng., 87: 303­ 313. HUGHES J.T., TATE K.G., 1982 ­ Dithiocarbamate spray residues on lettuces, tomatoes, berry fruits, and apples. ­ New Zeal. J. Exp. Agr., 10: 301­304. IBAMA, 2018 ­ Relatórios de comercialização de agrotóxi‐ cos. ­ Inst. Brasileiro do Meio Ambiente e dos Recursos Naturais Renováveis, Brasília, Distrito Federal, Brasil. IPCS, 1993 ­ Mancozeb pesticide residues in food: 1993 evaluations Part II Toxicology . ­ International Programme on Chemical Safety (IPCS), US Environmental Protection Agency, Washington, USA. JAMSHIDI B., MOHAJERANI E., JAMSHIDI J., 2016 ­ Developing a Vis/NIR spectroscopic system for fast and non‐destructive pesticide residue monitoring in agricul‐ tural product. ­ Measurement, 89: 1­6. JANKOWSKA M., LOZOWICKA B., KACZYŃSKI P., 2019 ­ Comprehensive toxicological study over 160 processing factors of pesticides in selected fruit and vegetables after water, mechanical and thermal processing treat‐ ments and their application to human health risk assessment. ­ Sci. Tot. Environ., 652: 1156­1167. JARDIM A.N.O., MELLO D.C., BRITO A.P., VAN DER VOET H., BOON P.E., CALDAS E.D., 2018 ­ Probabilistic dietary enabling large­scale individual analysis in 100% of the lettuces. The spectral behavior of other vegetable species certainly will differ from that of lettuce, as well as, different dithiocarbamate types (thiram, metiram, propineb, zineb, ziram, and maneb) tend to cause variations in the estimating models. Future research­ es will be performed in order of evaluating these effects when detecting and quantifying fungicide residues in vegetable crops. The results of these new studies will complement the findings of the present work, making this a more wide­ranging method. Acknowledgements The authors are grateful to the Foundation for Research Support of the Minas Gerais State (FAPEMIG) in Brazil, which provided funding to acquire the accessories for spectrometer and materi­ als for the experimental setup (grant number CAG­ APQ­01495­15). References ACHARYA U.K., SUBEDI P.P., WALSH K.B., 2012 ­ Evaluation of a dry extract system involving NIR spectroscopy (DESIR) for rapid assessment of pesticide contamina‐ tion of fruit surfaces. ­ Am. J. Anal. Chem., 3: 524­533. AGELET L.E., HURBURGH C.R. Jr., 2014 ­ Limitations and current applications of near infrared spectroscopy for single seed analysis. ­ Talanta, 121: 288­299. ANDERSON M., 2009 ­ A comparison of nine PLS1 algo‐ rithms. ­ J. Chemometrics, 23: 518­529. ANVISA, 2018 ­ Monografias de agrotóxicos. ­ Agência Nacional de Vigilância Sanitária (ANVISA), Brasília, Distrito Federal, Brasil. ARMBRUSTER D.A., PRY T., 2008 ­ Limit of blank, limit of detection and limit of quantitation. ­ Clin. Biochem. Rev., 29: 49­52. BARNES R.J., DHANOA M.S., LISTER S.J., 1989 ­ Standard normal variate transformation and de‐trending of near‐ infrared diffuse reflectance spectra. ­ Appl. Spectrosc., 43: 772­777. BELPOGGI F., SOFFRITTI M., GUARINO M., LAMBERTINI L., CEVOLANI D., MALTONI C., 2002 ­ Results of long‐term experimental studies on the carcinogenicity of ethyl‐ ene‐bis‐dithiocarbamate (mancozeb) in rats. ­ Annals of New York Academy of Sciences, 982: 123­136. CALDAS E.D., MIRANDA M.C.C., CONCEIÇÃO M.H., SOUZA L.C.K.R., 2004 ­ Dithiocarbamates residues in Brazilian food and the potential risk for consumers. ­ Food. Chem. Tox., 42: 1877­1883. Adv. Hort. Sci., 2020 34(2): 223­232 232 risk assessment of triazole and dithiocarbamate fungi‐ cides for the Brazilian population. ­ Food Chem. Tox., 118: 317­327. JEON H.R., EL­ATY M.A., CHO S.K., CHOI J.H., KIM K.Y., PARK R.D., SHIM J.H., 2007 ­ Multiresidue analysis of four pesticide residues in water dropwort (Oenanthe javanica) via pressurized liquid extraction, supercritical fluid extraction, and liquid‐liquid extraction and gas chromatographic determination. ­ J. Sep. Sci., 30: 1953­ 1963. JHA S.N., 2010 ­ Nondestructive evaluation of food quality: theory and practice. ­ Springer Science & Business Media, New York, USA. KEPPEL G.E., 1971 ­ Collaborative study of the determina‐ tion of dithiocarbamate residues by a modified carbon disulfide evolution method. ­ J. AOAC, 54: 528­532. KRAMER R., 1998 ­ Chemometric techniques for quantita‐ tive analysis. ­ CRC Press, New York, USA. LIU Y., YING Y., 2005 ­ Use of FT‐NIR spectrometry in non‐ invasive measurements of internal quality of “Fuji” apples. ­ Post. Biol. Tech., 37: 65­71. LOPES D.C., STEIDLE NETO A.J., 2018 ­ Classification and authentication of plants by chemometric analysis of spectral data. ­ In: BARCELÓ D., J. LOPES, and C. SOUSA (eds.) Comprehensive analytical chemistry, vibrational spectroscopy for plant varieties and cultivars character‐ ization. Elsevier, Amsterdam, Netherlands, pp. 316. LÓPEZ­FERNÁNDEZ O., RIAL­OTERO R., GONZÁLEZ­ BARREIRO C., SIMAL­GÁNDARA J., 2012 ­ Surveillance of fungicidal dithiocarbamate residues in fruits and vegetables. ­ Food Chem., 134: 366­374. LÓPEZ­FERNÁNDEZ O., RIAL­OTERO R., SIMAL­GÁNDARA J., 2013 ­ Factors governing the removal of mancozeb residues from lettuces with washing solutions. ­ Food Contr., 34: 530­538. MAKINO Y., OSHITA S., MURAYAMA Y., MORI M., KAWAGOE Y., SAKAI K., 2009 ­ Nondestructive analysis of chlorpyrifos on apple skin using UV reflectance. ­ Trans. ASABE, 52: 1955­1960. MOURA L.O., LOPES D.C., STEIDLE NETO A.J., FERRAZ L.C.L., CARLOS L.A., MARTINS L.M., 2016 ­ Evaluation of tech‐ niques for automatic classification of lettuce based on spectral reflectance. ­ Food Anal. Meth., 9: 1799­1806. PEREIRA S.I., FIGUEIREDO P.I., BARROS A.S., DIAS M.C., SANTOS C., DUARTE I.F., GIL A.M., 2014 ­ Changes in the metabolome of lettuce leaves due to exposure to mancozeb pesticide. ­ Food Chem., 154: 291­298. PIZZUTTI I.R., KOK A., SILVA R.C., ROHERS G.N., 2017 ­ Comparison between three chromatographic (GC‐ECD, GC‐PFPD and GC‐ITD‐MS) methods and a UV‐Vis spec‐ trophotometric method for the determination of dithio‐ carbamates in lettuce. ­ J. Braz. Chem. Soc., 28: 775­ 781. SALGUERO­CHAPARRO L., GAITÁN­JURADO A.J., ORTIZ­ SOMOVILLA V., PEÑA­RODRÍGUEZ F., 2013 ­ Feasibility of using NIR spectroscopy to detect herbicide residues in intact olives. ­ Food Contr., 30: 504­509. SÁNCHEZ M.­T., FLORES­ROJAS K., GUERRERO J.E., GUARRIDO­VARO A., PÉREZ­MARÍN D., 2010 ­ Measurement of pesticide residues in peppers by near‐ infrared reflectance spectroscopy. ­ Pest Manage. Sci., 66: 580­586. SARANWONG S., KAWANO S., 2005 ­ Rapid determination of fungicide contaminated on tomato surfaces using the DESIR‐NIR: a system for ppm‐order concentration. ­ J. Near Inf. Spect., 13: 169­175. STEIDLE NETO A.J., LOPES D.C., PINTO F.A.C., ZOLNIER S., 2017 ­ Vis/NIR spectroscopy and chemometrics for non‐ destructive estimation of water and chlorophyll status in sunflower leaves. ­ Biosyst. Eng., 155: 124­133. STEIDLE NETO A.J., MOURA L.O., LOPES D.C., CARLOS L.A., MARTINS L.M., FERRAZ L.C.L., 2016 ­ Non‐destructive prediction of pigment content in lettuce based on visi‐ ble‐NIR spectroscopy. ­ J. Sci. Food Agr., 97: 2015­2022. WILLMOTT C.J., 1981 ­ On the validation of models. ­ Phys. Geog., 2: 184­194. WILLMOTT C.J., MATSUURA K., 2005 ­ Advantages of the mean absolute error (MAE) over the root mean square error (RMSE) in assessing average model performance. ­ Clim. Res., 30: 79­82. WOLD S., SJÖSTROM M., ERIKSSON L., 2001 ­ PLS‐regres‐ sion: a basic tool of chemometrics. ­ Chem. Intell. Lab. Sys., 58: 109­130. XING J., BRAVO C., MOSHOU D., RAMON H., DE BAERDEMAEKER J., 2006 ­ Bruise detection on ‘Golden Delicious’ apples by vis/NIR spectroscopy. ­ Comput. Electron. Agr., 52: 11­20. YIP G., ONLEY J.H., HOWARD S.F., 1971 ­ Residues of maneb and ethylenethiourea of field‐sprayed lettuce and kale. ­ J. AOAC, 54: 1373­1375.