Moisture Influence Reducing Method for Heavy Metals Detection in

Jun 17, 2017 - Because it was hard to obtain the matrix-matched certified reference material (CRM), we calibrated the preprocessed LIBS signal to the ...
5 downloads 12 Views 1MB Size
Download PDF
Article pubs.acs.org/ac

Moisture Influence Reducing Method for Heavy Metals Detection in Plant Materials Using Laser-Induced Breakdown Spectroscopy: A Case Study for Chromium Content Detection in Rice Leaves Jiyu Peng,†,∥ Yong He,†,∥ Lanhan Ye,† Tingting Shen,† Fei Liu,*,† Wenwen Kong,‡ Xiaodan Liu,† and Yun Zhao§ †

College of Biosystems Engineering and Food Science, Zhejiang University, Hangzhou 310058, China School of Information Engineering, Zhejiang A & F University, Linan 311300, China § School of Information and Electronic Engineering, Zhejiang University of Science and Technology, Hangzhou 310023, China ‡

S Supporting Information *

ABSTRACT: Fast detection of heavy metals in plant materials is crucial for environmental remediation and ensuring food safety. However, most plant materials contain high moisture content, the influence of which cannot be simply ignored. Hence, we proposed moisture influence reducing method for fast detection of heavy metals using laser-induced breakdown spectroscopy (LIBS). First, we investigated the effect of moisture content on signal intensity, stability, and plasma parameters (temperature and electron density) and determined the main influential factors (experimental parameters F and the change of analyte concentration) on the variations of signal. For chromium content detection, the rice leaves were performed with a quick drying procedure, and two strategies were further used to reduce the effect of moisture content and shot-to-shot fluctuation. An exponential model based on the intensity of background was used to correct the actual element concentration in analyte. Also, the ratio of signal-to-background for univariable calibration and partial least squared regression (PLSR) for multivariable calibration were used to compensate the prediction deviations. The PLSR calibration model obtained the best result, with the correlation coefficient of 0.9669 and root-mean-square error of 4.75 mg/kg in the prediction set. The preliminary results indicated that the proposed method allowed for the detection of heavy metals in plant materials using LIBS, and it could be possibly used for element mapping in future work.

H

a novel atomic emission spectroscopy, which has advantages of fast analytical speed, no or little sample preparation, and in situ or stand-off detection capability.5 By analyzing the spectral signal emitted from laser plasma, LIBS have been used to characterize sample features and quantify elemental contents in various kinds of samples (gases, liquids, and solids).6−8 However, the detection capability of LIBS is limited by the moisture content in the plant materials. It has been reported that moisture content in samples might reduce the emission intensities and worsen the signal stability.9−11 Two strategies have been used to deal with it. One is to freeze the samples with a freezer,12 and the other one is to reduce the moisture content with the drying process.13,14 The frozen samples are likely to thaw when analyzing, and it is not suitable for practical application. For dried samples, drying, grinding, and pressing are usually preferred, and good repeatability and accuracy have been achieved for the detection of nutrient elements in plants

eavy metals contamination has become a globe problem due to the deposition and accumulation in soils and water resources.1 People might be exposed to heavy metals from polluted air, food, and water. As a consequence, chronic exposure to heavy metals can cause severe healthy problems to humans, plants, and animals if critical levels are exceeded.2,3 Some strategies such as conventional physicochemical or more recently biological treatments have been developed to decrease the concentration of heavy metals.4 However, one of the important points in remediation is to real-time monitor the concentration of heavy metals in contaminated creatures for indication and biosorption. What’s more, there is also a need for developing robust, rapid methods to determine the heavy metals content and to ensure the food safety. Currently, atomic absorption spectrometry (AAS), inductively coupled plasma optical emission spectroscopy (ICPOES) and inductively coupled plasma with mass spectrometry (ICPMS) are the conventional methods to detect heavy metals. However, these methods are limited by complex sample preparations and cannot meet the demands of real-time monitoring. Laser-induced breakdown spectroscopy (LIBS) is © 2017 American Chemical Society

Received: April 18, 2017 Accepted: June 17, 2017 Published: June 17, 2017 7593

DOI: 10.1021/acs.analchem.7b01441 Anal. Chem. 2017, 89, 7593−7600

Article

Analytical Chemistry materials.15,16 However, the complex sample preparation might be a limitation for further application, and it is not available when in situ measurement or element mapping is required. In this study, we aimed to propose the moisture influence reducing method to quickly detect heavy metals in plant materials using LIBS. In order to clarify the role that moisture content played in LIBS analysis, we first investigated the influences of moisture content on signal intensity, stability, and plasma parameters (temperature and electron density) using the rice leaves with various moisture contents and tried to find the main influential factors on the variations of signal. We reduced the moisture content using a quick drying procedure, and two strategies were used to compensate the effect of moisture content and shot-to-shot fluctuation. An exponential equation based on the intensity of background was developed to correct the reference value of element concentration in preprocessed samples. Then the ratio of signal-to-background for univariable analysis and partial least-squares regression (PLSR) for multivariable analysis were used to compensate the deviations caused by moisture content and shot-to-shot fluctuation.

until reaching constant weights, and the weights were recorded as W2. Finally, the moisture content ε of plant materials were calculated with the following equation. ε=

W1 − W2 × 100% W1

(1)

The moisture content of first sampling (dried at 60 °C in an oven for 0, 10, 30, 60, 120 min) was measured in the range of 1.1%−79.2%, and it was 4.8%−15.7% for the second sampling (dried at 60 °C in an oven for 20 min). Reference Method to Determine Chromium Content. Chromium content in rice leaves were determined using ICPMS after microwave digestion. In total, 100 mg of dried leaves was weighed in the TFM vessels, and 4 mL of 65% HNO3 and 1 mL of 30% H2O2 were added. After digestion, it was transferred to 25 mL volumetric flasks and diluted with deionized distilled water to the mark. Finally, the solution was analyzed by ICPMS (ELAN DRC-e, PerkinElmer). LIBS Data Acquisition. The LIBS analysis was carried out with a self-assembled LIBS device, which mainly consisted of a Q-switched pulse laser for ablating samples, spectrograph for dispersing light, and intensified CCD camera for recording signal. The detailed description of the device is given elsewhere.18,19 LIBS data acquisition was performed using single shot scanning, i.e., one shot in each position of leaf surface. In this case, the laser was operated at the second-harmonics wavelength with the energy of 60 mJ and the repetition rate of 1 Hz. With the help of plano-convex lens ( f = 100 mm), the laser beam was focused 2 mm below the sample surface to produce a relatively stable plasma. In addition, the laser beam ablated one side of the leaves about 2 mm away from midribs, and a total of 25 successive shots at different positions were performed for each leaf. In order to obtain a good signal-to-noise ratio and reduce the effect of background, the delay time and integration time of intensified CCD camera were set at 4 and 20 μs with gain factor of 2000. Before the experiment, the wavelength of spectrograph and the spectral intensity of camera were calibrated by a mercury argon lamp (HG-1, Ocean Optics) and a deuterium tungsten halogen source (DH-2000-BALCAL, Ocean Optics), respectively. Theory. On the basis of the assumption that the existence of local thermal equilibrium (LTE) at the plasma and optically thin at the observed spectral line, the spectral line radiant intensity can be described by20

■

EXPERIMENTAL SECTION Sample Preparation. Plant materials rice (Oryza sativa L.) leaves with different degrees of chromium stress were used in this study. Rice seeds with genotype Chunyou 84 were obtained from China National Rice Research Institute (Hangzhou, Zhejiang 310058, China). Rice seeds were first sterilized with 1% NaClO for 30 min, rinsed with tap water, soaked in the dark at 25 °C for 2 days, and then geminated at 35 °C for 1 day. The well geminated seeds were placed in a plate that was floating in a container filled with half-strength nutrient solution. The nutrient solution was prepared according to Yoshida et al.17 and renewed every 4 days. After 1 week, rice seedlings were transplanted into 10 L containers with 11 holes, and two seedlings were placed in each hole with complete nutrient solution. The seedlings were growing in a controlled growth chamber with a photoperiod of 12 h light/12 h dark and a light intensity of 225 μmol m−2 s−1. Temperatures of light and dark were set at 30/22 °C, and the humidity was maintained at 65%. After growing in complete nutrient solution for 7 days, five treatments were adopted in this experiment, i.e., control group and experiment groups of 25, 50, 75, 100 μM chromium stress (prepared by K2Cr2O7 solution). Two independent samplings were conducted after the chromium stress for 4 and 8 weeks. The first sampling (n = 66) could be considered as a proof-of-concept experiment to explore the effect of moisture content on LIBS signal, while the second one (n = 42) was an independent confirmatory study that was used for the verification of previous conclusion and detection of chromium content in rice leaves. In order to explore the effect of moisture content, the fresh leaves were dried at 60 °C in an oven for 0, 10, 30, 60, 120 min, respectively. We carried this procedure to enlarge the range of moisture content. For the second sampling, all the fresh leaves were dried in an oven for 20 min. In this case, the samples might maintain in low moisture content, and they were not easily cracked due to the laser ablation. Determination of Moisture Content. The moisture content of rice leaves was determined with gravimetric method according to Chinese national standard GB5009.3-2010. Before LIBS measurement, the weights of samples were recorded immediately as W1. Then all samples were dried in an oven

Iki = CsFgk Aki

e(−Ek / KBT ) Zs(T )

(2)

where Iki is the spectral line radiant intensity corresponding to the transition between the upper level k and lower level i; Cs is the concentration of emitting species; Aki is the emission transition probability from upper level k to lower level i; gk is the statistical weight of the upper level k; Ek is the upper level energy of the transition; KB is the Boltzmann constant; T is the plasma temperature; Zs(T) is the partition function of temperature T; and F is an experimental parameter with the consideration of optical efficiency, laser-matter coupling, plasma density, and etc. The values of elemental concentration Cs in plant materials measured by LIBS are in fresh weight, so there is a need to translate it into the values Cref that are in dry weight, which are usually described in traditional methods. 7594

DOI: 10.1021/acs.analchem.7b01441 Anal. Chem. 2017, 89, 7593−7600

Article

Analytical Chemistry

Figure 1. Representative spectra of the leaves with different moisture contents under chromium stress (Mhigh (0 μM, 78.3%; 100 μM, 78.5%) and Mlow (0 μM, 2.1%; 100 μM, 2.5%)).

Cref =

Wele Wele Cs × 100% = × 100% = W2 (1 − ε)W1 (1 − ε)

routine that was based median absolute deviation (MDA) was used to detect outliers that might be due to the unevenness of leaves surfaces and the fluctuation of laser energy.23 Compared with other spectroscopy technologies, the repeatability of LIBS is mediocre, as the RSDs of shot-to-shot range from 20 to 30%.24 Hence, there is a need to improve the repeatability of LIBS, and we removed the outliers and averaged the rest of the spectra. It has been demonstrated that removing several abnormal spectra could improve the precision and repeatability of the LIBS measurements.25 In this case, we used the peak intensity of emission line CN 388.29 nm as the variable to detect outliers, because the CN molecule bands often appear in organic samples and the intensity of that is relatively stable. First, the median and median absolute deviation of the peak intensity (CN 388.29 nm) from different positions were calculated. The spectrum was considered as an outlier when the difference value between its intensity of CN 388.29 nm and median was beyond 2.5 times the MDA. We ran the procedure until no outliers were detected or the number of remaining spectra were less than 60% of the total number. The remaining number was chosen according to Pablo et al., with the consideration of the trade-off between the repeatability and the necessity to keep enough spectra for averaging.25 After removing the outliers, all those remaining spectra were averaged to reduce the effect of sample inhomogeneity and the random noise. All the preprocessing operations were done in Matlab 2014 b (The Mathworks Inc., Natick). Two calibration methods including univariate calibration and multivariable calibration were used for quantifying the chromium content in rice leaves. Both two strategies have been widely used in LIBS analysis.26,27 Univariate calibration method (it is also named calibration curve method) is a traditional calibration method, which establishes the model by relating signal to the reference value of analyte. Because it was hard to obtain the matrix-matched certified reference material

(3)

where eq 1 is used in eq 3, W1 is the weight of LIBS measured samples, Wele is the weight of emitting species, W2 is the weight of dried samples, and Cs is the concentration of emitting species. The relation between Cref and Cs in eq 3 is based on the assumption that the plasma composition is the same as the sample composition, and it is the case when analyzing the samples with different moisture contents. This hypothesis is often applied in the calibration of the laser-ablation technique.21 Iki = (1 − ε)Cref Fgk Aki

e(−Ek / KBT ) Zs(T )

(4)

Hence, the correlation between emission intensity and the concentration could be expressed in eq 4. For the plant materials with different moisture contents, the atomic parameters Aki, Ek, gk in eq 2 keep the same, which can be found in National Institute of Standards and Technology (NIST) database.22 In contrast, experimental parameter F must be assumed to change among different samples, especially for the plant materials. The optical efficiency and plasma features differ for the samples with different moisture contents.9 For plasma temperature T, the effect of moisture content is unsure. Hence, some efforts should be done to explore and compensate the variations of F and T among plant materials. LIBS Data Analysis. Prior to establishing a calibration model for heavy metal detection, several preprocessing methods were applied for the collected LIBS spectra. The first few variables in the beginning of spectra were discarded due to the large noise, resulting in a total of 21346 variables with the range of 240−880 nm. In addition, a self-developed 7595

DOI: 10.1021/acs.analchem.7b01441 Anal. Chem. 2017, 89, 7593−7600

Article

Analytical Chemistry

Figure 2. Representative spectra in the range of emissions for (a) hydrogen, (b) oxygen, and (c) chromium.

to determine the main influential factors on the variations of emissions. Figure 1 shows raw spectra of representative leaf samples with different moisture contents, including control group (Mhigh = 78.3%, Mlow = 2.1%) and experiment group of 100 μM chromium stress (Mhigh = 78.5%, Mlow = 2.5%). All emission lines were identified based on National Institute of Standards and Technology database (http://physics.nist.gov/ PhysRefData/ASD/lines_form.html). It was noticeable that most emission lines from high moisture content samples have lower intensity. In addition, some emission lines that exist in the spectra of low moisture content samples disappeared or had similar intensity level as noise. For example, the spectral profile of CN molecule bands changed greatly in high moisture content samples, the intensities of which decreased. In addition, some emission lines Ca I (428.88, 429.95, 430.32, 430.81 nm) and Mg I (516.79, 517.33, 518.43 nm) disappeared in high moisture content samples. We also explored the relations between the moisture content and specific elements (hydrogen, oxygen, and chromium), which are shown in Figure 2. In general, the peak intensity of these emission lines and the background decreased as the increase of moisture content. Obviously, the disadvantages that come from moisture content were more than the benefits, because the decrease of background was ignorant compared with the spectral intensity. In previous literature, it has been reported that there were positive correlations between moisture content and the signal of hydrogen10 and oxygen.29 However, in Figure 2a,b, the intensities decrease as the increase of moisture content. It might be credited to the long delay time (4 μs) of intensified CCD camera. This phenomenon was also observed in ref 9, which mentioned that the intensities of H and O decreased as the increase of moisture content at the delay time larger than 1 μs.9 For the spectral range around chromium emission lines (425.44 and 427.48 nm), the peak intensity decreased as the increase of moisture content. It was hard to observe the chromium emission lines of fresh leaves (the moisture content of leave equaled to 78.3%). In other words, it was difficult to detect heavy metal of fresh leaves even if it was of high concentration. Therefore, it was of great importance to provide an approach to eliminate the effect of moisture content, which was the key point for the realization of the quantification and visualization of heavy metals in plant materials. Signal Stability Analysis. Signal stability is one of the important figures of merit in analysis, which directly relates to the precision and repeatability. In this case, we explored the

(CRM), we calibrated the preprocessed LIBS signal to the ICPMS measured chromium content. For multivariable analysis, full spectrum or several selected variables combined with chemometric methods are often used for calibration. Unlike univariable calibration, multivariable analysis has the ability to extract useful information from raw signal, which includes the element concentration and other fingerprint features corresponding to the variations of experimental parameters and ablation reaction. Also, by establishing the models between the useful information and response, it could predict the element concentration correctly even with other interruptions (e.g., matrix effect, overlapped emission lines, shot-to-shot fluctuation). To investigate the calibration ability of multivariable analysis, PLSR model was employed to determine the chromium content. PLSR is a widely used chemometric method, which projects the independent variables (X) into latent variables and relates them to the dependent variable (Y).28 In our experiments, the dependent variable was the ICPMS measured chromium content. For both univariable and multivariable analysis, the samples were rearranged according to the reference value from low to high and each of the two samples were assigned to the calibration set with an interval of three, and the rest were grouped into a prediction set. Performance Evaluation. The calibration results were evaluated with correlation coefficient (R), root-mean-square error of calibration (RMSEC), and root-mean-square error of prediction (RMSEP). The higher value of R indicates better correlation between predicted values and measured values, and lower root-mean-square error (RMSE) indicates better accuracy of the model. The indicators were used in both calibration and prediction sets. In addition, limit of detection (LOD) was used to evaluate the sensitivity of proposed methods in univariate calibration, which is calculated with following equation:15 LOD = 3.3s /b

(5)

where s is the standard deviation of the background near the selected emission line and b is the slope of the calibration curve.

■

RESULTS AND DISCUSSION Raw Spectra Analysis. In our study, we first performed a proof-of-concept experiment (i.e., the first sampling) to explore the effect of moisture content on the LIBS analysis, including emissions intensity, signal stability, and plasma parameters and 7596

DOI: 10.1021/acs.analchem.7b01441 Anal. Chem. 2017, 89, 7593−7600

Article

Analytical Chemistry

different moisture contents were calculated. As seen in Figure 4, the values of temperature ranged from 6500 to 9300 K, most of

effect of moisture content on signal stability, and relative standard deviation (RSD) was used to evaluate the signal stability. The spectra from different positions were used to calculate the values of RSD, and the RSDs of two samples from the experiment group with 100 μM chromium stress were calculated. Figure 3 shows the RSDs of the intensities of main emission lines. The red column and black column indicated fresh leaf

Figure 4. (a) Electron density and (b) temperature of the plasma with various moisture content.

which are around 7800 K. For electron density, the values were more scattered with the range from 3.05 × 1016 to 5.73 × 1016 cm−3. In order to investigate the correlation between moisture content and the plasma parameters, Pearson’s correlation was performed using SPSS (ver. 20.0, SPSS Inc., Chicago). Both temperature and electron density were slightly correlated with moisture content, with correlation coefficients of 0.329 (P < 0.01) and −0.360 (P < 0.01), respectively. Therefore, the effect of moisture content on the plasma parameters was limited, and the variation of emissions might be mainly caused by the experimental parameters F and the change of analyte concentration. In addition, McWhirter criterion was used to verify the validity of local thermal equilibrium (LTE), and the threshold of electron density is given as32

Figure 3. Relative standard deviation of the intensities of main elements.

sample and dried leaf sample, respectively. Obviously, the emission lines from the dried leaf sample had lower RSDs, which indicated that moisture content would worsen the repeatability of LIBS analysis. It could be credited to the increase of instability of the plasma, as the evaporation of generated water might splash melted particles.9 In addition, larger RSDs were observed among the emission lines from C, H, O, N, which were the main constituent elements of the atmosphere. Because the fluence used in this experiment was higher than the ablation threshold of atmosphere, the laser beam ablated the atmosphere as well as the samples. Hence, the emission lines might be from leaf samples and the atmosphere. This to some extent increased the instability of the plasma, which then resulted in the high RSDs. For the emission lines of chromium (Cr I 425.44, 427.48 nm), maintaining the low moisture content in samples could help to improve the repeatability of analysis, which was important for quantification. Plasma Parameters Analysis. We further explored the effect of moisture content on the plasma parameters (electron density and temperature) based on optical emission features of LIBS. Because CN emissions often appear in organic samples, the profile of which might change due to the variance of temperature. Hence, temperature was obtained by spectrum simulation of the CN emission spectra around 388 nm using LIFBASE2.1 software.30 For electron density, Stark broadening of Hα was used, which might increase as a function of density:31 Ne = 8.02 × 1012(ΔStark /α1/2)3/2 , cm−3

Ne > 1.6 × 1012T1/2(ΔE)3 , cm−3

(7)

where ΔE corresponds to the energy gap of the first resonance transition of Cr I, i.e., the value of ΔE is around 2.9 eV. The fulfillment of the McWhirter criterion was considered as a necessary condition for the existence of LTE, which was crucial for quantification of heavy metals in plant materials. In this case, all the values of electron density achieved the threshold of McWhirter criterion. Correlation Analysis between Moisture Content and Spectra. Although there was a certain correlation between moisture content and the profile of spectra, we sought to quantify the relationship and determine the relative emission lines using PLS regression. As seen in Figure 5a, the moisture contents in plant materials are strongly correlated with LIBS signal, with the R of 0.97 in the cross-validation set. However, the RMSE was relatively low, which might be credited to the variances of elements concentration in samples. It indicated that LIBS combined with chemometrics has the potential to predict moisture content in plant materials, and further research should be carried on reducing the prediction error.

(6)

−3

where Ne is the electron density (cm ), α1/2 is a weak function of temperature and electron density, and the Δstark is the Stark broadening of Hα. All the spectra of first sampling were used for analysis. In this case, the values of temperature and electron density with 7597

DOI: 10.1021/acs.analchem.7b01441 Anal. Chem. 2017, 89, 7593−7600

Article

Analytical Chemistry

Figure 5. (a) Reference vs predicted value of moisture content. (b) Regression coefficients plot of PLS regression.

Table 1. Univariable Calibration Results Based on Different Variables calibration variables ICr 425.44 nm, Cref ICr 427.48 nm, Cref ICr 425.44 nm, Cs ICr 427.48 nm, Cs ICr 425.44 nm/Ibackground, ICr 427.48 nm/Ibackground, ICr 425.44 nm/Ibackground, ICr 427.48 nm/Ibackground,

Cref Cref Cs Cs

prediction

Rc

RMSEC

LOD

Rp

RMSEP

0.9133 0.8943 0.9166 0.8980 0.9393 0.9321 0.9397 0.9330

9.4416 10.5957 8.6146 9.6752 7.7331 8.2288 7.1839 7.6193

4.4665 5.2346 4.1505 4.8613 4.7078 5.5340 4.3856 5.1602

0.9421 0.9381 0.9439 0.9398 0.9526 0.9490 0.9526 0.9487

8.4292 9.0686 7.7388 8.3339 6.7331 7.3837 6.2561 6.8631

Detection of Chromium Content in Plant Materials. In this study, two strategies were used to compensate the effect of moisture content when detecting heavy metals using LIBS. One is to correct the element concentration in analyte using eq 3. In this case, an exponential model between the emission intensity of background and moisture content was established and used in eq 3.

Figure 5b shows the regression coefficients corresponding to each variable in spectrum. Greater absolute value of regression coefficients indicated the greater importance of the variable. As seen, the variables contributed to the PLS regression were main emission lines, and the top three important variables were K I 766.49, 769.90 nm, C I 247.86 nm. In addition, PLS regression based on top three important variables was performed, with R of 0.93 in the cross-validation set. It indicated that the important variables were sensitive to the variance of moisture content. It has been mentioned previously that the effect of moisture content on emission intensity was mainly the variation of experimental parameter F and analyte concentration. Hence, the correlation between moisture content and emission intensity might be nonlinear. Scatter plot of moisture content against top three important variables and background are shown in the Supporting Information, Figure S-1. We fitted the data with various functions, including linear, quadratic polynomial, power, and exponential function and found exponential and quadratic polynomial functions obtaining a good curve fitting result (see the Supporting Information, Table S-1). Because the peak intensity of emission was a certain constant when the moisture content reached 0 or 100%, it might be more proper to choose an exponential function. In addition, although the intensity of K I 766.49, 769.90 nm as well as background have a good relationship with moisture content, the emission intensity from K I might also be affected by the concentration of potassium. Therefore, the exponential function based on the background was used to describe the moisture content in the confirmatory study.

Cs = Cref (1 − ε) = Cref (1 − a ebI )

(8)

where I is the intensity of background and a and b are constants obtained from best curve fitting. The other one was to compensate the variance of experimental parameter F caused by moisture content and shot-to-shot fluctuation. The ratio of analytical signal to background and PLS regression were used in univariable calibration and multivariable calibration, respectively. Table 1 shows the univariable calibration results based on different input and output variables. With adjusted reference chromium content, the calibration models had higher R values and lower RMSEs and LODs. It indicated that moisture content to some extent reduce the proportion of elemental concentration, and calibrating the reference values with proposed model could improve the calibration models. In addition, using the ratio of signal to background could greatly improve the linearity and accuracy of the models, while the sensitivity of that was slightly decreased. It has been proven that the ratio could compensate the shot-to-shot shifting of the background level which is related to the deviations of the analytical line.33 Also the deviations could be caused by the variations of moisture content as well as the experimental parameters. The models with two different emission lines Cr I 7598

DOI: 10.1021/acs.analchem.7b01441 Anal. Chem. 2017, 89, 7593−7600

Article

Analytical Chemistry Table 2. Multivariable Calibration Results Based on Different Variables calibration

cross-validation

prediction

variables

latent variables

Rc

RMSEC

Rcv

RMSECV

Rp

RMSEP

I240−880 nm, Cref I425−428 nm, Cref I425−428 nm, Cs I425−428 nm/Ibackground, Cref I425−428 nm/Ibackground, Cs

3 3 3 1 1

0.8862 0.9530 0.9541 0.9480 0.9485

9.8079 6.4134 5.9129 6.7406 6.2536

0.7719 0.9320 0.9337 0.9323 0.9327

13.6933 7.6786 7.0739 7.8293 7.2784

0.8850 0.9660 0.9669 0.9606 0.9611

10.52 5.1560 4.7500 5.5922 5.1924

Figure 6. Relationship between reference value and LIBS measured value that predicted by (a) univariate model of ICPMS measured value and intensity of Cr I 425.44 nm, (b) univariate model of corrected reference value and the ratio of intensity of Cr I 425.44 nm to background, and (c) PLSR model of corrected reference value and variables in the range of 425−428 nm.

other elements. It indicated that irrelevant variables might have a detrimental effect on multivariable analysis, and feature selection methods could be used in further study. As seen in Figure 6, the methods proposed for univariable and multivariable calibration greatly improve the analytical performance. After compensation, the samples in both calibration and prediction sets distributed closely to the linear fit lines, which indicated that the deviations caused by shot-toshot fluctuation and moisture content were reduced. Although using the ratio of signal-to-background could reduce the detrimental influences and improve the performance to some extent (see Figure 6b), it was hard to compensate completely. Because the deviations were caused by the mixture of various factors, and many of that could not be compensated with a single variable. However, better compensation performance was observed in PLS calibration which indicated that it was more proper to choose multivariable calibration in this case.

425.44, 427.48 nm were compared. The results indicated that the peak intensity of emission line 425.44 nm has better capability to predict chromium content, which obtained higher R values and lower RMSEs and LODs in all calibration models. Hence, the model with adjusted reference content and the ratio (intensity of Cr I 425.44 nm to background) obtained the best performance, with a correlation coefficient of 0.9526, RMSE of 6.2561 mg/kg in the prediction set, and LOD of 4.3856 mg/kg. Compared with univariable calibration, multivariable analysis has the capability to deal with matrix effect, shot-to-shot fluctuation, and to extract useful information for prediction.26 In this study, PLS regression was used to predict chromium content with different variables as inputs and outputs (see Table 2). In order to avoid the risk of overfitting, full crossvalidation was performed to select the number of latent variables. PLS models with the variables in the range of 425− 428 nm and corrected reference values obtained the best result (Rp = 0.9669, RMSEP = 4.7500 mg/kg). Similar to traditional univariate analysis, calibrating with adjusted reference values could help to improve the model performance. However, no improvement or even worse performance was observed when the signal was normalized to background. As mentioned previously, PLS could reduce shot-to-shot fluctuation and extract valuable information. The variables in the range of 425− 428 nm not only contained the information to predict chromium content but also had the specialized features that might be caused by the variance of experimental parameters, laser-to-sample interaction, and matrix effect. In addition, the decrease of calibration performance and slight overfitting were observed when the full spectrum was included. Although other emissions of chromium except Cr I 425.44, 427.48 nm might appear in the entire range of spectra, the emissions were so weak that it might be overwhelmed by noise or emissions from

■

CONCLUSION In this experiment, we have proposed moisture influence reducing method for heavy metal detection in plant materials. We demonstrated that moisture content in plant materials could greatly reduce the signal intensity and worsen the signal stability. The variation of emissions might be mainly caused by the experimental parameters F and the change of analyte concentration. It was feasible to quantify the chromium content in rice leaves with a quick drying. Two strategies were used to compensate the effect of moisture content and shot-to-shot fluctuation. It has shown that an exponential function based on the intensity of background could be used to correct the actual element concentration in plant materials. In addition, using the ratio of signal-to-background for univariate analysis and PLSR for multivariate analysis could eliminate the detrimental effects 7599

DOI: 10.1021/acs.analchem.7b01441 Anal. Chem. 2017, 89, 7593−7600

Article

Analytical Chemistry

(5) Peng, J.; Liu, F.; Zhou, F.; Song, K. L.; Zhang, C.; Ye, L. H.; He, Y. TrAC, Trends Anal. Chem. 2016, 85, 260−272. (6) Fortes, F. J.; Moros, J.; Lucena, P.; Cabalin, L. M.; Laserna, J. J. Anal. Chem. 2013, 85 (2), 640−69. (7) Harmon, R. S.; Russo, R. E.; Hark, R. R. Spectrochim. Acta, Part B 2013, 87, 11−26. (8) Singh, V. K.; Rai, A. K. Lasers Med. Sci. 2011, 26 (5), 673−87. (9) Chen, M.; Yuan, T.; Hou, Z.; Wang, Z.; Wang, Y. Spectrochim. Acta, Part B 2015, 112, 23−33. (10) Paris, P.; Piip, K.; Lepp, A.; Lissovski, A.; Aints, M.; Laan, M. Spectrochim. Acta, Part B 2015, 107, 61−66. (11) Wisbrun, R.; Schechter, I.; Niessner, R.; Schroeder, H.; Kompa, K. L. Anal. Chem. 1994, 66 (18), 2964−2975. (12) Kaiser, J.; Galiová, M.; Novotný, K.; Č ervenka, R.; Reale, L.; Novotný, J.; Liška, M.; Samek, O.; Kanický, V.; Hrdlička, A.; Stejskal, K.; Adam, V.; Kizek, R. Spectrochim. Acta, Part B 2009, 64 (1), 67−73. (13) Gomes, M. D.; Santos, D.; Nunes, L. C.; de Carvalho, G. G. A.; Leme, F. D.; Krug, F. J. Talanta 2011, 85 (4), 1744−1750. (14) Arantes de Carvalho, G. G.; Moros, J.; Santos, D., Jr.; Krug, F. J.; Laserna, J. J. Anal. Chim. Acta 2015, 876, 26−38. (15) de Souza, P. F.; Santos, D.; de Carvalho, G. G. A.; Nunes, L. C.; Gomes, M. D.; Guerra, M. B. B.; Krug, F. J. Spectrochim. Acta, Part B 2013, 83−84, 61−65. (16) Trevizan, L. C.; Santos, D.; Samad, R. E.; Vieira, N. D.; Nunes, L. C.; Rufini, I. A.; Krug, F. J. Spectrochim. Acta, Part B 2009, 64 (5), 369−377. (17) Yoshida, S.; Forno, D. A.; Cock, J. H.; Gomez, K. A. Laboratory Manual for Physiological Studies of Rice. International Rice Research Institute: Los Banos, Philippines, 1976. (18) Peng, J.; Liu, F.; Kong, W.; Zhang, C.; Yu, K.; He, Y. Trans. ASABE 2017, 60 (1), 19−28. (19) Peng, J.; Song, K.; Zhu, H.; Kong, W.; Liu, F.; Shen, T.; He, Y. Sci. Rep. 2017, 7, 44551. (20) Ciucci, A.; Corsi, M.; Palleschi, V.; Rastelli, S.; Salvetti, A.; Tognoni, E. Appl. Spectrosc. 1999, 53 (8), 960−964. (21) Tognoni, E.; Cristoforetti, G.; Legnaioli, S.; Palleschi, V. Spectrochim. Acta, Part B 2010, 65 (1), 1−14. (22) Kramida, A., Ralchenko, Yu., Reader, J.NIST ASD Team. NIST Atomic Spectra Database, National Institute of Standards and Technology: Gaithersburg, MD, 2015. (23) Leys, C.; Ley, C.; Klein, O.; Bernard, P.; Licata, L. J. Exp. Soc. Psychol. 2013, 49 (4), 764−766. (24) Galbacs, G. Anal. Bioanal. Chem. 2015, 407 (25), 7537−62. (25) Sobron, P.; Wang, A.; Sobron, F. Spectrochim. Acta, Part B 2012, 68, 1−16. (26) Hahn, D. W.; Omenetto, N. Appl. Spectrosc. 2012, 66 (4), 347− 419. (27) El Haddad, J.; Canioni, L.; Bousquet, B. Spectrochim. Acta, Part B 2014, 101, 171−182. (28) Haaland, D. M.; Thomas, E. V. Anal. Chem. 1988, 60 (11), 1193−1202. (29) Liu, Y.; Gigant, L.; Baudelet, M.; Richardson, M. Spectrochim. Acta, Part B 2012, 73, 71−74. (30) Luque, J.; Crosley, D. R. LIFBASE: Database and Spectral Simulation Program, SRI International Report MP 99-009, 1999. (31) Griem, H. R. Plasma Spectroscopy; McGraw-Hill Book Company: New York, 1964. (32) McWhirter, R. W. P. Plasma Diagnostic Techniques; Academic Press: New York, 1965. (33) Zorov, N. B.; Gorbatenko, A. A.; Labutin, T. A.; Popov, A. M. Spectrochim. Acta, Part B 2010, 65 (8), 642−657.

and improve the calibration performance. Best calibration result was achieved by the PLSR model that relating the variables in the range of 425−428 nm with corrected reference values, with the Rp of 0.9669 and RMSEP of 4.7500 mg/kg. In addition, the best calibration result of PLSR outperformed that in univariable calibration, which indicated that multivariable analysis was preferred in a complex situation. The proposed approach provided the first proof-of-principal data for application of LIBS for quantifying the heavy metals in high-moisture containing plant materials using a quick drying procedure, and the effect of moisture could be compensated to some extent. The proposed approach is simple and efficient, and it is available for element mapping in plant materials. However, further advances beyond our study are still needed. The approach proposed here is based on laboratorial study, and the operations (e.g., drying time, sampling shots, experimental parameters, etc.) could be optimized for specialized samples. In practical application, the drying processing can be performed using an accessory that integrated into the LIBS system. In addition, since the multivariable methods has proven efficient for calibration in complex situations, more samples with other chemometric methods could be involved in future work. It might help to develop a robust model as well as increasing the detection accuracy.

■

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.analchem.7b01441. Scatter plots of moisture content vs intensity of K I 766.49, 769.90 nm, C I 247.86 nm, and background and table of the determination coefficients of curve fitting results between peak intensity of K I 766.49, 769.90 nm, C I 247.86 nm, and background and moisture content (PDF)

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: fl[email protected]. Phone: +86-571-88982825. Fax: +86571-88982143. ORCID

Fei Liu: 0000-0003-0266-6896 Author Contributions ∥

Jiyu Peng and Yong He contributed equally to this work.

Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS This work was supported by Natural Science Foundation of China (Grants 31671579, 61605173), China Postdoctoral Science Foundation (Grants 2016M600466, 2017T100431), and Zhejiang Provincial Natural Science Foundation of China (Grant LY15C130003).

■

REFERENCES

(1) Wang, J.; Chen, C. Biotechnol. Adv. 2006, 24 (5), 427−451. (2) Jarup, L. Br. Med. Bull. 2003, 68, 167−182. (3) Khan, S.; Cao, Q.; Zheng, Y. M.; Huang, Y. Z.; Zhu, Y. G. Environ. Pollut. 2008, 152 (3), 686−692. (4) Hashim, M. A.; Mukhopadhyay, S.; Sahu, J. N.; Sengupta, B. J. Environ. Manage. 2011, 92 (10), 2355−88. 7600

DOI: 10.1021/acs.analchem.7b01441 Anal. Chem. 2017, 89, 7593−7600