Determination of Cadmium (II) in Aqueous Solutions by In Situ MID-FTIR-PLS Analysis Using a Polymer Inclusion Membrane-Based Sensor: First Considerations

Environmental monitoring is one of the most dynamically developing branches of chemical analysis. In this area, the use of multidimensional techniques and methods is encouraged to allow reliable determinations of metal ions with portable equipment for in-field applications. In this regard, this study presents, for the first time, the capabilities of a polymer inclusion membrane (PIM) sensor to perform cadmium (II) determination in aqueous solutions by in situ visible (VIS) and Mid- Fourier transform infrared spectroscopy (MID-FTIR) analyses of the polymeric films, using a partial least squares (PLS) chemometric approach. The influence of pH and metal content on cadmium (II) extraction, the characterization of its extraction in terms of the adsorption isotherm, enrichment factor and extraction equilibrium were studied. The PLS chemometric algorithm was applied to the spectral data to establish the relationship between cadmium (II) content in the membrane and the absorption spectra. Furthermore, the developed MID-FTIR method was validated through the determination of the figures of merit (accuracy, linearity, sensitivity, analytical sensitivity, minimum discernible concentration difference, mean selectivity, and limits of detection and quantitation). Results showed reliable calibration curves denoting systems’ potentiality. Comparable results were obtained in the analysis of real samples (tap, bottle, and pier water) between the new MID-FTIR-PLS PIM based-sensor and F-AAS.


Introduction
Heavy metals are persistent toxic metals or metalloids present at low concentrations, in all parts of the environment. Human activity affects the natural geological and biological redistribution of these metals through pollution of air, water, and soil. Therefore, highly accurate and sensitive spectrophotometric and spectrometric analytical techniques are used for their measurement in complex matrices, e.g., atomic absorption (Flame atomic absorption spectroscopy (F-AAS), Graphite furnace atomic absorption spectroscopy (GF-AAS), Hydride generation atomic absorption spectroscopy (HG-AAS), and Cold vapor atomic absorption spectroscopy (CV-AAS)), emission (Inductively coupled plasma optical emission spectroscopy (ICP-OES)), and mass (Inductively coupled plasma mass spectrometry (ICP-MS)) methods. At present, analytics and environmental monitoring are among the most dynamically developing branches of chemical analysis. The general trends in both areas can be classified into two basic groups-(i) development of new methodical procedures, and (ii) new achievements in the construction of measuring instruments (instrumentation) [1]. Examples of the first group developed in the pursuit of obtaining complex information on environmental quality are the introduction of solventless techniques to the analytical practice for sample preparation and multidimensional techniques, while examples of the second are the design of new sensors and detectors for
It is interesting to note that from 1 × 10 −3 mol dm −3 and on, the metal was practically not extracted in the system, due to a saturation phenomenon of the extracting phase. Further experiments were then performed, maintaining cadmium (II) concentrations within the range of 5 × 10 −4 to 1 × 10 −4 mol dm −3 . In addition, the characterization of the systems in terms of its absorption capacity was studied. It is interesting to note that from 1 × 10 −3 mol dm −3 and on, the metal was practically not extracted in the system, due to a saturation phenomenon of the extracting phase. Further experiments were then performed, maintaining cadmium (II) concentrations within the range of 5 × 10 −4 to 1 × 10 −4 mol dm −3 . In addition, the characterization of the systems in terms of its absorption capacity was studied.

Adsorption Isotherm
The amount of metal in the membrane phase, qe [mmol g −1 ] was plotted with respect to its equilibrium concentration in the aqueous phase, Ce [mmol cm -3 ] (Figure 3A), for a range of aqueous concentrations from 6.94 × 10 −7 to 3.82 × 10 −4 mol dm −3 . As a Langmuir type isotherm was observed (Equation 1), linearization of the data (Ce/qe = f(Ce)) was applied to determine the adsorption constant, KL (cm 3 mmol −1 ), and the maximum adsorption capacity, qmax (mmol g −1 ), parameters.
After performing the analysis, a careful inspection of the data pointed toward two different regions, depending on the metal content, high (1.63 × 10 −6 to 5.16 × 10 −5 (mmol cm −3 )) and low (1.11 × 10 −6 to 9.33 × 10 −7 (mmol cm -3 )), in which the degree of fit of the model was better performed, assuming different model parameters for each region ( Figure 3B).

Adsorption Isotherm
The amount of metal in the membrane phase, q e [mmol g −1 ] was plotted with respect to its equilibrium concentration in the aqueous phase, C e [mmol cm −3 ] (Figure 3A), for a range of aqueous concentrations from 6.94 × 10 −7 to 3.82 × 10 −4 mol dm −3 . As a Langmuir type isotherm was observed Equation (1), linearization of the data (C e /q e = f(C e )) was applied to determine the adsorption constant, K L (cm 3 mmol −1 ), and the maximum adsorption capacity, q max (mmol g −1 ), parameters.  In Table 1, both sets of model parameters are reported. It can be observed that qmax decreased with a diminishing Ce, as it probably became independent of the metal content [45], increasing the data dispersion. This behavior was further analyzed through computation of the separation coefficient, RL, defined by After performing the analysis, a careful inspection of the data pointed toward two different regions, depending on the metal content, high (1.63 × 10 −6 to 5.16 × 10 −5 (mmol cm −3 )) and low (1.11 × 10 −6 to 9.33 × 10 −7 (mmol cm −3 )), in which the degree of fit of the model was better performed, assuming different model parameters for each region ( Figure 3B).
In Table 1, both sets of model parameters are reported. It can be observed that q max decreased with a diminishing C e , as it probably became independent of the metal content [45], increasing the data dispersion. This behavior was further analyzed through computation of the separation coefficient, R L , defined by Values of R L > 1 were indicative of a non-favorable adsorption; R L = 1 indicates a linear adsorption, while 0 < R L < 1 were observed in favorable adsorption. An irreversible adsorption was present in such a case where R L = 0 [46], in the high concentration range of the studied PIM system 0 < R L < 1, pointing out a favorable adsorption [47]. On the contrary, in the low range, R L ≈ 0, denoted an irreversible adsorption [46]. This observation agreed well with the two distinct regions observed in the adsorption isotherm. Comparing q max to other cadmium (II) sorbents, Fan et al. reported q max = 0.545 mmol g −1 for P. Simplicissimum [48], while Chakravarty et al. reported q max = 0.0948 mmol g −1 for Areca catechu [49], and Singh et al. reported q max = 9.43 × 10 −4 mmol g −1 for Trichoderma viridae [50]. Some inorganic sorbents like activated alumina CNT nanoclusters, oxidized CNTs, and Fe 3 O 4 @TA showed maximum cadmium (II) adsorption capacities of 229.9, 11.01, and 286 mg g −1 [51] (equivalent to 2.04, 0.098, and 2.54 mmol g −1 , respectively). These results showed that a wide interval of maximum sorption capacities for the metal could be obtained, depending on the type of sorbent, pH, temperature, ionic force, among other factors. The obtained q max values then lies within the reported ranges.

Enrichment Factor
The enrichment factor, E, defined by is a measure of the preconcentration efficiency of the system. When plotting [Cd(II)] membrane (mmol g −1 ) as a function of [Cd(II)] initial (mmol dm −3 ) within the interval 6.94 × 10 −7 a 3.82 × 10 −4 mol dm −3 a linear relationship was observed, denoting a constant value of this parameter. From the slope, a value of E = 29.2 was determined. This result guaranteed the application of the sorption in the PIM as an adequate preconcentration method. The initial amount of cadmium (II) in the aqueous phase could be predicted from the amount of cadmium (II) in the membrane phase and the constant value of E in the metal concentration range, in which this constant value was attained. From the comparison of the E value with other membrane-based cadmium (II) preconcentration methods, the PIM once again presented an acceptable intermediate value, as Castro et al. reported E = 17.9, with the use of liquid membranes containing 2-APHB in toluene as extractant [52], while Peng et al. reported E = 387 using hollow fibers with dithizone dissolved in a mixture of 1-octanol and oleic acid [53]. Evidently, the E value should be dependent on the type of sorbent, cadmium (II) content, pH, temperature, ionic force, among other factors.

Stoichiometry of the Extracted Complex
Experiments in which Kelex 100 concentration was varied (1.9, 2.8, 3.6, 4.6, 7.1, 9.3, 10.9, and 12.3 w/w%), maintaining constant amounts of CTA and NPOE, were performed to determine the stoichiometry of the extracted complex through conventional graphical slope analysis. According to Aguilar et al. [42], cadmium (II) extraction with Kelex 100 proceeded through the reaction in which HL stands for the extractant, CdH n−1 L n NO 3 for the extracted species, the bar denotes species in the membrane phase, and n = 1 and 2, depending on the nature of the ionic medium. From the extraction equilibrium constant, K ext , defined by once the distribution coefficient, D, is considered where Cd(II) and [Cd(II)] are total membrane and aqueous phases equilibrium concentrations, respectively. From the plot logD = f ([HL]) pH a value of n ≈ 2. Such results perfectly agree with that reported by Aguilar et al. [42] for the extraction of the analyte with Kelex 100 in a solvent extraction system, using kerosene as solvent in nitrate medium. The determined extraction constant is logK ext = 0.02.

PLS modeling of VIS and FTIR data
As it was observed that VIS and FTIR information varied with cadmium concentration ( Figure 4A,B), the obtained spectral data were submitted to PLS regression analysis to correlate the two data matrices, the X matrix (the VIS and FTIR spectra) and Y matrix (the property, i.e., cadmium content).
The employed concentration range was selected so that a Langmuir-type absorption of the metal ion by the PIM and a favorable preconcentration factor were attained. In the beginning, the complete spectral range was employed. However, from the analysis of the regression coefficients and model parameters, an improvement in the regression parameters was observed when the FTIR range was restricted to 700-410 cm −1 , and, consequently, further processing was performed using this interval. In the first instance, it was verified that all samples were representative of the same population, using a Hotelling T 2 test, in conjunction with an F-residual plot. Once no outliers were detected, the analyses results were interpreted. From Figure 5A it was observed that 96% of variability in the VIS spectra was Molecules 2020, 25, 3436 7 of 15 explained using 2 factors; similarly, in the case of FTIR spectra ( Figure 5B), the variability explained by the two first factors almost reached 100% for the spectral data.
using kerosene as solvent in nitrate medium. The determined extraction constant is = 0.02.

Multivariate Regression Analysis
2.5.1. PLS modeling of VIS and FTIR data As it was observed that VIS and FTIR information varied with cadmium concentration ( Figures  4A and 4B), the obtained spectral data were submitted to PLS regression analysis to correlate the two data matrices, the X matrix (the VIS and FTIR spectra) and Y matrix (the property, i.e., cadmium content). The employed concentration range was selected so that a Langmuir-type absorption of the metal ion by the PIM and a favorable preconcentration factor were attained. In the beginning, the complete spectral range was employed. However, from the analysis of the regression coefficients and model parameters, an improvement in the regression parameters was observed when the FTIR range was restricted to 700-410 cm −1 , and, consequently, further processing was performed using this interval. In the first instance, it was verified that all samples were representative of the same population, using a Hotelling T 2 test, in conjunction with an F-residual plot. Once no outliers were detected, the analyses results were interpreted. From Figure 5A it was observed that 96% of variability in the VIS spectra was explained using 2 factors; similarly, in the case of FTIR spectra ( Figure 5B), the variability explained by the two first factors almost reached 100% for the spectral data. While the model for VIS data required just two latent variables, the model for FTIR data incorporated a third one, as selected according to a leave-one-out cross-validation process. This result was a direct consequence of the differences in complexity between the VIS and FTIR spectra. The accuracy of the models was quantitatively measured through the RMSEC, the RMSECV, and the  The employed concentration range was selected so that a Langmuir-type absorption of the metal ion by the PIM and a favorable preconcentration factor were attained. In the beginning, the complete spectral range was employed. However, from the analysis of the regression coefficients and model parameters, an improvement in the regression parameters was observed when the FTIR range was restricted to 700-410 cm −1 , and, consequently, further processing was performed using this interval. In the first instance, it was verified that all samples were representative of the same population, using a Hotelling T 2 test, in conjunction with an F-residual plot. Once no outliers were detected, the analyses results were interpreted. From Figure 5A it was observed that 96% of variability in the VIS spectra was explained using 2 factors; similarly, in the case of FTIR spectra ( Figure 5B), the variability explained by the two first factors almost reached 100% for the spectral data. While the model for VIS data required just two latent variables, the model for FTIR data incorporated a third one, as selected according to a leave-one-out cross-validation process. This result was a direct consequence of the differences in complexity between the VIS and FTIR spectra. The accuracy of the models was quantitatively measured through the RMSEC, the RMSECV, and the While the model for VIS data required just two latent variables, the model for FTIR data incorporated a third one, as selected according to a leave-one-out cross-validation process. This result was a direct consequence of the differences in complexity between the VIS and FTIR spectra. The accuracy of the models was quantitatively measured through the RMSEC, the RMSECV, and the slope, intercept, and determination coefficient (R 2 ) from the reference vs. predicted values of the property during calibration and cross-validation ( Figure 6A,B and Table 2).  Table 2).    As observed good performance was accomplished by the models and no systematic variations were detected based on the slope (b 1 ) and intercept values (b 0 ) of the regression equations, i.e., the joined F-test for both statistical parameters gave no significant differences at 95% confidence between these values and the expected ones for the slope (β 1 = 1) and intercept (β 0 = 0) (p-values were 0.9590 and 0.3703 for VIS data for calibration and cross-validation, respectively, and 0.9733 and 0.2731 for FTIR data for calibration and cross-validation, respectively) according to the statistics: where S 2 e error mean square; n, number of data points; x, mean of the reference values; and x 2 i /n, mean sum of squares of the reference values. The high values of the determination coefficients give information about the goodness of fit of the models, as this parameter is a statistical measure of how well the regression line approximates the real data points. In addition, the CV-determination coefficients showed a good predictive ability. The agreement between model predictions and ideal behavior is clearly seen in Figure 6A,B from the closeness of the data with the ideal reference line. The regression coefficients ( Figure 7A,B) showed similarities with spectral data, which was simpler with VIS than with FTIR data. While the VIS regression coefficient showed a curve profile with a maximal contribution typical of the 8-hydroxyquinoline-Cd(II) complex [54], the FTIR regression coefficient profile included Kelex 100 characteristic IR vibrational bands at about 687 cm −1 and 459 cm −1 , related to C-H out-of-plane bending and to C-O in-plane bending, respectively [55], as expected from the coordination properties of the extractant through its oxygen group.

Figures of Merit of the MID-FTIR-PLS PIM-Based Sensor
At this point it is important to mention that both models gave very good results, being slightly better for FTIR than for the VIS data, as indicated by the higher values of R 2 and lower values of RMSEC, RMSECV. Based on these observations, the MID-FTIR model was further characterized to extend its application to complex natural waters. Due to the specific spectroscopic signals between the metal ion and the extractant, a minimal or no effect caused by the impurities and the suspended particles is expected, and an accurate analysis could be performed. The unnecessary use of a chromophore agents is an additional advantage of the method. In Table 3

Figures of Merit of the MID-FTIR-PLS PIM-Based Sensor
At this point it is important to mention that both models gave very good results, being slightly better for FTIR than for the VIS data, as indicated by the higher values of R 2 and lower values of RMSEC, RMSECV. Based on these observations, the MID-FTIR model was further characterized to extend its application to complex natural waters. Due to the specific spectroscopic signals between the metal ion and the extractant, a minimal or no effect caused by the impurities and the suspended particles is expected, and an accurate analysis could be performed. The unnecessary use of a chromophore agents is an additional advantage of the method. In Table 3, the figures of merit of the FTIR-developed model are presented (linearity, evaluated from RMSEC = n i=1 (y i −ŷ i ) 2 n−1 , cross validation RMSE (RMSECV), determination coefficient (R 2 ), cross-validation R 2 (CV-R 2 ), slope, and intercept of the cadmium (II) observed vs. predicted results; sensitivity, as sen = ||s * k || = 1 ||b || ; analytical sensitivity as γ = sen |δx| ; minimum discernible concentration difference, γ −1 = |δx| sen ; and limits of detection and quantitation (LD = 3.3δx 1 sen and LQ = 10δx 1 sen , respectively), where y i andŷ i are the estimated and reference values, respectively, of the I, sample, n the total number of samples, ||s k || stands for the norm of the sensitivity coefficients of the spectra containing the analyte k at unit concentration and ||s * k || for that corresponding to its NAS, where x i is a sample spectrum after preprocessing and b is a column vector of the PLS regression coefficients, ||b || is the norm of the vector of regression coefficients of the calibration model, and δx is the instrumental noise [56][57][58]. Overall, good performance characteristics were observed. The low value of multivariate selectivity (4.03%) was anticipated according to the employed experimental conditions, as this parameter was a measure of the fraction of the spectrum that was related to the cadmium content in the PIM, and it should be considered that the Kelex 100: cadmium(II) ratio in the PIM was very high (approximately, a ratio of 21).

Application of the MID-FTIR-PLS PIM-Based Sensor
As the model's accuracy was dependent on the presence of interferences, the application of the model to three different natural waters (tap, bottle, and pier water) representing complex matrices due to the presence of different ions (i.e., calcium, magnesium, sodium, potassium, chloride, nitrate, sulfate, bicarbonate, fluoride, among others [59,60]) and particulates (e.g., dissolved organic compounds) was evaluated. Samples were spiked with different analyte concentrations and the results of the FTIR method was compared with F-AAS analysis. As observed from Table 4, the non-specific character of 8-hydroxyquinoline, i.e., Kelex 100, which could form complexes with Na (I), Ca (II), and Mg (II), among other ions [61,62], was perfectly compensated by the Cd (II)-Kelex 100 complex specific information contained within the analyzed FTIR spectral region and the pH value selected for the analysis. The joined F-test for the slope and intercept values of the regression equation between the reference vs. the determined values gave no significant differences at the 95% confidence between these values and the expected ones (p-value = 0.0774). Consequently, comparable results were obtained between both methods, even in the analysis of a challenging medium like pier water. Table 4. Results of the analysis of cadmium (II) in real water samples spiked with the analyte.

Sample
Reference Value (F-AAS) × 10 4 mol dm −3 Determined Value (MID-FTIR) × 10 4 mol dm −3  One major advantage of the developed MID-FTIR-PLS PIM-based method was that it did not require the presence in the membrane of a chemical reagent with special properties, either a chromophore species that is able to complex the metal ion, i.e., acting as ionophore [19], or a mixture of an ionophore and a chromophore in the same PIM [17], or a fluorescent reagent [18]. Consequently, there was no need to optimize the PIM composition for chromophore/ionophore/support compatibility [25], so that in practice the methodology is transferable to any PIM system reported up till now. Future work will be addressed toward extending the range of application of the methodology to-(i) a lower analyte concentration range by a careful selection of the dielectric nature of the medium and the dipole moment of the bond associated with IR vibrations of the extracted complex (variation in PIM component's composition and nature); the larger the dipole moment change and the smaller the position change of the atoms (i.e., of bond lengths or bond angles), the higher the band intensities [63]; and (ii) different analytes, either alone or in a mixture, to fit the purpose of environmental monitoring. In this regard, taking into account a similar behavior of Kelex 100 to its parent structure 8-hydroxyquinoline  [64], it is evident that the influence of the presence of other heavy metal ions is a challenge to handle. Chemometric selectivity based on specific absorption bands for the different metal ions might represent a promising alternative to be investigated in future applications. Overall, this article showed the potentiality of the proposed methodology and allowed a proof of the concept for the target purpose.
Extraction experiments were carried out with a model 75 Wrist ActionTM shaker (Burrell Scientific Inc, Pittsburgh, Pa., USA). The spectrometers Perkin Elmer 3100, Perkin Elmer Lambda 2 and Perkin Elmer Spectrum GX (Waltham, Mass., USA) were used for F-AAS, VIS, and FTIR determinations, respectively. A Metrohm 620 pH-meter (Herisau, Switzerland) was employed for pH measurement and adjustment. A Fowler IP54 micrometer (Fowler High Precision, Newton, Mass., USA) was used for measuring PIM thickness. The Unscrambler 10.5.1 software (Camo Analytics, Oslo, Norway) was employed for PLS analyses.
PIMs were prepared by dissolving the weighted amounts of CTA, NPOE, and Kelex 100 in dichloromethane. The mixture was stirred for 1 h on a magnetic plate with a stirring bar. The solution was then casted in a 5 cm diameter Petri dish and rested for 24 h for solvent evaporation. Finally, the membrane was carefully peeled off and the whole piece was used in all experiments. PIMs were transparent films with an average thickness of (46 ± 11) µm, average weight of (112.3 ± 0.0056) mg and an average diameter of (4.89 ± 0.0322) cm.
As for the biphasic solid-liquid extraction experiments, the membranes were introduced in 50 mL polypropylene Falcon tubes, in the presence of 30 mL of aqueous solution, with cadmium (II) at fixed concentrations. The tubes were shaken for regular time intervals and the aliquots of 400 µL were taken and diluted to 2 mL before F-AAS analysis, using the conditions recommended by the manufacturer (λ 228.8 nm, 7 nm slit, air/acetylene flame). Experiments were performed on a duplicate basis with an average RSD of 5%. No analyte elution step was required as direct analysis of the PIMs was performed. This was opposite to the traditional three-phases configuration (feed, membrane, strip) usually employed for metal ion removal. The employed set-up was commonly found when the PIMs were used for sensing in chemical analysis [25].
Chemometric analyses were conducted using a training set of 15 different concentrations (in duplicate), ranging from 6.94 × 10 −7 to 3.82 × 10 −4 mol dm −3 . The concentration of the calibration standards was determined by F-AAS (Perkin Elmer 3100, Waltham, Mass., USA). The same PIMs were analyzed by VIS and MID-FTIR spectroscopies. VIS spectra were recorded in transmission mode by triplicates, in the range 500-390 nm, after sandwiching the membrane between two Petri dishes to avoid wrinkles and movement. FTIR spectra were recorded in transmission mode by triplicates in the range of 4000-400 cm −1 . The PIM was mounted in the transmission accessory of the equipment and scanned 45 times to record the spectrum. The six spectra for each concentration in VIS and IR modes were then averaged for multivariate analysis, using the spectra calculator of the software. The best results were obtained after mean-centering the spectra and, in the case of the FTIR data, applying the unit vector normalization. Chemometric analyses were validated through cross-validation procedures. An in-house made MATHLAB program was used for the figures of merit determination, using the outputs of the Unscrambler 10.5.1 (Camo Analytics, Oslo, Norway) software. The applicability of the method was tested by spiking with reference concentrations bottle, tap, and pier (Cuemanco, Xochimilco, Mexico) water, after filtration of the samples.

Conclusions
The results showed that sorption coupled with direct spectroscopic analysis, using a PLS chemometric approach in a PIM, constituted a potential sensor for metal ion determination in waters with results comparable to F-AAS. On-site analysis with portable equipment was anticipated through the proof of the concept of the methodology in the case of the measurement of cadmium (II) in aqueous solutions, using the commercial extractant Kelex 100 immobilized in the membrane. MID-FTIR showed to be an adequate technique for cadmium (II) analysis in the membrane, once the metal was preconcentrated, which was barely used for quantitative purposes in this type of application. The specificity of bonding between the metal and the extractant, together with the optimization of the uptake conditions, allowed the selectivity of the system toward competing ions, while the chemometric treatment of the spectral data allowed the selectivity of the metal signal toward the organic components conforming the membrane. A wide range of future applications is anticipated to target different metal ions with specific extractants immobilized in PIMs, as no chromophores or fluorescent reagents are needed in this novel type of application. The simplicity of the system is expected to optimize the PIM composition for chromophore/ionophore/support compatibility, and in its transferability, as all organic extractants present active functional groups in the MID-IR region.