Rapid and Quantitative Determination of Sildenafil in Cocktail Based on Surface Enhanced Raman Spectroscopy

Sildenafil (SD) and its related compounds are the most common adulterants found in herbal preparations used as sexual enhancer or man’s virility products. However, the abuse of SD threatens human health such as through headache, back pain, rhinitis, etc. Therefore, it is important to accurately detect the presence of SD in alcoholic beverages. In this study, the Opto Trace Raman 202 (OTR 202) was used as a surface-enhanced Raman spectroscopy (SERS) active colloids to detect SD. The results demonstrated that the limit of detection (LOD) of SD was found to be as low as 0.1 mg/L. Moreover, 1235, 1401, 1530, and 1584 cm−1 could be qualitatively determined as SD characteristic peaks. In a practical application, SD in cocktail could be easily detected using SERS based on OTR 202. Also, there was a good linear correlation between the intensity of Raman peaks at 1235, 1401, 1530, and 1584 cm−1 and the logarithm of SD concentration in cocktail was in the range of 0.1–10 mg/L (0.9822 < R2 < 0.9860). The relative standard deviation (RSD) was less than 12.7% and the recovery ranged from 93.0%–105.8%. Moreover, the original 500–1700 cm−1 SERS spectra were pretreated and the partial least squares (PLS) was applied to establish the prediction model between SERS spectra and SD content in cocktail and the highest determination coefficient (Rp2) reached 0.9856. In summary, the SD in cocktail could be rapidly and quantitatively determined by SERS, which was beneficial to provide a rapid and accurate scheme for the detection of SD in alcoholic beverages.


Introduction
Sildenafil (SD) and its related compounds are the most common adulterants found in herbal preparation, which can be used as sexual enhancer or man's virility products [1]. Its pharmacological effect is to inhibit the metabolism of the second messenger cyclic guanosinc monophosphate (cGMP), promote the relaxation of cavernous artery smooth muscle, and then improve the symptoms of erectile dysfunction (ED) [2,3]. However, the usage of SD is controlled through medical supervision due to their harmful side-effects such as headache, dyspepsia, back pain, rhinitis, flu syndrome, etc. [4]. In recent years, SD, through illegal business, has illegally added to Chinese patent medicines It can be clearly seen that the average diameter of OTR 202 was about 30 nm. As shown in Figure  1b, the UV/Visible characteristic absorption peaks of OTR 202 was at 533 nm. Beside this, the Raman spectrum of OTR 202 only had a faint signal at 1630 cm −1 (Figure 1c), suggesting that OTR 202 themselves had no strong Raman characteristic peaks and did not have an interferential effect on experimental results. Therefore, the OTR 202 was suitable as SERS substrate to detect SD in this paper.

The SD Molecule and its Assignment of Raman Peaks
The molecular structure of SD (molecular formula: C22H30N6O4S) is shown in Figure 2a. Density functional theory (DFT), as a common method for molecular geometry optimization and frequency vibration calculation, can describe the ground state physical properties of atoms and molecules [26]. In this paper, DFT was applied to calculate SD molecule and optimize its structure in Gaussian.v09 software. In this software, the vibrational form of relevant chemical bonds calculated by Hartree-Fock wave function can be obtained [27]. Figure 2b is the RS of SD simulated by DFT, Figure 2c is the RS of solid SD. Besides, in order to verify the necessity of using OTR 202 reinforcement, the SERS of SD was analyzed. Figure 2d shows the SERS spectra of SD with OTR 202. As seen in Figure 2, the positions of SD spectral bands and its intensities were basically consistent with the experiment-detected Raman spectra of SD (Raman shifts < 5 cm −1 ), which indicated that the position of Raman peaks detected by SRES spectra based on OTR 202 were feasible and reliable. Except for the differences between experiment-detected Raman spectra of SD and the DFT-calculated Raman spectra of SD at 812 and 1487 cm −1 , the DFT-calculated Raman spectra of SD were basically similar to the experiment-detected Raman spectra of SD whose Raman shifts (less than 10 cm −1 ) were It can be clearly seen that the average diameter of OTR 202 was about 30 nm. As shown in Figure 1b, the UV/Visible characteristic absorption peaks of OTR 202 was at 533 nm. Beside this, the Raman spectrum of OTR 202 only had a faint signal at 1630 cm −1 (Figure 1c), suggesting that OTR 202 themselves had no strong Raman characteristic peaks and did not have an interferential effect on experimental results. Therefore, the OTR 202 was suitable as SERS substrate to detect SD in this paper.

The SD Molecule and its Assignment of Raman Peaks
The molecular structure of SD (molecular formula: C 22 H 30 N 6 O 4 S) is shown in Figure 2a. Density functional theory (DFT), as a common method for molecular geometry optimization and frequency vibration calculation, can describe the ground state physical properties of atoms and molecules [26]. In this paper, DFT was applied to calculate SD molecule and optimize its structure in Gaussian.v09 software. In this software, the vibrational form of relevant chemical bonds calculated by Hartree-Fock wave function can be obtained [27]. Figure 2b is the RS of SD simulated by DFT, Figure 2c is the RS of solid SD. Besides, in order to verify the necessity of using OTR 202 reinforcement, the SERS of SD was analyzed. Figure 2d shows the SERS spectra of SD with OTR 202. It can be clearly seen that the average diameter of OTR 202 was about 30 nm. As shown in Figure  1b, the UV/Visible characteristic absorption peaks of OTR 202 was at 533 nm. Beside this, the Raman spectrum of OTR 202 only had a faint signal at 1630 cm −1 (Figure 1c), suggesting that OTR 202 themselves had no strong Raman characteristic peaks and did not have an interferential effect on experimental results. Therefore, the OTR 202 was suitable as SERS substrate to detect SD in this paper.

The SD Molecule and its Assignment of Raman Peaks
The molecular structure of SD (molecular formula: C22H30N6O4S) is shown in Figure 2a. Density functional theory (DFT), as a common method for molecular geometry optimization and frequency vibration calculation, can describe the ground state physical properties of atoms and molecules [26]. In this paper, DFT was applied to calculate SD molecule and optimize its structure in Gaussian.v09 software. In this software, the vibrational form of relevant chemical bonds calculated by Hartree-Fock wave function can be obtained [27]. Figure 2b is the RS of SD simulated by DFT, Figure 2c is the RS of solid SD. Besides, in order to verify the necessity of using OTR 202 reinforcement, the SERS of SD was analyzed. Figure 2d shows the SERS spectra of SD with OTR 202. As seen in Figure 2, the positions of SD spectral bands and its intensities were basically consistent with the experiment-detected Raman spectra of SD (Raman shifts < 5 cm −1 ), which indicated that the position of Raman peaks detected by SRES spectra based on OTR 202 were feasible and reliable. Except for the differences between experiment-detected Raman spectra of SD and the DFT-calculated Raman spectra of SD at 812 and 1487 cm −1 , the DFT-calculated Raman spectra of SD were basically similar to the experiment-detected Raman spectra of SD whose Raman shifts (less than 10 cm −1 ) were As seen in Figure 2, the positions of SD spectral bands and its intensities were basically consistent with the experiment-detected Raman spectra of SD (Raman shifts < 5 cm −1 ), which indicated that the position of Raman peaks detected by SRES spectra based on OTR 202 were feasible and reliable. Except for the differences between experiment-detected Raman spectra of SD and the DFT-calculated Raman spectra of SD at 812 and 1487 cm −1 , the DFT-calculated Raman spectra of SD were basically similar to the experiment-detected Raman spectra of SD whose Raman shifts (less than 10 cm −1 ) were within a reasonable range. Combined with related literature [23], the assignments of Raman peaks of SD are listed in Table 1. For the SERS of SD, 552 cm −1 was the carbonyl stretching and phenetole breathing deformable vibration; 647 cm −1 was the carbonyl stretching, phenetole deformable vibration, and C-S stretching in sulfamide, and 831 cm −1 belonged to the pyrazole pyridine stretching; 922 cm −1 was assigned to the C-C deformable vibration and C-H stretching in pyrazole pyridine group, 1235 cm −1 was the C-H stretching vibration in carbonyl, 1305 cm −1 was C-H stretching vibration in ethyl. Further, 1401 cm −1 was the C-H deformable vibration in methyl piperazine.;1488, 1530, and 1584 cm −1 were the C-H deformable vibration in pyrazole pyridine. Table 1. The proposed assignment of Raman peaks of SD.

Limit of Detection of SD
To investigate the sensitivity and stability of the OTR 202 substrates for the detection of SD in cocktail, six different SD concentrations (0, 0.1, 0.5, 2, 5, and 10 mg/L) were collected and the corresponding SERS are shown in Figure 3A.
breathing deformable vibration; 647 cm −1 was the carbonyl stretching, phenetole deformable vibration, and C-S stretching in sulfamide, and 831 cm −1 belonged to the pyrazole pyridine stretching; 922 cm −1 was assigned to the C-C deformable vibration and C-H stretching in pyrazole pyridine group, 1235 cm −1 was the C-H stretching vibration in carbonyl, 1305 cm −1 was C-H stretching vibration in ethyl. Further, 1401 cm −1 was the C-H deformable vibration in methyl piperazine.;1488, 1530, and 1584 cm −1 were the C-H deformable vibration in pyrazole pyridine.

Limit of Detection of SD
To investigate the sensitivity and stability of the OTR 202 substrates for the detection of SD in cocktail, six different SD concentrations (0, 0.1, 0.5, 2, 5, and 10 mg/L) were collected and the corresponding SERS are shown in Figure 3A.
According to Figure 3A,B, the SERS intensity decreased gradually with the decrease of SD concentration from 10 mg/L to 0 mg/L. It was found that the characteristic peaks at 1235, 1401, 1530, and 1584 cm −1 of SD in cocktail were still identified even when the SD solution concentration was as low as 0.1 mg/L. It can be seen that there was faint SERS signal when the SD was 0 mg/L. This might belong to the SERS signal of some substances in cocktails. Generally, the LOD of SD in cocktail reached 0.1 mg/L and 1235, 1401, 1530, and 1584 cm −1 could be qualitatively determined as SD characteristic peaks.  According to Figure 3A,B, the SERS intensity decreased gradually with the decrease of SD concentration from 10 mg/L to 0 mg/L. It was found that the characteristic peaks at 1235, 1401, 1530, and 1584 cm −1 of SD in cocktail were still identified even when the SD solution concentration was as low as 0.1 mg/L. It can be seen that there was faint SERS signal when the SD was 0 mg/L. This might belong to the SERS signal of some substances in cocktails. Generally, the LOD of SD in cocktail reached 0.1 mg/L and 1235, 1401, 1530, and 1584 cm −1 could be qualitatively determined as SD characteristic peaks.

Detection of SD in Cocktail
In this study, to investigate the accuracy and stability of the OTR 202 substrate for the detection of SD in cocktail, SD in cocktail with the concentrations ranging from 10 mg/L to 0.1 mg/L were detected using SERS. The SERS spectra of SD in cocktail were obtained, and the representative SERS spectra are given in Figure 4.

Detection of SD in Cocktail
In this study, to investigate the accuracy and stability of the OTR 202 substrate for the detection of SD in cocktail, SD in cocktail with the concentrations ranging from 10 mg/L to 0.1 mg/L were detected using SERS. The SERS spectra of SD in cocktail were obtained, and the representative SERS spectra are given in Figure 4.

Detection of SD in Cocktail
In this study, to investigate the accuracy and stability of the OTR 202 substrate for the detection of SD in cocktail, SD in cocktail with the concentrations ranging from 10 mg/L to 0.1 mg/L were detected using SERS. The SERS spectra of SD in cocktail were obtained, and the representative SERS spectra are given in Figure 4.   According to Figure 5, SERS spectra of SD in cocktail mixed with the OTR 202 are concentration dependent. The peaks at 1235, 1401, 1530, and 1584 cm −1 could be regarded as a marker band for SD in cocktail determination owing to its drastic intensity change with varying SD concentration. There was a good linear correlation between Raman peak intensity and logarithm of SD concentration in cocktail in each linear regression equation ranged from 0.1 mg/L to 10 mg/L (0.9822 < R 2 < 0.9860), which demonstrated that the SERS can accurately and quantitatively analyze SD in cocktail. To verify the accuracy of this method, first, eight different SD concentration in cocktail (0.3, 0.5, 0.7, 0.9, 3, 5, 7, and 9 mg/L) were prepared and each concentration contained three samples. Second, all the samples were detected by SERS based on OTR 202 of three consecutive days. Third, the linear regression equations at 1235, 1401, 1530, and 1584 cm −1 were used to predict the SD concentration in cocktail. Table 2 presents the precision and accuracy of method for the determination of sildenafil in cocktail. According to Table 2, the intra and inter day relative standard deviation (RSD) were less than 12.7% and 11.8%, respectively. The intra and inter day accuracy% ranged between 93.0%-105.6% and 93.6%-105.8%, respectively. From the results, one can infer that precision and accuracy were within the acceptable limits.

Determination of SD in Cocktail with PLS
Considering that the Raman peaks of SD in cocktail were mainly distributed in the range of 500-1700 cm −1 , the partial least squares (PLS) prediction model was established based on 500-1700 cm −1 SERS spectra. The SERS spectra of SD in cocktail of 155 samples were obtained and then pretreated with Savitzky-Golay (S-G), detrend (DT), standard normal variation (SNV), and 1st-derivative (1st-Der), respectively, and then modeled by PLS. The sample set portioning based on the joint x-y distance (SPXY) [28] method was used to separate the soil samples into calibration set and validation set at the ratio of 2:1. The PLS modeling results based on the 500-1700 cm −1 spectra under different pretreatments are presented in Table 3. The scatter plot between the predicted values and the measured values of the correction set and the prediction set sample are shown in Figure 6. a Rc 2 (the coefficient of determination of the calibration set); Rp 2 (the coefficient of determination of the prediction set); RMSEC (root mean square error of the calibration set); RMSEP (root mean square error of the prediction set).
Molecules 2019, 24, x 7 of 12 pretreatments are presented in Table 3. The scatter plot between the predicted values and the measured values of the correction set and the prediction set sample are shown in Figure 6.   It can be seen that the predictive effect of SD in cocktail was great (0.9896 < Rc 2 < 0.9948, 0.216 < RMSEC < 0.310; 0.9760 < Rp 2 < 0.9856, 0.354 < RMSEP < 0.434). Moreover, the modeling effect was similar after using different pretreatment methods. Among them, the SERS original spectra performed a slightly better modeling effect compared with SNV and 1st-Der. On the one hand, it was shown that the background noise had little effect on the original spectrum and the PLS model with good effect could be established through the original spectrum. On the other hand, although the SNV and 1st-Der eliminated the influence of SRES background noise to some extent, it might weaken the spectral resolution and made it difficult for quantitative analysis.

Experimental Instruments and Reagents
In this study, the experimental instruments included: (1)

Sample Preparation
The process of specific sample preparation was as follows. First, the standard of SD was diluted to 1000 mg/L with methanol. Second, the standard solution of 1000 mg/L was diluted to 0 to 1 mg/L (0.1 mg/L per gradient, 11 concentrations) and 1.2 to 10 mg/L (0.2 mg/L per gradient, 44 concentrations) with cocktail. There were 3 samples for each concentration. A total of 155 samples were prepared.

SERS Measurement
Before Raman spectra acquisition, the instrument should be calibrated using a 785 nm excitation wavelength. The parameters were set as follows: A power of 200 mw, a scanning range of 200 to 3300 cm −1 , an optical resolution of 2 cm −1 , an integration time of 10 s, and an average spectral value of 3 times. The solid SD RS collection was that SD powder was in quartz plate with glass slides flattened and the spectra were acquired with matching microscope platform. When collecting the SERS of samples, 500 µL OTR 202, 100 µL test solution, and 100 µL OTR 103 were added in turn into a 2 mL quartz bottle, then it was placed at a liquid sample pool.

Density Functional Theory (DFT)
Density functional theory (DFT), as a tool for calculating molecular energy and analyzing properties, has been widely used in the field of physics and chemistry. It provides the computational strategies for obtaining information about the energetics, structure, and properties of atoms and molecules [29]. The performance of Becke three-parameter Lee-Yang-Parr (B3LYP) functional in combination with various basis sets has been extensively tested for molecular geometries, vibrational frequencies, ionization energies and electron affinities, dipole and quadrupole moments, atomic charges, infrared intensities, and magnetic properties [27]. Among the various functions and basis sets in DFT, the hybrid functional B3LYP with the 6-31G (d,p) basis set has been commonly used in the Raman spectroscopic calculation of biological molecules [30]. In this paper, B3LYP/6-31G (d,p) was used for the theoretical simulation and calculation of SD molecules.

Spectral Preprocessing Methods
The SERS spectra could be affected by instrument resolution, laser energy, instrument parameters, environmental factors, scattering light on quartz bottle surface, and optical path change. Thus, removing background fluorescence from Raman signals is essential for analyzing Raman signals accurately. Here in the PLS model, each original SERS spectrum was processed by the S-G [31] 5 points smoothing filter, Detrend, SNV, and 1st-Der, respectively. S-G smoothing can reduce the noise introduced by samples, instrument state, surrounding environment, and human operation. The principle of SNV [32] algorithm is that the absorbance values of each wavelength point satisfies a certain distribution in each spectra, and the spectral correction was performed according to this assumption, which can eliminate the influence of scattering light and path change on SERS spectrum of quartz bottle surface in liquid detection. The idea of detrend (DT) [33] algorithm is that the spectral absorbance and wavelength are first fitted into a trend line d according to the polynomial, and then the trend line d is subtracted from the original spectra x to achieve the effect of the trend. 1st-derivation (1st-Der) can distinguish overlapping peaks and eliminate interference from other backgrounds, which improves spectral resolution, sensitivity, and the signal-to-noise ratio of the spectra [34].

Partial Least Squares Model
PLS is a commonly-used calibration model for spectral data analysis, which reflects the relationship between spectra and attribution information due to its flexibility and reliability [35,36]. When PLS is applied to dealing with spectral data, the spectral matrix is decomposed first and the main principal component variables are obtained, then the contribution of each principal component is calculated. The flexibility of PLS makes it able to interpret the dependent and independent variables well by establishing regression models. In this study, the PLS model was established with the SERS spectral data as X and the content of SD as Y, whose best principal factor was determined by the root mean square error of cross validation (RMSECV). In addition, all above-mentioned data analysis in this study were performed on OMNIC v8.2 (Thermo Nicolet Corp., Madison, WI, USA), MATLAB R2014a (The MathWorks, Inc., Natick, MA, USA), and Gaussian.v09 (Gaussian, Inc., Wallingford, CT, USA).

Model Evaluation Index
In this experiment, the modeling effect was evaluated by the coefficient of determination (R 2 ), the root mean square error (RMSE), relative standard deviation (RSD), and recovery rate. The coefficient of determination R 2 reflects the level of intimacy between variables and the RMSE reflects the model accuracy. The lower the RMSE, and the closer the R 2 is to 1, the better the performance of the prediction model. In this study, Rc 2 and Rp 2 represent the coefficient of the determination of the calibration set and the prediction set respectively, while RMSEC and RMSEP represent the root mean square error of the calibration set and the prediction set, respectively [37,38]. In addition, relative standard deviation (RSD) reflects the degree of discretization between individuals in the reflection group and recovery rate reflects the degree of coincidence between the results of the reaction and the true value. The recovery rate ranges closer to 100%, and the lower the RSD, the better the reliability of the model [39].

Conclusions
In this paper, we reported the rapid and quantitative determination of SD in cocktail based on SERS with OTR 202. According to the results, we found that there was a good linear correlation between Raman peak intensity at 1235, 1401, 1530, and 1584 cm −1 and logarithm of SD concentration in cocktail with R 2 from 0.9822 to 0.9860 in the range of 0.1-10 mg/L and the LOD could reach 0.1 mg/L. Also, the determination coefficient (Rp 2 ) for SD in cocktail in PLS model was great (0.9760 < Rp 2 < 0.9856). It was indicated that the rapid detection of SD by SERS was feasible and reliable. Overall, the SERS method with OTR 202 enhancement developed through this study provide a novel, rapid, and accurate approach to quantitatively determine SD in cocktail, which could meet the requirements of analysis and detection of SD in other alcoholic beverages.