Optimization Studies
To optimize the TFME approach for identifying 18 metabolites present in the saliva samples, the type of column employed in the chromatographic separation was selected, the type of internal standard used in the quantification process was determined, and a range of mobile phase conditions was evaluated. Because the metabolites analyzed show polar properties, separation was performed using columns with high polarity. For this purpose, the separation efficiencies of the C8, HILIC, and Inertsil SIL 100 columns were compared. The experiments were performed in standard mix solutions with a concentration of 500 ng mL− 1. After testing the columns under reversed-phase and normal-phase conditions, 18 metabolites were selectively separated on the Inertsil 100 column with high peak resolution.
Different compositions of mobile phase solution in isocratic and gradient conditions were also tested during separation on the column. The results reveal that, the isocratic operation gave better results. Selection of the internal standard is also another critical point and for this purpose, Ornidazole Sildenophyll Citrate, Depoxetine Hydrochloride, and Tadalafil which are not present in urine were tested for use as internal standards. Mixtures of known concentrations of the targeted analytes were prepared in triplicate in the presence of 1 µg mL− 1 of each internal standard. Using the Ornidazole internal standard resulted in the acquisition of more accurate linear responses across a wider range of concentrations for all targeted compounds. Supplemental Fig. 1 shows the ion chromatograms of the retention times of each analyte with Ornidazole.
To optimize the extraction parameters of the TFME method, including pH and type of desorption solutions, an experimental design was constructed with a two-factor mixed level (two levels for the type of the desorption solution and six for the pH). Table 1 shows the factor levels and responses.
Table 1
Actual levels of the factors and the seven responses.
Desorption Solution Type | | pH | | Proline | | Histidine | | Phenylalanine | | Tyrosine | | Betaine | | Creatine | | 2-Methoxy benzoic acid |
1 | | 3 | | 0.26 | | 2.87 | | 0.13 | | 0.35 | | 2.44 | | 16.34 | | 0.64 |
2 | | 3 | | 0.20 | | 4.83 | | 0.06 | | 0.47 | | 3.57 | | 16.16 | | 1.46 |
1 | | 4 | | 0.32 | | 2.31 | | 0.13 | | 0.23 | | 3.08 | | 9.44 | | 0.49 |
2 | | 4 | | 0.43 | | 10.49 | | 0.30 | | 1.58 | | 6.58 | | 22.41 | | 0.66 |
1 | | 5 | | 0.67 | | 0.53 | | 0.36 | | 1.84 | | 3.51 | | 21.20 | | 1.37 |
2 | | 5 | | 5.87 | | 43.23 | | 2.66 | | 8.07 | | 7.27 | | 65.50 | | 3.29 |
1 | | 6 | | 2.11 | | 7.39 | | 1.54 | | 5.68 | | 3.88 | | 33.22 | | 1.11 |
2 | | 6 | | 5.85 | | 41.64 | | 3.43 | | 2.21 | | 10.87 | | 123.27 | | 1.70 |
1 | | 7 | | 7.45 | | 5.70 | | 1.81 | | 2.53 | | 3.18 | | 70.26 | | 1.27 |
2 | | 7 | | 17.47 | | 56.87 | | 4.00 | | 15.10 | | 11.61 | | 157.59 | | 3.01 |
1 | | 9 | | 7.53 | | 0.00 | | 1.25 | | 8.69 | | 3.94 | | 65.30 | | 6.89 |
2 | | 9 | | 14.56 | | 101.14 | | 2.79 | | 38.21 | | 16.28 | | 174.40 | | 4.99 |
Triplicate determination has been done on each desorption type and pH setting in Table 1 for the 18 metabolites and the average of these triplicate measurements was used as the corresponding response. Among these 18 metabolites, only seven of them produced measurable responses in all these 36 runs, and therefore response surface regression models were constructed for these 7 metabolites. pH of the solution was varied between 3 to 9 in the experiments. The desorption solution type 1 was water containing 0.5% formic acid and type 2 was an aqueous solution of 5.0 mM ammonium formate + 0.5% formic acid and 5.0% acetonitrile.
A pseudo-second-order polynomial model was selected in which only a linear term was chosen for the desorption type because it has two levels. The same model was used for all seven responses. Eq. 1 represents the general regression model.
y = b 0 + b1 dt + b2 pH + b22 pH pH + b12 dt pH + e Eq. 1 (dt: Desorption type)
Table 2 shows the regression coefficients and corresponding p-values along with the R2 and adjusted R2.
Table 2
The regression coefficients and corresponding p-values along with the R2 and adjusted R2
| Proline | Histidine | Phenylalanine | L-Tyrosine | Betaine | Creatine | 2-Methoxy benzoic acid |
Term | Coef. | p-val. | Coef. | p-val. | Coef. | p-val. | Coef. | p-val. | Coef. | p-val. | Coef. | p-val. | Coef. | p-val. |
b0 | 6.282 | 0.002 | 25.0032 | < 0.001 | 2.27978 | <0.001 | 5.33651 | 0.025 | 6.83608 | < 0.001 | 76.3982 | < 0.001 | 1.73514 | 0.006 |
dt | 2.416 | 0.032 | 22.5887 | < 0.001 | 0.71931 | 0.009 | 4.62516 | 0.009 | 3.32274 | < 0.001 | 31.9121 | 0.001 | 0.22539 | 0.481 |
pH | 2.200 | 0.002 | 7.84 | 0.754 | 0.42025 | 0.004 | 3.65689 | 0.001 | 1.12949 | < 0.001 | 20.516 | < 0.001 | 0.76311 | 0.001 |
pH*pH | -0.080 | 0.747 | 0.1734 | < 0.001 | -0.15037 | 0.024 | 0.74068 | 0.065 | -0.02692 | 0.575 | -1.2416 | 0.402 | 0.18971 | 0.049 |
dt*pH | 0.740 | 0.145 | 7.9171 | < 0.001 | 0.15139 | 0.171 | 2.29857 | 0.009 | 0.92919 | < 0.001 | 9.8454 | 0.007 | -0.16151 | 0.321 |
R2 | 0.823 | 0.9695 | 0.8514 | 0.894 | 0.9883 | 0.9375 | 0.8253 |
Adj. R2 | 0.7218 | 0.9521 | 0.7665 | 0.8335 | 0.9817 | 0.9018 | 0.7255 |
Table 2 represents the effects of both type desorption solution and pH on the extraction efficiency of the metabolites. These results show that the effect of pH is generally significant with p values less than 0.05 at the 95% confidence level. The interaction terms pH square and dt pH of Proline were not statistically significant, with values of 0.747 and 0.145. Only the pH term with a value of 0.754 remained at a high value for Histidine. The dt pH interaction value (0.171) for Phenylalanine was above the confidence limit. In the results for tyrosine betaine and creatine, the square of pH was above the required p-value. The dt and dt pH interaction values of 2-methoxybenzoic acid (2-MBA) was also not significant with p-values higher than 0.05.
Figure 1 illustrates the counter plots of the two factors as functions of responses along with the overlaid counter plot of all seven responses in a single curve. As can be seen from the counter plots, only for 2-MBA, desorption solution type 1 is better suited for higher desorption efficiency at pH 9. In the remaining six responses, the highest desorption yields were obtained with desorption solution type 2 and a pH of 9. In addition, the overlaid counter plot of all seven responses indicates that when the acceptable range of each response was defined as given on the right legend of the graph, a suitable extraction yield was obtained around pH 8–9 and desorption type 2.
Analytical characteristics of the proposed method
The analytical figure of merits including linearity, sensitivity, precision, and accuracy of the optimized method were evaluated. The linearity of the calibration plots for the metabolites was tested by analyzing a series of standard mix solutions at concentrations between 0.025 and 4.0 µg/mL in the presence of internal standard. The results demonstrated a robust regression for all analytes, with correlation coefficients (R2) ranging from 0.9975 to 0.9841. The limit of detection (LOD) and quantification (LOQ) values ranged from 0.014 to 0.97 µg mL− 1 and 0.046 to 3.20 µgmL− 1, respectively. The precision of the method was evaluated by determining the percent relative standard deviation (%RSD) value from the replicate analysis of the lowest concentration level through the application of the method. The %RSD values for all analytes were within the acceptable range (less than 20%) for the proposed method. Saliva samples obtained from controls were used in recovery studies. A standard mix solution was spiked to the saliva sample at concentrations of 1 µg mL− 1, 2 µg mL− 1, and 3 µg mL− 1, and a blank sample study was also performed in triplicate analysis. The analytical performance values are presented in Table 3.
Table 3
Analytical Characteristics of the Method
Metabolite Name | LOD (ug ml− 1) | LOQ (ug ml− 1) | RSD (%) (n = 3) | Calibration Equation Regression Coefficient | Recovery Values Spiked concentrations |
1µgmL− 1 | 2µgmL− 1 | 3µgmL− 1 |
L-Alanine | 0.20 | 0.66 | 4.9–11 | y = 0.4469x + 0.0479 R² = 0.9975 | 100.1 ± 5 | 100.9 ± 2 | 100 ± 0.7 |
L-Proline | 0.05 | 0.17 | 1.2–17 | y = 4.33x – 1.0282 R² = 0.9961 | 107.7 ± 20 | 102.9 ± 10 | 100.3 ± 10 |
L-Valine | 0.02 | 0.07 | 1.7–15 | y = 1.0714x − 0.6684 R² = 0.9936 | 104.8 ± 10 | 104.8 ± 20 | 100.4 ± 10 |
L-Histidine | 0.05 | 0.17 | 8.5–15 | y = 1.0739x − 0.7284 R² = 0.9940 | 100 ± 0.8 | 100.6 ± 2 | 100 ± 1 |
L- Phenylalanine | 0.05 | 0.17 | 3.4–15 | y = 0.8865x − 0.4104 R² = 0.9911 | 109.1 ± 20 | 100.5 ± 5 | 100.7 ± 6 |
L- Tyrosine | 0.18 | 0.59 | 5.5–12 | y = 1.0152x − 0.0522 R² = 0.9966 | 100 ± 7 | 100.6 ± 5 | 100.4 ± 3 |
L-Tryptophane | 0.04 | 0.13 | 9.2–16 | y = 3.7974x – 1.2578 R² = 0.9893 | 100.5 ± 8 | 100.6 ± 10 | 100.7 ± 1 |
Betaine | 0.02 | 0.07 | 2.9–8.4 | y = 20.389x + 1.4813 R² = 0.9971 | 100 ± 13 | 100.8 ± 7 | 100 ± 2 |
Taurine | 0.07 | 0.23 | 0.3–11 | y = 0.0863x + 0.0776 R² = 0.9945 | 107.9 ± 20 | 100 ± 0.2 | 100.1 ± 3 |
Hypoxanthine | 0.18 | 0.59 | 0.5–10 | y = 0.4456x − 0.0816 R² = 0.9930 | 100.7 ± 40 | 100 ± 15 | 102.1 ± 10 |
Pipecolic acid | 0.014 | 0.046 | 3.8–10 | y = 0.9086x − 0.3773 R² = 0.9975 | 100.9 ± 8 | 101.3 ± 10 | 100.7 ± 6 |
Creatine | 0.06 | 0.20 | 1.6–13 | y = 5.5529x + 0.691 R² = 0.9968 | 102.3 ± 30 | 100.7 ± 7 | 100 ± 0.6 |
2-Methoxy benzoic acid | 0.02 | 0.07 | 8.1–16 | y = 24.476x + 7.2921 R² = 0.9926 | 100.1 ± 8 | 102.1 ± 10 | 100.4 ± 6 |
Vaniline | 0.03 | 0.10 | 0.5–10.8 | y = 36.433x + 11.95 R² = 0.9927 | 100.2 ± 2 | 100.2 ± 9 | 100 ± 4 |
Carnitine | 0.18 | 0.59 | 1.5–7.2 | y = 0.7359x − 0.0375 R² = 0.9975 | 107.3 ± 30 | 105.5 ± 20 | 109.3 ± 20 |
Taurocholic acid | 0.97 | 3.20 | 9.8–17.4 | y = 0.0116x − 0.0087 R² = 0.9841 | 102.7 ± 20 | 100.2 ± 10 | 103.9 ± 10 |
Hippuric acid | 0.06 | 0.20 | 12–17 | y = 8.8765x – 19.909 R² = 0.9843 | 100.4 ± 2 | 100.6 ± 2 | 100.8 ± 5 |
Nicotinamide | 0.06 | 0.20 | 0.7–9.5 | y = 5.4016x – 1.4146 R² = 0.9962 | 100.2 ± 7 | 100.5 ± 8 | 100.5 ± 3 |
Metabolic pathway analysis
To calculate the impact of pathway topology and class enrichment, both metabolic set enrichment and metabolite topology analysis were performed using both the Small Molecule Pathway Database (SMPDB) (Fig. 2a-2b) and Kyoto Encyclopedia of Genes and Genomes (KEGG) (Fig. 3a-3b) for a better understanding of metabolic alteration among cancer patients. In both analyses, the phenylalanine and purine metabolism showed some modifications. Node color and size are based on the p-value and pathway impact value.
Metabolic pathway analysis results show that for both the SMPDB and KEGG databases, the phenylalanine, tyrosine, purine, and histidine pathways are of great importance for lung cancer patients. For the SMPDB database, the methionine pathway is also found as an important metabolite (Fig. 3).
The detailed pathways that show modifications are shown in Fig. 4. Furthermore, the heatmap of metabolites enriched in metabolic pathways in cancer patients (n = 40) compared with healthy metabolites (n = 38), with corresponding metabolisms (right) (Fig. 5).
The concentration alterations of metabolites revealed that phenylalanine, histidine, and hypoxanthine were upregulated in lung cancer patients. On the other hand, Tyrosine was upregulated in lung cancer patients (Fig. 4).
Figure 5 shows the Heat map analysis of significantly altered and biologically relevant metabolites found in cancer patients (1.1–1.40) and healthy samples (2.1–2.38). Blue and red colors refer to significantly downregulated and upregulated metabolites relative to the mean expression level within each group, respectively. The heat map results show the relative alteration of the concentration of 18 different metabolites. The results reveal that purine metabolism, primary bile acid biosynthesis, glycine, serine, threonine metabolism, and phenylalanine are of great importance for lung cancer patients (Fig. 5).
Biomarker discovery
MetaboAnalyst 5.0 was used for identifying and discovering the features of potential biomarkers using the receiver operating characteristic (ROC) curve-based approach. AUC was selected as AUC > 0.79. The ROC curves and related results are shown in Fig. 6.
ROC curve analysis was performed to calculate AUC at 95% confidence intervals. Figure 6 describes the biomarkers identified using the normalized data with their corresponding p-values < 0.05 from the t-tests and AUC > 0.79 with their corresponding box plots and ROC curves. The results reveal that proline (p-value = 7.3371E-6, AUC = 0.946), hypoxanthine (p-value = 1.1161E-13, AUC = 0.933), phenylalanine (p-value = 8.449E-13, AUC = 0.905), valine (p-value = 0.87171, AUC = 0.876), and alanine (p-value = 3.8555E-6, AUC = 0.799) can serve as potential biomarkers for cancer as they appeared significantly increased compared with the control group.
Upon comparison of these data with those from studies using salivary biomarkers for the diagnosis of LC, it was found that the area under the curve (AUC) values for the biomarkers in this study were the greatest for distinguishing the control group from the LC patient cohort. Takamori et al. identified salivary biomarkers for distinguishing LC (41 patients) from benign lung lesions (21 controls) and found that tryptophan, diethanolamine, cytosine, lysine, and tyrosine metabolites exhibited the greatest discriminatory ability, with a maximum value of 0.729 [25]. A comparable study has also been conducted for LC [35]. in the context of control of 71 and LC of 109 cases. Salivary metabolites were evaluated using a machine learning method with an area under the receiver operating characteristic curve of 0.744. In another study, patients with LC were discriminated from the healthy group with a sensitivity of 97% and specificity of 92% using the panel of feature metabolites [24]. AUC values of biomarkers were calculated slightly higher than other studies due to the efficiency of the new extraction method developed in the study.
In this study, phenylalanine metabolism and purine metabolism metabolites (i.e hypoxanthine) were found to change in abundance in the cancer samples for both SMPDB and KEGG database results. Phenylalanine is an essential amino acid that is the precursor of dopamine and phenylethylamine, and phenylalanine metabolism regulates T-cell proliferation and activation [36, 37]. Furthermore, Yang and coworkers demonstrated that phenylalanine metabolism is crucial for T-cell immune response suppression [38]. In this study, the other important metabolism is purine metabolism. Because purine metabolites and enzymes are critical for tumor cell proliferation and uncontrolled cell proliferation is a hallmark of cancer, targeting the cell cycle is of great importance for cancer therapy strategy [39]. Purines play a key role in modulating the immune cell response and releasing cytokines [39]. In addition, purine metabolism regulates DNA repair and therapy in cancer cells [40]. The proposed TFME method was also demonstrated that Proline was a promising biomarker in saliva metabolites for lung cancer diagnosis.