3.1 The feature of subjects
The age for 30 IAI patients is 58.87 ± 24.45 (mean ± standard deviation), while age for 20 healthy volunteers is 53.05 ± 15.65. IAI group consists of 10 women and 20 men, while healthy group is composed of 10 women and 10 men. The causes of IAI cases include 4 stomach ulcer, 2 duodenum ulcer, 2 small intestinal trauma, 2 colon diverticular perforation, 6 acute pure appendix, 5 acute suppurative appendix, 4 gangrene perforation appendix, 3 acute cholecystitis and 2 acute pancreatitis.
3.2 Serum sample
The TIC of IAI patients (Fig. 1A, C) and healthy volunteers (Fig. 1B, D) is analyzed. TIC pictures show the difference in peak intensities of many metabolites be-tween two groups in positive model and negative model, suggesting the existence of differential metabolites. We perform unsupervised PCA for serum samples from patients, volunteers and quality controls in positive (Fig. 1E) and negative model (Fig. 1F). In PCA, every plot represents a sample. The larger difference of the metabolites in samples is, the greater distance between plots is, meanwhile the smaller difference, the closer distance. In PCA, the distance among samples within each group is close, suggesting the difference of each group is small and the distance among three groups is great, indicating the difference between groups is large. In addition, the aggregation of QC is good, proving the credibility of experimental data. Comparing IAI patients with volunteers by t test (p < 0.05), we find great difference for 431 metabolites in positive mode and 216 ones in negative mode (supplementary sheet1). We employ PLS-DA for further sifting and identify 50 metabolites (VIP > 1) in positive and negative models. (Fig. 2, B) (supplementary sheet2). The PLS-DA plot reveals that two groups could be separated clearly. (Fig. 2, A)
According to 16 metabolites, hierarchical clustering analysis is used to access the similarity of differential metabolites (Dendritic structure on picture’s left) and sample (Dendritic structure above picture). (Fig. 2, C) The color intensity represents the content of metabolites in samples and red means the most content. This picture shows that 16 metabolites have enough ability to divide 50 samples into infect group and health group.
SVM is performed to filter further and we identify 20 metabolites with 100% weight in positive (11) and negative (9) models. Based on the HMDB (http://www.hmdb.ca), we identify 16 metabolites meeting our requirement, including Myristic acid, PGP(18:0/18:3(6Z,9Z,12Z)), TG(14:1(9Z)/16:1(9Z)/20:3n6), Decanoylcholine, MG(i-15:0/0:0/0:0), TG(14:0/i-16:0/i-12:0), DG(22:5n6/0:0/22:5n6), PS(18:0/20:0), PE(16:0/20:2(11Z:14Z)), FAHFA(18:1(9Z)/6-O-18:0), 8,9 Epoxyeicosatrienoic acid, 5a-Dihydrotestosterone sulfate, LysoPC(0:0/16:0), LysoPC(0:0/18:2(9Z,12Z)), Plate-let-activating factor, PS(22:1(13Z)/15:0). (Table1) Based on 16 metabolites, we utilized fisher distinguish analysis to classify 28/30 IAI into IAI and 20/20 health people into health people, therefore it has sensitivity of 93.3% and specificity of 100%. (Table2)
Table 1
16 metabolites identified that differentiate IAI and health people
Ionization | mass-to-charge ratio | Chemical formula | Metabolite |
ESI+ | 246.2441 | C14H28O2 | Myristic acid |
ESI+ | 285.1718 | C44H72O15P2 | PGP(18:0/18:3(6Z,9Z,12Z)) |
ESI+ | 301.2031 | C53H92O6 | TG(14:1(9Z)/16:1(9Z)/20:3n6) |
ESI+ | 318.3017 | C15H32NO2+ | Decanoylcholine |
ESI+ | 334.2961 | C18H36O4 | MG(i-15:0/0:0/0:0) |
ESI+ | 362.3278 | C49H94O6 | TG(14:0/i-16:0/i-12:0) |
ESI+ | 758.5726 | C47H72O5 | DG(22:5n6/0:0/22:5n6) |
ESI+ | 784.587 | C44H80NO10P | PS(18:0/20:0) |
ESI+ | 806.5728 | C41H78NO8P | PE(16:0/20:2(11Z:14Z)) |
ESI- | 281.2498 | C36H68O4 | FAHFA(18:1(9Z)/6-O-18:0) |
ESI- | 339.2343 | C20H32O3 | 8,9-Epoxyeicosatrienoic acid |
ESI- | 369.1748 | C19H30O5S | 5a-Dihydrotestosterone sulfate |
ESI- | 540.3326 | C24H50NO7P | LysoPC(0:0/16:0) |
ESI- | 564.3316 | C26H50NO7P | LysoPC(0:0/18:2(9Z,12Z)) |
ESI- | 568.3637 | C26H54NO7P | Platelet-activating factor |
ESI- | 802.5634 | C43H82NO10P | PS(22:1(13Z)/15:0) |
Table 2
Performance of 16 identified metabolites in differentiating IAI and healthy people
predict outcome | actual conditions | | | | |
| IAI | health | summation | sensitivity | specificity |
IAI | 28 | 2 | 30 | | |
health | 0 | 20 | 20 | 93.30% | 100% |
summation | 28 | 22 | | | |
ROCs involving 16 metabolites are drawn and we choose 9 metabolites with AUC > 0.85 or AUC < 0.15, including, PGP (18:0/18:3(6Z,9Z,12Z)), 8,9-Epoxyeicosatrienoic acid, 5a-Dihydrotestosterone sulfate, LysoPC(0:0/16:0), LysoPC(0:0/18:2(9Z,12Z)), Platelet activating factor (Fig. 3,A) (Table3). PCT, a biomarker for systematic inflammation, is a protein produced by C cells of the thyroid gland and a precursor of calcitonin. We collect the PCT results of 50 subjects and make the ROC whose AUC is 0.89, while the maximum AUC among 16 metabolite is 0.96 (LysoPC(0:0/18:2(9Z,12Z))) that possesses better predicting performance. (Fig. 3, C)
Table 3
the AUC of 16 metabolites and PCT between two groups
Variable | AUC | standard error | sig | 95% confidence intervals |
| | | | lower | upper |
Myristic acid | 0.203 | 0.062 | 0.000 | 0.082 | 0.325 |
MG(i-15:0/0:0/0:0) | 0.838 | 0.055 | 0.000 | 0.730 | 0.947 |
PGP(18:0/18:3(6Z,9Z,12Z)) | 0.092 | 0.043 | 0.000 | 0.007 | 0.176 |
Decanoylcholine | 0.238 | 0.069 | 0.002 | 0.103 | 0.373 |
TG(14:1(9Z)/16:1(9Z)/20:3n6) | 0.798 | 0.068 | 0.000 | 0.666 | 0.931 |
TG(14:0/i-16:0/i-12:0) | 0.195 | 0.062 | 0.000 | 0.074 | 0.316 |
PS(18:0/20:0) | 0.242 | 0.068 | 0.002 | 0.109 | 0.375 |
PE(16:0/20:2(11Z:14Z)) | 0.785 | 0.064 | 0.001 | 0.660 | 0.910 |
FAHFA(18:1(9Z)/6-O-18:0) | 0.798 | 0.068 | 0.000 | 0.664 | 0.933 |
8,9-Epoxyeicosatrienoic acid | 0.127 | 0.060 | 0.000 | 0.008 | 0.245 |
5a-Dihydrotestosterone sulfate | 0.063 | 0.032 | 0.000 | 0.000 | 0.127 |
LysoPC(0:0/16:0) | 0.077 | 0.037 | 0.000 | 0.004 | 0.149 |
LysoPC(0:0/18:2(9Z,12Z)) | 0.040 | 0.023 | 0.000 | 0.000 | 0.085 |
Platelet activating factor | 0.083 | 0.040 | 0.000 | 0.006 | 0.161 |
DG(22:5n6/0:0/22:5n6) | 0.177 | 0.067 | 0.000 | 0.046 | 0.307 |
PS(22:1(13Z)/15:0) | 0.270 | 0.074 | 0.006 | 0.125 | 0.415 |
PCT | 0.890 | 0.053 | 0.000 | 0.786 | 0.994 |
3.3 Urine sample
The calculate procedure of urine samples is same as serum ones. TIC suggests content difference of some metabolites in urine sample at same retention time and different ion model. (Fig. 4, A, B, C, D) In PCA, we observe that difference within group is small, meanwhile variance between groups is large. The position of QC is consistent, revealing that test data is believable. (Fig. 4E, F) After t test, we select significant difference for 269 metabolites in positive model and 409 ones in negative model. (supplementary sheet3). We utilize PLS-DA for further screening and choose 63 metabolites whose VIP > 1 in positive and negative mode (supplementary sheet4). (Fig. 5, B) In PLS-DA plot, two groups overlap partly, but they could be divided generally. (Fig. 5, A) The hierarchical clustering analysis exhibits 11 metabolites’ discrimination is excel-lent. (Fig. 5, C)
SVM is applied to select 12 metabolites whose weight accesses to 100%. Then, we im-port mass-to-charge ratio of the metabolites into HMDB and 11 metabolites meet requirements, including Tryptamine, Etiocholanolone, Deoxycholic acid 3-glucuronide, Margaroylglycine, Norepinephrine, Etiocholanolone glucuronide, 4-Hydroxyproline, Dopamine glucuronide, Riboflavin, Pregnanetriol, 5-alpha-Dihydrotestosterone glucuronide (Table4). Based on 11 metabolites, we utilize SVMDA to distinguish samples and the result indicates that they have specificity of 85%, sensitivity of 86.7%. Finally, we protract ROC about 11 metabolites and metabolites with AUC > 0.9 or AUC < 0.1 are chosen (Fig. 3, B), including Deoxycholic acid 3-glucuronide and Margaroylglycine (Table5). The mean Plots of 8 differential metabolites is made to observe the change trend of metabolites’ content between two groups. (Fig. 6)
Table 4
Metabolites identified that differentiate IAI and health people
Ionization | mass-to-charge ratio | Chemical formula | Metabolite |
ESI+ | 221.1656 | C10H12N2 | Tryptamine |
ESI+ | 273.2219 | C19H30O2 | Etiocholanolone |
ESI+ | 285.1716 | C30H48O10 | Deoxycholic acid 3-glucuronide |
ESI+ | 334.296 | C19H37NO3 | Margaroylglycine |
ESI+ | 152.0709 | C8H11NO3 | Norepinephrine |
ESI+ | 484.2907 | C25H38O8 | Etiocholanolone glucuronide |
ESI- | 150.0567 | C5H9NO3 | 4-Hydroxyproline |
ESI- | 350.0884 | C14H19NO8 | Dopamine glucuronide |
ESI- | 375.1307 | C17H50N4O6 | Riboflavin |
ESI- | 449.2552 | C21H36O3 | Pregnanetriol |
ESI- | 931.5049 | C25H38O8 | 5-alpha-Dihydrotestosterone glucuronide |
Table 5
the AUC of 11 metabolites and PCT between two groups
Variable | AUC | standard error | sig | 95% confidence intervals |
| | | | lower | upper |
4-Hydroxyproline | 0.725 | 0.081 | 0.005 | 0.567 | 0.883 |
Riboflavin | 0.195 | 0.073 | 0 | 0.052 | 0.338 |
Pregnanetriol | 0.155 | 0.069 | 0 | 0.019 | 0.291 |
5-alpha-Dihydrotestosterone glucuronide | 0.19 | 0.069 | 0 | 0.054 | 0.326 |
Dopamine glucuronide | 0.2 | 0.065 | 0 | 0.073 | 0.327 |
Tryptamine | 0.817 | 0.06 | 0 | 0.699 | 0.934 |
Etiocholanolone | 0.127 | 0.059 | 0 | 0.012 | 0.241 |
Deoxycholic acid 3-glucuronide | 0.968 | 0.02 | 0 | 0.929 | 1.008 |
Margaroylglycine | 0.007 | 0.007 | 0 | -0.007 | 0.021 |
Norepinephrine | 0.706 | 0.076 | 0.007 | 0.557 | 0.855 |
Etiocholanolone glucuronide | 0.155 | 0.067 | 0 | 0.023 | 0.287 |
PCT | 0.890 | 0.053 | 0.000 | 0.786 | 0.994 |