br Significant associations for pre operative protein with
Significant associations for pre-operative protein with clinical char-acteristics were found between loss of heterogeneity of CDH and SMAD4 (p = 0·0005) and between GCNT1 and TNM stage (p = 0·0072).
3.6. Alteration of serum protein levels before and after surgery
To check whether protein abundances can be affected by tumour re-section, levels of proteins before operation and one to two weeks after were compared. The univariate analysis with FDR correction for multiple tests showed that 184 of 316 (58·2%) proteins were significantly changed after surgery (88 decreased and 96 increased, Supplementary Table 5). Volcano plots in Fig. 2G-H display the most significantly altered proteins in the comparisons between pre and post groups and between control and post. The biological enrichment analysis by FunRich revealed that the proteins decreased after surgery are predominantly involved in cell adhesion, whereas the proteins increased are tend to be associated with Norfloxacin (Supplementary Fig. 8). Notably, the changes for the protein levels after operation were not consistent in all the patients, but following the same trend in most patients as illustrated for each protein and each patient in Supplementary Fig. 9. We then eval-uated the clinical associations of proteins differentially changed after sur-gery, and found associations between CYR61 with LOH of CDH, between GCNT1 with Lauren classification, and proteins associated with tumour site including ERBB2/3/4, GPNMB, ITGAV, ITGB5, LYPD3, NTRK2/3, SEZ6L, XPNPEP2, and TNFRSF19 (Supplementary Table 6).
3.7. Optimal proteins combination for distinguishing gastric cancer patients from controls
Multivariate analysis ENLR was performed to further select promis-ing serum biomarkers and generate a diagnostic model for gastric can-cer. With cross-validation, the optimal α = 0.4 was chosen for penalty proportion for the model according to the lowest misclassification error and the best accuracy (Supplementary Table 7). After ten-fold cross-validation, 27 proteins were retained to have all non-zero coeffi-cients at each cross-validation step (Fig. 3A). By comparing the ROC curves of different combinations of the proteins that ranked from the largest to the smallest absolute regression coefficients, the combination
A
Serum
Tissue
Serum
Detectable
Detectable
Tissue
Undetectable
Undetectable
C Tumor tissue vs. Normal tissue
N
T
E Serum_Pre vs. Post vs. Ctrl
Ctrl
Pre
Post
Serum_Pre vs. Post
SCF
ITGAV
Pvalues)
LRRN1
LEP
NTRK2
SYND1
LYPD3
CAV1
IGF1
MMP1
TRANCE CD6
HMOX1
adj
CTSV
TNFRSF6B
ICOSLG
ENRAGE
(FDR
OMG
FGFBP1
CEACAM5
ESM1
TXLNA
MK
log2 Fold Change (Pre/Post)
B
D Tumor tissue vs. Normal tissue
CPXM1
WISP1
MCP3
SPARC
ESM1
VEGFD
CCL4CXCL6
IL8
OSM
SULT2A1
h
CXCL1MCP2
IL6 IL1 alpha
DNER
MMP1
P
CFC1
ERBB4
TNFRSF6B
MICA/B
OPG
MSLN
DNAJB1
Serum_Pre vs. Ctrl
MMP1
CA9
MCP3
P
MSLN
CEACAM5
AR
ENRAGE
DCTN2
IL8
log2 Fold Change (Pre/Ctrl)
H Serum_Ctrl vs. Post
NTRK2
ENRAGE MMP1
NTRK3
SCF
ITGAV
SYND1
DNER
ITGA1
IL7
MCP3
adj
SIGLEC10
DNAJB1
TRANCE CTSV
CXCL1
CCL20
MK
LRRN1
IGF1
ICOSLG
AR
log2 Fold Change (Post/Ctrl)
Table 2
Clinical significance of proteins expressed in gastric cancer tumour tissue.
Variables
r>
Protein
Padj
MSI status
BOC
0·0403