br Fig Multiplex PEA results of proteins measured in
Fig. 2. Multiplex PEA results of Fasudil measured in GC tissues and serum specimen. (A) Venn diagram showing the number of proteins detectable or undetectable in serum and in tissue specimen. (B) Comparison of subcellular location between proteins detectable in serum but not in tissue and proteins detectable in tissue but not in serum by the FunRich software. The 13 proteins detectable in GC tissue but not in serum are AGR3, ARTN, CAMKK1, IL1A, IL20RA, IL22RA1, IL24, IL33, JUN, LIF, NCLN, NRTN, PAK4. The 14 proteins detectable in GC serum but not in tissue are CRX, DKKL1, FAM19A5, FCRLB, FGF23, IL10, IL10RA, IL2, LYPD1, OPTC, SEZ6L, SLITRK2, TCL1B, WNT9A. (C) Principal component analysis (PCA) plot illustrating the sample distribution of 100 gastric tumour tissues (T, blue) and matched adjacent noncancerous tissues (N, red), based on 245 proteins levels. Each dot represents an individual sample.
(D) Volcano plot showing the 245 proteins levels in GC tissues compared to matched non-tumour tissues. The dashed line represents the cutoff line with indicated significance criteria. Points having absolute log fold-change ≥2 and FDR adjusted p-value b0·05 are shown in red, with absolute log fold-change b2 and p-value ≥0·05 are in gray, and the rest are in black.
(E) PCA plot illustrating the distribution of 50 serum samples from controls (Ctrl, red), 100 GC preoperative serum samples (Pre, green) and matched 100 postoperative samples (Post,
blue), based on 316 proteins levels. (F\\H) Volcano plots showing the 316 protein levels in preoperative GC serum samples versus controls (F), between preoperative and postoperative ones (G), and between postoperative samples and controls (H). Points having absolute log fold-change ≥0·5 and FDR adjusted p-value b0·05 are shown in red, with absolute log fold-change b0·5 and p-value b0·05 are in black, with absolute log fold-change b0·5 and p-value ≥0·05 are in gray, and the rest are in orange.
Protein Coef. Combined AUC Sen Spe ROC test
CCL20 ADAMTS15 FLI1
IGF1 APBB1IP KAZALD1
MSLN CEACAM5 TGFA SCFKITLG DDAH1 ZBTB17
Fig. 3. Diagnostic capacity for gastric cancer of the identified 19 serum protein signature by elastic-net logistic regression. (A) Diagnostic performances of different protein combinations. Proteins are sorted according to the absolute coefficient from the largest to the smallest. “ROC test P" is the p-value of the comparison of ROC curves generated from successive protein combinations with one more protein added at a time. Coef., coefficient. ROC, receiver operator characteristics. *, p b 0·05. (B) Overlaid ROC curves of each of the combinations from two to 19 serum proteins. Comb, combination. (C) Protein-protein interactions among the 19 proteins assessed with the STRING database.
and the AUCs for clinically measured biomarkers CEA, CA19–9, and GC is often either asymptomatic or causing nonspecific symptoms in
Serum protein biomarkers identified by elastic-net logistic regression (ENLR) for gastric cancer diagnosis.
Identified proteins Short Ctrl (Mean Pre (Mean Pre vs. FDR Cut-off Sen Spe AUC (95%CI) T vs. FDR Padj
name ± SD) ± SD) Ctrl Padj
member 1-interacting protein
thrombospondin motifs 15
I: increase. D: decrease. FDR Padj: P values were tested by non-parametric Mann-Whitney-Wilcoxon and adjusted multiple tests with false discovery rate. CI: confidence interval. Coef.: coefficient calculated by ENLR. T: tumour tissue. N: adjacent normal tissue. –: Proteins not significantly altered.