966P Activation of the Wnt/PCP signaling pathway is an adverse prognostic predictor in patients with low grade glioma (LGG)

Previous studies have shown that Wnt/ planar cell polarity (PCP) pathway activation negatively correlates with survival in glioblastoma (GBM) patients. However, the role of the Wnt/PCP pathway in LGG development has not been elucidated. Although LGG patients have better prognosis compared to GBM patients, a small proportion of LGG patients show poor survival and no associated molecular/pathological factors have been found. We integrated data of 510 LGG patients from The Cancer Genome Atlas (TCGA) and 375 LGG patients from the Chinese Glioma Genome Atlas (CGGA) database. All transcriptomic data were unified to remove batch effects and normalized as transcript per million base (TPM). Gene Set Enrichment Analysis (GSEA) of Wnt/PCP singnaling pathway were evaluated by ssGSEA. Multivariate Cox model was used for survival analysis. Differential expression (DE) analysis was performed using edgeR, DE genes were identified based on if FC>2 and q value < 0.05. Feature selection was done using LASSO with λ equal to 0.082. Logistic regression was trained on TCGA dataset with CV, nfolds was as large as the sample size. CGGA dataset was used as an independent validation set. Statistical differences between groups was evaluated using Wilcoxon signed-rank test. Top DE genes between Wnt/PCP high/low groups were enriched in blood-brain barrier (BBB) related pathways (q value < 0.05). We found DKK1, SFRP1 and SFRP2 in Wnt antagonist Dickkopf and sFRP gene families were significantly overexpressed in the Wnt low group (P= 0.05). Survival analysis on the TCGA dataset showed that activity of the Wnt/PCP pathway significantly affected LGG patient survival. The survival difference remained significant (P = 0.01) when evaluated by multivariate Cox model adjusting for known confounders, including IDH mutations, MGMT methylation status and cancer grade. LASSO identified 14 genes that mostly contribute to Wnt activation. Logistic regression model was trained on 358 TCGA samples using these selected genes, the model showed high accuracy in training set (AUC = 0.924) as well as in an independent CGGA validation set (AUC = 0.93). The median survival was 113 and 51 months for the Wnt low and high groups, respectively. Our analysis indicated that activity of Wnt may be an independent adverse predictor of LGG patient survival. In addition, a set of key genes were selected and can be conveniently used in the clinical setting for prognosis of LGG patients.