A new molecular subclassification and in silico predictions for diagnosis and prognosis of papillary thyroid cancer by alternative splicing profile

Abstract

Introduction: Papillary thyroid cancer (PTC) is the most common endocrine malignancy. However, different PTC variants reveal high heterogeneity at histological, cytological, molecular and clinicopathological levels, which complicates the precise diagnosis and management of PTC. Alternative splicing (AS) has been reported to be potential cancer biomarkers and therapeutic targets. Method: Here, we aim to find a more sophisticated molecular subclassification and characterization for PTC by integrating AS profiling. Based on six differentially expressed alternative splicing (DEAS) events, a new molecular subclassification was proposed to reclassify PTC into three new groups named as Cluster0, Cluster1 and Cluster2 respectively. Results: An in silico prediction was performed for accurate recognition of new groups with the average accuracy of 91.2%. Moreover, series of analyses were implemented to explore the differences of clinicopathology, molecular and immune characteristics across them. It suggests that there are remarkable differences among them, but Cluster2 was characterized by poor prognosis, higher immune heterogeneity and more sensitive to anti-PD1 therapy. The splicing correlation networks proved the complicated regulation relationships between AS events and splicing factors (SFs). An independent prognostic indicator for PTC overall survival (OS) was established. Finally, three compounds (orantinib, tyrphostin-AG-1295 and AG-370) were discovered to be the potential therapeutic agents. Discussion: Overall, the six DEAS events are not only potential biomarkers for precise diagnosis of PTC, but also the probable prognostic predictors. This research would be expected to highlight the effect of AS events on PTC characterization and also provide new insights into refining precise subclassification and improving medical therapy for PTC patients.

Keywords: alternative splicing (AS); diagnosis; in silico prediction; papillary thyroid cancer (PTC); prognosis; subclassification.

FIGURE 1

Summarization of AS events in PTC Cohort (A) The number of AS events and their parental genes (B) The upset plot of the intersection among seven types of AS events.

FIGURE 2

(A) Results of K-means clustering analysis was visualized by t-SNE algorithm (B) PCA of three distinct clusters was shown in a scatter plot (C) Differential analysis on PSI values of six DEAS events among three clustering groups (****: p < 0.0001) (D) Sankey Diagram showing comparisons between our clusters, PTC variants and subtypes based on DNA promoter methylation.

FIGURE 3

Performance of the SVM based classifier. The boxplots show the distributions of ACC and F1 values of 100 different testing sets produced by performing 100 round 10-fold cross-validations.

FIGURE 4

Clinicopathological characteristics and immune microenvironment features across DEAS-based clusters (A) Kaplan-Meier survival analysis of patients within three clustering subtypes of OS (B) A total of 442 PTC patients ordered by distinct clusters with annotations with cliniaopathological characteristics and immune features were visualized in a matrix heatmap (C) Immune and stromal scores of each DEAS-based cluster (D) Tumor purity of three clusters (E) Density curve of immune and stromal scores of all PTC patients (F) Comparisons on the proportions of immune infiltrating cells between three clusters (G) Expressions of immune checkpoints across the three clusters (H) Tumor mutation burden of three clusters (I) Responses to anti-PD1 therapy. The color in the cells represent the p values.

FIGURE 5

Molecular signatures associated with three DEAS-based clusters. Top 10 significantly differential signatures including KEGG pathways, reactome, molecular function, cellular component and biological process GO terms of three clusters were visualized in a matrix heatmap. The color (blue to red) in the matrix heatmap represents GSVA scores of biological pathways and GO terms of each PTC patient.

FIGURE 6

The AS-SF regulatory networks of Cluster0 (A), Cluster1 (B) and Cluster2 (C). Red circles are AS events associated with survival times and Green triangles are SFs related with corresponding AS events. The red/green lines represent positive/negative correlations between PSI values of prognostic AS events and expressions of SFs.

FIGURE 7

The prognostic significance of DEAS events (A) Kaplan-Meier curve analysis for OS (B–D) ROC validation of the predictive signature for predicting outcomes of PTC at 1-year, 3-year and 5-year, respectively (E) Forest plot summary of univariable analysis of sex, age, stage and risk score (F) Forest plot summary of multivariable analysis of stage and risk score.

FIGURE 8

Workflow of selecting potential compounds for PTC therapy and structures of three bioactive chemicals that share common PDGFR receptor inhibitor.