Computed tomography-based radiomics prediction of CTLA4 expression and prognosis in clear cell renal cell carcinoma

Abstract

Objectives: To predict CTLA4 expression levels and prognosis of clear cell renal cell carcinoma (ccRCC) by constructing a computed tomography-based radiomics model and establishing a nomogram using clinicopathologic factors.

Methods: The clinicopathologic parameters and genomic data were extracted from 493 ccRCC cases of the Cancer Genome Atlas (TCGA)-KIRC database. Univariate and multivariate Cox regression and Kaplan-Meier analysis were performed for prognosis analysis. Cibersortx was applied to evaluate the immune cell composition. Radiomic features were extracted from the TCGA/the Cancer Imaging Archive (TCIA) (n = 102) datasets. The support vector machine (SVM) was employed to establish the radiomics signature for predicting CTLA4 expression. Receiver operating characteristic curve (ROC), decision curve analysis (DCA), and precision-recall curve were utilized to assess the predictive performance of the radiomics signature. Correlations between radiomics score (RS) and selected features were also evaluated. An RS-based nomogram was constructed to predict prognosis.

Results: CTLA4 was significantly overexpressed in ccRCC tissues and was related to lower overall survival. A higher CTLA4 expression was independently linked to the poor prognosis (HR = 1.458, 95% CI 1.13-1.881, p = 0.004). The radiomics model for the prediction of CTLA4 expression levels (AUC = 0.769 in the training set, AUC = 0.724 in the validation set) was established using seven radiomic features. A significant elevation in infiltrating M2 macrophages was observed in the RS high group (p < 0.001). The predictive efficiencies of the RS-based nomogram measured by AUC were 0.826 at 12 months, 0.805 at 36 months, and 0.76 at 60 months.

Conclusions: CTLA4 mRNA expression status in ccRCC could be predicted noninvasively using a radiomics model based on nephrographic phase contrast-enhanced CT images. The nomogram established by combining RS and clinicopathologic factors could predict overall survival for ccRCC patients. Our findings may help stratify prognosis of ccRCC patients and identify those who may respond best to ICI-based treatments.

Keywords: CTLA4; biomarker; clear cell renal cell carcinoma; machine learning; radiomics signature.

FIGURE 1

Flowchart of the radiomics process.

FIGURE 2

CTLA4 is overexpressed and related to poor prognosis in patients with ccRCC. (A) The expression of CTLA4 in tumor and normal tissues in the TCGA‐KIRC cohort. (B) Heatmap of the association between CTLA4 mRNA expression and clinicopathological features. (C) Kaplan–Meier analysis of overall survival in the TCGA‐KIRC cohort stratified by CTLA4 expression. (D) Violin plots of the correlation between CTLA4 expression and tumor‐infiltrating immune cells. Univariate (E) and Multivariate (F) analyses of the TCGA‐KIRC cohort.

FIGURE 3

GESA‐based GO annotation and KEGG pathway enrichment analyses. (A) biological process, (B) cellular component, (C) molecular function, and (D) KEGG analysis.

FIGURE 4

Establishment and assessment of radiomics signature to predict CTLA4 expression. (A) Features selected for model development and their importance. (B) ROC curves and AUC values for radiomics signature. (C) RS in groups with high and low CTLA4 expression. Kaplan–Meier (D) and landmark (E) analyses of overall survival stratified by RS. (F) Heatmap of the association between RS and clinicopathological features. (G) Violin plots of the correlation between RS and tumor‐infiltrating immune cells.

FIGURE 5

Construction and evaluation of RS‐based nomogram. (A) Nomogram used to predict the overall survival probability. (B) ROC curves and AUC values for RS‐based nomogram. (C) Calibration curve of the nomogram.