CTLA4 (Cytotoxic T-Lymphocyte-Associated Protein 4) is a critical immune checkpoint receptor expressed on T cells. When activated, CTLA4 binds to B7 ligands on antigen-presenting cells and transmits inhibitory signals that dampen T cell activation, allowing tumors to evade immune destruction. Ipilimumab, an anti-CTLA4 antibody, has demonstrated survival benefit in combination with anti-PD-1 therapy in advanced RCC.
Despite therapeutic relevance, CTLA4 expression in tumors varies widely and cannot be easily assessed without tissue sampling. Elevated CTLA4 in the tumor microenvironment has been associated with advanced tumor stage, poor grade, and worsened prognosis in clear cell renal cell carcinoma (ccRCC), making it a biomarker of interest. However, routine biopsy for CTLA4 testing is invasive and not universally feasible.
CT-based radiomics offers a non-invasive alternative. By extracting quantitative image features from pre-operative CT scans, it may be possible to predict CTLA4 expression levels and use that prediction to inform prognosis and treatment selection. This approach would democratize access to immune profiling without the need for molecular testing in every patient.
This study used a dual-dataset design. Transcriptomic and clinical data for 493 ccRCC patients were obtained from The Cancer Genome Atlas (TCGA-KIRC), while CT imaging data for 102 patients with matched TCGA identifiers were sourced from The Cancer Imaging Archive (TCIA). This design enabled correlation of imaging features with CTLA4 gene expression.
Radiomics features were extracted from manually segmented tumor regions on arterial-phase CT images. Feature categories included shape, histogram-based statistics, texture (GLCM, GLSZM, GLRLM), and wavelet-transformed features. After excluding redundant and low-variance features, a final set was passed to feature selection algorithms.
Support Vector Machine with Recursive Feature Elimination (SVM-RFE) was used to identify the most discriminative features for predicting CTLA4 expression status (high vs. low, based on median split). This method systematically removes the least important features and re-trains the model, converging on the smallest subset that maintains classification accuracy.
Seven radiomics features were selected. A Radiomics Score (RS) was computed as a linear combination of these features weighted by their SVM coefficients. Separately, a prognostic nomogram was constructed combining RS with clinical variables to predict one-year, three-year, and five-year overall survival in the TCGA cohort.
The 7-feature radiomics model achieved an AUC of 0.769 in the training set and 0.724 in the validation set for classifying CTLA4 expression status. While not perfect, these AUCs demonstrate that CT imaging features carry meaningful information about the molecular immune state of ccRCC tumors.
Kaplan-Meier survival analysis stratified by RS showed significant separation in overall survival between RS-high and RS-low patients in the TCGA cohort (log-rank p less than 0.05). Patients in the high-RS group had shorter median OS, consistent with the hypothesis that high CTLA4-associated imaging phenotypes confer worse prognosis.
The prognostic nomogram integrating RS with age, tumor stage, and grade achieved time-dependent AUCs of 0.826, 0.805, and 0.760 for predicting one-year, three-year, and five-year OS respectively. These values compare favorably to standard staging alone, suggesting that radiomics adds prognostic value beyond conventional clinical variables.
Calibration curves demonstrated good agreement between nomogram-predicted and observed survival rates across the time horizons analyzed. Decision curve analysis confirmed a positive net benefit of the nomogram over a wide range of threshold probabilities, supporting clinical utility for individualized risk communication.
Analysis of the 493 TCGA patients confirmed that high CTLA4 expression was significantly associated with higher WHO/ISUP histological grade and more advanced clinical stage. This aligns with the known biology of CTLA4 as an immune checkpoint upregulated in response to increased immune activation pressure in aggressive tumors.
Immune deconvolution using CIBERSORT revealed that CTLA4-high tumors contained significantly elevated proportions of M2 macrophages. These alternatively activated macrophages suppress cytotoxic immune responses and promote tumor progression through pro-angiogenic and tissue-remodeling activities, creating a hostile environment for effective immune control.
The enrichment of M2 macrophages in CTLA4-high tumors is particularly relevant because it suggests that patients with high radiomics scores may have both immune checkpoint upregulation and macrophage-mediated immunosuppression, representing a doubly immunosuppressed microenvironment that may be especially refractory to single-agent immunotherapy.
Gene set enrichment analysis showed upregulation of immune evasion and angiogenesis pathways in CTLA4-high tumors. This convergence of imaging phenotype, immune landscape, and molecular pathway enrichment validates that the radiomics score captures clinically meaningful tumor biology.
The combination of CT-derived radiomics and CTLA4 expression patterns has direct clinical implications. Patients identified as CTLA4-high by radiomics-based scoring might be candidates for combination checkpoint blockade regimens, such as nivolumab plus ipilimumab, which targets both the PD-1 and CTLA4 pathways simultaneously.
Conversely, the enrichment of M2 macrophages in CTLA4-high tumors suggests that anti-CTLA4 therapy alone may be insufficient without simultaneously addressing the macrophage-driven immunosuppression. This provides a rationale for clinical trials combining CTLA4 blockade with macrophage-targeting agents such as CSF1R inhibitors in high-RS ccRCC patients.
The non-invasive nature of the radiomics prediction framework is a key clinical advantage. In patients where biopsy is contraindicated or where genomic testing is unavailable, CT-based RS could provide an accessible proxy for tumor immune profiling, enabling treatment stratification decisions at diagnosis without additional invasive procedures.
This study demonstrates a proof-of-concept for using CT radiomics to predict immune checkpoint expression non-invasively in ccRCC. The CTLA4 radiomics score adds prognostic value beyond clinical staging and provides a window into the tumor's immune microenvironment without biopsy.
The integration of TCGA molecular data with TCIA imaging data represents a powerful methodological framework for radiogenomics research. By linking CT image features to gene expression profiles, this approach can potentially extend to any molecularly characterized target, enabling imaging-based molecular phenotyping across diverse cancer types.
Limitations include the relatively small size of the imaging cohort (n=102), retrospective design, and absence of independent external imaging validation. Future work should prospectively validate the RS in larger cohorts with matched imaging and molecular data, and test its predictive value for actual immunotherapy response outcomes rather than surrogate expression markers.
If validated prospectively, a CT-based CTLA4 radiomics score could become part of a routine pre-treatment imaging analysis pipeline, contributing to a non-invasive precision oncology workflow for kidney cancer management.