Machine learning enables pan-cancer identification of mutational hotspots at persistent CTCF binding sites

Area of Science:

• Genomics
• Epigenetics
• Cancer Biology

Background:

• CCCTC-binding factor (CTCF) is a key insulator protein regulating 3D genome architecture and transcription.
• CTCF binding sites (CTCF-BSs) are frequently mutated in cancer.
• A subclass of CTCF-BSs, termed persistent CTCF binding sites (P-CTCF-BSs), are resistant to CTCF knockdown and maintain constitutive 3D chromatin architecture.

Purpose of the Study:

• To investigate the mutation rates of P-CTCF-BSs in various cancer types.
• To develop a machine learning tool (CTCF-INSITE) for predicting P-CTCF-BSs.
• To assess the functional impact of P-CTCF-BS mutations on CTCF binding and genome organization.

Main Methods:

• Analysis of ICGC sequencing data for mutation rates in breast and prostate cancer.
• Development and application of the CTCF-INSITE machine learning tool to predict P-CTCF-BSs.
• In vitro binding assays to validate the functional impact of mutations on CTCF binding.

Main Results:

• P-CTCF-BSs exhibit significantly elevated mutation rates in breast and prostate cancer compared to all CTCF-BSs.
• Predicted P-CTCF-BSs show increased mutational burden across 12 tested cancer types.
• Mutations in P-CTCF-BSs predicted to disrupt CTCF binding and chromatin looping were validated to reduce CTCF binding in vitro.

Conclusions:

• A novel subclass of cancer-specific CTCF-BS DNA mutations has been identified.
• These P-CTCF-BS mutations are prevalent across multiple cancer types and affect genome organization.
• The findings highlight the importance of P-CTCF-BSs in cancer development and genome regulation.