The foundation of this research lies in the recognition that R-loop dysregulation has long been implicated in genomic instability in colorectal cancer. However, the precise association between specific R-loop-related transcriptional programs and the distinct malignant cell states within the tissue context remained largely undefined. To address this gap, the team utilized a multi-omic approach, combining single-cell, spatial, and bulk transcriptomic data from CRC samples. R-loop-associated transcriptional scores were calculated using multi-method gene-set scoring, a technique designed to capture the complex, heterogeneous nature of R-loop activity across different cell types and spatial locations. Furthermore, the study rigorously evaluated cell-state dynamics, the spatial organization of tumor cells, and the intricate microenvironmental communication between malignant cells and surrounding stromal components using advanced trajectory, communication, colocalization, and co-expression network analyses. This holistic integration of data sources allowed the researchers to move beyond isolated gene expression patterns to understand the systemic impact of R-loop activity on tumor biology.

The methodological core of the study involved the application of two distinct machine-learning stages. The initial stage utilized five complementary feature-selection algorithms—specifically Random Forest, Boruta, LASSO, gradient boosting machine, and XGBoost—to prioritize genes that were significantly associated with R-loop activity. This robust feature selection process was crucial in filtering the vast amount of transcriptomic data down to a manageable and biologically relevant set of candidates. This rigorous selection process successfully identified six shared R-loop-associated candidate genes across the different analytical frameworks. Among these critical genes, nucleolin (NCL) emerged as a focal point, demonstrating significant enrichment within malignant epithelial cells. Crucially, NCL was further associated with a significantly poorer overall survival rate among CRC patients, thereby elevating its importance as a potential biomarker for aggressive disease.

The second stage of the machine learning strategy focused on prognostic modeling. The researchers compared 101 different survival-model combinations to construct a robust prognostic signature specifically from NCL-positive malignant epithelial cells. This extensive combinatorial analysis led to the identification of a forward stepwise Cox plus Ridge model, which proved highly effective in stratifying patient survival within both the TCGA-COAD and GSE41258 cohorts. Beyond survival prediction, the high-risk groups identified by this model also exhibited distinct predicted drug responses when evaluated against several pharmacogenomic datasets, including GDSC2, CTRP, and PRISM. This suggests that the R-loop/NCL signature not only predicts who might die but also hints at how these aggressive cells might respond to specific therapeutic interventions, opening avenues for personalized medicine.

The functional validation of NCL’s role in CRC was performed through cellular assays. The researchers successfully silenced NCL in colorectal cancer cell lines and meticulously assessed the consequences of this silencing on key malignant behaviors. Using standard assays such as qRT-PCR, CCK-8 for viability, Transwell for migration, and wound-healing assays, they observed that NCL silencing resulted in a measurable reduction in proliferation, migration, and invasion. Furthermore, virtual NCL knockout simulations suggested a significant perturbation of proliferative signaling pathways within the malignant cells. These experimental results provide strong functional evidence that NCL is not merely a correlative marker but an active participant in driving the aggressive phenotype of CRC.

The spatial analysis provided a critical layer of context to these findings. The study revealed that R-loop-high malignant cells were not randomly distributed but were specifically enriched within tumor nests. Moreover, these high-risk cells showed increased inferred communication with various components of the tumor microenvironment, including fibroblasts, endothelial cells, and myeloid cells. This finding is highly significant as it suggests that the aggressive nature of R-loop-associated malignant cells is not confined to the epithelial layer but actively involves and modulates interactions with the surrounding stromal and immune cells, a process vital for tumor progression and metastasis.

The integration of these complex datasets—transcriptomics, spatial organization, and survival outcomes—highlights the power of consensus machine learning in modern cancer biology. By linking R-loop-associated gene selection with NCL-positive epithelial prognostic modeling, the study provides a unified view of how epigenetic dysregulation (R-loops) translates into clinical malignancy (NCL expression and poor prognosis). This two-stage strategy moves beyond traditional single-gene analyses, offering a more comprehensive and clinically relevant picture of CRC progression.

In conclusion, this research strongly supports the notion that NCL functions as a critical marker of R-loop-associated malignant progression in colorectal cancer. The resulting prognostic model and the identification of specific drug response patterns provide a solid basis for future exploratory studies aimed at developing novel therapeutic strategies. By pinpointing NCL as a functional marker linked to R-loop activity, the study offers a tangible target for intervention, potentially allowing clinicians to stratify patients more accurately and guide treatment decisions based on the underlying molecular drivers of their cancer.

Subject of Research: The use of integrated single-cell and spatial transcriptomics with machine learning to link R-loop activity to nucleolin-driven malignancy in colorectal cancer.

Article Title: Consensus machine learning integrates single-cell and spatial transcriptomics to map R-loop-associated transcriptional scores and NCL-associated malignant progression in colorectal cancer

Article References: Sun, S., Chen, S., Chen, K., Li, W., Li, J., & Lv, B. (2026). Consensus machine learning integrates single-cell and spatial transcriptomics to map R-loop-associated transcriptional scores and NCL-associated malignant progression in colorectal cancer. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-09144-4

Image Credits: AI Generated

DOI: 10.1186/s12967-026-09144-4

Keywords: Colorectal Cancer, R-loops, Nucleolin, Single-Cell Transcriptomics, Machine Learning, Cancer Prognosis, Malignant Progression, Tumor Microenvironment, Consensus, machine, learning, integrates