Clear cell renal cell carcinoma (ccRCC) is the most common form of kidney cancer, making up about 70% of all renal malignancies. Most cases are driven by loss of the VHL tumor suppressor gene, which causes a chain reaction that activates proteins called HIFs (hypoxia-inducible factors), leading cancer cells to behave as if they are starved of oxygen even when they are not.
ccRCC is exceptionally complex because changes happen at multiple levels inside the cell - not just in the DNA sequence itself, but also in how genes are silenced by chemical tags (DNA methylation), how genes are read into messenger RNA (transcription), and how RNA is converted into proteins (translation). This layered complexity means that looking at just one type of data, such as gene expression alone, often fails to reveal why a particular cancer behaves the way it does.
Treatments for ccRCC have had limited success partly because doctors and researchers have lacked tools to understand which layer of gene regulation is most disrupted in a given patient. Without knowing where the problem originates - at the DNA, RNA, or protein level - it is difficult to design drugs that will work consistently across patients.
Researchers developed SiRCle (Signature Regulatory Clustering), a computational method that integrates three types of biological data simultaneously: DNA methylation data (which genes are chemically silenced), RNA-sequencing data (which genes are actively read), and proteomics data (which proteins are actually produced). The method follows the central dogma of biology - information flows from DNA to RNA to protein - to trace where a gene's dysregulation first appears.
SiRCle groups genes into clusters based on the regulatory layer where their problem originates. For example, a gene might have its activity suppressed by abnormal DNA methylation rather than by a change in its protein-coding sequence. By identifying this starting point, SiRCle can pinpoint the molecular mechanism driving each gene's abnormal behavior in cancer.
The method was paired with a variational autoencoder (VAE), a type of machine learning model that compresses complex multi-omics data into a simplified representation while preserving the most important patterns. This VAE component allowed researchers to compare different groups of patients - for example, early-stage versus late-stage cancer - and identify features that change systematically with disease progression.
Applying SiRCle to a cohort of ccRCC patients, researchers found that glycolysis - the process cancer cells use to generate energy from sugar - is abnormally elevated due to DNA hypomethylation (reduced chemical silencing of the relevant genes). This means the regulatory switch that normally keeps glycolysis controlled has lost its chemical tag, allowing these energy-production genes to run unchecked.
Simultaneously, mitochondrial enzymes and components of the cellular respiratory chain - the machinery that normally generates energy through oxygen-dependent processes - were found to be suppressed at the translational level. This means the instructions for building these proteins were present, but the cell was blocking their production. The transcription factor HIF1A was identified as a likely molecular driver behind the upregulation of glycolytic enzyme genes, consistent with the known role of the VHL-HIF pathway in ccRCC.
The VAE analysis revealed a stage-dependent pattern: as ccRCC advances, genes specific to the proximal renal tubule (the cell type from which kidney cancer arises) become progressively silenced. This suggests that as cancer cells grow and spread, they lose characteristics of their original cell type - a process called loss of cellular identity. This finding helps explain why advanced ccRCC behaves so differently from normal kidney tissue.
A key strength of SiRCle is its ability to distinguish between genes that look similar on one measurement but are dysregulated through entirely different mechanisms. For example, two genes might both be underexpressed in tumor tissue, but one may be suppressed by DNA methylation while another is blocked at the protein production stage. These two genes would require different therapeutic strategies to correct.
The researchers also applied SiRCle to a broader dataset spanning multiple cancer types (a pan-cancer cohort), finding that some regulatory signatures were shared between ccRCC and other cancers, while others were unique to kidney cancer. This suggests there may be opportunities to repurpose drugs developed for one cancer type in kidney cancer treatment, and vice versa.
The study also identified metabolic enzymes associated with patient survival - genes whose activity at the protein level, but not necessarily at the RNA level, predicts how well a patient will do. These represent potential biomarkers or drug targets that would have been invisible to analyses looking only at gene expression data.
SiRCle provides a new framework for understanding why ccRCC behaves as it does - and for identifying which patients might respond to which treatments. By pinpointing the regulatory layer where each gene's problem begins, it enables more precise targeting of cancer's root causes rather than its symptoms.
For patients, this research represents progress toward precision medicine in kidney cancer - treatment plans tailored to the specific molecular profile of each person's tumor. Rather than applying the same drug to all patients and hoping for the best, oncologists may eventually be able to use tools like SiRCle to match each patient with the therapy most likely to work for their particular pattern of gene dysregulation.
SiRCle is freely available as an open-source software package, allowing cancer researchers worldwide to apply this method to other cancer types and patient datasets. This broad availability increases the likelihood that insights from this work will translate into clinical benefits for patients in the years ahead.