Colorectal cancer (CRC) is one of the most common and deadly cancers worldwide, with over 1.85 million new cases and roughly 850,000 deaths each year. Radiotherapy is one of the most important treatment strategies for CRC, particularly for rectal cancer, where it is used both before and after surgery to reduce the risk of local cancer recurrence.
Despite its widespread use, radiotherapy frequently fails because cancer cells develop resistance to radiation. Fewer than one in five rectal cancer patients achieves a complete pathological response - meaning the tumor is fully destroyed - following preoperative chemoradiotherapy. This resistance not only reduces treatment benefit but also raises the risk of cancer returning and spreading.
Understanding the molecular causes of radiation resistance is therefore critical. If scientists can identify specific genes or proteins that drive resistance, those molecules become potential targets for new drugs or treatment combinations that could restore cancer cells' sensitivity to radiation.
This study focused on a protein called MTDH (Metadherin), a transmembrane protein made of 582 amino acids that has been linked to cancer progression and treatment resistance in several cancer types including breast, pancreatic, and lung cancer. Whether MTDH plays a similar role in CRC radiation resistance was the central question this research set out to answer.
To identify genes most responsible for radiation resistance, the researchers applied a powerful computational approach: running 14 different machine learning algorithms simultaneously on gene expression data from CRC patients who either responded or did not respond to radiotherapy. The dataset used for initial model building (GSE68204) contained 59 rectal cancer patients, including 32 radiation-resistant and 27 radiation-sensitive cases.
The machine learning models tested included a wide range of approaches: GlmBoost, XgBoost, Ranger, Stepglm, Random Forest, Naive Bayes, Ridge, Elastic Net, Support Vector Machine (SVM), K-nearest neighbor, Lasso, multinomial logistic regression (multiNom), Linear Discriminant Analysis (LDA), and Logistic Regression. Each algorithm was trained and tested using 5-fold cross-validation to ensure the results were reliable and not just a product of chance.
To make the machine learning models interpretable - meaning to understand not just what they predicted but why - the team used a technique called SHAP (SHapley Additive exPlanations). SHAP assigns each gene a numerical score reflecting how strongly it influences the model's predictions. Genes with the highest SHAP values are the most important drivers of the model's ability to distinguish resistant from sensitive tumors.
A total of 5,963 genes were initially identified as differentially expressed between radiosensitive and radioresistant CRC tissues. This list was then cross-referenced with a database of known radio-resistance genes, yielding 21 candidate genes that went into the machine learning models. Across all 14 algorithms, MTDH consistently emerged with the highest predictive contribution, establishing it as the top radio-resistance gene candidate.
The team validated MTDH expression using multiple independent data sources. Analysis of single-cell RNA sequencing (scRNA-seq) data from 12 CRC patients confirmed that MTDH levels were significantly higher in tumor tissue compared to adjacent normal tissue (p less than 0.01). Broader transcriptomic analysis across 33 cancer types using the TCGA database confirmed MTDH overexpression specifically in both colon and rectal adenocarcinomas.
A particularly informative finding emerged from examining MTDH expression within specific subpopulations of tumor cells. Using a computational tool called CopyKat, cells were classified as diploid (normal chromosome number) or aneuploid (abnormal chromosome number, a hallmark of cancer malignancy). MTDH expression was significantly higher in the aneuploid, more malignant subpopulation, suggesting it is especially important in the most aggressive cancer cells.
To understand what biological processes MTDH controls, the team performed gene pathway enrichment analysis on 694 genes whose expression patterns were associated with MTDH levels. These genes were strongly linked to pathways known to be involved in radiation resistance: DNA repair regulation, DNA damage response, epithelial-mesenchymal transition (EMT), and stem cell differentiation. Each of these processes helps cancer cells survive the genetic damage that radiation is designed to cause.
In 82 rectal cancer patients who received neoadjuvant chemoradiotherapy, MTDH expression was measured directly in tumor biopsies taken before treatment. Patients whose tumors had higher MTDH showed a significantly lower response rate to chemoradiotherapy, with 39% of the non-response group having high MTDH compared to only 38% in the response group. An ROC analysis gave an AUC of 0.690, indicating MTDH has meaningful predictive value for distinguishing responders from non-responders.
Beyond its relationship with radiotherapy response, MTDH levels were linked to several indicators of disease severity in the 82-patient clinical cohort. Tumors at stage III had significantly higher MTDH expression than stage I-II tumors, indicating that MTDH rises as the disease becomes more advanced.
Statistical analysis using the chi-square test revealed that high MTDH expression was significantly associated with worse tumor differentiation (meaning the cancer cells look less like normal cells, which is a sign of more aggressive disease), larger tumor size (tumors greater than 5 cm), and lymph node involvement (cancer that has spread to nearby lymph nodes). All of these are established indicators of poor prognosis in rectal cancer.
Notably, MTDH expression was not significantly associated with patient age or gender, suggesting that its overexpression is a property of tumor biology rather than a reflection of patient demographics. This makes MTDH a more specific marker of tumor aggressiveness rather than simply a reflection of who gets the cancer.
Taken together, these clinical findings position MTDH as a potential prognostic biomarker - a measurable indicator that can predict patient outcomes before or after treatment. Identifying patients with high MTDH before radiotherapy begins could allow clinicians to consider alternative or intensified treatment strategies for those most at risk of treatment failure.
To confirm the machine learning and clinical findings in living cancer cells, the researchers created radioresistant variants of two CRC cell lines (SW480 and HT29) by repeatedly exposing the cells to radiation until they became resistant. The resulting radioresistant sublines (SW480-R and HT29-R) showed higher viability after radiation, formed more colonies, underwent less apoptosis, and had fewer DNA damage markers (gamma-H2AX foci) compared to their radiation-sensitive parent cells.
MTDH protein and mRNA levels were consistently higher in the radioresistant SW480-R and HT29-R cells than in the original sensitive cells, directly linking MTDH upregulation to acquired radiation resistance. This confirmed the association seen in patient samples at the cellular level.
To test whether MTDH actually causes the resistance rather than just correlating with it, the team used siRNA gene silencing - a technique that uses small RNA molecules to block production of a specific protein. When MTDH was silenced in the radioresistant cells and those cells were then exposed to radiation, they showed significantly reduced proliferation and increased rates of apoptosis compared to cells with normal MTDH levels.
Immunofluorescence imaging of gamma-H2AX foci - a standard marker for DNA double-strand breaks caused by radiation - showed that MTDH-silenced cells accumulated more DNA damage after radiation. This means that without MTDH, the cancer cells lose much of their ability to repair the radiation-induced damage that would normally kill them, effectively reversing their resistance.
The enrichment of MTDH-related genes in DNA repair and DNA damage response pathways offers a mechanistic explanation for how MTDH promotes radiation resistance. Radiation works by creating breaks in the DNA double helix that, if unrepaired, cause cancer cells to die. High MTDH expression appears to enhance the cell's ability to detect and repair this damage, allowing it to survive doses of radiation that would otherwise be lethal.
The enrichment in EMT pathways is also significant. Epithelial-mesenchymal transition is a process by which cancer cells take on a more mobile, stem-like identity. Cells that have undergone EMT are generally more resistant to therapy, including radiation, and are more capable of spreading to other organs. MTDH's association with EMT suggests it may simultaneously promote both radiation resistance and metastatic potential in CRC.
The link to stem cell differentiation pathways is similarly important. Cancer stem cells are a subpopulation of tumor cells that are particularly resistant to treatment and capable of regenerating the tumor after therapy. If MTDH helps maintain or expand the cancer stem cell pool, this could explain both the radiation resistance and the tendency for tumors with high MTDH to recur after treatment.
These multiple mechanisms - DNA repair, EMT, and stem cell maintenance - suggest that MTDH functions as a master regulator of treatment resistance rather than acting through a single pathway. This makes it a particularly attractive therapeutic target, since blocking MTDH could simultaneously disrupt several resistance mechanisms at once.
This study establishes MTDH as a validated key gene in CRC radiation resistance, identified through a rigorous multi-step process combining computational machine learning screening, multi-omics data validation, clinical patient cohort analysis, and direct laboratory experiments. The convergence of evidence from these independent approaches makes the case for MTDH's importance particularly strong.
The clinical data showing that MTDH expression before treatment predicts radiotherapy response suggests a practical application: measuring MTDH levels in pre-treatment tumor biopsies could help identify patients who are unlikely to respond well to standard chemoradiotherapy. These patients could then be considered for alternative treatment intensification or novel therapeutic approaches.
From a therapeutic perspective, MTDH represents an actionable target. While no MTDH-specific drugs exist yet, the demonstration that siRNA silencing of MTDH restores radiosensitivity provides proof-of-concept that reducing MTDH activity is a viable strategy. Development of small molecule inhibitors or RNA-based therapeutics targeting MTDH in combination with radiotherapy could significantly improve outcomes for resistant rectal cancer patients.
Future work should focus on expanding validation to larger and more diverse patient cohorts, investigating the exact molecular mechanisms through which MTDH regulates DNA repair and EMT in CRC, and testing MTDH-targeting strategies in animal models before advancing toward clinical trials. The ultimate goal is to translate this discovery into a clinically useful tool that improves survival for the many rectal cancer patients who currently do not benefit from radiotherapy.