A new study demonstrated how combining spatial transcriptomics, pseudotime analysis, and machine learning identified reliable biomarkers for prostate cancer that could outperform traditional prostate-specific antigen testing. Researchers from the Karolinska Institute and international collaborators developed an approach that achieved an area under the curve of 0.92 for detecting prostate cancer using urine biomarkers.

The study addressed a fundamental challenge in cancer diagnosis: identifying reliable biomarkers that can be measured using routine clinical methods. Current prostate cancer screening approaches rely heavily on prostate-specific antigen (PSA) testing, which has limited sensitivity and specificity, leading to unnecessary biopsies or missed diagnoses.

"We hypothesized that reliable biomarkers that can be measured with routine methods could be identified by exploiting three facts: the same tumor can have multiple grades of malignant transformation, these grades and their molecular changes can be characterized using spatial transcriptomics, and these changes can be integrated into models of malignant transformation using pseudotime," the study authors wrote.

Methodology and Key Findings

The research team analyzed spatial transcriptomics data from 12 samples across 6 prostate cancer patients. Using pseudotime analysis, they tracked molecular changes during malignant transformation and identified 45 genes highly correlated with cancer progression, including 17 positively correlated and 28 negatively correlated genes.

The researchers validated these biomarkers through multiple approaches:

• Single-cell RNA analysis of 21,743 cells from 18 samples confirmed that 29 of 31 differentially expressed candidate biomarkers matched the direction predicted by pseudotime analysis.

• Immunohistochemistry of Spondin-2 (SPON2), one of the key identified biomarkers, showed increasing expression with higher cancer grades.

• Analysis of bulk RNA from 714 tumor tissue samples and 174 normal samples demonstrated that 33 of 36 differentially expressed biomarkers aligned with the predictions.

• Urine proteomic analysis confirmed that biomarkers identified in tissue were detectable in urine with high accuracy.

Machine learning models based on these biomarkers in urine samples achieved a mean area under the curve (AUC) of 0.92 for detecting prostate cancer, substantially outperforming PSA (AUC = 0.63) and even randomly selected proteins (AUC = 0.88).

Clinical Implications

The researchers demonstrated that these biomarkers may also help predict cancer grade. The study authors noted: "We found that our model (R² = 0.295, root mean square error [RMSE] = 1.458, correlation = 0.588) outperformed the baseline model based on serum PSA (R² = 0.191, RMSE = 1.561, correlation = 0.445), thus providing potential clinical utility in the detection of clinically significant prostate cancer."

Lead study author Martin Smelik, and colleagues indicated the need for prospective studies to validate these findings. The researchers have made their data and methods freely available to encourage application to other cancer types.

This approach represented an advancement in biomarker discovery by systematically addressing the heterogeneity and complexity of cancer. The integration of spatial transcriptomics with pseudotime analysis provided a mechanistic understanding of why specific biomarkers are relevant, potentially improving diagnostic reliability across different clinical settings.

Source: Cancer Research