An international study has identified potential urine biomarkers for prostate cancer screening. By combining spatial transcriptomics, digital modelling of tumors, and machine learning, researchers identified a suite of biomarkers that indicate the presence and severity of prostate cancer with a high degree of precision.
In the study, published in Cancer Research, the researchers analyzed the micro RNA activity of all human genes in thousands of individual cells in prostate tumors using spatial transcriptomics. Knowing the position and degree of cancer of each cell, the team then constructed digital models of prostate cancer using a technique called pseudotime.
The models were analyzed with AI to identify proteins that could be used as biomarkers. Finally, the biomarkers were analyzed in the blood, prostate tissue, and urine of almost 2000 patients.
Here, lead author Martin Smelik, of The Karolinska Institutet, Stockholm, outlines the key findings and their implications for prostate cancer diagnostics.
What are the unmet needs in prostate cancer diagnostics that inspired this study?
Currently, the most common test to diagnose prostate cancer is based on the blood level of PSA. While this is a great tool, it lacks specificity, which might result in many false positive cases – leading to unnecessary biopsies for patients. This limitation inspired us to identify new biomarkers which might be easily measured in urine.
What motivated the approach of integrating spatial transcriptomics, pseudotime analysis, and machine learning for biomarker discovery in prostate cancer?
Spatial transcriptomics is a new technology that allows us to study the prostate with great resolution. As we were interested in the development of the cancer, pseudotime comes as a natural choice of methodology.
We used machine learning approaches successfully in our previous studies. For this study, our aim was to use this experience in a slightly different setting and find a way to effectively combine it with pseudotime and spatial transcriptomics.
How would you explain pseudotime modeling to the uninitiated, and how did it enhance your ability to identify reliable biomarkers for prostate cancer?
Pseudotime is essentially a digital model of malignant transformation. In other words, we used pseudotime to model the development of the cancer and identified genes that were associated with this development.
Of the 45 candidate biomarkers you identified, were there any that particularly stood out in terms of diagnostic performance or clinical relevance?
Indeed, there were several biomarkers that have been already studied in the context of prostate cancer. Some examples include TIMP1, which promotes proliferation of cancer cells in vivo, and S100A6, which is a calcium-binding protein implicated in a variety of biological functions as well as tumorigenesis.
Your study reports an AUC of 0.92 for urine-based biomarkers – significantly higher than that of serum PSA. What are the implications of this for non-invasive prostate cancer screening in clinical practice?
The main implication of our study is that the screening tests might potentially be more precise, if biomarkers are measured in urine, as opposed to blood, which is the current practice.
Variability in biomarker expression between patients is a well-known challenge. How did your approach address inter- and intra-patient heterogeneity?
We addressed the intra-patient heterogeneity by analyzing multiple prostate cancer samples from the same patients with a various level of cancer involvement. The inter-patient heterogeneity was addressed in the way we prioritized the biomarkers. Specifically, we selected those biomarkers that were consistently highly correlated with pseudotime across samples from multiple patients.
Looking ahead, how might this biomarker discovery pipeline be adapted to other cancers or therapeutic contexts?
We published all our codes to the online repositories where other researchers might access and re-use them. While we were focused specifically on prostate cancer, the methodology used in our study can be applied for other cancers which might potentially result in relevant biomarkers.
Newsletters
Receive the latest analytical scientist news, personalities, education, and career development – weekly to your inbox.
