Roses are red, tumors play chess,
your pee may be sending progress reports, I guess.

The Bathroom Sample With Main Character Energy

OK, so I have read this paper more times than is emotionally healthy, and I think the big idea is this: for biliary tract cancer, a notoriously unpleasant family of cancers involving the bile ducts and gallbladder area, the researchers asked whether urine proteins could help predict who benefits from immune checkpoint inhibitors.

That sounds almost too convenient. Cancer immunotherapy is complex, tumor biopsies are invasive, and the tumor microenvironment behaves like a tiny hostile office where every cell has a different agenda. Meanwhile, urine is just sitting there like, "I contain clues, actually."

The study, published in Gut, analyzed 211 urine samples from 97 treatment-naive patients with biliary tract cancer receiving immune checkpoint inhibitor-based therapy. The team used mass spectrometry to measure urinary proteins, trained a machine learning model on baseline protein signals, then connected those urine patterns to single-cell and spatial transcriptomics from tumor tissue. Correct me if I am wrong, but that is basically building a translator between "what the tumor neighborhood is doing" and "what leaves the body through the least glamorous exit ramp" Wang et al., 2025.

Why Immunotherapy Needs Better Crystal Balls

Immune checkpoint inhibitors, such as PD-1 and PD-L1 blockers, work by releasing brakes on immune cells so they can attack cancer. In theory: elegant. In practice: biology immediately asks for seventeen exceptions, a permissions form, and a snack.

For biliary tract cancer, adding immunotherapy to chemotherapy has improved outcomes. TOPAZ-1 showed durvalumab plus gemcitabine and cisplatin improved survival in advanced disease Oh et al., 2022. KEYNOTE-966 found a survival benefit from pembrolizumab plus gemcitabine and cisplatin Kelley et al., 2023. Good news, yes. But most patients still do not get long-lasting benefit.

That leaves clinicians with the awkward question: who is likely to respond?

Traditional biomarkers like PD-L1 expression, tumor mutation burden, and MSI status can help sometimes, but they are not exactly a GPS. More like a weather app that says "possible immune activity, bring umbrella?" A 2024 systematic review found PD-L1 may predict survival in biliary tract cancer patients receiving PD-1/PD-L1 therapy, but its value for predicting actual tumor response is less reliable and depends on testing details Kim et al., 2024.

The Four-Protein Urine Panel

Here is the part where I had to re-read the methods and whisper "please be a simple classifier" to myself.

The researchers built a machine learning model from baseline urinary proteomic features and landed on a four-protein panel:

• PTPN13
• SUB1
• MICAL-L1
• VARS1

This panel predicted durable clinical benefit and early response. Then they tested it in an independent cohort of 24 patients using parallel reaction monitoring mass spectrometry, a more targeted protein measurement method. That validation step matters because biomarker papers can sometimes behave like souffles: impressive in one kitchen, tragic everywhere else.

The paper also reports that patients with durable benefit showed urinary protein signals linked to immune activation and systemic inflammatory pathways. Patients without durable benefit showed patterns tied to pro-tumor processes. If I am reading this right, the urine was not merely saying "cancer exists." It was hinting at which biological mood the tumor ecosystem was in.

The Tumor Neighborhood, Now With Street Addresses

The clever twist is the integration with single-cell and spatial transcriptomics. Single-cell RNA sequencing tells you what individual cells are doing. Spatial transcriptomics adds the "where," because in tumors, location matters. A T cell standing beside a malignant cell is a different story from a T cell trapped across the room behind fibroblast bureaucracy.

Spatial transcriptomics has become a major tool for mapping gene expression inside tissue architecture, while single-cell methods reveal cell states that bulk tissue averages can smear into biological soup Palla et al., 2024. This study used those approaches to connect urinary signals with tumor microenvironment remodeling.

The standout character was PTPN13-positive malignant cells. The authors suggest these cells may help regulate pro-apoptotic tumor microenvironment states, contributing to sustained immunotherapy responsiveness. In normal-person terms: some tumor cells may be associated with a neighborhood where cancer cells are more prone to die and immune therapy has a better shot. I think. I am leaving room here because cancer biology loves making anyone sound overconfident and then immediately proving them rude.

Why This Could Matter

If this approach holds up, it could give doctors a non-invasive way to predict and monitor immunotherapy response over time. That is the key phrase: over time. A biopsy is a snapshot. Urine sampling could become more like checking the tumor's group chat every few weeks, minus the privacy violations and with more mass spectrometry.

This is especially attractive in cancers where repeated tissue sampling is hard, risky, or just deeply unpopular among people who enjoy not being biopsied. A dynamic urine test could help identify early non-responders, monitor tumor evolution, and maybe guide treatment changes before imaging makes the situation obvious.

But let us not sprint directly into confetti. The study is promising, not final. The patient numbers are still modest, the validation cohort was small, and clinical deployment would require larger, multi-center testing across treatment regimens, sample handling conditions, and patient populations. Machine learning models also have a classic bad habit: they can look brilliant in familiar data and then panic in the wild like a Roomba encountering stairs.

Still, the concept is genuinely neat: urine proteomics as a window into immunotherapy response, cross-checked against single-cell and spatial biology. It is not magic. It is measurement, modeling, and a lot of patient samples. Which, honestly, is better than magic because magic rarely comes with supplementary tables.

References

1. Wang S, Guo Z, Sun B, et al. Dynamic urinary proteomics integrates single-cell and spatial transcriptomics to reveal tumour microenvironment and predict immunotherapy response in biliary tract cancer. Gut. 2025. DOI: 10.1136/gutjnl-2025-335513. PMID: 41151791.

2. Oh DY, He AR, Qin S, et al. Durvalumab plus gemcitabine and cisplatin in advanced biliary tract cancer. NEJM Evidence. 2022. DOI: 10.1056/EVIDoa2200015.

3. Kelley RK, Ueno M, Yoo C, et al. Pembrolizumab in combination with gemcitabine and cisplatin compared with gemcitabine and cisplatin alone for advanced biliary tract cancer (KEYNOTE-966). The Lancet. 2023;401:1853-1865. DOI: 10.1016/S0140-6736(23)00727-4.

4. Kim TH, Kim H, Lee SH, et al. The predictive value of PD-L1 expression in response to anti-PD-1/PD-L1 therapy for biliary tract cancer: a systematic review and meta-analysis. Frontiers in Immunology. 2024. DOI: 10.3389/fimmu.2024.1321813.

5. Palla G, Fischer DS, Regev A, Theis FJ. Dissecting the tumor microenvironment in response to immune checkpoint inhibitors with single-cell and spatial transcriptomics. Nature Reviews Cancer. 2024. PMCID: PMC11374862.

Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.