High-dimensional, irregularly sampled longitudinal plasma metabolomics in small immunotherapy cohorts is the bottleneck this paper tries to kick out of the oncology lab.

That phrase sounds like it was assembled from spare grant-review parts, so let’s translate: doctors can collect blood from cancer patients over time, measure a pile of tiny molecules floating around in it, and ask, “Can this chemical soup tell us who will benefit from immune-checkpoint inhibitors?” The trouble is that the data are messy, time-stamped unevenly, and packed with more variables than a group chat planning dinner. Standard models tend to either panic or pretend they are more confident than they should be. Neither is ideal when the stakes are cancer treatment.

Suissa, Fidelle, Reich and colleagues went big: 4,336 plasma samples from 1,714 patients, spanning five tumor types and 16 cohorts across Europe and North America. They mixed targeted metabolomics, metagenomics, clinical variables, mouse experiments, diet questionnaires, and a machine-learning framework based on DynForest, a random-forest approach built for longitudinal survival data Suissa et al., 2026; Devaux et al., 2023.

The Immune System Has Brakes, Snacks, and Drama

Immune-checkpoint inhibitors, like anti-PD-1, anti-PD-L1, and anti-CTLA-4 drugs, try to release the brakes on T cells so they can attack tumors. Great idea. Slight catch: only some patients respond, and biology does not hand out neat labels like “Responder, definitely” or “Do not bother, try plan B.”

This paper looks at metabolism as part of that answer. Metabolomics is basically chemical receipt-reading: amino acids, fatty acids, and other small molecules reveal what the body, tumor, immune system, and gut microbes have been up to. It is less glamorous than a sci-fi scanner, but probably more useful than asking the tumor nicely to disclose its intentions.

The model integrated 154 metabolites with clinical features and landed on a compact signal: histidine, succinate, several long-chain fatty-acid-related molecules, age, BMI, and renal-function-related information. In the paper’s main report, the model reached an AUC of 0.88 in training and 0.73 in a 30-patient validation cohort. That 0.73 is promising. It is also the part where the skeptical eyebrow goes up, because 30 patients is not exactly a stadium crowd.

Histidine Walks In Looking Suspiciously Helpful

The headliner is histidine, an essential amino acid. Higher plasma histidine was associated with better progression-free survival across several cohorts, including melanoma and renal-cell carcinoma datasets. In mouse models, histidine supplementation enhanced antitumor immune activity and seemed to help T cells avoid some exhaustion-associated metabolic failure.

Wait, really? An amino acid from protein-rich foods might help immunotherapy work better? Maybe. But this is where the fine print starts doing push-ups.

The study does not say “everyone on immunotherapy should buy histidine supplements before lunch.” It says histidine looks like a favorable biomarker and possibly a modifiable pathway worth testing prospectively. The distinction matters. Biology loves making simple ideas weird. It is the friend who says “quick question” and then opens a spreadsheet.

The gut microbiome complicates the story. In some patients, especially those with dysbiosis, gut microbes may chew through histidine and produce imidazole propionate, or ImP, a metabolite linked here with poorer outcomes. So the question is not just “How much histidine did you eat?” It is also “What did your microbial roommates do with it after you swallowed it?” A related 2026 review makes the same larger point: microbial metabolites can shape both innate and adaptive immune responses, but the mechanisms vary by context Toner-Bartelds et al., 2026.

The Villains: Succinate and Fatty Acids, Maybe

Histidine got the flattering lighting. Succinate and long-chain fatty acids got the suspicious soundtrack. The model linked higher succinate and certain lipid-related metabolites with worse outcomes. That fits a broader immunometabolism literature where metabolites are not just fuel, but signals that can make immune cells sharper, sleepier, or weirdly bureaucratic Trefny et al., 2025.

Other recent work points in the same direction. A 2024 Nature Communications study tied microbiota-associated metabolic reprogramming to melanoma outcomes in the MIND-DC trial Alves Costa Silva et al., 2024. Another Nature Medicine study tracked gut microbiome changes during checkpoint blockade in advanced melanoma Björk et al., 2024. And a JITC paper connected metabolomic patterns with PD-1 inhibitor plus chemotherapy outcomes in advanced non-small-cell lung cancer Zheng et al., 2024.

The Catch, Because There Is Always a Catch

This paper is impressive because it does not stop at “the model found a thing.” It validates across external cohorts, checks stool and plasma metabolites, looks at immune phenotypes, and tests histidine in mice. That is a lot more satisfying than a heatmap wearing a lab coat.

Still, the authors are careful: the training and validation cohorts were modest and heterogeneous, the feature space was large, and false discovery remains possible. Also, a clinical-only comparator performed strongly in the validation setup, which means the metabolite signature still has to prove its added clinical value in prospective trials.

If the result holds up, the impact could be practical: blood and stool profiling might help identify patients more likely to benefit from checkpoint inhibitors, monitor metabolic resistance, or select people for diet or microbiome-based add-ons. Not a magic amino-acid smoothie. More like a better dashboard for a very complicated machine.

References

1. Suissa, D., Fidelle, M., Reich, E. et al. Metabolic determinants of cancer immunotherapy outcomes identified by plasma profiling. Nature Medicine (2026). DOI: 10.1038/s41591-026-04481-9. PMID: 42350644.
2. Toner-Bartelds, C., Mimpen, I. L., Parra-Martinez, M. et al. Microbiota-derived metabolites as modulators of cancer immunotherapy response. Nature Communications 17, 5274 (2026). DOI: 10.1038/s41467-026-72178-1.
3. Trefny, M. P., Kroemer, G., Zitvogel, L. & Kobold, S. Metabolites as agents and targets for cancer immunotherapy. Nature Reviews Drug Discovery 24, 764-784 (2025). DOI: 10.1038/s41573-025-01227-z. PMID: 40571788.
4. Alves Costa Silva, C. et al. Influence of microbiota-associated metabolic reprogramming on clinical outcome in patients with melanoma from the randomized adjuvant dendritic cell-based MIND-DC trial. Nature Communications 15, 1633 (2024). DOI: 10.1038/s41467-024-45357-1. PMID: 38395948.
5. Björk, J. R. et al. Longitudinal gut microbiome changes in immune checkpoint blockade-treated advanced melanoma. Nature Medicine 30, 785-796 (2024). DOI: 10.1038/s41591-024-02803-3. PMID: 38365950.
6. Zheng, L. et al. Association of metabolomics with PD-1 inhibitor plus chemotherapy outcomes in patients with advanced non-small-cell lung cancer. Journal for ImmunoTherapy of Cancer 12, e008190 (2024). DOI: 10.1136/jitc-2023-008190. PMID: 38641349.
7. Devaux, A., Helmer, C., Genuer, R. & Proust-Lima, C. Random survival forests with multivariate longitudinal endogenous covariates. Statistical Methods in Medical Research 32, 2331-2346 (2023). DOI: 10.1177/09622802231206477.

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.