Gut Microbe Metabolites Sharpen Prediction of Stroke Worsening After Thrombolysis

When a patient receives intravenous thrombolysis for an acute ischemic stroke, the treatment can dissolve the clot and restore blood flow, yet a troubling minority of patients deteriorate neurologically within the first day instead of improving. This phenomenon, known as early neurological deterioration or END, is strongly linked to poor long-term outcomes, and clinicians currently have limited tools for identifying who is most at risk before the decline begins. A new prospective observational study from a single Chinese stroke center now reports that adding gut microbiota-derived tryptophan metabolites and an inflammatory cytokine to conventional bedside variables substantially improves the prediction of END in thrombolyzed stroke patients.

The research, conducted at the Second People’s Hospital of Yibin City-West China Yibin Hospital of Sichuan University between October 2021 and February 2026, enrolled adults with acute ischemic stroke who received intravenous thrombolysis with either tenecteplase or alteplase within 4.5 hours of symptom onset. Of 757 patients screened, 666 remained in the final analysis after exclusions for urokinase treatment, stroke mimics, and loss to follow-up. END, defined as a sustained increase of at least 4 points on the National Institutes of Health Stroke Scale within 24 hours of admission, occurred in 75 patients, an incidence of 11.26 percent.

The study’s central finding concerns the performance of two predictive nomograms. The first, built from conventional risk factors alone, used systolic blood pressure and fasting blood glucose, both of which remained independently associated with END after multivariable adjustment. This base model achieved an area under the receiver operating characteristic curve of 0.719 (95 percent confidence interval 0.658 to 0.780), a level the authors characterize as moderate discrimination. The second model added three emerging serum biomarkers: indole-3-lactic acid (ILA) and indole-3-propionic acid (IPA), both gut-derived tryptophan metabolites, and interferon gamma, a pro-inflammatory cytokine. The extended model reached an AUC of 0.909 (95 percent confidence interval 0.879 to 0.939), with a sensitivity of 83.9 percent and a specificity of 85.9 percent at the optimal cutoff probability of 13.4 percent.

The improvement was formally quantified using reclassification statistics. Compared with the conventional model, the biomarker-enhanced model produced a categorical Net Reclassification Improvement of 0.601 (95 percent confidence interval 0.477 to 0.725), meaning that 60.1 percent of individuals were correctly reassigned to risk categories, and a continuous NRI of 1.369. The Integrated Discrimination Improvement was 0.197 (95 percent confidence interval 0.162 to 0.233), with both comparisons reaching statistical significance through bootstrap testing with 1,000 replicates. Internal validation by the same bootstrap procedure yielded a C-index of 0.905 for the extended model, closely matching its apparent performance and supporting the stability of the result within this cohort.

The biological rationale for including these particular markers rests on the gut-brain axis. Tryptophan, an essential dietary amino acid, is metabolized along serotonin, kynurenine, and indole pathways, and gut bacteria generate indole compounds that act as signaling molecules for the aryl hydrocarbon receptor, a route implicated in immune regulation and neuroinflammation in stroke. IPA is an indole derivative produced exclusively by gut bacteria, and prior studies reviewed by the authors suggest it exerts neuroprotective effects in experimental cerebral ischemia. ILA, another indole-derived metabolite, has been reported to mitigate brain injury after ischemia in animal work, apparently through aryl hydrocarbon receptor activation that inhibits neuronal ferroptosis.

In the new cohort, the laboratory measurements told a consistent story. Serum concentrations of ILA and IPA, measured by commercial ELISA kits before thrombolysis, were significantly lower in patients who went on to deteriorate than in those who did not, while levels of matrix metalloproteinase 9, lipopolysaccharide, and interferon gamma were elevated; trimethylamine N-oxide showed no difference between the groups. In multivariable logistic regression, higher ILA (odds ratio 0.973 per unit, 95 percent confidence interval 0.963 to 0.981), higher IPA (odds ratio 0.989, 95 percent confidence interval 0.983 to 0.994), and higher interferon gamma (odds ratio 1.052 per unit in the biomarker-only analysis, 95 percent confidence interval 1.005 to 1.107) were each independently associated with END. Matrix metalloproteinase 9 and lipopolysaccharide did not retain independent significance in adjusted models.

The conventional predictors behaved as expected. Patients who developed END had markedly higher systolic blood pressure after thrombolysis (mean 153.15 versus 140.91 mmHg) and higher fasting glucose (median 6.91 versus 5.87 mmol/L) than patients without deterioration. In adjusted analysis, each 1 mmHg increase in systolic blood pressure carried an odds ratio of 1.046 for END and each 1 mmol/L increase in glucose an odds ratio of 1.079. The authors also examined the stress hyperglycemia index, a glucose-to-glycated hemoglobin ratio intended to capture relative hyperglycemia, but found it was not significantly associated with END after adjustment, apparently because it correlated strongly with absolute glucose in their data. They note that a formal stratification by diabetic status was not feasible given the small number of END events among diabetic patients, leaving open whether admission glucose predicts deterioration differently in diabetic and non-diabetic populations.

Several methodological safeguards shaped the analysis. Because symptomatic intracranial hemorrhage overlaps definitionally with END and occurs only after thrombolysis, the authors excluded it from the multivariable model and ran a sensitivity analysis removing all 17 patients with symptomatic hemorrhage; the base model’s performance remained essentially unchanged (AUC 0.719 versus 0.724), and all subsequent biomarker analyses were conducted on this hemorrhage-free cohort. A further sensitivity test showed that adding matrix metalloproteinase 9 to the extended model raised the AUC only marginally, from 0.909 to 0.912, with a statistically significant but tiny Integrated Discrimination Improvement of 0.019, so the final nomogram retained five predictors: systolic blood pressure, glucose, ILA, IPA, and interferon gamma. Calibration was assessed with Hosmer-Lemeshow tests and calibration curves, overall accuracy with a Brier score of 0.07, and clinical usefulness with decision curve analysis, all of which supported the extended model.

The study’s limitations are explicitly acknowledged and matter for interpretation. Recruitment came from a single center, and the model was evaluated only internally, without external validation, so generalizability to other populations remains untested. The ELISA-based measurement of ILA and IPA, rather than liquid chromatography-tandem mass spectrometry, may reduce the accuracy of absolute concentrations, although relative group differences held up in regression. The authors also note that residual confounding from unmeasured factors is possible and that the biomarker panel may not capture all relevant prognostic information. They frame the nomogram as a tool for early risk stratification that could guide intensified neurological monitoring in high-risk patients, while emphasizing that the model will need testing on datasets from multiple external institutions before its real-world applicability can be judged. Within those boundaries, the study offers a concrete demonstration that serum markers linking gut microbial metabolism and inflammation to stroke progression can add measurable predictive value beyond blood pressure and glucose in thrombolyzed stroke patients.

Subject of Research: Prediction of early neurological deterioration after intravenous thrombolysis in ischemic stroke using gut-derived tryptophan metabolites and inflammatory biomarkers

Article Title: Gut microbiota-derived tryptophan metabolites and inflammatory factors enhance prediction of early neurological deterioration after intravenous thrombolysis in stroke patients: a comparative model study

Article References: Li, W., Feng, Z., Zeng, C., Cao, Z., Bai, Y., Wang, H., & Huang, J. (2026). Gut microbiota-derived tryptophan metabolites and inflammatory factors enhance prediction of early neurological deterioration after intravenous thrombolysis in stroke patients: a comparative model study. Frontiers in Neurology, 17, Article 1934189. https://doi.org/10.3389/fneur.2026.1934189

Image Credits: AI Generated

DOI: 10.3389/fneur.2026.1934189

Keywords: ischemic stroke, thrombolysis, early neurological deterioration, gut microbiota, tryptophan metabolites, indole-3-propionic acid, indole-3-lactic acid, interferon gamma, nomogram, predictive model, biomarkers, neuroinflammation