Spine Surgery and Related Research
Online ISSN : 2432-261X
ISSN-L : 2432-261X
Metabolite-Based Unsupervised Clustering Classifies Adult Spinal Deformity into TPI-Centered Phenotypes with Distinct Curve-Type and Clinical Features
Mitsuru Yagi, Ryo Mizukoshi, Keitaro Ito, Norihiro Isogai, Haruki Funao, Retsu Fujita
Author information
JOURNAL OPEN ACCESS Advance online publication
Supplementary material

Article ID: 2026-0303

Details
Abstract

Introduction: Adult spinal deformity (ASD) is a heterogeneous disorder in which radiographic deformity, disability, and systemic biological vulnerability interact. Although the SRS-Schwab classification is clinically useful, it does not directly account for metabolic or inflammatory states that may influence symptom burden and disease phenotype. The trimethylamine N-oxide pathway index (TPI), a composite score reflecting microbiome-linked pathway activity, may provide complementary information. However, whether global metabolomic profiles form TPI-centered ASD phenotypes remains unclear.

Methods: This secondary cross-sectional analysis included 120 adults with primary ASD from a prospective single-center multi-omics cohort. Ward hierarchical clustering was applied to 212 standardized plasma metabolites. TPI and clinical variables were excluded from clustering; TPI was used only post hoc to order and label clusters. Candidate k = 2-5 solutions were evaluated using internal validation metrics. Cluster characteristics, metabolite signatures, SRS-Schwab curve type, clinical outcomes, and age-adjusted sensitivity analyses were assessed.

Results: Clustering identified Low-TPI (n = 56), Intermediate-TPI (n = 43), and High-TPI (n = 21) phenotypes. Cluster membership was associated with SRS-Schwab curve type (chi-square = 16.71, p = 0.010): N-type curves were enriched in the high-TPI cluster, whereas T-type curves were absent. However, age differed both among clusters (p = 0.049) and across curve types (p = 1.02 × 10−7). In an age-adjusted sensitivity analysis focused on N-type versus non-N-type curves, High-TPI status was not independently associated with N-type classification (OR 1.77, 95% confidence interval [CI] 0.62-5.07, p = 0.290). The High-TPI phenotype was enriched for gamma-glutamyl, branched-chain amino acid, and glycation-associated metabolites and depleted in uracil, Pro-Hyp, and succinic acid. Adjusted models showed borderline lower Oswestry Disability Index scores in the Intermediate-TPI versus High-TPI cluster (beta = −9.08, p = 0.050). TPI correlated most strongly with the gut/trimethylamine N-oxide axis (rho = 0.75, q = 1.4 × 10-22), followed by Glycation/Carbonyl and ROS/Redox axes.

Conclusion: Metabolite-based clustering identified exploratory TPI-centered ASD groups with distinct metabolic signatures. TPI appears to track broader multi-axis metabolic variation, whereas cluster–curve-type associations should be interpreted cautiously because they are partly age- and disease-heterogeneity related.

Content from these authors
© 2026 The Japanese Society for Spine Surgery and Related Research.

SSRR is an Open Access journal distributed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License. Anyone may download, reuse, copy, reprint, or distribute articles published in the journal for not-for-profit purposes if they cite the original authors and source properly. If you remix, transform, or build upon the material, you may not distribute the modified material.
https://creativecommons.org/licenses/by-nc-nd/4.0/
Previous article Next article
feedback
Top