Immune-metabolic trajectories delineate subgroups in paediatric long COVID

Most children and adolescents recover rapidly from SARS-CoV-2 infection, yet a subset develops paediatric long COVID (LC). How immune ontogeny shapes LC biology and heterogeneity remains unclear. We deeply phenotype a two-visit cohort with severe LC ( n  = 74) and controls ( n  = 27) spanning up to 3.2 years post index infection. Symptom burden remains high and neurofilament light chain (NfL) percentiles inversely associate with functional status (Bell score; r  = −0.3536, P  = 0.0060). Cardiopulmonary assessment and serology are unremarkable. Conventional autoantibodies are not enriched, whereas anti-DFS70 supports subgrouping. Immune features are temporally structured; SARS-CoV-2–associated mediators decline within 1 year, while innate-weighted, Th2-skewed cytokines persist. Metabolomics (43 metabolites) recapitulate the identified subgroups and align with EBV serostatus, disease phase (<1 year versus years 1–3.2), and anti-DFS70 positivity. In EBV-naïve LC, higher haemoglobin concentration (MCHC) tracks worse function, whereas higher IL-12p40, thiamine and basophils track milder impairment (all P  ≤ 0.0170). These data delineate immune-metabolic and haematological axes of paediatric LC heterogeneity and support biomarker-guided stratification.