Background
One in 10 infections result in long COVID (WHO, 2023), but little is known about subtypes, causes or treatments.

Methods
Clinical data and plasma were collected from 719 adults, 6 months after hospitalisation via the PHOSP-COVID study. 360 plasma proteins were measured using Olink and penalised logistic regression used to identify predictors of symptoms.

Findings
Markers of monocyte-related inflammation and complement activation predicted all symptoms (fig1). IL1R2 and MATN2 predicted all symptoms except cognitive impairment (CI). COLEC12 predicted respiratory symptoms, fatigue and anxiety/depression, while patients with CI and GI symptoms had distinct inflammatory profiles. Although C1QA predicted both symptoms, markers of neural growth and neuroinflammation predicted CI whilst SCG3, which indicates changes in the brain-gut axis distinguished GI patients. Elevated sCD58, an immunosuppressive factor, was associated with reduced odds of long COVID.

Discussion
We found persistent systemic inflammation in patients with long COVID and different inflammatory profiles with distinct symptom complexes, suggesting these syndromes may be driven by different mechanisms. These findings suggest that trials of therapy for long COVID need to be tailored to individuals based on symptom clusters and patterns of plasma markers.