Oral microbial signatures of head and neck cancer patients highlight diverse longitudinal patterns of oral mucositis severity

Oral mucositis is a painful complication common in head and neck cancer care. Shifts in the oral microbiome have been associated with oral mucositis, motivating analyses of microbial signatures that may inform future research. We aimed to (i) characterise longitudinal microbial patterns aligned with mucositis severity, (ii) identify clinically meaningful patient clusters by severity trajectory, and (iii) identify cluster-specific microbial candidates. Weekly symptom ratings and buccal swabs (16S rRNA) were collected longitudinally from 140 head and neck cancer patients. An oral mucositis symptom composite score was derived by applying non-negative sparse principal component analysis to all symptoms. Functional data analysis captured individual severity curves, and hierarchical clustering grouped patients by trajectory. Partial least squares knockoff analysis was used as an exploratory screen for taxa associated with the score within each cluster, and the resulting candidates were assessed using repeated-measures linear mixed models with patient random effects. Three clusters emerged. One showed a rapid rise in mucositis severity soon after treatment initiation, whereas two displayed more gradual increases. Age and weight differed descriptively across trajectory clusters but did not remain significant after multiple-comparison adjustment. The exploratory screen selected Prevotella OTUs across clusters and Alloprevotella OTUs in the rapid-progression cluster and one gradual-progression cluster. After repeated-measures correction, genera linked to oral health, including Rothia and Actinomyces , were negatively associated with severity, and one Alloprevotella OTU showed a positive association confined to the rapid-progression cluster; several screened taxa, including the most prominent Prevotella OTU, did not reach the corrected threshold. Distinct mucositis trajectories were accompanied by differing microbial profiles and descriptive demographic patterns. Because the microbial screen treated repeated samples as independent, these microbial findings are hypothesis-generating and require confirmation in studies designed for longitudinal microbiome inference.