The healthy pediatric oropharyngeal colonization landscape: Mycoplasma pneumoniae as a near-specific disease signal at shotgun depth and load-stratified DNA viruses in pneumonia

Abstract Interpreting a pediatric pneumonia metagenome requires knowing which organisms colonize the healthy oropharynx. We re-screened 246 pediatric oropharyngeal shotgun metagenomes from a Shenzhen cohort (171 healthy, 75 with pneumonia) and 1,803 runs from twelve external pediatric cohorts using read-level, dual-verification mapping with an a priori ≥ 2-read positivity rule. The healthy landscape was streptococcus-dominated ( Streptococcus pneumoniae –mitis complex 99.4%; Haemophilus influenzae 93.0%). Against this backdrop, Mycoplasma pneumoniae (MP) broke the colonization pattern as a near-specific disease signal: 29/51 (56.9%) oropharyngeal samples in MP pneumonia versus 1/170 (0.6%) in healthy children (Fisher p = 2.6×10⁻²¹; odds ratio 214, exact 95% CI 32–9,147) and 4/24 (16.7%) in non-MP pneumonia (0/24 at ≥ 5 reads), and 0/802 quality-control-passed external carriage samples (95% CI 0–0.5%) — no detectable healthy carriage at shotgun depth. High-load samples were P1-1 and CARDS-toxin positive; the highest carried 23S A2063G in all 16 site-covering reads. Adenovirus was enriched in MP pneumonia (66.7% vs 1.2% healthy; most detection reflects oropharyngeal shedding), with no healthy detection above relative abundance 10⁻⁴. At 56.9% sensitivity and 99.4% specificity, oropharyngeal detection is a strong rule-in signal (positive predictive value 97.6% at 30% prevalence) but cannot rule out disease. Read-level screening calibrated against healthy carriage distinguishes infection from background in public metagenomic archives. Importance Metagenomic sequencing is entering routine diagnosis of childhood pneumonia, but “detected” is too often misread as “causing disease”, because many microbes colonize the healthy airway. We mapped the healthy pediatric oropharyngeal colonization landscape and then asked which detections break it. Streptococci and Haemophilus influenzae were near-universal healthy colonizers, but Mycoplasma pneumoniae was not: across 802 community and carriage samples from healthy children on three continents, we found no detectable carriage at sequencing depth. At that depth, a metagenomic report of this organism in a sick child’s airway sample strongly supports infection over background carriage. The same colonization-calibrated logic separates clinically meaningful adenovirus detection from harmless trace shedding: only high-level detection signals meaningful virus. Because oropharyngeal sampling underestimates lower-airway infection, detection rules in but non-detection cannot rule out.