A Database-Derived Phthalate Ester-Ankylosing Spondylitis Signature Identifies an AP-1/CXCL8 Inflammatory...

Objectives: We tested whether a database-derived phthalate ester (PAE)-ankylosing spondylitis (AS) candidate panel identifies an inflammatory transcriptional program and characterized its transcription-factor (TF) architecture. Methods: Machine learning prioritization and repeated nested cross-validation were followed by locked-model transfer from GSE73754 to GSE25101. Donor-level single-cell analyses included 96,746 peripheral blood mononuclear cells from 10 AS and 29 healthy-control donors (GSE194315), with corroboration assessed in 25 additional baseline AS patients (GSE277117). Results: Within classical monocytes, the continuous signature covaried with panel-excluded TNF-α/NF-κB signaling (ρ = 0.571; q = 0.000729), inflammatory response (ρ = 0.447; q = 0.0108), and eight of 10 AP-1-related TF activities. Target-excluded TF activities covaried with CXCL8 (8/10) and IL1B (10/10). AP-1-gene/CXCL8 co-expression received partial corroboration in the additional AS cohort (JUN-CXCL8: ρ = 0.92). Among 16 genes prioritized from 473 shared candidates, CXCL8, IL2RB, STAT5B and TNF were stable. The locked 14-gene model achieved an area under the receiver-operating-characteristic curve of 0.770 (DeLong 95% confidence interval, 0.596-0.943), supporting partial rank transportability. Secondary analyses showed a lower CD56bright fraction among natural killer (NK) cells in AS (-2.699 percentage points; 95% confidence interval, -4.575 to -0.611; q = 0.0498) and reduced NK-cell JUN expression (log₂ fold change = -0.995; adjusted p = 0.0497). Conclusions: The PAE-AS signature identifies an AP-1/CXCL8-associated classical-monocyte inflammatory program, with partial cross-cohort corroboration and secondary NK alterations, providing candidates for exposure-informed mechanistic studies.