Background/Purpose: Periodic Fever, Aphthous Stomatitis, Pharyngitis, and Adenitis (PFAPA) is the most common periodic fever presenting with frequent episodes of pain and impaired function. While its etiology remains unknown, evidence suggests a genetic component and dysregulation of innate immunity. Our objective is to develop transcriptomic insights into the molecular pathways predicting PFAPA flares.

Methods: Ten PFAPA patients, fulfilling consensus patient characteristics, had blood samples collected during flare and non-flare times. Differences between groups were analyzed with the paired t-test. Differentially Expressed Genes (DEGs) were analyzed with standard RNA sequencing pipelines, including quality control, alignment to the reference genome (hg39), and quantification of transcript levels (STAR). Genes with significant differential expression were identified (DESeq2) and analyzed for pathway enrichment (gseapy). We trained a random forest (RF) classifier on gene expression counts to distinguish between flare and non-flare states, obtaining a final model with area under the ROC curve of 1. Top-ranked features’ biological relevance to understand key pathways driving disease flares were evaluated.

Results: The PFAPA cohort’s age of onset was 32 (IQR; 12-78) months, maximum temperature of 104.9 (IQR 103.4-105.2) degrees Fahrenheit, fever duration of ~4 (IQR:3-4.25) days and an 8 out of 10 in severity (IQR: 6-8,9) with how an episode feels and affects function. During flares, 100% of patients had pharyngitis, 80% cervical lymphadenopathy, and 50% oral aphthae. Lab abnormalities with flare included leukocytosis, neutrophilia, monocytosis, lymphopenia, and increased CRP (P < 0.01). A clear distinction between flare and non-flare gene expression was identified by RNA-sequencing Principal Component Analysis with multiple DEGs (Fig. 1). The top 50 DEGs by adjusted p-value included upstream regulators of immune pathways (CD177, GPR84, C1QC, and CARD17P). Gene Set Enrichment Analysis identified neutrophilic, interferon, response to organism and protein processing as key inflammatory pathways. Differentially regulated pathway signatures are depicted in Fig. 2. The top 20 features from the RF of flare versus non-flare were related to cell cycle arrest and microtubule polymerization. The highest-ranking feature was microRNA 5195 (downstream effectors Fig. 3), which at low levels was most predictive of flare.

Conclusion: We identified distinct gene expression profiles between PFAPA states demonstrating a transcriptomic signature of immune dysregulation associated with flares. The flare state reflects neutrophilic activation (modulated by corticosteroids aborting fever), interferon signaling, innate immune dysregulation and heightened cellular activity manifesting as cell cycle arrest as well as microtubule polymerization (modulated by colchicine). The MicroRNA 5195 gene was identified as the highest-ranking predictive feature of PFAPA flares. Future work incorporating machine learning models in prioritizing biologically relevant markers in comparison to other cohorts will enable future development of a unique potential PFAPA biomarker.

Figure 1: (Left) Principal Component Analysis Showing a Clear Separation and (Right) MA Plot of Log Fold Change Between PFAPA Flare and Non-Flare Plotted Against Mean Gene Expression. Significantly altered genes at p value < 0.05 are marked in color .

Figure 2: Gene Expression Signatures between PFAPA Flare and Non-Flare for Gene Ontologies Associated with Inflammation

Figure 3: Network Motif of MICRO-RNA 5195: Downstream Effectors

« Back to ACR Convergence 2025

ACR Meeting Abstracts - https://acrabstracts.org/abstract/transcriptomics-and-machine-learning-unraveling-the-molecular-drivers-of-pfapa-flares/