Recent techniques for the automated detection of online misinformation typically rely on ML models trained with features extracted from content analysis and/or general-purpose Knowledge Graphs (KGs). These techniques often fail to consider the interplay between misinformation and polarization. To bridge this gap, we introduce PARALLAX, a methodology that enhances misinformation detection by infusing polarization knowledge into existing classifiers. Polarization knowledge is represented in terms of Polarization Knowledge Graphs (PKG). PARALLAX constructs PKGs in an unsupervised way, and uses them to enrich articles with polarization knowledge. A Flexible Knowledge-aware Graph Neural Network (FlexKGNN) is trained on these enriched representations. We tested our methodology on three misinformation datasets, demonstrating that it achieves approximately a 15% improvement in performance over baseline classifiers and consistently outperforms other KGs, which typically reach baseline levels only.