FC07 Transcriptomic profiling across tissues and machine learning reveal molecular endotypes of psoriasis phenotypes and signatures of disease severity during biologic therapy

Abstract Despite advances in understanding the pathogenesis of psoriasis, important knowledge gaps exist including (i) whether molecular endotypes of stable plaque psoriasis underlie distinct clinical phenotypes (e.g. subjects with specific comorbidities) and (ii) the positive drivers and negative molecular regulators of disease severity across tissue compartments. We performed a comprehensive RNA-seq study of skin and blood samples (n = 718) from deeply phenotyped discovery (n = 89) and refinement (n = 57) psoriasis cohorts, during targeted treatment with a TNF inhibitor (adalimumab) or an interleukin (IL)-12/23 inhibitor (ustekinumab). As gene expression is controlled by multiple factors across different cell types, network analysis of bulk RNAseq data across large numbers of clinical samples takes into account the underlying latent factors and modules that result in the coordination of gene expression and can more accurately reflect the relationship between transcriptional signatures and clinical phenotypes compared with analysis of individual genes. Using two complementary methods for dimensionality reduction, weighted gene correlation network analysis (WGCNA) and Independent Component Analysis (ICA), we defined distinct but interconnected gene modules and factors within skin and blood that were significantly positively or negatively associated with disease severity as measured by Psoriasis Area Severity Index (PASI). To understand which modules and factors in lesional psoriasis skin contribute to the prediction of disease severity, we applied Gaussian process regression followed by SHAP (SHapley Additive exPlanations). This analysis highlighted 2 principal gene modules [cytokine signalling (267 genes) and Wnt signalling (207 genes)] that explained ∼50% of the variance of PASI during clinical response to both adalimumab and ustekinumab. We derived a further gene signature (14 genes), linked to body mass index in nonlesional skin at baseline, that in lesional skin was significantly negatively associated with PASI. Notably, PASI-associated signatures in blood were only seen following exposure to adalimumab whereas ustekinumab PASI-associated signatures were confined to skin. These highly drug-specific tissue compartmentalization effects indicate a differential systemic impact of adalimumab compared with ustekinumab, in line with differences in side-effect profiles. In contrast, a gene signature in blood closely linked to HLA-C*06:02 status was independent of disease severity (PASI). The systematic nature of the dataset across skin and blood and its integration with key clinical features and outcomes to therapy provides a comprehensive and global perspective of psoriasis that are not apparent from previous studies. Specifically, our findings delineate, across skin and blood, distinct gene-environmental and genetic effects on the psoriasis transcriptome linked to clinical phenotypes and disease severity endotypes.