Fig. 1

Network analysis of synovial RNA sequencing in early RA reveals gene-gene interactions uniquely linked to the lympho-myeloid pathotype. A Analytical pipeline using network approach to extract informative networks and predictive gene pairs from RNA-seq profiles. Having defined subgroups of patients, an extensive network of interactions is built using merged KEGG pathways enriched with micro-RNAs and transcription factors. The network is replicated for each subgroup and the average expression level of each gene in a subgroup is used to infer a weight on each network node. A first filtering step removes, from each network, nodes (genes) whose weight (subgroup average expression level) is below an optimal threshold obtained via percolation analysis. The second filtering step pull out links (gene-gene interactions) overlapping two or more networks. Robust linear regression with interaction term is used to extract significant gene-gene links. A logistic regression model is built for each significant gene-gene pair to predict response. Ability to predict response is tested by receiver operating characteristic (ROC) curve analysis. B Network of unique active interactions in the lympho-myeloid pathotype. Clusters LM1-LM4. Selected clusters of interest. Labels are determined by gene ontology (GO)/pathway enrichment analysis. Percentages indicate the number of cluster genes included in the associated GO/pathway term. Cluster LM1. Cluster of chemokines needed for leukocyte recruitment (93.5% enrichment). Cluster LM2. Antigen processing and presentation with T cell activation genes (100% enrichment). Cluster LM3. Group of focal adhesion genes comprising collagens, integrins and laminins (93.9% enrichment). Cluster LM4. TNF signaling through mTOR (48.8% enrichment). Cluster LM5. Interferon regulation signaling (87.5% enrichment). Cluster LM6. Genes of the intrinsic and extrinsic apoptotic pathways (50.8% enrichment). C Correlation plots showing differential gene-gene correlations with interactions associated with pathotype. Statistical analysis by robust linear regression model. p-value of the gene to pathotype interacting term is shown. Correlation plots of gene pairs CD28 and PIK3R1, CD79A and LYN, and TNC and ITGB7 across different pathotypes