All animal experiments were approved and carried out in accordance with the guidelines by the institutional care and use committee (IACUC) at Emory University. FLS were prepared either from wild type or Sox4^fl/fl11^fl/fl12^fl/fl (SoxC ^fl/fl) mice according to a well-established protocol (5, 22). Briefly, fore limbs were separated at the radiocarpal joint while hind limbs were separated at the tibiotalar joint. Skin and nail were removed carefully, and phalanges were separated as to keep the interphalangeal and metacarpophalangeal joints intact. This was followed by a first Collagenase IV (Sigma-Aldrich, 2mg/mL for 30 min) digestion of the isolated joints to remove skin fibroblasts and other surface cells. A second digestion with Collagenase IV (1mg/mL for 2h) was carried out to digest the synovium covering the phalangeal joints. The digests were filtered in a 50-micron cell strainer to remove the undigested bones and debris. The resulting cell suspension containing >90% FLS were cultured in DMEM with 10% fetal bovine serum (Corning) and 1% penicillin/streptomycin for a period of 16h, followed by washing with PBS and changing cell culture medium to remove unattached and dead cells. Upon reaching confluency, the FLS were sub-cultured and utilized between passage 3.

RNA-seq was performed in triplicates of independent preparations of FLS. SoxC^fl/fl FLS at passage 3 were infected with AdCMV5eGFP (control) or AdCMV5Cre (SoxC-knockout) adenovirus (UI Viral Vector Core) at a concentration of 200 pfu/cell, for a period of 24h. FLS were treated with or without 10ng/mL recombinant human TNF (R&D Systems) for the last 16h. Total RNA was extracted and purified using Direct-zol RNA MicroPrep (Zymo Research). Only samples with an RNA integrity number (RIN) >8 were used. Libraries were generated from 250 ng RNA using TruSeq Stranded Total RNA Sample Prep Kit (Illumina). Paired-end 150-base sequence reads at a sequencing depth of 50 million reads per sample were obtained using Illumina HiSeq 2500 System (Novogene). FASTQ files were analyzed using Strand NGS 4.0 software and aligned to the mm10 mouse genome assembly. Quality of sequencing was ensured by plots for measuring per base sequence quality (Q30 = 99%). Duplicate reads were filtered, followed by quantification and normalization using DESeq method.

Chromatin was prepared and immunoprecipitated according to the following procedure. Triplicates of wild type mouse FLS containing 40 million cells per replicate were treated with or without 10ng/mL TNF for 16 h, followed by fixation with 1.5% paraformaldehyde. Chromatin was extracted and sheared into 100- to 500-bp fragments using a Bioruptor (Diagenode). 10 μg of antibodies against p65 (Active Motif) or SOX4 (Diagenode), coupled to 20 μl Dynabeads (Life Technologies) were used for immunoprecipitating chromatin from each replicate and DNA was purified by phenol/chloroform/isoamyl alcohol extraction and ethanol precipitation. DNA pooled from the triplicates was used for library preparation. Single-end 50-base sequence reads were obtained at a sequencing depth of 30 million reads per sample using Illumina HiSeq 2500 System and analyzed using ChIP-seq pipeline on Strand NGS 4.0 software. Quality of sequencing was ensured by plots for measuring per base sequence quality (Q30 = 97.8%). Reads were mapped to the mm10 mouse genome build. Peak calling and gene annotation was performed using MACS with an upstream padding distance of 50kb. De novo motif enrichment analysis was performed using MEME-ChIP suite (23).

Wildtype mouse FLS were treated with TNF (10ng/mL) for 16 h. Nuclear extracts were prepared from FLS using NE-PER™ Extraction Reagents (Thermo Scientific) protocol. Immunoprecipitation was carried out by conjugating 5 μg of p65 antibody (Active Motif) or rabbit non-immune IgG (Millipore) to 20 μl Dynabeads (Life Technologies). Cell extracts were added to the conjugated beads overnight and eluted protein was analyzed by western blotting using antibodies against SOX4 (Diagenode), p65 (Cell Signaling) or HDAC1(Cell signaling).

ChIP-seq enhancer peak regions assigned to Sox4, Sox11, Il15 and Mapk1 were amplified from mouse genomic DNA and cloned into pBV- Luc, a luciferase reporter plasmid with a minimal promoter and very low basal activity (24). These reporter plasmids were co-transfected with 3X-FLAG SOX4 or 3X-FLAG SOX11 expression plasmids (14) into HEK293 cells using Viromer Red reagent (Origene). pSV2bgal plasmid was used as a control for transfection efficiency. Cells were treated with 10ng/mL recombinant TNF (R&D systems) for the last 16h of the 36h transfection period. Luciferase and β-galactosidase activities were assayed using the Dual-light combined reporter gene assay system (Applied Biosystems) using Synergy H1 Multi-Mode Plate Reader (Biotek). Reporter activities were normalized for transfection efficiency and reported as fold change in luciferase activity. Assays were performed as triplicates.

Experiments were performed in triplicates. Differential gene expression in changes in RNA-seq were calculated by Audic Claverie and Benjamini Hochberg false discovery rate for multiple testing correction. p-value cut-off of was set at 0.05. Statistical significance in reporter assays was calculated by student’s t-test.

We previously showed that the transcription factor SOX4 plays a role in promoting FLS transformation and thereby TNF-induced arthritogenesis (14). To understand the mechanism underlying the role SOX4 in TNF-induced gene expression we investigated whether SOX4 and the canonical NF-κB signaling transcription factor RELA/p65 are part of the same transcriptional complex. We predicted a potential physical interaction between and SOX4 and RELA and tested it by a co-immunoprecipitation assay. We immunoprecipitated RELA from the nuclear extracts of wildtype mouse FLS that were treated with or without TNF ( Figure 1A ). As reported earlier the level of SOX4 protein was increased upon TNF-treatment. Interestingly, we found that SOX4 co-immunoprecipitated with RELA in mouse FLS both in the presence and absence of TNF. However, the interaction was higher under TNF-treated condition likely due to the expected increase in the nuclear localization of RELA. To determine whether SOX4 and RELA interaction occurs at the genome wide level and to identify the genes co-regulated by SOX4 and RELA, we developed an experimental design involving ChIP-seq and RNA-seq assays ( Table 1 ). We first performed ChIP-seq using either SOX4 or RELA antibodies. We found that TNF treatment increased the number of SOX4 and RELA binding peaks by 3-fold. The enrichment of sequencing reads from the around the transcription start sites ( Figure 1B ) and box plots ( Figure 1C ) show increased binding of both SOX4 and RELA antibody on the chromatin from TNF-treated FLS in comparison with the untreated FLS. At the genomic level the SOX4 and RELA peaks were predominantly (50-60%) located upstream of TSS. About 20% percent of the peaks were located in the intronic regions. A small percentage (2-6%) of them were localized the 3’ or 5’ untranslated regions and even smaller percentage (>1%) of binding was detected in the coding sequence ( Figure 1D , Supplementary Tables S1 – S4 ). Based on the physical interaction between RELA and SOX4 we speculated that these proteins might bind in close proximity to each other at a genome wide level. We overlapped the peak summits identified from SOX4 and RELA ChIP-seq experiments under TNF-treated condition to find that 70.4% of the SOX4 peak summits were present within 50bp of a RELA peak summit which we labelled as Group 1. The peaks with SOX4 only or RELA only peaks were labeled as Groups 2 and 3, respectively ( Figure 1E , Supplementary Table S5 ). The TSS profile plots and the read density heatmaps show that while read density for Group 1 peaks with SOX4-RELA co-binding is enriched in SOX4 and RELA ChIP-seq and as expected the Groups 2 and 3 peaks show a higher enrichment around TSS only in either SOX4 or RELA peaks respectively ( Figures 1F, G ). The genomic tracks for ChiP-seq peaks at representative examples of Groups 1, 2 and 3 further demonstrate the co-binding and differential binding of SOX4 and RELA ( Figure 1H ). De novo motif search by MEME-ChIP discovered SOX binding motifs as the most frequently identified motif in the SOX4-RELA overlapping peaks, followed by a partial NF-κB as reported for CyclinD1 and Il12β genes (25), SIX2 and SP2 binding motifs. Contrastingly motif analysis of the RELA only peaks did not indicate the enrichment of SOX motifs but were instead enriched with AP-2 and NF-kB binding motifs ( Figure 1I ). To determine the functional roles of the genes co-bound by SOX4 and RELA, we compiled a list of 600 genes that were assigned to the RELA-SOX4 overlapping peaks. Network analysis by Ingenuity Pathway Analysis (IPA) revealed a potential crosstalk between the SOX4 and NF-κB signaling regulated genes ( Supplementary Figure S1A ). At the functional level, the SOX4-RELA genes were predicted to regulate pain signaling pathways, xenobiotic stress and tryptophan metabolism, nitric oxide signaling as well as macrophage, fibroblasts, and endothelial cell activities in rheumatoid arthritis ( Supplementary Figure S1B ).

We next investigated whether SOX4 is required for regulating the global TNF-responsive transcriptome by RNA-seq. We generated SoxC knockout FLS by infecting SoxC ^fl/fl with Cre recombinase-expressing adenovirus. Their corresponding controls were generated by infection with GFP-expressing adenovirus. We utilized total SoxC knockout background instead of a Sox4 knockout background because the SoxC family genes (Sox4, Sox11 and Sox12) are functionally redundant and possesses a highly identical HMG-box domain allowing them bind to the same SOX-binding sites on DNA (17). Therefore, deletion of Sox4 alone may be compensated for by Sox11 and Sox12. We defined TNF-responsive genes as those that were differentially expressed by ≥ 1.5-fold by TNF treatment in the AdeGFP-SoxC^fl/fl FLS. Interestingly, 60% of the TNF-responsive genes remained unchanged or exhibited a reversed regulation in the AdeCre-SoxC^fl/fl FLS indicating that the effect of TNF on FLS is significantly altered in the absence of SoxC genes ( Figure 2A , Supplementary Figure S2A and Supplementary Table S6 ). Pathway analysis revealed that that the upregulated TNF-responsive transcriptome which was either SOXC-dependent and independent is predicted to play a role in inflammatory disease pathways, cell migration and tumorigenesis. Interestingly, the downregulated TNF-responsive genes that are SOXC-dependent were predicted to regulate organogenesis and cell survival ( Figure 2B ). We integrated the differential gene expression data with the ChIP-seq results to find that only a small proportion i.e., 17 of 638 SoxC-dependent TNF-responsive genes were assigned to RELA-SOX4 peaks and 66 genes were assigned to RELA only peaks ( Figure 2C ). The genes with RELA-SOX4 peaks regulated inflammation mediators such as interleukin-15 (Il15), chromatin remodeling factors such as histone deacetylase 4 (Hdac4), lysine acetyltransferase 6b (Katb6), SET-binding protein 1 (Setbp1) and nuclear receptor coactivator 1 (Ncoa1) and Activating protein-1 (AP-1) signaling components such as mitogen-activated protein kinase 1 (Mapk1/ERK2), mitogen-activated protein kinase 8 (Map3k8/MEKK) and JunB proto-oncogene (Junb/AP-1) ( Figure 2D ). To understand the clinical significance of the SOXC-dependent TNF-responsive genes we analyzed a published RNA-seq data set from freshly sorted CD45- Podoplanin+ FLS from the synovia from OA and RA patients (26). We found that the expression levels of SOX4, and SOX11 remained unchanged between the highly inflammatory FLS from leukocyte rich RA patients than in OA patient FLS with low level of inflammation. However, the SoxC-dependent TNF-responsive genes were upregulated in the leukocyte rich RA FLS ( Supplementary Figure S2 ). These data suggest that the SOXC/RELA molecular axis may play a critical role in the pathology of highly inflammatory forms of RA.

Using representative examples of SOX4-RELA co-bound genes, such as Il15 and Mapk1 we showed the extent of overlap between RELA and SOX4 peaks ( Figure 2E ). Their SoxC-dependance was demonstrated by showing that the knockout of SoxC genes prevented their TNF-induced upregulation, whereas the levels of Sox4, Sox11 and Sox12 remained unchanged ( Figure 2F and Supplementary Figure S3A ). The SoxC-dependent regulation of Il15 ( Figure 2G ) and Mapk ( Supplementary Figure S3B ) were additionally demonstrated in a luciferase reporter assay, where SOX4 and SOX11 overexpression in the presence of TNF significantly increased the activity of a luciferase reporter gene encompassing the SOX4-RELA ChIP-seq peak. Mutation of the two SOXC binding sites in the Il15 ChIP-seq peak sequence (position 52: AATCAA to AGATCGA and position 168: AAACAAT to AGACAGT) resulted in a loss of SOXC-dependent activation of the Il15-reporter in the absence and presence of TNF. Similarly, examples of genes assigned to RELA only peaks such as Sequestosome 1 (Sqstm1), a ubiquitin binding protein that plays a role in autophagy and NF-κB inhibitor beta (Nfkbib), critical intermediate in the canonical NF-κB signaling were upregulated by TNF only in the presence of SoxC genes, but the neighboring gene Ccer remained unchanged by TNF or SoxC knockout. Mgatb4 remained unexpressed in the FLS ( Figures 2H, I ).

We previously showed that TNF increases SOX4 and SOX11 protein levels, at least in part by protein stabilization without significantly effecting the mRNA levels (14). Here we addressed the additional mechanisms that might contribute to the TNF-mediated regulation of Sox4 and Sox11 expression. In ChIP-seq assays we found that the 3’UTR of Sox4 and Sox11 genes were bound by RELA under basal conditions, suggesting that RELA may function as an upstream regulator of Sox4 and Sox11 expression. Notably, the RELA binding was lost upon TNF-treatment and this loss corresponded with a gain in SOX4 binding to the same genomic region, suggesting a switch from RELA-mediated expression to an autoregulatory mode of expression ( Figures 3A, B ). We generated luciferase reporter constructs composing the ChIP-seq peak region in the 3’ UTR of Sox4 and Sox11 genes to find that a combination of TNF-treatment and transient over expression of SOX4 was necessary for increasing the reporter activity ( Figure 3C ). In consistence with our previous results, TNF-treatment did not result in the level of Sox4 or Sox11 expression ( Figure 2F ), indicating that a switch in RELA to SOX4 binding did not alter the overall gene expression, but rather helped in maintaining a consistent level of gene expression in the presence of inflammation.