Rheumatoid arthritis (RA) is a debilitating joint disease characterized by chronic inflammation, pathologic alteration of fibroblast-like synoviocytes (FLS), destruction of cartilage and bone, and the formation of an invasive pannus. RA-FLS exhibit increased proliferation and resistance to apoptosis. The retinoid X receptor (RXR) has a role in regulating cell cycle, differentiation and apoptosis, and agonism of RXR has been investigated as a treatment strategy in several types of cancer. However, there is little research on the effects of RXR agonism in other diseases. Bexarotene is a novel selective RXR ligand used in the treatment of T-cell lymphoma. In the present study, bexarotene was used to investigate the involvement of RXR in tumor necrosis factor-α (TNF-α)-induced RA conditions in human FLS. To the best of our knowledge, this is the first time that RXR has been demonstrated to be expressed in FLS and to be downregulated in response to TNF-α stimulation. The present study also demonstrated that bexarotene exerted an anti-inflammatory effect by downregulating expression of interleukin (IL)-6, IL-8, monocyte chemoattractant protein-1, and high mobility group box-1. Notably, bexarotene also rescued the TNF-α-induced downregulation of the anti-inflammatory cytokines IL-4 and transforming growth factor-β1. Bexarotene treatment exhibited a potential protective effect against cartilage degradation by downregulating the expression of matrix metalloproteinase (MMP)-1, MMP-3 and MMP-13. In addition, the present results demonstrated that the effects of bexarotene were mediated through the p38 mitogen-activated protein kinase/nuclear factor-κB pathway, via inhibition of p38 protein and the inhibitor α of κB phosphorylation. Taken together, the present findings demonstrated the potential of RXR agonism using bexarotene as a treatment against the development and progression of RA.

Abstract. We propose a new deep learning network, i.e. deep synergetic neural network (DSNN), for object recognition. DSNN is constructed by a top-down manner, and it can overcome the problem of pseudo-state with the traditional neural network. To verify the performance of DSNN in object recognition, experiments are performed on two famous databases, i.e. ORL face library and MNIST handwritten library. The experimental results show that the proposed DSNN outperforms the same class of algorithm DBN.

No abstract