Leflunomide (LEF) is a first-line disease-modifying antirheumatic drug (DMARD) for rheumatoid arthritis (RA). However, there are still a few nonresponders. It is logical to suggest that employing combinations including LEF that produce synergistic effects in terms of pharmacological activity is a promising strategy to improve clinical outcomes.

We propose a novel approach for predicting LEF combinations through investigating the potential effects of drug targets on the disease signaling network. We first constructed an RA signaling network with disease-associated driver genes. Thousands of available FDA-approved and investigational compounds were then selected based on a drug-RA network, which was generated using an algorithm model named synergistic score that combines chemical structure, functional prediction and target pathway. We then validated our predicted combination in a prospective clinical trial.

Ligustrazine (LIG), a key component of the Chinese herb Chuanxiong and an approved drug in China, ranked first according to synergistic score. In the clinical trial, after 48 weeks, the American College of Rheumatology (ACR) 20 response rate was significantly lower (P < 0.05) in the LEF group [58.8% (45.4%, 72.3%)] than in the LEF + LIG group [78.7% (68.5%, 89.0%)]. Consistently, the erosion score was lower in patients treated with LEF + LIG than in those treated with LEF (0.34 ± 0.20 vs 1.12 ± 0.30, P < 0.05).

Our algorithm combines structure and target pathways into one model that predicted that the combination of LEF and LIG can reduce joint inflammation and attenuate bone erosion in RA patients. To our knowledge, this study is the first to apply this paradigm to evaluate drug combination hypotheses.

The online version of this article (10.1186/s13020-019-0247-8) contains supplementary material, which is available to authorized users.