Background: Because of a relative dearth of longitudinal studies, the directionality of the relationship between mood and inflammation among patients with bipolar disorder (BD) is still unclear. We aimed to investigate the longitudinal associations of pro-inflammatory markers with mood symptom severity in BD. Methods: Hundred and thirty-two adult patients with BD were enrolled. At the baseline and 1-year follow-up visit, all participants received mood assessment with Montgomery Åsberg depression rating scale (MADRS) and Young mania rating scale, and underwent blood draws to quantify metabolic profile and serum levels of the pro-inflammatory markers, including soluble interleukin-6 receptor, soluble tumor necrosis factor-α receptor type 1 (sTNF-αR1), monocyte chemoattractant protein-1, and C-reactive protein. A four-factor model of MADRS, consisting of sadness, negative thoughts, detachment, and neurovegetative symptoms, were applied. Results: At baseline, 65 patients with BD were in depressed state, and 67 patients with BD were in euthymic state. Among patients in depressed state, baseline MADRS total score positively correlated with sTNF-αR1 level at follow-up. While baseline sTNF-αR1 level positively predicted sadness symptom in euthymic patients with BD who later developed depression (n = 22), sadness in patients with bipolar depression predicted later increase in serum sTNF-αR1 level even after remission (n = 17). Moreover, lithium had a stronger effect of lowering peripheral sTNF-αR1 level as compared with other mood stabilizers. Conclusion: Our results indicate the bidirectional inflammation-depression relationship in BD.