• Low-dose IL-2 is efficacious in steroid-refractory cGVHD, with objective responses in .50% of patients, and durable disease control.• IL-2 initiation earlier after cGVHD onset, prior to severe impairment of Treg:Tcon ratios, improves likelihood of clinical response.Chronic graft-versus-host disease (cGVHD) is associated with inadequate reconstitution of tolerogenic CD4 1 CD25 1 FOXP3 1 regulatory T cells (Tregs). Previous phase 1 studies identified a low daily dose of interleukin-2 (IL-2) that was well tolerated, did not exacerbate alloimmunity, augmented Treg in vivo, and was associated with improvement of active cGVHD. In the current phase 2 study, 35 adults with steroid-refractory cGVHD received daily IL-2 (1 3 10 6 IU/m 2 ) for 12 weeks. Median time from transplantation and cGVHD onset was 616 days (range, 270-2145 days) and 317 days (range, 28-1880 days), respectively. Two patients withdrew and 5 required IL-2 dose reductions due to side effects. Twenty of 33 evaluable patients (61%) had clinical responses at multiple cGVHD sites (liver, skin, gastrointestinal tract, lung, joint/muscle/fascia). Three patients (9%) had progressive cGVHD. Compared with pretreatment levels, Treg and natural killer cell counts rose >fivefold (P < .001) and >fourfold (P < .001), respectively, without significant change in conventional CD4 T cells (Tcons) or CD8 T cells. The Treg:Tcon ratio rose >fivefold (P < .001). Clinical responders initiated IL-2 earlier (508 vs 917 days after transplantation, P 5 .005; 249 vs 461 days after cGVHD onset; P 5 .03). Treg:Tcon ratios ‡0.07 at baseline and ‡0.2 at week 1 also predicted clinical response (P 5 .003; P 5 .0003, respectively). After a 4-week treatment hiatus, clinical responders were eligible to continue IL-2 therapy indefinitely. During 2 years of extended IL-2 therapy, clinical and Treg immune responses persisted, while Tcon count and Treg:Tcon ratio gradually normalized. Lowdose IL-2 provides durable clinical improvement in active cGVHD and extended therapy is well-tolerated. (Blood. 2016;128(1):130-137)