Another trial-and-error candidate in chronic GVHD

In this issue of Blood, Koshy at al¹ report the results of a phase 2 trial evaluating abatacept in the treatment of advanced chronic graft-versus-host disease (cGvHD), demonstrating an overall response rate (ORR) of 58% and a failure-free survival rate at 3 years of 66%. The response was accompanied by a favorable toxicity profile, confirming the results of a prior phase1 trial.² Interesting to note is that, aside from oral (ORR 42%), gastrointestinal (ORR 50%), and hepatic (ORR 54%) involvement, bronchiolitis obliterans syndrome (BOS; usually associated with relatively low success rates) was the manifestation with the highest response rate (57%), including 4 patients who improved from moderate to mild severity, and 1 patient who improved from having a severe to having a moderate symptom lung score. Improvement in BOS was observed previously in the phase 1 trial, with 3 patients with moderate BOS achieving organ complete remission, an effect that also was described within a retrospective multicenter analysis.³ 

The results reported in this trial for BOS are striking; so, should all lung manifestations be treated primarily with abatacept? The answer is certainly no, as the total number of patients with BOS treated within the 2 trials is low (n = 26). Phase 2 trials are not powered to produce robust response rates, and often, the response rates are lower in larger studies. Although the phase 1 and 2 trials of abatacept in cGvHD used the National Institutes of Health (NIH) response assessment criteria, challenges comparing response rates across different cohorts remain, as shown by a meta-analysis by an Italian group.⁴ Most importantly, almost half of the patients with BOS either did not respond or had progression of their condition. Moreover, the approved therapeutics, such as ruxolitinib⁵ and belumosudil,⁶ show activity in BOS as well. Although the overall ORR was lower, compared to that in the REACH 3 trial evaluating ruxolitinib (ORR 76.4%),⁵ a direct comparison of ruxolitinib to abatacept would be needed to determine the ORR for each, using similar patients in a clinical trial. For example, the REACH 3 trial included only second-line patients, and the median number of prior treatment lines in the current phase 2 trial was 3 (maximum: 8!). This also explains in part the lack of complete responses, which are rarely seen after failure of multiple treatment lines. A more appropriate comparable trial would be that of ibrutinib with an ORR of 67%, taking into account that the majority of patients in the latter trial had steroid-dependent mucocutaneous cGVHD, and high-risk organ manifestations such as hepatic involvement and BOS were present in a minority of patients only.⁷ Belumosudil led to an even higher ORR of 74%, including response in pulmonary manifestations; but again, comparing response rates of phase 2 trials is fraught with difficulties. A common theme in treatment of cGVHD is the “trial-and-error” approach. Even an ORR of 74%, as in the ROCKstar trial, indicates the lack of response in a quarter of patients. The reported failed prior treatment modalities, including ruxolitinib and ibrutinib in studies such as the ROCKstar trial (belumosudil),⁶ suggest that treatment failure is due to the underlying heterogenous disease biology.⁸ In fact, 4 patients in the current abatacept trial failed prior treatment with ruxolitinib. An urgent need exists for a biomarker that predicts response to different classes of agents, to enable biology-driven treatment designed to prevent progression of cGvHD toward nonreversible changes.⁸ Although the presented trial performed a selected correlative immune analysis, the biomarker showed changes associated with treatment but did not predict response—a common theme also observed in other trials.

Traditional trial designs that ignore the biological heterogeneity of cGvHD will most likely fail to resolve the problem of having to use trial-and-error. Future trials should combine clinical and biological response assessment, including an unbiased multiomics approach (until suitable biomarkers are identified), including microbiome assessment,^8,9 with sampling before the start of new treatment, and repeated analysis at 4-to 8 weeks thereafter, as suggested by the NIH consensus 2020. Such an approach would require private–public partnerships (combining industry-sponsored trials with public-sponsored biomarker programs) or third-party nonprofit support.¹⁰ Until predictive biomarkers are available, abatacept will remain another candidate in the system of trial and error, but it certainly deserves further evaluation within clinical trials.

Conflict-of-interest disclosure: D.W. received honoraria from Novartis, Incyte, Sanofi, and Mallinckrodt.