Introduction: Danburstotug has demonstrated promising efficacy in patients with R/R ENKTL. However, the relationship between PD-L1 expression and treatment outcomes in ENKTL remains poorly defined, partially due to disease's complex and heterogeneous immune microenvironment. To better characterize the tumor immune landscape and identify potential predictive biomarkers, this study incorporated both quantitative AI–based analysis of PD-L1 expression at the single-cell level and manual classification of the tumor immune microenvironment (TIME) based on immune cell markers. We aimed to determine whether the intensity and subcellular localization of PD-L1 staining, as well as TIME subtypes, are associated with clinical response to danburstotug. These findings may offer novel insights into patients' stratification and immune resistance mechanisms in ENKTL.

Methods:Patients received danburstotug (20 mg/kg every 2 weeks) without prior biomarker-based selection. Tumor responses were assessed every 12 weeks using the lymphoma response to immunomodulatory therapy criteria (Cheson et al. 2016).

A deep learning model utilizing the Lunit SCOPE universal IHC subcellular compartment analysis was applied to segment the membrane, cytoplasm, and nucleus, and to quantify protein expression within each compartment on a continuous scale (0–100). Membrane specificity (MS) was defined as the ratio of membrane intensity to the total intensity across all three compartments, classified to MS-High and MS-Low. Additionally, manual immune subtyping of the TIME was performed to classify tumors into four subgroups—Immune Tolerance (IT), Immune Evasion A (IE-A), Immune Evasion B (IE-B), and Immune Silenced (IS)—based on PD-L1, FoxP3, and CD68 expression. Clinical outcomes were analyzed in relation to MS and TIME classification.

Results:All 23 enrolled patients received the study drug. Median treatment duration was 17.1 months (range, 0.5–26.4). The objective response rate (ORR) was 79%, with a complete response (CR) rate of 63%. Median progression-free survival (PFS) was 29.4 months (95% CI, 12.0–46.9), with a 2-year PFS rate of 62%. Median overall survival (OS) was 40.2 months (95% CI, 25.1-55.4), with a 2-year OS rate of 78%.

PD-L1 membrane specificity (AI-based analysis): Among 18 evaluable patients, 11 were classified as MS-High and 7 as MS-Low. The MS-High showed superior outcomes, with an ORR of 82% vs. 14% (p=0.013) and CR rate of 64% vs. 0% (p=0.013). Median PFS and OS were not reached in the MS-High, whereas the MS-Low had a median PFS of 2.8 months and OS of 7.9 months. The 2-year PFS and OS rates were 61% and 78% in the MS-High, vs. 0% and 43% in the MS-Low, respectively. A significant PFS difference was observed (p=0.019), while OS difference was not statistically significant (p=0.181).

TIME subgroup analysis: Of 20 patients with available TIME classification, 1, 10, and 9 patients were classified as IT, IE-A, and IE-B, respectively; no patients were classified as IS. The IE-B showed numerically higher CR (67%) and ORR (67%) compared to the IT/IE-A group (CR: 36%, ORR: 64%), although not statistically significant (CR: p=0.370; ORR: p=1.000). Median PFS was not reached in the IE-B vs. 8.5 months (95% CI, 0.0-24.5) in the IT/IE-A group, with a significant difference in 2-year PFS rate (100% vs. 40%, p=0.019). Median OS was not reached in either group. The 2-year OS rate was 100% in the IE-B vs. 67% in the IT/IE-A group without statistical significance (p=0.132).

One IT patient had MS-High and achieved CR. In MS-High, the ORR and CR rates were 71% and 43%, respectively, in the IT/IE-A group, and 100% for both ORR and CR in the IE-B, suggesting consistently superior efficacy across IT/IE-A and IE-B in MS-High. In MS-Low, the ORR and CR rates were 33% and 0% in IT/IE-A, and both 0% in IE-B. Notably, patients in the IT/IE-A subgroup with MS-High showed the most favorable outcomes, suggesting that membrane specificity may better predict efficacy than TIME.Conclusions: Danburstotug demonstrated robust and durable efficacy in patients with R/R ENKTL. Clinical benefit was observed across tumor immune subtypes. Importantly, the combination of TIME classification and AI-based PD-L1 membrane specificity analysis identified a subgroup with the greatest clinical benefit, suggesting this combined model may serve as a predictive biomarker.

This content is only available as a PDF.
2025
Sign in via your Institution