Artificial intelligence-powered spatial analysis of endothelial cells and tumor-infiltrating lymphocytes to predict response to axitinib in adenoid cystic carcinoma

  • Kim, Dong Hyun; 
  • Hwang, Woochan; 
  • Lee, Seungeun; 
  • Lim, Yoojoo; 
  • Ali, Siraj Mahamed; 
  • ... Kang, Eun Joo; 
  • 외 10명

초록

Background: Despite the limited systemic options for adenoid cystic carcinoma (ACC), VEGFR inhibitors remain a clinical mainstay. Although high stromal tumor-infiltrating lymphocyte (TIL) density has been identified as a predictive biomarker for improved progression-free survival (PFS), its predictive power remains to be fully optimized. We hypothesized that baseline vascular architecture, represented by endothelial cell (EC) density, might modulate the efficacy of axitinib. Methods: We performed a post-hoc exploratory analysis on H&E-stained whole-slide images (WSI) from 27 patients with R/M ACC treated with axitinib in a multicenter phase II trial (NCT02859012). An updated AI-powered analyzer (Lunit SCOPE IO), capable of multiple component TME profiling, was used to quantify the density (cells/mm²) of TILs and ECs within the tumor epithelium and stroma. Patients were stratified into subgroups based on the median values. The clinical impact of integrated immune and vascular architecture was evaluated by analyzing PFS and OS. Results: The analyzed cohort (N=27) had a best objective response of stable disease in 25 patients (92.6%) while 16 patients showed tumor shrinkage (59.3%). Stratification revealed that patients with concurrent High EC and High TIL density in the tumor stroma (n=9) derived exceptional clinical benefit compared to all other patients (N=18). The High EC/High TIL subgroup achieved a median PFS of 19.6 months compared to 11.1 months in the comparator group (HR 0.30; 95% CI: 0.11–0.87; P=0.026). Furthermore, this subgroup demonstrated significantly prolonged OS (median NR vs. 24.4 months; HR 0.12; 95% CI: 0.02–0.95; P=0.044). Stratification based on intratumoral densities of EC and TILs showed a similar trend with the High EC/High TIL subgroup (n=10) reporting prolonged PFS (HR 0.32; 95% CI: 0.12-0.87; P=0.025) but not OS (HR 0.46; 95% CI: 0.12-1.74; P=0.251). The individual biomarkers based on median TIL and EC showed a trend towards prolonged survival but were not statistically significant. Conclusions: The co-enrichment of stromal ECs and TIL is associated with significantly prolonged PFS and OS, suggesting that this unique TME architecture may identify R/M ACC patients who derive greater clinical benefit from axitinib. This AI-based spatial analysis using H&E slides offers a practical, scalable biomarker strategy to guide treatment selection in this rare cancer.

제목
Artificial intelligence-powered spatial analysis of endothelial cells and tumor-infiltrating lymphocytes to predict response to axitinib in adenoid cystic carcinoma
저자
Kim, Dong Hyun; Hwang, Woochan; Lee, Seungeun; Lim, Yoojoo; Ali, Siraj Mahamed; Kang, Eun Joo; Ahn, Myung-Ju; Lee, Keun-Wook; Kwon, Jung Hye; Yang, Yaewon; Choi, Yoon Hee; Kim, Min Kyoung; Ji, Jun Ho; Yun, Tak; Kim, Sung-Bae; Keam, Bhumsuk
DOI
10.1200/JCO.2026.44.16_suppl.6122
발행일
2026-05-30
학회명
2026 ASCO Annual Meeting (American Society of Clinical Oncology Annual Meeting)
개최지
Chicago, USA
개최국가
미국
학회 개최일
2026-05-29 ~ 2026-06-02