Preclinical evaluation of a dnTGFbetaR-armored GPC3 CAR-T (AZD5851) in pediatric solid tumor PDX models - A report from the Pediatric Preclinical In Vivo Testing (PIVOT) Consortium

Introduction GPC3 is a heparan sulfate proteoglycan aberrantly expressed in pediatric tumors such as hepatoblastoma (HBL), Wilms tumor (WT), germ cell tumors (GCT), malignant rhabdoid tumor (MRT), and rhabdomyosarcoma (RMS). AZD5851 is an armored GPC3-targeted CAR-T incorporating a dominant-negative TGF-beta receptor II (dnTGFbetaRII) to resist TGF-beta-mediated inhibition and enhance functionality. AZD5851 was evaluated in pediatric patient-derived xenograft (PDX) models to assess antitumor activity and associations between GPC3 and TGF-beta expression. Methods PDX models were characterized by IHC for GPC3 (H-score) and TGF-beta (% cells ≥1+ staining in tumor and stroma). Stromal TGF-beta was classified as low (≤30%), moderate (31-60%), or high (>60%). Nine PDX models (3 WT, 2 HBL, 2 GCT, 1 MRT, 1 RMS) were treated in NSG MHC DKO mice (Jackson Laboratory) with 5 × 10⁶ AZD5851 or untransduced T cells intravenously. Tumor responses were categorized per established PIVOT metrics (Houghton et al., 2007) using an event definition of >4-fold increase in tumor volume. For EFS analyses, a complementary definition of tumor volume >1,000 mm³ or moribund status was also applied to account for delayed tumor regression typical of CAR-T therapies. EFS distributions were estimated by Kaplan-Meier analysis and compared using the log-rank test. Results PDX models demonstrated variable GPC3 expression (median H-score = 90, range 0-254). TGF-beta staining was low in tumor cells (<5-20%) but prominent in stroma (40-90%, median ~80%). Only models with GPC3 H-scores >200 (2 HBL, 1 RMS) showed partial responses (PR) or stable disease (SD). One HBL model exhibited initial progression followed by regression after day 21. Models with H-scores <200 showed only PD. EFS analysis showed that AZD5851 significantly prolonged EFS in models with high GPC3 expression (H-score >200; p = 0.002-0.01, log-rank) compared with controls, while models with lower expression (≤100) showed no significant differences ( p ≥ 0.2). Responses occurred in models with both moderate (40%) and high (90%) stromal TGF-beta, suggesting dnTGFbetaRII armoring may protect CAR-T cells from TGF-beta-mediated suppression. Conclusions AZD5851 demonstrated tumor-regressing activity in pediatric PDX models with high GPC3 expression (RMS, HBL). Activity across models was heterogeneous, with evidence of delayed responses following initial progression. As seen with other CAR T-cell therapies, delayed activity may reflect in vivo proliferation and expansion of infused cells, and standard event-based metrics may not fully capture this effect. These findings support clinical evaluation of GPC3 CAR T cells with dnTGFbetaR armoring for GPC3-expressing pediatric cancers such as RMS and HBL and suggest that GPC3 expression by IHC may serve as a predictive biomarker of response.