Loop-structured CD38/CD123 bispecific CAR-T cells for antigen-heterogeneous acute myeloid leukemia

Chimeric antigen receptor T (CAR-T) cell therapy has transformed the treatment of B-cell malignancies, but its application to acute myeloid leukemia (AML) remains limited by heterogeneous antigen expression and the scarcity of leukemia-restricted targets. CD38 and CD123 are expressed across overlapping but non-identical AML compartments; combined recognition may therefore broaden target coverage. We generated two reciprocal loop-structured CD38/CD123 CARs, Loop 123 × 38 CAR and Loop 38 × 123 CAR, and compared them with the monovalent CD38 CAR and CD123 CAR. Two public single-cell RNA-sequencing datasets, GSE241989 and SciLifeLab 23715648, were reanalyzed to assess CD38 and IL3RA expression across malignant cell states, individual AML samples, molecular subtypes, and normal hematopoietic progenitor compartments. Antileukemic activity was evaluated against CD38/CD123-expressing AML cell lines and primary AML samples by cytotoxicity, degranulation, and cytokine assays. CAR-T cell phenotype, T cell numbers, apoptosis, and subset-specific CD38 expression were also examined. Reciprocal single-antigen-loss variants were tested in short-term assays and in a mixed-target co-culture assay followed through day 12. Antitumor efficacy was assessed in cell line-derived, patient-derived, and mixed-antigen-loss xenograft models. Hematopoietic toxicity effects were evaluated by colony-forming assays using human umbilical cord blood (UCB) CD34 + progenitors. Across the two single-cell datasets, CD38 and IL3RA showed substantial interpatient heterogeneity and partially complementary transcript coverage. Longitudinal analysis of the CAR-T cell products showed altered expansion and apoptosis in CD38-targeting products, together with selective depletion of CD38-expressing cells within the live GFP + CAR-T cells. Both loop CAR-T cells recognized both CD38 and CD123 and showed activity against CD38 + CD123 + AML cell lines, primary AML samples, and reciprocal single-antigen-loss cells. During the 12-day mixed-target co-culture, the monovalent CAR-T products could not recognize the corresponding single-antigen-loss population, resulting in accumulation of this uncontrolled population, whereas both loop CAR-T products suppressed the outgrowth of both reciprocal antigen-loss populations. Loop CAR-T cell treatment reduced leukemia burden and prolonged survival in xenograft models. In colony formation assay of UCB-derived CD34 + progenitors, there was no significantly reduced colony formation after 24 h co-cultured with either loop CAR-T cells. However, after co-culturing with CAR-T cells for 72 h, the Loop 123 × 38 CAR-T cell treatment group showed a decrease in BFU-E and CFU-GM formation, while Loop 38 × 123 CAR-T cells did not show a significant decrease compared to that of vector-T cells. BFU-E and CFU-GM formation was higher in Loop 38 × 123 CAR-T co-cultured group than that in Loop 123 × 38 CAR-T co-cultured group ( P < 0.01 and P < 0.0001, respectively). CD38/CD123 dual recognition retained activity in antigen-heterogeneous AML models, including reciprocal single-antigen-loss populations. Among the two loop configurations, Loop 123 × 38 CAR-T cells showed a stronger overall antileukemic profile, but this was accompanied by greater suppression of BFU-E and CFU-GM formation after prolonged exposure.