Integrated multi-omic profiling identifies genomic subtypes associated with differential outcomes after car-t therapy in large b-cell lymphoma
Authors
Silvia Escribano Serrat¹, Sandeep S. Raj¹, Aaron Gillmor¹, Sigrun Einarsdottir,¹ Marina Gomez-Llobell¹, Evgenia Alekseeva², Nazar Aryutyunian², Alexander Nesmelov², Andrey Suponin², Pavel Zemskiy², Dmitrii Snitkin², Gunjan Shah¹, M. Lia Palomba¹, Connie Lee Batlevi¹, Craig Sauter¹, Miguel-Angel Perales¹, Marcel van den Brink³, Gilles Salles¹, Roni Shouval¹
- Memorial Sloan Kettering Cancer Center, New York, NY, USA
- BostonGene Corporation, Waltham, MA, USA
- City of Hope National Medical Center, Duarte, CA, USA
Abstract
Background:
Outcomes after CD19-directed CAR-T cell therapy in large B-cell lymphoma (LBCL) remain heterogeneous and are not fully explained by established clinical prognostic models. Tumor genomic architecture and microenvironmental states may influence CAR-T efficacy, but their integrated impact in real-world cohorts remains incompletely defined. We performed a multi-omic analysis to identify biological determinants of clinical variability after CAR-T therapy.
Aims:
To define tumor-intrinsic genomic and transcriptomic features associated with resistance to CAR-T cell therapy in LBCL.
Methods:
We conducted a retrospective cohort study of patients with LBCL treated with commercial CAR-T therapy between 2016 and 2022. Pre-CAR-T tumor specimens underwent whole-exome sequencing (WES) and RNA sequencing (RNA-seq), followed by integrated genomic and transcriptomic analyses using a standardized multi-omic computational pipeline (BostonGene, Waltham, MA). Genomic subtypes were assigned using the established LymphGen framework (Wright et al., Cancer cell, 2020) and Lymphly, a hierarchical genomic classification algorithm developed to improve molecular stratification of LBCL. The tumor microenvironment (TME), and double hit signature (DHIT) were classified according to Kotlov et al. (Cancer Discovery, 2021), dark zone (DZsig) signature in accordance with Harris et al. (Leukemia Lymphoma, 2025). Associations between molecular features and progression-free survival (PFS), 6-month PFS (PFS6), and complete response (CR) were evaluated. Subgroups with fewer than 10 cases were excluded from statistical comparisons but retained for visualization.
Results: Among 112 patients with available molecular data, median follow-up was 66 months (IQR 46-73). CAR-T products included axicabtagene ciloleucel (49%), tisagenlecleucel (38%), and lisocabtagene maraleucel (13%). Fifty-four percent of tumors were germinal center B-cell type, 46% activated B-cell type, and 17% were DHIT-positive.
WES data were available in 99 patients. BostonGene’s Lymphly classifier assigned genomic subtypes to a larger proportion of evaluable cases compared with LymphGen and demonstrated improved outcome discrimination (Fig. 1A). Inferior outcomes were primarily driven by the BN2 subtype, which was associated with worse PFS (16%, 95% CI 6-46, p=0.017), lower PFS6 (p=0.009), and reduced CR rates (p=0.041) compared with other evaluable subgroups (Fig. 1B). In multivariable Cox models adjusting for relevant clinical covariates, BN2 remained independently associated with inferior PFS and PFS6 (both p<0.001).
RNA-seq data were available for 55 patients. RNA-defined TME subtypes and DHIT status were not significantly associated with PFS in univariable analyses. The inflammatory TME subtype demonstrated numerically inferior outcomes, with a 1-year PFS of 19% (95% CI 7-52) compared with 36% and 41% in the mesenchymal and depleted subtypes, respectively, whereas DHIT-positive cases showed numerically favorable 1-year PFS (46% vs 31%) (Fig. 1C). Within the GCB subgroup, the Dzsig showed similar PFS between DZsig-positive and DZsig-negative cases (Fig 1D); formal statistical testing was not performed due to limited sample size.
Summary/Conclusion: Multi-omic profiling identifies biologically defined high-risk subgroups after CAR-T therapy, highlighting BN2 as a genomically driven resistance phenotype. Genomic classification may refine patient selection and enable mechanism-based therapeutic strategies in LBCL.
Outcomes after CD19-directed CAR-T cell therapy in large B-cell lymphoma (LBCL) remain heterogeneous and are not fully explained by established clinical prognostic models. Tumor genomic architecture and microenvironmental states may influence CAR-T efficacy, but their integrated impact in real-world cohorts remains incompletely defined. We performed a multi-omic analysis to identify biological determinants of clinical variability after CAR-T therapy.
Aims:
To define tumor-intrinsic genomic and transcriptomic features associated with resistance to CAR-T cell therapy in LBCL.
Methods:
We conducted a retrospective cohort study of patients with LBCL treated with commercial CAR-T therapy between 2016 and 2022. Pre-CAR-T tumor specimens underwent whole-exome sequencing (WES) and RNA sequencing (RNA-seq), followed by integrated genomic and transcriptomic analyses using a standardized multi-omic computational pipeline (BostonGene, Waltham, MA). Genomic subtypes were assigned using the established LymphGen framework (Wright et al., Cancer cell, 2020) and Lymphly, a hierarchical genomic classification algorithm developed to improve molecular stratification of LBCL. The tumor microenvironment (TME), and double hit signature (DHIT) were classified according to Kotlov et al. (Cancer Discovery, 2021), dark zone (DZsig) signature in accordance with Harris et al. (Leukemia Lymphoma, 2025). Associations between molecular features and progression-free survival (PFS), 6-month PFS (PFS6), and complete response (CR) were evaluated. Subgroups with fewer than 10 cases were excluded from statistical comparisons but retained for visualization.
Results: Among 112 patients with available molecular data, median follow-up was 66 months (IQR 46-73). CAR-T products included axicabtagene ciloleucel (49%), tisagenlecleucel (38%), and lisocabtagene maraleucel (13%). Fifty-four percent of tumors were germinal center B-cell type, 46% activated B-cell type, and 17% were DHIT-positive.
WES data were available in 99 patients. BostonGene’s Lymphly classifier assigned genomic subtypes to a larger proportion of evaluable cases compared with LymphGen and demonstrated improved outcome discrimination (Fig. 1A). Inferior outcomes were primarily driven by the BN2 subtype, which was associated with worse PFS (16%, 95% CI 6-46, p=0.017), lower PFS6 (p=0.009), and reduced CR rates (p=0.041) compared with other evaluable subgroups (Fig. 1B). In multivariable Cox models adjusting for relevant clinical covariates, BN2 remained independently associated with inferior PFS and PFS6 (both p<0.001).
RNA-seq data were available for 55 patients. RNA-defined TME subtypes and DHIT status were not significantly associated with PFS in univariable analyses. The inflammatory TME subtype demonstrated numerically inferior outcomes, with a 1-year PFS of 19% (95% CI 7-52) compared with 36% and 41% in the mesenchymal and depleted subtypes, respectively, whereas DHIT-positive cases showed numerically favorable 1-year PFS (46% vs 31%) (Fig. 1C). Within the GCB subgroup, the Dzsig showed similar PFS between DZsig-positive and DZsig-negative cases (Fig 1D); formal statistical testing was not performed due to limited sample size.
Summary/Conclusion: Multi-omic profiling identifies biologically defined high-risk subgroups after CAR-T therapy, highlighting BN2 as a genomically driven resistance phenotype. Genomic classification may refine patient selection and enable mechanism-based therapeutic strategies in LBCL.
Latest publications
