Single Cell Analysis of Pre-Treatment Peripheral Immune Composition Can Predict CAR-T Outcomes in Patients with Diffuse Large B-Cell Lymphoma

Background: Chimeric Antigen Receptor (CAR)-T cell therapy has revolutionized the treatment landscape for relapsed/refractory Diffuse Large B-cell Lymphoma (DLBCL). However, CAR-T failure remains a challenge, and an effective pre-treatment prognostic score is yet to be established.

Aim: To gain a deeper understanding of the cellular and molecular determinants of response to CAR-T treatment by applying single cell multiomics, which will facilitate early response prediction, address resistance mechanisms, and refine patient selection criteria. This is particularly crucial with the introduction of CAR-T cells to earlier treatment lines and for high-risk patients.

Methods: We performed single-cell RNA sequencing of peripheral blood (PB) samples at the time of leukapheresis from DLBCL patients treated with anti-CD19 CAR-T: Axicabtagene ciloleucel (Axi-cel) or Tisagenlecleucel (Tis-cel), and correlated their molecular and cellular features with treatment outcome. Response was defined as complete remission at 3 months post-infusion.

Results: The cohort included 43 patients (Axi-cel, n=19; Tis-cel, n=24), among whom 17 patients whose disease transformed, while 26 were diagnosed with de novo DLBCL. Six patients received CAR-T following the first line of treatment, 20 patients as a third line, and 17 received fourth line treatment or higher. Patient outcomes were tracked for up to 40 months (median 11.3 months). Eleven out of 43 (25.5%) patients were refractory to CAR-T treatment, 12/43 (27.9%) relapsed within the follow-up period and 13 patients (30.2%) are in ongoing remission.

We analyzed single cell data of 85,589 high quality cells and observed several striking features in responding patients. Paradoxically, all patients (Axi-cel, n=3; Tis-cel, n=9) with high B-cells in the PB at the time of leukapheresis, achieved complete remission. B-cells exhibited both healthy-like B-cell phenotypes (Naïve and memory B-cells) (n=7), and malignant signatures (n=5). The malignant signatures included B-cell development (TCF4), anti-apoptotic regulators (BCL2) and glucocorticoid response genes (FKBP5). We inferred copy number variations from the single cell RNA data and found that the malignant B-cells demonstrated aberrations characteristic of DLBCL and pre-transformed disease clones, such as amplifications in BCL2 and TCF4 genes, and deletion of TP53 gene. Finally, the malignant B-cells, representing circulating lymphoma cells, had clonal expression of the immunoglobulin κ and λ light chains, which we further validated using flow cytometry.

In the myeloid compartment, we identified enrichment of CD16 (non-classical) monocytes as a good prognostic marker. In the non-responder patients, we identified enrichment of CD14 (classical) monocytes and activation of hypoxia (HIF1A), cellular stress (HBEGF) and inflammation (SDC2) pathways across several myeloid populations. The cellular composition did not differ significantly in responders versus non-responders in the T and Natural Killer (NK) compartment. However, gene expression analysis revealed upregulation of pathways related to cellular activation and stress in CD8 Effector Memory and NK subsets of non-responders, further supporting the association linking inflammation to inferior prognosis.

To translate our findings into a clinical decision support framework, we developed a machine learning model for CAR-T response prediction. The model achieved a positive predictive value (PPV) of 0.96 and a negative predictive value (NPV) of 0.71, with an overall ROC AUC of 0.88. We subsequently evaluated the model using an external dataset derived from 89,393 cells across 20 patients (Haradhvala et al., Nat Med, 2022) reaching an ROC AUC of 0.84.

Conclusions: Our findings demonstrate that high-resolution profiling of patients' PB has the potential to develop effective pre-treatment biomarkers of response. This approach includes identifying healthy-like and malignant circulatory B-cells, CD16 monocytes, and naïve T and myeloid cell phenotypes. These observations highlight potential mechanisms to enhance CAR-T efficacy in high-risk resistant patients, which we will discuss further in the presentation. Our work offers a framework for personalized treatment and improved CAR-T prognosis.