Single-cell atlas of circulating immunity identifies shared specific DLBCL signatures predicting response to R-CHOP and CD19 CAR-t therapies Free

Background

While the immunological heterogeneity of diffuse large B cell lymphoma (DLBCL) involves both the tumor microenvironment and peripheral immune system, blood-based immune profiling to assess the systemic status and predict treatment response is lacking.

Methods

Longitudinal blood samples were prospectively collected from a newly diagnosed DLBCL cohort, an R/R DLBCL cohort treated with anti-CD19 CAR-T cells (sampled for up to 365 days) and healthy controls (HCs). The samples were evaluated via 22-marker single-cell spectral flow cytometry, algorithm-guided analysis and ELISAs for cytokine quantification to profile circulating immunity, develop a machine learning model for refractory distinction, and investigate treatment-related immune variations.

Results

Over one million cells were collected to accurately profile the circulating immune compartment in HCs and DLBCLs. After preprocessing, immune cells were clustered with FlowSOM and visualized via UMAP, identifying 39 clusters. Compared to HCs, DLBCL patients exhibited peripheral immune subset dysregulation, characterized by a reduced innate cell and T cell imbalance. Notably, the frequencies of four cell clusters— HLA-DR^hi PD-1^hi CD127^- Tfh1 cells, CXCR5⁺ CXCR3⁺ CD127^- CD8⁺ T cells, HLA-DR⁺ CXCR3⁺ CD73^- CD8 T cells and CD4⁺ NKT cells —were found to be very low in HCs but markedly increased in DLBCL patients, indicating that these four cell populations may be distinctive features of DLBCL.

In the de novo DLBCL cohort, to convert the intricate immune signatures for clinical application, we used a random forest machine learning method to differentiate primary refractory DLBCL patients from responders based on 69 peripheral immune features. A 3-feature signature—HLA-DR^hi PD-1^hi CD127^- Tfh1 cells, classical monocytes, and Tregs—showed high accuracy and an AUC of 0.84 in distinguishing between the two groups. Analysis of cytokine levels associated with the top three immune cell subsets indicated significantly elevated levels of IL-21 and IL-10. Additionally, changes in these subsets correlated with treatment effectiveness.

Furthermore, in the R/R DLBCL cohort receiving CD19 CAR-T cell therapy, we performed a comprehensive analysis of dynamic changes in 39 defined subpopulations, comparing patients with longer PFS (≥6 months) to those with shorter PFS (<6 months). An analysis of immune reconstitution dynamics revealed both shared and divergent patterns. All patients exhibited sustained expansion of CXCR3⁺ CD4⁺ Th1 cells from pre-lymphodepletion through day 365, indicative of Th1 polarization following CAR-T cell activation. Terminally differentiated CD8⁺ TEMRA cells reached their peak at day 90, after which a decline was observed, indicative of transient effector mobilization and potential exhaustion. The distinct reconstitution patterns revealed that long-PFS patients exhibited CXCR3⁺ CD8 Tem and CCR6⁺ CXCR3⁺ CD4 Tem recovery after day 30, in contrast to the irreversible decline seen in the short-PFS group. These findings indicated a more effective immune homeostasis recovery in patients with prolonged PFS.

After identifying the immune changes over time, we analyzed cell populations with the most significant differences between long-PFS and short-PFS patients at the same time point to identify predictive biomarkers. Pre-apheresis and D90 blood immunophenotypes (elevated HLA-DR^hi PD-1^hi CD127^- Tfh1 cells, classical monocytes, Tregs, and pDCs but reduced CXCR3⁺ CCR6⁺CD4⁺ Tems and CD8⁺ naïve T cells) were linked to poorer PFS, paralleling primary refractory cohort findings.

Conclusions

The present systemic analysis of peripheral immune signatures revealed refractory DLBCL patients may exhibit shared immune profiles. These new systemic immunity-based liquid biomarkers can be utilized to predict and monitor responses in a clinical context.