Tumor immune micro-environment (TME) dynamics in longitudinal diffuse large B-cell lymphoma (DLBCL) cases from diagnosis to RCHOP relapse

DLBCL gene expression signatures from single-nucleus analyses have recently identified new tumor microenvironment (TME) archetypes (LymphoMAPs) linked to RCHOP and CAR-T cell sensitivity. While some archetypes are enriched at diagnosis or relapse, their stability during disease progression remains unclear. We analyzed longitudinal DLBCL cases using LymphoMAP signatures and histologic annotation to assess archetype switching after RCHOP. Methods: We utilized two independent well-annotated DLBCL cohorts with paired diagnosis/relapse biopsies:23 in-house (46 samples) and 44 published cases (88 samples). Gene expression in our cohort was profiled using the NanoString Cancer Immune Panel on fixed (FFPE) specimens. The external dataset comprised whole-transcriptome from FFPE or fresh-frozen samples pre- and post-RCHOP. LymphoMapR predicted the archetypes from normalized counts: enriched in fibroblast and tumor-associated macrophages (FMAC), naïve and memory T cell (LN), activated macrophages and exhausted T cell (TEX). Samples with classification probability <0.7 were excluded. Additionally, we correlated the archetypes with histologic annotation from our cohort. Immunohistochemistry (IHC) supported T cell markers (CD3, CD4, CD8) and PDL1 expression. Multiplex OPAL imaging quantified macrophage markers (CD68, CD163) with immune checkpoint (CD47), and chemokines (CXCL9 and CXCL10) with receptor (CXCR3). Results: In the in-house cohort, 8 patients relapsed following only RCHOP, others received ≥1 relapse regimen. One patient per cohort was excluded for low LymphoMAP confidence. Notably, archetype switching occurred in 13 of 22 patients (59%, 95% CI: 36-79%) in-house and 18 of 43 patients (42%, 95% CI: 28-57%) externally. To gain statistical power, the cohorts were combined, yielding an overall switch rate of 48% (95% CI: 36-60%). Switching was less frequent for patients with LN at diagnosis (38%, 95% CI: 23-57%) than FMAC (59%, 95% CI: 36-78%) or TEX (57%, 95% CI: 33-79%), but no consistent direction of switch or correlation with relapse timing was observed. Germinal center B-cell-like (GCB) DLBCLs were predominantly LN archetype (58%), whereas non-GCB encompassed all LymphoMAP archetypes, suggesting an association with cell-of-origin. Histology was consistent with LymphoMAP prediction: T cell abundance (CD3, CD4, CD8) and CXCR3 expression were highest in TEX, followed by LN and FMAC ( p <0.05). IHC also demonstrated a significant decrease in CD3 and CD8 T cells at relapse compared with diagnosis. Conclusions: Based on our observations, the immune environment of DLBCL is dynamic. Although relapse samples exhibited more T cell-depleted environments, changes in TME archetypes frequently occurred, without a dominant pattern. In the context of patient selection for CAR-T cell therapy, relapsed TME should not be inferred from TME at diagnosis.