Bone Marrow Spatial Signatures Predict Survival Outcomes and Toxicities after CAR-T Therapy in Patients with Relapsed/Refractory Multiple Myeloma

Introduction

Chimeric antigen receptor (CAR) T cell therapy has shown remarkable response rates in patients with relapsed/refractory multiple myeloma (RRMM). However, patients ultimately experience disease progression partially due to tumor-intrinsic alterations and an immunosuppressive microenvironment. Previous research primarily utilized sequencing and flow cytometry to investigate relapse mechanisms, but the influence of bone marrow (BM) spatial organization remains underexplored. This study employs imaging mass cytometry (IMC) with a novel immune-cell-focused panel to investigate how BM architecture impacts survival and toxicity following CAR-T cell therapy in 50 RRMM patients.

Methods

We analyzed 61 BM trephine biopsies from 50 RRMM patients treated with ide-cel (n=39) or cilta-cel (n=11) between 08/2021 and 05/2023. We assessed the spatially orchestrated cellular niche using a panel of 36 antibodies (HyperionXTi, Standard BioTools). After quality control, 60 slides were retained: 43 at baseline and 17 one month post-CAR-T infusion, with 11 matched to baseline. Cells were segmented using CellProfiler, and batch correction was performed using FastMNN. Cell types were annotated based on marker expression, and a structural pattern-based extraction algorithm identified adipocytes. Cellular neighborhoods (CNs), defined as the microenvironment within a 10 μm radius surrounding each cell, were identified with the Mini Batch K-means algorithm. Compositional data was isometric log-ratio transformed.

Results

We detected 511,756 single cells across all samples, representing 24 cell subtypes. Functional markers, including Ki-67, PD-L1, and granzyme B (GZMB), allowed for further identification of cell states. Validation of IMC-measured plasma cell (PC) infiltration against pathologists’ reports showed a strong correlation (Pearson r=0.72, p<0.001). For the 43 baseline samples, spatial clustering algorithms identified 20 distinct CNs with varying cellular compositions. Two CNs were enriched with PCs, indicating tumor-bearing areas. CN0 was tumor-exclusive with 83% of PCs, while CN10 was more heterogeneous, with 50% of PCs and 25% of myeloid-derived suppressor cells (MDSCs) and monocytes. PCs were further classified into Ki-67^pos, PD-L1^pos, and KI-67/PD-L1^neg. Of note, patients with Ki-67^pos and PD-L1^pos PCs above the median had reduced progression-free survival (mPFS: PD-L1^pos 6.4 vs. PD-L1^neg 19.2 mo; Ki-67^pos 6.6 vs. Ki-67^neg 19.2 mo; each p(log-rank)<0.001). In contrast, the amount of KI-67/PD-L1^neg did not impact PFS, suggesting that tumor burden alone might be an inferior predictor for treatment response.

Among other neighborhoods, two CNs were independently associated with a decreased mPFS (CN1: high 5.3 vs. low 15.6 mo, p(log-rank)=0.002; CN6: high 6.4 vs. low 19.2 mo, p(log-rank)=0.002). CN1 exhibited an M2-like macrophage (M2-mac)-monocyte signature, encompassing 64% of cells in this neighborhood. A pairwise cell-type interaction analysis based on spatial localization revealed for CN1 that 92% of PCs interacted with MDSCs or M2-macs. In contrast, CN6 showed the strongest enrichment of all neighborhoods in CD8 T-cell subsets (23%), including cytotoxic GZMB^pos T-memory cells (T-mem, 12%) and phenotypically exhausted T-cells (T-ex, 5%). Notably, over one-third of GZMB^pos T-mem interacted with M2-Macs/MDSCs, while only 3.6% of GZMB^pos T-mem were in spatial proximity to PCs. Moreover, 29.1% of GZMB^pos T-mem interacted with T-ex, implying that CN6 might display a spatial area of active T-cell exhaustion processes.

In addition to survival outcomes, patients with BM samples enriched with CN17 had a higher cumulative incidence of a severe cytokine release syndrome (CRS)(p=0.005). CN17 had high proportions of CD4 T-cell subtypes (dysfunctional CD4 T-cells 21%, CD4 T-mem 12%), consistent with the role of CD4 T-cells in CRS development.

Conclusions

This pioneering study leverages IMC to explore human BM samples in the context of CAR-T cell therapy. Our findings validate the feasibility of using the IMC panel to describe BM spatial architecture, highlight the potential of IMC in predicting post-CAR-T therapy survival outcomes and toxicities, and suggest novel spatial mechanisms driving relapse in RRMM patients. Ongoing studies aim to elucidate these mechanisms further, paving the way for improved therapeutic strategies.