Spatial proteomics of the bone marrow reveals distinct Tumor–Immune niches in smoldering and relapsed myeloma and their remodeling in response to bispecific antibody and CAR-T therapy

Abstract INTRODUCTION T cell–redirecting therapies, including CAR T cells and bispecific antibodies (BsAbs), have transformed the treatment of relapsed/refractory multiple myeloma (RRMM) and are being explored in earlier stages, including high-risk smoldering myeloma (SMM). Emerging evidence suggests spatial tissue organization modulates immunotherapy outcomes. Here, we used spatial proteomics to (1) map the bone marrow (BM) microenvironment across disease stages, (2) assess associations between baseline spatial features and response, and (3) characterize longitudinal remodeling in response to immunotherapy. METHODS We analyzed 121 BM biopsies from 47 patients with high-risk SMM treated with teclistamab or cilta-cel, and 19 patients with RRMM treated with BsAbs (14 teclistamab, 2 elranatamab, 3 talquetamab) between 2022–2024. Thirty-six patients had ≥1 post-treatment biopsy. A total of 196 regions of interest (ROIs) were profiled using a 32-marker imaging mass cytometry (IMC) panel (Hyperion XTi, Standard BioTools). Cell segmentation was performed with cellpose-sam. Marker-based annotation identified 27 phenotypes across 1,116,993 single cells. Cellular neighborhoods were defined as the 10 μm microenvironment around each cell. Spatial co-localization was evaluated via permutation testing. P values were FDR-adjusted. RESULTS SMM and RRMM cohorts showed expected differences: all RRMM patients had prior BCMA therapy, 42% had extramedullary disease, and median prior lines were five (range 2–13); SMM patients were mainly treatment-naïve. High-risk cytogenetics were similarly distributed (p=0.7). PC infiltration in ROIs showed a moderate correlation with pathologist estimates based on whole-slide review (R=0.61, p<0.001). In contrast, intra-sample consistency across ROIs was high (R=0.91–1.0). Regarding overall composition, SMM samples exhibited higher PC infiltration (q=0.0008) but fewer proliferating (Ki67⁺) PCs (q<0.001). T cells were more abundant in SMM (q=0.03), whereas RRMM was enriched for activated (HLA-DR⁺) and central memory (CD45RO⁺CCR7⁺) CD4⁺ T cells (Tact, Tcm; both q<0.001), as well as exhausted (PD1⁺TIM3⁺; Tex, q<0.001) and effector memory (CD45RO⁺; Tem, q=0.005) CD8⁺ T cells—indicative of an immunosuppressed T cell milieu. Regulatory T cells (Tregs) were similar between groups (q=0.3). A cellular neighborhood clustering approach at baseline revealed two PC neighborhoods in SMM (35% and 23% PCs) versus one in RRMM (44% PCs). Most non-PC cell types were similarly distributed across SMM PC neighborhoods. However, one PC neighborhood was enriched in T cells (16%), whereas the second (8%) possessed T cell levels similar to those of the RRMM PC neighborhood (9%). The T cell–enriched SMM PC neighborhood showed significantly higher levels of Tex and Tregs (both q<0.001), suggesting localized tumor-T cell immune processes unique to the SMM microenvironment. To assess treatment-induced BM changes, we analyzed matched samples at 6 months (n=20) and 12 months (n=5). PC proportions declined rapidly, while T cells (6m q=0.12, 12m q=0.06) and Granzyme B⁺ cells (6m q=0.06, 12m q=0.04) increased over time. Within CD4⁺ T cells, Tact and Tcm trended upward, whereas Tregs declined (q=0.15). CD8⁺ T cells were more dynamic, with transient Tem and Tex increases (both q=0.005). Additionally, macrophages showed a transient rise (q=0.05). Subsequently, analyzing spatial cell-cell interactions, we observed a decline between PCs and CD4⁺ and CD8⁺ memory T cell interactions over time. PC–CD8⁺ Tex interactions initially decreased but rebounded by 12 months, while T cell–DC interactions followed the opposite pattern, increasing early and then declining (q<0.05). These temporal patterns indicate an early phase of BsAb-driven immune remodeling, which appears to subside by one year, likely due to reduced tumor burden and limited ongoing T cell engagement. Finally, to assess whether baseline spatial interaction scores predict treatment response, we stratified RRMM patients by median PC–Treg and PC–Tex interaction levels. Kaplan–Meier analysis revealed significantly shorter progression-free survival in patients with enriched PC–immune interactions (Treg: p=0.003; Tex: p=0.006).CONCLUSIONS This study provides a framework for leveraging spatial profiling to dissect the bone marrow microenvironment in myeloma across disease stages and proposes a model in which spatial PC–T cell interactions inform the efficacy of T cell–redirecting therapies.