Spatially resolved transcriptomics reveals immunosuppressive niches and clonal diversity in extramedullary disease in multiple myeloma refractory to immunotherapy

Abstract Introduction: Extramedullary disease in multiple myeloma (EMD) is a clinically aggressive variant of multiple myeloma (MM), frequently associated with resistance to therapy and poor outcomes. Despite advances in immunotherapies such as CAR T-cell and bispecific T cell engager (BiTE) therapies, their effectiveness in EMD remains limited, partly due to the distinct biological characteristics and resistance mechanisms within EMD lesions. A detailed understanding of both the clonal composition and spatial organization of the tumor microenvironment in EMD is essential to guide the development of tailored treatment strategies and novel therapeutic targets. Methods We applied high-resolution spatial transcriptomics to fifteen extramedullary plasmacytoma samples collected from ten relapsed/refractory MM (RRMM) patients who had progressed following CAR T-cell or BiTE therapy (8 CAR-T, 2 BiTE EMD lesions) and five newly diagnosed MM patients. Tumors were sampled from diverse anatomical regions including liver, soft tissue (located in neck, supraclavicular region and thigh), paramedullary region (located in femur, iliac crest and rib), and additional nodal locations. Spatial gene expression profiling was conducted using the CosMx SMI 6k panel (NanoString Technologies), the Xenium and the Visium HD platforms (10X Genomics). Results Cellular profiling identified a total of 388,250 single cells across all lesions (range: 6,070–155,648 cells/sample; median: 26,056), encompassing immune (T/NK cells, myeloid cells) and malignant plasma/B cell populations. The number of genes detected per cell spanned 114 to 2,437 in all three datasets (CosMx, Xenium and VisiumHD). Analysis of the CosMx dataset revealed plasma cell clusters with heterogeneous gene expression signatures and three dominant transcriptional programs observed across patients at disease progression after CAR-T cell or BiTE therapy: (1) a highly proliferative cluster marked by MKI67, HMGB2, and STMN1; (2) a transcriptionally active, pro-survival cluster upregulating BRAF and BCL2; and (3) a relatively quiescent cluster with low transcriptional activity. Neighborhood analysis identified 18 neighborhoods across EMD samples from RRMM patients, with two niches present in all samples. Cell type compositional analysis determined a high frequency of plasma cells with a transcriptionally active, pro-survival signature in one niche while the other was composed of myeloid, T cells and plasma cells. Spatial mapping of the immune microenvironment in all three datasets uncovered consistent regional restriction of antigen-presenting cells (APCs) and T cells in all EMD lesions. CD3+ T cells represented under 5% of the total cellular population in all datasets, indicating severe T cell depletion comparable to cold solid tumors (less than 10%). This immune exclusion was especially pronounced in the liver EMD lesion, where APCs and CD8+ T cells were confined to a small niche, while over 80% of the tissue comprised myeloma cells—reflecting a profound lack of immune infiltration. Further cell–cell interaction analysis revealed enriched immunosuppressive signaling, particularly involving myeloid–myeloid (e.g., S100A9/CD68, S100A9/ITGB2) and myeloid–CD8+ T cell (e.g., S100A9/ITGB2, B2M/KLRD1) interactions. These interactions may hinder immune cell migration and spatial distribution, reinforcing localized immune suppression. Conclusion Spatial transcriptomic analysis of EMD lesions highlights striking inter- and intra-lesional heterogeneity in both tumor clonality and immune contexture. The presence of proliferative tumor cell states alongside immune deserts, focal T cell/APC clusters, and suppressive myeloid signaling circuits reveals key barriers to effective anti-tumor immunity. These findings support the incorporation of spatially resolved biomarkers and combination immune-modulatory approaches for improving outcomes in EMD.