Spatial Dissection of the Bone Marrow Microenvironment in Multiple Myeloma By High Dimensional Multiplex Tissue Imaging

Introduction

Multiple myeloma (MM) is shaped by interactions between immune, stromal and tumor cells in the bone marrow (BM) microenvironment (BMME). Our understanding of these processes is based on high parametric flow cytometry and single cell transcriptomics, which lack spatial resolution. Thus, crucial aspects of the in situ tumor ecosystem, including cell-cell interactions, can only be predicted. Novel high dimensional, multiplex tissue imaging methods are able to capture similar granularity while also preserving spatial organization. By optimizing antibody selection, as well as staining, imaging and data processing protocols we have established a BMME-targeted CODEX workflow based on a 60 marker panel. Here, we conducted the first comprehensive multiplex protein imaging survey of the BMME across the spectrum of MM and precursor stages and leveraged paired FISH and whole genome sequencing (WGS) to elucidate the interplay of tumor intrinsic features and the BMME.

Methods

We imaged 489 BM trephine biopsies in tissue microarray format on the CODEX (Akoya Biosciences) platform. After quality control (≥750 cells per core, <25% unclassifiable cells) we retained data from 12 non-malignant controls, 13 MGUS, 20 smoldering MM, 349 uniformly treated newly diagnosed MM (NDMM) patients from a phase III clinical trial, as well as 22 relapsed MM samples with a median of 3 (range 1-6) tissue cores per sample, comprising a total of 3.1 million single cells. CODEX data was processed and analyzed using custom scripts. Cellular neighborhoods were computed by clustering the cell type composition of sliding windows (n=10 nearest neighboring cells) across the tissue. WGS (n=212) and FISH (n=309) were performed on CD138-purified BM mononuclear cells and analyzed using in-house pipelines.

Results

We identified 40 cell types, encompassing the relevant lineages (myeloid, lymphoid, stromal, endothelial, tumor), subtypes (e.g. T, B, NK cells) and functional states (e.g GRZB, Ki67, and PD1) of the BMME. The degree of plasma cell infiltration determined by CODEX correlated with the clinical pathologist’s quantification based on classical immunohistology (r=0.77, p<0.0001), and the kappa/lambda status of MM cells matched the reported light chain restriction, supporting the validity of our data.

At the compositional level, we observed shifts in relative cell type abundance from precursor to MM stages. Major changes in the BMME included enriched stromal, endothelial and CD8+ T cells as well as depleted polymorphonuclear (PMN), erythropoietic and CD4+ T cells in advanced stages. The highest proportion of monocytoid dendritic cells, monocytes, mast cells and cytotoxic CD8+ T cells was seen in relapsed MM.

Leveraging the spatial resolution of the CODEX data, we identified multiple distinct cellular neighborhoods (CNs) defined by recurrent local cell type composition. CNs enriched for plasma cells in combination with different immune cell subsets including exhausted T cells increased towards the MM stage, while CNs dominated by PMNs, erythropoietic cells and CD206 negative macrophages decreased.

To investigate links between MM subtypes and BMME patterns, we correlated CODEX data with MM-initiating and driver events. While patients with t(4;14) showed a significant increase in eosinophils, patients with gain1q had higher levels of endothelial and stromal subsets colocalizing in the same CN. Distinct types of stromal cells were also enriched in patients with double hits (≥2 FISH high risk features). RAS mutations and biallelic inactivation of tumor suppressor genes were associated with altered T cell subset composition.

Further corroborating the importance of BMME architecture, a CN with colocalization of plasma and exhausted CD8 T cells was associated with unfavorable prognosis in NDMM patients.

Conclusions

High dimensional, multiplex tissue imaging of the BMME enables interrogation of cellular interactions and tissue architecture in situ in clinical samples at single cell resolution. To our knowledge this is the first highly multiplexed spatial study of the human BMME in a large clinical cohort and the first such study in multiple myeloma. We track characteristic shifts in tissue composition associated with disease progression, reveal genotype-phenotype associations in the BMME as a correlate of tumor-microenvironment co-evolution and detect prognostically relevant tissue architectural features.