Baseline and on-Treatment Bone Marrow Microenvironments Predict Myeloma Patient Outcomes and Inform Potential Intervention Strategies

Introduction

The multiple myeloma (MM) tumor microenvironment (TME) strongly influences patient outcomes as evidenced by the success of immunomodulatory therapies. To develop precision immunotherapeutic approaches, it is essential to identify and enumerate TME cell types and understand their dynamics.

Methods

We estimated the population of immune and other non-tumor cell types during the course of MM treatment at a single institution using gene expression of paired CD138-selected bone marrow aspirates and whole bone marrow (WBM) core biopsies from 867 samples of 436 newly diagnosed MM patients collected at 5 time points: pre-treatment (N=354), post-induction (N=245), post-transplant (N=83), post-consolidation (N=51), and post-maintenance (N=134). Expression profiles from the aspirates were used to infer the transcriptome contribution of immune and stromal cells in the WBM array data. Unsupervised clustering of these non-tumor gene expression profiles across all time points was performed using the R package ConsensusClusterPlus with Bayesian Information Criterion (BIC) to select the number of clusters. Individual cell types in these TMEs were estimated using the DCQ algorithm and a gene expression signature matrix based on the published LM22 leukocyte matrix (Newman et al., 2015) augmented with 5 bone marrow- and myeloma-specific cell types.

Results

Our deconvolution approach accurately estimated percent tumor cells in the paired samples compared to estimates from microscopy and flow cytometry (PCC = 0.63, RMSE = 9.99%). TME clusters built on gene expression data from all 867 samples resulted in 5 unsupervised clusters covering 91% of samples. While the fraction of patients in each cluster changed during treatment, no new TME clusters emerged as treatment progressed. These clusters were associated with progression free survival (PFS) (p-Val = 0.020) and overall survival (OS) (p-Val = 0.067) when measured in pre-transplant samples. The most striking outcomes were represented by Cluster 5 (N = 106) characterized by a low innate to adaptive cell ratio and shortened patient survival (Figure 1, 2). This cluster had worse outcomes than others (estimated mean PFS = 58 months compared to 71+ months for other clusters, p-Val = 0.002; estimate mean OS = 105 months compared with 113+ months for other clusters, p-Val = 0.040). Compared to other immune clusters, the adaptive-skewed TME of Cluster 5 is characterized by low granulocyte populations and high antigen-presenting, CD8 T, and B cell populations. As might be expected, this cluster was also significantly enriched for ISS3 and GEP70 high risk patients, as well as Del1p, Del1q, t12;14, and t14:16. Importantly, this TME persisted even when the induction therapy significantly reduced the tumor load (Table 1). At post-induction, outcomes for the 69 / 245 patients in Cluster 5 remain significantly worse (estimate mean PFS = 56 months compared to 71+ months for other clusters, p-Val = 0.004; estimate mean OS = 100 months compared to 121+ months for other clusters, p-Val = 0.002).

The analysis of on-treatment samples showed that the number of patients in Cluster 5 decreases from 30% before treatment to 12% after transplant, and of the 63 patients for whom we have both pre-treatment and post-transplant samples, 18/20 of the Cluster 5 patients moved into other immune clusters; 13 into Cluster 4. The non-5 clusters (with better PFS and OS overall) had higher amounts of granulocytes and lower amounts of CD8 T cells. Some clusters (1 and 4) had increased natural killer (NK) cells and decreased dendritic cells, while other clusters (2 and 3) had increased adipocytes and increases in M2 macrophages (Cluster 2) or NK cells (Cluster 3). Taken together, the gain of granulocytes and adipocytes was associated with improved outcome, while increases in the adaptive immune compartment was associated with poorer outcome.

Conclusions

We identified distinct clusters of patient TMEs from bulk transcriptome profiles by computationally estimating the CD138- fraction of TMEs. Our findings identified differential immune and stromal compositions in patient clusters with opposing clinical outcomes and tracked membership in those clusters during treatment. Adding this layer of TME to the analysis of myeloma patient baseline and on-treatment samples enables us to formulate biological hypotheses and may eventually guide therapeutic interventions to improve outcomes for patients.

View largeDownload slide

View largeDownload slide

Close modal