Bone marrow microenvironments that contribute to patient outcomes in newly diagnosed multiple myeloma: A cohort study of patients in the Total Therapy clinical trials

Abstract

Background: The tumor microenvironment (TME) is increasingly appreciated as an important determinant of cancer outcome, including in multiple myeloma (MM). However, most myeloma microenvironment studies have been based on bone marrow (BM) aspirates, which often do not fully reflect the cellular content of BM tissue itself. To address this limitation in myeloma research, we systematically characterized the whole bone marrow (WBM) microenvironment during premalignant, baseline, on treatment, and post-treatment phases.

Methods and findings: Between 2004 and 2019, 998 BM samples were taken from 436 patients with newly diagnosed MM (NDMM) at the University of Arkansas for Medical Sciences in Little Rock, Arkansas, United States of America. These patients were 61% male and 39% female, 89% White, 8% Black, and 3% other/refused, with a mean age of 58 years. Using WBM and matched cluster of differentiation (CD)138-selected tumor gene expression to control for tumor burden, we identified a subgroup of patients with an adverse TME associated with 17 fewer months of progression-free survival (PFS) (95% confidence interval [CI] 5-29, 49-69 versus 70-82 months, χ2 p = 0.001) and 15 fewer months of overall survival (OS; 95% CI -1 to 31, 92-120 versus 113-129 months, χ2 p = 0.036). Using immunohistochemistry-validated computational tools that identify distinct cell types from bulk gene expression, we showed that the adverse outcome was correlated with elevated CD8+ T cell and reduced granulocytic cell proportions. This microenvironment develops during the progression of premalignant to malignant disease and becomes less prevalent after therapy, in which it is associated with improved outcomes. In patients with quantified International Staging System (ISS) stage and 70-gene Prognostic Risk Score (GEP-70) scores, taking the microenvironment into consideration would have identified an additional 40 out of 290 patients (14%, premutation p = 0.001) with significantly worse outcomes (PFS, 95% CI 6-36, 49-73 versus 74-90 months) who were not identified by existing clinical (ISS stage III) and tumor (GEP-70) criteria as high risk. The main limitations of this study are that it relies on computationally identified cell types and that patients were treated with thalidomide rather than current therapies.

Conclusions: In this study, we observe that granulocyte signatures in the MM TME contribute to a more accurate prognosis. This implies that future researchers and clinicians treating patients should quantify TME components, in particular monocytes and granulocytes, which are often ignored in microenvironment studies.

Fig 1. On-treatment MM project overview.

(A) Samples from 436 patients with NDMM were taken during the course of treatment, resulting in 867 total samples. (B) The putatively pure tumor CD138⁺ samples were combined with the LM22 leukocyte signature matrix and 5 additional BM-specific cell types to determine the cell-type mix in the WBM samples. Illustration based on [18]. (C) The proportion of gene expression from the tumor was computationally removed from the WBM samples, which were then assigned to one of the 5 unique microenvironment clusters shown on the left side of the panel. One of these clusters (termed “low-granulocyte” Cluster 5) was associated with lower patient PFS/OS as shown on the right side of the panel. (D) Cluster 5 had a distinct microenvironment characterized by unusual levels of various cell types. (E) Follow-up studies on predisease SMM and MGUS patients revealed that this microenvironment developed as the disease progressed. As the 3 pClusters on the left contain an ever-greater portion of patients with SMM, they come to resemble the microenvironment for Cluster 1–4 patients with NDMM. Cluster 5 NDMM patients show an extension of the trends observed as the disease develops. (F) Machine learning revealed that patients with low levels of granulocytes had consistently worse outcomes, even after accounting for high-risk status using GEP-70 and ISS stage. (G) As predicted, patients with RRMM in the STRATUS trial were significantly enriched for the low-granulocyte phenotype. BM, bone marrow; CD, cluster of differentiation; del, deletion; GEP-70, 70-gene Prognostic Risk Score; IMiD, immunomodulatory drug; ISS, International Staging System; LM22, leukocyte matrix 22; MGSM, myeloma genome signature matrix; MGUS, monoclonal gammopathy of undetermined significance; MM, multiple myeloma; NDMM, newly diagnosed multiple myeloma; OS, overall survival; pCluster, premalignant microenvironment cluster; PFS, progression-free survival; RRMM, relapsed/refractory multiple myeloma; SMM, smoldering multiple myeloma; STRATUS, Evaluation of Safety of Pomalidomide in Combination With Dexamethasone (Low Dose) in Patients With Refractory or Relapsed and Refractory Multiple Myeloma; t, translocation; TME, tumor microenvironment; UAMS70, University of Arkansas for Medical Sciences 70-gene signature; WBM, whole bone marrow.

Fig 2. Pseudo-CD138^– gene-expression clusters: Clustering of 867 samples with tumor signature removed based on deconvolution.

(A) Columns show the 867 patient samples; rows show the high-variance genes, with relatively low expression shown in white and high expression shown in blue. Hierarchical clustering and Bayesian information criteria analysis revealed 5 large clusters, with all other samples (shown in purple) classed as orphan clusters. (B) Cluster membership is distributed across treatment time points. (C) A Sankey diagram showing the cluster population shifts that occur during treatment. Colored lines are scaled to show the number of patients in each cluster at each time point. Gray lines indicate transitions between time points. Portions of colored lines with no attached gray lines had no samples in the immediately preceding or following time point. *Proportions significantly different to pretreatment time point (i.e., binomial false discovery rate < 0.05). CD, cluster of differentiation; TME, tumor microenvironment.

Fig 3. Clinical outcomes and cellular components of microenvironment clusters: Kaplan–Meier curves for pretreatment sample modeling.

(A) PFS for the “low-granulocyte” Cluster 5 (orange; n = 66/106; rmean: 59 ± 5 months) versus Clusters 1–4 (blue; n = 124/248; rmean = 76 ± 3 months). (B) OS for the “low-granulocyte” Cluster 5 (orange; n = 50/106; rmean: 106 ± 7 months) versus Clusters 1–4 (blue; n = 103/248; rmean = 121 ± 4 months). (C) Average cytogenetic and clinical characteristic enrichment for patients in each of the 5 clusters. (D) Average deconvolved tumor and stromal cell types present in each of the 5 clusters. FDRs were calculated using Wilcoxon signed-rank test and Benjamini–Hochberg correction. FDRs > 0.05 are white. (E, F) Univariate Cox proportional HRs for PFS and OS (respectively). The analysis considered the upper quartile versus the lowest quartile (or high versus low if binary) for variables in pretreatment samples. Those cell types, ratios, and clinical characteristics marked with an asterisk remain after multivariate Cox proportional hazards lasso regression, implying that they contain independent information about patient outcomes. CD, cluster of differentiation; del, deletion; FDR, false discovery rate; GEP-70, 70-gene Prognostic Risk Score; HR, hazard ratio; ISS, International Staging System; NK, natural killer; OS, overall survival; PFS, progression-free survival; t, translocation.

Fig 4. Pretreatment regression models for cell-type, clinical, and cytogenetic characteristics.

(A, B) Univariate Cox proportional HRs for PFS and OS (respectively) from 286 pretreatment samples with full cytogenetics and tumor burden data. Those cell types, ratios, and cytogenetic and clinical characteristics marked with an asterisk remain after multivariate Cox proportional hazard lasso regression, implying that they contain independent information about patient outcomes. Myeloma refers to the pathologist estimates of tumor cell percentages (rather than the deconvolution estimates). Lenalidomide indicates those 9 patients who received lenalidomide during treatment, testing whether this was skewing the results. (C, D) Variables added during stepwise Cox regressing using variable quartiles as predictors. CD, cluster of differentiation; CI, confidence interval; del, deletion; GEP-70, 70-gene Prognostic Risk Score; HR, hazard ratio; ISS, International Staging System; NK, natural killer; OS, overall survival; PFS, progression-free survival; Q, quartile; t, translocation.

Fig 5. Mast cells, patient outcomes, and TME.

(A) Cox proportional hazards conditional inference tree that selected an optimal combination of attributes from the cell types and clinical characteristics for the 290 pretreatment samples with known cytogenetics and tumor burden estimates. The algorithm first selects the strongest attribute (the cell-type or clinical characteristic with the lowest p-value), in this case, the GEP-70 high-risk flag (p < 0.001). As expected, those 34 patients (27 of whom had an event) with GEP-70 high-risk tumors have particularly bad outcomes (estimated mean PFS = 42 months), and no other attribute can significantly split the patient population. The algorithm then selects the attribute that best divides the remaining patient population, which is ISS stage III status (p < 0.001). These 68 patients (40 of whom had an event) without GEP-70 high-risk tumors but who are ISS stage III cannot be significantly divided by any other attribute and have relatively poor outcomes (estimated mean PFS = 54 months). The remaining 188 patients (86 of whom had an event) are best divided by deconvolution-based mast cell estimates (p = 0.001). Those 40 patients (26 of whom had an event) with predicted mast cell levels of 8.97% or lower also have relatively short PFS (estimated mean PFS = 61 months). The remaining 148 patients had a relatively long PFS of 82 months. Importantly, the model does not select any more attributes that can pass a cutoff of p = 0.01, implying that of these deconvolved immune cell estimates and clinical characteristics, GEP-70, ISS stage III, and mast cell estimates are the strongest predictors. Clinically estimated tumor burden, which is inversely correlated with deconvolved mast cell percentages, was not selected by the tree and is, therefore, less informative. (B) Pretreatment samples with known ISS stage, GEP-70 status, and mast cell signature scores (n = 320). The 8 rows show all possible combinations of GEP-70 high risk versus not high risk, ISS stage III versus not stage III, and mast cell signature as high versus low (≤ 8.97%). GEP-70, 70-gene Prognostic Risk Score; ISS, International Staging System; OS, overall survival; PFS, progression-free survival; TME, tumor microenvironment.

Fig 6. Progression of immune states through pClusters to elevated risk.

(A) Shown are the deconvolved cell-type estimates for 131 samples progressing through the 3 SMM/MGUS clusters (pCluster 1: MGUS [n = 27], SMM [n = 11]; pCluster 2: MGUS [n = 24], SMM [n = 40]; pCluster 3: MGUS [n = 4], SMM [n = 25]). Also shown are 354 patients with NDMM in Cluster 5 (n = 106) and in other clusters (n = 248). The cells outlined in blue in the SMM/MGUS clusters show a significant difference (Wilcoxon FDR ≤ 0.05) between patients with SMM and those with MGUS. The cells outlined in green show a significant difference between patients with SMM and those with NDMM. The plasma cell estimates outlined in dotted red show that the overall plasma cell estimates (tumor plasma cells, plasma cells, and B cells) are very similar between SMM pCluster 3 (25.56%) and NDMM clusters that are not “low-granulocyte” Cluster 5 (27.85%). (B) Shown are the aggregate tumor and microenvironment factors that characterize the high-risk microenvironment. The top section shows cells whose proportion is elevated (blue) or reduced (red) in the high-risk microenvironment. The bottom section shows a plausible mechanism of interaction between the tumor and microenvironment involving VCAM1 and other factors shown in S7 Fig. APOBEC3B, apolipoprotein B MRNA editing enzyme catalytic subunit 3B; CD, cluster of differentiation; FAM133A, family with sequence similarity 133 member A; FDR, false discovery rate; GBP1, guanylate-binding protein 1; IFI44L, interferon-induced protein 44-like; IFI6, interferon-inducible protein 6; IFIT1/3, interferon-induced protein with tetratricopeptide repeats 1/3; ISG15, interferon-stimulated gene 15; MAD2L1, mitotic spindle assemble checkpoint protein; MGUS, monoclonal gammopathy of undetermined significance; MITD1, microtubule interacting and trafficking domain 1; NDC80, kinetochore protein NDC80 homolog; NDMM, newly diagnosed multiple myeloma; NK, natural killer; pCluster, premalignant microenvironment cluster; RASSF6, Ras association domain family member 6; SMAD, homologies to the Caenorhabditis elegans SMA ("small" worm phenotype) and Drosophila MAD ("Mothers Against Decapentaplegic") family of genes; SMM, smoldering multiple myeloma; VCAM1, vascular cell adhesion protein 1.