Most patients with multiple myeloma (MM), a bone marrow (BM) malignancy, experience relapse after initial therapy (1). Relapsed/refractory (R/R) MM remains a challenge (2), and there is a high unmet medical need for more targeted therapies in this setting. Response to MM therapy can vary markedly even among patients at the same disease stage and with comparable clinical characteristics, and tumor intrinsic features such as chromosomal abnormalities, mutations, and gene expression patterns are well-established risk factors (3). An immunosuppressive tumor microenvironment (TME) is essential for MM disease establishment and progression and is implicated in protection of MM cells against anti-tumor therapies (4). Features of the TME, such as CD8+ cells, tumor cells and osteoclasts, may have a considerable impact on the immunologic and clinical efficacy of therapy in MM (5–7). Integration of transcriptomic and proteomic data from both the tumor and the TME can help elucidate the underlying features of this molecular heterogeneity, refining our understanding of the tumor biology (8). Improved understanding of the TME and its impact on therapeutic efficacy may support selection of targeted therapies for specific patient subgroups.

Programmed death-1 (PD-1) is an inhibitory receptor expressed by regulatory T cells (Treg), B cells and some myeloid cell populations (9), and is a validated target for cancer immunotherapy. However, two recent phase 3 studies of PD-1 inhibition via the monoclonal antibody pembrolizumab in patients with treatment-naïve and R/R MM were halted early due to an unfavorable benefit-risk profile (10, 11). Programmed death-ligand 1 (PD-L1), a PD-1 ligand, is expressed on MM cells, with higher levels of expression observed after relapse and with advanced disease (12). PD-L1–expressing immune cells in the TME have been shown to promote MM cell survival and potential immune escape (13). PD-L1 is also expressed on dendritic cells and myeloid-derived suppressor cells in MM patients (14–16); therefore, the anti-PD-L1 monoclonal antibody atezolizumab (Atezo) may have therapeutic application in MM (17). Unlike PD-1 inhibition, PD-L1 inhibition with Atezo may exert a different biologic influence on the TME with more favorable efficacy and tolerability.6 However, PD-L1 inhibition alone is insufficient to induce a durable clinical response (18).

Daratumumab (Dara) is an anti-CD38 monoclonal antibody with multiple modes of immune activity that is approved as a single agent and in combination with proteasome inhibitors (PIs) and immunomodulatory drugs (IMiDs) for patients with new and R/R MM (19, 20). Lenalidomide (Len) and pomalidomide (Pom) are IMiDs that induce myeloma cell death and promote immunity through B- and T-cell activation, dendritic cell maturation and other activities (21, 22). Combinations of these agents with Atezo may be advantageous due to their immunomodulatory activities.

We present data from GO29695 (NCT02431208), an open-label phase 1b study to evaluate the safety and efficacy of Atezo alone or in combination with Len, Dara, Dara and Len, or Dara and Pom in patients with MM who are R/R or post-autologous stem cell transplantation (ASCT) with residual disease ( Figure 1 ). Exploratory biomarker analyses were performed in patients receiving Atezo monotherapy or Atezo+Dara to validate the pharmacodynamics and mode of action of Atezo (23) and to assess whether Atezo+Dara is additive/synergistic for CD8+ T-cell activation. To gain biological and mechanistic insight into R/R MM in patients receiving Atezo monotherapy or Atezo+Dara, high-dimensional phenotyping was performed and the resultant multiscale data were analyzed integratively using a novel, unbiased, unsupervised machine-learning algorithm based on a patient similarity network (24). These results were compared with clinical data such as patient selection criteria and outcome to investigate correlations between tumor, immune and clinical features.

This study showed that Atezo alone and in combination with Dara and IMiDs was safe in myeloma patients. There was no safety signal and the observed safety profiles of the combinations were consistent with those of the individual agents. The correlative analysis showed that Atezo had the expected immunologic effects in the T cell compartment even in the absence of clinical activity, and multi-omic SNF analysis of cellular and molecular data from the TME recapitulated the cohorts with high fidelity and revealed novel correlations with clinical outcome.

Biomarker data from Cohort A suggest that Atezo promotes T-cell activation as early as C1D15, and all Dara-naïve patients showed an on-treatment increase in T-cell activation and proliferation in blood, although the same effect was not observed in patients with Dara-refractory disease. Evaluation of BMAs and osteoclasts allowed stratification of non-responders and responders, and analysis of baseline osteoclast levels was able to differentiate Dara-naïve patients from those with Dara-refractory disease. This suggests that osteoclasts may contribute to the immunosuppressive TME that inhibits T-cell function as reported previously (28), consistent with the observed lack of T-cell activation observed in Dara-refractory disease.

Patients in the Dara-containing Cohorts (D1, E1 and F1) demonstrated deep and durable responses, however, patients with Dara-refractory disease did not respond to Dara + Atezo (Cohort D3). Similarly, patients in D2 did not recapitulate the promising responses seen in D1, although differing patient selection criteria may have biased this cohort to older patients and greater frequency of high-risk cytogenetics. Furthermore, baseline CyTOF data revealed that patients in the Dara-refractory cluster had higher numbers of the senescent-type CD57+ T cells, which are known to be unable to inhibit growth of malignant cells and may have negatively affected the immune response to tumor antigens (29, 30). The TME-specific mRNA gene expression data layer enriched for patients in Cohort D2 in cluster CD138nRNA3.1 which had significantly higher expression of integrin signal pathway genes, indicating that integrin signaling may represent a mechanism of resistance to Dara+Atezo and a potential therapeutic target. Integrins are expressed on TME cells, such as fibroblasts and endothelial cells, which participate in the cross-talk relevant to tumor progression, specifically functioning as mediators of the interaction between tumor cells, cancer-associated fibroblasts and the matricellular proteins relevant to tumor progression (31). Although integrin-targeting cancer therapies are not effective alone, there is emerging evidence that a combination approach together with anti-PD-1 can improve response. (32) These findings show that the selection criteria reflected not only the clinical status of these patients, but also the state of the TME.

High-dimensional immune profiling and integrative clustering analysis of baseline and on-treatment samples revealed novel associations with outcome in both Dara-naïve and Dara-refractory R/R MM treated with Atezo+Dara. Pairwise analysis comparing matched treatment samples to baseline identified potential biomarkers, such as NKG2D and T cell activation/exhaustion genes, that were differentially expressed across patient subgroups. Similarly, the one-vs-all comparisons of the hallmark gene sets in the CD138–negative gene expression data layer revealed that cluster 1 is enriched for the T cell gamma delta gene signature and the T effector 6 gene signature. This is in line with previous studies, in which these gene signatures have been associated with response to cancer immunotherapy treatment (33, 34). A recent study of gene expression in patients with R/R MM revealed that a number of immune response-regulating genes were the most downregulated pathways in Dara-resistant vs Dara-naïve patients, further supporting the hypothesis that changes in the immune compartment of the TME may play a role in resistance to Dara (35). An integrated multi-omics approach has previously been proposed in patients with R/R MM, in which patients were treated based on their tumor DNA and RNA sequencing data, with results demonstrating feasibility and early efficacy (36). The authors noted that the addition of profiling of the TME (using single-cell RNAseq or mass cytometry) into their platform may further improve the specificity of the approach. SNF analyses in this study confirm that integration of high-dimensional tumor and TME data is feasible and can identify potential biomarkers of resistance or response in the setting of combination immunotherapy.

Given the complexity of the immune response to cancer and the proliferation of immunotherapeutic strategies in MM, it is critical to identify predictive and prognostic biomarkers to guide treatment. High-dimensional single-cell and whole transcriptomic approaches profiling the tumor and the microenvironment as applied in this study are being more readily applied clinically (37). Conventional single cell diagnostics such as FISH and flow cytometry are limited in their ability to capture the diversity of cell types and states. High-dimensional flow cytometry and mass cytometry (CyTOF) have made it possible to detect significantly larger numbers of proteins in thousands of cells, allowing deeper molecular characterization of cell populations in detail in an increasingly cost-effective manner. These assays bring their own set of challenges; for example, gold standard methods of analyzing flow cytometry data are inefficient and unreliable when applied to high-dimensional data. The FlowCAP (Flow Cytometry: Critical Assessment of Population Identification Methods) Consortium has recognized that high-dimensional assays such as CyToF require algorithm-based clustering approaches that have the advantages of being unbiased and automated (38). Biomarkers derived from high dimensional assays and data analyses may eventually inform treatment (39).

These findings demonstrate that anti-PD-L1 therapy with Atezo (as monotherapy and in combination) is safe and has the expected immunologic activity in MM patients. Integrated SNF analysis of high-dimensional immunophenotyping data demonstrated that patient selection based on clinical criteria had a significant impact on the immunologic state of the TME at baseline. Novel correlations with clinical outcome suggest that T-cell associated factors may predict response to treatment. With the rapid introduction of novel immune therapies in MM including CAR-T cells and bispecific antibodies, addressing mechanisms of immune escape is even more critical. Therefore, we believe that the safety and efficacy data reported here support further investigation of PD-L1 checkpoint inhibition combinations in myeloma. This rational immune-oncology approach may enhance understanding of myeloma-TME biology and promote biomarker-driven treatment.

The data presented in the study are deposited in European Genome-Phenome Archive (EGA) repository, accession number EGAS00001007286.

The study was conducted in accordance with the International Conference on Harmonization guidelines for Good Clinical Practice, and the protocol was approved by the ethics committees of all participating centers and registered at ClinicalTrials.gov (NCT02431208). All patients provided written informed consent. The patients/participants provided their written informed consent to participate in this study.

HC, SB, TN, EW-F and HH contributed to the conception and design of the study. SW, LCJ, NN, and RV contributed to the provision of study materials or patients. HH, HC, SB, TN, and EW-F collected and assembled the data. All authors analyzed and interpreted the data, contributed to the writing of the manuscript, and provided final approval of the manuscript.