Myelodysplastic Syndrome Revealed by Systems Immunology in a Melanoma Patient Undergoing Anti–PD-1 Therapy

Antibodies targeting the programmed cell death-1 (PD-1) receptor and its ligand (PD-L1) have demonstrated remarkable activity in a variety of solid tumor malignancies and Hodgkin disease (1–5). Agents such as nivolumab and pembrolizumab block this key immune checkpoint to generate effective CD8⁺ T cell–mediated antitumor immune responses (6). Although generally well tolerated, anti–PD-1/PD-L1 agents have also induced grade 3–4 immune-mediated adverse events (∼5% of patients; refs. 1, 3, 7). The ability to monitor the immune system as a whole is needed in clinical settings where therapies are designed to alter the interaction between multiple cell types, such as the blockade of the PD-1/PD-L1 interaction. Mass cytometry has emerged as an important platform for comprehensive and minimally biased immune profiling (8). This approach has been applied effectively in the context of healthy human bone marrow, blood, and tonsil tissue (9–12). An important element of high-dimensional mass cytometry is the use of unsupervised computational data analysis tools that reveal and characterize cells with unusual phenotypes (13–15).

To examine the effects of immune checkpoint inhibition on peripheral blood cell subsets, we performed serial blood acquisition in patients with advanced melanoma being treated with pembrolizumab. Multiplexed mass cytometry analysis of peripheral blood characterized the phenotypic evolution of all major immune cell and blast populations. Here, we present a case report of a single patient from this study who experienced clinical emergence and progression of myelodysplastic syndrome (MDS) while undergoing treatment with pembrolizumab.

The patient gave consent to an Institutional Review Board (IRB)-approved research protocol to collect serial blood prior to treatment (baseline) and following pembrolizumab (at 3 weeks, 6 weeks, 12 weeks, and 24 weeks). Peripheral blood was obtained during regularly scheduled laboratory blood draws in accordance with the Declaration of Helsinki following protocols approved by Vanderbilt University Medical Center (VUMC) IRB.

Peripheral blood mononuclear cells (PBMC) were isolated from 20 mL of freshly drawn blood using sodium heparin anticoagulant and Ficoll–Paque PLUS (GE Healthcare Bio-Sciences) centrifugation. Freshly isolated PBMCs were immediately cryopreserved in FBS (Life Technologies) containing 12% DMSO (Fischer Scientific).

Purified antibodies from Biolegend were labeled using MaxPar DN3 kits (Fluidigm Sciences) and stored at 4°C in antibody stabilization buffer (Candor Bioscience GmbH). PBMC were first incubated with a viability reagent (25 μmol/L cisplatin, Enzo Life Sciences) as previously described (16). For staining, 2 × 10⁶ cells were washed in PBS containing 1% BSA and stained in 50 μL PBS/1% BSA containing antibody cocktail. Cells were stained for 30 minutes at room temperature using antibodies listed (Supplementary Table S1). Cells were washed twice in PBS/1% BSA and then fixed with 1.6% paraformaldehyde (PFA, Electron Microscopy Sciences). Cells were washed once in PBS and permeabilized by resuspending in ice-cold methanol. After incubating overnight at −20°C, cells were washed twice with PBS/1% BSA and stained with iridium DNA intercalator (Fluidigm Sciences) for 30 minutes at room temperature. Finally, cells were washed twice with PBS and twice with deionized water before being resuspended in 1× EQ Four Element Calibration Beads (Fluidigm Sciences) according to the manufacturer's instructions and collected on a CyTOF 1.0 mass cytometer (Fluidigm Sciences) at the Vanderbilt Flow Cytometry Shared Resource. Events were normalized prior to analysis.

Analysis was performed on Cytobank using published techniques, including viSNE (13, 17). Each file was pre-gated on CD45^lo events. All CD45^lo events were used and BTLA, CD69, HLA-DR, CD45RO, CD44, CD27, CD3, CD45RA, CXCR3, CD33, CD16, PD-L1, CD4, CD8a, CD43, ICOS, CD20, CD38, CCR4, CD45, CCR7, CD25, CXCR5, CD57, PD-1, and CD56 were used to create the t-SNE axes of the viSNE map. Natural killer (NK) cells were gated as CD16⁺CD3^–CD19^–HLA-DR^– to allow consistent gating between this study and other immune monitoring studies where CD56 is not available. For this patient, equivalent NK-cell frequencies were observed with and without inclusion of CD56 in the gating scheme.

A 72-year-old female patient with a past medical history of breast cancer treated with mastectomy and chemotherapy (24 years prior) developed swelling in her thigh. CT scan showed a 2 × 4 × 3 cm complex, cystic mass in the soft tissues of the thigh. She underwent a resection that revealed melanoma with involved lateral margins. No primary site was identified, and she received 2,800 cGy via external beam therapy to the resection margin in the thigh (American Joint Committee on Cancer [AJCC] stage TxNxM1a). A surveillance PET/CT scan performed 18 months later showed multiple new pulmonary nodules, and she received ipilimumab, 3 mg/kg for four doses. Repeat CT scan showed disease progression in the lungs, and she was started on pembrolizumab, 2 mg/kg every 3 weeks (pembrolizumab expanded access program). Her initial surveillance scans in 3 months showed a mixed response in lung nodules but overall stable disease. Her subsequent CT scan 6 months into therapy confirmed disease progression, primarily in two lung nodules (Fig. 1A). She underwent radiation to those nodules and has continued pembrolizumab.

Prior to starting pembrolizumab, the patient was noted to have mild thrombocytopenia (platelet count, 95 × 10⁹/L), which had been present but stable for the previous 18 months, and normal white blood cell count and hemoglobin. Her monocyte count was slightly elevated (1.27 × 10⁹/L), but otherwise she had a normal differential. One month after starting pembrolizumab, she developed a clinically detectable but small circulating myeloid blast population on her complete blood count (CBC) differential that remained at <1% for the next 2 months (Fig. 1B). Blasts were not detected in her pretreatment sample on her CBC differential, but were detected pretreatment by CyTOF analysis (Fig. 2). After treatment, her peripheral blast percentage rose to 4.4%, and her platelets and hemoglobin fell to 72 × 10⁹/L and 9.4 g/dL, respectively, prompting a bone marrow biopsy. This revealed refractory anemia with excess blasts-2 (RAEB-2) with 8.5% blasts. During this workup, her peripheral blast percentage ranged from 7% to 11% and she began decitabine treatment. Since initiating decitabine, she has felt well and her peripheral blast count has stabilized.

Samples of peripheral blood from prior to anti–PD-1 therapy, 3 weeks, 12 weeks, and 6 months following the initiation of therapy were analyzed by a 33-parameter mass cytometry panel. Major cell populations were revealed using traditional biaxial plots (Fig. 2). As a subset of PBMCs, blasts ranged in abundance from 1.16% pretherapy to 4.65% at 6 months after the start of treatment (Fig. 2A and C, left). In comparison with mature myeloid and monocyte cell populations, the majority of peripheral blasts expressed lower CD45 and CD33, higher CD38, and comparable HLA-DR (Fig. 2B). Four additional cell populations were monitored using mass cytometry over the course of anti–PD-1 therapy (Fig. 2C, right). Myeloid cells (HLA-DR⁺CD19^–CD3^–CD16^–) increased over the course of anti–PD-1 therapy, while T cells (CD3⁺CD19^–), B cells (CD19⁺CD3^–), and NK cells (CD16⁺CD3^–CD19^–HLA-DR^–) decreased.

To determine which cells could be modulated by pembrolizumab, mass cytometry was used to monitor PD-1 protein expression on all major peripheral blood subsets. In the blood of healthy donors, 1.8% ± 0.68% of HLA-DR^posCD19^negCD3^negCD16^neg myeloid lineage PBMCs expressed PD-1 (n = 5). In contrast, elevated levels of PD-1⁺ myeloid cells were seen in all pre- and posttreatment samples collected from the melanoma patient (Fig. 3A). PD-1⁺ myeloid cells decreased over therapy from 4.04% pretreatment to 2.81% 6 months after the start of pembrolizumab. The percentage of PD-1⁺ CD4⁺ T cells in PBMCs was high prior to therapy (4.93% in the patient prior to therapy vs. 3.31% ± 1.28% in healthy donors, n = 5). However, by week 3 after therapy, the frequency of both PD-1⁺ CD8 and CD4 T cells in PBMCs decreased (1.94% and 2.29%, respectively, in the patient 6 months after start of therapy vs 4.44% ± 2.15% and 3.31% ± 1.28%, respectively, in healthy donors, n = 5). B cells, NK cells, and blasts were very rarely PD-1⁺ (Fig. 3A, < 2% PD-1⁺ at all observed times).

Mass cytometry was also used to monitor expression of PD-L1, a key ligand for PD-1 and potential biomarker of anti–PD-1 treatment response (18). PD-L1 expression was largely absent; no cell subset was observed to express PD-L1 on >1.5% of cells (data not shown). Monocytes were the only peripheral cell subset to exhibit increased PD-1 expression, and a large increase in activation marker CD45RA was seen on monocytes over the course of therapy (Fig. 3B; ref. 19).

To assess the phenotypic changes of blasts over the course of anti–PD-1 therapy, blast cells from each time point were gated based on immunophenotype and analyzed in a single viSNE plot (Fig. 4A). viSNE approximates high-dimensional relationships using a two-dimensional (2D) scatter plot, or map, where each dot represents a single cell (13). Cells close together on a viSNE map are phenotypically similar across all parameters used to make the map. Gates were used to assess the percentage of cells in the most dense regions of the viSNE map, and the fold change in population frequency was compared over therapy (Fig. 4B). Before therapy, blasts fell mostly within population 1, defined by high expression of HLA-DR, CD33, CD38, CD43, and CD44 and intermediate expression of CD45RA. By 6 months after the start of therapy, the cells had shifted phenotype and largely fell into population 7, defined by expression of CXCR3 in addition to the phenotypic characteristics of population 1. Each population was distinguished by key phenotypic differences (Fig. 4C). Populations 3 and 5 were distinguished by a lack of HLA-DR and CD33 expression along with high expression of the T-cell costimulatory molecule CD28 and the chemokine trafficking marker CXCR3, respectively. Additionally, population 6 expressed both myeloid markers (HLA-DR and CD33) and neural cell adhesion molecule (NCAM/CD56), a marker expressed on NK and neural lineage cells, as well blasts in MDS and acute myelogenous leukemia (20).

We report here a case study of a patient who experienced progression of MDS while undergoing anti–PD-1 therapy for advanced melanoma. Subclinical MDS was detected by mass cytometry prior to treatment and blasts became clinically apparent and progressed during treatment. Given the importance of the immune checkpoint inhibitor treatment class for melanoma and other malignancies and the unexpected nature of the emergence and progression of MDS, we described the clinical features and immune evolution of this case. As anti–PD-1 directed therapies are expected to be approved and tested in numerous malignancies, cellular response profiles of the immune system as a whole must be explored. When paired with unsupervised computational analysis tools (13), mass cytometry provides a complete platform for immune monitoring that effectively addresses the heterogeneity and complexity of primary tissue biopsies by revealing and characterizing known and novel cell populations.

In this case study, a population of myeloid peripheral blasts emerged and was readily detected by both computational and human expert analysis. Both complete blood counts and mass cytometry were able to identify these cells, ranging from approximately 1% to 5% of PBMCs. Of interest, although no peripheral blasts were identified prior to treatment on clinical analysis, a clear population was detected by mass cytometry. Lymphocytes decreased while myeloid cells increased, indicative of disrupted hematopoiesis. Additionally, mass cytometry effectively characterized PD-1 and PD-L1 expression on all major peripheral blood immune subsets. Phenotypes of immune cells and blasts were tracked over the course of therapy, revealing an increase in PD-1⁺ monocytes and the emergence of a CXCR3⁺ blast population, respectively. This phenotypic evolution of peripheral blasts may be shaped by the presence of anti–PD-1, or may simply reflect the natural history of MDS.

Peripheral blasts were largely negative for PD-1 and PD-L1 protein expression, underscoring the premise that if anti–PD-1 is used to treat MDS or myeloid leukemia in the future, additional treatments may be needed to augment therapy in cases where a PD-L1^– peripheral blast population is present (21). Notably, the frequency of PD-1⁺ T cells decreased following anti–PD-1 treatment. This observation might be explained by occupation or internalization of surface PD-1 mediated by pembrolizumab (1), natural variation over time, treatment-induced changes in the blood T-cell milieu, and trafficking of PD-1⁺ cells to another location, such as tumor. The presence of myeloid blasts may alter the outcome of immune checkpoint inhibitors for PD-1–expressing T cells.

In this case, it is unclear whether pembrolizumab contributed to MDS progression. Increasingly, the role of the immune system in the development and progression of MDS is being elucidated. Inflammatory cytokines and Th17-cell subsets are predominant in low-risk disease, but in high-risk MDS (as in our patient), regulatory subsets begin to dominate, giving rise to an overall suppressive immune microenvironment (22, 23). Myeloid-derived suppressor cells (MDSC) and associated suppressive cytokines TGFb and IL10 are increased in MDS bone marrow and may promote progression of disease and suppress normal hematopoiesis through immune suppression and dysregulation (24). From our data, it is clear that multiple immune regulatory processes are involved in advanced MDS, and rescue of exhausted T cells through PD-1 blockade may only solve part of the problem. Perhaps inhibition of the MDSCs or cytokine responses could serve as a synergistic approach together with PD-1 blockade.

This patient is the second reported case of the development and progression of MDS while undergoing anti–PD-1 therapy (25). The other reported patient had Hodgkin lymphoma and received multiple prior chemotherapies. Our patient had also received both chemotherapy and radiotherapy, and MDS was likely precipitated by these other factors. These cases indicate that it will be valuable to monitor the myeloid compartment in anti–PD-1–treated patients.

This case study highlights both the complexity of immune monitoring in the context of cancer treatment and the power of high-dimensional tools to provide a comprehensive and minimally biased view of dynamic cellular systems. It is clear that high-dimensional immune monitoring will be of significant value in cases where immune checkpoint inhibitors are used in the context of MDS, myeloid leukemia, or other myeloid bone marrow failure syndromes. These results also underscore the critical need to characterize mechanisms of response to immune checkpoint inhibitors, as these mechanisms may depend greatly on the disease context in which they are applied.

D.B. Johnson reports serving as a consultant/advisory board member for Bristol-Myers Squibb and Genoptix. M.R. Savona reports receiving commercial research support from Astex and Celgene and is a consultant/advisory board member for Astex and Celgene. J.M Irish is the cofounder and a board member for Cytobank Inc., reports receiving commercial research support from Incyte Corp., and has received speakers bureau honoraria from Novartis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A.R. Greenplate, D.B. Johnson, M.R. Savona, I. Puzanov, P.B. Ferrell Jr, J.M. Irish

Development of methodology: A.R. Greenplate, D.B. Johnson, M.R. Savona, J.M. Irish

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.R. Greenplate, M.R. Savona, I. Puzanov

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.R. Greenplate, D.B. Johnson, M. Roussel, M.R. Savona, J.A. Sosman, I. Puzanov, P.B. Ferrell Jr, J.M. Irish

Writing, review, and/or revision of the manuscript: A.R. Greenplate, D.B. Johnson, M.R. Savona, J.A. Sosman, I. Puzanov, P.B. Ferrell Jr, J.M. Irish

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.R. Greenplate, I. Puzanov, P.B. Ferrell Jr, J.M. Irish

Study supervision: M.R. Savona, P.B. Ferrell Jr, J.M. Irish

This study was supported by NIH/NCI R00 CA143231-03 (J.M. Irish and A.R. Greenplate), F31 CA199993 (A.R. Greenplate), K12 CA090625 (P.B. Ferrell Jr and D.B. Johnson), and the Vanderbilt-Ingram Cancer Center (VICC, P30 CA68485). ACS Melanoma Professorship (J.A. Sosman).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.