Research ResourcesTUMOR IMMUNOLOGY

Defining the emergence of myeloid-derived suppressor cells in breast cancer using single-cell transcriptomics

See allHide authors and affiliations

Science Immunology  21 Feb 2020:
Vol. 5, Issue 44, eaay6017
DOI: 10.1126/sciimmunol.aay6017

Getting a hold on MDSCs

How myeloid-derived suppressor cells (MDSCs) arise and whether they can be therapeutically targeted akin to exhausted T cells are both areas of active investigation. A persistent challenge in studying MDSCs has been the identification of MDSC-specific cell surface markers that can facilitate their isolation and characterization. Here, by carrying out scRNAseq in a mouse model of breast cancer, Alshetaiwi et al. have defined gene signatures that distinguish MDSCs from other myeloid and granulocytic cells. By mining these datasets, they have identified CD84 to be a robust cell surface marker for identification of MDSCs in both human and murine breast cancer. Whether their findings can be extended to MDSCs in other cancer settings remains to be seen.


Myeloid-derived suppressor cells (MDSCs) are innate immune cells that acquire the capacity to suppress adaptive immune responses during cancer. It remains elusive how MDSCs differ from their normal myeloid counterparts, which limits our ability to specifically detect and therapeutically target MDSCs during cancer. Here, we sought to determine the molecular features of breast cancer–associated MDSCs using the widely studied mouse model based on the mouse mammary tumor virus (MMTV) promoter–driven expression of the polyomavirus middle T oncoprotein (MMTV-PyMT). To identify MDSCs in an unbiased manner, we used single-cell RNA sequencing to compare MDSC-containing splenic myeloid cells from breast tumor–bearing mice with wild-type controls. Our computational analysis of 14,646 single-cell transcriptomes revealed that MDSCs emerge through an aberrant neutrophil maturation trajectory in the spleen that confers them an immunosuppressive cell state. We establish the MDSC-specific gene signature and identify CD84 as a surface marker for improved detection and enrichment of MDSCs in breast cancers.


Breast cancer is one of the most prevalent types of cancer with more than 260,000 new cases and more than 40,000 deaths in 2018 in the United States (1). During tumor development, breast cancer cells secrete various cytokines such as granulocyte-macrophage colony-stimulating factor (GM-CSF), which exert systemic effects on hematopoiesis and myeloid cell differentiation promoting the development of myeloid-derived suppressor cells (MDSCs) (2, 3). These MDSCs are a heterogeneous population of neutrophil- and monocyte-like myeloid cells, which are increasingly recognized as key mediators of immunosuppression in various types of cancer (3, 4). In patients with cancer, increased numbers of MDSCs in circulation correlate with advanced clinical stages, increased metastatic progression, and immunosuppression (5). MDSCs can mediate immunosuppression through multiple mechanisms including the production of reactive oxygen species (ROS) and depletion of key amino acids required for T cell proliferation through expression of arginase (Arg) and indoleamine 2,3-dioxygenase (68). In addition, MDSCs produce a range of immunosuppressive and cancer-promoting cytokines including interleukin-10 (IL-10) and transforming growth factor–β (9). Besides their immunosuppressive function, MDSCs may also actively shape the tumor microenvironment through complex cross-talk with breast cancer cells and surrounding stroma, resulting in increased angiogenesis, tumor invasion, and metastasis (8, 10, 11).

The unique molecular features of MDSCs are now unclear, and it remains elusive whether MDSCs represent a unique subpopulation of myeloid cells that differ from their normal, healthy counterparts. This limits our ability to determine specific MDSC functions as opposed to bulk-level changes in neutrophils or monocytes during cancer. In mice, MDSCs are defined through the expression of CD11b+Gr1+ and can be further classified into CD11b+Ly6ClowLy6G+ granulocytic MDSCs (G-MDSCs) and CD11b+Ly6C+Ly6G monocytic MDSCs (M-MDSCs) (12). In humans, G-MDSCs are defined as CD11b+CD14CD15+ or CD11b+CD14CD66b+, and M-MDSCs are defined as CD11b+CD14+CD15, followed by additional functional characteristics such as T cell suppression and ROS assays (12). However, these markers overlap with those defining healthy neutrophils and monocytes, which makes it challenging to distinguish MDSCs from normal cells to advance our understanding of MDSC biology and, ultimately, to establish novel therapeutic avenues to interfere with their tumor-promoting and immunosuppressive roles.

Here, we used single-cell RNA sequencing (scRNAseq) to delineate the molecular features of MDSCs in the MMTV-PyMT (mouse mammary tumor virus–polyomavirus middle T antigen) mouse model of breast cancer. Our computational analysis of 14,646 single-cell transcriptomes revealed a unique MDSC gene signature, which is largely shared between G-MDSCs and M-MDSCs but strongly differs from their normal myeloid counterparts. Focusing on G-MDSCs, our pseudotemporal analysis delineates the emergence of MDSCs as an aberrant differentiation state that forms a separate branch during the transition of neutrophil progenitors into mature neutrophils. Further interrogation of the distinct MDSC gene expression signature identified several surface markers [e.g., CD84 and junctional adhesion molecule (JAML)] for faithful MDSC detection and prospective enrichment.


The spleen is the predominant site of MDSC accumulation in tumor-bearing mice

Mice expressing the PyMT driven by the MMTV promoter (13) develop breast tumors that closely resemble human pathogenesis (14) and give rise to MDSCs during tumor progression (2). Here, we used the MMTV-PyMT transgenic mouse model of breast cancer to explore the role of MDSCs during breast cancer progression. We first sought to confirm the most reliable organ site of MDSC accumulation for further molecular studies of this cell population. In agreement with previous reports in other murine models of cancer (2), we observed that later stages of cancer progression were associated with an expansion of CD11b+Gr1+ myeloid cells in bone marrow, blood, spleens, lungs, brains, and primary tumors (fig. S1, A to C) and an enlargement of the spleen of tumor-bearing PyMT mice compared with wild-type (WT) controls (fig. S1, D and E). To functionally confirm whether MDSCs are present in expanded populations of CD11b+Gr1+ cells in tumor-bearing mice, we isolated CD11b+Gr1+ cells from various organs of tumor-bearing and control mice by fluorescence-activated cell sorting (FACS) and cocultured these with isolated T cells to measure suppression of T cell proliferation induced by CD3/CD28 costimulation (2) and ROS formation as a readout for MDSC function (fig. S2A) (12). We found that CD11b+Gr1+ cells sorted from spleens of tumor-bearing mice significantly suppressed CD4+ and CD8+ T cell proliferation (fig. S2, B and C), whereas CD11b+Gr1+ cells from control spleens showed no measurable effect on T cell proliferation. CD11b+Gr1+ cells sorted from bone marrow (fig. S2, D and E) and lungs (fig. S2, F and G) of tumor-bearing mice demonstrated nonsignificant suppression of T cell proliferation. These findings were further corroborated by ROS production assays as measured by flow cytometry using 2′,7′-dichlorofluorescein diacetate (H2DCFDA), which showed that only spleen-derived CD11b+Gr1+ cells from tumor-bearing mice exhibited significant oxidative burst formation as a hallmark for MDSCs (fig. S2, H and I). Together, these results establish the spleen as one of the major sites of MDSC emergence during breast tumor formation in PyMT mice.

Single-cell transcriptomics reveal MDSCs as distinct clusters within neutrophilic and monocytic lineages

To determine how MDSCs differ from their normal myeloid counterparts on a cellular and molecular level, we used scRNAseq to compare the molecular differences of spleen-derived myeloid cells in tumor-bearing mice against the respective cell population from WT mice on an individual cell basis. We used a scalable droplet-mediated scRNAseq platform (10x Genomics Chromium) to profile FACS-purified live (SYTOX-negative) CD45+CD11b+Gr1+ myeloid cells from the spleens of tumor-bearing PyMT and control WT mice (Fig. 1A). We profiled two samples from tumor-bearing PyMT mice (9155 cells) and WT control mice (5491 cells), respectively, for a total of 14,646 cells that were sequenced at an average depth of ~50,000 reads per cell. The two libraries were aggregated and aligned together using the Cell Ranger pipeline (10x Genomics) to compensate for minor differences in library complexity. After quality control filtering to remove cells with low gene detection (<500 genes) and high mitochondrial gene content (>8%), we performed clustering and cell type identification analysis of combined PyMT and WT datasets using Seurat (fig. S3A) (15). Using the canonical correlation analysis (CCA) method (16), we identified the main cell types based on expression of hallmark genes for myeloid subsets (Fig. 1B) and determined their marker genes (fig. S3C and table S1). Neutrophils formed the largest population encompassing numerous distinct clusters (C0, C2, C4, C5, C7, and C8) characterized by high levels of Ly6g and Cxcr2 expression (Fig. 1, B and C). Monocytes were less abundant and less diverse, forming one cluster (C1) that was marked by the expression of Csf1r and Ccr2 (Fig. 1, B and C). We also detected two minor cell types: T cells (C9) expressing Cd3, Cd4, and Cd8 and B cells (C6 and C3) expressing Cd19, Cd22, and Cd79a (Fig. 1, B and C).

Fig. 1 Identifying MDSC-specific gene expression signatures using scRNAseq.

(A) Approach overview for single-cell analysis of and (SYTOX Blue–negative) CD45+CD11b+Gr1+ cells were sorted from the spleen of control WT and tumor-bearing PyMT mice by FACS after droplet-enabled scRNAseq. (B and C) Combined Seurat analysis of a total of 14,646 cells from control and PyMT mice shown in tSNE projection results in various distinct clusters of splenic CD11b+Gr1+ cells. Main cell types (T cells, B cells, neutrophils, and monocytes) are outlined on the basis of hallmark gene expression. (C) Feature plots of characteristic markers of the four main cell types showing expression levels with low expression in gray to high expression in purple. (D) G-MDSCs were identified in cluster C1 by expression marker genes (Arg2 and Il1β) from the PyMT sample. (E) Subset analysis of monocyte cluster identified M-MDSCs. Three clusters were found; cluster M2 was identified as M-MDSCs (positive for Arg2 and Il1β). (F) Heat map displaying the scaled expression patterns of top marker genes within each G-MDSC and M-MDSC cluster compared with normal neutrophil and monocyte clusters from WT mice, respectively. Yellow, high expression; purple, low expression. (G) Venn diagram showing the number of statistically significant marker genes and overlap between G-MDSC and M-MDSC. (H) GO term analysis using Enrichr of curated MDSC signature. (I) Validation using qPCR of selected up-regulated MDSC genes and statistical analysis using unpaired t test (means ± SEM of n = 3), *P < 0.05.

Further interrogation of neutrophil heterogeneity revealed that cluster C0 was marked by high levels of genes associated with a mature neutrophil state such as Camp (17) and high Ly6g expression (18); cluster C2 was strongly enriched in tumor-bearing PyMT mice (fig. S3E) and displayed high expression of MDSC-related genes such as Il1β and Arg2, two major immunosuppressive factors previously used to define MDSCs in cancer models (Fig. 1D) (4, 19); clusters C4 and C5 displayed overlapping marker gene expression including genes such as Cebpe and Retnig; and clusters C7 and C8 exhibited high expression of cell cycle genes such as Tuba1b and Cdc20, indicating the existence of a proliferative pool of neutrophils in the spleen.

We next focused on the monocyte-restricted cluster C1, which showed diffuse expression of MDSC genes Arg2 and Il1b in the combined analysis (Fig. 1D), suggesting that M-MDSCs were present but did not cluster distinctly from monocytes because of the more substantial differences between different cell types in the combined analysis. Therefore, we performed a monocyte-only clustering analysis after the removal of contaminating cells such as neutrophils and B cells to identify three distinct states (clusters M0 to M2) including a distinct M-MDSC population in cluster M2, which was strongly enriched in tumor-bearing PyMT mice (Fig. 1E; fig. S3, B, D, and F; and table S2). These analyses formed the basis for a detailed molecular definition of G- and M-MDSCs as described below. Our dataset represents a valuable single cell–level transcriptome analysis of MDSCs, which revealed that G- and M-MDSCs form distinct clusters that are unique from their normal myeloid counterparts.

G- and M-MDSCs share a conserved immune cell activation program that differs from normal myeloid cells

We next used our scRNAseq dataset to reveal the unique molecular features of MDSCs and to unravel the distinct biological programs that define the MDSC state. We performed differential expression analysis in Seurat to determine how G- and M-MDSCs from tumor-bearing mice differ from their normal counterparts, namely, neutrophils and monocytes in WT mice (Fig. 1F). Our analysis revealed 642 differentially expressed genes in G-MDSCs compared with normal neutrophils (table S3) and 223 differentially expressed genes in M-MDSCs compared with normal monocytes (table S4), demonstrating that MDSCs differ substantially from their normal myeloid counterparts. There was a substantial overlap between gene signatures for G- and M-MDSCs (105 of 200 top marker genes; Fig. 1G and table S5), indicating that a similar immunosuppressive cell state can be acquired by both monocytes and neutrophils independently. Shared markers included genes involved in immunosuppression such as Il1b, Arg2, Cd84, and Wfdc17 (20). Interferon-induced transmembrane protein 1 (Ifitm1), which has been reported to be involved in the progression of colorectal cancer (21, 22) and inflammatory breast cancer cells (23), was up-regulated in MDSCs. Additional MDSC markers included myeloid-associated immunoglobulin-like receptor family (Cd300ld), C-type lectin domain family 4 member E and D (Clec4e and Clec4d), IL-1f9 (Il1f9), activating protein 1 transcription factor subunit (Junb), cathepsin D (Ctsd), phospholipase A2 group VII (Pla2g7), and cystatin domain containing 5 (BC100530).

We next performed Gene Ontology (GO) term analysis on our MDSC gene signature (table S5) using Enrichr (GO Biological Process 2018) (24). The top GO terms included “neutrophil mediated immunity,” “neutrophil activation involved in immune response,” and “neutrophil degranulation” (Fig. 1H and table S6). Those terms included genes encoding complement C5a receptor 1 (C5ar1), S100 calcium-binding protein A11 (S100a11), C-type lectin domain family 4 member D (Clec4d), chemokine receptor 2 (Cxcr2), and myeloid cell surface antigen CD33 (Cd33). Several of these factors promote recruitment of neutrophils and MDSCs as reported for C5ar1 (25, 26) and S100a8/9 (27, 28). In addition, Cd33 is specifically expressed in MDSCs (29) and is currently pursued as a target in therapeutic approaches (30). Another significant GO term “cytokine-mediated signaling pathway” (Fig. 1H) included genes such as Il1β, Ifitm1, Junb, and Myd88. In particular, Myd88 has been reported to promote expansion of immature Gr1+ cells and may be involved in mediating T cell–suppressing cell states (31). In addition, this pathway included genes associated with MDSC accumulation and trafficking such as Cxcr2 (32, 33), Csf3r (34), and Ccr1 (35), suggesting that MDSCs may be able to respond to recruitment signals from sites of inflammation such as the primary tumor or metastatic sites.

To validate the MDSC gene signature, we isolated CD11b+Gr1+ cells from spleens of WT and tumor-bearing PyMT mice by FACS and subjected them to quantitative polymerase chain reaction (qPCR). Our scRNAseq results were broadly confirmed in this targeted approach because a large proportion of MDSC signature genes were significantly up-regulated in CD11b+Gr1+ cells from PyMT compared with WT (Fig. 1I). Together, these analyses revealed that G- and M-MDSCs share a conserved gene signature that strongly differs from their normal myeloid counterparts. This shared MDSC marker gene list shows certain differences and overlap with previous bulk transcriptome-level analyses of MDSCs, and it would be interesting to systematically determine whether bulk-level changes in these myeloid cell populations mask certain specific programs underlying MDSC cell function.

The MDSC gene signature is highly expressed in human breast cancer–associated neutrophils

To determine whether this MDSC gene signature is generalizable and translatable into the human context, we explored a recently published scRNAseq immune cell map including T cells, B cells, monocytes, and neutrophils from primary tumor samples of patients with breast cancer (36). We performed clustering of this dataset in Seurat to reproduce cell type labels (Fig. 2A) and then carried out an unbiased gene signature scoring of all cell types, which revealed that, specifically, neutrophils and monocytes in the tumor microenvironment express high levels of MDSC signature genes (Fig. 2B). To assess whether there are distinct subsets of neutrophils with particularly high MDSC signatures, we also analyzed neutrophils separately using unbiased clustering yielding four distinct states (Fig. 2, C and D). Cluster N0 showed MDSC-related marker genes S100A9 and CCR2, suggesting that this subset of neutrophils represents G-MDSCs in the tumor microenvironment of patients with breast cancer. Gene scoring analysis using our MDSC gene signature in these neutrophil subclusters indeed showed, by far, the highest scores in cells from cluster N0 (Fig. 2E). In addition, monocytes were analyzed separately, showing three distinct clusters that scored more ambiguously for our MDSC signature (Fig. 2, F to H), suggesting that M-MDSCs may not be directly comparable between mouse and human. Together, these analyses confirmed that, at least in the context of G-MDSCs, our MDSC gene signature derived from a murine breast cancer model is translatable into human disease, indicating that the MDSC state is largely conserved between mice and human.

Fig. 2 Comparative analysis using MDSC signature in myeloid cells from human patients with breast cancer.

(A) Seurat analysis of previously published scRNAseq dataset comprising various immune cell populations in primary human breast tumor samples (36) projected in UMAP with cell type labels as indicated in different colors. pDC, plasmacytoid dendritic cell; mDC, myeloid dendritic cell. (B) Violin plot showing relative MDSC score of all cells in this dataset ordered by cell type showing the highest scores in neutrophils and monocytes. (C) Separate unbiased Seurat clustering analysis of neutrophil alone projected in UMAP yielded four distinct clusters of neutrophils in this dataset. (D) Heat map showing top 10 marker genes for each neutrophil cluster. (E) Violin plots showing relative MDSC score ordered by neutrophil subcluster, showing that cluster 0 exhibits the highest expression of MDSC gene signature. (F) Subset monocyte-specific Seurat clustering analysis projected in UMAP yielded three distinct clusters of monocytes in this dataset. (G) Heat map showing top 10 marker genes for each monocyte cluster. (H) Violin plots showing relative MDSC score ordered by monocyte subclusters.

Identification of cell surface markers for MDSC detection and isolation

Our scRNAseq data revealed several previously unknown specific cell surface markers for MDSCs including CD84 and Amica1/JAML. CD84 is a cell surface receptor of the signaling lymphocytic activation molecule family (37) and is expressed on some immune cell types (38, 39). Amica1/JAML is a junctional adhesion molecule known to mediate the transmigration of neutrophils and monocytes by interacting with coxsackie-adenovirus receptor (CAR) expressed by epithelia (40). We profiled CD84 and JAML expression using FACS on CD11b+Gr1+ cells from different organs in tumor-bearing PyMT mice and WT mice. We first used fluorescence minus one (FMO) and isotype controls to determine specific marker expression (fig. S4, A and B). Next, we characterized CD84 and JAML expression in the CD11b+Gr1+ population from various organ preparations [bone marrow, lung, spleen, mammary fat pad (MFP), and primary tumor] and compared control WT with tumor-bearing PyMT mice. Whereas CD11b+Gr1+ cells from bone marrow and lung were generally negative for CD84 (Fig. 3A) and JAML (Fig. 3D), we found a significant number of CD11b+Gr1+ cells from the spleen and primary tumors of PyMT mice that exhibited high expression of CD84 (Fig. 3B) and JAML (Fig. 3E) compared with the respective WT controls. This is particularly apparent when cells from all organs are plotted side by side (Fig. 3, C and F). This observation of high expression of CD84 and JAML in spleen and primary tumors correlates with high MDSC capacity of CD11b+Gr1+ cells in these sites.

Fig. 3 Identification of cell surface markers for MDSCs in breast cancer models.

(A) CD84 expression profiling in WT and tumor-bearing PyMT showing that only spleen and primary tumor (TM) from PyMT exhibit significant expression. SSC-A, side scatter area. (B) Combined results and statistical analysis using unpaired t test (means ± SEM of n = 10), *P < 0.05. (D) Profiling JAML expression in WT and PyMT showing only spleen and tumor from PyMT exhibit significant expression. (E) Combined results and statistical analysis unpaired t test (means ± SEM of n = 3), *P < 0.05. (C and F) Concatenate multiple flow samples to visualize CD84 and JAML expression in one feature plot across all samples including FMO, bone marrow (BM), lung, spleen, MFP, and tumor from WT and PyMT; significant expression was only observed in spleen and tumor from PyMT. (G) Overview of PBMC collection, culture condition, and FACS approach. (H) Concatenate multiple flow samples to visualize CD84 expression in G- and M-MDSCs in one feature plot across all samples including PBMC control and treated. (I and J) Statistical analysis using unpaired t test (means ± SEM of n = 3), *P < 0.05.

To determine how generalizable these markers are, we next explored whether CD11b+Gr1+ cells express CD84 in another mouse model of breast cancer: an orthotopic transplant model using 4T1 breast cancer cells in Balb/c mice. Similar to the MMTV-PyMT model, we observed that CD84 expression was elevated in spleens (~ 21.46%) and primary tumors (~ 8.49%) of 4T1 tumor-bearing animals, whereas CD11b+Gr1+ cells from bone marrow and lungs showed no detectable CD84 expression (fig. S4, C and D). In addition, we used in vitro generation of MDSCs by treating myeloid cells with GM-CSF (7). We found that after GM-CSF treatment, the CD11b+Gr1+ population exhibited a significant increase in CD84-positive cells (~ 24.45%; fig. S5, A and B) and JAML-positive cells (~ 11.26%; fig. S5, A and C). Last, we used previously established protocols for in vitro generation of human MDSCs by isolating peripheral blood mononuclear cells (PBMCs) and treating them with GM-CSF and IL-6 (Fig. 3G and fig. S5D) (41). We observed a significant up-regulation of CD84 in samples from in vitro–generated CD11b+/CD14+ M-MDSCs and CD11b+/CD15+ G-MDSCs compared with control cells (Fig. 3, H to J). To additionally verify the successful generation of MDSCs in these in vitro cultures, we measured the expression of lectin-type oxidized LDL receptor 1 (LOX-1) in CD11b+/CD15+ neutrophils, which was previously used to define G-MDSCs (42, 43), and found that this G-MDSC marker is concomitant with CD84 up-regulated in human neutrophils. We also measured expression of human leukocyte antigen–DR (HLA-DR) CD11b+/CD14+ monocytes in culture, which has been shown to be down-regulated in M-MDSCs (12), and found that HLA-DR is indeed down-regulated in in vitro generated M-MDSCs (fig. S5, D and E). Together, these experiments established CD84 as a robust and generalizable cell surface marker for G- and M-MDSCs.

CD84hi MDSCs exhibit T cell suppression and increased ROS production

To functionally validate whether CD45+CD11b+Gr1+CD84hi cells inhibit immune cell activation, we performed cocultures with activated T cells as described above (Fig. 4A). CD45+CD11b+Gr1+CD84hi cells from spleen of tumor-bearing mice suppressed CD4 and CD8 T cell proliferation significantly, whereas CD45+CD11b+Gr1+CD84lo cells from tumor-bearing mice showed no significant inhibition (Fig. 4, B and C). In primary tumors, both CD84hi and CD84lo exhibited T cell–suppressive capacity (Fig. 4, D and E), which likely indicates that additional factors exist in the tumor microenvironment that contribute to maturation of MDSCs (44). In direct comparison, CD84hi showed more robust immunosuppression compared with CD84−/lo cells (Fig. 4, D and E), which is in line with the notion that CD84 expression is a robust indicator of MDSC capacity.

Fig. 4 CD11b+Gr1+CD84hi cells exhibit potent capacity for T cell suppression and increased ROS production.

(A) Overview of FACS approach using two different tissues (spleen and primary tumor) from WT and PyMT were subjected to T cell activation, ROS formation, and qPCR assays. (B and C) Splenic CD11b+Gr1+CD84hi cells from tumor-bearing mice suppress T cell proliferation. Histogram overlay (B) and quantitative bar charts (C) showing CD4/CD8 T cell proliferation measured by FACS in control samples with T cells only (black), T cells activated by CD3/CD28 (blue), activated T cells plus CD11b+Gr1+ cells from control spleens (SPLN) (orange), activated T cells plus CD11b+Gr1+CD84−/lo cells (purple), and activated T cells plus CD11b+Gr1+CD84hi (red) from spleen of tumor-bearing mice. (C) Statistical analysis using unpaired t test (means ± SEM of n = 3) *P < 0.05. (D and E) T cell suppression analysis using CD11b+Gr1+CD84hi and CD84−/lo cells isolated from primary tumors. Histogram overlay (D) and quantitative bar charts (E) showing CD4/CD8 T cell proliferation measured by FACS in control sample T cells (black), T cells activated by CD3/CD28 (blue), activated T cells plus CD11b+Gr1+CD84−/lo cells (purple), and activated T cells plus CD11b+Gr1+CD84hi (red) from tumor of tumor-bearing mice. (E) Statistical analysis using unpaired t test (means ± SEM of n = 3) *P < 0.05. (F and G) CD11b+Gr1+CD84hi cells from tumor-bearing mice show increased ROS formation compared with CD11b+Gr1+CD84−/lo; PMA-treated cells were used as positive control. ROS was measured by FACS using H2DCFDA. (G) Statistical analysis of ROS assay using unpaired t test (means ± SEM of n = 3) *P < 0.05. DCF, 2′,7′-dichlorofluorescein diacetate. ns, not significant.

Next, we measured ROS production as a hallmark for MDSC function in CD11b+Gr1+CD84−/lo and CD11b+Gr1+CD84hi cells from spleens and primary tumors of tumor-bearing and control mice. We used H2DCFDA staining for ROS in combination with flow cytometry and observed that CD11b+Gr1+CD84hi cells produced significantly higher amounts of ROS compared with CD11b+Gr1+ cells from control mice, whereas CD11b+Gr1+CD84−/lo showed no statistically different ROS production (Fig. 4, F and G). In addition, ROS production was significantly increased in CD11b+Gr1+JAMLhi cells from spleen and tumor of PyMT tumor-bearing mice compared with CD11b+Gr1+JAML−/lo and with the control CD11b+Gr1+ cells (fig. S5F). Last, we used qPCR to interrogate selected genes from our MDSC signature and found elevated expression of the complete panel of MDSC-related genes in CD11b+Gr1+CD84hi cells compared with CD11b+Gr1+CD84−/lo (fig. S6A and table S7). These findings indicate that MDSCs capable of T cell suppression and ROS production can be faithfully detected and enriched on the basis of high CD84 expression.

To further distinguish the expression of CD84 and JAML in the subsets of MDSCs (G- and M-MDSCs), we used flow cytometry to investigate CD45+CD11b+Ly6C+Ly6G (M-MDSCs) and CD45+CD11b+Ly6CloLy6G+ (G-MDSCs) (fig. S6B) in combination with CD84 or JAML staining. CD84 was lowly detected in M-MDSCs from spleen (fig. S6C), whereas M-MDSCs from primary tumors showed a substantial increase in CD84 expression (~28.13%) compared with cells from control mammary glands (fig. S6D). G-MDSCs, on the other hand, exhibit high CD84 expression in both primary tumors (~30.46%) and spleens (~17.33%) from tumor-bearing mice (fig. S6, C and D). JAML expression was up-regulated in both M- and G-MDSCs in spleen and tumor from tumor-bearing mice (fig. S6, E and F). To further validate that increased CD84 expression is associated with functional MDSC capacity in both MDSC subsets, Ly6C+CD84hi or Ly6G+CD84hi cells were isolated from in vitro–generated MDSCs and subjected to T cell suppression assays. Ly6C+CD84hi and Ly6G+CD84hi suppressed CD4 and CD8 T cell proliferation (fig. S6, G and H) to a similar degree as we previously found in the combined CD11b+/Gr1+ cell populations (fig. S2). These findings indicate that G- and M-MDSCs subsets express both CD84 and JAML and are capable of suppressing T cells.

G-MDSCs emerge through aberrant differentiation trajectory during cancer

To reconstruct the maturation process leading to MDSC generation in the spleen and to determine their differentiation state relative to normal progenitor and mature neutrophil populations, we next performed Monocle for unsupervised pseudotemporal ordering of our scRNAseq dataset (45). We focused on the Ly6g+ neutrophil subset [clusters C0, C2, C4, C5, C7, and C8 in Fig. 1 (B and C)] because in contrast to M-MDSCs, we recovered sufficient numbers of total neutrophils and G-MDSCs in this analysis to ensure an interpretable result. We first generated a new Seurat-based clustering of this neutrophil subset and then performed Monocle using this newly defined set of marker genes (fig. S7A and table S8). This resulted in a three-branch trajectory with five distinct cell states (Fig. 5A). To interpret this trajectory, we compared our results with recent work using scRNAseq to define the signatures of the naïve hematopoietic stem, progenitor, and differentiated cell states in the bone marrow of mice, which revealed that neutrophil progenitors are marked by the genes Elane, Mpo, and Prtn3, whereas mature neutrophils expressed elevated levels of Camp, Ltf, and Lcn2 (17). Integrating these markers together with the MDSC signature established in our work (Fig. 1F), we were able to annotate the five states. First identified were neutrophil progenitors (state 4; Elane-hi) that show increased proliferation (fig. S7, B and C) and form the beginning of pseudotime. These progenitors then bifurcate into mature neutrophils (state 3; Camp-hi) on the one branch and MDSCs (state 1; Cd84-hi) on the other branch, as illustrated by gene plots over pseudotime (Fig. 5B), suggesting that G-MDSCs emerge from neutrophil progenitors via an alternative maturation process.

Fig. 5 G-MDSCs emerge through aberrant differentiation trajectory during cancer.

(A) Neutrophil-specific Monocle analysis on subset of Ly6g+ neutrophil clusters resulted in branched trajectory with five distinct Monocle states (color code for each state is indicated), which are named on the basis of respective gene expression profile. (B) Pseudotime plot illustrating expression of selected marker genes over pseudotime with the branch ending in state 1, shown with the dotted line, and the branch ending with state 3, highlighted by the solid line. Neutrophil progenitors are characterized by high levels of Elane, Mpo, and Prtn3 (state 4), which bifurcate into mature neutrophils (state 3; Camp, Ltf, and Lcn2) on the one branch and MDSCs (state 1; e.g., CD84) on the other branch. (C) Early G-MDSC transition was marked by high expression of Asprv1, Plscr1, and Pirb. (D) Summary schematic indicates that G-MDSCs emerge from neutrophil progenitor cells via an aberrant form of neutrophil differentiation rather than from mature neutrophils that are reprogrammed into immunosuppressive cells.

Monocle detected two additional cell states (2 and 5) around the beginning of the MDSC branch: Whereas state 5 was characterized by high ribosomal gene counts indicative of a translationally active cell state, state 2 represents the earliest phase of MDSC differentiation and was marked by high expression of Asprv1 (46), Plscr1 (47), and Pirb (paired immunoglobulin-like receptor b) (Fig. 5C) (48). It has been reported that neutrophils promote chronic inflammation using Asprv1 (46), suggesting the aspartic protease encoded by Asprv1 may functionally contribute to the emergence of MDSCs in the spleen. Furthermore, Pirb has been reported to regulate the suppressive function and fate of MDSCs, indicating that Pirb is required for MDSC generation (48). Together, these findings indicate that MDSCs emerge from neutrophil progenitor cells via an aberrant form of neutrophil differentiation in the spleen rather than from mature neutrophils that are reprogrammed into immunosuppressive cells (Fig. 5D and table S9). In addition, our work identified an early, transitional MDSC state characterized by a number of genes showing elevated expression only around the branching point and during MDSC differentiation but not during the normal progenitor or mature neutrophil trajectory. This may suggest that the transitional MDSC state could be targeted to block differentiation into mature MDSCs while not affecting normal neutrophil maturation and function.

We also performed a monocyte-specific subset Monocle analysis in a comparable way as described for neutrophils above, which yielded a similar three-branched trajectory (fig. S7, D, E and table S10), suggesting that a similar alternative maturation process occurs during M-MDSC maturation. In contrast to our neutrophil-specific trajectory analyses, no transitional states emerged in our monocyte-specific analysis, which likely is due to the lower cell numbers in monocytes and decreased coverage of potentially transitioning states.


Understanding the cellular and molecular mechanisms through which the tumor microenvironment can suppress an active antitumor immune response will be critical to improve current approaches for cancer immunotherapy such as checkpoint inhibition [e.g., programmed death 1 (PD-1) and cytotoxic T lymphocyte–associated antigen 4) or chimeric antigen receptor T cell treatments (49). MDSCs represent such microenvironmental components that commonly expand in patients with cancer and promote advanced tumor progression and T cell suppression in various different cancers including breast cancer (5). Here, we generate a single-cell transcriptomics map of MDSC maturation during cancer to dissect the unique molecular features of MDSCs in breast tumor–bearing mice and to elucidate how these immunosuppressive cells differ from their normal myeloid counterparts. Using this resource, we establish an MDSC-specific gene signature that is largely shared between G- and M-MDSCs but strongly differs from their normal myeloid counterparts, reconstruct their differentiation trajectory from neutrophil progenitors through an aberrant path of differentiation, and identify MDSC-specific cell surface markers for detection and prospective isolation of MDSCs.

The MMTV-PyMT mouse model for breast cancer is one of the most widely used model system for studying breast cancer, which closely resembles human pathogenesis (14) and is known to induce significant expansion of MDSCs during tumor progression (2). We show here that spleen is the major organ site in which MDSCs can be robustly detected. To identify unique molecular features associated with MDSC function, we used scRNAseq as a powerful, unbiased method to reveal hidden variation on a single-cell level in a population of FACS-isolated CD45+CD11b+Gr1+ cells from the spleen of WT control and tumor-bearing PyMT mice. This dataset not only provides single cell–level depiction of monocyte/neutrophil heterogeneity in the spleen under steady-state conditions (WT mice) but also elucidated how MDSCs emerge as distinct clusters in both monocytes and neutrophil-like cells, which allowed us to establish MDSC-specific gene signatures. There was significant overlap between G- and M-MDSCs, suggesting that both monocytes and neutrophils acquire similar immunosuppressive features. The MDSC signature includes various genes associated with immune regulations such as Arg2 and Cd84, as well as chemokine receptors (e.g., Ccr2 and Cxcr2), indicating that MDSCs are responsive to neutrophil/MDSC-recruiting chemokines guiding their migration to active sites of inflammation such as the primary tumor or metastatic foci (Fig. 6).

Fig. 6 Proposed model of aberrant neutrophil differentiation in the spleen during cancer.

Myeloid cells differentiate in bone marrow from hematopoietic stem cells through common myeloid progenitors. Common granulocyte/monocyte progenitors expand in the bone marrow and migrate to spleen as a marginated pool, where they give rise to normal neutrophil maturation and, in cancer, aberrant neutrophil differentiation into G-MDSCs. Our findings indicate that MDSC-specific gene signature is largely shared between G- and M-MDSCs but differs from their normal myeloid counterparts. This MDSC signature includes numerous chemokine receptors, which likely guide their migration toward primary tumor or metastatic sites (indicated by arrows), where they may shield tumor cells from antitumor immunity. G-CSF, granulocyte colony-stimulating factor.

Our findings indicate that G-MDSCs may emerge from neutrophil progenitors through an aberrant differentiation trajectory giving rise to a cell state that is not present in normal conditions. Interrogating our observed cell states using pseudotemporal ordering and comparing these to recent work that defined the signatures of various hematopoietic stem and progenitor cell states in single-cell resolution (17) allowed us to reconstruct MDSC form an aberrant trajectory from neutrophil progenitor cells that occurs at the cost of normal differentiation into mature neutrophil granulocytes, which are less abundant in tumor-bearing mice (Fig. 5A). Further interrogation of the initial transitional cell state that branches off into G-MDSCs revealed several genes that strongly increase in expression in this transitional phase (Fig. 5C), which may suggest that therapeutically interfering with these gene products could block MDSC differentiation before they become functionally active.

Although our study provides some previously unappreciated insights into the biology of MDSCs, there are potential limitations to our analysis here. One limitation is associated with the inability to clearly distinguish between monocytes and macrophages in scRNAseq data. Hence, it is conceivable that our flow cytometry approach included macrophages from the spleen, which, however, did not readily emerge as distinct clusters in our computational analyses. Therefore, we are unable to clarify ongoing discussions in the field about the nature of M-MDSCs as macrophages or monocytes. Second, although our findings indicate that the spleen might play a critical role in the generation and maturation process of MDSCs, the significance of the spleen should be addressed in future studies, for example, by carrying out splenectomy experiments in murine cancer models. Last, our studies have relied solely on a single cancer, breast cancer. It remains to be determined whether the findings reported here are applicable to other cancer settings.

A major problem for researchers studying MDSCs is the relative dearth of MDSC-specific cell surface receptors for detection and prospective isolation for functional interrogation. Here, we identify CD84 and JAML cell surface receptors on MDSCs, which can be used in combination with CD11b/Gr1 staining to detect the presence of MDSCs in various organs of tumor-bearing mice or in humans in combination with CD11b/CD14 or CD15. CD84 is involved in cell-cell interactions and modulation of the activation and differentiation of a variety of immune cells (50) and functions as a homophilic adhesion molecule on B cells; monocytes; and, on a lower extent, T cells where it enhances interferon-γ secretion and activation (38, 39). CD84 can regulate PD-1/programmed death ligand 1 expression and function in chronic lymphocytic leukemia resulting in suppression of T cell responses and activity (51), suggesting that CD84 may allow MDSCs to directly regulate immune checkpoints in patients with breast cancer.


Study design

The aim of this study is to define the molecular features of cancer-associated MDSCs using unbiased single-cell transcriptomics. To this end, we analyzed the most commonly used mouse model for breast cancer (i.e., MMTV-PyMT), which was previously shown to lead to MDSC generation in tumor-bearing animals. We first characterized various tissues for the presence of MDSCs by isolating CD11b+/Gr1+ cells and subjecting them to T cell suppression assays in vivo, which showed robust presence of MDSC capacity in the spleen of tumor-bearing mice. We performed single-cell transcriptomics on a total of 14,646 FACS-isolated CD11b+/Gr1+ cells harvested from spleens of three tumor-bearing and three control mice. We then performed computational analyses for cell type clustering to identify cells of neutrophilic and monocytic origin and focused our analyses on those clusters that were predominantly present in tumor-bearing mice. We performed orthogonal validation of MDSC-related expression changes using qPCR in additional tumor-bearing and control mice. Subset analysis for cell surface markers allowed us to identify MDSC-specific cell surface molecules (CD84 and JAML), which were validated and integrated with known cell surface markers for granulocytic (Ly6g or CD15) and monocytic (Ly6c or CD14) cells by flow cytometry in samples from control and tumor-bearing mice, as well as in primary in vitro–generated MDSCs from mouse bone marrow or human peripheral blood samples. Because of the retrospective and observational nature, no prior sample size calculation was performed.


All mouse experiments were approved by the Institutional Animal Care and Use Committee of University of California, Irvine, in accordance with the guidelines of the National Institutes of Health. Transgenic PyMT (MMTV-PyMT) mice were purchased from the Jackson laboratory (stock no. 002374), and breedings were maintained on FVB/n and PyMT (MMTV-PyMT) backgrounds. Littermates from control and transgenic mice were used for all experiments.

Tissue collection and cell isolation

Bone marrow

After mouse dissection, bone marrow was flushed from mouse tibia and femurs using a 28-gauge needle and plastic syringe and then kept in Hanks’ balanced salt solution (HBSS) (Corning, 21-023-CV). Bone marrow cells were centrifuged at 500g at 4°C for 5 min. Cells were incubated for 5 min at room temperature (RT) in 2 ml of red blood cell (RBC) lysis buffer. Cells were quenched with 10 ml of HBSS containing 2% fetal bovine serum (FBS) (Omega Scientific, FB-12) and centrifuged at 500g at 4°C for 5 min. Cells were resuspended in 3 ml of FACS buffer [1× phosphate-buffered saline (PBS) and 3% FBS], and total remaining live bone marrow cells were counted using the automated cell counter Countess II (Thermo Fisher Scientific, AMQAX1000).


The spleen was pushed through a 70-μm cell strainer and washed with RPMI to create a cell suspension of splenocytes. Cells were centrifuged at 500g at 4°C for 5 min and then incubated for 5 min in 5 ml of RBC lysis buffer at RT. Cells were quenched with 10 ml of RPMI (Corning, 10-040-CV) with 5% FBS and centrifuged at 500g at 4°C for 5 min. Cells were resuspended in 3 ml of FACS buffer (1× PBS and 3% FBS), and total remaining live cells were counted by Countess II and processed for FACS.

Lung, tumor, and MFP

Tissue samples were harvested from mice and mechanically dissociated using a razor blade. Tissues were placed in Dulbecco’s modified Eagle’s medium/F-12 (Corning, MT10090CV) complete medium containing insulin (5 μg/ml; Sigma-Aldrich, 11376497001), penicillin-streptomycin (50 μg/ml; HyClone, SV30010), and collagenase type IV (0.1 mg/ml; Sigma-Aldrich, C5138) and were digested at 37°C on a shaker for 45 min. Samples were centrifuged at 500g for 5 min at RT. Cells were resuspended in HBSS and centrifuged at 500g for 5 min at RT. Cells were resuspended in 25 μl of deoxyribonuclease I (Sigma-Aldrich, D4263) for 5 min at RT; then, 2 ml of 0.05% trypsin (Corning, 25-052-CI) was added, and samples were incubated at 37°C for 10 min. Samples were centrifuged at 500g for 5 min at RT and then resuspended in 5 ml of HBSS with 2% FBS. The cell suspension was filtered through a 70-μm cell strainer (Fisher Scientific, 22363548) and incubated for 5 min at RT in 3 ml of RBC lysis buffer. Cells were quenched with 10 ml of HBSS with 2% FBS and centrifuged at 500g for 5 min at RT. Cells were then resuspended in RPMI with 10% FBS, and total remaining live cells were counted by Countess II and processed for FACS.

Peripheral blood

Blood was collected using a 20-gauge needle and syringe from the chest cavity after the right atrium and left ventricle were punctured. Mice were perfused with 15 ml of 10 mM EDTA in 1× PBS, and blood was collected. Blood cells were centrifuged at 500g at 4°C for 5 min. Cells were resuspended in 5 ml of RBC lysis buffer and incubated at RT for 5 min. Cells were quenched in 5 ml of RPMI with 3% FBS and centrifuged at 500g at 4°C for 5 min. Cells were then resuspended in 3 ml of FACS buffer (1× PBS and 3% FBS), and total remaining live cells were counted by Countess II and processed for FACS.


Brain tissue was dissociated into a single-cell suspension using the Adult Brain Dissociation Kit (Miltenyi Biotec, 130-107-677), according to the manufacturer’s protocol, and a gentleMACS Octo Dissociator with Heaters (Miltenyi Biotec, 130-096-427). Briefly, the brain was dissected from the cranium; the meninges were removed, and the brain was chopped into 8 to 10 pieces. The chunks were transferred into a gentleMACS C Tube (Miltenyi Biotec, 130-093-237) containing enzymes A and P and then placed onto the gentleMACS. The brain was digested using the 37C-ADBK protocol on the instrument. After a 30-min heated digestion, the brain slurry was strained through a 70-μm nylon strainer and washed with 10 ml of ice-cold 1× PBS with 2% bovine serum albumin (BSA) (Sigma-Aldrich). The suspension was centrifuged at 300g for 10 min at 4°C and then mixed with 4 ml of 1× Debris Removal Solution provided in the kit and centrifuged for 10 min at 3000g at 4°C or RT. The myelin layer was removed, and the cells were washed with Dulbecco’s PBS (DPBS) and centrifuged for 10 min at 1000g. RBCs were lysed for 3 min on ice with 1 ml of 1× RBC lysis buffer provided in the kit. After quenching in 2 ml of DPBS with 2% BSA, the cells were pelleted at 500g for 3 min at 4°C, and total remaining live cells were counted by Countess II and processed for FACS.

In vitro generation of MDSCs


Bone marrow cells were collected as described above, and then, cells were cultured with RPMI and 10% FBS and treated with recombinant murine GM-CSF (20 ng/ml; PeproTech, 315-03) every other day. The cells were collected after 4 days in culture, and total remaining live cells were counted by Countess II and processed for FACS.


We followed established protocols for in vitro generation of human MDSCs (41). Briefly, human blood was incubated with 3% dextran (Sigma-Aldrich, 31392-10G) for 18 min, and supernatants were collected and followed by differential density gradient separation (Ficoll-Paque Plus; Neta Scientific, GHC-17-1440-02). Samples were centrifuged at 500 relative centrifugal force for 30 min at 20°C. PBMCs including granulocytes were collected and incubated with recombinant human cytokines (10 ng/ml) (GM-CSF, PeproTech, 300-03 and IL-6, PeproTech, 200-06) or without in RPMI contain 10% FBS. Cells were treated with these cytokines every day, and on day 4, cells were collected from cultures and total remaining live cells were counted by Countess II. Cells were blocked with Human TruStain FcX (Fc Receptor Blocking Solution) (BioLegend, 422301) on ice for at least 10 min. Cells were then centrifuged at 500g for 5 min at 4°C and washed once with FACS buffer (1× PBS with 3% FBS). Cells were incubated for 30 min at 4°C with preconjugated fluorescence-labeled antibodies with the following combinations: CD45 (2D1) [Thermo Fisher Scientific, 48945942 (eFluor)], CD11b (ICRF44) [BioLegend, 301335, (Brilliant Violet 650)], CD14 (61D3) [Thermo Fisher Scientific, 45014942 (PerCp-Cyanine5.5)], CD15 (HI98) [BioLegend, 301923 allophycocyanin (APC) or 301923 (PE-cy7)], CD84 (CD84.121) [BioLegend, 326007 phycoerythrin (PE)], LOX-1 (15C4) [BioLegend, 358605 (APC)], and HLA-DR (L243) [BioLegend, 307645 (Brilliant Violet 510)]. SYTOX Green dye (Thermo Fisher Scientific, S34860) was added to stained cells to assay for viability. Cells were analyzed using BD FACSAria Fusion.

Fluorescence-activated cell sorting

Tissue samples were harvested from mice and mechanically dissociated to generate single-cell suspensions as described above. Cells were blocked with anti-mouse FcγR (CD16/CD32) (BioLegend, 101301) on ice for at least 10 min. Cells were then centrifuged at 500g for 5 min at 4°C and washed once with FACS buffer (1× PBS with 3%FBS). Cells were incubated for 30 min at 4°C with preconjugated fluorescence-labeled antibodies with the following combinations: CD45 (30-F11) [BioLegend, 103112 (APC) or 103115 (APC-cy7)], CD11b (M1/70) [BioLegend, 101206 (fluorescein isothiocyanate) or 101212 (APC)], Gr1 (Rb6-8C5) [BioLegend, 101206 (PE) or 108439 (BV605)], Ly6C (HK1.4) [BioLegend, 128017 (PE/Cy7)], Ly6G (1A8) [BioLegend, 127613 (APC)], CD84 (mCD84.7) [BioLegend, 122805 (PE)], and JAML (4e10) [BioLegend, 128503 (PE)]. SYTOX Blue dye (Life Technologies, S34857) was added to stained cells to assay for viability. Cells sorted using BD FACSAria Fusion and desired populations were isolated for different experiments. Human PBMCs were prepared as described above; cells were blocked with Human TruStain FcX (BioLegend, 422301) on ice for 10 min. Then, cells were centrifuged at 500g for 5 min at 4°C and washed once with FACS buffer (1× PBS with 3%FBS). Cells were incubated for 30 min at 4°C with the following anti-human, preconjugated fluorescence-labeled antibodies: CD45 (eFluor 450) (Thermo Fisher Scientific, 48-9459-42), CD11b (BV650) (BioLegend, 101206), CD14 (PerCP-Cy5.5) (Thermo Fisher Scientific, 45-0149-42), CD15 (APC) (BioLegend, 301907), and CD84 (PE) (BioLegend, 326007). SYTOX Green (Life Technologies, S34860) was added to the stained cells to assay for viability. Cells were analyzed using BD FACSAria Fusion.

T cell suppression assay

Spleens were dissected, filtered into a single-cell suspension, and depleted of RBCs using tris–acetic acid chloride. T cells were isolated from the spleen using the EasySep Mouse T Cell Isolation Kit (STEMCELL Technologies, 19851) according to the manufacturer’s instructions. Isolated T cells were washed once with PBS and resuspended at 15 × 106 per ml in staining buffer (0.01% BSA in PBS). T cells were stained with proliferation dye eFluor 670 (Thermo Fisher Scientific, 65-0840-85) using 5 mM dye per 1 × 107 cells and incubated in a 37°C water bath for 10 min. Last, T cells were washed and resuspended at 1 × 106 per ml in RPMI 1640 with Hepes plus l-glutamine (Gibco, 22400-105) complete medium containing 10% FBS (Atlanta Biologicals, S11150), 1× nonessential amino acids (Gibco, 11146-050), penicillin (100 U/ml)–streptomycin (100 μg/ml) (Gibco, 15140163), 1 mM sodium pyruvate (Gibco, 11360-070), and 55 μM β-mercaptoethanol (Gibco, 21985-023). eFluor 670–labeled T cells were plated (50 × 103 per well) in a U-bottom 96-well plate (VWR, 10062-902) and activated with plate bound anti-Armenian hamster immunoglobulin G (30 μg/ml; Jackson ImmunoResearch, 127-005-099), with CD3 (0.5 μg/ml; Tonbo Biosciences, 70-0031) and CD28 (1 μg/ml; Tonbo Biosciences, 70-0281). Sorted CD11b+ Gr1+ cells from PyMT or WT mice were added to T cells in 1:1 ratio (50 × 103 T cells and 50 × 103 CD11b+ Gr1+ cells). After 4 days of culture, cells were collected, blocked with anti-mouse CD16/32 (BioLegend, 101302), and stained with Zombie live/dead dye (BioLegend, 423105) and the following fluorescence-conjugated antibodies: CD4 (BioLegend, 100512; clone RM4-5) and CD8 (BioLegend, 100709; clone 53-6.7). Single-stained samples and FMO controls were used to establish photomultiplier tube voltages, gating, and compensation parameters. Cells were processed using the BD LSR II or BD LSRFortessa X-20 flow cytometer and analyzed using FlowJo software version 10.0.7 (Tree Star Inc.).

ROS production assay

Cells were harvested from respective tissues and processed to single-cell suspensions as described above. Cells were stained with CD45, CD11b, Gr1, and CD84 or JAML antibodies as described above. After staining, cells were resuspended in FACS buffer, and 10 mM H2DCFDA (Sigma-Aldrich, D6883) was added and incubated for 30 min at RT. Positive control cells were treated with 100 nM phorbol myristate acetate (PMA) (Sigma-Aldrich, P1585-1MG). Cells were then processed on the BD FACSAria Fusion and analyzed using FlowJo software version 10.0.7 (Tree Star Inc.).

Quantitative real-time PCR

CD11b+Gr1+ cells, CD11b+Gr1+CD84hi cells, and CD11b+Gr1+CD84–/lo cells were sorted by FACS, and RNA were extracted using Quick-RNA Microprep Kit (Zymo Research, R1050) following the manufacturer’s instructions. RNA concentration and purity were measured with a Pearl nanospectrophotometer (Implen). Quantitative real-time PCR was conducted using PowerUp SYBR Green Master Mix (Thermo Fisher Scientific, A25742), and primer sequences were found in Harvard PrimerBank and obtained from Integrated DNA Technologies (table S10). Gene expression was normalized to glyceraldehyde phosphate dehydrogenase housekeeping gene. For relative gene expression, 2−ΔΔCt values were used, and for statistical analysis, ΔCt was used. The statistical significance of differences between groups was determined by unpaired t test using Prism 6 (GraphPad Software Inc.).

Single-cell RNA sequencing

FACS-isolated CD11b+/Gr1+ cells from the spleens of control WT (five mice pooled), and tumor-bearing PyMT mice (three mice pooled) were washed once in PBS with 0.04% BSA, resuspended to a concentration of about 1000 cell/μl, and loaded onto the 10x Genomics Chromium platform for droplet-enabled scRNAseq according to the manufacturer’s instructions. Library generation was performed following the Chromium Single Cell 3′ Reagents Kits version 2 user guide (CG00052 Rev B). Each library was sequenced on the Illumina HiSeq 4000 platform to achieve an average of 48,488 reads per cell. Alignment of 3′ end counting libraries from scRNAseq analyses was completed using 10x Genomics Cell Ranger 2.1.0. Each library was aligned to an indexed mm10 genome using Cell Ranger Count. The “Cell Ranger Aggr” function was used to normalize the number of confidently mapped reads per cell across the two libraries.

Cluster identification of mouse data using Seurat

The Seurat pipeline (version 2.3.1) was used for cluster identification in scRNAseq datasets. Data were read into R (version 3.5.0) as a count matrix, scaled by a size factor of 10,000, and log-transformed. We set gene expression cutoffs at a minimum of 500 and a maximum cutoff of 5000 genes per cell for each dataset. In addition, cells with a percentage of total reads that aligned to the mitochondrial genome (referred to as percent mito) greater than 8% were removed. Using Seurat’s CCA, cells from WT and PyMT mice were integrated together into a single analysis. For tSNE (t-distributed stochastic neighbor embedding) projection and clustering analysis, we used the first 20 principal components. Specific markers for each cluster identified by Seurat were determined using the “FindAllMarkers” function. For gene scoring analysis, we compared gene signatures and pathways in subpopulations using Seurat’s “AddModuleScore” function. For cell type subset analyses (monocytes and neutrophils), clusters with high expression of cell type markers (Csf1r and Ly6g, respectively) that were determined to not be of the monocyte or neutrophil cell type in their analyses were subset out, and standard Seurat workflow was applied on each. In the case of the neutrophil-specific analysis, a population of cells that grouped together and expressed a set of markers associated with neutrophil progenitors was manually labeled and treated as a distinct cluster for analysis (17).

Human breast cancer–associated immune cell analysis using Seurat

Data from Azizi et al. (36) were downloaded with associated metadata and processed in R version 3.6.0 and Seurat version 3.0.2. No trimming of cells or features was performed to maintain consistency with the published analysis, and the full dataset was labeled by the cell types denoted in the manuscript. All cells in the analysis were scored using the conserved MDSC signature (table S5) using Seurat’s “AddModuleScore” function. Neutrophils and monocytes were specifically subset out to their own analyses and new dimensionality reduction via uniform manifold approximation and projection (UMAP), and clustering via the “FindClusters” function was generated, respectively.

Reconstruction of differentiation trajectories using Monocle

Using the R package Monocle (version 2.8.0), differentiation hierarchies within the neutrophil and monocyte compartments were reconstructed. All cells in the analysis were scored using Seurat’s “AddModuleScore” function using a cell cycle gene list published by Giladi et al. (17). For the neutrophil-specific analysis, starting with all cells from the WT and PyMT combined analysis, neutrophils were specifically subset out. Once subset, contaminating cell types were removed, and the cells were reclustered to explore additional heterogeneity within the neutrophils compartment. Using marker genes of these clusters, the top 20 unique genes per cluster were used to order cells along a pseudotemporal trajectory. Because cells that expressed markers associated with neutrophil progenitors (17) localized to a single branch, that branch was chosen as the start of pseudotime for further analysis. For the monocyte-specific analysis, cells corresponding to cluster 1 were subset out, and contaminating cell types were removed. Subsequent clustering was performed as described above. Using genes differentially expressed by cluster via Monocle’s “differentialGeneTest” function, subsets to those that had a Q value less than 1 × 10−10 and expressed in greater than 20 cells were used for ordering. The start of pseudotime was chosen to be the branch containing most cells from the WT library.

Statistical analysis

For statistics pertaining to the scRNAseq data analysis, P values from the function “FindAllMarkers” in Seurat, and when comparing scoring as a result of the “AddModuleScore” function, we used an unpaired Wilcoxon rank sum test with Bonferroni correction. For the differential expression test through Monocle’s “differentialGeneTest” function, significance and Q value calculation are derived from the VGAM package implementation of the “vglm” and its test for control of the false discovery rate (FDR). Data from FACS and qPCR are expressed as means ± SEM or SD, and statistical significance was determined using an unpaired t test using the Prism 6 software (GraphPad Software Inc.). P values were considered to be significant when P < 0.05.


Fig. S1. Expansion of CD11b+Gr1+ cells during tumor progression in PyMT mice.

Fig. S2. MDSCs emerge predominantly in spleen of tumor-bearing mice.

Fig. S3. Sample labels and marker genes from scRNAseq analysis.

Fig. S4. CD84 is a generalizable MDSC marker in different breast cancer models.

Fig. S5. CD84 and JAML are up-regulated by in vitro–generated mouse and human MDSCs.

Fig. S6. Characterization and validation of myeloid cell subsets for CD84 and JAML expression.

Fig. S7. Reconstruction of MDSC differentiation trajectory in neutrophils and monocytes.

Table S1. Marker genes from combined Seurat analysis.

Table S2. Marker genes from Seurat analysis of monocytes only.

Table S3. Gene signature from G-MDSCs-versus-neutrophils comparison.

Table S4. Gene signature from M-MDSCs-versus-monocytes comparison.

Table S5. Combined MDSC signature gene list.

Table S6. GO terms (Biological Process 2018) MDSC gene signature.

Table S7. qPCR primer sequences.

Table S8. Marker genes from Seurat analysis of neutrophils only.

Table S9. Neutrophil-specific Monocle state marker genes.

Table S10. Monocyte-specific Monocle state marker genes.


Acknowledgments: We thank D. Lawson for constructive feedback on experimental design and analysis. We thank R. T. Davis and L. M. Hosohama for careful review and feedback of the manuscript. We thank J. Atwood for assistance with cell sorting in the flow cytometry core at the University of California, Irvine. Funding: This study was supported by funding from NIH/NCI (1R01CA234496 and 4R00CA181490 to K.K.); the American Cancer Society (132551-RSG-18-194-01-DDC to K.K.); the University of Hail, Hail, Saudi Arabia for the Ph.D. Fellowship (to H.A.); the Canadian Institutes of Health Research (CIHR) Postdoctoral Fellowship (to D.M.); and pilot funds from the U54 Center for Cancer Systems Biology (CA217378). Author contributions: H.A., L.L.M., D.M., Q.N., K.N., J.A.R., G.H., K.E., L.T., and A.S. performed research. K.K. supervised research. C.W. developed reagents and analytic tools needed for the study. N.P., H.A., and K.K. performed bioinformatic analyses. H.A. and K.K. wrote the paper, and all authors discussed the results and provided comments and feedback. Competing interests: The authors declare that they have no competing financial or nonfinancial interests. Data and materials availability: Sequencing datasets have been deposited in the GEO database under the accession number GSE139125.

Stay Connected to Science Immunology

Navigate This Article