Role of radiation-induced circulating myeloid-derived suppressor cells on systemic lymphopenia after chemoradiotherapy for glioblastoma

Abstract

Severe and prolonged lymphopenia frequently occurs in glioblastoma patients after standard chemoradiotherapy and has been associated with significantly worse survival, but its biological mechanism is not well understood. To address this we performed a correlative study designed to prospectively collect and evaluate peripheral blood of glioblastoma patients treated with chemoradiotherapy using genomic and immune monitoring technologies. The RNA sequencing analysis of the peripheral blood mononuclear cells (PBMC) showed an elevated level of myeloid-derived suppressor cells (MDSC) regulatory genes in the lymphopenic patients (LP) when compared to non-lymphopenic patients (NLP) after chemoradiotherapy. Additional analysis using flow cytometry and single-cell RNA sequencing further confirmed increased number of circulating MDSC in LP when compared to NLP after chemoradiotherapy. Preclinical murine models were also established to recapitulate this phenomenon and demonstrated a causal relationship between radiation-induced MDSC and systemic lymphopenia using transfusion and depletion experiments. Pharmacological inhibition of MDSC using arginase-1 inhibitor (CB1158) or phosphodiesterase-5 inhibitor (tadalafil) during RT successfully abrogated radiation-induced lymphopenia and improved survival in the preclinical models.

One Sentence Summary:

Role of myeloid-derived suppressor cells in lymphopenia in GBM

INTRODUCTION

Glioblastoma (GBM) is the most common malignant primary brain tumor and one of the most devastating cancers, with over 11,000 cases per year (1). Despite the optimal multi-modality therapy of surgery, radiation therapy (RT), and temozolomide (TMZ), the median overall survival (OS) for GBM remains dismal at approximately 14–16 months in recent clinical trials (2–5). Progress in immunotherapy has generated excitement in developing novel approaches to treat this devastating disease, but recent randomized studies with PD-1 inhibitors or tumor vaccines for GBM have yielded negative results (6, 7). The disappointing results have been attributed to the complex immune-suppressive properties of GBM to evade antitumor immunity, and one of the major mechanisms may be systemic lymphopenia (8, 9). Systemic lymphopenia (total lymphocyte count below 500 per μL) can occur after RT and TMZ in approximately 30-40% of GBM patients, which is independently associated with poor OS (10–12). Our research group has previously shown irradiated brain volume, concurrent TMZ, and overall corticosteroid exposure during RT are all contributing factors to the lymphopenia phenomenon (13–15), and others have also demonstrated that GBM can induce T cell sequestration in the bone marrow to exacerbate lymphopenia (16). Systemic lymphopenia after RT is commonly observed across multiple tumor types, regardless of the inclusion of lymphatic tissues or bone marrow in the radiation field. Notably, irradiation of the tumors while avoiding bone marrow or lymphatic tissue elicits approximately 60% reduction of total lymphocyte counts in the peripheral blood (17–20).

The biological mechanisms of lymphopenia in GBM and its association with poor survival are still not clearly understood. To delineate the mechanisms underlying radiation-induced lymphopenia in GBM, we conducted a correlative study to prospectively analyze patient samples using multiple genomic and immune monitoring assays. We developed a preclinical GBM mouse model to evaluate the hypothesis derived from the clinical observations.

RESULTS

Lymphopenia after chemoradiotherapy is associated with increased number of myeloid-derived suppressor cells (MDSC)

To delineate the underlying mechanisms of lymphopenia in GBM, we conducted a prospective correlative clinical study using multiple genomic and immune monitoring technologies such as RNA sequencing and flow cytometry. We evaluated 20 GBM patients receiving standard RT and concurrent TMZ for six weeks (Fig. 1A). We collected peripheral blood from GBM patients before (baseline) and during chemoradiotherapy (week 2 and week 6). Of the 20 patients, 10 developed systemic lymphopenia (Lymphopenic patients; LP) during or within 4 weeks of completing chemoradiotherapy, and 10 did not develop lymphopenia (Non-lymphopenic patients; NLP). The consort flow diagram is given in Fig. S1A and the baseline clinical characteristics are summarized in supplemental Table S1. With a relatively small sample size, LP demonstrated non-significantly worse progression-free survival and overall survival than NLP (Fig. S1B & C). As seen in Fig. 1B, the lymphocyte counts were not significantly different between LP and NLP at baseline in whole blood. However, the lymphocyte counts of LP were significantly lower than NLP at week 2 and week 6 (p=0.005 & p<0.001). The neutrophil counts of LP were significantly higher than NLP at baseline (p=0.02) and remained steadily elevated without significant increase at week 2 and 6, while the monocyte counts were not significantly different between LP and NLP at all-time points (Fig. S1D & E). Bulk RNA sequencing (RNA-seq) of peripheral blood mononuclear cells (PBMC) was performed on 15 patients (7 NLP and 8 LP). Fig. 1C shows selected gene signatures related to T cells, monocytes, myeloid-derived suppressor cells (MDSC), and MHC class II. In humans, the MDSC is generally defined as the cells that express CD11b and the common myeloid marker CD33 but lack the expression of markers of mature myeloid and lymphoid cells and the MHC class II molecule HLA-DR (21, 22). LP had elevated expression of MDSC regulatory genes at baseline and increased further after RT when compared to NLP.

Figure 1. T cell lymphopenia after radiation therapy is associated with increased number of MDSC.

A. Illustration depicting the treatment strategy, blood sample collection at different time points, and PBMC isolation for RNA sequencing, scRNASeq, and flowcytometric analysis. B. Peripheral lymphocyte level in non-lymphopenic (NLP) and lymphopenic (LP) patients at baseline (week 0), 2 weeks, and 6 weeks. C. Heat map showing differential expression of MDSC and T cells specific genes in non-lymphopenic (NLP, n=6) and lymphopenic (LP, n=8) at baseline, 2 weeks, and 6 weeks. t-SNE representation of single-cell gene expression demonstrating the distribution of different immune cell types (D), MDSC subtypes (E), and expression and number of M-MDSC (F) in LP (n=1) compared to NLP (n=1). The size and color of each circle represents the proportion of cells within the group expressing each transcript (F). Bar graph indicating the number or m-MDSC RNA expression in LP and NLP (G). Flow cytometric analysis of NLP (n=8) and LP (n=8) PBMC revealed the changes in G-MDSC (H), and M-MDSC (I) at baseline and week 6. J Flow cytometric gating strategy demonstrating for G-MDSC and M-MDSC on CD15⁺CD11b⁺Lox-1⁺ and CD14⁺CD11b⁺HLADR⁻ cells at week 6 between NLP and LP. Data are shown as mean ± S.D. P-values were determined by an unpaired t-test for B and two-way ANOVA with a post hoc Tukey’s test for H and I.

To further explore this, single-cell RNA sequencing (scRNA-seq) was performed for paired PBMC samples at baseline and week 6 for two selected phenotypical patients (Fig. S1G). Uniform manifold approximation and projection (UMAP) analysis was used to identify the various immune subsets including MDSC subtypes of LP and NLP at baseline and week 6 (Fig. 1D). The integrated distribution of immune cells including monocytes and T cells are shown in Fig. S1G. We found a reduction in T cells (CD4 and CD8) and an increase in B cells, monocytes, reticulocytes, and platelets in the LP when compared to the NLP at week 6 (Fig. 1D). We further evaluated the distribution of different monocyte subsets and MDSC as shown in Fig. 1E & F. M-MDSCs formed a cluster of the monocytic population, which was characterized by high expression of CD14 and S100A8/9, low expression of HLA-DR genes, and other previously reported MDSC markers. scRNA-seq data have confirmed that the LP had higher CD14⁺ monocytes and M-MDSC count at baseline than the NLP, and the difference is further exacerbated at week 6 of RT (Fig. 1F & G). To confirm the bulk RNA-seq and scRNA-seq findings, we performed flow cytometry to evaluate MDSC in the PBMC from 8 LP vs 8 NLP. The LP had significantly increased number of G-MDSC (CD15⁺CD11b⁺Lox-1⁺) and M-MDSC (CD14⁺CD11b⁺HLADR⁻) at week 6 when compared to NLP (p=0.04 & p=0.001), even though the number of baseline MDSC was not significantly different between the two cohorts (Fig. 1H, I & J). Together these data suggest that the expansion of circulating MDSC is associated with lymphopenia in GBM patients after chemoradiotherapy.

Lymphopenia after chemoradiotherapy is associated with decreased number of T cells

We also evaluated the impact of chemoradiotherapy on different subtypes of T cells using our bulk RNAseq and scRNA-seq data sets. We found decreases in gene signatures related to T cells, NK cells, and effector cells in the LP after RT at week 6 (Fig. 2A). scRNA-seq also demonstrated a dramatic reduction of different CD4 and CD8 T cells lineages including; naïve, terminally effector, central memory, and effector memory T cells in the phenotypical LP after RT, which was not observed in the phenotypical NLP (Fig. 2B & C). Flow cytometry analysis confirmed that the LP developed a non-significant reduction of CD4 and CD8 T cells after RT, which was not observed among the NLP (Fig. 2D, E & F). A similar trend was also observed for other T cell subsets, such as effector, naïve, central memory, effector memory CD4 and CD8 T cells (Fig. S2 A, B & C). These data confirm that lymphopenia in GBM after chemoradiotherapy affects all subtypes of T cells.

Figure 2. RNA sequencing and single-cell RNA seq analysis showing changes in T cells.

Heat map showing differential expression MDSC, T cells with other myeloid and lymphoid cells specific genes in NLP and LP at baseline and week 6 (A). t-SNE representation of single-cell gene expression of the distribution of different T cell subtypes with markers expression (B). T_N - naive, T_CM - central memory, T_EM - effector memory, T_EMRA - terminal effector, T_REG - regulatory T cells. The size and color of each circle represents the proportion of cells within the group expressing each transcript (C). Flow cytometric analysis of NLP and LP PBMC revealed the changes in CD4 (D), and CD8 T cells (E). Flow cytometric gating strategy demonstrating for CD4 and CD8 T cells on CD3 cells at week 6 between NLP and LP (F). Data are shown as mean ± S.D. P-values were determined by two-way ANOVA with post hoc Tukey’s test.

Orthotopic GBM mouse model to study immune suppression

We subsequently developed a preclinical mouse model to evaluate the association between cranial irradiation, circulating MDSC, and lymphopenia. A syngeneic GBM mouse model was established using immunocompetent mice (C57BL/6) orthotopically implanted with a GBM cell line (GL261). The mice were then treated with cranial irradiation with various RT fractionation: 2Gy/day x 5 fractions, 4Gy x 5, and 10Gy x 3 (Fig. S2A). After clinically relevant fractionation of 2Gy x 5 (RT, Fig. 3A), tumor-bearing mice showed a significant reduction in CD4 and CD8 T cells by flow cytometry on day 14 and day 21 (Fig. 3B: p=0.006 & p=0.03, C: p=0.002 & p=0.0001). However, only a slight reduction in CD4 and CD8 T cells was observed in the irradiated control mice without implanted tumors (Fig 3B & C). Correspondingly, tumor-bearing mice (but not tumor-free mice) exhibited a significant increase of G-MDSC (CD11b⁺ Ly6G⁺Ly6C⁻) and M-MDSC (CD11b⁺ Ly6C⁺Ly6G⁻) on day 14 and day 21 after irradiation (Fig. 3D: p<0.0001 & p<0.0001, E: p<0.0001 & p<0.0001,). The other irradiation regimens (4Gy x 5 and 10Gy x 3) also showed similar effects on T cells and MDSC (Fig. S2B & C), so the clinically relevant fractionation of 2Gy x 5 was selected for subsequent experiments. As expected, the tumor-bearing mice treated with RT had smaller tumors when compared to tumor-bearing mice without RT (Fig. 3F & G). Since GBM patients are treated with RT and concurrent TMZ, we also performed the experiment by treating the mice with RT and TMZ (Fig S4 A–E). The addition of TMZ in combination with RT did not significantly alter T cells, G-MDSC, and M-MDSC counts when compared to mice treated with RT alone (Fig S4 D & E). Hence, for our subsequent preclinical experiments, we treated the mice with RT alone. To test if irradiating normal tissue away from the tumor would induce a similar effect on MDSC, we irradiated the thorax of the orthotopic GBM-bearing mice with 2Gy x 5, so the brain containing GBM was not irradiated (Fig. S5 A). Thoracic radiation had a minimal effect on T cells and MDSC in the tumor-bearing mice (Fig. S5 B–E). These data indicate that our murine model can replicate the radiation-induced lymphopenia and MDSC phenomenon observed in GBM patients, and such phenomenon is dependent on the context of the tumor and the irradiation of the tumor.

Figure 3. Orthotopic GBM mouse model to study immune suppression.

A. To study the mechanism of radiation-induced lymphopenia, the C57BL6 mice were injected with GBM tumor cells (GL261), and treated with five fractions of 2 Gy cranial irradiation (RT). The PBMC and MDSC were analyzed on various days as indicated in the following groups; no tumor, no tumor+RT, tumor, and tumor+RT. B-E. The absolute counts of peripheral blood CD4, CD8, G-MDSC, and M-MDSC cells. F & G. The quantitative radiance value and their representative images over time. Data are shown as mean ± S.D. (n=5 mice/group). P-values were determined by two-way ANOVA with post hoc Tukey’s test. The indicated p-values are between the tumor and tumor+RT cohorts at 14 and 21 days post tumor implantation.

RT enhances myelopoiesis in the GBM tumor-bearing mice

To test the hypothesis that the increased circulating MDSC in blood after irradiation may be due to radiation-induced aberrant production of myeloid cells in the bone marrow or spleen, the bone marrow and spleen tissues from tumor-free control mice and tumor-bearing mice were harvested on day 14 after either RT (2Gy x 5) or sham treatment (Fig. 4A). Mice implanted with GBM tumors and treated with RT had enhanced total myeloid cells (CD11b⁺), G-MDSC, and M-MDSC in the bone marrow (Fig. 4B & D) and spleen (Fig. S6A, B C & D) as compared to the other groups. Furthermore, mice implanted with tumor and treated with RT had significantly higher granulocyte-monocyte progenitor (GMP: p=0.001), common myeloid progenitor (CMP: p=0.03), and multipotent progenitor (MPP: p=0.05) cells (but not short-term hematopoietic stem cells or ST-HSC) when compared to the other groups (Fig. 4E, F G, H & I). Our data confirm that irradiation and tumor have a synergistic effect on the aberrant myelopoiesis in the bone marrow and spleen to produce MDSC.

Figure 4. Effect of RT on the bone marrow of GBM mice.

A. To investigate the source of MDSC generation, the C57BL6 mice were injected with GBM tumor cells (GL261) and treated with five fractions of 2 Gy cranial irradiation (RT). The bone marrow HSPC and MDSC were analyzed on day 14 in the following groups; no tumor, no tumor with radiation (no tumor+RT), tumor, and tumor+RT. B-D. The changes in bone marrow total myeloid cells (CD11b⁺), G-MDSC, and M-MDSC. E-H. The absolute cell counts of ST-HSC, MPP, CMP, and GMP. I The flow cytometric gating strategy of GMP, CMP, and MEP on Lineage⁻c-kit⁺ cells (LK⁺; top panel) and the following MPP, ST-HSC, and LT-HSC in Lineage⁻Sca-1⁺c-Kit⁺ cells (LSK⁺; bottom panel). Data are shown as mean ± S.D. (n=5 mice/group). P-values were determined by one-way ANOVA with post hoc Tukey’s test.

Functional assay of radiation-induced MDSC

To evaluate if the radiation-induced MDSC can functionally inhibit peripheral T cell proliferation and activation, we used the in-vitro functional assay described by Burger et. al. (23). Human CD33⁺ myeloid cells from the previously described LP and NLP GBM groups were isolated and incubated with T cells from healthy donors (human) in the presence of a T cell stimulus, CD3/CD28 beads (Fig. 5A). The T cells were then analyzed for T cell proliferation by measuring Ki67⁺ cells and T cell activation by measuring TCR zeta. The CD33⁺ myeloid cells from LP more efficiently suppressed T cell proliferation (Fig. 5B: p=0.02) and T cell activation (Fig. 5C: p=0.02) when compared to CD33⁺ myeloid cells from NLP. To test if radiation-induced circulating MDSC can directly induce systemic lymphopenia, we isolated G-MDSC (CD11b⁺Ly6G⁺Ly6C⁻) from the tumor-bearing and tumor-free mice treated with RT and infused it into a naïve unirradiated tumor-free recipient mice (Fig. 5D). The G-MDSC from the irradiated tumor-bearing mice significantly reduced CD4 and CD8 T cells in the naïve mice on day 3 and day 7 after infusion, whereas the G-MDSC from the irradiated tumor-free mice did not (Fig. 5E: p=0.01, F: p=0.01).

Figure 5. Increased circulating MDSC can drive systemic lymphopenia.

A. Illustration depicting the patient (NLP and LP) sorted MDSC were co-cultured with healthy pan T cell and the (B) T cell proliferation (Ki67⁺) and (C) T cell activation (TCR zeta⁺) were evaluated. D. Illustration depicting infusion of sorted G-MDSC (CD11b⁺Ly6G⁺Ly6C⁻) from GBM-bearing mice treated with or without RT sorted G-MDSC infused into non-tumor recipient mice and evaluated peripheral T cells 24 hours post MDSC infusion in recipient naïve mice. The MDSC from irradiated mice significantly reduced (E) CD4 and (F) CD8 T cells when compared to the unirradiated control. G. Illustration depicting GBM-bearing mice treated with anti-Ly6G antibody (α Ly6G) or isotype control during and after RT. The peripheral T cell and MDSC were analyzed at time points indicated in the graph. I. Depletion of G-MDSC, and (J) No change in M-MDSC during α Ly6G treatment. Depleting of G-MDSC led to rescuing (K) CD4 and (L) CD8 and (H) enhanced survival. Data are shown as mean ± S.D. (n=3 patients/group and n=5 mice/group). P-values were determined by two-tailed with unpaired student’s t-test in B and C, two-way ANOVA with post hoc Tukey’s test in I-L, and generalized Wilcoxon test in H. The indicated p-values in H-L are a comparison between isotype+RT and Ly6G+RT cohorts.

To further validate the association between circulating MDSC and lymphopenia, we depleted G-MDSC and neutrophils by treating the tumor-bearing mice with 5 doses of αLy6G antibody every 3 days beginning at day 7 (post tumor implantation) concurrently with fractionated RT (2Gy x5) or sham treatment. We evaluated the impact of depletion on circulating T cells and MDSC post tumor implantation and irradiation (Fig. 5G). Depleting G-MDSC and neutrophils significantly improved the survival of mice with or without RT (Fig. 5H: p=0.0001). As expected, αLy6G only depleted circulating G-MDSC and not M-MDSC (Fig. 5I & J). Notably, G-MDSC/neutrophil depletion was associated with a dramatic CD4 T cell recovery after RT and a more delayed and modest recovery of the CD8 T cell. The CD8 cells only began to recover on day 35 and continued to increase on day 42 (Fig. 5K & L). These data suggest that GBM tumors treated with RT can induce aberrant myelopoiesis to produce MDSC that are immunosuppressive, which can directly contribute to systemic reduction of T cells.

Radiation-induced MDSC has increased expression of Arg-1

MDSC is known to inhibit the T cells by expressing Arg-1, so we evaluated the Arg-1 level in MDSC in GBM patients and our tumor model. The bulk RNAseq of the patient PBMC samples showed increased Arg-1 expression in LP compared to NLP after chemoradiotherapy (Fig. 6A & B). Flow cytometric analysis of GBM patients also confirmed an increased Arg-1 expression of G-MDSC and M-MDSC in LP compared to NLP (Fig. 6C & D). Similar patterns of increased Arg-1 expression in the circulating G-MDSC and M-MDSC after cranial irradiation were also observed in our murine GBM model (Fig. 6E & F). These data demonstrate the functionally suppressive MDSC after RT in GBM is associated with increased Arg-1 expression.

Figure 6. Radiation-induced MDSC increases the expression of Arg-1.

RNA sequencing analysis reveal the changes in Arg-1 gene expression in LP (n=8) and NLP (n=6) patient samples. Heat map (A) and box plot showing the Arg-1 expression (B). Flow cytometric analysis of LP and NLP patient samples showing the changes of Arg-1 expression (MFI) in G-MDSC (C) and M-MDSC (D) cells. Arg-1 expression (MFI) from tumor and tumor+RT groups were represented in peripheral G-MDSC (F) and M-MDSC (G) cells. Data are shown as mean ± S.D. (n=8 patient samples/group and n=5 mice/group for flow cytometric analysis). P-values in C and D were determined by two-way ANOVA with post hoc Tukey’s test and in F and G two-tailed with unpaired student’s t-test.

MDSC inhibitors improved survival and lymphopenia in GBM tumor-bearing mice.

To further examine the role of radiation-induced MDSC in driving lymphopenia and tumor progression, in-vivo experiments with pharmacological inhibition of MDSC were performed with two drugs: CB1158 and tadalafil. CB1158 is an oral small-molecule inhibitor of Arg-1 that has been shown to inhibit MDSC function (24). Tadalafil is an oral and generic PDE5 inhibitor that is approved by the FDA to treat erectile dysfunction and pulmonary hypertension. The PDE5 inhibitor down-regulates the mRNA expression of Arg-1 and nitric oxide synthase (iNOS) to inhibit MDSC and has been shown to reduce circulating and intra-tumoral MDSC in prior clinical trials with solid tumor patients (25–27).

Two orthotopic syngeneic glioma models (GL261 and CT2A) were used to evaluate the impact of pharmacological inhibition with CB1158 and tadalafil. The tumor-bearing mice were treated with either CB1158 (100mg/kg) or tadalafil (2mg/Kg) for 10 days alone or in combination with sham or RT (2Gy x 5 to the head), and the mice were followed for survival for up to 120 days (Fig. 7A). Tumor-bearing mice treated with CB1158 plus RT had a significantly improved survival than RT alone for both GL261 and CT2A models (median survival of 89 days compared to 36 days for both models; p= 0.0028, Fig. 7B & D). Similarly, tadalafil plus RT also improved survival than RT alone for both models (median survival of 89 days compared to 36 days for both; p= 0.0028, Fig. 7C & E). However, CB1158 or tadalafil alone did not improve survival (Fig. 7B–E).

Figure 7. Blocking MDSC with Arg-1/PDE5 inhibitors during radiation therapy enhanced the survival of GBM-bearing mice.

A. Illustration depicting GL261 and CT2A implantation, five fractionated 2Gy radiation treatments (RT) to head along with CB1158 and Tadalafil treatments and monitoring the survival of C57BL/6 mice. The survival curve of GBM mice (B. GL261 and D. CT2A) treated with CB1158 or tadalafil (C. GL261 and E. CT2A) with fractionated cranial irradiation. Survival (n=5/group) was assessed by a generalized Wilcoxon test and p-values are depicted. Indicated p-values represent comparisons between the RT and RT+drug cohorts.

Next, we evaluated the level of MDSC subsets upon RT and drug treatments in the tumor-bearing mice (Fig. 8A). The administration of either CB1158 or tadalafil with RT successfully abrogated radiation-induced increase of G-MDSC and M-MDSC in the blood, spleen, and tumor (Fig. 8B & C). Notably, RT increased MDSC more in the spleen than in the tumor, suggesting the effect is more systemic rather than solely due to the local tumor micro-environment effect. However, neither CB1158 nor tadalafil alone significantly altered the existing G-MDSC or M-MDSC in the blood, spleen, and tumor of the tumor-bearing mice, thus suggesting these drugs predominantly targeted the radiation-induced MDSC (Fig. 8 B & C).

Figure 8. Treatments with Arg-1/PDE5 inhibitors and radiation decreased MDSC counts and rescued T cells in GBM-bearing mice.

A. Illustration depicting GL261 cell implantation, five fractionated 2 Gy radiation treatment (RT) to head along with CB1158 and Tadalafil treatments, MDSC and T cell profiling in peripheral blood, spleen, and tumor. Quantification of G and M-MDSC (B and C), CD4 and CD8 T cells (D and E) in peripheral blood, spleen, and tumor. Data are shown as mean ± S.D. (n=5 mice/group). P-values were determined by one-way ANOVA with post hoc Tukey’s test.

Corresponding to the effect on MDSC, administration of either CB1158 or tadalafil with RT generally prevented the radiation-induced reduction of CD4 and CD8 T cells in the blood, spleen, and tumor (Fig. 8 C & D), though an improvement of CD4 T cells after tadalafil plus RT was not observed in the tumor. Radiation-induced lymphopenia was more pronounced in the spleen than in the tumor, again suggesting a systemic effect. CB1158 or tadalafil alone did not significantly increase CD4 T cells in the blood, spleen, and tumor, but the drug alone increased CD8 T cells in the blood and tumor (Fig. 8 C & D). We then evaluated the effects of CB1158 and tadalafil on the bone marrow (Fig. S7A). As previously shown in Figure 4, RT enhanced G-MDSC, neutrophil, pre-neutrophil, GMP, and MPP in the bone marrow, but CB1158 or tadalafil alone or their addition to RT did not alter the production of these cells in the bone marrow (Fig. S7 B–G). These data suggest that both drugs can reduce radiation-induced peripheral MDSC without affecting radiation-induced myelopoiesis in the bone marrow.

DISCUSSION

Although clinical studies have demonstrated an independent association between RT and lymphopenia after chemoradiotherapy in GBM (12, 13, 20), it is not clear whether RT directly kills circulatory lymphocytes or indirectly suppresses T cell production to induce lymphopenia in these patients (28, 29). Combining clinical data from a prospectively designed correlative study and preclinical in-vivo data using syngeneic mice implanted with orthotopic GBM tumors, we demonstrated that cranial irradiation of GBM tumors leads to increased myelopoiesis in the bone marrow to generate circulating MDSC with enhanced suppressive activities. The resulting MDSC appears to contribute to systemic lymphopenia and worse tumor control after RT. Pharmacological inhibition of radiation-induced MDSC using clinically available drugs prevented lymphopenia and improved the survival of mice implanted with GBM tumors. Thus, our data provide a mechanistic explanation of how cranial irradiation of GBM tumors may increase circulating MDSC and suppress T cells in a subset of GBM patients, leading to worse tumor control and survival. Although RT represents one of the most important therapies to improve the survival of GBM patients, this radiation-induced myelopoiesis process may represent a mechanism of treatment resistance of certain GBM to chemoradiotherapy and a potential target for novel therapeutics to improve outcomes of GBM.

MDSC represents a major mechanism for tumors to evade the immune response. They are immature myeloid cells that have been identified in cancer patients as well as mice with cancers (30, 31). Studies have shown that GBM patients have relatively high number of MDSC in the peripheral blood at diagnosis as compared to age-matched normal volunteers or patients with other solid tumors (32). GBM tumors produce proinflammatory factors such as GM-CSF, M-CSF, IL6, IL-10, TGF-β, PGE-2, S100A8/A9, and CCL-2 that can block myeloid cell maturation and promote differentiation into MDSC (33). MDSC can also be induced in the tumor micro-environment by ionizing radiation through the STING pathway or the CSF-1 pathway (34, 35). There are several mechanisms of how MDSC can suppress T cells. MDSC produces Arg-1 and iNOS that depletes l-arginine, which is essential for T cell function (36). MDSC produces reactive oxygen species (ROS) and peroxynitrite (ONOO−), which modify receptors for antigens and chemokines on T cells and impair their function (37). MDSC produces cytokines such as IL-10 and TGF-β that impair effector T cells and NK cells and convert DCs into regulatory DCs, which can further impair T cell function. MDSC can also bring about T cell dysfunction through the downregulation of TCR ζ-chain (38, 39).

Our data supports that GBM tumors can create an environment for aberrant myelopoiesis that is further exacerbated with irradiation to induce an expansion of MDSC, and the resulting radiation-induced MDSC can drive T cell lymphopenia and poor survival. Although there has been a preponderance of data previously demonstrating irradiation of tumors can increase recruitment of MDSC to the tumor microenvironment, possibly through CSF-1 and HIF-1(35, 40), our data suggest that irradiation of GBM tumors can also increase the systemic production of MDSC by activating myelopoiesis in the bone marrow, which has not been extensively described to our knowledge. MDSC is thought to arise in the bone marrow (and also spleen in mice) in response to a variety of proinflammatory signals produced by the tumors and host cells in the tumor microenvironment (33, 41). Some key proinflammatory mediators include IL-1β, IL6, TNFα, and prostaglandin E2 (41–44). Irradiation of glioma cells and immune cells has been shown to increase the expression and secretion of these cytokines (45, 46), which likely further exacerbates the dysregulated myelopoiesis promoted by the tumor. Interestingly, our data suggest that irradiation of the brain in the absence of the tumor does not lead to a significant increase in circulating MDSC or myelopoiesis, suggesting that the interaction between irradiation and tumor microenvironment is necessary for the increased myelopoiesis. Furthermore, even in the hosts with intracranial tumors, irradiating normal tissue away from the tumor (such as thorax) does not increase the myelopoiesis as compared to irradiating the tumor directly. Although prolonged TMZ administration can contribute to the severity and duration of lymphopenia among glioma patients treated with RT(15), our preclinical data suggest TMZ does not contribute significantly to the radiation-induced myelopoiesis phenomenon. This explains the paradox that even though lymphopenia is prognostic for worse OS and that TMZ increases lymphopenia, concurrent TMZ significantly improves survival. Thus, our data suggest that radiation-induced myelopoiesis, rather than post-treatment lymphopenia, may be the true driver of the worse prognosis of lymphopenic GBM patients. As seen in our clinical trial, not all GBM patients would develop increased circulating MDSC or systemic lymphopenia after RT, so certain GBM tumors are more effective in generating MDSC to resist chemoradiotherapy. Additional investigation is currently ongoing to further elucidate the crucial signaling mechanisms that these GBM tumors use to upregulate myelopoiesis and how they are different from the GBM tumors that do not upregulate myelopoiesis.

Given circulating MDSC can directly contribute to radioresistance and poor survival, pharmacological inhibition of MDSC may be a rational approach to enhance the efficacy of RT. Since our data have demonstrated increased Arg-1 expression of MDSC after irradiation and that arginine depletion by Arg-1 is a key mechanism for MDSC to suppress T cells, we first examined the effect of Arg-1 inhibition using CB1148 during RT. CB-1158 has been shown to decrease the infiltration of CD68 myeloid cells while increasing the infiltration of CD8 T cells and NK cells into the tumor microenvironment (24). When administered concurrently with RT, our data show that concurrent administration of CB-1158 with RT can reduce the radiation-induced increase of MDSC and can improve lymphopenia and survival. Our finding is consistent with a prior study that showed the myeloid-specific knockout of Arg-1 in mice also improves tumor control after RT (47). PDE5 inhibitor can inhibit MDSC by suppressing the expression of Arg-1 and iNOS, which effectively block the two major immune-suppressive mechanisms of MDSC. PDE5 inhibitors can also down-regulate the expression of proinflammatory factors such as IL-4Rα, IL-1β, GM-CSF, and IL6 (27, 31, 48, 49), which in turn regulate MDSC production through myelopoiesis. Previous clinical studies have demonstrated that PDE5 inhibitors such as tadalafil can reduce circulating and intra-tumoral MDSCs and improve T cell responsiveness in patients with solid tumors (25–27). We found a similar effect in orthotopic GBM mice treated with either tadalafil or CB1158 in combination with RT. However, neither tadalafil nor CB1158 by itself were able to reduce MDSC in orthotopic GBM mice without RT, suggesting their effect is dependent on the context of RT. Furthermore, both drugs appear to only decrease MDSC in the peripheral organs without affecting radiation-induced myelopoiesis in the bone marrow. Taken together, our in-vivo experiments with CB1148 and tadalafil provide a proof-of-concept that the pharmacological inhibition of MDSC during RT may not only prevent radiation-induced lymphopenia but also improve tumor control. The data suggest that MDSC blockade may be more effective when combined with RT rather than administered alone. Notably, tadalafil is a generic drug and has an excellent safety profile including concurrent administration with RT for prostate cancer (50, 51).

The primary limitations of our study includes its small patient sample size and that not all patients yielded evaluable data for bulk RNAseq and flow cytometry. The technical limitations of scRNAseq,processing also prevented analysis of G-MDSC. Furthermore, orthotopic GBM mouse models do not reflect the complexity of human GBM, so our preclinical finding that the inhibition of radiation-induced MDSC may improve lymphopenia and survival require clinical validation. Based on our preclinical data, we have designed and opened a phase 1B study to combine tadalafil with standard chemoradiotherapy for newly diagnosed IDH-wildtype GBM patients (NCT04757662). The study is designed to evaluate the safety of combining tadalafil with chemoradiotherapy and to confirm the on-target effect of tadalafil to block the radiation-induced increase of circulating MDSC in GBM patients. The clinical results and the subsequent correlative studies may provide more insights on the potential of targeting MDSC as an anti-cancer treatment.

MATERIALS AND METHODS

Study design and sample processing.

A correlative study (#201611017) was designed to collect peripheral blood samples at baseline, week 2, and week 6 from GBM patients receiving a standard 6-week course of RT and TMZ (Fig. 1). The study was approved by the Institutional Review Board and was conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. Blood samples were collected into heparin-containing tubes. Peripheral blood mononuclear cells (PBMC) were also isolated and viably cryopreserved in dimethylsulfoxide at a final concentration of 10% and were batch processed for bulk RNA sequencing (RNA-seq), single-cell RNA sequencing (scRNA-seq), and flow cytometry. Lymphocytes, neutrophils, and monocytes in the whole blood of GBM patients were also quantified using the clinical hematology analyzer as part of the complete blood count assay. The raw subject-level data are provided in the supplemental materials.

RNA-seq, data processing, and gene analysis.

RNA-seq processing and data analysis: Total RNA from the patient PBMC were extracted using RNeasy Micro Kit (Qiagen) according to the manufacturer’s protocol. Ribosomal RNA (rRNA) was removed using the RiboMinus kit (Life Technologies) and custom-designed rRNA probes. rRNA-depleted total RNA was used as a template for RNA-seq library construction using the NEBNext mRNA Library Prep kit (New England BioLabs). Double-stranded cDNA was synthesized from rRNA-depleted total RNA, end-repaired, and then ligated to standard Illumina adaptor oligonucleotides. Adaptor-ligated cDNA libraries were loaded into HiSeq 2000 (Illumina) for sequencing. Raw RNA-seq data (in fastq format) were made available after the sequencing runs, which were parsed to align each sequence read to the original RNA sample. RNA-seq single-end reads were aligned using Kallisto v0.42.4 with default parameters to GENCODE v23 transcripts. Gene expression was measured in transcripts per million (TPM) and transformed as log2 (1 + TPM). Several types of transcripts were removed: histone- and mitochondria-related transcripts, the noncoding RNA, IGH/K/L- and TCR-related. As a result, 20,062 protein-coding genes were analyzed. Gene signature scores were calculated using the ssGSEA algorithm in the R package GSVA. Gene signatures are listed in Supplementary Table 1. Raw gene scores were medium-scaled to (−3, 3) range. Medium scaling was calculated as: Medium scale (ij) = (Xij - median (Xi))/MAD (Xi), where Xij is the score of signature i in sample j, MAD (Xi) is the median absolute deviation of scores of signature i.

scRNA-seq library construction and processing.

The PBMC from selected patient samples were used to prepare the RNA as per the 10X Genomics protocol (52). Briefly, the cryopreserved PBMC were rapidly thawed and resuspended in phosphate buffer saline with 0.5% BSA. The cells were processed to remove dead cells using Cell Removal MicroBeads and MACS Separator column. Following quantification (1000 cells/μl) the cells were used for the preparation of scRNA-seq library construction.

Single-cell data analysis

Kallisto-bustools (53) were used to align and quantify the single-cell reads. The data were cleaned of empty droplets, doublets, detected by scrublet algorithm, and cells with greater than 20% of counts aligned to the mitochondrial genome. The expression data for the remaining cells were normalized to a library size of 10000 and log-transformed. The data from 4 samples were integrated into a single analysis (Fig. S1E). To make the neighborhood graph for the possible batch effect, we used the BBKNN algorithm implemented in the scanpy package (54). The initial clustering was performed using the Leiden algorithm, and marker genes for each cluster were detected with rank_genes_groups. To perform a more detailed classification of the myeloid cell and T cell populations, cells identified as myeloid and T cells at the previous step were isolated, and they were separately analyzed using the principal component analysis (PCA), neighborhood graph construction, and clustering for this separate group (Fig. S1E).

Myeloid cell-T cell co-culture assay.

The pan T cells were sorted from healthy donor PBMC (Stem cell technologies) and mixed with human T cell activator CD3/CD28 Dynabeads (Thermo fisher scientific) at a 2:1 T cell/bead ratio. Further, CD33⁺ myeloid cells were sorted from patient PBMC (Stem cell technologies) and mixed with a 1:1 ratio of T cells and myeloid cells in RPMI 1640 medium. Cells were incubated for 72 hours, and proliferation was determined by flow cytometrically using Ki67 expression (55). The percent of Ki67 expression was compared between groups.

Cell culture.

GL261 (NCI) and CT2A murine glioma cells, transduced with luciferase reporter, were confirmed to be mycoplasma free by PCR. Cells were maintained in DMEM F12 (Gibco, Grand Island, NY, USA), supplemented with 10% fetal bovine serum (Gibco) and 1% penicillin-streptomycin (Gibco).

Animals, intracranial glioma implantation, and bioluminescence imaging (BLI).

Six to eight-week-old female C57BL/6 mice were obtained from Charles River and maintained under specific pathogen-free conditions at the Washington University School of Medicine. All animal work was conducted in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and approved by Institutional Animal Care and Use Committee, Washington University Division of Comparative Medicine (#21-0337).

On the day of surgery, 70% confluent GL261 or CT2A cells were trypsinized, washed, and prepared as a single cell suspension. Then, 4×10⁴ glioma cells in 5 μl were stereotactically injected into the right frontal cortex/striatum of mice under 2% isoflurane anesthesia. To confirm tumor engraftment and growth, bioluminescence imaging was performed approximately 1-week post tumor implantation using a Xenogen IVIS-100 system (Perkin Elmer, USA). Animals were injected with D-luciferin (150 mg/kg) in 100μl of phosphate buffer saline intraperitoneally and kept in a light-tight imaging chamber under 2% isoflurane anesthesia after 10 minutes of injection. Images were acquired and analyzed in living image software (Perkin Elmer). Following BLI, mice were randomized to each treatment arm such that each group had similar averages of tumor size per group.

Animal irradiation and drug treatments.

Mice were irradiated with 2 Gy/day for 5 days (unless otherwise specified) to the head by shielding the rest of the body with a lead shield. TMZ (Sigma-Aldrich, MO, USA) powder was reconstituted in dimethyl sulfoxide at a concentration of 50 mg/ml, aliquoted, and stored at −20C for further use. Mice were treated with an intraperitoneal injection of TMZ (34 mg/kg) for five consecutive days, 45 minutes prior to irradiation. The arginase-1 (Arg-1) inhibitor (CB1158, Incyte, and Calithera) was dissolved in saline and administered by oral gavage 45 minutes before RT at a concentration of 100 mg/kg/day. phosphodiesterase-5 (PDE5) inhibitor (tadalafil, Sigma) was dissolved in dimethyl sulfoxide and intraperitoneally administered 45 minutes before RT at 2 mg/kg/day. For survival studies, mice were followed until severe neurologic morbidity or death. For the thorax irradiation experiment, mice with intra-cranially implanted GBM (GL261) were irradiated with 2 Gy/day for 5 days to the thorax by shielding the rest of the body and the head with a lead shield.

In-vivo G-MDSC and neutrophil depletion assay.

Anti Ly6G (αLy6G; clone 1A8, catalog no. BE0075-1) and corresponding isotype control (clone 2A3, catalog no. BP0089) were purchased from Bio X Cell and injected intraperitoneal during and after RT. The αLy6G dosing was 500 μg/mice 2 hours after RT on the first day and then 250 μg/mice every 3 days for 4 additional doses. Depletion was confirmed by flow cytometry.

Adoptive MDSC transfer assay.

G-MDSC (CD11b⁺Ly6G⁺Ly6C⁻) were sorted from the bone marrow and spleen of GBM tumor-bearing mice treated with RT (2Gy x 5) and no tumor-bearing mice treated with the same RT doses for five days. The isolated cells were mixed and injected (1.5×10⁶/mouse) intravenously into the naïve mice. A total of three doses were administered on days 0, 3, and 5 into naïve mice. The changes in peripheral CD4 and CD8 T cells in the recipient mice were measured using flow cytometry and compared with the baseline number prior to the infusion.

Murine peripheral blood and organ processing for flow cytometry.

Blood (50-100 μl) was collected from the sub-mandibular vein using EDTA tubes at various time points. The blood was stained with various antibodies and processed with ACK buffer to remove RBC. Brain, spleen, and bone marrow were harvested two days after the final RT. The spleen was crushed on a 70μm strainer with a syringe plunger into FACS buffer (1X PBS, 0.5% BSA+5mM EDTA). Cells were pelleted (400 × g, 10 minutes, 4°C), lysed with ACK buffer (Thermo-Fisher, USA), washed twice, and finally resuspended in FACS. Similarly, bone marrow cells were isolated by flushing the femur and tibia in FACS buffer, followed by treatment with ACK buffer, washed twice, and resuspended in FACS buffer for further antibody staining. Tumors were gently isolated from the brain and incubated with collagenase buffer followed by ficol separation as described previously (56). The isolated tumor immune cells were used for antibody staining.

Flowcytometry.

Multiparameter flow cytometry was performed to evaluate different T cells and MDSC in human and murine samples. Data were acquired using a MACSQuant Analyzer 10 instrument (Miltenyi Biotec) and analyzed with FlowJo (v10.6.1) software.

Human sample:

For extracellular surface staining, dead cells were depleted using a dead cell removal kit (Miltenyi Biotec, USA) and washed with FACS buffer. Cells were then incubated with predetermined concentrations of antibodies for 1 hour at 4°C and were washed twice with FACS buffer. The following fluorophore-conjugated antibodies were used for different T cell subpopulation staining: CD3 (Order no. 130-113-129, Clone: BW264/56), CD4 (Order no. 130-121-336, Clone: M-T321), CD8 (Order no. 130-110-683, Clone: REA734), CD45RO (Order no. 130-113-561, Clone: REA611), CCR7 (Order no. 130-120-468, Clone: REA546), and CD127 (Order no. 130-113-413, Clone: REA614), all obtained from Miltenyi Biotech. We used the following gating strategy to identify different CD4 and CD8-T cell subsets in PBMC: effector-CD4 (CD3⁺CD4⁺CCR7⁻CD45RO⁻), naïve-CD4 (CD3⁺CD4⁺CCR7⁺CD45RO⁻), central memory-CD4 (CD3⁺CD4⁺CCR7⁺CD45RO⁺), effector memory-CD4 (CD3⁺CD4⁺CCR7⁻CD45RO⁺), effector-CD8 (CD3⁺CD8⁺CCR7⁻CD45RO⁻), naïve-CD8 (CD3⁺CD8⁺CCR7⁺CD45RO⁻), central memory-CD8 (CD3⁺CD8⁺CCR7⁺CD45RO⁺), and effector memory-CD8 (CD3⁺CD8⁺CCR7⁻CD45RO⁺). For MDSC staining, the following fluorophore-conjugated antibodies were used: CD14 (Catalog no. 555399, Clone: M5E2), CD15 (Catalog no. 562371, Clone: W6D3), CD33 (Catalog no. 340474, Clone: P67.6), HLA-DR (Catalog no. 339194, Clone: L243), CD11b (Catalog no. 557743, Clone: ICRF44), and Lox-1 (Catalog no. 358610, Clone: 15C4), all obtained from BD Biosciences except Lox-1 was obtained from BioLegend. For intracellular staining, cells were fixed and permeabilized using fixation/permeabilization (BD) solution, stained for 30 minutes with Arg-1 (R16-715, E-biosciences), and then resuspended with FACS buffer. We used the following gating strategy to identify different MDSC subsets in PBMC: G-MDSC (CD15⁺CD11b⁺Lox-1⁺), and M-MDSC (CD14⁺CD11b⁺HLADR⁻).

Murine sample:

The cells were stained in FACS buffer with different surface antibodies. For T cell subpopulation, the following fluorophore-conjugated antibodies were used: CD3 (Catalog no. 555274 and 564008, Clone: 17A2), CD19 (Catalog no. 562701, Clone: 1D3), CD4 (Catalog no. 565650, Clone: RM4-5), and CD8 (Catalog no. 553036, Clone: 53-6.7), all obtained from BD Biosciences. We used the following gating strategy to identify different T cell subsets: CD4-T cells (CD3⁺CD4⁺CD8⁻CD19⁻), and CD8-T cells (CD3⁺CD8⁺CD4⁻CD19⁻). For MDSC staining, the following fluorophore-conjugated antibodies were used: CD11b (Catalog no. 550993, Clone: M1/70), Ly6G (Catalog no. 560599, Clone: 1A8), and Ly6C (Catalog no. 560596, Clone: AL-21), all obtained from BD Biosciences. The intracellular staining was again performed with Arg-1 (R16-715) as described earlier. We used the following gating strategy to identify different MDSC subsets: G-MDSC (CD11b⁺Ly6G⁺Ly6C⁻), and M-MDSC (CD11b⁺Ly6C⁺Ly6G⁻). The hematopoietic stem and progenitor cells (HSPC) from the bone marrow were stained using the following fluorophore-conjugated antibodies: lineage negative cocktail (CD11b, CD45R/B220, Ly6G/Ly6C, CD3e, TER-119/Erythroid cells; Catalog no. 101226, 103224, 108424, 100330 and 116223; Clone M1/70, RA3-6B2, RB6-8C5, 145-2C11, and TER-119), CD117 (Catalog no. 105809, Clone: 2B8), Sca-1 (Catalog no. 122512, Clone: E13-161.7), CD135 (Catalog no. 135313, Clone: A2F10), CD34 (Catalog no. 560238, Clone: RAM34) and CD16/32 (Catalog no. 562896, Clone: 2.4G2), all obtained from BioLegend except CD34 and CD16/32 were obtained from BD Biosciences. We used the following gating strategy to identify different HSPC and lineage subsets in bone marrow: long-term-hematopoietic stem cells (LT-HSC; Lineage⁻C-kit⁺Sca⁻1⁺CD135⁻CD34⁻), short-term-hematopoietic stem cells (ST-HSC; Lineage⁻C-kit⁺Sca⁻1⁺CD135⁻CD34⁺), multipotent progenitor cells (MPP; Lineage⁻C-kit⁺Sca⁻1⁺CD135⁺CD34⁺), common myeloid progenitor (Lineage⁻C-kit⁺Sca⁻1⁻CD16/32^intCD34^int), granulocyte-monocyte progenitor (Lineage⁻C-kit⁺Sca⁻1⁻CD16/32^highCD34^high), megakaryocyte-erythrocyte progenitor (MEP; Lineage⁻C-kit⁺Sca⁻1⁻CD16/32⁻CD34⁻), common lymphoid progenitor (CLP; Lineage⁻C-kit^intSca1^intIL-7R⁺), pre-neutrophil (Pre-Neu; CD11b⁺Gr-1^int), and neutrophil (Neu; CD11b⁺Gr-1⁺).

Statistical analysis.

Patient characteristics were compared using the chi-square test (or Fisher’s exact testing for smaller cell counts) for categorical variables or the Mann-Whitney U test for continuous variables. Progression-free survival and overall survival were estimated using the Kaplan-Meier method and compared using the log-rank testing, with the times to event determined from the start of chemoradiotherapy. The data from the preclinical experiments were expressed as means ± SD. Between-group comparisons for the preclinical experiments were performed using a two-tailed t-test, whereas multiple-group comparisons were performed using analysis of variance (ANOVA) followed by the post hoc turkey’s test. All tests were 2-sided, and a p-value of <0.05 was considered significant. Statistical analyses were performed with GraphPad Prism version 9 (GraphPad Software, Inc., San Diego, CA, USA) and the Statistical Package for Social Sciences, version 23.0 (IBM SPSS Statistics, Armonk, NY).