Expansion of suppressive myeloid cell subsets in the peripheral blood of glioma patients
To determine the role of circulating immune parameters related to immune suppression in the clinical outcome of glioma patients, we set out to perform a detailed immunophenotypical analysis of circulating myeloid cells by multicolor flow cytometry in 134 patients undergoing surgery for a suspect glioma. To this aim peripheral blood samples were prospectively collected from 3 cohorts of patients and control HDs, as detailed in Figure 1, and parameters analyzed in each of the 3 cohorts are indicated in Table S1. As previously reported by us and others (31-33), GBM patients had a significant increase in the percentage of CD14+ circulating monocytes as compared to age and sex matched HDs, as well as a significant increase of PMNs, defined as CD15+ cells (Fig. 2a). When we analyzed lower grade gliomas we observed that also these patients were characterized by a significant expansion of both monocytes and PMNs, with a significant increase going from lower grade to GBM (Fig. 2b). Although some of these glioma patients underwent preoperative dexamethasone and it is known that this treatment increases PMNs (34, 35), no clear data exist regarding its effect on monocytes (36). However, we did not find a significant difference in CD14+ cells from patients with or without preoperative dexamethasone (Supplementary Fig. S1). Collectively, these results indicate that the increase in CD14+ monocytes in glioma patients is not due to dexamethasone, and suggest a dysregulation of the myeloid compartment. In the same blood samples, we also evaluated MDSCs levels. Previous studies analyzed MDSCs in glioma patients, but few of them analyzed more than one MDSC subset at a time, and most studies evaluated MDSCs among PBMCs, thus excluding the potential presence of PMN-MDSC subsets present in whole blood, discarded after a ficoll gradient. In addition, also cryopreservation is well known to influence the evaluation of PMN-MDSC. Based on these considerations, we decided to use a standardized 7-color panel, stemming from our previous experience (7, 8, 37), to detect in fresh whole blood the simultaneous presence of the following subsets: M-MDSC1 (CD14+/IL4Rα+), PMN-MDSC2 (CD15+/IL4Rα+), e-MDSC3 (Lin-/HLA-DR-/CD11b+/CD33+) and M-MDSC4 (CD14+/HLA-DRlow/-). All glioma patients had a significant expansion of circulating MDSC1, 2 and 4 subsets in comparison to age-matched HDs (Fig. 2 panels c, d and f). On the contrary, MDSC3 levels were significantly decreased in glioma grade II and IV patients with respect to HDs and in GBM as compared to grade III (Fig. 2e). Interestingly, MDSC4 were significantly increased in GBM compared to grade III gliomas, raising the possibility to use this marker in longitudinal studies to monitor evolution of grade III gliomas into GBM (Fig. 2f), since it is well known that GIII gliomas eventually evolve to GBM.
From these results, it appears that detection of circulating myeloid cells might represent a new tool for the follow-up of glioma patients, and that an altered myelopoiesis is associated with glioma grading progression.
Activation of STAT3/PD-L1 axis in circulating monocytes from glioma patients
We next investigated in circulating monocytes the activation of STAT3, one of the key players regulating tolerogenic activities of tumor-associated myeloid cells (21), by analyzing p-STAT3 expression in CD14+ cells and found that its intensity significantly increased in all glioma grade compared to HD (Fig. 2g), thus suggesting an active involvement of this transcription factor in the modulation of immune suppression.
Once phosphorylated STAT3 moves to the nucleus, where it can induce the expression of PD-L1 by binding to its promoter and activating its transcription (38). Thus, to further analyze the regulation carried out by STAT3, we evaluated the expression of PD-L1 on circulating monocytes, and found that the percentage of monocytes expressing PD-L1 was significantly increased on grade III and IV gliomas, but not in grade II (Fig. 2h).
All together these observations indicate that increased glioma grading is associated with a rise in CD14+ cells expressing activated STAT3 and PD-L1, suggesting their potential use as blood biomarkers.
ARG1 activity increases with tumor stage
It has been clearly demonstrated that ARG1, an enzyme constitutively expressed in PMNs and stored within intracellular granules, is a downstream target of activated STAT3, and that this regulation is also present in circulating MDSCs from cancer patients, in which it controls the immune suppressive activity (21). We thus investigated the presence of ARG1 in circulating PBMCs from glioma patients, by using confocal microscopy (Fig. 3a) and flow cytometry analysis (Fig. 3b). With both techniques, we observed the presence of a fraction of CD14+ monocytes expressing ARG1 (18.8% by flow cytometry analysis and 21.4% by confocal microscopy evaluation) localized in the cytoplasm of the cells.
Previously, elevated circulating levels of ARG1 in GBM patients have been associated with PMN degranulation and immunosuppression (24). In addition, we recently demonstrated that high serum levels of ARG1 in pancreatic ductal adenocarcinoma patients were associated with high ARG1 activity (20). We thus measured ARG1 levels in the plasma samples obtained from 64 glioma patients (15 GII, 10 GIII and 39 GBM) and found that ARG1 levels in glioma patients were significantly higher compared to HD control (Fig. 3c). We then assessed ARG1 enzymatic activity at both pH 7.2 and pH 9.5 (Fig. 3d and 3e). In both conditions, serum from glioma patients showed a significant increase in ARG1 activity that peaked in grade IV gliomas. Interestingly, ARG1 activity at pH 7.2 (determined as urea measured at pH 7.2) positively correlated with tumor grade, increasing from a median activity of 29.6 mg/hour of urea in HD to 85.4 mg/hour in GII, 93.9 and 152.8 mg/hour in GIII and GBM, arguing for ARG1 activity as a potential marker of glioma progression from grade III to GBM, although longitudinal studies are required to confirm and strengthen this conclusion.
Correlation between immune suppressive markers
To define the association of the different circulating parameters, we performed a PCA (Fig. 4) on the three groups of markers previously determined. When myeloid parameters considered in cohort 1 were evaluated, all variables displayed a moderate to strong positive correlation (r Pearson from 0.27 to 0.72), with the exception of MDSC3 that was negatively correlated with all the other variables (r Pearson from -0.26 to -0.42). Activation of STAT3 and PD-L1 expression were not correlated (r Pearson = -0.09) each other, while ARG1 activity showed a positive moderate correlation at the two different pH values (r Pearson = 0.45), but its quantity was not correlated from the activity both at pH 7.2 and pH 9.5 (r Pearson = 0.14).
The hierarchical clustering on PCA, used to identify groups of markers with similarities, divided the subjects of cohort 1 into three main clusters, with one of them containing 35 out of 36 HDs (cluster 1), while glioma patients were divided in 2 groups (Cluster 2 and 3) (Fig. 4 upper panel). Cluster 1 was characterized by lower levels of CD14+, CD15+ MDSC1, 2 and 4 and higher levels of MDSC3, in line with values obtained in HDs and, conversely, the opposite distinguished cluster 3 in which 32 out of 75 GBM were grouped (Fig. 4, upper panel). The proposed clustering classified HD and the glioma patients with a sensitivity of 79.8%, specificity of 97.2% and accuracy of 84.3%, with 21 false negatives and 1 false positive, suggesting that the model can be useful for ruling in glioma patients.
As far as it concerns the markers expressed by suppressive monocytes such as PD-L1 and activated STAT3 reported in Fig. 2 g-h, in a cohort of 63 glioma patients and 23 HD, two clusters could be identified. The first group comprised 87% of HD, and was characterized by lower percentage and intensity of activated STAT3, while higher levels were observed in group 2, in which expression of PD-L1 among monocytes was similarly expressed throughout the clustering (Fig. 4, central panel). This categorization classified HD and glioma patients with a sensitivity of 100%, specificity of 0% and accuracy of 73.3%.
Finally, the parameters related to ARG1 quantity and activity gave rise to 3 clusters (Fig. 4, lower panel). Cluster 1 was characterized by the lowest levels and activity of ARG1, cluster 2 had the samples with the highest quantity of ARG1, and an intermediate activity; cluster 3 contained samples with a low level of enzyme, but endowed with the highest activity. This clustering classified the HD and the glioma patients with a sensitivity of 100%, specificity of 0% and accuracy of 78.1%. Interestingly, all HDs fall in group 1, while gliomas distributed in the 3 clusters, but the majority of GBM fell in group 3, and low-grade gliomas were mainly present in cluster 1. It thus appears that the amount and activity of this enzyme characterize disease stage, and it has a significantly higher activity in high-grade gliomas.
Development of a diagnostic score to identify biomarkers associated with disease and with disease stage.
In order to define a diagnostic score, each biomarker was categorized according to high and low levels, and the association with the disease was tested first by univariate analysis (Table S2 and S3), followed by multiple logistic regression models considering in the stepwise model selection all markers within each cohort, together with age and sex. Analysis was performed including glioma patients as a single group, in a case control study (univariate in Table S2 and multivariate in Table 2a), or divided on the basis of pathological stage, from grade II to GBM (univariate in Table S3 and multivariate in Table 2b). When cell-associated myeloid markers were considered as biomarkers to differentiate HD from glioma patients, levels of CD15+ cells, MDSC1, and MDSC3 emerged as independent factors predicting the presence of disease (Table 2a), with an overall accuracy of 87.1%. In detail, high levels of CD15+ cells and MDSC1 were significantly associated with a high risk of disease (adjusted OR: 7.2, 95%CI: 2.2─24.1 and adjusted OR: 40.1, 95%CI: 10.1─160.4, respectively), while high levels of MDSC3 showed a significantly lower probability to develop disease (adjusted OR: 0.1, 95%CI: 0.02─0.4). The logit transformation of the probability of glioma (any grade) risk was calculated as follows:
Of note, when glioma patients were classified according to their grade, CD15+ cells, MDSC1, and MDSC3 remained independent significantly predictors of GBM, and MDSC1 was the biomarker significantly associated to all grades of disease (Table 2b, accuracy of 72.9%). In this case the formula for the risk score was:
Regarding PD-L1 and p-STAT3 expression, only the shift of intensity of expression of activated STAT-3 in monocytes (p-STAT3) remained an independent factor predicting the presence of disease (Table 2a, cohort 2), with an overall accuracy of 79.1%. In detail, high levels of expression of p-STAT3 were significantly associated with a high risk of disease (adjusted OR: 13.8, 95% CI: 4.3─44.3), and the final score was
When glioma patients were considered according to their grade, intensity of p-STAT3 was confirmed to be an independent significant predictor of all grades of disease (Table 2a, accuracy of 68.6%). From these results it thus appears that high levels of STAT3 activation in monocytes is not only a marker of immune suppression, but also a biomarker of disease.
The same analysis performed with soluble biomarkers identified ARG1 activity at physiological pH (Urea pH 7.2) as an independent risk factor of disease (adjusted OR: 255, 95% CI: 26.7─2434), with an overall accuracy of 94% and
When this analysis considered glioma patients according to their grade, ARG1 activity at pH.7.2 remained an independent significant predictor of all grades of disease, and, interestingly, significantly discriminated HD from low-grade gliomas, thus indicating that it is an early biomarker of glioma disease (Table 2b, cohort 3), with an accuracy of 72%.
Development of a GBM prognostic model
We next evaluated the prognostic role of the myeloid-associated biomarkers present in this study to predict the outcome of GBM patients by performing a univariate analysis (Table S4), and then using multiple survival analyses (Table 3). Both analyses examined a cohort of 67 patients for myeloid cell markers (cohort 1), of 45 patients for PD-L1 and p-STAT3 (cohort 2), and of 32 patients for ARG1 markers (cohort 3) with available clinical and follow-up data (Table S5). At an estimated median follow-up time of 33.2 months (95% CI 30.5─49.1) for cohort 1, 27.6 months (95% CI 19.3─30.5) for cohort 2, and 33.2 months (95% CI 28.9─37.7) for cohort 3, median OS times were 12.6 months (95% CI 10.6─18), 12.5 months (95% CI 6.2─16.7), and 12.0 months (95% CI 6.5─28.5), respectively.
In the multiple Cox regression model for the myeloid cell-associated markers, elevated MDSC2 levels remained significantly associated with worse OS (hazard ratio (HR) =1.8, 95% CI: 1.0─3.4), in addition to ECOG PS, surgery and MGMT methylation (Table 3; C-index=0.74). The prognostic index derived from the model was:
Among STAT3 and PDL-1 markers, elevated levels of expression of p-STAT3 were significantly associated with worse overall survival (HR=4.43, 95% CI: 1.7─11.6), in addition to surgery, Stupp and MGMT methylation (Table 3; C-index=0.75), The multiple prognostic model for ARG1 identified high levels of ARG1 activity at alkaline pH 9.5 as a risk factor for survival (HR=3.7, 95%CI: 1.4─9.9), in addition to Stupp treatment (Table 3; C-index=0.76),