Peripheral blood samples were prospectively collected from three independent cohorts of patients with different stages of pancreatic ductal adenocarcinoma admitted at the Unit of General and Pancreatic Surgery of the Azienda Ospedaliera Universitaria Integrata of Verona before surgical resection or Healthy Donors (HD). Clinic-pathologic features of patients were reported in Tables 1 and 2 and included age, gender, tumor location and TNM stage. No subject had a prior history of cancer or was undergoing therapy at the time of sample collection. BM aspirates were subjected to lysis to remove red blood cells, with a hypotonic solution of ammonium chloride. Cells were plated (2 × 10⁶ cells/well) into a 24-well tissue culture plate (BD, Franklin Lakes, NJ, USA) in IMDM (Lonza, Visp, Switzerland) supplemented with with 10% FBS (Euroclone, Milano, Italy), 100 U/ml penicillin/streptomycin (Euroclone, Milano, Italy), β-mercaptoethanol (Sigma-Aldrich, Milan, Italy) and 10 mM HEPES (Euroclone, Milano, Italy) in presence of 40 ng/ml of G-CSF and GM-CSF (Miltenyi Biotec) for 4 days at 37 °C, 8% CO₂, obtaining BM-MDSC as previously reported [16].Tab1

Clinical characteristics of the study population

PBMCs were isolated from leukocyte-enriched buffy coats from healthy volunteers (Transfusion Center, University and Hospital Trust of Verona, Verona, Italy) by Ficoll-Hypaque (GE Healthcare, Uppsala, Sweden) gradient centrifugation. PBMCs were then counted, frozen at − 80 °C and stored in liquid nitrogen. PBMCs were recovered, washed in IMDM medium (Lonza, Visp, Switzerland), supplemented with 10% FBS (Euroclone, Milano, Italy), 100 U/ml penicillin/streptomycin (Euroclone, Milano, Italy), β-mercaptoethanol (Sigma-Aldrich, Milan, Italy) and 10 mM HEPES (Euroclone, Milano, Italy), resuspended at a final concentration of 10⁷ cells/ml in PBS and stained with 1 μM as final working concentration of CellTrace Violet stock solution (Thermo Fisher Scientific, Waltham, MA, USA), followed by 5 min’ incubation at 37 °C, protected from light. Labelled “target” PBMCs were stimulated with coated 0.6 μg/ml anti-CD3 (clone OKT-3, eBioscience, Thermo Fisher Scientific, Waltham, MA, USA) and 5 μg/ml soluble anti-CD28 (clone CD28.2, eBioscience, Thermo Fisher Scientific, Waltham, MA, USA) for 4 days and co-cultured with “effectors” M-MDSCS (CD14⁺ cells) or PMN-MDSC (CD66b⁺ cells) cells at 0.5:1, 1:1, 3:1, 6:1 ratio (effector:target) in 384 flat bottom well plates (BD, Franklin Lakes, NJ, USA). Cell cultures were incubated at 37 °C and 8% CO₂ in arginine and glutamine–Free-RPMI (Biochrom AG, Berlin, Germany), supplemented with 2 mM L-glutamine (Euroclone, Milano, Italy), 150 μM arginine (Sigma-Aldrich, St. Louis, MO, USA), 10% FBS (Superior, Merck, Darmstadt, Germany), 10 U/ml penicillin and streptomycin (Euroclone, Milano, Italy), and 0.1 mM HEPES (Euroclone, Milano, Italy). At the end of the culture, cells were stained with PE-Cy7 conjugated anti-CD3 (UCHT1, eBioscience, Thermo Fisher Scientific, Waltham, MA, USA), and CellTrace signal of gated lymphocytes was analyzed. TruCount™ tubes (BD, Franklin Lakes, NJ, USA) were used to determine the absolute cell number of CD3⁺ cells in the samples. Data were analyzed by FlowJo software (Tree Star, Inc. Ashland, OR, USA).

Blood was collected into EDTA-treated tubes (BD Biosciences, NJ, USA) and processed fresh. For each donor, 450 μL of whole blood or 10⁶ frozen PBMCs were taken for MDSC characterization by flow cytometry. Sample tubes were washed in phosphate-buffered saline (PBS), incubated with Fc receptor (FcR) Blocking reagent (Miltenyi Biotec) for 10 min at 4 °C to saturate FcR and then stained with fluorochrome-conjugated antibodies (Additional file 1: Supplementary methods). For tumor-infiltrating leukocytes evaluation, tumor biopsies were minced and incubated for 2 h at 37 °C with shaking with an enzymatic cocktail. Normal tissues were detected by pathological analysis and isolated from patient biopsies. 5 × 10⁵ cells were washed with PBS supplemented with 2 Mm EDTA, incubated with FcR Blocking reagent (Miltenyi Biotec) for 10 min at 4 °C and then stained with fluorochrome-conjugated antibodies (Additional file 1: Supplementary methods).

Total RNA was isolated using TRIzol reagent (Life technology, CA, USA) and RNA integrity assessed using Agilent-2100-Bioanalyzer (Agilent Technologies, CA, USA). RNA from human CD14⁺ cells was further purified with RNeasy MinElute Cleanup kit (Qiagen, Venlo, Netherlands) and cDNA was synthesized and amplified from total purified RNA with RETROscript® (Life technology, CA, USA). All the samples were hybridized to Affymetrix U133 PLUS 2.0 arrays and scanned with an Affymetrix GCS 3000 7G scanner.

All statistical analysis was carried out using SigmaPlot (Systat Software) and R/Bioconductor. For statistical comparison of two groups, non-parametric Mann-Whitney Wilcoxon test was used. Data are shown as mean ± SD or mean ± SEM as indicated in the figures legends. Receiver operator characteristic (ROC) analysis was performed to determine the performance of MDSC percentage in distinguishing patients with metastatic carcinoma. The optimal cutoff threshold for MDSC percentage was obtained based on the maximization of the Youden’s statistics J = sensitivity+specificity+ 1 by using an R-based software as described [17]. Statistical analyses were performed using SPSS Statistics 22 (IBM Corporation, Somers, NY, USA), GraphPad Prism software program (version 6.0; GraphPad Software, San Diego, CA), and the statistical language R.

The immune composition of PDAC has been shown to have prognostic implications, with high number of CD8⁺ T lymphocytes associated with good outcome while accumulation of myeloid cells with poor prognosis [18, 19]. However, our knowledge on the immune heterogeneity of the PDAC microenvironment is still limited and need to be further investigated. To dissect this complexity, using a multicolour flow cytometry approach, we analysed infiltrating leukocytes isolated from 29 tumor samples from treatment-naïve PDAC patients (Table 1) and 5 normal pancreatic biopsies, obtained from tumor-free tissues of some patients. Among alive CD45⁺ cells, we focused on T lymphocytes (CD3⁺ cells), effector T lymphocytes (CD3⁺CD8⁺ cells), helper T lymphocytes (CD3⁺CD4⁺ cells), regulatory T lymphocytes (CD3⁺CD4⁺ CD25⁺FoxP3⁺ cells, Tregs), B lymphocytes (CD3⁻CD19⁺ cells), regulatory B cells (CD3⁻CD19⁺CD25⁺FoxP3⁺ cells, Bregs), myeloid-dendritic cells (CD11b⁺CD11c⁺HLA-DR⁺ cells, DCs), plasmacitoid DCs (CD11b⁺CD11c⁻CD123⁺, pDCs), macrophages (CD14⁺HLA-DR⁺CD68⁺CD206⁺ cells), granulocytes (PMNs, CD14⁻CD15⁺CD11b⁺ cells) as well as two MDSC subsets: e-MDSCs (Lin⁻HLA-DR⁻CD11b⁺CD33⁺ cells) and M-MDSCs (CD14⁺HLA-DR^−/lo cells) (Additional file 1: Figure S1). Notably, we found that PDAC tissues have a higher CD45⁺ cell infiltrate than their normal counterpart, likely reflecting the ability of the tumor or surrounding stroma to release soluble factors attracting immune cells [20, 21] (Fig. 1a). Among the CD45⁺ cells, we identified a high frequency of several myeloid cells, such as PMNs, MDSCs and macrophages (Fig. 1b) and several T cell subsets, supporting the current hypothesis that PDAC is not an immune “desert” [22, 23]. While we did not find expansion in Bregs (0.052 ± 0.012) and pDCs (0.073 ± 0.018), we observed a higher frequency of several myeloid cells, such as PMNs (28.89 ± 4.693), M-MDSCs (0.969 ± 0.167), e-MDSC (1.235 ± 0.198) and macrophages (8.832 ± 2.265) and Tregs (1.092 ± 0.196) (Fig. 1b), supporting the concept that PDAC is a tumor with an immune-hostile tumor microenvironment [24]. Indeed, a significant inverse correlation between T cell numbers with either PMN or M-MDSCs, but not with macrophages and e-MDSCs, could be detected (Fig. 1c), which is in line with recent reports. Of note, a significant inverse correlation between both PMNs and B cells, as well as between PMNs and different T cell subsets including effectors T cells, helper T cells and Tregs emerged (Additional file 1: Figure S2). Interestingly, we identified a significant direct linear correlation between T cells and Tregs as well as a trend between M-MDSCs and Tregs (Additional file 1: Figure S2). Collectively, these results suggest that accumulation of myeloid cells, such as MDSCs, in PDAC is detrimental for T cell infiltration.Fig1Fig. 1

Immune characterization of PDAC tumor microenvironment. a Leukocytes infiltration (CD45⁺ cells) in normal pancreas (n = 5) and PDAC tissue (n = 29) biopsies. Statistical analysis was performed by ANOVA test. b Immune populations abundance (% of CD45⁺ cells) in PDAC tissues. c Correlation between tumor-infiltrating T cells with either macrophages, PMNs, M-MDSCs or e-MDSCs within PDAC tissues. Correlation analysis was performed by Spearman’s rank correlation

Preclinical data suggest that PDAC effects on the immune system are not limited to local microenvironment but might cause systemic alterations, fuelling an “emergency” myelopoiesis that favours the accumulation of circulating MDSCs [13]. To assess systemic changes in PDAC patients, we evaluated the presence of different MDSC subsets in the peripheral blood, following recently published guidelines [25]. Three independent patient cohorts (Table 2) were enrolled to define the MDSC baseline at diagnosis: the first cohort comprised 21 PDAC patients (stage III-IV) and 8 age- and sex-matched healthy donors (HD) (Fig. 2a), the second cohort comprised 23 PDAC patients (stage III-IV) and 9 HDs (Fig. 2b); the last cohort consisted of 73 PDAC patients, including resectable (stage I-II, n = 21) and non-resectable (stage III-IV, n = 52) tumors and 28 HDs (Fig. 2c). MDSC analysis was performed using both fresh whole blood (WB) cells (Fig. 2a-b) and frozen peripheral blood mononuclear cells (PBMCs) (Fig. 2c). In fresh WB we analyzed the frequency of monocytes (CD14⁺CD15⁻CD11b⁺) and granulocytes (PMNs, CD15⁺CD14⁻CD11b⁺) as well as the presence of MDSC1 (CD14⁺IL-4Rα⁺), MDSC2 (CD15⁺IL-4Rα⁺), MDSC3 (Lin⁻HLA-DR⁻CD33⁺) and MDSC4 (CD14⁺HLA-DR^−/low) subsets (Additional file 1: Figure S3). Among frozen PBMCs, we only discriminate monocytic (MDSC1 and MDSC4) and early-stage (MDSC3) MDSCs; the assessment of PMN-MDSCs is not accurate and probably even misleading, since granulocytes (including PMN-MDSCs) are typically lost during freezing/thawing procedure. We detected a significant increase in circulating M-MDSC subsets (MDSC1 and MDSC4) in PDAC patients compared to the control group in the three independent analyses (for CD14⁺IL-4Rα⁺ cells, median value 0.19% vs. 0.57%, p < 0.001 in the first cohort, 0.18% vs. 0.59%, p < 0.001 in the second cohort and 2.2% vs. 4.3%, p = 0.002 in the third cohort; for CD14⁺HLA-DR^−/low cells, median value 0.19% vs. 0.31%, p = 0.033 in the first cohort, 0.08% vs. 0.32%, p = 0.042 in the second cohort and 1.78% vs. 3.25%, p < 0.001 in the third cohort). Moreover PMN-MDSC subset (MDSC2) was significantly increased in PDAC patients: CD15⁺IL-4Rα⁺ cells, median value 1.53% vs. 4.89%, p = 0.006 in the first cohort, 1.89% vs. 6.78%, p < 0.001 in the second cohort. Interestingly, PDAC patients showed an increased frequency in WB of both monocytes and granulocytes compared to HDs: for monocytes, median value was 0.94% vs. 3.15%, p < 0.001 in the first cohort, 0.98% vs. 3.95%, p < 0.001 in the second cohort; for granulocytes, median value 44.82% vs. 56.23%, p = 0.006 in the first cohort, 47.89% vs. 62.45%, p < 0.001 in the second cohort. Finally, we did not observe any alteration in circulating e-MDSCs (MDSC3) between HDs and cancer patients in any of the analysed cohorts.Fig2Fig. 2

Blood-circulating MDSC enumeration in PDAC patients. a-b Flow cytometry analysis of circulating myeloid cells in whole blood of two independent cohorts of PDAC patients (b PDAC n = 21, HD = 8; c PDAC n = 23, HD = 9): monocytes (CD14⁺CD15⁻), MDSC1 (CD14⁺IL-4Rα⁺), MDSC4 (CD14⁺HLA-DR^low/−), granulocytes (CD15⁺CD14⁻), MDSC2 (CD15⁺IL-4Rα⁺) and MDSC3 (LIN⁻HLA-DR⁻CD33⁺SSC^high). c Flow cytometry analysis of circulating M-MDSCs (MDSC1, CD14⁺IL-4Rα⁺; MDSC4, CD14⁺HLA-DR^low/−) and e-MDSCs (MDSC3, LIN⁻HLA-DR⁻CD33⁺SSC^high) in PDAC patients (n = 73) compared to healthy donors (HD; n = 28). M-MDSC percentages were evaluated on frozen PBMCs, whereas e-MDSCs on the whole blood. Statistical analysis was performed by ANOVA test

We then evaluated in vitro the immunosuppressive properties of PMNs (isolated as CD66b⁺ cells) and monocytes (isolated as CD14⁺ cells) freshly purified from blood samples of the second PDAC patients cohort (n = 10) to confirm their MDSC-associated functional activity (cell purity was above 95% after cell isolation, Fig. 4a). Isolated cells were co-cultured in presence of activated, cell trace-labelled allogeneic PBMCs for 4 days. As reported in Fig. 4b, at the highest T cells:myeloid cells cell ratio (1:6) both myeloid cell subsets showed suppressive activity, whereas only monocytes were able to restrain T cell proliferation at lower cell ratio (i.e. at 1:1 ratio; p = 0.021 myeloid cells/PBMCs), in agreement with previous preclinical reports [6, 26]. Therefore, these data suggest that in PDAC patients, the per cell based suppressive capacity of neutrophils is lower than the one of monocytes, as it was already observed for other tumors [27].Fig4Fig. 4

Circulating monocytes from PDAC patients are able to restrain T cell proliferation in vitro. a Freshly isolated PMNs (CD66b⁺ cells, orange box) and monocytes (CD14⁺ cells, blue box) from PDAC patients analysed by flow cytometry and haematoxylin-eosin staining. b Functional assay reflecting the different ability of PMNs and monocytes to affect T cells proliferation when co-cultured in vitro with CD3/CD28-activated-PBMCs at different ratios. All values are normalized on the activated PBMCs in the absence of myeloid cells (grey bar) and reported as percentage of Cell Trace⁺CD3⁺ cells. Statistical analysis was performed by ANOVA test. c Functional assay performed (at 1:3 ratio of PBMCs:CD14⁺ cells) on monocytes of PDAC patients (n = 26) compared to HDs (n = 8), reported as percentage of CD3⁺ proliferating cells (right panel) and graphed as proliferation peaks of Cell Trace⁺CD3⁺cells after the co-culture (left panel). Among all PDAC patients, “Suppressive CD14⁺ cells” (blue) and “Non-suppressive CD14⁺ cells” (red) were grouped based on the quantitative analysis of the in vitro immunosuppressive function. Statistical analysis was performed by ANOVA test. d Different ability of suppressive and non-suppressive monocytes to limit CD3⁺ T cell proliferation at different cell ratios. Statistical analysis was performed by ANOVA test. e Pearson correlation between MDSC4 and MDSC1 among CD14⁺ cells of PDAC patients. f Pro-metastatic potential of suppressive CD14⁺ cells. Statistical analysis was performed by Pearson Chi-Square test

In order to define the molecular network relevant for CD14⁺ cell immunosuppression, we performed a genome-wide mRNA expression profiling on purified monocytes isolated from 3 suppressive and 4 non-suppressive PDAC patients. First of all, we compared the gene profiles of PDAC-derived monocytes against three independent public datasets of normal circulating CD14⁺ cells isolated from HDs (GSE60601, GSE64480 and GSE13899) demonstrating a specific cancer-related signature, as the hierarchical gene clustering revealed different patterns of expression between the two groups (Additional file 1: Figure S5A). Indeed, by using the gene set enrichment analysis (GSEA) of hallmarks of cancer, the differentially expressed genes were enriched in categories involved in: TNFα signaling via NF-κB, inflammatory response, IL6 JAK/STAT3 signaling and apoptosis categories (Additional file 1: Figure S5B). These results are in agreement with our recent findings [12] and indicate how cancer cells alter the normal monopoiesis favouring the development of CD14⁺ cells with cancer-related imprint. To elucidate further this cancer-driven reprogramming, we compared the gene profile of suppressive towards non suppressive PDAC monocytes by clustering genes according to their expression levels, demonstrating that immunosuppressive monocytes had a distinctive gene signature (Fig. 5a). The comparative analysis identified differences in the expression of genes involved in metabolism, cell cycle, signaling and structural components (Fig. 5b). Considering the structural constituents category, suppressive CD14⁺ cells showed a greater expression in FBN2, TSPAN16, LEPR, CLTA and CD163 that are normally associated with classic monocytes. In particular, CD163 expression is strongly regulated by IL-6 and IL-10 that are two of the main inflammatory mediators in PDAC patients’ sera [12, 21]. Moreover, the CD163 cleaved form (sCD163), released by monocytes/macrophages, was reported to inhibit T cell proliferation, underlying its potential involvement in immune evasion [28]. Suppressive monocytes showed also an altered cell cycle-associated gene signature, as well as a complex signaling-related gene enrichment. Among cell cycle cluster, we found the expression of CASP2, recently described as a regulating key of myeloid progenitor differentiation [29]; AKAP9, involved in c-AMP-dependent suppression on LPS-activated macrophages and NLRP1, described to impair T cell responses [30]. In the signaling category, we identified the expression of several zinc-finger protein-coding genes (ZFP3, ZNF585B, ZNF320, ZNF329, ZNF148, ZNF137P, ZNF573, ZNF776 and ZNF441), as well as the different pattern of expression of MAP 3 K3, PRKRA, JAK2 and different components of the STAT family (STAT1, STAT2, STAT3, STAT5A, STAT5B and STAT6) that have been already defined as MDSC-associated transcriptional factors [9]. In the metabolism group, we identified several genes potentially linked to immunosuppression [31]: fatty acid and lipoprotein metabolism-related genes, such as CD36, LYPLA1 and CERS5; energy (ATP) metabolism-associated genes, such as ATP51C, ATP5G2 and SDHB; glucose metabolism-associated genes, such as PDK4 and GXYLT1, as well as hormone and water soluble vitamins metabolism-associated genes (i.e. HSDL2 and PCCA respectively). Finally, we identified different genes involved in both amino acid metabolism, such as ERICH1, GLS, CTSC and ARG1 and amino acid modifying enzymes, such as NAT2, UST and OXR1. To understand the depth of cancer-induced monocyte reprogramming towards MDSCs, we compared suppressive CD14⁺ cell gene profiles to gene signatures of human bone marrow (BM) derived MDSCs (BM-MDSCs, n = 8 independent donors) obtained by in vitro differentiation of BM cells in presence of a cytokine cocktail composed by G-CSF and GM-CSF, as previously reported [13] (Additional file 1: Figure S5C). Despite the phenotypic differences and the expected variances in their isolation and generation, BM-MDSCs and cancer patient immunosuppressive monocytes displayed a shared signature (not-differentially expressed genes, Fig. 5c) characterized by genes such as PTGS2, SOCS2, TNF, IDO1, CD38 and ARG1, all related to immune regulation. Interestingly, they also shared expression levels of AKT3, JAK1, JAK3, STAT1, STAT4, STAT5, STAT6 and STAT3, suggesting a common signaling network among these myeloid cells, as well as the same expression of CFLAR, which we recently reported as an important candidate for driving the acquisition of the immunosuppressive program in monocytes [12].Fig5Fig. 5

Gene profiling of suppressive CD14⁺ cells isolated from PDAC patient. a Supervised clustering of suppressive and not suppressive monocytes arrays using 1119 differentially expressed genes (FDR < 0.05 and absolute fold change > 2). b Clustering of cell cycle, structure, signaling and metabolism in suppressive- and not suppressive monocytes (absolute fold change > 2; FDR < 20%). c Difference in expression between suppressive monocytes isolated from PDAC patients and human BM-MDSCs samples for genes in JAK/STAT Signaling Pathway. d Dot plot of log fold change demonstrating common (yellow plots) or different (purple plots) gene expression modulation between differentially expressed signature of either tumor-educated or suppressive monocytes to related controls. e miRNAs-expression profile of suppressive and non-suppressive CD14⁺ cells isolated from PDAC patients using 19 differentially expressed miRNAs (FDR < 0.05 and absolute fold change > 2)