In the present study, a total of 155 patients were enrolled. Patients were aged between 40 and 84 years (median, 58.0 years) and had histologically confirmed breast cancer that was treated in the Department of Organ Regulatory Surgery, Fukushima Medical University (Fukushima, Japan) between January 2011 and June 2016. A total of 110 preoperative patients, including 18 patients with stage I disease, 39 with stage II, 17 with stage III, and 36 with stage IV cancer; 23 postoperative patients, including 5 with stage I, 10 with stage II, 7 with stage III, and 1 with stage IV cancer in accordance with National Comprehensive Cancer Network (NCCN) Clinical Practice Guidelines in Oncology: Breast Cancer Screening and Diagnosis (18); 22 recurrent patients that had received a regimen of chemotherapy (6 cycles of 1,000 mg/m² gemacitabine on days 1 and 8 for 21 days); and 18 healthy volunteers were enrolled. None of the preoperative patients had received any anticancer treatment. A total of 177 blood samples were collected from the 155 patients with breast cancer, including double sampling in recurrent patients (pre- and post-chemotherapy), and from the 18 healthy volunteers. Peripheral blood mononuclear cells (PBMCs) were isolated using the Ficoll density gradient centrifugation method (Ficoll-Hypaque; GE Healthcare Life Sciences, Chalfont, UK) from 20 ml venous blood collected in EDTA tubes. Aliquots of PBMCs were cryopreserved in freezing medium (BLC-1; Juji-Field Co. Ltd, Tokyo, Japan). The plasma was separated by centrifugation at 1,500 × g for 10 min in a refrigerated centrifuge at 4°C and kept at −80°C until analysis. Double samples (pre- and post-chemotherapy) from pleural effusions were examined. A total of 1×10⁶ cells from 100-ml plural effusion samples were isolated by centrifugation at 400 × g for 30 min at 4°C using the Ficoll density gradient centrifugation method, and subjected to flow cytometric analysis as described. The present study was approved by the Ethics Committee of Fukushima Medical University (2011–2016). Written informed consent was obtained from all enrolled patients and healthy donors.

A total of 1×10⁶ frozen PBMCs were thawed rapidly and washed three times with PBS. A three-color flow cytometric analysis was performed with a cocktail of antibodies, consisting of fluorescent isothiocyanate (FITC)-conjugated anti-CD14 (cat. no. ab28061; 20 µl; Abcam, Cambridge, UK), phycoerythrin (PE)-conjugated anti-CD11b (cat. no. IM-2581 U; dilution, 0.2 mg/ml; Beckman Coulter, Inc., Brea, CA, USA), and phycoerythrin cyanin (PC)5.1-conjugated anti-CD33 (cat. no. FAB1137F; 20 µl; Beckman Coulter, Inc.). Following incubation with PBS for 30 min at 4°C, samples were washed with fluorescence-activated cell sorting (FACS) buffer. Pellets were subsequently resuspended in 800 µl FACS buffer. Data acquisition and analysis were performed on a FACS Aria II flow cytometer (BD Biosciences, San Jose, CA, USA; Fig. 1) using Flow Jo v10.2 software (Tree Star, Inc., Ashland, OR, USA). The labeled cells were first gated (R1) based on their expression of CD14; R1 was composed of CD14⁻ cells. The fraction of cells in this population that expressed the myeloid markers CD11b and CD33 was then determined. Therefore, in the present study, MDSCs were defined as CD14⁻, CD11b⁺ and CD33⁺ cells. The percentage of MDSCs was calculated as a percentage of the total PBMCs.

To assess cytokine production, 1×10⁶ PBMCs were cultured in 1 ml RPMI-1640 medium (Wako Pure Chemical Industries, Ltd., Osaka, Japan) containing 10% heat-inactivated fetal calf serum (FCS; Gibco; Thermo Fisher Scientific, Inc., Waltham, MA, USA) and 100 µg/ml phytohemagglutinin (PHA; Sigma-Aldrich; Merck KGaA, Darmstadt, Germany) for 24 h in 5% CO₂ at 37°C. Subsequently, 1×10⁶ PBMCs were seeded into a 96-well plate (Costar, Lowell, MA, USA) in the presence of PHA. Following cultivation, IL-6 and IFN-γ in supernatants were measured using a Quantikine ELISA kit (Human IL-6 immunoassay, cat. no. D6050; Human IFN-γ immunoassay, cat. no. DIF50), according to the manufacturer's protocol (R&D Systems, Inc., Minneapolis, MN, USA).

At the same time, 1×10⁶ PBMCs were cultured in 1 ml RPMI-1640 medium containing 10% heat-inactivated FCS and 0.075%/vol of Staphylococcus aureus Cowan1 (SAC; Sigma-Aldrich; Merck KGaA) for 24 h in 5% CO₂ at 37°C. PBMCs (1×10⁶) were then seeded into a 96-well plate in the presence of SAC. Following cultivation, IL-12 in supernatant was measured using a Quantikine ELISA kit (Human IL-12 immunoassay, cat. no. DP400) according to the manufacturer's protocol (R&D Systems, Inc.). Absorbance at 450 nm was measured using an ELISA micro plate reader.

Lymphocyte proliferation assays were performed using A total of 1×10⁶ cells/ml were suspended in RPMI-1640 medium containing 20% fetal bovine serum (FBS; Kojin Bio Co., Ltd., Saitama, Japan). Following the addition of 100 µg/ml PHA into the PBMC culture wells, the culture was kept at 37°C in a 5% CO₂ atmosphere and PHA mitogenesis was observed 3 times for 80 h by a light microscope (magnification, ×10). A total of 10 µl ³H-thymidine (Japan Radioisotope Association, Tokyo, Japan) was added to the wells (18.5 Beq/well) for the last 72 h of incubation. Cells were harvested using an auto cell harvester with a UniFilter Plate (Bio Tec Co., Ltd. Tokyo, Japan) and ³H-thymidine incorporation was determined using a liquid scintillation counter (PerkinElmer, Inc., Waltham, MA, USA) and was expressed as counts per minute (cpm). The stimulation index (SI) was obtained by calculating total cpm/control cpm. The controls were PBMCs that had not been subjected to PHA addition.

All statistical analyses were performed using Excel Statics 2012 software for Windows (Social Survey Research Information Co., Ltd. Tokyo, Japan). The differences between groups were determined using Student's t-tests. Relationships between two variables were quantified using Spearman's rank correlation coefficient tests. All data are presented as the mean ± standard deviation. P<0.05 was considered to indicate a statistically significant difference. When measuring overall survival (OS), follow-up data that were not reached as of the last follow-up date or at 1,500 days were censored. The prognoses of the patients were analyzed using Kaplan-Meier method curves and the log-rank test was used to determine the univariate significance of the variables. In 110 patients with preoperative breast cancer, multivariate Cox regression analysis of survival was performed according to MDSC levels, tumor size, molecular subtype of the tumor, Ki-67 status, lymph node status, and histological grading in accordance with NCCN Clinical Practice Guidelines in Oncology (18). A Cox proportional-hazards model was used to simultaneously examine the effects of multiple covariates on survival. The effect of each individual variable was described by the hazard ratio, with a 95% confidence interval.

MDSCs were successfully detected by flow cytometry. Representative flow cytometric data was obtained for normal healthy volunteers (Fig. 1A), preoperative patients (Fig. 1B), postoperative patients (Fig. 1C), patients with recurrent breast cancer (Fig. 1D), and patients receiving gemcitabine (Fig. 1E). The percentage of MDSCs in the PBMCs of 177 samples from 155 patients with breast cancer and from 18 healthy volunteers was analyzed. The percentage of MDSCs in the PBMCs from 110 preoperative patients with breast cancer was significantly increased compared with the 18 healthy volunteers (2.27±1.39 and 1.11±0.54%, respectively; P<0.05; Fig. 2). Of the 110 preoperative patients, the MDSC levels of patients with stage I, stage II, stage III, and stage IV diseases were 1.94±1.23, 2.91±0.35, 1.26±0.41 and 2.98±0.40%, respectively, and those of stage I, stage II, and stage IV were significantly higher than the levels of healthy volunteers (Fig. 2). The MDSC levels of 23 postoperative patients, 22 recurrent patients, and 22 recurrent patients who received chemotherapy were 1.31±0.31, 1.91±2.08, and 1.21±0.52%, respectively (Fig. 2). The MDSC levels were also significantly higher in recurrent patients prior to chemotherapy treatment compared with in healthy volunteers (P<0.05; Fig. 2). The MDSC levels of postoperative patients were significantly lower than those of preoperative patients (P<0.05; Fig. 2) and were equivalent to the range in healthy volunteers. The MDSC levels of the patients with recurrent breast cancer following chemotherapy were lower than those in patients with recurrent breast cancer prior to chemotherapy and were not significantly different from the levels in healthy volunteers. In summary, the MDSC levels were increased in preoperative patients, decreased following removal of the tumor, and increased again with recurrence. In addition, MDSC levels of patients with recurrent breast cancer decreased following chemotherapy (Fig. 2).

The levels of MDSCs in preoperative patients were correlated with immune suppression, in particular Th2 polarization, malnutrition and inflammation. Regarding Th2 polarization, the MDSC levels of preoperative patients were significantly positively correlated with IL-6 production (r=0.49, P<0.01; Fig. 3A) and were significantly inversely correlated with IFN-γ (r=−0.49, P<0.01; Fig. 3B) and IL-12 (r=−0.46, P<0.01; Fig. 3C) production. Regarding malnutrition, the MDSCs levels of preoperative patients were significantly negatively correlated with the short turnover protein: retinol binding protein (r=−0.53, P<0.01; Fig. 4A), prealbumin (r=−0.59, P<0.01; Fig. 4B), and transferrin (r=−0.52, P<0.01; Fig. 4C). In terms of inflammation and the SI, the MDSC levels of preoperative patients were significantly positively correlated with the neutrophil/lymphocyte ratio (NLR; r=0.54, P<0.01; Fig. 5A) and C-reactive protein (CRP; r=0.59, P<0.01; Fig. 5B) and were significantly negatively correlated with the SI (r=−0.50, P<0.01; Fig. 5C).

The preoperative patients were separated into 2 groups based on whether their circulating MDSC levels were higher or lower than 1.0% of total PBMCs. The OS of these 2 groups was then compared. In patients with MDSC levels >1.0%, the OS of stage IV disease was significantly shorter than that of stage I, II, or III disease (P<0.01; Fig. 6A), and no difference in survival was observed between stages I, II, and III. On the other hand, there were no differences in OS between stage I–IV breast cancer patients with MDSC levels <1.0% (Fig. 6B). The OS of stage IV patients with MDSC levels that were >1.0% was significantly shorter than that of patients with MDSC levels <1.0% (P<0.05; Fig. 7).

It has previously been reported that gemcitabine treatment of tumor-bearing mice significantly inhibited tumor growth, reduced splenomegaly, and significantly decreased the proportion of MDSCs in the spleen (19). Patients with recurrent breast cancer in the present study had received 6 cycles of gemcitabine (1,000 mg/m²). It was therefore determined whether gemcitabine therapy had altered MDSC levels in these patients. MDSC levels were analyzed in these patients prior to chemotherapy and 14 days following the last cycle of chemotherapy. There was a significant decrease in MDSC levels following chemotherapy compared with those prior to chemotherapy (1.91±1.50 and 1.18±0.71%, respectively; P=0.01; Fig. 8).

A 69-year-old female who had undergone mastectomy and axillary dissection two years previously was revealed to have new lung metastasis and pleural effusion. MDSCs were detected in the pleural effusion using the same experimental procedure as that described above for PBMCs. The effect of chemotherapy on MDSC levels was then examined in the pleural effusion of this patient. MDSC levels in the pleural effusion were decreased from 34.5% prior to chemotherapy (Fig. 9A) to 11.0% following 6 cycles of gemcitabine treatment (Fig. 9B). The MDSC levels in the peripheral blood were also decreased following chemotherapy in this case. Thus, the level of MDSCs in the pleural effusion and in the PBMCs decreased in parallel in response to chemotherapy.

The present study then examined whether the prognostic value of an MDSC level of >1.0% remained statistically significant in a multivariate analysis in the total cohort. A multivariate Cox regression model, which included a level of MDSCs <1.0%, tumor size, molecular subtype of the tumor, including Luminal A, Luminal B, human epidermal growth factor receptor 2 or triple negative breast cancer, Ki-67 status, lymph node status and histological grade was applied. Apart from an MDSC level of <1.0%, none of the other characteristics were significantly and independently prognostic in this model (Table I).