A total of 100 consecutive patients with MM who underwent ASCT as part of a front-line treatment at our institution between January 2013 and December 2016 were enrolled in this analysis. General ASCT procedures are summarized in the supplemental data (Additional file 1) [18].

Blood samples for the analysis of MDSC frequency were collected at diagnosis and pre- and post-ASCT. Pre-ASCT sampling was performed before conditioning chemotherapy, and post-ASCT sampling was done one day after neutrophil engraftment. PBMCs were freshly isolated from whole blood (30 mL) and were processed immediately for flow cytometric analysis.

MDSCs were phenotypically divided into two categories, M-MDSCs and E-MDSCs. E-MDSCs immunophenotyped as the HLA-DR⁻Lin⁻ CD11b⁺CD33⁺ population and M-MDSCs as the HLA-DR⁻CD14⁺ population were quantitated as a percentage of PBMCs (Additional file 4: Figure S1). Monoclonal antibodies for the identification of E- and M-MDSCs and isolation of MDSCs from PBMCs are summarized in the supplemental data (Additional file 1).

One microgram of total RNA was reverse transcribed into cDNA. Quantitative assessment of target mRNA levels was performed by real-time PCR with a CFX96 Real-Time PCR Detection System (Bio-Rad, Hercules, CA, USA). Primer sequences were as previously described (Additional file 2: Table S1) [19].

MDSCs and T cells were isolated from PBMCs of MM patients. Isolated MDSCs were cocultured with CFSE-labelled autologous T cells (MDSC:T cell ratio 1:1). T cell stimulation was provided by 2 μg/ml of anti-CD3/CD28 (eBioscience, San Diego, CA, USA) and 5 ng/ml of recombinant human IL-2 (R&D Systems, Minneapolis, MN, USA). After five days of incubation, the cells were stained with anti-CD4, anti-CD8, and anti-CD56 (eBioscience). Proliferation of T cells was analysed using LSRII (BD Pharmingen, San Jose, CA, USA) and Flowjo software (Ashland, OR, USA).

CFSE-labelled IM-9, RPMI 8266, OPM2 cell lines and primary MM cells were cultured with or without isolated MDSCs (MM cell:MDSC ratio 1:1) in the presence of human M-CSF. The cocultured CFSE-positive cells were then incubated with or without 10 uM melphalan and 500 nM BLZ945 (Additional file 1). After incubation for 48 h, the cells were harvested, stained with Annexin V-APC and propidium iodide (PI), and examined by flow cytometry. Data obtained from flow cytometry were analysed using Flowjo software.

RNA extraction, cDNA library preparation, and bioinformatics analysis of the sequencing data are summarized in the supplemental data (Additional file 1).

OS from transplantation was defined as the time from ASCT to death from any cause, and surviving patients were censored at the last follow-up. PFS was measured as the time from ASCT to disease progression or death (regardless of cause), whichever came first. We wanted to observe the effect of circulating MDSCs on disease progression after ASCT. Therefore, time to progression (TTP) was calculated as time from ASCT to disease progression, with deaths due to causes other than progression censored. Statistical analyses are summarized in the supplemental data (Additional file 1).

A total of 100 patients, 59 males and 41 females, with a median age of 56 years (range, 33–67 years) were analysed in this study (Additional file 3: Table S2). Median disease duration before ASCT was 7.0 months (range, 2.9–12.3 months). The International Staging System (ISS) stages II, II, and III at diagnosis comprised 29, 44, and 23% of subjects, respectively, with 4% unknown ISS) [20]. After induction chemotherapy, 43 (43%), 35 (35%), and 22 (22%) patients had complete response, very good partial response (VGPR), and PR, respectively. The median follow-up was 36 months (95% CI, 30.6–42.5) for survivors. A total of 14 (14%) patients died, and 42 (42%) patients had disease progression. The 3-year OS and PFS were 84.8 ± 4.6% and 42.2 ± 6.3%, respectively (median OS and PFS were not reached and 26.6 months, respectively), and the 3-year TTP was 43.8 ± 6.3% (median TTP was 26.6 months).

Figure 1a shows serial changes in MDSC phenotypes through induction chemotherapy and ASCT. At diagnosis, absolute number of E-MDSC phenotype was 0.9 ± 0.2 × 10⁶/L, which significantly increased to 2.4 ± 0.3 × 10⁶/L (P = 0.002) after induction chemotherapy. In contrast, absolute number of M-MDSC phenotype was significantly decreased after induction chemotherapy, from 31.6 ± 6.0 × 10⁶/L at diagnosis to 21.3 ± 4.6 × 10⁶/L (P < 0.001). When absolute numbers of pre-and post-ASCT MDSC phenotypes were compared, there was no difference in E-MDSC phenotypes (P = 0.757), whereas M-MDSC phenotype increased after ASCT (P < 0.001). The frequency of E-MDSC phenotypes at time of diagnosis was not significantly different among the three groups divided by the ISS, whereas a higher frequency of M-MDSCs was significantly associated with a higher ISS stage (Additional file 5: Figure S2).Fig1Fig. 1

Clinical relevance of MDSCs during induction chemotherapy and ASCT. Serial changes in MDSC phenotypes through induction chemotherapy and ASCT (a). The data are presented as the mean ± SEM. *P < 0.05; **P < 0.01; ***P < 0.001. The 100 patients were grouped (low versus high) according to median frequency value of each E- (0.21 for pre-ASCT, 0.85 for post-ASCT) and M-MDSC phenotype (0.15 for pre-ASCT, 1.04 for post-ASCT). The 3-year time to progression (TTP) between the low and high pre-ASCT E-MDSC groups (b, top left) and M-MDSC groups (b, top right). The 3-year TTP according to post-ASCT MDSC phenotype groups are shown at the bottom

Next, we evaluated how each MDSC phenotype in both pre- and post-ASCT correlated with the 3-year TTP. The patients were grouped according to median frequency value of each MDSC phenotype. First, the association of pre-ASCT MDSCs with the 3-year TTP was analysed and showed that there was no difference between the high and low E-MDSCs groups (52.2% vs. 32.3%, P = 0.352) (Fig. 1b, top left). In contrast to the E-MDSCs groups, the 3-year TTP was significantly lower in the high M-MDSC group compared with the low M-MDSC group (34.3 vs. 52.9, P = 0.049, Fig. 1b top right). Second, we also analysed the effect of post-ASCT MDSCs on the 3-year TTP, which showed that neither E- or M-MDSC phenotype correlated with the 3-year TTP (Fig. 1b, bottom). Ultimately, after adjusting for potential risk factors (immunoglobulin type and serum calcium level at diagnosis), multivariate analysis revealed that the high M-MDSC group pre-ASCT was associated with a lower TTP, with an HR of 0.49 (95% CI, 0.24 to 0.99, P = 0.045) (Table 1).Tab1

Predictive factors for time to progression

To investigate the functional characterization of each MDSC phenotype in both pre- and post-ASCT, we isolated E- and M-MDSC phenotypes from patients’ PBMCs collected pre- and post-ASCT. And then, we tested autologous T- and NKT-cell suppression mediated by each MDSC phenotype (Fig. 2). Both pre-ASCT E- and M-MDSC phenotypes had similarly suppressed autologous T- and NKT- cell proliferation. In contrast, E- and M-MDSC phenotypes post-ASCT did not show suppressive effects on autologous T- and NKT- cells, which indicates these cells are not MDSCs but rather monocytes. It has been shown that MM-associated macrophages protect MM cells from chemotherapy drug-induced apoptosis in vitro [21]. M2-polarized macrophages also mainly upregulate CD200R and CD206 and downregulate CD14 [22]. CD200R and CD206 were expressed in M2 macrophages but not in pre-transplant isolated MDSC phenotypes (Additional file 6: Figure S3).Fig2Fig. 2

Suppressive function of pre- and post-ASCT MDSC phenotypes. We isolated E- and M-MDSC phenotypes from six patients’ PBMCs collected pre- and post-ASCT and tested autologous CD4, CD8 T-, and NKT-cell suppression mediated by each MDSC phenotype. The top figures are representative and individual data from independent experiments using MDSC phenotypes isolated from the six patients, as shown in the bottom figure. Both pre-ASCT E- and M-MDSC subsets had similarly suppressed autologous CD4 (left), CD8 T- (middle), and NKT-cell (right) proliferation. In contrast, post-ASCT E- and M-MDSC phenotypes did not show suppressive effects on those immune cells. The data are presented as the mean ± SEM. **P < 0.01; ***P < 0.001

To better understand the mechanisms leading to poor prognosis mediated by the pre-ASCT, but not post-ASCT, M-MDSC phenotype, we tested the influence of E- and M-MDSCs isolated from patient PBMCs on in vitro melphalan-induced cytotoxic assay according to time before and after ASCT. First, in the test using the MM cell line (IM-9) (Fig. 3a), pre-ASCT M-MDSCs inhibited melphalan-induced cytotoxic effects more strongly than pre-ASCT E-MDSCs. In contrast, isolated cells of M- and E-MDSC phenotype post-ASCT did not have any inhibitory effect on melphalan-induced cytotoxic activity. Next, primary CD138⁺ cells taken from patients’ BM were examined for melphalan-induced cytotoxicity in the presence of E- and M-MDSCs isolated from another patient at the time before and after ASCT (Fig. 3b). Similarly, pre-ASCT M-MDSCs were capable of reducing the cytotoxic activity of melphalan on primary myeloma cells more strongly than pre-ASCT E-MDSCs, whereas the post-ASCT MDSC phenotypes did not show the inhibitory effect.Fig3Fig. 3

The influence of pre- and post-ASCT MDSC phenotypes on in vitro melphalan-induced cytotoxic assay. MM cell line, IM-9 cells (a) or primary MM cells (b) were cultured with or without MDSCs isolated from pre- and post-ASCT samples (MM cell:MDSC ratio 1:1) in the presence of human M-CSF. The top figures are representative staining with Annexin V-APC and PI after incubation with or without melphalan. In the bottom figure, individual data from independent melphalan-induced cytotoxic assay by E- and M-MDSC phenotypes isolated from five patients were compared. The label of post-ASCT MDSCs on the figure means the cells expressing each MDSC phenotype. The data are presented as the mean ± SEM. *P < 0.05; **P < 0.01; ***P < 0.001

As we found a negative impact of pre-ASCT M-MDSCs on TTP and in vitro melphalan-induced cytotoxicity, we were interested in which genes were differentially expressed between pre-ASCT E- and M-MDSCs. Using transcriptome resequencing, we analysed KEGG pathways for 533 differentially expressed genes between E- and M-MDSC populations using a threshold of a 2-fold change and P-value < 0.05. We found that the most remarkable difference was osteoclast differentiation in pre-ASCT M-MDSCs versus E-MDSCs (Fig. 4a). In contrast, no difference in expression of osteoclast differentiation was observed between post-ASCT E- and M-MDSC phenotypes (Fig. 4b). Next, we investigated the differentially expressed genes associated with osteoclast differentiation. Among them, CSF1R was a highly expressed gene in pre-ASCT M-MDSCs compared to other phenotypes of MDSCs (Fig. 4c). These results were confirmed using qRT-PCR in isolated peri-ASCT E- and M-MDSC phenotypes. mRNA expression of CSF1R was much higher in pre-ASCT M-MDSCs (n = 9) than pre-ASCT E-MDSCs (n = 8) (P < 0.001; Fig. 4d, left), whereas there was no difference between post-ASCT M- (n = 11) and E-MDSC phenotypes (n = 12) (Fig. 4d, right).Fig4Fig. 4

Transcriptome profiling analysis of isolated E- and M-MDSC phenotypes. The top 20 KEGG pathways for 533 differentially expressed genes between pre-ASCT E- and M-MDSC populations (a) and for 65 differentially expressed genes between post-ASCT E- and M-MDSC phenotypic populations (b), using a threshold of a 2-fold change and P-value < 0.05. The most remarkable difference was osteoclast differentiation in pre-ASCT M- versus E-MDSCs, which was not observed in post-ASCT M- versus E-MDSC phenotypes. Among the genes associated with osteoclast differentiation, CSF1R was the most significant (c) and was confirmed using qRT-PCR in isolated peri-ASCT E- and M-MDSC phenotypes (d). The data are presented as the mean ± SEM. *P < 0.05. Next, M-CSF and IL-34, which are known to trigger CSF-1R signalling in patient sera (n = 75 for M-CSF, n = 82 for IL-34), were measured, and the correlation between these factors and the frequency of pre-ASCT (e) and post-ASCT (f) MDSC phenotypes was analysed. The Spearman correlation coefficient was used to evaluate association for continuous variables. The label of post-ASCT MDSCs on the figure means the cells expressing each MDSC phenotype

Next, we measured factors such as M-CSF and IL-34, which are known to trigger CSF-1R signalling in patient sera (n = 75 for M-CSF, n = 82 for IL-34) [23]. In correlation analysis between these factors and the frequency of pre-ASCT MDSCs, M-CSF was correlated with M-MDSCs frequency (R² = 0.1049, P = 0.005), verifying that M-CSF enhanced M-MDSCs proliferation (Fig. 4e, top left), but IL-34 did not correlate with M-MDSCs (Fig. 4e, top right). On the other hand, there was no relationship between these factors and the frequency of pre-ASCT E-MDSCs (Fig. 4e, bottom). Furthermore, neither post-ASCT E- (Fig. 4f, top) nor M-MDSC phenotype (Fig. 4f, bottom) was related to level of M-CSF or IL-34.

Finally, to determine whether a CSF1R inhibitor can recover melphalan-induced cytotoxicity attenuated by pre-ASCT M-MDSCs, we examined the influence of BLZ945, a human CSF1R inhibitor, on cell death induced by melphalan (Fig. 5a). Presence of CSF1R inhibitor reversed the protective effect of pre-ASCT M-MDSCs on IM-9 cells. However, BLZ945-treated pre-ASCT E-MDSCs did not affect survival of the MM cells. The effect of post-ASCT E- and M-MDSC phenotypes on melphalan-induced cytotoxicity was not affected by BLZ945 treatment. Similar results were obtained using the RPMI 8226 and OPM2 cell lines (Additional file 7: Figure S4). Taken together, these results demonstrate that inhibition of CSF1R signalling results in recovery of anti-MM activity by melphalan, which is attenuated by pre-ASCT M-MDSCs. In addition, we measured several cytokines in the culture supernatants and compared them according to BLZ945 treatment because cytokines are major proliferative factors for malignant plasma cells. Although the concentrations of IL-6, IGF1, and VEGF with pre-ASCT M-MDSCs were higher than those with pre-ASCT E-MDSCs, BLZ945 treatment did not have an effect on their concentrations in the presence of pre-ASCT M-MDSC phenotypes (Fig. 5b, top). Concentrations of those cytokines in culture supernatants were not changed in the presence of post-ASCT MDSC phenotypes (Fig. 5b, bottom). Importantly, only M-CSF concentration was significantly decreased in the culture supernatants with pre-ASCT M-MDSCs after BLZ945 treatment (Fig. 5c).Fig5Fig. 5

The influence of CSF1R inhibition on melphalan-induced cytotoxicity attenuated by pre-ASCT M-MDSCs. The influence of BLZ945, a human CSF1R inhibitor, on cell apoptosis induced by melphalan was tested (a). IM-9 cells were cultured with or without MDSCs isolated from pre- and post-ASCT patients (MM:MDSC ratio 1:1) in the presence of human M-CSF, as shown in Fig. 3. The top figures are representative staining with Annexin V-APC and PI after incubation with vehicle, 10 uM melphalan with or without 500 nM BLZ945. In the bottom figure, individual data from independent experiments by E- and M-MDSC phenotypes isolated from five patients were compared. (b) Cytokines (IL-6, IGF1, VEGF, and M-CSF) in culture supernatants with pre- and post-ASCT MDSC phenotypes were measured, and the effects of BLZ945 treatment were compared. The concentrations of IL-6, IGF1, and VEGF in culture supernatants with pre-ASCT MDSCs (top) and post-ASCT MDSC phenotypes (bottom) are shown. (c) M-CSF concentrations in culture supernatants with pre-ASCT MDSCs (left) and post-ASCT MDSC phenotypes (right) are shown. MDSC phenotypes were isolated from six patients. The label of post-ASCT MDSCs on the figure means the cells expressing each MDSC phenotype. The data are presented as the mean ± SEM. *P < 0.05; **P < 0.01; ***P < 0.001