Circulating M‐MDSC Expansion During Therapy Associates With Treatment Response in Multiple Myeloma: A Longitudinal Observational Study

ABSTRACT

Dynamic changes in myeloid‐derived suppressor cells (MDSCs) during multiple myeloma (MM) treatment remain poorly understood. In this longitudinal observational study, circulating monocytic (M‐MDSCs) and polymorphonuclear MDSCs (PMN‐MDSCs) were measured by flow cytometry in 99 newly diagnosed MM patients at diagnosis, post‐induction, and Day 100 after autologous stem cell transplantation (ASCT). Primary analyses were restricted to the bortezomib‐lenalidomide‐dexamethasone (VRd)‐treated cohort (n = 78) to avoid regimen‐specific confounding. VRd, but not daratumumab‐based regimens, significantly expanded M‐MDSCs (median Δ +0.98%, p < 0.0001), which normalized after ASCT. In the VRd cohort, ΔM‐MDSCs (post‐induction minus baseline) were higher in patients achieving ≤ partial response (PR) than in those with ≥ very good partial response (VGPR) (median +1.56% vs. +0.50%, p = 0.0083). In univariable analysis, ΔM‐MDSCs, R2‐ISS stage, and MRD positivity each met the p < 0.20 threshold; however, MRD was excluded from multivariable modeling due to collinearity with response. In a two‐variable model adjusted for R2‐ISS, ΔM‐MDSCs were not an independent predictor of suboptimal response (OR 1.28 per 1% increase, 95% CI 0.95–1.74, p = 0.109), and bootstrap validation yielded confidence intervals overlapping unity. Receiver operating characteristic analysis showed an AUC of 0.730, but the derived cutoff was not used for modeling. VRd induction is associated with transient M‐MDSC expansion that correlates with response depth, but ΔM‐MDSCs did not retain independent predictive value after adjustment. These hypothesis‐generating findings require validation in larger independent cohorts.

Study Highlights

• What is the current knowledge on the topic?

  • ○
    The role of myeloid‐derived suppressor cells (MDSCs) in multiple myeloma (MM) immunosuppression is established, but their kinetic changes throughout induction therapy and autologous stem cell transplantation (ASCT) and their potential association with treatment response have not been fully characterized in an observational framework.
• What question did this study address?

  • ○
    In this observational study, we explored whether the change in circulating monocytic MDSCs (M‐MDSCs) from diagnosis to post‐induction (ΔM‐MDSCs) is associated with the depth of treatment response in newly diagnosed MM patients.
• What does this study add to our knowledge?

  • ○
    This exploratory analysis suggests that a larger increase in ΔM‐MDSCs may be associated with a lower likelihood of achieving a deep response (≥ VGPR) in the VRd‐treated subset. Notably, M‐MDSCs expanded during VRd induction despite tumor reduction, yet returned to baseline levels after ASCT—an observation that warrants further investigation in independent cohorts.
• How might this change clinical pharmacology or translational science?

  • ○
    These findings highlight the distinct immunomodulatory patterns associated with induction therapy and ASCT. The post‐ASCT “immune reset” observed here generates the hypothesis that timing immunotherapeutic interventions (e.g., CAR‐T, bispecific antibodies) after ASCT, when the immune microenvironment appears less immunosuppressed, could be advantageous. This hypothesis requires prospective evaluation in dedicated clinical trials.

1. Introduction

Multiple myeloma (MM) accounts for approximately 1% of all malignancies and 10% of hematologic cancers, ranking as the second most common blood cancer after lymphoma [1]. MM is a highly heterogeneous disease characterized by frequent relapse and remains incurable despite substantial therapeutic progress. The incidence of MM in China has risen steadily in recent decades. Current standard therapies include immunomodulatory drugs (IMiDs), proteasome inhibitors (PIs), and autologous stem cell transplantation (ASCT), while allogeneic transplantation (allo‐HSCT) is less frequently applied. Emerging immunotherapies—including monoclonal antibodies, bispecific antibodies, immune checkpoint inhibitors, antibody–drug conjugates, and chimeric antigen receptor (CAR) T‐cell therapy—have further improved outcomes [2], leading to debate about the continued necessity of ASCT. Nevertheless, clinical evidence consistently shows that ASCT deepens response and prolongs survival, and it remains a standard of care for transplant‐eligible newly diagnosed MM (NDMM) patients [3].

Immune dysregulation is increasingly recognized as a key driver of MM progression and treatment resistance [4]. Within the immunosuppressive bone marrow microenvironment, myeloid‐derived suppressor cells (MDSCs) play a central role by inhibiting T‐cell cytotoxicity and promoting immune escape [5]. Although elevated baseline MDSC frequencies have been linked to advanced disease stage and poor prognosis [6, 7, 8, 9, 10, 11], their dynamic changes throughout standard therapy—particularly across the ASCT continuum—remain poorly characterized, and existing longitudinal data are limited in scope.

Although ASCT profoundly reconstitutes adaptive immunity, its effects on the myeloid compartment remain largely undefined [12, 13]. Similarly, the impact of induction therapy on MDSCs is controversial. Some studies have reported an expansion of MDSCs following bortezomib‐ or IMiD‐based regimens, implying a compensatory immunosuppressive response [9, 14], whereas others have shown reduced monocytic MDSCs (M‐MDSCs) or inhibitory effects of IMiDs on MDSCs expansion in vitro [6, 11, 15]. These inconsistencies highlight unresolved questions regarding the direction and clinical significance of MDSCs modulation during therapy. Moreover, longitudinal studies tracking MDSCs kinetics from diagnosis through induction and into immune reconstitution after ASCT—particularly in Chinese patients—remain lacking. Whether treatment‐driven fluctuations in MDSCs have predictive value for therapeutic response also remains unknown.

Several longitudinal or semi‐longitudinal studies have begun to explore MDSC dynamics in MM, but key gaps remain. Lee et al. tracked M‐MDSCs around first‐line ASCT but did not include later post‐transplant follow‐up [6]. Bae et al. conducted a cross‐sectional comparison of M‐MDSCs and PMN‐MDSCs across response groups, without capturing within‐patient kinetics from baseline [7]. Tyrinova et al. provided the most comprehensive longitudinal data to date, tracking MDSCs subsets from diagnosis through 12 months post‐ASCT and comparing graft content with peripheral blood. However, the association between the magnitude of MDSCs change during induction (ΔM‐MDSCs) and the depth of induction response was not systematically quantified [8]. Consequently, the integrated kinetic profile of MDSCs across the entire ASCT continuum—from diagnosis through induction to post‐transplant immune reconstitution—remains incompletely defined. Furthermore, none of these studies specifically addressed the Chinese NDMM population, in which disease biology, treatment practices, and host immune backgrounds may differ. The question of whether therapy‐induced changes in MDSCs carry independent prognostic information, or instead primarily reflect a pharmacodynamic response to treatment, also remains unresolved.

To address these gaps, we conducted a longitudinal study using standardized flow cytometry [9] to quantify circulating MDSC subsets at diagnosis, post‐induction, and Day 100 post‐ASCT in a Chinese NDMM cohort. We aimed to define the kinetic profile of the monocytic (M‐MDSCs) and polymorphonuclear (PMN‐MDSCs) subsets and to evaluate the association between the pre‐ASCT change in M‐MDSCs (ΔM‐MDSCs, post‐induction minus baseline) and the depth of treatment response.

2. Materials and Methods

2.1. Patients and Sample Selection

This observational study enrolled 99 treatment‐naïve patients with MM from the Department of Hematology, Beijing Jishuitan Hospital, Capital Medical University, between June 2022 and May 2025. Detailed inclusion and exclusion criteria for patient enrollment are provided in Table S1. All patients were diagnosed, staged, and assessed for treatment response according to the International Myeloma Working Group (IMWG) consensus criteria [16]. Patients with active infections at diagnosis were excluded. For comparison, 20 age‐ and sex‐matched healthy donors were recruited as controls. Demographic characteristics of the healthy donors are summarized in Table S2. The study protocol was approved by the Institutional Review Board and Ethics Committee of Beijing Jishuitan Hospital (approval No. K2023‐318) on August 29, 2023. Written informed consent was obtained from all participants in accordance with the Declaration of Helsinki.

MDSC subsets were analyzed in peripheral blood using multiparameter flow cytometry (MFC). High‐risk cytogenetic abnormalities were defined as the presence of one or more of the following: 1q21 gain/amplification, del(17p), t(4;14), t(14;16), or t(14;20).

2.2. Flow Cytometric Analysis of MDSCs

The relative frequencies of polymorphonuclear MDSCs (PMN‐MDSCs; CD11b⁺CD14⁻CD15⁺HLA‐DR^−/lowCD33⁺) and monocytic MDSCs (M‐MDSCs; CD11b⁺CD14⁺CD15⁻HLA‐DR^−/lowCD33⁺) were quantified in freshly collected peripheral blood samples by MFC, following consensus recommendations by Bronte et al. [17].

The following anti‐human monoclonal antibodies were used: CD45 (PerCP, Lifangjiye, China), CD14 (APC‐Cy7, Miltenyi Biotec, Germany), CD33 (PE, BD Biosciences, USA), CD11b (APC, BioLegend, USA), CD15 (FITC, BioLegend, USA), and HLA‐DR (AF700, BioLegend, USA). Isotype‐matched antibodies conjugated to corresponding fluorochromes served as negative controls. Isotype controls and fluorescence‐minus‐one (FMO) controls were used to establish negative boundaries and accurately define the HLA‐DR⁻/low population.

The gating strategy (Figure S1) was as follows: nucleated cells (CD45⁺) were identified, HLA‐DR⁻/low cells were selected, and the CD33⁺CD11b⁺ population was gated. Within this population, PMN‐MDSCs and M‐MDSCs were defined as CD15⁺CD14⁻ and CD15⁻CD14⁺ subsets, respectively.

Absolute counts were calculated using the formula:

$$\begin{array}{c}\text{Absolute count}\left(\times {10}^9\text{cells}/\mathrm{\mu L}\right)\hfill \\ =\left(\text{Percentage of MDSCs subset}\times \text{total}\mathrm{WBC}\text{count}\right)/100.\hfill \end{array}$$

2.3. Fluorescence In Situ Hybridization (FISH)

CD138‐positive plasma cells were isolated from bone marrow aspirates of newly diagnosed MM patients using magnetic microbeads (Miltenyi Biotec, Germany). Enriched cells were subjected to hypotonic treatment, fixation, and slide preparation for interphase FISH analysis.

Commercially available probes (Anbiping, Guangzhou, China) were used to detect 17p13 (P53/CSP17), 1q21 (CKS1B), t(11;14) (CCND1/IGH), t(4;14) (FGFR3/IGH), t(14;16) (MAF/IGH), and t(14;20) (MAFB/IGH). Procedures followed the manufacturer's protocols. Signal enumeration and image acquisition were performed using a Zeiss fluorescence microscope.

Cutoff values for positive FISH results were defined as 8% for del(17p13) and 6% for all other probes based on normal reference controls.

2.4. Statistical Analysis

All statistical analyses were performed using SPSS version 21.0 (IBM Corp., Armonk, NY, USA) and GraphPad Prism version 9.5.1 (GraphPad Software, San Diego, CA, USA). Firth's penalized logistic regression was conducted as a sensitivity analysis using R version 4.0.5 (R Foundation for Statistical Computing, Vienna, Austria) with the “logistf” package (version 1.26.1). The Shapiro–Wilk test was used to assess the normality of continuous variables. Data are presented as mean ± standard deviation (SD) for normally distributed variables, or as median and interquartile range (IQR) for non‐normally distributed variables. Categorical variables are summarized as frequencies and percentages.

Between‐group comparisons were conducted using the Mann–Whitney U test or the Kruskal–Wallis test for continuous variables, and the χ ² test or Fisher's exact test for categorical variables, as appropriate. Paired comparisons between time points were analyzed using the Wilcoxon signed‐rank test or the paired t‐test, depending on the distribution of the data.

2.5. Transparency Statement Regarding the Analytic Plan

This was a retrospective observational study without a preregistered protocol; all analyses are therefore exploratory and post hoc in nature. The initial manuscript submission included a saturated multivariable model with a data‐derived dichotomized ΔM‐MDSCs threshold. In response to peer review, the primary analysis strategy was revised to restrict to the VRd‐treated cohort (n = 78), treat ΔM‐MDSCs as a continuous variable, employ univariable screening (p < 0.20) for covariate selection, and construct a parsimonious two‐variable logistic regression model (ΔM‐MDSCs plus R2‐ISS stage) with bootstrap internal validation (1000 resamples). Additionally, at the editor's request, Firth's penalized logistic regression was added as a further sensitivity analysis to assess potential small‐sample bias. The receiver operating characteristic (ROC) analysis was part of the original exploratory plan, but the derived optimal cutoff was not used in the primary multivariable model to avoid overfitting.

Given the limited number of patients with a suboptimal response (≤ PR, n = 13 in the VRd cohort) and the strong confounding effect of the daratumumab‐based regimen (n = 21, of whom 3 had ≤ PR), the primary multivariable analysis was restricted to the VRd‐treated cohort to ensure a homogeneous treatment background and to avoid model instability. To minimize overfitting, a two‐step approach was employed: first, univariable logistic regression was performed for each candidate predictor; second, a parsimonious binary logistic regression model was constructed. To internally validate the stability of the estimates, bootstrap resampling with 1000 iterations was applied, and bias‐corrected 95% confidence intervals were reported. As an additional sensitivity analysis, Firth's penalized likelihood logistic regression was performed to address potential small‐sample bias.

Receiver operating characteristic (ROC) curve analysis was performed to explore the potential discriminative capacity of ΔM‐MDSCs for response; however, the derived threshold was not used for multivariable adjustment in the primary analysis.

All statistical tests were two‐sided, and p < 0.05 was considered nominally significant. No formal adjustment for multiple comparisons was performed; therefore, all results should be interpreted as exploratory signals and not as confirmatory evidence.

3. Results

3.1. Baseline Characteristics of NDMM Patients

A total of 99 patients with newly diagnosed multiple myeloma (NDMM) were included in the final analysis. A CONSORT‐style patient flow diagram detailing the screening, enrollment, and follow‐up process is presented in Figure S2. The cohort was stratified by frontline therapy: 21 patients received daratumumab‐based (D‐based) regimens, while 78 received bortezomib, lenalidomide, and dexamethasone (VRd). Treatment response was evaluated after three cycles for transplant‐eligible patients and after four cycles for those ineligible for autologous stem cell transplantation (ASCT). In the D‐based group, 18 of the 21 patients (85.7%) achieved at least a very good partial response (≥ VGPR), with the remaining 3 patients (14.3%) achieving a partial response (PR). In the VRd cohort, 65 of the 78 patients (83.3%) achieved ≥ VGPR, and 13 (16.7%) achieved ≤ PR. Ultimately, 46 patients proceeded to and completed ASCT, while 53 did not undergo transplantation. The baseline demographic, clinical, and disease characteristics of the entire cohort are summarized in Table 1.

TABLE 1.

Baseline clinical and laboratory characteristics of newly diagnosed multiple myeloma patients.

Characteristics	All patients (n = 99) (%)
Male sex, M/F (%)	57 (57.6)/42 (42.4)
Age at diagnosis, median [IQR]	58 (55–63)
Serum M‐protein
Ig G kappa/Ig G lambda	26 (26.3)/13 (13.1)
Ig A kappa/Ig A lambda	13 (13.1)/6 (6.1)
Light chain, kappa/lambda	17 (17.2)/16 (16.2)
Others	8 (8.1)
BM monoclonal plasma cells, median [IQR]	6.985 (1.49–17.0225)
Circulating plasma cells ≥ 0.07%, n (%)	32 (32.3)
Circulating plasma cells ≥ 0.2%, n (%)	4 (4.0)
Cytogenetics
1q21	50 (51.5)
P53	13 (13.1)
t(4;14)	14 (14.1)
t(11;14)	25 (25.3)
t(14;16)	1 (1.0)
ISS stage, n (%)
I/II/III	38 (38.4)/30 (30.3)/31 (31.3)
RISS stage, n (%)
I/II/III	27 (27.3)/61 (61.6)/11 (11.1)
R2‐ISS stage, n (%)
I/II/III/IV	14 (14.1)/31 (31.3)/48 (48.5)/6 (6.1)
β2m, median [IQR] (mg/L)	3.86 (2.745–6.9)
Correction Calcium, median [IQR] (mmol/dL)	2.272 (2.179–2.415)
Hemoglobin
Low hemoglobin, n (%)	69 (69.7)
Hemoglobin, median [IQR] (g/dL)	109 (87.5–128)
Serum albumin
Low albumin, n (%)	44 (44.4)
Albumin, median [IQR] (g/dL)	41.0 (34.5–43.6)
Creatinine, median [IQR] (mg/dL)	71 (60.5–82)
Serum LDH
High LDH, n (%)	8 (8.1)
LDH, median [IQR] (U/L)	170 (143–203.5)
Induction therapy
Dara‐based/VRD	21 (21.2)/78 (78.8)
Auto‐HSCT
Yes/No	46 (46.5)/53 (53.5)
Response to induction therapy
VRD, ≤ PR/VGPR/CR	13 (16.7)/44 (56.4)/21 (26.9)
Dara‐based, ≤ PR/VGPR/CR	3 (14.3)/10 (47.6)/8 (38.1)

Abbreviations: β2‐MG, β2‐microglobulin; IQR, interquartile range; ISS, international staging system; LDH, lactate dehydrogenase; R‐ISS, revised ISS; R2‐ISS, second revision ISS.

3.2. MM Patients Exhibited Elevated Circulating MDSCs

Compared with healthy donors (HD, n = 20), patients with NDMM demonstrated a significant expansion of circulating myeloid‐derived suppressor cells (MDSCs). Flow cytometric comparison across HD and NDMM patients at diagnosis (Dx, n = 99), post‐induction (PI, n = 99), and Day 100 post‐ASCT (ASCTd100, n = 46) revealed distinct patterns in the two major MDSC subsets: monocytic M‐MDSCs (CD11b⁺CD14⁺CD15⁻HLA‐DR^−/lowCD33⁺) and polymorphonuclear PMN‐MDSCs (CD11b⁺CD14⁻CD15⁺HLA‐DR^−/lowCD33⁺) (Figure S1).

M‐MDSCs frequencies were markedly higher in NDMM patients than in HD (median 0.65% [IQR 0.265–1.13%] vs. 0.28% [0.13–0.59%]; p = 0.0170), further increased after induction (1.23% [0.62–2.26%]; p < 0.0001 vs. HD), and remained elevated at ASCTd100 (0.72% [0.32–1.10%]; p = 0.0236 vs. HD) (Figure 1, Table S3). PMN‐MDSC frequencies showed no significant intergroup differences (all p > 0.05).

FIGURE 1.

Quantification and disease‐associated expansion of circulating MDSCs. (a) Frequencies of monocytic MDSCs (M‐MDSCs) and (b) polymorphonuclear MDSCs (PMN‐MDSCs) in peripheral blood from healthy donors (HD, n = 20) and multiple myeloma patients at diagnosis (NDMM, n = 99), post‐induction (PI, n = 99), and Day 100 post‐autologous stem cell transplantation (ASCTd100, n = 46).

3.3. Dynamic Changes in MDSCs During Induction Therapy and ASCT

To investigate treatment‐specific effects on MDSC kinetics, patients were stratified by induction regimen: VRd (n = 78) and D‐based (n = 21).

As shown in Figure 2a,b, M‐MDSCs displayed distinct regimen‐dependent responses. In the VRd cohort, M‐MDSCs frequency and absolute count significantly increased post‐induction (median Δ +0.585%, p < 0.0001; median Δ +26.75 cells/μL, p < 0.0001). In contrast, D‐based therapy produced no significant changes in M‐MDSCs (median Δ +0.07%, p = 0.34; mean Δ +3.64 cells/μL, p = 0.57).

FIGURE 2.

Induction therapy–specific effects on MDSC subsets. Longitudinal paired changes in (a and b) M‐MDSCs and (c and d) PMN‐MDSCs during VRd and D‐based induction therapy. Box plots show medians and IQRs.

For PMN‐MDSCs (Figure 2c,d), VRd therapy induced significant reductions in frequency (median Δ −2.37%, p = 0.028) and count (median Δ −580.5 cells/μL, p = 0.005), whereas D‐based regimens maintained stable levels (all p ≥ 0.05).

Given the limited and variable PMN‐MDSCs changes, subsequent longitudinal analyses focused on M‐MDSCs in the 46 transplant‐eligible patients. Across diagnosis, post‐induction, and Day 100 post‐ASCT, M‐MDSCs exhibited a characteristic triphasic pattern (Figure 3). Frequencies peaked after induction (median Δ +0.455% vs. Dx, p < 0.0001), declined post‐ASCT (median Δ −0.510% vs. PI, p = 0.0003), and normalized to baseline (median Δ +0.020% vs. Dx, p = 0.97; Figure 3a). Absolute counts followed a parallel trajectory (p < 0.001; Figure 3b). Collectively, these data indicate that ASCT reverses the therapy‐induced M‐MDSCs expansion.

FIGURE 3.

Longitudinal kinetics of M‐MDSCs across diagnosis, post‐induction, and post‐ASCT. (a) M‐MDSC frequencies (% of CD45⁺ cells). (b) Absolute counts (×10⁹cells/μL).

3.4. Variations in M‐MDSCs Correlate With Treatment Response

Given the marked differences in treatment mechanisms between daratumumab‐based and VRd regimens, and the small number of patients in the daratumumab‐based subgroup (n = 21, of whom 3 had ≤ PR), the primary analysis was restricted to the VRd‐treated cohort (n = 78) to avoid treatment‐related confounding and to minimize model instability. Among these patients, 13 (16.7%) achieved ≤ PR and 65 (83.3%) achieved ≥ VGPR. The median ΔM‐MDSCs (post‐induction minus baseline) was significantly higher in the ≤ PR group than in the ≥ VGPR group (median + 1.56% [IQR: 0.28–4.12] vs. +0.50% [IQR: −0.44–1.88]; p = 0.0083, Mann–Whitney U test; Figure 4a).

FIGURE 4.

ΔM‐MDSCs is associated with the depth of response to induction therapy. (a) Validation of the ΔM‐MDSCs association with response depth in the VRd‐treated subgroup (≥ VGPR, n = 65; PR, n = 13). (b) Receiver operating characteristic (ROC) analysis of ΔM‐MDSCs in the VRd cohort. The area under the curve (AUC) is 0.730 (95% CI: 0.593–0.867, p = 0.0027). CR, complete response; VGPR, very good partial response; PR, partial response.

To identify potential independent predictors of suboptimal response (≤ PR), univariable logistic regression was performed for ΔM‐MDSCs (continuous), R2‐ISS stage, age, bone marrow plasma cell percentage, and minimal residual disease (MRD) status. Three variables met the pre‐specified liberal threshold of p < 0.20: ΔM‐MDSCs (OR 1.30, 95% CI: 0.97–1.76; p = 0.082), R2‐ISS stage (OR 0.50, 95% CI: 0.23–1.09; p = 0.083), and MRD positivity (OR 20.50, 95% CI: 2.51–167.62; p = 0.005) (Table S4). MRD was not advanced to the multivariable model because of its intrinsic collinearity with depth of response and to preserve model parsimony given the limited number of events (≤ PR, n = 13).

In the subsequent two‐variable binary logistic regression model incorporating ΔM‐MDSCs and R2‐ISS stage, neither variable retained statistical significance as an independent predictor. The odds ratio for ΔM‐MDSCs (per 1% increase) was 1.28 (95% CI: 0.95–1.74; p = 0.109), and for R2‐ISS stage (per one‐stage increase) was 0.52 (95% CI: 0.24–1.16; p = 0.108). Bootstrap internal validation (1000 resamples) yielded bias‐corrected 95% confidence intervals of 0.96–2.61 for ΔM‐MDSCs and 0.16–1.19 for R2‐ISS, both overlapping unity, confirming the absence of an independent association (Table 2). Firth penalized logistic regression yielded similar estimates: the odds ratio for ΔM‐MDSCs was 1.26 (95% CI, 0.96–1.73, p = 0.091) and for R2‐ISS stage was 0.60 (95% CI, 0.28–1.26, p = 0.176). The likelihood ratio test for the overall model was not significant (χ ² = 5.07, df = 2, p = 0.079). These findings confirm that neither variable was an independent predictor of suboptimal response in this conservative sensitivity analysis (Table S5). These revised multivariable analyses were performed in response to peer review.

TABLE 2.

Multivariable logistic regression for suboptimal response (≤ PR) with bootstrap validation.

Variable	Original OR (95% CI)	Original p value	Bootstrap bias‐corrected OR (95% CI) a
ΔM‐MDSCs (continuous)	1.28 (0.95–1.74)	0.109	0.96–2.61
R2‐ISS (I–IV, continuous)	0.52 (0.24–1.16)	0.108	0.16–1.19

Abbreviation: R2‐ISS, second revision ISS.

^a Bootstrap bias‐corrected 95% confidence intervals were derived from 1000 resamples by exponentiation of the confidence limits for the B coefficients. Confidence intervals crossing 1 indicate a lack of independent statistical significance.

For descriptive purposes, receiver operating characteristic (ROC) analysis of ΔM‐MDSCs in the VRd cohort showed an area under the curve (AUC) of 0.730 (95% CI: 0.593–0.867). However, the optimal cutoff derived from this analysis was not used for multivariable adjustment, in order to avoid overfitting and overestimation of predictive accuracy (Figure 4b). In the daratumumab‐based subgroup (n = 21, three with ≤ PR), no formal comparative analysis was undertaken due to the small event count and the recognized direct effect of anti‐CD38 therapy on MDSC frequencies.

4. Discussion

A pivotal step in elucidating the pathophysiology of MM is deciphering the dynamic interplay between malignant plasma cells and the bone marrow microenvironment. As the disease progresses, this interaction establishes a permissive ecosystem that fosters MM cell proliferation, survival, and the emergence of drug resistance. A hallmark of this dysregulated milieu is the expansion of myeloid‐derived suppressor cells (MDSCs), particularly the monocytic subset (M‐MDSCs), in both the peripheral blood and bone marrow of MM patients [18]. This relationship is bidirectional: MM cells promote the differentiation and activation of MDSCs, which in turn facilitate immune evasion, disease progression, and therapy resistance [19, 20]. Although elevated levels of both PMN‐ and M‐MDSCs have been documented in MM, findings stratified by genetic background, disease stage, or tumor burden remain inconsistent. Considering the inherent complexity of immune regulation, longitudinal analyses within the same individual may yield more biologically meaningful insights than cross‐sectional comparisons.

This longitudinal study comprehensively characterizes the dynamics of circulating MDSCs throughout the autologous stem cell transplantation (ASCT) course in newly diagnosed multiple myeloma (NDMM). Our findings are threefold. First, induction therapy—particularly the bortezomib, lenalidomide, and dexamethasone (VRd) regimen—induced a marked expansion of M‐MDSCs. Second, this therapy‐driven expansion was reversed following ASCT, with M‐MDSCs levels returning to baseline by Day 100 post‐transplant. Third, the magnitude of M‐MDSCs elevation during induction (ΔM‐MDSCs) was associated with inferior response depth in univariable analysis, although this association did not retain independent predictive value after adjustment for established clinical risk factors.

Patients who received daratumumab‐based regimens exhibited no significant change in M‐MDSCs frequency or absolute count following induction, consistent with prior evidence that daratumumab targets CD38‐expressing MDSCs and suppresses their proliferation in vitro [21]. Conversely, the expansion of M‐MDSCs observed after immunomodulatory drug (IMiD)‐based induction appears paradoxical when compared with in vitro models, which often show reduced M‐MDSCs following lenalidomide or bortezomib exposure [6, 8, 11]. Several mechanisms may contribute to this increase. First, previous studies have demonstrated that dexamethasone directly promotes MDSC expansion and enhances their suppressive function [22, 23, 24], which may partly account for the increased M‐MDSCs observed in our VRd cohort. Second, bortezomib may trigger the release of inflammatory signals that recruit and activate MDSCs. It is important to recognize that such models primarily assess direct drug effects, whereas in vivo immune regulation is multifactorial. Our real‐world clinical data are in agreement with recent reports demonstrating increased M‐MDSCs or monocyte‐derived subsets after lenalidomide‐based regimens, suggesting that IMiDs may simultaneously activate both stimulatory and inhibitory immune pathways [9, 14]. This supports the hypothesis that potent immune activation triggers compensatory induction of suppressive cells, including MDSCs and regulatory T cells, to maintain homeostasis. The post‐ASCT decline in M‐MDSCs suggests that high‐dose chemotherapy with stem cell rescue may offer more extensive immunomodulatory effects than simple T‐cell reconstitution. We hypothesize that it could also contribute to the restoration of a broader immune equilibrium [12], though this concept requires prospective validation.

In the VRd‐treated cohort, ΔM‐MDSCs was significantly higher in patients with a suboptimal response (≤ PR) than in those achieving ≥ VGPR (median + 1.56% vs. +0.50%, p = 0.0083), and receiver operating characteristic analysis showed a modest discriminative capacity (AUC = 0.730). However, when we constructed a parsimonious multivariable logistic regression model incorporating ΔM‐MDSCs and R2‐ISS stage, neither variable emerged as a statistically significant independent predictor of suboptimal response. The odds ratio for ΔM‐MDSCs (per 1% increase) was 1.28 (95% CI: 0.95–1.74, p = 0.109), and for R2‐ISS stage was 0.52 (95% CI: 0.24–1.16, p = 0.108). Bootstrap internal validation (1000 resamples) confirmed the instability of these estimates, with bias‐corrected 95% confidence intervals broadly overlapping unity. There are several plausible explanations for this attenuation of the univariable signal. First, our study had limited statistical power, with only 13 suboptimal response events in the VRd cohort available to support a multivariable model. Second, R2‐ISS stage—which integrates ISS, LDH, and high‐risk cytogenetic abnormalities—captures a dominant component of baseline risk that may subsume the prognostic information carried by ΔM‐MDSCs. Third, therapy‐induced MDSCs expansion may function primarily as a marker of immune activation rather than as an independent driver of treatment resistance. In this hypothesized pharmacodynamic role, ΔM‐MDSCs reflects the intensity of the host immune response to VRd, which is itself correlated with—but not mechanistically independent of—disease biology and tumor burden.

A major limitation of our study concerns the quantification of PMN‐MDSCs. In contrast to several earlier studies reporting elevated PMN‐MDSCs in newly diagnosed, relapsed/refractory, or even remission‐stage MM [10, 25, 26, 27, 28], our data did not show significant variation in this subset. A key methodological challenge in MDSC research is the accurate discrimination of PMN‐MDSCs from conventional neutrophils using surface markers alone. Our initial panel, based on widely reported phenotypes [6, 21], confirmed this limitation, as the CD11b⁺CD14⁻CD15⁺HLA‐DR⁻/lowCD33⁺ gate likely included non‐suppressive bystander cells. Without functional suppression assays or the inclusion of additional discriminatory markers such as LOX‐1 or CD10, the identification of bona fide PMN‐MDSCs remains unreliable. Given the clearer phenotypic definition and stronger association with clinical outcomes observed for M‐MDSCs, we focused our analysis on this subset. This strategic focus not only enhances data reliability but also aligns with the goal of developing a clinically feasible assay suitable for routine laboratory implementation.

Additional limitations should be considered. First, the absence of long‐term survival follow‐up precludes determination of whether ΔM‐MDSCs predicts progression‐free or overall survival. Second, no functional assays (e.g., T‐cell suppression or arginase‐1 activity) were performed. Therefore, we cannot definitively prove that the phenotypically defined M‐MDSCs are genuinely immunosuppressive. This limitation precludes any causal or mechanistic claim, and our findings should be interpreted as correlational and hypothesis‐generating. Additionally, MDSC content within stem cell grafts was not analyzed, which may affect post‐ASCT immune reconstitution. Third, the daratumumab‐based subgroup was excluded from the primary multivariable analysis because of the direct pharmacodynamic effect of anti‐CD38 therapy on MDSCs and the small event count (n = 3 ≤ PR); therefore, the generalizability of our findings to D‐based regimens cannot be assumed. Fourth, the single‐center design and modest overall sample size further constrain the precision and external validity of our estimates. Future research should validate the association between ΔM‐MDSCs and response in larger, multi‐center cohorts with integrated survival analysis and explore mechanistic interactions between MDSCs and T‐cell subsets within the marrow niche. Extending this investigation to novel immunotherapies—such as CAR‐T cells and bispecific antibodies—is also essential, as baseline immune suppression may critically shape response to these emerging modalities [29, 30, 31].

In conclusion, this study delineates the kinetic trajectory of MDSCs during ASCT in NDMM, demonstrating an induction therapy‐driven expansion followed by post‐transplant normalization. While a larger ΔM‐MDSCs was associated with inferior response depth in univariable analysis, this association did not retain independent predictive significance after adjusting for R2‐ISS stage in a conservative, bootstrap‐validated model. These findings suggest that ΔM‐MDSCs may serve as a marker of immune modulation rather than a standalone predictive biomarker, and they underscore the need for validation in larger, independent cohorts before any clinical application is considered. The post‐ASCT normalization of M‐MDSCs further prompts a critical consideration for clinical decision‐making in the era of novel immunotherapies: given this apparent “immune reset”, should potent agents (e.g., CAR‐T cells, bispecific antibodies) be deployed earlier, when the immune microenvironment is less immunosuppressed, or reserved for later lines after multiple relapses? This hypothesis requires prospective evaluation to optimize the treatment sequencing and maximize patient benefit.

Author Contributions

Q.Y. and B.C. wrote the manuscript. L.B. designed the research. Q.Y., M.L., Y.C., M.W., Q.X., L.F., and N.J. performed the research. Q.Y., M.L., S.G., and L.S. analyzed the data.

Funding

This work was supported by the Beijing Natural Science Foundation (Grant 7242075) and the Yangfan plan of Beijing Hospitals Authority (YGLX202315).

Ethics Statement

This study was approved by the Institutional Review Board and Ethics Committee of Beijing Jishuitan Hospital (Approval No. K2023‐318). Written informed consent was obtained from all participants prior to enrollment.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Sequential gating: Live CD45⁺ cells → HLA‐DR⁻/low → M‐MDSCs (CD11b⁺CD33⁺CD14⁺CD15⁻) and PMN‐MDSCs (CD11b⁺CD33⁺CD15⁺CD14⁻). FSC‐A/FSC‐H doublet exclusion applied. SSC/FSC Debris removal applied.

Figure S2: Patient flow diagram. A total of 131 newly diagnosed plasma cell disorders were screened. After excluding patients who did not receive induction at our center (n = 11), lacked cytogenetic data (n = 9), had solitary plasmacytoma (n = 3), had an uncertain diagnosis (n = 2), or had missing follow‐up data (n = 6), 99 patients with confirmed multiple myeloma (MM) were included in the final analysis. Among them, 78 received VRd induction and 21 received daratumumab‐based regimens. Fifty‐five patients (50 VRd, 5 daratumumab) proceeded to stem cell collection. Of these, 5 did not complete day‐100 evaluation and 4 were harvested but not infused; thus, 46 patients (42 VRd, 4 daratumumab) completed ASCT and had evaluable day‐100 post‐transplant samples. No other missing data were present for the primary timepoints (diagnosis and post‐induction) in the 99 included patients.

Table S1: Inclusion and exclusion criteria for multiple myeloma patients.

Table S2: Demographic characteristics of healthy donors.

Table S3: Expression characteristics of MDSCs in different disease states.

Table S4: Univariable logistic regression for suboptimal response (≤ PR) in the VRd Cohort (n = 78).

Table S5: Firth penalized likelihood logistic regression for suboptimal response (≤ PR) in the VRd cohort (n = 78).