Bone marrow immunosuppressive states associate with survival after guadecitabine and atezolizumab therapy in HMA-R/R MDS

Key Points

• •Baseline marrow inflammation and senescence limit immune activation after epigenetic and immune checkpoint therapy.
• •Durable survival is marked by interferon-driven immune reinvigoration and depletion of suppressive myeloid cells.

Visual Abstract

Abstract

Approximately half of patients with myelodysplastic syndromes (MDS) relapse or are refractory to hypomethylating agents (HMAs) and experience poor clinical outcomes. We previously reported improved survival in a phase 1/2 clinical trial combining the HMA guadecitabine with a programmed death-ligand 1 (PD-L1) inhibitor (atezolizumab) in HMA-relapsed or HMA-refractory (HMA-R/R) MDS, yet the biological features associated with durable responses to this combined epigenetic and immunotherapy approach remain unclear. Here, we performed integrated bulk and single-cell transcriptomic profiling of bone marrow samples from patients treated on this trial to identify molecular and cellular features associated with survival. Long-term survival was associated with the presence of immunosuppressive myeloid cells and primed dendritic cells at baseline, together with therapy-induced immune reinvigoration characterized by interferon pathway activation and expansion of effector lymphocytes. In contrast, short-term survival was associated with persistent senescence-associated inflammatory programs in CD34⁺ bone marrow cells and elevated expression of immunosuppressive immune checkpoint molecules. These findings suggest that chronic inflammatory and senescent microenvironmental states constrain effective immune activation despite combined epigenetic and immune checkpoint therapy. Here, we identify distinct bone marrow microenvironments associated with patient survival after combined epigenetic and immune checkpoint therapy and suggest candidate biomarkers to guide patient stratification in HMA-R/R MDS.

Introduction

Myelodysplastic syndromes (MDS) are clonal hematopoietic disorders characterized by ineffective hematopoiesis and a risk of progression to acute myeloid leukemia.1, 2, 3, 4 Hypomethylating agents (HMAs) are the standard of care for patients with higher-risk MDS who are ineligible for allogeneic stem cell transplantation, yet approximately half do not respond or relapse after treatment.^5,6 These patients with HMA-relapsed or HMA-refractory (HMA-R/R) MDS have a median survival of only 4 to 6 months, highlighting a major unmet clinical need.7, 8, 9, 10

HMAs can induce a viral mimicry response through reactivation of transposable elements (TEs) and interferon signaling, a process previously associated with clinical response in MDS.11, 12, 13 However, HMA therapy can also promote adaptive immune resistance by increasing expression of immune checkpoint molecules in both T cells¹⁴ and malignant myeloid cells.¹⁵ These observations provided a rationale for combining HMAs with immune checkpoint blockade.16, 17, 18

We recently reported results from a phase 1/2 trial combining guadecitabine with the programmed death-ligand 1 (PD-L1) inhibitor atezolizumab in patients with HMA-R/R MDS.¹⁹ The study enrolled 33 patients with a median age of 73 years (range, 54-85), of whom 70% were refractory and 30% had relapsed after receiving at least 4 cycles of azacitidine. No dose-limiting toxicities were observed during phase 1 dose escalation, and treatment was associated with manageable immune-related adverse events, consistent with expected cytopenias. The median overall survival was 15.1 months (95% confidence interval, 8.5-25.3), exceeding historically reported outcomes for this population. Although the cohort size was modest, the observed survival benefit motivated further investigation into the biological features associated with durable patient outcomes.

Most prior correlative studies in MDS have focused on tumor-intrinsic genetic or molecular features of CD34⁺ myeloid cells, with inconsistent clinical associations.^5,7,20, 21, 22, 23 These findings suggest that tumor-extrinsic mechanisms within the bone marrow microenvironment may also influence therapeutic outcomes. Because the marrow contains complex interactions among malignant progenitors, residual normal hematopoietic cells, and immune populations, single-cell profiling offers an opportunity to define cellular and transcriptional states associated with response or resistance. Here, we performed integrated bulk and single-cell transcriptomic profiling of bone marrow aspirates from patients with HMA-R/R MDS treated with guadecitabine and atezolizumab. By combining tumor-focused and microenvironment-focused analyses, we sought to identify molecular and cellular features associated with survival after combined epigenetic and immune checkpoint therapy.

Methods

Bulk total RNA-seq of bone marrow CD34⁺ cells

Bone marrow mononuclear cells (BMMCs) were isolated and cryopreserved, as previously described.¹⁹ Van Andel Institute received only coded data and samples under IRB 16014. Bone marrow aspirates were collected at baseline, after cycle 2, and after cycle 4 of therapy. BMMCs were thawed in Dulbecco's Phosphate-Buffered Saline containing 5% Fetal Bovine Serum, DNase I, and MgCl₂, and CD34⁺ cells were purified by magnetic selection (Miltenyi CD34 UltraPure). Total RNA was extracted using TRIzol. Libraries were prepared using QIAseq FastSelect and the KAPA RNA HyperPrep Kit and sequenced on an Illumina NovaSeq 6000. Healthy donor bone marrow CD34⁺ RNA-seq data were generated separately and deidentified before analysis.

Total RNA-seq libraries were adapter-trimmed (TrimGalore, https://zenodo.org/records/7598955), filtered in silico for ribosomal RNA,²⁴ downsampled to 35 million paired reads to normalize library size, and aligned using HISAT2.²⁵ Differential gene expression and pathway analyses were performed using DESeq2²⁶ (v.1.42.0, RRID:SCR_000154) and fgsea²⁷ (v.1.28.0). TE transcript quantification was performed using featureCounts,²⁸ as previously described.²⁹ Somatic variants were inferred from CD34⁺ RNA-seq BAM files using MuTect2 (GATK v4.6.2.0),^30,31 and MuTect2-reported allele fractions were quantified for previously identified MDS-associated mutations (supplemental Table 3).

Bone marrow and PBMC Cellular Indexing of Transcriptomes and Epitopes by sequencing

Cryopreserved BMMC samples collected pretreatment and after cycle 2 and peripheral blood mononuclear cell (PBMC) samples collected pretreatment, 8 days after treatment initiation in cycle 1 or 2, and after cycle 1 were thawed and resuspended, as previously described. Cells were labeled with TotalSeq-C hashtag antibodies, sorted for viability, pooled, and stained with a TotalSeq-C antibody panel. Single-cell gene expression and surface protein libraries were generated using the 10× Chromium Next GEM 5′ kit and sequenced on a NovaSeq 6000.

Gene and protein count matrices were generated with Cell Ranger (v4.0.0) and processed with CellBender³² (v0.3.0) to remove ambient or background noise. Data were normalized with SCTransform and integrated using Seurat³³ (v5.0.1) weighted nearest-neighbor analysis. Cell identities were assigned using Azimuth-guided annotation³⁴ (v0.5.0), followed by manual curation based on canonical markers. The FindMarkers function (Wilcoxon rank-sum test) was used for single-cell differential expression testing. Pseudobulk count matrices for each cell subtype were generated using SingleCellExperiment³⁵ (v1.24.0), and gene set enrichment analysis (GSEA) was performed with DESeq2 and fgsea packages. Cell subtypes with <10 cells detected in a sample were excluded from pseudobulk analysis. T-cell receptor analysis used TRUST4³⁶ (v.1.0.13), and ligand-receptor analysis used CellChat³⁷ (v2.1.2). All analyses were performed in R v4.3.0. No randomization, blinding, or formal power calculations were performed for these analyses.

Low-dimensional flow cytometry characterization of select patient bone marrow aspirates

Cryopreserved BMMC vials were thawed and resuspended, as previously mentioned. A total of 1 × 10⁶ live cells per sample were aliquoted, washed twice in Stain Buffer BSA (RRID:AB_2869007), and blocked in Human BD Fc Block (RRID:AB_2728082) for 10 minutes at room temperature. Surface staining was performed for 30 minutes on ice using the antibodies listed in supplemental Table 1, with a final staining volume of 80 μL. Cells were stained using FVS780 viability stain (RRID:AB_2869673) and subsequently fixed and permeabilized with Cytofix/Cytoperm fixation and Perm/Wash buffers (RRID:AB_2869008) according to the manufacturer’s instructions. Intracellular S100A8/9 staining was performed for 30 minutes on ice; the samples were then washed twice with Perm/Wash buffer, twice with 3 mL stain buffer, and finally resuspended in 250 μL stain buffer. Antibody-based single-color controls were prepared on Anti-Mouse Ig, κ CompBeads (RRID:AB_10051478), and all except S100A8/9 Alexa Fluor 647 were exposed to Cytofix/Cytoperm buffer to account for preparation-related spectral changes. Samples and single-color controls were acquired using a Cytek Aurora spectral cytometer (SpectroFlo v 3.0.3), data were unmixed with autofluorescence extraction, and minor spillover corrections were made with SpectroFlo v3.0.3 (correction matrix can be found in supplemental Table 1). Gating for analysis was performed with FlowJo v10.10.0 (RRID:SCR_008520).

Results

Immunosuppressive signatures demarcate STS bone marrow CD34⁺ cells and associate with limited interferon response after treatment

We collected bone marrow biopsies and PBMCs from short-term survivors (STS; <15 months survival) and long-term survivors (LTS; >15 months survival) before treatment and after 2 and/or 4 cycles of therapy (Figure 1A; supplemental Table 2). Patients were grouped according to whether their survival duration was shorter or longer than the median overall survival observed in this clinical trial (15.1 months). In prior work,¹⁹ targeted sequencing of MDS-associated genes (supplemental Table 3) did not identify mutation-based differences in myeloid cells associated with survival.

Figure 1.

Bulk transcriptomic comparisons of bone marrow CD34⁺ cells of patients with HMA-R/R MDS. (A) Treatment overview of the VAI-SU2C MDS clinical trial combining guadecitabine and atezolizumab in the HMA-R/R MDS cohort. Bone marrow CD34⁺ cells were collected at pretreatment, post–cycle 2, and post–cycle 4 time points. The number of samples profiled by bulk RNA-seq at each time point is shown for STS and LTS. (B) Hallmark GSEA comparing STS or LTS CD34⁺ cells across treatment time points (adjusted P value <.01). Enrichment directionality is color-coded, as indicated in the legend. (C) GSEA of senescence-associated pathway, SenMayo, in bulk bone marrow CD34⁺ cells at pretreatment time point. (D) Distribution of immune checkpoint gene expression in CD34⁺ cells isolated from healthy donor, STS, and LTS bone marrow at pretreatment time point. Statistical significance tested with 2-tailed Welch t test: ∗P < .05; ∗∗P < .01. MOS, months overall survival; Pre, pretreatment; PC2, post–cycle 2; PC4, post–cycle 4.

We isolated CD34⁺ cells from healthy donor and patient bone marrow aspirates and performed bulk total RNA-seq. Healthy donor CD34⁺ samples were transcriptionally homogeneous, whereas MDS CD34⁺ samples demonstrated marked interpatient heterogeneity (supplemental Figure 1A-B). Differential gene expression analysis identified few genes that passed the false discovery rate threshold across most pairwise comparisons of patient survival and treatment time points (supplemental Figure 1C-D; supplemental Table 4). Next, we assessed pathway-level differences using GSEA.³⁸ At baseline, STS CD34⁺ cells were enriched for tumor necrosis factor α signaling via the nuclear factor κB pathway, transforming growth factor β signaling, reactive oxygen species, and apoptosis pathways (Figure 1B). These features are consistent with aging-associated immune dysfunction and cellular senescence.39, 40, 41, 42 Indeed, the senescence-associated “SenMayo” pathway⁴³ was also significantly enriched in STS CD34⁺ cells at baseline, supporting the presence of a senescent and immunocompromised marrow state. After treatment, STS CD34⁺ cells failed to mount strong interferon gamma responses and instead indicated induction of MYC-associated programs. In contrast, therapy in patients with LTS was associated with enhanced proliferation and strong activation of interferon gamma response pathways.

Consistent with prior reports,¹³ viral mimicry genes were elevated in patient-derived CD34⁺ cells at baseline relative to healthy donor controls, together with increased expression of TEs, particularly intergenic long interspersed nuclear elements and long terminal repeats (supplemental Figure 2B). After treatment, interferon pathway activation became uncoupled from TE transcription in both STS and LTS groups (supplemental Figure 2C-D). Furthermore, relative to healthy donor CD34⁺ cells, patient-derived CD34⁺ cells also demonstrated reduced epithelial-mesenchymal transition and collagen-containing extracellular matrix programs at baseline, which were selectively increased after therapy in LTS CD34⁺ cells (supplemental Figure 3).

Because HMAs can induce immune checkpoint expression in hematopoietic cells,¹⁵ we also examined checkpoint genes in bone marrow CD34⁺ cells at baseline (supplemental Figure 4). PD-1 gene expression was low across groups, whereas PD-L1 was modestly elevated in both STS and LTS relative to healthy donors (Figure 1D). Notably, STS CD34⁺ cells exhibited broader upregulation of immunosuppressive checkpoint molecules^44,45 compared with both healthy donors and patients with LTS. In contrast, LTS CD34⁺ cells exhibited an inverse pattern of OX40 and OX40L expression, with reduced OX40 and increased OX40L levels relative to STS and healthy controls. In addition, LTS CD34⁺ cells showed modest upregulation of the immunomodulatory genes CD160 and TDO, although the functional significance of these changes remains unclear. Collectively, these findings indicate that STS CD34⁺ cells are marked by senescence-associated inflammation and elevated immunosuppressive programs that may constrain response to combined HMA and immune checkpoint blockade.

Distinct cellular composition and dynamics mark STS and LTS bone marrow

To evaluate both tumor-intrinsic and tumor-extrinsic features, we profiled bone marrow samples by CITE-seq^46,47 in a subset of available specimens from 3 patients who were STS and 10 patients who were LTS collected before treatment and after 2 cycles of therapy (supplemental Table 5; supplemental Figure 5A). The number of profiled samples reflects the availability of suitable cryopreserved bone marrow specimens. CITE-seq enables simultaneous quantification of whole-transcriptome gene expression and ∼150 to 200 cell surface proteins at single-cell resolution (supplemental Tables 6 and 7). Using multimodal integration and marker-based annotation,^33,34 we identified 30 hematopoietic progenitor and immune cell populations across 61 487 cells (Figure 2A-B; supplemental Figures 5-8).^34,48,49 Low-dimensional flow cytometry on additional samples supported the broad compositional trends identified by CITE-seq (supplemental Figures 8C and 9).

Figure 2.

Distinct immune compositions in the bone marrow of patients with HMA-R/R MDS before and after combination therapy. (A) Weighted nearest neighbor–based Uniform Manifold Approximation and Projection (wnnUMAP) projection of the 30 cell subtypes identified in the bone marrow microenvironment and peripheral blood of patients with MDS. (B) Gene expression (top) or protein abundance (bottom) of cell type–specific markers quantified in each single cell detected in the bone marrow microenvironment, which are projected on wnnUMAP. (C) The distribution of cell subtype fractions detected in the bone marrow of patients with STS or those with LTS before and after combination therapy. Statistical significance tested with 2-tailed Welch t test: ∗P < .05; ∗∗P < .01. (D) Heat map reflecting the changes in bone marrow cellular fractions for each patient after combination therapy. CTL, cytotoxic; EPC, erythroid precursor cells; gdT, γδ T; MAIT, mucosal-associated invariant T; MEP, megakaryocyte erythroid progenitor; NK, natural killer; pDC, plasmacytoid dendritic cell; TCM, T central memory; TEM, T effector memory; TEMRA, terminally differentiated effector memory; Treg, regulatory T cell.

At baseline, hematopoietic stem cells (HSCs), lymphoid-primed multipotent progenitors (LMPPs), and granulocyte-monocyte progenitors (GMPs) were more abundant in LTS bone marrow relative to STS bone marrow (Figure 2C). These progenitor populations include malignant and disease-initiating cells, consistent with differences in tumor burden or differentiation state at baseline. STS bone marrow contained higher fractions of naive CD4⁺ T cells and CD4⁺ central memory T cells, whereas LTS bone marrow was enriched for antigen-presenting populations, including myeloid dendritic cells (mDCs) and plasmacytoid dendritic cells. After combination therapy, patients who were LTS exhibited expansion of effector lymphocytes, including natural killer cells, γδ T cells, CD4⁺ T cells, and CD8⁺ T cells, accompanied by a reduction in hematopoietic stem and progenitor cell (HSPC) populations (HSCs, LMPPs, and GMPs) (Figure 2D). Patients who were STS instead showed modest decreases in CD4⁺ and CD8⁺ T cells and relative increases in HSC and LMPP fractions. These results indicate that baseline marrow composition and treatment-induced cellular remodeling differ substantially between patients who are STS and those who are LTS.

STS bone marrow demonstrates elevated inflammation and senescence-like signatures in HSPCs

To identify transcriptional programs associated with early treatment failure, we performed pseudobulk analyses of single-cell RNA-seq data (DESeq2²⁶) to mitigate patient-to-patient variability and imbalanced cell numbers.⁵⁰ Pseudobulk analysis of STS CD34⁺ cells, aggregating megakaryocyte-erythroid progenitors, HSCs, LMPPs, and GMPs, recapitulated the inflammatory and oncogenic signatures seen in bulk RNA-seq (supplemental Figure 8D). Although therapy induced interferon signaling in STS CD34⁺ cells, these responses were not accompanied by increased proliferative capacity, as shown in LTS CD34⁺ cells. Extending GSEA to individual HSPC subtypes showed consistent enrichment of inflammatory and senescence-associated pathways across STS megakaryocyte-erythroid progenitor, HSC, and LMPP populations at baseline and after treatment (Figure 3A). These signatures were accompanied by immune cell proliferation without coordinated interferon activation after therapy (supplemental Figure 10).

Figure 3.

Enhanced inflammation and senescent-like signatures in bone marrow HSPCs of STS. (A) Pseudobulk GSEA of hallmark pathways comparing STS and LTS bone marrow CD34⁺ cell subtypes across pretreatment (pre-txt) and posttreatment (post-txt) time points (adjusted P value <.05). (B) Volcano plots illustrating single-cell differential gene expression analysis comparing STS and LTS HSPCs at study initiation. Red points represent differentially expressed genes with adjusted P value <.01 and log₂ fold change >1.0. (C) GSEA of the senescence pathway, “SenMayo,” detected in LMPP cells of STS.

Single-cell differential expression analysis (Wilcoxon rank-sum test;³⁴ supplemental Table 8) identified CDKN2C, CDKN2D, CXCL2, CXCL3, and CXCL8 as significantly upregulated in STS HSPCs relative to LTS (Figure 3B). CDKN2C (p18) and CDKN2D (p19) are cyclin-dependent kinase inhibitors associated with cell-cycle restriction and senescence,51, 52, 53 whereas CXCL2, CXCL3, and CXCL8 (IL-8) are proinflammatory senescence-associated secretory phenotype cytokines that contribute to immune suppression and chronic inflammation.^43,54 These genes were broadly expressed across STS HSPCs, particularly in LMPPs (supplemental Figure 11A-C). Consistent with this pattern, pseudobulk GSEA confirmed persistent enrichment of the “SenMayo” pathway in STS LMPPs before and after therapy (Figure 3C). Ligand-receptor inference using CellChat³⁷ revealed that TNF and IL-1 signaling interactions (which associate with chronic inflammation and senescence⁵⁵) were specifically detected in STS HSPC subpopulations, but not in LTS HSPCs (supplemental Figure 12).

We also observed increased expression of CDC-like kinase 3 (CLK3) in STS HSPCs (supplemental Figure 11D), most prominently in the patient harboring a DDX41 mutation. DDX41 has been linked to increased CLK3 expression, and both proteins are implicated in RNA splicing regulation and MDS pathogenesis.⁵⁶ Although the role of CLK3 in therapy resistance remains unclear, CLK inhibitors are under investigation as potential therapeutic strategies in MDS.⁵⁷ Notably, we did not detect elevated expression of ITGA5, previously associated with HMA resistance⁵⁸ (supplemental Figure 11E). Together, these data indicate that STS marrow is characterized by persistent inflammation and senescence-like programs in HSPCs that are not reversed by therapy.

Combination therapy reduces immunosuppressive monocytes in LTS bone marrow

Because LTS bone marrow showed both expansion of effector lymphocytes and reduction of HSPC populations, we investigated whether this reflected loss of malignant clones or broader transcriptional remodeling within persistent disease. Variant allele frequency analysis of bulk CD34⁺ RNA-seq from paired samples did not show consistent reductions in previously identified MDS-associated mutations after treatment (supplemental Table 8), suggesting that the observed transcriptional changes largely reflect altered cell state rather than replacement by residual normal hematopoiesis.

Pseudobulk GSEA comparing pretreatment and posttreatment samples showed robust activation of interferon pathways across multiple immune and progenitor populations in LTS bone marrow, responses that were largely absent in STS samples (Figure 4A; supplemental Figure 11). At baseline, LTS bone marrow also contained higher proportions of GMPs and myeloid cells than STS marrow (Figure 2C). Transcriptomic analysis revealed preferential expression of the alarmin genes S100A8 and S100A9 in GMPs and classical monocytes from patients who were LTS (Figure 4B-C; supplemental Figure 13A). S100A8/9 signaling has also been implicated in the induction of PD-1 expression on HSPCs^59,60 (supplemental Figure 14) and the recruitment of myeloid-derived suppressor cells (MDSCs) to the bone marrow,61, 62, 63 thereby establishing an immunosuppressive, tumor-promoting microenvironment. MDSCs further propagate alarmin signaling pathways by becoming an additional source of S100A8/9 production in the bone marrow.⁶⁴ Indeed, classical monocytes from LTS marrow also demonstrated increased expression of MDSC marker genes⁶⁵ at baseline (Figure 4D). After therapy, GMPs and classical monocytes were markedly reduced in patients who were LTS, coinciding with expansion of effector immune populations and induction of interferon signaling (Figure 2D). These findings suggest that therapy-associated remodeling in LTS marrow includes depletion of immunosuppressive myeloid populations.

Figure 4.

LTS bone marrow microenvironments are enriched with S100A8/9-expressing myeloid cells and CD1-expressing mDCs. (A) Pseudobulk GSEA of hallmark pathways comparing pretreatment and posttreatment samples of LTS cell subtypes (adjusted P value <.05). (B) Single-cell differential gene expression analysis comparing STS and LTS GMPs at pretreatment time point. Red points represent differentially expressed genes with adjusted P value <.01 and log₂ fold change >1.0. (C) Single-cell gene expression distribution of S100A8/9 in HSPCs and classical monocytes. (D) Distribution of MDSC marker gene expression in classical monocytes of STS or LTS at pretreatment time point. (E) Single-cell differential gene expression analysis comparing STS and LTS mDCs at pretreatment time point. Red points represent differentially expressed genes with adjusted P value <.01 and log₂ fold change >1.5. (F) Pseudobulk gene expression profiles of differentially expressed CD1D and CD1E in HSPCs and myeloid cells.

Noncanonical dendritic cell antigen presentation associates with patient survival

Effective interactions between dendritic cells and T cells are essential for T-cell activation and differentiation during immune-mediated tumor clearance.⁶⁶ STS marrow contained higher frequencies of less-differentiated T-cell populations (naive CD4⁺ and CD4⁺ central memory T cells) at baseline, which coincided with lower abundance of myeloid and plasmacytoid dendritic cells relative to LTS (Figure 2C). In contrast, LTS mDCs exhibited higher expression of major histocompatibility complex class II genes and CD48 at baseline, molecules involved in CD4⁺ T-cell and natural killer cell activation⁶⁷ (Figure 4E). LTS mDCs also expressed higher levels of CEBPD, a transcription factor linked to inflammatory responses and regulation of S100A8/9.⁶⁸ These features were associated with greater T-cell receptor clonal diversity (TRUST4³⁶) and expansion after therapy in patients who were LTS (supplemental Figure 15A-D).

Notably, members of the CD1 family of lipid antigen-presenting molecules, including CD1D and CD1E, were preferentially expressed in LTS mDCs (Figure 4F). CD1D was also elevated in LTS LMPPs and LTS GMPs (supplemental Figure 13B). Because CD1D can present lipid antigens to γδ T cells,^69,70 these observations raise the possibility that noncanonical antigen presentation contributes to antitumor immune activation in patients who are LTS. Consistent with this model, γδ T cells modestly expanded in LTS marrow after therapy (Figure 2D).

Evaluating peripheral blood as a surrogate for bone marrow microenvironment dynamics

To determine whether PBMCs could serve as a surrogate for bone marrow microenvironment dynamics, we performed CITE-seq profiling on available PBMC samples (supplemental Figure 8A-B). At baseline, PBMC immune composition broadly mirrored corresponding bone marrow samples, aside from the expected absence of HSPC populations from peripheral blood (Figure 5A). Expression of canonical immune checkpoint molecules on immune cells was also similar between PBMCs and bone marrow (supplemental Figure 16). However, treatment-induced changes in immune cell composition observed in bone marrow were not reflected in PBMCs (Figure 5B), indicating that posttreatment bone marrow sampling remains necessary to assess therapeutic remodeling.

Figure 5.

CITE-seq profiling of PBMCs from patients with HMA-R/R MDS treated with combination therapy. (A) Heat maps illustrating the fraction of cell subtypes detected in the bone marrow and PBMC samples of patients who were STS and those who were LTS across treatment time points. (B) Heat map reflecting the change in cellular fractions in the bone marrow or PBMC samples for select patients with MDS after combination therapy. (C) Pseudobulk expression distribution of IFNG in PBMC-immune subtypes.

We further assessed viral mimicry and interferon responses in PBMCs but did not observe associations between viral mimicry activation and survival (supplemental Figure 17A). Instead, basal and treatment-induced IFNG expression in effector lymphocytes distinguished patients who were STS and those who were LTS (Figure 5C). Senescence-associated transcriptional features identified in marrow were also detectable in PBMCs, with elevated senescence-related genes in peripheral CD34⁺ cells from patients who were STS and increased S100A8/9 expression in monocytes from those who were LTS (supplemental Figure 17B-C). These data suggest that PBMCs capture selected systemic immune features but do not adequately reflect treatment-driven bone marrow remodeling.

Discussion

Effective therapies remain limited for patients with HMA-R/R MDS. In a prior phase 1/2 trial, we reported improved survival with guadecitabine plus atezolizumab,¹⁹ prompting investigation of the biological features associated with durable benefit. Although the cohort size limits definitive conclusions, the present correlative study provides insight into bone marrow states associated with survival after combined epigenetic and immune checkpoint therapy. Although guadecitabine has not advanced further in clinical development for MDS because it did not demonstrate superiority over existing therapies, it functions as a prodrug of decitabine. As such, the biological mechanisms identified in this study likely extend to other DNA methyltransferase (DNMT) inhibitors used in MDS, particularly decitabine, which shares the same active metabolite and epigenetic mechanism of action. In prior work, we associated longer survival with ASXL1 mutations in myeloid and T cells¹⁹ and demonstrated that ASXL1 loss in CD8 T cells enhances their self-renewal and responsiveness to immune checkpoint blockade.⁷¹ However, CITE-seq did not reliably capture ASXL1 expression across immune cell types in this study, precluding mutation-specific analysis. Instead, our integrated bulk and single-cell analyses highlight the bone marrow microenvironment as a key determinant of survival after combined epigenetic and immune checkpoint therapy in HMA-R/R MDS.

By integrating bulk and single-cell transcriptomic analyses, we identified distinct marrow microenvironments associated with short-term and long-term survival. Bulk RNA-seq of bone marrow CD34⁺ cells revealed minimal differentially expressed genes in comparing treatment and survival groups. This likely reflects substantial interpatient variability, which limits statistical power for detecting individual differentially expressed genes. However, gene pathway–level analysis revealed that patients who were STS were characterized by inflammatory and senescence-like programs in HSPCs, including TNF-α and NF-kB signaling, cyclin-dependent kinase inhibitors, and senescence-associated cytokines, such as CXCL2, CXCL3, and CXCL8. These states persisted despite treatment and were accompanied by limited interferon activation and expansion of HSPC populations after therapy, consistent with continued disease progression. These observations support a model in which chronic inflammatory and senescent marrow states constrain effective immune reinvigoration even after combination therapy, raising the possibility that senescence-targeting strategies, such as senolytic therapies,⁷² could represent alternative or complementary approaches for this patient subset.

In comparison, patients who were LTS demonstrated therapy-associated interferon pathway activation across multiple bone marrow cell types together with expansion of effector lymphocytes and reduction of GMP and classical monocyte populations. At baseline, these myeloid populations expressed higher levels of S100A8/9 and MDSC-associated programs, suggesting that LTS marrow initially contains immunosuppressive myeloid compartments that are subsequently depleted or remodeled during therapy. These findings are consistent with a model in which checkpoint blockade relieves immune suppression mediated by MDSC-like myeloid cells, enabling cytotoxic immune activation within the marrow microenvironment (supplemental Figure 18A).

Our data also suggest a potential role for noncanonical antigen presentation. LTS mDCs exhibited increased expression of major histocompatibility complex class II genes and CD1 family molecules, including CD1D and CD1E. Because CD1D can present lipid antigens to γδ T cells, these features may contribute to superior activation of cytotoxic lymphocytes in LTS bone marrow (supplemental Figure 18B). The modest expansion of γδ T cells after treatment is consistent with this possibility, although functional validation will be required.

Several limitations should be acknowledged. The cohort was modest in size and imbalanced between STS and LTS groups, limiting statistical power and precluding causal inference. In addition, although the study identifies strong associations between bone marrow states and outcome, mechanistic validation of specific cellular interactions will require functional studies that remain challenging in MDS. Nevertheless, detailed molecular data sets from patients with HMA-refractory MDS are rare, and these analyses provide a useful framework for hypothesis generation and future validation.

In summary, our study identifies distinct baseline bone marrow microenvironment states associated with survival in patients with HMA-R/R MDS treated with combined HMA and immune checkpoint blockade. Persistent senescence-associated inflammation characterized STS, whereas immune reinvigoration and reduction of immunosuppressive myeloid cells were associated with durable survival. These findings support the incorporation of bone marrow microenvironment profiling into future clinical trials. Although PBMCs reflected baseline immune states, they failed to capture treatment-induced remodeling, indicating that bone marrow sampling remains essential. Nevertheless, interferon signaling, senescence-associated inflammation, and alarmin expression in peripheral blood may serve as minimally invasive biomarkers for patient stratification.

Conflict-of-interest disclosure: J.-P.J.I. served as a consultant for Daiichi and Astex. C.L.O. and K.G. received nonfinancial support from Astex and Genentech during the conduct of the clinical study. P.A.J. served on the scientific advisory board for Zymo Research, EpiGenOnco, and Cancer Research UK; and was a scientific review council member for the Cancer Prevention and Research Institute of Texas. The remaining authors declare no competing financial interests.

Declaration of generative AI and AI-assisted technologies in the writing process: During the preparation of this work the authors used generative artificial intelligence in order to assist in identifying and correcting grammatical and stylistic errors. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.