To the Editor:

Somatic mutations in the splicing factor SF3B1 occur in about one-third of all myelodysplastic neoplasms (MDS) and define a subgroup of patients characterized by ring sideroblasts (RS), ineffective erythropoiesis, and an indolent disease course in lower-risk (LR) MDS [1]. They are typically heterozygous missense substitutions, most commonly (>50% in MDS) involving p.K700E (SF3B1 NM_012433.4: c.2098A>G (p.Lys700Glu), hereafter referred to as SF3B1K700E), and have been shown to induce mis-splicing of key genes throughout erythroid differentiation [2, 3]. Surprisingly, although SF3B1 mutations are known to target multipotent lymphomyeloid hematopoietic stem cells and clonally propagate to myeloid progenitors [4], their impact on mature immune cells remains largely unexplored. Clinically, SF3B1 mutations are associated with high response rates to the erythroid maturation agent luspatercept and lower response to immunosuppressive treatment (IST) [5,6,7].

In this study, we performed multiplex immunophenotyping in conjunction with machine learning-based analytical approaches on bone marrow (BM)/peripheral blood (PB) samples from newly diagnosed or disease-modifying treatment-naïve SF3B1mut or SF3B1wt MDS patients (experimental cohort: Supplementary Table S1; Fig. S1) and healthy donors (HD) to identify genotype-immunophenotype correlations. Initial gene expression profiling of 730 immune-related genes in SF3B1mut versus SF3B1wt MDS BM mononuclear cells (BM-MNCs) revealed a predominantly myeloid cell-related innate immune gene signature (e.g., CYBB, CSF1R) lacking signs of overt myeloid-driven inflammation (i.e. IL1B, CXCL5), whereas lymphoid-related genes were underrepresented (e.g., CD3D, CD79A) (Supplementary Table S4, Fig. S2). These results are consistent with the reported lower proportion of lymphocytes in BM [8], mild myeloid dysplasia [9], and our previous finding of significantly lower IL1B mRNA in BM monocytes from SF3B1mut LR-MDS [10]. As IL-1β protein levels in paired BM plasma samples were often below the detection limit, we could not determine whether lower mRNA levels correspond to lower cytokine levels.

Next, we conducted high-dimensional mass cytometry (CyTOF) on BM-MNCs and analyzed data using the Tracking Responders EXpanding (T-REX) algorithm to identify immunophenotypic differences associated with LR-MDS and SF3B1K700E LR-MDS in particular. As expected, LR-MDS (SF3B1mut and SF3B1wt) showed several immunophenotypic changes consistent with an activated immune response (Fig. 1A, Supplementary Fig. S3/S4), in particular specific clusters resembling terminally differentiated effector memory CD8+ T cells (TTE/TEMRA, cluster 1295), mature CD57+ NK cells (cluster 2495), CD27+ IgD memory B cells (cluster 795), and γδ T cells with an exhausted immunophenotype (cluster 1395). LR-MDS exhibited dysregulated T-cell homeostasis, with fewer naïve CD4+ and CD8+ T cells, and memory phenotype skewing toward CD8+ effector memory (TEM) and TTE cells (Supplementary Fig. S4). This is consistent with progressive memory differentiation entailing loss of survival, which could contribute to impaired long-term antitumor immunosurveillance.

Fig. 1: Monocytes with HLA-DRlow/neg immunophenotype emerge frequently in the BM of SF3B1mut MDS.
figure 1

A T-REX plot of regions of significant change on Uniform Manifold Approximation (UMAP) axes for CD45+ BM-MNCs stained for CyTOF showing distinct LR-MDS-specific (dark red, ≥95% of cells are contributed by LR-MDS samples) and HD-specific (dark blue, ≥95% of cells are contributed by HD) cell clusters. 14 LR-MDS (mean age = 74 years, 4 women, 10 men) and 4 HD (mean age = 58 years, all men) were included in the analysis. LR-MDS group comprises SF3B1K700E (n = 5, orange dots; mean age = 75 years, 2 women, 3 men) and SF3B1wt (n = 9, blue dots; mean age = 74 years, 2 women, 7 men) patients. Top 10 Marker Enrichment Modeling (MEM) labels with enrichment scores are shown for statistically significant LR-MDS-specific clusters (cutoff  >2000 cells) indicated on T-REX plot. B T-REX analysis of CD45+ BM-MNCs stained for CyTOF showing distinct SF3B1K700E-specific (dark red) and SF3B1wt-specific (dark blue) cell clusters. Top 10 MEM labels are shown for statistically significant and trend clusters (cutoff  >1000 cells) indicated on T-REX plot. A, B Labels on T-REX plot indicate major immune cell subsets (myeloid cells, NK cells, γδ T cells, CD4+ and CD8+ T cells, B cells). C T-REX analysis of CD33+ CD14+ pre-gated monocytes showing SF3B1K700E-specific (dark red) and SF3B1wt-specific (dark blue) clusters. Cluster 795 depicts a distinct HLA-DRlow/neg monocyte subset in SF3B1K700E LR-MDS (NOTE: this cluster is not related to cluster 795 shown in (A)). HLA-DR expression was projected onto UMAP axes. AC Two-sided Mann–Whitney-U-test/Wilcoxon rank-sum test was performed for indicated clusters (p < 0.05 was considered significant; p-values are shown in brackets). Box plots depict median, IQR (lower and upper hinges), and 1.5 times the IQR (lower and upper whiskers extend to values within 1.5 times the IQR from the hinge). (D) Percentage of CD33+ CD14+ BM monocytes with HLA-DRlow/neg immunophenotype in HD (median = 9.7, IQR = 8 [n = 9, mean age = 69 years, 6 women, 3 men]), SF3B1mut (median = 37.9, IQR = 34.9 [n = 17; orange dots, K700E; light orange-filled circles, nonK700E including one K666R, one E622D, one H662Y, and one Y623C; mean age = 71 years, 6 women, 11 men]), and SF3B1wt (median = 6.3, IQR = 11.2 [n = 16, mean age = 66 years, 8 women, 8 men]) MDS assessed by diagnostic FCM of freshly stained BM samples (Kruskal–Wallis test with Dunn’s post-hoc test [Bonferroni adjusted p-values]). E Representative HLA-DR staining on CD33+ CD14+ BM monocytes. The black line indicates the set threshold distinguishing low or negative from high HLA-DR expression. F Percentage of CD33+ CD14+ BM monocytes with HLA-DRlow/neg immunophenotype in four SF3B1K700E MDS patients over time. Patients #4, #10, and #11 harbor an isolated SF3B1K700E mutation.

We then compared SF3B1K700E to SF3B1wt LR-MDS using the T-REX pipeline, which identified a SF3B1K700E-specific cluster comprising CD33+ CD14+ monocytes (cluster 495, p < 0.01) (Fig. 1B, Supplementary Fig. S5). Further analysis of CD33+ CD14+ BM-MNCs showed that a remarkable proportion of the monocytes in SF3B1K700E LR-MDS adopt a HLA-DRlow/neg phenotype (Fig. 1C). Importantly, retrospective analysis of diagnostic flow cytometry data (Fig. 1D–E) and external validation in two independent cohorts comprising combined 130 MDS (118 LR-MDS) patients (Supplementary Fig. S6) confirmed an increased frequency of HLA-DRlow/neg monocytes in SF3B1mut (both SF3B1K700E and SF3B1nonK700E) compared to SF3B1wt MDS and HD. The external data support our observation that staining cryopreserved BM-MNCs may underestimate the actual frequency of HLA-DRlow/neg monocytes. Additionally, we found a strong correlation between HLA-DRlow/neg monocyte frequencies in BM and PB (Supplementary Fig. S7). HLA-DRlow/neg monocytes in SF3B1mut MDS were classical monocytes (CM) based on the lack of CD16 surface expression (Supplementary Fig. S7). Analysis of longitudinal data from four SF3B1K700E MDS patients showed a consistently high frequency of HLA-DRlow/neg monocytes (Fig. 1F).

To the best of our knowledge, the only other study directly investigating immunophenotypic features in BM of SF3B1mut MDS reported lower expression of CD11b, CD36, and CD64 on monocytes [8]. Another study found a higher frequency of thrombomodulin-expressing CM in MDS subtypes with <5% blasts and RS [11]. The association of SF3B1 mutations with lower monocyte surface HLA-DR expression identified here may be of clinical relevance, for example in view of the predicted poor response of SF3B1mut MDS to IST [6, 7]. Overall, the frequency of HLA-DRlow/neg monocytes showed no correlation with blood hemoglobin levels, the Revised International Prognostic Scoring System (IPSS-R) risk classifications, or the mutational burden of SF3B1 and co-mutated TET2 or DNMT3A (Supplementary Fig. S7). HLA-DRlow/neg monocyte frequencies were comparable between transfusion-dependent and –independent SF3B1mut LR-MDS (Supplementary Fig. S6). However, HLA-DRlow/neg monocytes have known immunoregulatory properties via multiple mechanisms, including effector T-cell inhibition, decreased antigen presentation, and defective dendritic cell maturation [12]. A possible scenario is that the early acquisition of SF3B1 mutations [13] and the presence of inflammation foster the emergence of HLA-DRlow/neg monocytes, which then contribute to counteract and balance inflammatory responses in established SF3B1mut MDS. In this context, T-REX also identified a cluster of naïve CD4+ T cells specific to SF3B1K700E LR-MDS with low expression of the co-stimulatory molecule CD27 (cluster 3895, MEM score CD27+1) (Fig. 1B, Supplementary Fig. S4/S5). Thus, although disease-related shifts in CD4+/CD8+ T-cell differentiation were noticeable irrespective of SF3B1 mutation status, naïve CD4+ T cells in SF3B1K700E LR-MDS displayed subtle immunophenotypic differences indicative of less recent activation.

As CD14+ monocytes lose HLA-DR expression, they become functionally deactivated, which can contribute to the transition to a more immunosuppressed state. To investigate whether this is the case for CM from SF3B1K700E LR-MDS, we studied their global gene expression profile using RNA-seq. Overall, we found 545 up- and 812 downregulated genes in the clonally involved CM from SF3B1K700E LR-MDS compared to HD (Supplementary Table S5). Importantly, these patients harbored an isolated K700E mutation and no confounding cytogenetic aberrations. Upregulated genes were enriched in genes involved in oxygen transport (e.g., HBB, HBA1/2), probably due to erythrocyte impurities or enhanced phagocytosis of damaged erythrocytes by CM in SF3B1K700E LR-MDS. Downregulated genes were significantly enriched in genes related to cytokine signaling, including cytokine receptors (e.g., IL6R, IL10RA, IL7R, TNFRSF1A), TREM1, signaling kinases (e.g., MAP3K7, MAP3K8, PIK3CG), and NF-κB signaling modulators (e.g., NFKBIB, IKBKG, RELA/B) (Fig. 2A). Ingenuity pathway analysis (IPA) of DEG identified enriched pathways pertaining to inflammatory cytokine signaling (i.e. NF-κB signaling, IL-6 signaling, acute phase response signaling, PI3K/AKT signaling) and inflammatory conditions (i.e. hepatic fibrosis signaling pathway) that could be affected in SF3B1K700E CM (Fig. 2B). Expression levels of the NF-κB targets IL1B and TNF were, however, variable between individual patients (Supplementary Fig. S8). Notably, IPA-based analysis of DEG in CM from SF3B1wt LR-MDS patients, of whom 2 out of 3 carried somatic mutations in TET2, brought to the fore different inflammatory pathways predicted to be more active compared to HD (Fig. 2B).

Fig. 2: Classical monocytes (CM) from SF3B1K700E LR-MDS exhibit dysregulated immune gene expression and splicing.
figure 2

A Metascape pathway and process enrichment analysis of up- and downregulated genes in peripheral blood CM from SF3B1K700E LR-MDS (n = 3) compared to HD (n = 3). The top 10 and 20 representative terms are shown for up- and downregulated genes, respectively. B IPA core pathway analysis showing the predicted activity (cutoff z-score of >|0.5|) of overrepresented annotations (p-value < 0.05 [right-tailed Fisher’s exact test]) based on the list of DEG (PostFC ≥ 2 or ≤0.5, PPDE > 0.95) in SF3B1K700E or SF3B1wt compared to HD CM (n = 3 per group). C Alternative splicing (AS) signature in SF3B1K700E CM: Volcano plot highlighting differentially spliced genes (DSG) with inclusion level difference (ILD) > |0.2| and pie chart showing distribution of differential splicing event types detectable in SF3B1K700E LR-MDS compared to HD CM using rMATS. Shown below is the pathway and process enrichment analysis of DSG using Metascape (top 20 enriched terms across input DSG). D Cytokine secretion of LPS-stimulated CM was determined by Luminex analysis. Heatmap depicts Log2-transformed normalized median fluorescence intensity values for the indicated cytokines produced by HD, SF3B1K700E, or SF3B1wt (n = 3 per group) LR-MDS classical monocytes following in vitro LPS stimulation. The variant allele frequency (VAF) of SF3B1K700E mutation in CM is shown on the left side. AS alternative splicing, A3SS alternative 3’ splice site, A5SS alternative 5’ splice site, DEG differentially expressed genes, DSG differentially spliced genes, MXE mutually exclusive exon, RI retained intron, SE skipped exon.

In addition, we analyzed alternative splicing in SF3B1K700E versus HD CM using rMATS (Supplementary Table S6). Among the more robust differentially spliced genes (DSG) were various genes previously reported as mis-spliced in SF3B1mut cells, such as BRD9, COASY, and TMEM214 (Fig. 2C, Supplementary Table S6). We could also confirm the previously reported cryptic 3’ splice site for MAP3K7 predicted to undergo nonsense-mediated RNA decay [14], along with decreased MAP3K7 transcript levels in SF3B1K700E CM (Supplementary Fig. S9, Table S5). We did not observe a clear association of the longer IRAK4 isoform with SF3B1K700E (Supplementary Fig. S9), as has been reported previously [15]. DSG were enriched in genes involved in the regulation of defense response and cytokine signaling, next to mRNA metabolism, apoptotic signaling, and mitotic cell cycle (Fig. 2C, Supplementary Table S7). Importantly, 369 out of the 834 DSG were also differentially spliced in SF3B1K700E versus SF3B1wt LR-MDS CM (Supplementary Table S8).

Based on the RNA-seq data pointing to dysregulated cytokine signaling in SF3B1K700E CM, we then assessed their cytokine secretion following in vitro stimulation with the Toll-like receptor 4 agonist lipopolysaccharide (LPS). We found that CM with a heterozygous mutation in SF3B1 (VAF  0.4) responded to LPS stimulation with adequate secretion of pro- (TNF, IL-1β, IL-6, IP-10, MCP-1) and anti-inflammatory cytokines (IL-10, IL-1RA) (Fig. 2D), except for one patient with an extremely high mutation burden (VAF = 0.86) (Fig. 2D). This patient exhibited the highest basal mRNA levels of TNF and IL-6 (Supplementary Fig. S8), which only marginally increased with LPS stimulation. Altogether, at the level of secreted cytokines, we did not observe markedly hyperactivated NF-κB signaling in SF3B1K700E CM following LPS exposure, although we can confirm mis-splicing and reduced mRNA expression of MAP3K7, previously linked to enhanced NF-kB activity [14]. In interpreting our findings, it is important to acknowledge the small sample size for functional assays as a limitation of our study. Therefore, further research with larger sample sizes will be required to address the functional and stimulation context-dependent deficits resulting from mis-splicing of the identified genes.

Phenotypically, the HLA-DRlow/neg CM resemble monocytic myeloid-derived suppressor cells (M-MDSCs), but markers associated with M-MDSC biology were not enriched in SF3B1K700E CM (Supplementary Fig. S8). However, HLA-DRlow/neg CM from one SF3B1K700E LR-MDS patient with co-mutations in TET2 and DNMT3A had a less stimulatory effect on the proliferative capability of autologous CD4+ T cells compared to their HLA-DRhigh counterparts (Supplementary Fig. S10). In light of this, the conversion to HLA-DRlow/neg CM may prevent excessive inflammatory reactions in the tissue driven partly by disproportionate T-cell activation. Further studies comparing HLA-DRlow/neg and HLA-DRhigh CM from patients with an isolated SF3B1K700E mutation will help to clarify their respective roles in the inflammation process.