Erythroid Differentiation Enhances RNA Mis-Splicing in SF3B1-Mutant Myelodysplastic Syndromes with Ring Sideroblasts

Myelodysplastic syndromes (MDS) are clonal myeloid malignancies in which one or multiple branches of hematopoiesis are disrupted due to ineffective differentiation of hematopoietic stem and progenitor cells (HSPC), generally resulting in cytopenia (1). MDSs with ring sideroblasts (MDS-RS) are a slowly progressing low-risk MDS subgroup comprising 16% to 24% of all MDS (2). MDS-RS primarily affect the erythroid lineage and are characterized by extensive dyserythropoiesis and perinuclear accumulation of mitochondria loaded with aberrant ferritin–iron complexes, reflected as the RS phenotype (3).

Hematopoietic stem cell-borne mutations in the SF3B1 gene (pre-mRNA-splicing factor 3b subunit 1) are the primary disease driver in more than 80% of MDS-RS cases (4, 5). Moreover, 80% of SF3B1^mt patients with MDS-RS display none or few comutations, confirming SF3B1^mt as the disease-driving molecular event (6). SF3B1 mutations cause extensive RNA mis-splicing (7, 8), leading to increased nonsense-mediated decay (NMD) of mis-spliced transcripts. Mis-splicing and downstream NMD of ABCB7 and PPOX transcripts in particular is mechanistically associated with RS development by disrupting iron processing (9, 10), which cascades into redox imbalance, obstructed differentiation, and increased apoptosis of erythroid precursors (11, 12).

Despite alternative splicing (AS) and NMD supposedly reducing the survivability of SF3B1^mt cells (particularly erythroid cells), the MDS-RS bone marrow (BM) is hyperplastic and usually displays increased erythropoiesis with abundant RS, which often comprise 50% or more of all erythroid cells. The molecular mechanisms that enable RS survival, expansion, and accumulation in the MDS-RS BM, therefore remain unclear.

Exploring how SF3B1^mt affects erythropoiesis has been pursued through diverse model systems (5, 13–16), each of which is unable to recapitulate full BM/erythroid biology and/or RS development. Crucially, the isolation of viable human RS has not yet been achieved, complicating direct studies into MDS-RS pathobiology.

By exploiting the accumulation of ferric iron inherent to the RS phenotype, we here isolate RS and investigate the entire course of SF3B1^mt erythropoiesis through an integrative multiomic approach. We identify several molecular pathways engaged to minimize the consequences of widespread mis-splicing and extreme oxidative stress, promoting RS survival and accumulation. Finally, we identify increased RNA mis-splicing, decreased NMD, and transcriptome–proteome uncoupling during erythroid differentiation, which altogether demonstrate a role for SF3B1^mt erythropoiesis itself in driving disease pathogenesis.

BM and/or peripheral blood (PB) samples were collected from 36 patients with MDS-RS and 3 patients with MDS non-RS evaluated at Karolinska University Hospital, Huddinge, Sweden. Diagnostic procedures were performed according to the European LeukemiaNet recommendation and WHO classification for myeloid neoplasms (17, 18). As the specific purpose was to dissect the pathobiology of SF3B1^mt MDS-RS, all patients with MDS-RS belonged to the SF3B1^α category (6). RS presence was quantified according to standard clinical practice. Additional samples were collected from a total of 40 healthy normal BM (NBM) donors for control purposes. A deidentified donor and experiment index including clinical and mutational status (5) is provided in Supplementary Data S1. All source material was provided with written informed consent for research use, given in accordance with the Declaration of Helsinki, and the study was approved by the Ethics Research Committee at Karolinska Institutet (2010/427–31/3, 2017/1090–31/4).

Samples were separated by Lymphoprep (STEMCELL Technologies) density gradient centrifugation at 400 × g for 30 minutes, room temperature, to derive a mononuclear cell layer (MNC) and an erythrocyte-rich high-density layer (HD). MNC were cryopreserved in 40% RPMI1640 Glutamax (Thermo Fisher Scientific), 50% inactivated FBS (Thermo Fisher Scientific) and 10% dimethyl sulfoxide (DMSO; Sigma-Aldrich). Cryopreserved MNC were thawed in RPMI1640 Glutamax + 20% FBS + 100 U/mL DNase I (Sigma-Aldrich). HD cells were washed with PBSAG [PBS (Sigma-Aldrich) + 1 mg/mL BSA (Sigma-Aldrich) + 2 mg/mL glucose (Sigma-Aldrich)] and stored at 4°C for a maximum of 1 week.

For RS isolation, packed HD cells were diluted 1:10 with auto-magnetic-activated cell sorting (MACS) Running Buffer (MACS buffer; Miltenyi Biotec) at 4°C and distributed at 5 mL per LS column (Miltenyi Biotec), or 5 × 10⁶ BM MNC were thawed, resuspended in 5 mL MACS buffer, and distributed in one LS column. LS columns were washed with 20 mL MACS buffer and eluted with 5 mL MACS buffer. For antibody-mediated separation, the manufacturer's protocol (Miltenyi Biotec) was followed for separation using CD34, CD235a/GPA, or CD71 MicroBeads.

Cells were analyzed on a CytoFlex S (Beckman Coulter) or analyzed and sorted on a FACS ARIA II Fusion (Becton Dickinson) at the MedH FACS facility of Karolinska Institutet. All steps were performed in FACS buffer (PBS + 2% FBS + 1 mmol/L EDTA) kept at 4°C. All experiments included fluorescent-minus-one (FMO) and single-stained controls. Antibodies/fluorescent reagents utilized are listed in Supplementary Table S1, including the respective RRIDs. Data were analyzed using FlowJo v. 10.7.2 (Becton Dickinson, RRID: SCR_008520).

Cells were cytospun, fixed in methanol for 15 minutes at room temperature, air-dried, and submitted to Karolinska University Hospital for iron staining. Brightfield micrographs were acquired using a Pannoramic MIDI II slide scanner (3D Histech) at 40× with a Hitachi HV-F22 3CCD SXGA camera (Hitachi Kokusai Electric) using Pannoramic Scanner v. 1.17 (3D Histech). Image analysis was performed using QuPath v. 0.2.0m9 (19) and Fiji v. 2.3.0/1.53f51 (20). Fixed-cell immunofluorescence is detailed in Supplementary Materials and Methods.

Droplet digital PCR (ddPCR) was performed with probes for the detection of SF3B1 mutations K700E and K666N (Bio-Rad) as described previously (5). QuantaSoft analysis software v.1.7.4 (Bio-Rad) was used to calculate variant allele frequencies (VAF) based on the Poisson distribution. At least one known mutated sample, one wild-type sample, and one H₂O sample were included as controls in every run.

CD34⁺ HSPC samples, mixed GPA⁺ erythroblast samples, and CD71⁺ PB reticulocyte samples (Ret^PB) were isolated through MACS. RS and siderocytes were obtained through MACS+FACS. Cells were lysed in RLT (Qiagen) + 40 mmol/L dithiothreitol (Sigma-Aldrich) and RNA extraction was performed with RNeasy Micro Kit (Qiagen) with RNase-free DNase treatment according to the manufacturer's protocol. RNA integrity numbers (RIN) were estimated using Agilent RNA 6000 Pico Kits (Agilent Technologies). A minimum RIN value of 6.5 was considered adequate. Additional details are provided in Supplementary Materials and Methods. The count matrix and sample metadata are provided in Supplementary Data S2.

Four experiments were conducted: two 10x Genomics experiments (5 SF3B1^mt MDS-RS patients and 3 NBM donors), one Smart-seq3 (21) experiment (2 MDS-RS donors) and one Smart-seq3xpress (22) experiment (1 MDS-RS donor). Additional details on sample preparation, sequencing, and downstream analysis are provided in Supplementary Materials and Methods. A complete experimental design for the 10x experiments is provided in Supplementary Fig. S1.

Erythroid culture was initiated by thawing and seeding BM MNCs in polyurethane scaffolds as described previously (14). GDF15 levels were measured in both neat (1:1) and diluted (1:10) culture supernatant using a Human GDF15 ELISA Kit (LSBio), based on a standard curve of known GDF15 concentrations.

CD34⁺ HSPCs from 3 NBM donors and 5 SF3B1^mt patients with MDS-RS were treated with RPMI plus 100 ng/mL recombinant GDF15 (NBP2–76204–20ug; Novus Biologicals) or 0.5% v/v sterile H₂O for 1 hour. After preincubation, GDF15 or H₂O were added to a projected concentration of 100 ng/mL, cells were resuspended in MethoCult (H4434; STEMCELL Technologies) and plated in culture dishes. NBM and MDS cells were plated at 4,000 cells/dish and 40,000 cells/dish (due to differential CFU-forming ability; ref. 23), cultured as described previously (5) and counted using a Leica DM inverted microscope (Leica Microsystems).

FACS-separated samples were snap frozen in liquid nitrogen. Cell pellets were lysed with 4% SDS lysis buffer and prepared for MS using a modified version of the SP3 protein clean up and digestion protocol (24). Additional details are provided in Supplementary Materials and Methods and Supplementary Table S2. Expression matrix (centered against a pool of all samples) files, processed files, and downstream analyses are provided in Supplementary Data S3.

Statistical comparisons were performed through ANOVA, multiple comparison-corrected paired and unpaired t tests (Benjamini–Hochberg), and Fisher exact test. Correlation coefficients were calculated and tested for association through Pearson product–moment correlation. Statistical methods specific to the analysis of each high-throughput data format are detailed in their respective sections. Single-cell RNA sequencing (scRNA-seq) analysis was performed with RStudio Server v.1.3.1056 (RRID:SCR_000432) and R v.3.6.3 (RRID:SCR_001905; ref. 25). All other statistical analyses were performed with RStudio v.1.4.1767 (RRID:SCR_000432), R v.4.0.5 (RRID:SCR_001905), Excel v.2204 (RRID:SCR_016137), and GraphPad Prism v.9.4.0 (RRID:SCR_002798).

Preprocessed and deidentified RNA-seq/proteomic data and downstream results generated in this study are available within the article and its supplementary data files. Deidentified raw data for all high-throughput datasets, the corresponding metadata, and intermediate analysis files have been deposited on the Swedish National Data Service's research data repository with restricted access in accordance with the European Union General Data Protection Regulation and are accessible upon reasonable request to the Swedish National Data Service through the respective DOI links (online data descriptions are available in both English and Swedish). Other raw data are available from the corresponding author upon request.

SND-ID 2023–118: Tandem mass tag proteomics of erythroblasts from healthy donors and erythroblasts/ring sideroblasts from patients with MDS-RS. DOI: https://doi.org/10.48723/f2xa-4005

SND-ID 2023–120: Bulk RNA sequencing of BM cells from patients with MDS-RS and healthy donors. DOI: https://doi.org/10.48723/41wa-yv42

SND-ID 2023–121: 10x single-cell RNA sequencing of BM cells from patients with MDS-RS and healthy donors. DOI: https://doi.org/10.48723/nq2a-1e03

SND-ID 2023–122: Smart-seq3 and Smart-seq3xpress single-cell RNA sequencing of BM cells from patients with MDS-RS. DOI: https://doi.org/10.48723/0f0c-p816

To investigate RS development during erythropoiesis, we first pursued erythroblast staging in MNC from MDS-RS and healthy BM samples through co-detection of Band 3 and integrin α4 (Fig. 1A; ref. 26). MDS-RS samples displayed a clear increase in total erythroid frequency, but with relative accumulation at early precursor stages (Pro/Basophilic) and matching depletion at the Orthochromatic stage (Fig. 1B). Morphologic analysis identified a progressive increase in RS frequency with erythroid maturation (Fig. 1C; Supplementary Figs. S2A and S2B, S3, and S4). Assessing mitotic activity showed significantly increased DNA in MDS with no change in Ki-67 expression, indicating uncoupled cell-cycle dynamics corresponding to a dysplastic phenotype (Fig. 1D). Although RS decreased with increasing Ki-67 signal, the median Ki-67 population had equal RS frequencies to diagnostic BM smears (Fig. 1E). Interestingly, these experiments also identified aberrantly decreased CD71 surface expression by MDS precursors as a potential mechanism to avoid iron overload (Fig. 1F). Finally, erythroid enucleation ability was assessed through 3D cell culture of MDS-RS cells (14). RS were found to enucleate at equal rates to normoblasts from the same patients (Fig. 1G; Supplementary Fig. S5), fully recapitulating progression through erythropoiesis.

Further assessment of RS biology was challenging in heterogeneous cell populations. However, as hemozoin-rich red blood cells (RBC) are commonly separated by reagent-free magnetic isolation (MACS; ref. 27), we hypothesized that iron burden could similarly be exploited to isolate SF3B1^mt RS/siderocytes (Fig. 2A). MACS was performed on MNC (viably frozen) and high-density (HD, fresh) fractions and achieved significant RS enrichment in both (Fig. 2B). FACS of HD-isolated cells further enriched a high-purity (99.3%) RS population at significantly higher abundance than MNC (Fig. 2C; Supplementary Fig. S6) and was the method utilized throughout this study unless otherwise specified. Isolated RS significantly correlated with morphologic BM RS frequencies (Fig. 2D) and were confirmed as fully clonally involved (Fig. 2E). Validating our previous flow cytometric and morphologic analysis of MDS-RS erythroblasts, purified RS displayed highly significantly decreased surface CD71 (Fig. 2F; Supplementary Figs. S7A and S7B) as well as identifiable nuclear localization of Ki-67 (Fig. 2G; Supplementary Fig. S8).

Reagent-free magnetic isolation was tested in PB samples to detect circulating siderocytes. Surprisingly, M+FACS assessment of PB HD cells identified a substantial number of RS (Fig. 3A). No previous reports have described appreciable numbers of circulating RS in MDS-RS PB (3). However, peripheral RS were a ubiquitous observation correlating well in abundance with BM RS isolates (Fig. 3B). Where matched clinical data were available, PB and BM RS isolates were positively correlated with diagnostic smear RS (Fig. 3C) and serum erythropoietin (Fig. 3D), and negatively correlated with hemoglobin (Fig. 3E). Interestingly, the identification of PB RS in FC also detected increased DNA content RS in erythropoiesis-stimulating agent (ESA)-treated patients (Fig. 3F). This was morphologically validated in several cases (Fig. 3G and H), identifying a positive effect of ESAs on RS (increasing survival of dysplastic cells or increasing mitosis).

We next investigated SF3B1^mt erythropoiesis through RNA-seq, comparing MDS-RS cell populations against healthy individuals at bulk (Fig. 4A; Supplementary Table S3) and single-cell level (Fig. 4B; Supplementary Figs. S1, S9, S10A and S10B, S11, and S12). We first focused on CD34⁺ HSPC, as these constitute the compartment of origin in MDS-RS (5). As reported previously (9, 28), differential gene expression (DE) analysis comparing MACS-enriched CD34⁺ cells between MDS-RS and healthy individuals identified overexpression of erythroid genes and underexpression of known SF3B1^mt targets (Fig. 4C, left; Supplementary Data S4). However, no increased erythroid transcripts were observed in transcriptomically identified HSPCs (Fig. 4C, right). AUCell analysis (29) was then conducted to assess a potential proerythroid signature (top 100 erythroblast development genes; ref. 30) and similarly demonstrated no difference between MDS-RS and NBM HSPC clusters (Fig. 4D).

In light of the efficiency of RS isolation through reagent-free MACS, we investigated whether antibody-mediated MACS could induce sample contamination. Indeed, CD34⁺-enriched MDS-RS cells displayed a Lin⁻CD34⁻GPA⁺ population averaging ∼90% RS/siderocytes, which was absent in MDS samples without RS (Fig. 4E–G), highlighting important limitations in interpreting magnetically enriched cell data from MDS-RS. (7–10) To circumvent these, we evaluated the molecular signature of CD34 transcript-positive HSPC in scRNAseq (Fig. 4H; Supplementary Fig. S13; Supplementary Data S5). Transcriptomically defined HSPC subset frequencies were unchanged (Fig. 4I), matching previous data of phenotypical HSPC in MDS-RS (5). Interestingly, few DE genes were detected in CD34⁺ HSPCs despite these maintaining RNA mis-splicing, and the only enrichment identified was a protein/ER stress-associated response (Fig. 4J). Given these limited HSPC results, we next focused our approach on specifically investigating erythroid biology.

RS were used as the primary point of comparison against NBM erythroblasts due to being phenotypically and clonally engaged. Bulk RNA-seq identified extensive underexpression of cell-cycle-related genes, matching our initial characterization (Fig. 5A). Most importantly, overexpressed genes in bulk RNA-seq mapped to an MDS-specific scRNA-seq cluster, which was overrepresented in RS and absent in healthy donors (Fig. 5B). This cluster presented a more aberrant transcriptomic phenotype in purified HD-RS versus MNC-RS (Fig. 5C). However, and beyond this MDS-specific cluster, RS clearly encompassed all stages of erythroid differentiation.

DE analysis was then performed between RS/NBM within each scRNA-seq transcriptomic cluster to define differentiation-independent alterations (Fig. 5D; Supplementary Data S6). This approach identified a major enrichment of stress response pathways (Fig. 5E), particularly related to autophagy, proteasomal processing, and mitophagy. Metabolic changes were also evident, particularly the significant enrichment of several antioxidant pathways (glutamine/ROS response); together, these indicate an active transcriptional state in RS to maintain cell homeostasis against overactive mis-splicing and oxidative stress.

Erythroid-specific genes were also heavily affected in this comparison, with RS displaying loss of expression of hemoglobin chains HBA2 and HBD, a hallmark for iron deficiency anemia (Supplementary Fig. S14; ref. 31), and decreased HBB expression in cells with higher RS-associated transcripts (Supplementary Fig. S15). Conversely, erythroid gene expression was higher in peripherally circulating RS, suggesting that this population is either more differentiated or less phenotypically affected (Supplementary Figs. S16A and S16B and S17).

Focusing on bulk expression of ABCB7, the major SF3B1^mt-induced NMD-targeted event in SF3B1^mt MDS-RS (9, 28), we found that its abundance was evidently decreased in HSPCs (matching previous reports) but significantly increased in RS/siderocytes both in bulk and scRNA-seq (Fig. 5F; Supplementary Fig. S18) and identified lower levels of the cryptic ABCB7 splice site in RS as compared with CD34⁺ and GPA⁺ cells (Fig. 5G).

Both RNA-seq datasets identified overexpression of Growth Differentiation Factor 15 (GDF15) as a major occurrence (Fig. 5H). GDF15 is a stromal/erythroid stress factor previously shown to be increased in MDS-RS patient sera (32) and reported to suppress HSPC growth in culture (33). When in vitro culture was used to induce stress erythropoiesis, cultured SF3B1^mt cells still secreted more than 9× GDF15 than normal cells, highlighting GDF15 overexpression is disease-related (Fig. 5I). A colony-forming unit (CFU) experiment to evaluate GDF15 effects on HSPC suppression displayed variable results in MDS-RS samples, but identified a consistent and significant decrease in NBM erythroid/total CFU-forming ability (Fig. 5J), linking GDF15 secretion by RS to an effect on upstream HSPC biology.

Given the surprising patterns of ABCB7 expression/splicing in RS (Fig. 5F–H), we next focused on a broader analysis of AS in the bulk RNA-seq data comparing healthy individuals and SF3B1^mt MDS-RS [HSPC, RS, vs. erythroblasts (nucleated erythroid, N) and siderocytes vs. reticulocytes (anucleate erythroid, A)]. RS presented substantially more differential AS events; 63 genes were AS in both HSPC/RS, and only 7 genes were AS in all three groups (Fig. 6A).

Shared AS events between HSPC/RS were enriched in RNA splicing-associated genes (Fig. 6B), in line with previous SF3B1^mt studies (9, 34). However, genes that were specifically AS in each population significantly enriched for cell type identity, reflecting distinct transcriptomic programs. AS event types also progressively shifted toward predominantly exon skipping events during erythroid differentiation, compared with the initial skew of alternative 3′-splicing and intron retention events in HSPCs (Fig. 6C; Supplementary Figs. S19A and S19B).

Focusing on literature-validated SF3B1^mt AS events (9, 35), we found that decreased mis-splicing of ABCB7 was a singular case in RS; in fact, every other AS event displayed a constant or increased median level of the cryptic transcript in RS/erythroblasts as compared with HSPC, including known NMD-targeted AS (Fig. 6D). Similarly, although new AS sites affecting several genes implicated in MDS (36) and congenital sideroblastic anemia (CSA; SLC25A38; ref. 37) were detected in HSPC at very low levels, cryptic mis-spliced transcripts of these genes increased during erythropoiesis to become the predominant transcript form in RS, with many of these inducing frameshifts and/or inclusion of premature termination codons (PTC; Fig. 6E).

Functional analysis of cryptic 3′ site locations in HSPCs confirmed their enrichment between −30 and −10 bp upstream of the canonical 3′ site (Fig. 6F), as reported previously (38). Conversely, RS displayed events in the same interval but had a much larger distance distribution, indicating reduced fidelity for splice site recognition. This is functionally supported by differentially altered RS nucleotide frequencies upstream of and surrounding the cryptic 3′ splice site (Fig. 6G), as well as a much larger proportion of frameshift/PTC events (Fig. 6H and I).

We next hypothesized that the highly aberrant AS profile in RS could be caused by altered splicing/NMD dynamics in erythropoiesis. RNA velocity analysis (39) of spliced transcript percentages in the 10x clusters identified increased splicing rates already in early erythroblasts (Fig. 6J). We were concerned that the high erythroid transcript burden of ribosomal/globin genes could skew these data; however, this trend remained after discarding these transcripts from analysis (Fig. 6K). Importantly, NBM and MDS samples behaved identically, highlighting the role of differentiation and not disease in this process. As velocity analysis of the 10x samples was based solely on clusters identified through computational prediction, we then confirmed the same splicing dynamics in FACS-purified HSPC/erythroblast populations through Smart-seq3xpress (Fig. 6L and M; Supplementary Figs. S20, S21, and S22A–S22C).

Cycloheximide treatment of HSPC and erythroblasts was then pursued to evaluate NMD dynamics as originally reported in HSPC by Shiozawa and colleagues (8). Overall, NMD-targeted cryptic transcripts (either SF3B1^mt-associated or of endogenous origin; ref. 40) were increased in frequency by cycloheximide treatment in HSPC alone (Fig. 6N), through which we determine erythroid differentiation displays decreased NMD activity. This is consistent with the significantly higher proportions of NMD-targeted AS events in RS (Fig. 6D). Thus, taken together, we conclude that erythroid differentiation magnifies the downstream effects of SF3B1^mt.

Semiquantitative proteomics of healthy and MDS-RS erythroblasts/RS were performed and integrated with RNA-seq to functionally assess the outcome of SF3B1^mt expression/splicing (Fig. 7A; Supplementary Figs. S23A and 23B). This analysis identified four major signatures of differential RNA/protein expression, which validated our previous scRNA-seq results (Supplementary Table S4). The same pro-survival pathways from scRNA-seq were validated as overexpressed, together with a highly significant decrease in cell cycle–associated proteins; however, the proteomic dataset also identified a specific and significant decrease in NMD target-degrading proteins (Supplementary Fig. S24). Importantly, GDF15 was significantly overexpressed in both erythroblasts and RS, further supporting previous findings (Supplementary Fig. S25).

Cases of RNA/protein uncoupling, that is, genes with increased RNA expression and negative or no correlation between RNA and protein abundance, were particularly interesting as candidate targets of dysfunctional AS events. Indeed, this gene subset encompassed several known SF3B1^mt-affected genes (Fig. 7B; refs. 9, 38), as well as several of the newly identified RS-enriched AS events affecting driver genes implicated in MDS/CSA pathogenesis (Fig. 6E). In adition, pathway analysis of this gene set identified specific enrichment of the TP53 pathway, including several proapoptotic effectors; and also the detection of EIF2AK1, identified in a previous report to be highly overexpressed in single-cell RNA-seq data of MDS-RS erythroblasts (41) and yet here underexpressed at protein level in both MDS-RS erythroblasts and RS, potentially due to multiple mis-splicing events (Supplementary Figs. S26A–S26D).

Investigating ABCB7 and MAP3K7 protein levels to further dissect this mechanism demonstrated significant underexpression of both proteins in MDS-RS erythroblasts and RS in comparison with healthy erythroblasts. However, both were significantly increased in RS as compared with MDS-RS erythroblasts (Fig. 7C). Therefore, the increased RNA mis-splicing detection in RS resulting from decreased NMD is partially compensated through RNA overexpression. However, this process fails to completely normalize protein abundance and likely also retains dysfunctional proteins in the cell generated via translation of the nondegraded mis-spliced transcripts.

MDM2, BAX, and FAS were selected as major elements for assessing the downstream consequences of TP53 pathway mis-splicing due to their key roles in apoptosis. All were confirmed to be overexpressed in 10x scRNA-seq (Fig. 7D, top). However, diverse mis-splicing events were identified in both RNA-seq datasets as exclusive to RS (rMATS FDR < 10⁻⁵; Fig. 7D, bottom; Supplementary Fig. S27). Several would canonically undergo NMD, and thus are detrimental to production of functional proteins; however, the cryptic transcripts increase in RS to even become the main form in the case of FAS. We thus conclude that RNA mis-splicing affects apoptotic regulation in SF3B1^mt RS, and for the first time identify SF3B1^mt-associated mis-splicing of TP53 pathway genes.

Although in vitro and in vivo studies of SF3B1^mt erythropoiesis have proven successful in modeling HSPC pathobiology and recreating RS generation, the molecular mechanisms driving the growth and survival of SF3B1^mt RS in the MDS-RS BM and the competitive advantage of SF3B1^mt cells have remained elusive (5, 14, 15). Through an integrative multiomics approach, we investigated SF3B1^mt MDS-RS erythropoiesis at a cell and molecular level to address these open questions.

We established a novel method for viable RS and siderocyte isolation from patient BM samples, enabling exploration of their disrupted erythroid development and rendering us the unique possibility to explore the entire process of SF3B1^mt erythroid development from HSC to RS. This method was also validated in PB samples, identifying circulating RS as a relevant clinical observation correlating in abundance with BM RS burden and hemoglobin/serum EPO levels.

An important outcome associated with reagent-free magnetic isolation of RS was the finding that CD34⁺ MACS enrichment (a core experimental approach for HSPC isolation; refs. 7–10, 28) indirectly isolates a significant number of RS and siderocytes due to their high iron content, creating an artificial erythroid subpopulation, which was absent in matched MDS samples without RS. This subpopulation fundamentally underlies the bulk transcriptomic erythroid signature conventionally associated with MDS-RS HSPCs, which we validated in our bulk RNA-seq data and yet failed to verify in a detailed scRNA-seq analysis including enriched HSPCs. Indeed, despite showing a clear functional impairment in long-term culture-initiating cell and CFU assays (5), true MDS-RS HSC/MPPs displayed few alterations in gene expression and no detectable differences in commitment, subset frequencies, or erythroid identity. However, a previous study focusing on scRNA-seq and scATAC-seq analysis of MDS-RS HSPCs identified an increase in erythroid versus megakaryocyte commitment of megakaryocyte-erythroid progenitors occurring through differential chromatin usage (41). We thus speculate that combining these insights with additional data regarding RNA splicing and protein production at HSPC subset resolution will be instrumental towards understanding the functional consequences of SF3B1^mt and downstream RNA mis-splicing on early hematopoietic development.

In evaluating erythroblast and RS biology, we demonstrated that RS constitute a living, differentiating, and molecularly active population, with decreased cell-cycle progression as the major sign of dysfunction and which may explain the marked abundance of RS in the MDS-RS BM. In comparison with approaches focusing on the general and highly heterogeneous erythroblast population, the single-cell and proteomic analysis of purified RS provides a much increased granularity to the diverse homeostatic mechanisms engaged to survive oxidative and RNA splicing stress, including the surprising protein rescue of known mis-spliced and NMD-targeted genes ABCB7 and MAP3K7 (9).

We explored the molecular mechanisms underlying these changes and define an intensified SF3B1^mt mis-splicing panorama in RS. This includes the novel identification of PTC induction and general mis-splicing and dysregulation of several MDS-implicated driver genes (36), each of which may constitute a target for further studies. We thus speculate that modified RNA splicing and NMD dynamics in erythroid cells combine to switch “quality control” from RNA to protein, supported by the finding that RS sensitivity to proteasome inhibition by bortezomib has been demonstrated in a clinical trial (42).

Importantly, we demonstrate that the observed uncoupling of RNA and protein expression enhances detection and functional validation of RNA mis-splicing events and in turn allows us to directly associate SF3B1 mutations with disruption of proapoptotic genes in the TP53 pathway. In addition to providing a potential explanation for why RS accumulate in the BM, this finding may also help to further dissect a major question in the study of MDS: what explains the increased risk conferred by SF3B1 mutations in TP53-mutant settings such as MDS^5q−, AML, and other malignancies including chronic lymphocytic leukemia, in comparison with the relatively benign course of MDS-RS with isolated SF3B1 mutations (43, 44)? More specifically, could SF3B1^mt mis-splicing of TP53 pathway modulators and proapoptotic genes specifically enhance the cell's adaptability when TP53 function is already compromised? These are complex questions that beckon the need to evaluate how complex genetic profiles and RNA splicing act in tandem to potentiate new interactions beyond the effects of a single driver mutation.

An additional major finding of our multiomics studies was the repeated identification of SF3B1^mt RS as a key source of GDF15. Previously reported as a marker of increased erythropoiesis in MDS-RS (45), we here instead confirmed a detrimental impact of this cytokine on wild-type HSPC biology. Chronic myeloproliferative neoplasms develop over decades, and we recently reported preliminary data indicating this is also the case in MDS-RS (46, 47). On the basis of our data, we hypothesize that SF3B1^mt cell effects on their surroundings may contribute to clonal expansion and clinical disease evolution. It is thus tantalizing to speculate that RS could function as disease-augmenting “foot soldiers,” similarly to Reed–Sternberg cells in Hodgkin lymphoma (48), and further exploration of how SF3B1^mt cells affect their surrounding microenvironment and wild-type cells will be essential.

In conclusion, our characterization of SF3B1^mt erythropoiesis constitutes a unique platform for the study of MDS-RS, providing novel insights into the unexpectedly active biology of the “dead-end” RS and enabling further investigation of disease pathogenesis and treatment avenues.

P.L. Moura reports grants from Cancerfonden (Swedish Cancer Society) during the conduct of the study. W.W. Kretzschmar reports personal fees from PerkinElmer/Revvity, Inc. outside the submitted work. M. Hagemann-Jensen reports holding shares in Xpress Genomics AB. C. Ziegenhain reports holding shares in Xpress Genomics AB. V. Lundin reports grants from Vetenskapsrådet (Swedish Research Council) during the conduct of the study. P.S. Woll reports grants from Knut and Alice Wallenberg Foundation during the conduct of the study and grants from Cancerfonden and Radiumhemmets Forskningsfonder outside the submitted work. R. Sandberg reports holding shares in Xpress Genomics AB. E. Hellström-Lindberg reports grants from Cancerfonden (Swedish Cancer Society), Knut and Alice Wallenberg Foundation, and Vetenskapsrådet (Swedish Research Council) during the conduct of the study. No disclosures were reported by the other authors.

P.L. Moura: Conceptualization, data curation, software, formal analysis, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. T. Mortera Blanco: Formal analysis, investigation, writing–review and editing. I.J. Hofman: Investigation, writing–review and editing. G. Todisco: Software, formal analysis, investigation, writing–review and editing. W.W. Kretzschmar: Software, writing–review and editing. A. Björklund: Investigation, writing–review and editing. M. Creignou: Investigation, writing–review and editing. M. Hagemann-Jensen: Software, investigation, writing–review and editing. C. Ziegenhain: Software, investigation, writing–review and editing. D. Cabrerizo Granados: Investigation, writing–review and editing. I. Barbosa: Investigation, writing–review and editing. G. Walldin: Investigation, writing–review and editing. M. Jansson: Investigation, writing–review and editing. N. Ashley: Investigation, writing–review and editing. A.J. Mead: Supervision, investigation, writing–review and editing. V. Lundin: Supervision, writing–review and editing. M. Dimitriou: Validation, writing–review and editing. T. Yoshizato: Resources, software, writing–review and editing. P.S. Woll: Resources, supervision, validation, writing–review and editing. S. Ogawa: Supervision, writing–review and editing. R. Sandberg: Supervision, methodology, writing–review and editing. S.E.W. Jacobsen: Resources, supervision, validation, writing–review and editing. E. Hellström-Lindberg: Conceptualization, resources, data curation, supervision, funding acquisition, validation, writing–original draft, project administration, writing–review and editing.

The authors would like to thank the patients and healthy donors for their willingness to participate in this research. The authors would like to acknowledge support from the Karolinska Institutet MedH Flow Cytometry Core Facility for assistance with FACS experiments; the KI Bioinformatics and Expression Analysis Core Facility for assistance with scRNAseq experiments; the facility of Clinical Proteomics Mass Spectrometry at Karolinska University Hospital and Science for Life Laboratory for assistance in mass spectrometry and data analysis; the KI Single-Cell Core Facility (SICOF) for assistance with Smart-seq3 scRNAseq; the KI Research Data Office for assistance with data management and deposition; and support from the National Genomics Infrastructure in Stockholm funded by Science for Life Laboratory, the Knut and Alice Wallenberg Foundation and the Swedish Research Council, and SNIC/Uppsala Multidisciplinary Center for Advanced Computational Science for assistance with massively parallel sequencing and access to the UPPMAX computational infrastructure. This project received grant support from Cancerfonden / Swedish Cancer Society (grant number 21 0340 to P.L. Moura and 19 0200 to E. Hellström-Lindberg); Vetenskapsrådet/Swedish Research Council (grant number 2020–01902 to V. Lundin and 211133 to E. Hellström-Lindberg); and from the Knut and Alice Wallenberg Foundation (grant number 2017.0359 to E. Hellström-Lindberg).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.