Smad2/3‐pathway ligand trap luspatercept enhances erythroid differentiation in murine β‐thalassaemia by increasing GATA‐1 availability

Abstract In β‐thalassaemia, anaemia results from ineffective erythropoiesis characterized by inhibition of late‐stage erythroid differentiation. We earlier used luspatercept and RAP‐536 protein traps for certain Smad2/3‐pathway ligands to implicate Smad2/3‐pathway overactivation in dysregulated erythroid differentiation associated with murine β‐thalassaemia and myelodysplasia. Importantly, luspatercept alleviates anaemia and has been shown to reduce transfusion burden in patients with β‐thalassaemia or myelodysplasia. Here, we investigated the molecular mechanisms underlying luspatercept action and pSmad2/3‐mediated inhibition of erythroid differentiation. In murine erythroleukemic (MEL) cells in vitro, ligand‐mediated overactivation of the Smad2/3 pathway reduced nuclear levels of GATA‐1 (GATA‐binding factor‐1) and its transcriptional activator TIF1γ (transcription intermediary factor 1γ), increased levels of reactive oxygen species, reduced cell viability and haemoglobin levels, and inhibited erythroid differentiation. Co‐treatment with luspatercept in MEL cells partially or completely restored each of these. In β‐thalassaemic mice, RAP‐536 up‐regulated Gata1 and its target gene signature in erythroid precursors determined by transcriptional profiling and gene set enrichment analysis, restored nuclear levels of GATA‐1 in erythroid precursors, and nuclear distribution of TIF1γ in erythroblasts. Bone marrow cells from β‐thalassaemic mice treated with luspatercept also exhibited restored nuclear availability of GATA‐1 ex vivo. Our results implicate GATA‐1, and likely TIF1γ, as key mediators of luspatercept/RAP‐536 action in alleviating ineffective erythropoiesis.


and in disorders
with erythroid maturation defect (EMD), such as β-thalassaemia and myelodysplastic syndromes (MDS). 3,4 IE in β-thalassaemia is caused by mutations in the β-globin gene leading to defective haemoglobin production, 5 whereas IE in MDS is caused by varied mutations in haematopoietic lineage cells. 6 Despite the diversity of factors underlying IE, the common outcome in the two aforementioned diseases is inhibition of terminal erythroid differentiation and accumulation of immature erythroid precursors in erythropoietic tissues. 3,4 Currently, the mainstay supportive treatment for patients with IE is repeated blood transfusion, which leads to progressive iron accumulation in multiple tissues and complications from iron overload despite iron chelation therapy. 7 The TGF-β superfamily plays a critical role in regulating haematopoiesis in normal and disease states by controlling cellular proliferation, differentiation and apoptosis. 8,9 This superfamily comprises several dozen ligands, including TGF-β isoforms, activins, growth differentiation factors (GDFs) and bone morphogenetic proteins (BMPs), which engage promiscuously with multiple combinations of receptors to produce signals of remarkable complexity. 10 In brief, these ligands trigger formation of heteromeric complexes between specific type I and type II transmembrane receptors, leading to phosphorylation of cytoplasmic Smad proteins, such as Smad2/3 or Smad1/5/8. 11 Such activated Smads (pSmads) then form an oligomeric complex with Smad4 (co-Smad) and enter the nucleus to directly alter gene transcription in combination with transcription factors, chromatin-remodelling complexes and histone-modifying enzymes. 8,12 Importantly, Smad signalling occurs along two main branches that often mediate opposing functional outcomes. Activins, GDF8, and GDF11 signal through Smad2 or Smad3 (Smad2/3 pathway), whereas BMPs and other GDFs typically signal through Smad1, Smad5, or Smad8 (Smad1/5/8 pathway). TGF-β signals primarily through Smad2/3 but can also use Smad1/5/8 in certain contexts. 12 Smad2/3 signalling has emerged as an important regulator of erythropoiesis, exerting an inhibitory influence under normal steady-state conditions. 8,9,13,14 Additionally, overactivation or dysregulation of Smad2/3 signalling has been implicated in diseases characterized by impaired erythroid differentiation and IE, 9,[15][16][17][18] There is evidence for further branching of the Smad2/3 pathway because of the ability of pSmad2/3 to bind alternatively to Smad4 or TIF1γ (transcription intermediary factor 1γ), also known as TRIM33 (tripartite motif-containing 33). In human haematopoietic/progenitor cells, where TGF-β inhibits proliferation and stimulates erythroid differentiation, pSmad2/3-TIF1γ complexes mediate the differentiation response while pSmad2/3-Smad4 complexes mediate the anti-proliferative response. 19 Evidence has also emerged that formation of pSmad2/3-TIF1γ complexes can be regulated by TIF1γ phosphorylation 20 and that TIF1γ can influence Smad4 availability thorough its activity as a Smad4 ubiquitin ligase. [21][22][23] Importantly, TIF1γ has been shown to stimulate expression of the master erythroid transcription factor GATA-1 and its downstream erythroid-signature genes in both mice and zebrafish. [24][25][26] Thus, competition between TIF1γ and Smad4 for pSmad2/3 mediates TGF-β-induced commitment of haematopoietic stem cells to the erythroid lineage, raising the possibility that a related mechanism may act at later stages to co-ordinate erythroid differentiation.
We previously investigated the role of Smad2/3 signalling in terminal erythroid differentiation using luspatercept (ACE-536)-a modified extracellular domain of human activin receptor type IIB (ActRIIB) attached to a human IgG1 Fc domain-and its murine analogue. These agents produce sustained elevations of haemoglobin levels in a wide variety of settings, including normal rodents, nonhuman primates and healthy volunteers 18,27 as well as in murine models of MDS or β-thalassaemia and patients with these diseases. 17,18,28 Luspatercept and RAP-536 elevate RBC levels by a mechanism distinct from that of erythropoietin because they enhance maturation of erythroid precursors without first increasing numbers of erythroid progenitors. These agents sequester and neutralize several ligands of the Smad2/3 signalling pathway, including activin B, GDF8 and GDF11, thus leading to inhibition of Smad2/3 signalling. 18 However, the downstream molecular effects of these agents remain to be identified.
In the present study, we investigated the molecular mechanism by which luspatercept-mediated inhibition of Smad2/3 signalling promotes erythroid differentiation. Our results indicate that ligand-mediated overactivation of the Smad2/3 pathway in murine erythroleukemic cells impedes their differentiation, whereas co-treatment with luspatercept restores nuclear GATA-1 levels together with multiple indicators of differentiation, likely through a mechanism favouring nuclear localization of TIF1γ.
Corroborating evidence was obtained in β-thalassaemic mice, in which RAP-536 up-regulates Gata1 and its target gene signature in erythroid precursors as determined by transcriptomic and gene set enrichment analyses restores nuclear levels of GATA-1 in erythroid precursors.

| Murine model of β-thalassaemia
Hbb th1/th1 and Hbb th3/+ mice and C57BL/6 wild-type (WT) strains were obtained from Jackson Laboratories and maintained at Acceleron Pharma. Genotyping was carried out at Transnetyx Inc, TN. All procedures used in this study were performed according to protocols that were previously approved by the Acceleron Pharma Institutional Animal Care and Use Committee.

| Luspatercept and RAP-536
As previously described, luspatercept consists of a modified human ActRIIB ECD (residues 24-131) linked to the human IgG1 Fc domain. 18 RAP-536, the murine analogue of luspatercept, was generated similarly as described. 18

| Real-time quantitative PCR assay
Cells were placed into RNAprotect Cell Reagent (Cat. #76526, Qiagen), and RNA was isolated by using an Aurum total RNA mini kit (Cat. #732-6820, Bio-Rad). cDNA was synthesized by using an iScript cDNA kit (Cat. #172-5037, Bio-Rad). PrimePCR Primers (Bio-Rad) and housekeeping control genes were used for data acquisition. Data analysis was done with the CFX Manager Software (Bio-Rad).

| Transcriptome analysis
After cell sorting, sorted populations were used for RNA extraction with an RNeasy mini kit (Cat. #74106, Qiagen) as per the manufacturer's protocol. RNA quality was analysed via a BioAnalyzer and checked for RNA degradation prior to running RNA sequencing using Illumina paired-end sequencing approach. Sequencing libraries were generated from double-stranded cDNA using the Illumina TruSeq kit according to the manufacturer's protocol. Library quality control was checked using the Agilent DNA High Sensitivity Chip and qRT-PCR. High-quality libraries were sequenced on an Illumina HiSeq 2500 platform. To achieve comprehensive coverage for each sample, we generated 40-50 million paired-end reads.

| Statistical analyses
GraphPad Prism was used for one-way ANOVA, Tukey's, and nonpaired Student's t test when appropriate.

| Data analysis
The raw sequencing data were processed to remove any adaptor, PCR primers and low-quality transcripts using FASTQC and Trimomatic software. These high-quality, clean reads were aligned against mouse genome (10 mm) using tophat2 and bowtie2 packages (http://tophat.cbcb.umd.edu/). Gene expression measurement was performed from aligned reads by counting the unique mapped reads.
The read count-based gene expression data were normalized based on library complexity and gene variation using Bioconductor EdgeR package. The normalized count data were compared among groups using a negative binomial model to identify differentially expressed genes. The differentially expressed genes were identified based on multiple tests corrected P value and fold change. Genes were considered significantly differentially expressed if the adjusted P-value was < .05 and absolute fold change > 1.5. Unsupervised analysis was performed using principal component analysis (PCA), which projects multivariate data objects onto a lower dimensional space while retaining as much of the original variance as possible.

| Gene set enrichment analysis to understand molecular mechanism of RAP-536 treatment
As a complementary approach, we also performed analysis on normalized RNASEQ data using gene set enrichment analysis (GSEA) to determine whether a priori defined sets of genes showed statistically significant, concordant differences between transcriptome profile of control and RAP-536 treated samples. GSEA can be more powerful than single-gene methods for understanding effects of RAP-536 on pathways and biological gene set levels resulting in repairing defects in erythroid maturation in thalassaemia. GSEA was performed using the GSEA-R, a Bioconductor implementation of GSEA. 29 GSEA was performed on pre-ranked gene lists based on log fold change representing the effect of RAP-536. We have performed the enrichment analysis using the canonical pathways, biological processes and transcription factor targets gene sets derived from MSigDB2.0. 29,30 The gene sets with a nominal P-value (NPV) less than 5% after 1000 random permutations were considered significantly altered.

| Comparison of RAP-536 treatment signature with GATA-1-regulated genes
Genome-level GSEA demonstrated that RAP-536 treatment significantly alerted GATA-1-regulated genes. To further understand the role of GATA-1 in RAP-536 treatment, we performed a comparison of RAP-536 and GATA-1 transcriptome profiles using individual genebased, and gene set enrichment-based approaches. GATA-1-mediated gene activation and repression signatures were obtained from the previous report. 31 GSEA was run with 1000 permutations, and a classic statistic. Normalized enrichment score (NES) and nominal P values were measured to determine the significance of enrichment.

| Pathway enrichment analysis
Pathway enrichment analysis was performed to identify pathways that are regulated/co-regulated by luspatercept treatment and GATA-1 transcription factor. Ingenuity Pathway Analysis (IPA 8.0, Qiagen) was used to identify the pathways that are significantly affected by RAP-536 and GATA-1 co-regulated genes. The knowledge base of this software consists of functions, pathways and network models derived by systematically exploring the peer-reviewed scientific literature. A detailed description of IPA analysis is available at the Ingenuity Systems' web site (http//www.ingen uity.com). A Pvalue is calculated for each pathway according to the fit of users' data to the IPA database using one-tailed Fisher exact test. The pathways with P-values < .05 were considered significantly affected.
The genes from enriched pathways and GSEA gene set were merged into functional network modules on the interaction information obtain from public databases such as MIPS, DIPS and MsigDB 2.0. The network was developed and visualized using Cytoscape: An Open Source Platform for Complex Network Analysis and Visualization. 32

| Luspatercept inhibits Smad2/3 phosphorylation in murine erythroleukemic cells
To investigate the mechanism by which Smad2/3 signalling regulates erythroid differentiation, we used murine erythroleukemic (MEL) cells, a well-characterized model system that undergoes erythroid differentiation in vitro after exposure to DMSO. 33 We first tested the ability of TGF-β superfamily ligands to induce phosphorylation  Figure S1A). Co-treatment with luspatercept blocked pSmad3 induction by GDF11 at both time points and also reduced pSmad3 induction by activin B (Figure 1). As expected, co-treatment with luspatercept also blocked pSmad3 induction by GDF8 at 60 minutes after treatment but did not inhibit pSmad3 induction by activin A (Figure 1), consistent with the previously demonstrated inability of luspatercept to bind activin A in a cell-free system or inhibit activin A-mediated signalling in a cellular system. 18

| Luspatercept reverses pSmad2/3-mediated reduction in nuclear TIF1γ levels in differentiating MEL cells
Next, we studied the effect of luspatercept on GDF11-mediated changes in the subcellular localization of phosphorylated-Smad3, Smad4 and TIF1γ in MEL cells. When Smad2/3 is phosphorylated in the cytoplasm, it binds Smad4 and this complex translocates to the nucleus where it regulates transcription of target genes. 34 By western blotting analysis, we found increased nuclear localization of pSmad3 in GDF11 treated MEL cells at various time points up to 24 hours compared to control treated cells. Lamin B1 was used as a protein loading control for nuclear extracts ( Figure S1B).
Treatment of luspatercept in combination with GDF11 inhibited the nuclear localization of pSmad3 levels in a time-dependent manner ( Figure S1B). Consistent with the detection of pSmad3 in nuclear extracts, immunofluorescence microscopic analysis of MEL cells treated with GDF11 alone for 24 hours increased subcellular localization of Smad4 as determined by mean fluorescence intensity using DAPI counterstain to identify cell nuclei (Figure 2A). Nuclear Smad4 levels trended higher with GDF11 treatment alone compared to DMSO control and lower with co-treatment of GDF11 and luspatercept, narrowly missing statistical significance ( Figure 2B). Similarly, we also found increased levels of Smad4 by Western blot analysis in nuclear extracts following GDF11 treatment compared to control, and luspatercept treatment reduced pSmad3 and Smad4 levels in the nucleus ( Figure 2C). We then performed a similar experiment in MEL cells to determine effects of GDF11, with and without luspatercept, on the subcellular localization of TIF1γ. Previous studies have indicated that nuclear pSmad2/3-TIF1γ complexes promote commitment to erythroid differentiation in haematopoietic progenitor cells, 19 and we hypothesized that a related mechanism may influence later-stage differentiation of erythroid precursors. Analysis of MEL cells by immunofluorescence microscopy revealed that, under control conditions (DMSO only), TIF1γ was detectable in the nucleus in approximately 80% of such cells ( Figure 3A,B). GDF11 treatment caused a significant reduction in the percentage of cells with nuclear localization of TIF1γ ( Figure 3A,B). Photomicrograph analysis indicated that co-treatment with luspatercept increased the TIF1γ (magenta) immunofluorescence in the nucleus compared to that with GDF11 alone (Figure 3A,B). The findings were also corroborated by Western blot analysis in cytosolic and nuclear extracts ( Figure 3C).
Lamin B1 and GAPDH served as protein loading controls for nuclear and cytoplasmic extracts, respectively ( Figure 3C). GDF11 treatment for 24 hours reduced TIF1γ protein expression marginally in the cytoplasmic extracts but significantly in the nuclear extracts compared to control treatment ( Figure 3C). Co-treatment of luspatercept with GDF11 increased the TIF1γ protein levels in the nuclear F I G U R E 3 Smad2/3-pathway overactivation reduces, and luspatercept co-treatment promotes, nuclear localization of TIF1γ in MEL cells. A, Immunofluorescence microscopy showing effect of GDF11 (100 ng/mL, 24 h) alone or in combination with luspatercept (1 µg/mL) on TIF1γ levels in DMSO-pretreated MEL cells (control). Representative images depict TIF1γ (magenta/AF647). TIF1γ levels were lower overall and localized to a lesser degree in the nucleus under GDF11-treated conditions compared with control and luspatercept co-treatment. Scale bar, 4 µm. B, Percentage of cells with nuclear TIF1γ localization. Data are means ± SEM (n = 3 images per group), *P < .05 vs. DMSO or co-treatment. C, Western blot analysis showing effect of GDF11 (100 ng/mL) alone or in combination with luspatercept (1 μg/mL) on TIF1γ protein expression in cytosolic and nuclear fractions of MEL cells pretreated with 2% DMSO (control) to induce differentiation. GAPDH served as loading control for cytosolic extracts, and LaminB1 is used as nuclear protein loading control for nuclear extracts. D, Cell diameter measurement from panel A. Data are means ± SEM (n = 10 randomly selected cells per group), ****P < .0001 vs. control or co-treatment. Note larger size of GDF11-treated cells compared to control (indicated by arrowheads) extracts but not in cytosolic extracts ( Figure 3C). Together, these re-

| Smad2/3-pathway overactivation in MEL cells reduces viability and inhibits erythroid differentiation
Previous studies have indicated that increased nuclear pSmad2/3-Smad4 and/or TIF1γ deficiency inhibits erythroid differentiation. 19 We also examined effects of GDF11-induced Smad2/3-pathway overactivation on erythroid differentiation in MEL cells by assess-  Figure S2A,B). We then used immunofluorescence microscopy to assess pan cellular levels of haemoglobin (using antihaemoglobin-α antibody) as a measure of cellular differentiation ( Figure S2C). Consistent with its increase in cell size, GDF11 treatment for 24 hours reduced cellular haemoglobin levels compared to control conditions ( Figure 4A,B), thus providing additional evidence that erythroid differentiation in MEL cells was inhibited by GDF11.
Importantly, co-treatment with luspatercept increased cellular haemoglobin levels significantly compared to either group ( Figure 4A,B), thus indicating enhanced erythroid differentiation even compared to control conditions. Together, these results establish that erythroid differentiation of MEL cells responds to overactivation and inhibition of Smad2/3 signalling as expected from previous studies in other models. 18 Similarly, previous studies have also indicated that pS-mad2/3-Smad4 complexes mediate anti-proliferative responses in haematopoietic progenitor cells. 19 Therefore, we next characterized  Figure 4H). It remains to be confirmed that the increased levels of reactive oxygen species shown here with GDF11 treatment are causally related to reduced cell viability; however, our correlative finding is consistent with a previous report. 36

| Luspatercept reverses Smad2/3-mediated reduction in nuclear GATA-1 levels in differentiating MEL cells
Given the established role of GATA-1 in terminal erythroid differentiation as well as links between TIF1γ and GATA-1 expression, 26 Figure 5B), whereas luspatercept significantly increased levels of GATA-1 per cell compared to GDF11 alone but did not restore cellular GATA-1 to control levels ( Figure 5B). Note that this net effect of luspatercept co-treatment on GATA-1 protein expression reflects an increase in mean GATA-1 levels per cell (compared to GDF11 alone, Figure 5B). To further confirm the forgoing results, we then subjected nuclear extracts from treated MEL cells to Western blot analysis. By this method, GDF11 treatment reduced nuclear GATA-1 protein to barely detectable levels, whereas co-treatment with luspatercept restored GATA-1 to levels similar to control ( Figure 5C). In a different experiment, we found that GATA-1 levels trended towards lower levels in the cytosolic extracts with GDF11 treatment compared to control ( Figure S3B). Importantly, co-treatment with luspatercept restores nuclear availability of TIF1γ and GATA-1 to control levels and restores traits consistent with erythroid differentiation such as smaller cell size, higher haemoglobin content, reduced oxidative stress and partly improved cell viability.

| RAP-536 alleviates disease comorbidities in the Hbb th3/+ mouse model of β-thalassaemia
We have previously shown that RAP-536 alleviates anaemia as well as comorbidities in the Hbb th1/th1 mouse model of β-thalassaemia intermedia, 17 in which both copies of the β-globin major gene are deleted. Here, we investigated RAP-536 activity in the Hbb th3/+ mouse model of β-thalassaemia-in which both the β-globin major and β-globin minor genes have been eliminated in heterozygosity-to provide context for transcriptional profiling in this model (below).
Hbb th3/+ mice develop anaemia and symptoms of β-thalassaemia intermedia, including reduced RBC parameters (Table 1), increased spleen weight ( Figure 6A) and decreased bone mineral density ( Figure 6A) compared to wild-type mice. Similar to our previously reported results in the Hbb th1/th1 model, RAP-536 treatment (1 mg/kg) twice weekly for 2 months improved the phenotype in the Hbb th3/+ model by increasing RBC parameters (Table 1), increasing bone mineral density ( Figure 6B) and reducing splenomegaly ( Figure 6A  albeit not significantly, in late-stage erythroblasts from β-thalassaemic mice compared to erythroblasts from wild-type mice ( Figure 8B). Importantly, treatment of β-thalassaemic mice with RAP-536 increased Gata-1 expression significantly in late-stage erythroblasts compared to either vehicle-treated β-thalassaemic mice or wild-type mice ( Figure 8B). In these late-stage erythroblasts, expression of the GATA-1 downstream target genes Fech (ferrochelatase, involved in haem biosynthesis) and Bcl2l1 (Bcl-xL, anti-apoptotic) was non-significantly reduced in vehicle-treated β-thalassaemic mice compared to wild-type mice but was not significantly restored by RAP-536 treatment ( Figure 8C,D).
Together, these data demonstrate that overactivation of the  Could not perform significance test because of insufficient N.

| RAP-536 restores nuclear localization of TIF1γ in β-thalassaemic erythroblasts to a focal pattern typical of wild-type erythroblasts
bone marrow of wild-type mice or β-thalassaemic mice treated with RAP-536 (30 mg/kg) or vehicle for 16 hours (Figure 9). Nuclear levels of TIF1γ, as determined by DAPI counterstain, were reduced in erythroid cells from vehicle-treated β-thalassaemic mice compared to those from wild-type mice or β-thalassaemic mice treated with RAP-536. In addition, TIF1γ was concentrated in the nuclear region of erythroid cells from wild-type mice or β-thalassaemic mice treated with RAP-536, whereas TIF1γ displayed a distinct, perinuclear distribution in cells from vehicle-treated β-thalassaemic mice (Figure 9).
It is ambivalent if the nucleated erythroid precursors shown here ( Figure 9) are in the same stage of erythroid maturation, nevertheless these preliminary results do support the notion that the effects of luspatercept on IE are mediated through the subcellular distribution of TIF1γ in erythroid precursors and GATA-1 availability.  and reduced accumulation of unpaired α-globin on RBC membranes that we reported previously in β-thalassaemic mice. 17 Collectively, our findings highlight the importance of Smad2/3pathway overactivation in impaired erythroid differentiation underlying IE in β-thalassaemia. By sequestering Smad2/3-pathway ligands, luspatercept prevents overactivation of this pathway and increases nuclear availability of the master erythroid regulator GATA-1 in erythroid precursors, likely by favouring nuclear complexes of pS-mad2/3-TIF1γ over pSmad2/3-Smad4. Up-regulation of GATA-1 and its target gene signature, as well as heat shock factor and its protein quality pathways, in turn exert coordinated downstream actions that ameliorate oxidative stress and promote erythroid differentiation. β-thalassaemia is considered a paradigmatic example of IE 54 ;

| D ISCUSS I ON
therefore, our present mechanistic findings may have wider applicability to related diseases. It remains to be determined how activity by other components of the TGF-β superfamily 55 is integrated with mechanisms implicated here.

ACK N OWLED G EM ENTS
The authors would like to thank Lay-Hong at BIDMC for her confocal microscope expertise; G. Paradis and A. Parmelee at the Flow Cytometry Core of MIT and Tufts University, respectively, for their assistance with flow cytometry; and M. Alexander at Acceleron Pharma for editorial assistance with the manuscript. We would also like to thank the groups of Molecular Cell Biology-Protein Biochemistry and Preclinical Pharmacology at Acceleron Pharma.
This study was supported by Acceleron Pharma. Center.

CO N FLI C T O F I NTE R E S T
F I G U R E 9 RAP-536 treatment in vivo restores nuclear localization of TIF1γ in TER119 + bone marrow cells of Hbb th1/th1 β-thalassaemic mice to a focal pattern typical of wild-type mice. Immunofluorescence microscopy showing the levels and subcellular distribution of TIF1γ in TER119 + bone marrow cells from wild-type mice and Hbb th1/th1 β-thalassaemic mice (HOM) treated with a single dose of RAP-536 (30 mg/kg, i.p, 16 hours) or vehicle. Images depict TIF1γ (green/AF488), TER119 (red), DAPI-labelled nuclei (blue) and the merge of these three independent images. Scale bar, 2 µm. This merged image was further processed with Imaris software to better visualize the levels and subcellular distribution of TIF1γ by conversion of diffuse labelling to punctate labelling. As visible in the Imaris merged images, green punctate labelling representing TIF1γ is mainly distributed focally within the nuclei of erythroblasts from wild-type mice and β-thalassaemic mice treated with RAP-536. By contrast, TIF1γ labelling in the noticeably larger erythroblasts from vehicle-treated β-thalassaemic mice is localized primarily circumferentially in the perinuclear region

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.