Single-cell transcriptomics reveal long non-coding RNA dysregulation in PPP1R15B-EIF2A axis impairing erythroid maturation in myelodysplastic syndromes

Abstract

Abstract INTRODUCTION: Anemia is the main clinical manifestation of low-risk myelodysplastic syndromes (MDS), reflecting ineffective erythropoiesis. Although therapies such as erythropoietin and luspatercept are available, their effects are often transient, highlighting the need for new therapeutic agents. In the past years, long non-coding RNAs (lncRNAs) have emerged as critical regulators of gene expression in hematopoiesis and cancer, yet their role in ineffective erythropoiesis is poorly defined. The aim of this study was to identify lncRNAs contributing to erythroid dysregulation using del(5q) MDS as a model. METHODS Single-cell RNA-seq (10x Genomics) was performed on CD34+ bone marrow cells from 4 MDS-del(5q) patients and 3 age-matched healthy donors. Preprocessing was done using CellRanger and Kallisto (hg38, GENCODE v37), followed by quality filtering via emptyDrops and scDblFinder. Cell types were annotated based on canonical marker expression. To identify erythroid-relevant lncRNAs, ELATUS pipeline using recommended thresholds (ratio >3, CR >150, SI >0.1) was applied across the erythroid stages – megakaryocyte-erythroid progenitors (MEP), early (EE) and late erythroid (LE). Differential expression was assessed on pseudobulk data via DESeq2 (log2FC >1, padj <0.05) Pseudotime trajectories were inferred with Monocle3, and lncRNA expression dynamics were modeled applying generalized additive models (GAM), using binarized cell expression (>0 counts) as input. The significance of dynamic differences between conditions was tested using ANOVA by comparing full (condition-interacting) vs null (no condition-interacting) GAM. Candidate lncRNAs were defined as those meeting all three criteria: ELATUS selection, differential expression, and altered dynamics. Gene ranking was performed using a Z-score combining scaled log2FC and -log10(padj-GAM) and expression dynamics of axis-related genes were examined. RESULTS Among the 2,100 biologically relevant lncRNAs identified, 46 were differentially expressed between MDS and controls. Of these, 43 also displayed significantly altered expression dynamics, fulfilling all selection criteria. Literature review revealed that 10 of the 43 lncRNAs have cis-targets implicated in erythropoiesis. Among them, ENSG00000288934, an antisense (AS) transcript of PPP1R15B emerged as top candidate based on the previous prioritization score and dynamic profile. In healthy controls, PPP1R15B-AS showed a sharp, stage-specific pulse of expression during erythroid maturation, while PPP1R15B was concomitantly activated in a mirror-like pattern. In MDS, this lncRNA was downregulated, and PPP1R15B expression became flattened and temporally uncoupled from erythroid progression. In parallel, axis-related genes such as ATF4 showed a progressive increase during pseudotime, while pro-apoptotic effector CHOP (DDIT3) exhibited a marked late-stage peak coinciding with terminal maturation. This contrasts with the more restricted and temporally coordinated expression observed in healthy samples. CONCLUSIONS In our dataset, downregulation of PPP1R15B-AS coincided with a temporal shift and loss of coordination in PPP1R15B expression, suggesting a role as repressor in fine-tuning gene activation during specific erythroid stages. While PPP1R15B is known to resolve integrated stress response by promoting eIF2a dephosphorylation and repressing effectors such as ATF4 and CHOP, we observed that both transcription factors remained overexpressed despite having increased levels of PPP1R15B. These findings point toward a dysregulation of the PPP1R15B-eiF2a axis in MDS, in which PPP1R15B-AS may contribute to fine-timing of stress deactivation. Altogether, these observations support a model in which ineffective erythropoiesis may not only result from aberrant expressions, but also from disruption of transcriptional dynamics mediated by lncRNAs. This altered expression of PPP1R15B-AS, despite its genomic location outside 5q deletion, may reflect broader downstream remodeling. Ongoing in vivo studies are developing to validate these findings.