Ineffective erythropoiesis, the death of maturing erythroid cells, is a common cause of anemia. To better understand why this occurs, we studied the fates and adaptations of single erythroid marrow cells from individuals with Diamond Blackfan anemia (DBA), del(5q) myelodysplastic syndrome (del(5q) MDS), and normal controls, and defined an unhealthy (vs. healthy) differentiation trajectory, using velocity pseudotime and cell surface protein assessment. The pseudotime trajectories diverge immediately after the cells upregulate transferrin receptor (CD71), import iron, and initiate heme synthesis, although cell death occurs much later. Cells destined to die highly express heme-responsive genes, including ribosomal protein and globin genes. In contrast, surviving cells downregulate heme synthesis, while upregulating DNA damage response, hypoxia, and HIF1 pathways. Surprisingly, 24±12% of cells from controls follow the unhealthy trajectory, implying that heme also regulates cell fate decisions during normal red cell production. Del(5q) MDS (unlike DBA) results from somatic mutations, so many normal (unmutated) erythroid cells persist. By independently tracking their trajectory, we gained insight into why they cannot expand to prevent anemia. In addition, we show that intron retention is especially prominent during red cell differentiation. The additional information provided by messages with retained introns also allowed us to align data from multiple independent experiments and thus accurately query the transcriptomic changes that occur as single erythroid cells mature.

Cryo-preserved bone marrow mononuclear cells were thawed, resuspended in equivalent volume of FBS, washed in HBSS, then resuspended at 1×10⁷ cells per ml in PBS, 0.1% BSA, and 2 mM EDTA, and incubated with 1:10 dilution of biotinylated anti-CD3, anti-CD19, and anti-CD11b. Antibody-bound lineage-positive cells were depleted with streptavidin beads. Lineage-depleted marrow mononuclear cells were cultured in erythroid expansion media (IMDM, 5% pooled Human AB plasma, 330 μg/ml holo-transferrin, 1 μM dexamethasone, 160 mM monothioglycerol, 10 μg/ml Insulin, 2 U/ml preservative-free heparin, 2 U/ml EPO, 100 ng/ml SCF and 5 ng/ml IL3) at 2×10⁵ cells/ml. The cells were re-seeded at 2E5 cells/ml on day 3 with fresh media.

Cells were collected on days 0, 3, and 6.  Cells were washed and resuspended in PBS with 1% BSA at 5×10⁵ per ml.  Cells were incubated with 10 μg/ml of each DNA-labelled antibody (Biolegend TotaSeq or equivalent): CD3, CD11b, CD14, CD36, CD70, CD71/TfR, CD117/cKit, CD235a/GlyA.  Sample processing. Single-cell mRNA barcoding and library preparation were performed on the 10x Chromium controller according to the manufacturer’s instructions. A single GEM (gel beads-in-emulsion) well from each sample was used to generate both the gene expression and antibody capture libraries which were recombined for sequencing. The libraries were sequenced on Illumina NextSeq 550 or NovaSeq instruments and raw sequence data was processed with Cell Ranger (version 3).

Bam files produced by Cell Ranger were analyzed by Velocyto and transcript splice quantification saved as Loom files. Loom files were aggregated by ScanPy (concatenate function). Transcript counts were normalized to 1×10⁴ UMIs per cell and log-transformed (log1p function). The aggregated data was saved to a single HDF5 file.

Scanpy (https://scanpy.readthedocs.io/en/stable/)