Fetal‐hemoglobin‐expressing red blood cells (“F cells”) consist of three distinct types as revealed by single‐cell transcriptomic analysis of circulating reticulocytes

Human erythropoiesis switches from expressing fetal hemoglobin (HbF, with an α₂γ₂ chain composition) to adult hemoglobin (HbA‐α₂β₂ and HbA₂‐α₂δ₂) around birth. Some red blood cells seem to evade a complete switch, retaining significant amounts of HbF throughout life. These have been termed F cells ¹ and usually constitute between 1% and 7% of circulating erythrocytes. Little is known about when and how this apparent red‐cell sublineage ² diverges from common erythropoiesis, but F cells have attracted considerable interest for two main reasons. First, they show resistance to the adverse effects of the sickle cell and β‐thalassemia mutations by slowing the rate of the polymerization of sickle hemoglobin (HbS) in sickle cell disease (SCD) and reducing the excess of α‐globin in β‐thalassemia. Therefore, boosting HbF expression is the target of many therapeutic strategies for these conditions. It is also the main mechanism of action of hydroxyurea, a disease‐modifying agent in SCD, and to a lesser extent, thalassemia. Second, F cells are potentially useful biomarkers because they become more abundant in acute erythropoietic stress, some malignant hematological disorders, and during bone marrow regeneration. ³ Thus, their study promises to shed light on how erythropoiesis transitions between normal and abnormal conditions.

Investigating gene expression on a single‐cell level offers insight into the biology of heterogeneous cell populations, such as differences between erythroid cells containing high levels of HbF (F cells) and those with little or no HbF (non‐F cells). Reticulocytes, as the immediate precursors to mature erythrocytes, also encompass a distinct HbF‐containing fraction, “F reticulocytes.” ⁴ They still contain significant amounts of messenger RNA (mRNA), allowing us to dissect red blood cell heterogeneity in reticulocytes from three healthy volunteers and one patient with sickle cell anemia (HbSS, ethics board approval IRAS#92493 and 296705). To get initial insight into underlying biology, we also studied BEL‐A cells, a human erythroid cell line with a globin expression profile matching adult erythropoiesis. ⁵

We used the BD Rhapsody single‐cell multi‐omics platform with a panel of 118 genes that excluded α‐ and β‐globin genes (to prevent their transcripts from overwhelming the data) alongside seven erythroid surface protein markers. In brief, after Percoll enrichment from a fresh (<1 h) peripheral‐blood sample, reticulocytes (CD71+ erythrocytes) were flow‐sorted to >95% purity, then labeled with oligonucleotide‐conjugated antibodies for (“AbSeq”) CD71, CD36, CD45, CD44, CD34, CD235a, and CD49d. About 25,000 cells were captured into single‐cell cartridge wells and provided with a bead containing oligonucleotide tags for the cell, unique molecular identifiers, and a poly‐T tail (Figure 1A). Subsequent complementary DNA libraries were amplified and sequenced with Illumina NovaSeq 6000/PE150. Reads were mapped to genes with the Rhapsody Targeted Analysis Pipeline (SevenBridges) and transcriptomes analyzed with SeqGeq™/v1.7 (BD Biosciences). Targeting specific transcripts allowed us to amplify and sequence a defined fragment in all chosen mRNA species. For both γ‐globin genes, this was achieved with the amplification primer 5′‐GAACTTCAAGCTCCTGGGAAAT‐3′, targeting a specific 150‐bp sequence that contains the codon for the chain‐specific amino acid (GCA for alanine, in the ^Aγ‐chain, and GGA for glycine, in the ^Gγ‐chain), leading to correct HBG1‐versus‐HBG2 gene assignment for >99% of the γ‐globin mRNA molecules analyzed (Supporting Information).

Figure 1.

Identification of HBG1 and HBG2 expressing human erythroid cells through targeted single‐cell expression profiling. (A) Steps to generate and sequence the single‐cell messenger RNA (mRNA) library provide for highly specific mRNA quantification through sequencing of 118 erythroid gene transcripts (example shown: HBG1 and HBG2). From top: Bead extraction of poly‐A mRNA of a cell (1) and subsequent complementary DNA (cDNA) synthesis, (2) focuses analysis on the 3′ region of genes. In addition, amplification of the cDNA library utilizes specific primers that define the gene and gene region to be sequenced. For example, primer GAACTTCAAGCTCCTGGGAAAT amplifies both gamma globin genes. It is located 83‐bp upstream of the nucleotide specific for HBG1 (“C,” creating a codon for alanine of the ^Aγ‐chain), or for HBG2 (“G,” encoding glycine of the ^Gγ‐chain). After paired‐end sequencing of the library, analysis steps include sequence quality control, the pairing of two 150‐bp reads R1 (cell and molecule tags, poly‐T) and R2 (mRNA sequence) for each molecule sequenced, alignment to genome sequence (identifying the expressed gene), and condensing reads into molecule counts with using unique molecular identifier (UMI) tags. Reads missing the primer sequence for any of the 118 genes targeted are discarded. For the γ‐globin genes, this ensures that only reads covering the gene‐specific nucleotide are included in the analysis and ambiguities in assigning either HBG1 or HBG2 are avoided. (B) Single‐cell expression profiling of human reticulocytes: Clusters of HBG1 and HBG2 expressing cells (F cells) are evident among the reticulocytes of all four subjects (healthy donors S1, S2, S3, and a patient with sickle cell disease, HbSS genotype). BD Rhapsody‐generated transcriptome data were plotted after quality control, multiplet exclusion based on sample tags, principal component analysis, and subsequent dimensionality reduction (t‐SNE plots) following SeqGeq analysis protocols. Cells expressing the HBG genes are represented by colored dots: orange for HBG1, pink for HBG2, and red when both are expressed. F cell numbers based on this analysis are in the range expected for healthy individuals (1%–7%). For the sickle cell patient (24% HbF by HPLC), 45.5% of cells are contained within the F cell cluster.

We detected a distinct cluster of F reticulocytes (Figure 1B), analogous to previous reports for mature red blood cells. ⁶ Critically, we observed unexpected additional heterogeneity within this cluster regarding the expression of the two genes encoding the γ‐chains of HbF: HBG1, coding for the ^Aγ version, and HBG2, coding for ^Gγ. Among the reticulocytes of all subjects, we detected four distinct populations of cells: those containing ^Aγ‐mRNA and very little or no ^Gγ‐mRNA, which we term “A‐F cells,” an equivalent population containing mostly ^Gγ‐mRNA (“G‐F cells”), a third population expressing both genes (“AG‐F cells”), and a fourth, the largest population, with very little or none of either (“non‐F cells”) (Figure 2A–C). Notably, AG‐F cells appear to contain a similar amount of ^Aγ‐mRNA as A‐F cells and a similar amount of ^Gγ‐mRNA as G‐F cells. For example, in the patient with SCD, AG‐F reticulocytes contained 921 ^Aγ‐mRNA molecules per cell, on average, comparable to A‐F cells (1055 ^Aγ molecules), and 1580 ^Gγ molecules per cell, comparable to G‐F cells (1165 ^Gγ molecules). Overall, the AG‐F reticulocytes contained about double the amount (2501) of γ‐mRNA molecules compared to A‐F (1197) and G‐F cells (1250).

Figure 2.

Subpopulations of erythroid cells with distinct expression patterns for ^A γ ( HBG1 ) and ^G γ ( HBG2 ) messenger RNA (mRNA). Readings for ^Aγ‐ and ^Gγ‐mRNA (molecules counted) were normalized for each cell with the sum of non‐γ‐globin molecules detected. Due to the very low mRNA content of some reticulocytes, ^Aγ‐ and ^Gγ‐mRNA presence occasionally drops below the last molecule detectable in a cell. Since readings of “0” cannot be shown in a logarithmic graph, these cells were plotted onto the margins of the figure B–D. This allows these cells to be shown as belonging to the distinct fractions “G‐F,” “A‐F,” “AG‐F,” or “non‐F” subpopulations, expressing substantial amounts of ^Aγ‐mRNA, ^Gγ‐mRNA, both, or very little of either, respectively. (A) Circulating reticulocytes from the three healthy donors (S1–S3). The distribution of A‐F:G‐F:AG‐F:non‐F cells was 3.6%:0.4%:0.8%:95.1% for S1, 1.0%:1.4%:0.3%:97.3% for S2, and 0.8%:0.1%:0.11%:99.0% for S3. (B) The same dataset for S1 only, with reticulocyte maturity shown based on their distinct glycophorin A (GPA) surface expression. The distribution pattern of F cell subpopulations occurs irrespective of reticulocyte maturity. (C) Circulating reticulocytes from a patient with sickle cell disease. The distribution of A‐F:G‐F:AG‐F:non‐F cells was 2.8%:20.9%:21.77%:54.49%. (D) Human erythroid cell line BEL‐A. Cells were cultured and harvested during the expansion phase (mostly proerythroblasts) and on Day 2 of the differentiation phase (mostly basophilic erythroblasts). The two stages were discernible through a distinct gain of surface GPA during the transition between the stages.

To assess whether such groups of HbF‐expressing red blood cells can be detected earlier in erythropoiesis, we investigated BEL‐A cells, ⁵ which remain at the proerythroblast stage when cultured under “expansion” conditions. We detected four groups of erythroid cells according to their specific γ‐chain expression (Figure 2D), equivalent to those observed with reticulocytes (deposited at Gene Expression Omnibus under accession GSE287830).

We hypothesize that what is generally regarded as “F cells,” that is, red blood cells expressing detectable amounts of HbF, consists of three distinct subpopulations: A‐F cells, G‐F cells, and AG‐F cells. Since these groups were recognizable already at the proerythroblast stage, we postulate that they branch from the main erythropoietic trajectory as erythroid progenitors, that is, at the CFU‐e or BFU‐e stage. At that point, cells would be primed stochastically ² , ⁷ to activate and express either HBG1, HBG2, or both genes simultaneously, and maintain this distinction during terminal erythroid differentiation. Possible mechanisms include epigenetic modification, for example, DNA methylation, or asymmetric partitioning of ribosomes preloaded with HBG1 or HBG2 transcripts to daughter cells.

Our observation sheds important new light on how γ‐globin genes are controlled. ⁸ , ⁹ , ¹⁰ It has previously been shown that at birth, ^Gγ‐chains make up about 70% of all γ‐globin, reflecting fetal erythropoiesis, when all red blood cells carry HbF. Later, ^Aγ‐chains become dominant, usually reaching a ratio of 60/40 ^Aγ/^Gγ ⁸ in adult erythropoiesis. However, this is variable, with more ^Gγ present in some SCD patients, with hydroxyurea therapy and in other conditions with raised HbF or erythropoietic stress. ⁹ , ¹¹ It has long been proposed that cellular heterogeneity underlies this variability. ⁹ A shift in prevalence from the A‐F to the AG‐F and G‐F subpopulations might occur under stress conditions, alongside overall increases ² in F cell production and release. Outside hemoglobinopathy, an increase in the AG‐F population has probably little downstream physiological consequence, while its extent might be an indicator of the degree of erythropoietic stress present in an individual.

Novel strategies to induce HbF therapeutically, involving small molecules or gene therapy/editing, are presently a major research target or have recently entered clinical practice. The existence of different erythroid lineages with specific HbF expression potential would be an important consideration when studying the effect of these new therapies. For SCD and β‐thalassemia, the selective induction of AG‐F cells, with their double‐dose of γ‐globin mRNA, may be a particularly effective therapeutic approach, with substantially increased intracellular HbF concentrations, leading to better erythroid survival under conditions of ineffective erythropoiesis ¹² and a longer life in circulation. The investigation of more patients, and a careful quantification of the clusters, will show whether a marked presence of AG‐F and G‐F cells, as seems to be the case in our patient (Figure 2C), is typical for SCD. Overall F cell percentage was 45.5% in the patient (24% HbF), compared to 4.9%, 2.7%, and 1.0% in the healthy subjects S1, S2, and S3, respectively (Figure 2A). Uncovering the pathways and signals that boost AG‐F cell abundance will be desirable.

Measuring the ratio of ^Aγ/^Gγ‐globin protein in peripheral blood of our subjects by whole blood mass spectrometry, we found these to be closely correlated (r = 0.91, P < 0.05) with the ^Aγ‐mRNA/^Gγ‐mRNA ratio of reticulocytes measured in the single‐cell experiments. This provides some initial indication that the A‐F, AG‐F, and G‐F cell clusters observed on the mRNA level in reticulocytes might be preserved in mature red blood cells and reflected in the protein level. However, single‐cell protein analysis, distinguishing ^Aγ‐ and ^Gγ‐chains, will be required to confirm the existence of the four cell populations defined by protein.

Our findings suggest that the existence of A‐F, G‐F, and AG‐F sublineages could be a general feature of human erythropoiesis. We hope that presenting our initial observation will induce others to look for these cells in their datasets ⁶ , ¹³ , ¹⁴ and will spark further research into their origin, nature, and potential significance for erythropoiesis and health. The only other gene we found differentially expressed between the four erythroid sublineages, BGLT3, is active mainly in A‐F and AG‐F cells. Additional differentially expressed genes might have been missed due to their absence from our panel. Within the β‐globin gene cluster, BGLT3 is located downstream of HBG1 and encodes a long noncoding RNA involved in γ‐globin gene regulation. ¹⁵ It was previously found to be associated with HbF‐expressing erythroid precursors ¹⁰ and appears to be expressed alongside HBG1 (^Aγ) in our experiment. If little distinguishes the cellular make‐up of HbF‐expressing precursors from their non‐F counterparts, ¹⁰ critical differences are likely to occur upstream, at the junction between late progenitor stages and terminal erythroid differentiation. Dissecting this process and identifying modifying conditions and factors may help understand the precise mechanisms of persistence of HbF, open a window into the regulation of erythropoiesis under stress conditions, and point to new therapeutic avenues for hemoglobinopathies and anemia in general.

AUTHOR CONTRIBUTIONS

Helen Rooks: Conceptualization; methodology; data curation; formal analysis; investigation; visualization; project administration; writing—original draft; writing—review and editing. Cecilia Ng: Investigation; writing—review and editing. Spyros Oikonomopoulos: Conceptualization; methodology; data curation; investigation; formal analysis; visualization; writing—review and editing. Sara El Hoss: Methodology; investigation; writing—review and editing. Charles Turner: Methodology; data curation; validation; investigation; formal analysis; writing—review and editing; resources. Kar Lok Kong: Methodology; investigation; writing—review and editing. Syed Mian: Methodology; resources; writing—review and editing. Yvonne Daniel: Conceptualization; methodology; writing—review and editing. Oyesola O. Ojewunmi: Methodology; formal analysis; writing—review and editing. John Brewin: Investigation; resources; writing—review and editing. David Rees: Supervision; writing—review and editing. Ghulam J. Mufti: Resources; writing—review and editing; conceptualization. Jiannis Ragoussis: Conceptualization; supervision; resources; writing—review and editing. John Strouboulis: Resources; supervision; writing—review and editing. Stephan Menzel: Conceptualization; methodology; data curation; validation; formal analysis; supervision; funding acquisition; visualization; project administration; writing—review and editing.

CONFLICT OF INTEREST STATEMENT

Helen Rooks: none; Cecilia Ng: none; Spyros Oikonomopoulos: none; Sara El Hoss grant support: EU Horizon 2020 Marie Sklodowska‐Curie grant #101024970; Kar Lok Kong: none; Syed Mian: none; Oyesola O. Ojewunmi grant support: MRC project grant MR/T013389/1; Ghulam J. Mufti: research funding from Bristol Myers Squibb; Jiannis Ragoussis grant support: Genome Canada Genomic Technology Platform grant Canada Foundation for Innovation (#33408 and CFI‐MSI #35444); John Strouboulis grant support: MRC project grant MR/T013389/1 Newton/GCRF grant EP/X527920/1; David Rees: none; Yvonne Daniel: none; Charles Turner is a founder/director of SpOtOn Clinical Diagnostics Limited; John Brewin: none; Stephan Menzel grant support: MRC project grant MR/T013389/1 Newton/GCRF grant EP/X527920/1 LIBRA (Haematology Charity) and King's College Hospital Charity. Commercial sponsor: BD Biosciences provided in‐kind funding (microfluidic cartridges, amplification kits, and analysis software) for the piloting of reticulocyte and erythroid cell profiling on the BD Rhapsody platform.

FUNDING

Our erythroid biology work is supported by MRC MR/T013389/1, Newton/GCRF grant EP/X527920/1, by LIBRA and by King's College Hospital Charity (D3013/52022/Menzel/588), as well by a Genome Canada Genomic Technology Platform grant and the Canada Foundation for Innovation (#33408 and CFI‐MSI #35444) to J.R. BD Biosciences provided in‐kind funding (microfluidic cartridges, amplification kits, and analysis software) for the piloting of reticulocyte and erythroid cell profiling on the BD Rhapsody platform. S.E.H. received funding from the European Union's Horizon 2020 Research and Innovation Programme under the Marie Sklodowska‐Curie grant agreement number 101024970.

Supporting information

Supporting Information.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are openly available in Gene Expression Omnibus (GEO) at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE287830, reference number GSE287830.