A Novel Method for a Precise Identification of Human Erythroblast Subpopulations by Flow Cytometry

Abstract

Erythropoiesis is a complex multistep process encompassing the differentiation of hematopoietic stem cells (HSCs) to mature red blood cells (RBCs). Three distinct phases are identified: erythroid progenitors, terminal erythroid differentiation, and reticulocyte maturation. A detailed understanding of physiological and pathological erythropoiesis requires careful monitoring and characterization of these distinct differentiation stages. Recent technological advances such as single-cell transcriptomics have enabled unprecedented insights into cellular heterogeneity [1]. Nevertheless, these techniques are expensive and require specialized expertise, limiting their accessibility and broad use [1]. Consequently, flow cytometry (FACS) remains the gold standard for studying erythroid differentiation [2, 3].

Two staining and gating strategies, referred to as “waterfalls” are currently used to study human terminal erythroid differentiation. While both approaches utilize glycophorin A (GPA), one panel is based on the expression of CD49d and Band3 [2], while the other relies on CD105 [3]. As these strategies utilize different marker sets, it is likely that the identified populations do not completely overlap due to variations in sample sources, that is, peripheral or cord blood and bone marrow, and in experimental conditions used in different studies. As such, there is a need for establishing consensus immunophenotyping for terminally differentiating erythroblast subpopulations.

In this study, we performed in vitro erythroid differentiation with FACS monitoring, integrating both “waterfalls” within the same experimental setup to compare their ability to resolve different cell subsets and their limitations. Using bioinformatics tools for dimensionality reduction (UMAPs) and unsupervised clustering, we show that while the two “waterfalls” are complementary, they are also distinct. We report a new gating strategy combining GPA, CD105, and CD49d that enhances the resolution and ensures panel harmonization for studying erythroblast subpopulations during terminal erythroid differentiation.

We isolated CD34^Pos cells from three healthy donors and cultured them for 18 days using a four-phase protocol previously described [4]. Phenotypic expression profiles were monitored every other day using an antibody panel composed of CD123, CD49d, CD71, GPA, and Band3 to characterize both “waterfalls.” We then employed the gating strategy shown in Figure 1A to define the erythroblast subpopulations.

First, we uploaded clean single-cell data from FCS files on a single R based software (Omiq) to run all the algorithms ensuring reliable analysis. Second, we used Uniform Manifold Approximation and Projection (UMAP) [5] to represent the terminal erythroid differentiation across the culture timeline of CD34^Pos cells isolated from three healthy blood donors in a single graph (Figure 1B). Applying a supervised clustering by projecting the manually gated erythroblast subpopulations onto the UMAP provided a better definition of proerythroblast (ProE), early basophilic (EB), and late basophilic (LB) erythroblasts with the GPA/CD105 waterfall (Figure 1B, upper right panel). Conversely, polychromatic (PolyE) and orthochromatic (OrthoE/Retics) erythroblasts were better defined by the CD49d/Band3 waterfall (Figure 1B, upper left panel). Pairwise comparison of the erythroblast stages revealed that the ProE stage was the only stage that was similarly defined by both waterfalls, while all the other stage pairs showed partial overlap (Figure 1C). These results imply that the choice of surface markers combined with a manual gating can introduce bias and reproducibility issues. By incorporating markers from both waterfalls, we reanalyzed the data using an unsupervised clustering algorithm (FloSOM) which identified more clusters compared to the five groups defined by the supervised manual gating strategy (Figure 1B, lower panel). Projecting the expression of markers throughout differentiation, we found that CD105 had a greater dynamic range in the early stages (ProE to LB) compared to CD49d, while CD49d exhibited greater dynamic range during the terminal stages (PolyE and OrthoE/Retics) (Figure 1D). Indeed, CD105 expression discriminated three groups during the first days of culture, whereas CD49d identified only one. Conversely, in the last days of culture, CD49d expression discriminated two groups, while CD105 defined only one. These observations were further validated using pseudo-time analysis, which also showed that Band3 expression was very similar to GPA, indicating that monitoring one of these two markers is sufficient for accurate phenotyping (Figure 1E). Finally, the progressive decrease in CD71 expression during late-stage erythropoiesis highlighted its utility in distinguishing reticulocytes from OrthoE (Figure 1D).

Considering the dynamic ranges of these surface markers, we developed a new gating strategy using CD123, GPA, CD105, CD49d, and CD71 to distinguish the erythroblast subpopulations in an individualized non-overlapping manner. The strategy starts by gating live CD123^NegGPA^Pos cells, followed by the combination of GPA and CD105 to identify the first three stages of terminal erythropoiesis: ProE, EB, and LB (Figure 1F). The PolyE and OrthoE/Retics are gated within the CD105^neg population and discriminated based on CD49d expression levels. Finally, CD71 expression levels were used to discriminate between OrthoE and Retics (Figure 1F). To validate this gating strategy, we applied FloSOM on the live CD123^NegGPA^Pos cells and obtained 10 clusters (Figure S1). A clustered heatmap analysis of expression markers revealed proximity between several clusters, indicating shared expression profiles (Figure S1). Therefore, clusters 1, 2, and 4 were pooled into a single cluster, MK-NWF-421, while clusters 7, 6, and 10 were pooled into the cluster MK-NWF-7610 (Figure S1). Unsupervised analysis produced clusters closely matching manual gating: clusters 08, 05, and 03 corresponded to ProE, EB, and LB, while clusters 421, 7610, and 9 corresponded to PolyE, OrthoE, and Retics (Figure 1G and Figure S1).

To establish the efficacy of our strategy for studying ineffective erythropoiesis, we performed in vitro differentiation of CD34^Pos cells from sickle cell disease patients [6]. Both manual and unsupervised analyses showed a high level of resolution of specific non-overlapping erythroblast subpopulations, from the ProE to the OrthoE/Retics (Figure 1H).

Multicolor flow cytometry allows the study of markers differentially expressed during erythroid differentiation and maturation. Despite recent advances in the field, consensus on the optimal phenotypic markers to accurately discriminate all erythroblast stages during terminal erythroid differentiation has not yet been established. In the present study, we built a novel combination of surface markers to enhance the resolution in identifying all the differentiation stages. By combining the best attributes of the two commonly used staining strategies, we propose a 5-parameter flow cytometry panel that comprises CD123, GPA, CD105, CD49d, and CD71. We excluded Band 3 in our strategy as it showed limited resolution, and its exclusion reduces the risk of hemagglutination during the later stages of erythroid differentiation due to its very high expression levels. Furthermore, the inclusion of both CD49d and CD105 provides a safeguard against potential variability in expression levels caused by pathological conditions; if one of the markers exhibits aberrant expression, the other marker would ensure reliable analysis. Moreover, this simple backbone panel can be combined with other reported markers known to characterize erythroid progenitors, such as CD34, CD117, and CD36, or with nuclear staining (e.g., Syto16) and RNA staining (e.g., thiazole orange) to identify reticulocytes and erythrocytes. To improve reproducibility, we implemented the method with a pipeline using bioinformatics tools. In contrast to manual gating, which is subjective and time-consuming, the computational analyses we describe provide a standardized framework that can be easily applied across different datasets, reducing human error and enabling the generation of comparable data among different laboratories.

In addition to its utility for in vitro exploration of human erythropoiesis, this innovative immunophenotyping technique could provide a powerful tool for investigating ineffective erythropoiesis in vivo using bone marrow samples from patients with inherited or acquired erythropoietic disorders such as hemoglobinopathies, Diamond-Blackfan anemia, and myelodysplastic syndromes. It enables the identification and quantification of stage-specific abnormalities in a precise and unbiased manner, which would help establishing an accurate diagnosis at the cellular level and would improve the treatment and monitoring of the patients. This new immunophenotyping technique has the potential of replacing the manual method currently performed in clinics to classify the terminal erythroid differentiation stages based on May-Grünwald-Giemsa staining of bone marrow aspirates, improving the accuracy of diagnosis and subsequent patients' classification in multicenter clinical trials.

In summary, we report a strategy employing unbiased bioinformatics tools that provide a robust and reliable framework for a comprehensive identification of human erythroblast subpopulations by flow cytometry. It enables the accurate determination of their phenotype in both health and disease and enables the isolation of live cells by sorting for downstream applications.

The study was conducted in accordance with the declaration of Helsinki, the French blood donation regulations and ethics, and with approval from a medical ethics committee (GR-Ex/CPP-DC2016-2618/CNILMR01). Biological material was obtained from healthy donors and sickle cell disease patients after informed consent.

The authors declare no conflicts of interest.