Erythropoiesis is a complex process that ensures the production of about 200 billion erythrocytes every day, at a rate of almost 2 million per second in humans. Knowledge of the maturation process of red blood cells comes mostly from morphological analysis of bone marrow (BM) smears or biopsies, colony unit formation studies in cultures, and electron microscopy.1-3 Multiparameter flow cytometry (MFC) explorations added new information about the kinetics of expression and the role of various erythroid lineage-associated markers such as CD105 (endoglin), CD71 (transferrin receptor), and CD36 (thrombospondin receptor).4,5 A relatively limited number of MFC studies have been published concerning erythroid maturation,6,7 lately aiming at providing information about erythroid lineage alterations in early stages of unilineage myelodysplastic syndromes characterized by isolated anemia.8 The Red score, proposed in 2013, presented a bold approach using nonlysed BM samples in a panel where the addition of a DNA dye allowed to exclude from analysis mature red blood cells that have lost their nucleus.9 More recently, our group in Lund published a similar MFC approach, including multiparametric radar representation, to assess erythroid maturation with a single 7-color panel.10 This strategy allowed to delineate and confirm the differentiation pattern of erythropoiesis characterized by the following steps: (1) loss of CD117, (2) transient expression of CD105, and (3) appearance and decrease of CD36 and CD71. This highly reproducible representation, however, relied on traditional supervised sequential gating along the continuum of maturation.
Unsupervised analysis of cytomics data is an emergent and growing field mostly applied to mass cytometry but applicable to classical fluorescence-based MFC.11 Among the various software developed, the R-based (Bioconductor) Flow SOM (for flow-self organizing maps) solution was recognized as both robust and fast.12,13 This strategy has been applied by our group to the definition of normal hematopoiesis using 4 different antibody combinations. Merged list-mode files of 16 (for lymphoid markers) or 19 (for myeloid markers) normal BM samples were used to define reference patterns.14,15 All samples had been processed in the same harmonized16 lysis-no wash fashion. FlowSOM self-delineated 100 cell subset “nodes” in each of 4 so-called minimal spanning trees (MSTs). By coupling FlowSOM unsupervised segregation of these cell subsets and the versatile Kaluza software, a rapid identification of the size (number of cells) and immunophenotypic features of each node is easily done. Up to 24 different hematopoietic subsets (comprising different nodes) could be identified, including progenitors, granulocytic, monocytic, and lymphoid maturation.14,15
Here, we applied the same strategy to the list-mode files of 11 unlysed normal BM samples stained with our published so-called ERY (erythroid) tube.10 All 11 BM samples had normal erythropoiesis by cytomorphology and all individuals had normal complete blood counts (Supplemental Table 1, http://links.lww.com/HS/A120). Of note, the size of the panel used is limited by the broad spectrum of emission fluorescence of emission fluorescence of the Draq5 stain, used here primarily to set the acquisition gate on nucleated cells and then, during analysis, to define proliferating cell clusters. The files were subjected to compensation checking and fluorescence was normalized using lymphocytes as a reference as described in detail previously.15 Then, the normalized files were merged with the Kaluza merging tool. The resulting single merged file was submitted to unsupervised analysis by FlowSOM with the set-seed option15 allowing the software to generate 24 MST of 100 nodes. The nodes belonging to the erythroid lineage were located on the MST through back gating and coloring of a classical CD45/side scatter dot plot.15 The graphical MST providing the best grouping, yet clear separation of the nodes of interest, was then chosen (Figure 1). Each node was individually studied in Kaluza, recording the number of events and the mean fluorescence intensity (MFI) of each marker. The frequency of cells in each node was then calculated both as a fraction of the whole sample and of the erythroid population (Figure 2 and Table 1).
Table 1 -
Subsets of Erythropoietic Cells Identified in a Merged File of 11 Normal Bone Marrows.
BM = bone marrow, FSC = forward scatter, INT = intermediate, SSC = side scatter.
*In total BM/within erythropoiesis.
†% Draq5high positive events were considered the fraction of dividing cells.
In the chosen MST, 24 nodes qualified as belonging to various known stages of erythroid differentiation, built on the expression of CD45, CD117, CD105, CD36, and/or CD71. In turn, based on fluorescence intensity of each of these markers, these 24 nodes could be classified into 6 subsets of erythropoietic maturation (Figures 1 and 2 and Table 1). These subsets refined the known sequence of marker expression within erythroid maturation. The cutoffs for HIGH versus intermediate (INT) CD36 and CD71 normalized mean fluorescence intensity (MFI) were 4.1 and 4.59, respectively. The lowest normalized MFI for positive expression of CD36 and CD71 in normal BM were 3.83 versus 2.52, respectively. The cutoff for high Draq5 normalized MFI was 0.1, while very low Draq5 was between 0.01 and 0.06. Of interest, MFI of the DNA dye Draq5, which parallels DNA content, could be used more precisely than in our Radar approach10 to identify proliferative subsets and correlate this with their immunophenotype. Three subsets of dividing (DIV) cells (respectively dubbed EARLY DIV, INT DIV, and LATE DIV) were identified by their high percentage of Draq5hi cells (Table 1). They differed by the presence of CD117 (EARLY DIV), then by CD105 expression in the absence of CD117 (INT DIV), and finally by a CD117–/CD105– immunophenotype (LATE DIV). The normalized MFI of CD36 and CD71 was high in these populations as defined by the cutoffs. The next 2 subsets (dubbed HIGH 36/71 and INT 36/71) were consistently negative for CD117 and CD105. They also had a low fraction of Draq5hi cells suggesting that they nearly stopped dividing. The last small identified subset had very low Draq5 content (LOW DRAQ5) and could correspond to apoptotic cells or cells having just expelled their nucleus (reticulocytes).17
The EARLY DIV population, corresponding to the earliest erythroid progenitors (CD117+), had the highest CD45 expression and cell size measured by forward scatter. CD45 MFI was progressively lower in later stages of erythropoietic differentiation. Forward scatter was also progressively lower while side scatter remained similar in most subsets.
In conclusion, we report here on the use of unsupervised MFC analysis of normal BM, with an erythroid-specific panel and the no-lysis strategy, to delineate discrete subsets of erythroid maturation. Although the 24 nodes allocated to erythropoiesis could be ultimately grouped in 6 major populations, subtle differences, significant enough for FlowSOM to single them out, existed between each node within the erythroid compartment. These differences were not obvious enough to define the subsets by supervised analysis. Because eleven normal BM samples were used to obtain a representative pattern, differences between nodes related to individual variations were minimized. Thus, the robustness of FlowSOM (for unsupervised analysis) and of the panel used10 provide good confidence about this reference pattern. Of course, different panels using other erythroid-specific markers such as CD123, E-cadherin, EPO-R, or coxsackie adenovirus receptor could provide further information in a similar analytical approach. As previously shown with the radar approach,10 this new strategy is likely to provide important information for the analysis of diseases involving the erythroid lineage such as myelodysplastic syndrome, renal failure, vitamin B or iron deficiency, and other conditions. This is a currently ongoing investigation.
1. Dzierzak E, Philipsen S. Erythropoiesis: development and differentiation. Cold Spring Harb Perspect Med. 2013; 3:a011601.
2. Fibach E, Manor D, Oppenheim A, et al. Proliferation and maturation of human erythroid progenitors in liquid culture. Blood. 1989; 73:100–103.
3. Bessis M. [Erythroblastic island, functional unity of bone marrow]. Rev Hematol. 1958; 13:8–11.
4. Valent P, Büsche G, Theurl I, et al. Normal and pathological erythropoiesis in adults: from gene regulation to targeted treatment concepts. Haematologica. 2018; 103:1593–1603.
5. Okumura N, Tsuji K, Nakahata T. Changes in cell surface antigen expressions during proliferation and differentiation of human erythroid progenitors. Blood. 1992; 80:642–650.
6. Mao B, Huang S, Lu X, et al. Early development of definitive erythroblasts from human pluripotent stem cells defined by expression of glycophorin A/CD235a, CD34, and CD36. Stem Cell Reports. 2016; 7:869–883.
7. Machherndl-Spandl S, Suessner S, Danzer M, et al. Molecular pathways of early CD105-positive erythroid cells as compared with CD34-positive common precursor cells by flow cytometric cell-sorting and gene expression profiling. Blood Cancer J. 2013; 3:e100.
8. Westers TM, Cremers EM, Oelschlaegel U, et al.; IMDSFlow Working Group. Immunophenotypic analysis of erythroid dysplasia in myelodysplastic syndromes. A report from the IMDSFlow working group. Haematologica. 2017; 102:308–319.
9. Mathis S, Chapuis N, Debord C, et al. Flow cytometric detection of dyserythropoiesis: a sensitive and powerful diagnostic tool for myelodysplastic syndromes. Leukemia. 2013; 27:1981–1987.
10. Violidaki D, Axler O, Jafari K, et al. Analysis of erythroid maturation in the nonlysed bone marrow with help of radar plots facilitates detection of flow cytometric aberrations in myelodysplastic syndromes. Cytometry B Clin Cytom. 2020; 98:399–411.
11. Samusik N, Good Z, Spitzer MH, et al. Automated mapping of phenotype space with single-cell data. Nat Methods. 2016; 13:493–496.
12. Van Gassen S, Callebaut B, Van Helden MJ, et al. FlowSOM: using self-organizing maps for visualization and interpretation of cytometry data. Cytometry A. 2015; 87:636–645.
13. Weber LM, Robinson MD. Comparison of clustering methods for high-dimensional single-cell flow and mass cytometry data. Cytometry A. 2016; 89:1084–1096.
14. Lacombe F, Dupont B, Lechevalier N, et al. Innovation in flow cytometry analysis: a new paradigm delineating normal or diseased bone marrow subsets through machine learning. HemaSphere. 2019; 3:e173.
15. Lacombe F, Lechevalier N, Vial JP, et al. An R-derived FlowSOM process to analyze unsupervised clustering of normal and malignant human bone marrow classical flow cytometry data. Cytometry A. 2019; 95:1191–1197.
16. Lacombe F, Bernal E, Bloxham D, et al. Harmonemia: a universal strategy for flow cytometry immunophenotyping-A European LeukemiaNet WP10 study. Leukemia. 2016; 30:1769–1772.
17. Allan RW, Ansari-Lari MA, Jordan S. DRAQ5-based, no-lyse, no-wash bone marrow aspirate evaluation by flow cytometry. Am J Clin Pathol. 2008; 129:706–713.