IL-17A, IL-17AF, and IL-17F stimulate erythroid colony formation

We recently found that Il17ra, encoding the IL-17 receptor A chain (IL17RA), is expressed by early hematopoietic and erythroid progenitors [5], and that its ligand, IL-17A, markedly potentiates formation of CFU-E colonies from adult mouse or human bone marrow, with saturable kinetics and an EC₅₀ of ~60 pM. This effect is lost in cells from Il17ra^−/− mice [5]. IL-17RA is a common receptor chain in several heteromeric receptor complexes, each of which binds distinct members of the IL-17 cytokine family [23,24]. Here, we examined the remaining six IL-17 family ligands for potential erythropoietic activity using CFU-E colony-formation assays (Fig 1B). We found that, like IL-17A, the homodimeric IL-17F, and the heterodimeric IL-17AF, also enhanced CFU-E colony formation in response to Epo, and that none had an effect in the absence of Epo. By contrast, the remaining homodimeric IL-17 ligands, IL-17B, C, D, and E, had no effect. IL-17A, F and AF (each at 20 ng/ml) increased the response to Epo (0.5 U/ml) by 39% ± 1.4%, 21% ± 0.7%, and 46% ± 1.1%, respectively (mean ± SEM, adjusted p value (p_adj) = 0.02 for each cytokine, Wilcoxon signed-rank test paired by Epo concentration, with Benjamini–Hochberg (BH) correction across cytokines). IL-17A, F, and AF are the three known ligands for the IL-17RA/IL-17RC receptor complex [23,26], suggesting that this is the functional IL-17 receptor in erythroid progenitors. Both these receptor chains are expressed in erythroid progenitors (Fig 1C).

As a technical note, the efficacy of IL-17A, F, and AF in CFU-E assays was sensitive to the age of the lyophilized proteins and to the length of storage after reconstitution (S1A Fig). For all subsequent experiments, we used cytokines with lyophilization age less than 1 year to minimize the source of variation. The non-responsiveness to IL-17D and E could not; however, be explained by cytokine age. We focused our in vivo experiments on IL-17A, since there are no commercially available recombinant IL-17F or IL-17AF preparations from mammalian-cell systems.

IL-17A and Epo synergize in vivo to increase erythropoietic rate

We next examined whether IL-17A promotes erythropoiesis in vivo. We established an optimal dosing regimen by undertaking pharmacokinetic analysis of IL-17A (S2A Fig, [27]). We then treated 8–12-week-old Balb/c mice with either vehicle (phosphate-buffered saline, PBS), IL-17A (200 ng/g every 12 hours), Epo (0.25 U/g every 24 hours), or both Epo and IL-17A (n = 6–8 mice per treatment group, pooled from 3 independent experiments) (Fig 2A). Mice were sacrificed at 72 h, and blood, bone marrow, and spleen were collected for a comprehensive analysis of hematopoietic progenitors and mature blood cells (Figs 2, S2, S3 and S4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. IL-17A synergizes with Epo to increase erythropoietic rate in vivo.

(A) Experimental design to evaluate IL-17A erythropoietic stimulation in vivo for data in panels B to G. The data were pooled from 3 independent similar experiments, with a total of 6 to 8 mice per treatment. FACS gate definitions and full data are in S2, S3 and S4 Figs. (B) Representative images of freshly-isolated spleens at 72 hours and spleen to body weight ratios. p-values are from two-tailed t-tests. Here and in all subsequent boxplots, data points correspond to individual mice, box shows medians and the central quartiles (25th to 75th percentiles), and whiskers extend by 1.5 times the inter-quartile range. Data for this figure are in S2 Table. (C) Changes in blood reticulocyte fraction (CD71⁺ Draq5^- blood cells) and hematocrit. (D) Representative flow-cytometric 2D contour plots of Ter119 (erythroid-specific) and CD71 expression, identifying ProE and later erythroid precursors (Ter119^high gate). Erythroid precursors are sub-divided into increasingly differentiated erythroblast gates EryA, B, and C as in [29], shown in S2B Fig. (E) Summary data of bone marrow (BM) and spleen cells labeled with either a 10-color stain antibody-panel to identify CFU-E [5,30] (S3A Fig) or with the CD71/Ter119 panel to identify ProE (panel D). (F) Heat map summarizing the fold-change in the mean cell number (cells/g body weight) of n = 6 mice per condition relative to vehicle-treated mice, for each flow-cytometric subset in bone marrow (BM) and spleen. Full data for individual subset/mice is given in panel E, S2, S3 and S4 Figs, and S2 Table. Groups I–III indicate functionally distinct progenitor subsets. (G) Principal component (PC) analysis of the entire dataset, treating the 22 bone marrow and 22 spleen subsets as independent features of the data, for all mice in the 4 treatment groups. Individual datapoints correspond to individual mice. The y-axes are PC1 and PC2, which explain 35% and 18% of the between-mouse variance. Statistical significance was assessed with Kruskal-Wallis H-test across all treatments (one-way ANOVA on ranks; p = 0.0002 for PC1, p = 0.0147 for PC2) and a Conover test was used for post-hoc pairwise comparisons with Benjamini–Yekutieli (BY) correction: ***, p_adj < 10⁻³; **, p_adj < 10⁻². (H) Loadings of PC1 and PC2. Each dot corresponds to FACS subset in the dataset. FACS subsets are colored by the groups defined in panel F. PC1 is enriched for early hematopoietic progenitors (Group I) and erythroid progenitors and precursors (Group II; p = 1.45 × 10⁻⁶, Wilcoxon rank sum test comparing PC1 loadings for Groups I and II vs. Group III). PC2 is enriched for myeloid and lymphoid subsets (Group III, p = 0.029). See S2E and S2F Fig for the identity and tissue of each of the FACS subsets. Data for all figure panels are in S2 Table.

https://doi.org/10.1371/journal.pbio.3003462.g002

The mouse spleen is the principal extra-medullary site of expansion of hematopoietic and erythropoietic tissue in response to stress signals. We found that administration of IL-17A by itself had little effect on spleen size, and that, as expected, Epo increased spleen size by 2-fold. Remarkably, the combined Epo and IL-17A treatment increased spleen size by nearly 3-fold [Fig 2B; spleen to body weight ratio for vehicle, 0.34 ± 0.008% (mean ± SEM); for Epo, 0.67 ± 0.06%; and for Epo + IL-17, 1.0 ± 0.03%; the difference between Epo and Epo + IL-17A is significant at p = 0.0003, two-tailed t test].

Spleen enlargement suggested an increase in erythropoietic rate, which was confirmed when we examined blood reticulocytes (which are reflective of erythropoietic rate). Reticulocyte numbers were increased by Epo as expected, and further significantly increased, compared with Epo alone, by the combined Epo + IL-17A treatment (Fig 2C, p = 0.048). Given that the hematocrit was unaltered in any of the treatment groups (Fig 2C), the increase in reticulocytes was not a compensatory response to anemia/hypoxia. Of note, 72 hours of erythropoietic stimulation is too short a period for an increase in erythropoietic rate to result in a significant increase in hematocrit.

We proceeded with comprehensive examination of early hematopoietic, erythroid, myeloid and lymphoid cells in the bone marrow and spleen (Figs 2D–2H, S2–S4; S2 Table) using flow cytometry [28,29], including a recent flow-cytometric strategy that identifies highly pure BFU-E, CFU-E and erythroid/megakaryocytic/basophil-mast cell progenitors (EBMPs) (S3A Fig, the “10 color stain”) [5,30]. This analysis showed that administration of IL-17A alone has little or modest (<2 fold) effects on hematopoietic progenitors, including minimal effects on myeloid and lymphoid cells (Figs 2F, S2C, and S2D). In combination with Epo, however, IL-17A markedly increased the absolute numbers of BFU-E, CFU-E, erythroid precursors and early hematopoietic progenitors that contribute to the erythroid trajectory (EBMP, MEP); the expansion in these populations was well above that observed in response to Epo alone (Figs 2D–2F, S2B, and S3C–S3E). Peak response to the combined Epo + IL-17A treatment was in CFU-E, ProE, and EryA precursors; for example, spleen ProE increased 37 ± 12-fold in response to Epo alone, and 193 ± 58-fold in response to the combined Epo + IL-17A treatment (Figs 2D–2F and S2B).

Because changes in different cell sub-populations are not independent of one another, we carried out a statistical analysis of changes in progenitors after first binning them into three principal groups: early hematopoietic progenitors (group I); erythroid progenitors-and-precursors (group II); and myeloid-and-lymphoid progenitors (group III). We found a statistically significant increase in the absolute number of progenitors in groups I and II when comparing the combined treatment of Epo + IL-17A with treatment with Epo alone (p_adj = 0.017 in bone marrow, p_adj = 0.03 in spleen, Wilcoxon signed-rank test paired by FACS gate, with BH correction across comparisons). There was no significant difference between the combined treatment and treatment with Epo alone on group III (myeloid and lymphoid progenitors, Figs 2F, S2C and S2D). These same changes can also be appreciated without prior grouping of progenitors by principal component analysis (PCA) of the 44 flow-cytometric bone marrow and spleen subsets in each of the four treatment conditions. The first principal component (PC) explains 35% of the variance between mice from all conditions (Fig 2G) and corresponds to the response by early hematopoietic and erythroid subsets (Groups I and II), as seen from its loading coefficients (Fig 2H). This PC is most affected by the combined Epo + IL17A treatment, less so by Epo treatment alone, and least by IL-17A alone (Fig 2G). By contrast, the second PC (which explains 18% of the variance in the dataset) is enriched for myeloid and lymphoid subsets, including GMP progenitors, CD4⁺ and Gr⁺/Mac1⁺ precursors, and is responsive to treatment with IL17-A alone (Figs 2G, 2H, S2E and S2F).

Together, these results show that IL-17A synergizes with Epo: it is an erythropoietic-stimulating factor in vivo, but only in the context of elevated Epo, as might be seen in erythropoietic stress.

In these experiments, we also examined the fraction of cells that are undergoing apoptosis in each flow cytometric subset, using Annexin V binding (S4D Fig). As established previously [11,29], Epo administration reduced erythroblast apoptosis. The combined Epo + IL-17A treatment had a quantitatively similar effect to Epo alone, suggesting that IL-17A does not exert its synergistic effect through the regulation of erythroblast apoptosis.

IL-17A is dispensable for baseline erythropoiesis

Given the ability of IL-17A to stimulate erythropoiesis in combination with Epo, we examined whether IL-17A contributes to basal erythropoiesis. To test this, we generated mice with hematopoietic and erythroid-specific deletions of Il17ra. Mice with an Il17ra^f/f allele [31] were crossed with either Vav-iCre [32] or EpoR-Cre [33] mice. Quantitative-PCR (qPCR) showed poor deletion of Il17ra in EpoR-Cre mice (S5A Fig), however, possibly because Il17ra is expressed earlier than Epor during erythroid differentiation (Fig 1C). By contrast, Il17ra was deleted in over 99% of Kit⁺ and ProE cells on the Vav-iCre background, in both bone marrow and spleen (Figs 3A and S5A). There was no significant difference in hematocrit or in spleen size between IL-17ra^f/f/ Vav-iCre mice and control Vav-iCre mice (Fig 3B), nor in any of the erythroid progenitor and precursor subsets in either bone marrow or spleen (S5B Fig). We conclude that there is no requirement for IL-17RA in the maintenance of basal erythropoiesis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. The erythropoietic effect of IL-17A is autonomous to hematopoietic cells and accelerates the response to hypoxic stress.

(A) Deletion efficiency of Il17ra in the hematopoietic-specific Vav-iCre model, evaluated by genomic qPCR of the excised genomic locus (exons 2,3) normalized to an adjacent locus (exons 5,7). The cartoon of the Il17ra locus is not to scale. qPCR was performed on FACS-sorted Kit⁺ and ProE cells. Data points are independent biological replicates. Additional analysis confirming deletion using independent primers is in S5A Fig. Data for this figure panel are in S10 Table. (B) The IL-17RA receptor is not required for steady-state erythropoiesis. Spleen to body weight ratio and hematocrit in Il17ra^f/f/ Vav-iCre mice and in control Vav-iCre mice. Analysis of hematopoietic subsets for these mice are shown in S5B Fig. Data for this figure panel are in S10 Table. (C) Erythroid subsets (fold change) at 72 hours following cytokine injections, in control Vav-iCre and Il17ra^f/f/ Vav-iCre mice. Cytokine injections followed the schedule in Fig 2A. Data are pooled from 2 independent experiments, n = 6 mice per treatment. Color bands are 95% confidence intervals. p_adj values are one-tailed Wilcoxon signed-rank test paired by progenitor subset and then BH-corrected across the hypotheses, testing whether Epo + IL-17A results in larger progenitor populations than Epo alone. Data for this figure panel are in S3 Table. (D) Experimental design for evaluating IL-17A effect on hypoxia response. Triangles indicate injections of IL-17A or PBS. (E, F) Flow-cytometric subsets in spleen and bone marrow, respectively, at the indicated time points during hypoxia. Results are expressed as a fold-change in viable cells per gram body weight, relative to equivalent progenitor subsets in normoxia. Data pooled from four independent experiments (2 experiments each at 24 and 48 h), for a total of n = 6 mice for each time point for hypoxia ± IL-17A and n = 4 mice for normoxia. Datapoints are individual mice, lines are drawn through the means. P-values are Wilcoxon signed-rank test across all erythroid subsets at each time point and tissue, testing that erythroid subsets are larger in hypoxia + IL-17A than in hypoxia alone. BH correction was applied across comparisons. Data for figure panels E and F are in S4 Table.

https://doi.org/10.1371/journal.pbio.3003462.g003

Synergism between IL-17A and Epo is autonomous to hematopoietic cells

To determine whether the synergism in vivo between IL-17A and Epo is cell autonomous, we administered to IL-17ra^f/f/ Vav-iCre and control Vav-iCre mice either Vehicle, IL-17A, Epo or Epo + IL-17A for 72 hours, following the same experimental design we previously used in wild-type mice (n = 6 mice per treatment/genotype combination, Fig 3C). We found no significant difference between Epo and the combined Epo + IL-17A treatment in IL-17ra^f/f/Vav-iCre mice, while synergism was still observed in the response of both spleen and bone-marrow erythroid progenitors and precursors in control Vav-iCre mice (p_adj = 0.015, Fig 3C). We therefore conclude that the synergistic interaction between Epo and IL-17A is autonomous to hematopoietic progenitors.

IL-17A accelerates the erythropoietic response to hypoxia

Tissue hypoxia induces an erythropoietic stress response, driven by increased Epo secretion from the kidney. The resulting higher Epo concentration in blood accelerates erythropoietic rate by promoting survival and cycling of erythroid precursors (Fig 1A). We asked whether IL-17A could stimulate erythropoiesis in the context of physiological stress, without administration of exogenous Epo.

We previously modeled erythropoietic stress by placing mice in a reduced-oxygen environment (10% O₂, compared with 21% oxygen at sea level), finding that blood Epo levels increase and plateau at 0.03 U/ml within 24 hours of the onset of hypoxia, in turn increasing ProE and EryA in spleen and bone marrow [11]. To test whether IL-17A impacts this hypoxic stress response, we injected mice with either IL-17A (200 ng/g every 12 hours) or vehicle for 24 hours before either transferring them to a 10% O₂ environment for up to 48 hours or allowing them to remain in normoxia. We continued IL-17A administration up until the end of the experiment (Fig 3D).

Older reports based on colony formation assays show that in response to acute erythroid stress, bone-marrow erythroid expansion is delayed relative to the spleen, possibly because bone-marrow BFU-E initially migrate to the spleen, where they contribute to rapid erythroid expansion [34–36]. Here, our flow-cytometric assessment of BFU-E and CFU-E using the “10 color stain” [5,30] (S3A Fig) is in agreement with these older studies, showing an initial decline in bone-marrow BFU-E and a slower bone-marrow erythroid expansion relative to the spleen (Fig 3E and 3F; S4 Table).

Significantly, treatment with IL-17A accelerated the erythroid response in both tissues. At 24 hours in hypoxia, IL-17A treatment increased nearly all of the spleen progenitors and precursors on the erythroid differentiation trajectory when compared with hypoxia alone (early and late CFU-E, ProE, and EryA/B/C, p_adj = 0.004, Wilcoxon signed-rank paired by cell subset, with BH correction) (Fig 3E). As an example, splenic ProE increased by 1.76 ± 0.48 fold in hypoxia, but by 5.7 ± 1.9 fold in combined hypoxia with IL-17A. In the bone marrow, the response to hypoxia became apparent at 48 hours, with IL-17A treatment further increasing erythroid progenitors and precursors (Fig 3F, p_adj = 0.004, test as above). Late CFU-E, for example, increased by 1.9 ± 0.15 fold in response to hypoxia, but by 4.1 ± 1.0 in response to hypoxia with IL-17A. Taken together, these findings show that IL-17A accelerates the erythropoietic response to hypoxia.

Of interest, we found that mice placed in hypoxia for 24 hours had a significantly increased level of IL-17A in plasma compared with normoxia (5.49 ± 2.09 pg/mL versus 1.19 ± 0.52 pg/mL, p = 0.03, n = 31 mice, S5C Fig), suggesting a potential role for IL-17A in hypoxia responses. It is unclear, however, what the hypoxia-induced IL-17A levels would be in tissue and whether this would suffice to drive erythropoiesis.

IL-17A amplifies the transcriptional response of erythroid progenitors to Epo

Our flow cytometry results showed that IL-17A does not enhance Epo’s pro-survival effect on erythroblasts (S4D Fig). What then is the mechanism underlying the synergistic action of Epo and IL-17A? To address this, we undertook scRNA-seq of bone-marrow and spleen hematopoietic cells from mice treated for 72 hours with IL-17A alone, Epo alone, combined Epo and IL-17A or vehicle, using the same experimental design as in the flow cytometry experiments above (Fig 2A). To ensure representation of rare progenitor populations, we first enriched the tissues for Kit⁺ cells, and then pooled these with 10% unenriched cells (Fig 4A). We analyzed 28,681 cells from both spleen and bone marrow (n = 2 mice per treatment; cell counts and markers used for cell annotation in S5 Table and S6A Fig). This data set contains a representation of the complete erythroid trajectory, from MPP to reticulocytes, as well as eight other hematopoietic lineages derived from MPPs, consistent with previous work [5,37–39] (Fig 4B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Single-cell transcriptomics of the Epo/IL-17A synergism in erythropoiesis.

(A) Experimental workflow for scRNA-seq. Bone marrow and spleen samples (n = 2 mice per treatment) were enriched for Kit⁺ cells, labeled using MULTI-seq, and pooled for sequencing. (B) UMAP visualization of single-cell transcriptomes across conditions and tissues, colored by annotated cell states. MPP, multipotent progenitor; EBMP, erythroid-basophilic-megakaryocytic progenitor; BaMaP, basophil-mast cell progenitor; MegP, megakaryocyte progenitor; EosP, eosinophil progenitor; LyP, lymphoid progenitor; GMP, granulocyte monocyte progenitor; MoP, monocyte progenitor; imMo; immature monocyte; mMo, mature monocyte; NeuP, neutrophil progenitor; imNeu, immature neutrophil; mNeu, mature neutrophil; imB, immature B cell; proB, proB cell; T, T cell; cDC1/2, conventional dendritic cell 1/2; pDC, plasmacytoid dendritic cell; Mac, macrophage; EEP, early erythroid progenitor (BFU-e equivalent); CEP, committed erythroid progenitor (CFU-e equivalent); ProE, proerythroblast; BasoE, basophilic erythroblast; PolyE, polychromatic erythroblast; OrthoE, orthochromatic erythroblast; pyrenocytes, and reticulocytes. (C) Relative cell densities across the sampled RNA-Seq state space, evaluated in high dimensions and visualized on the UMAP plot. The plots are colored by the log2-transformed ratio of cell densities in the neighborhood of each sampled transcriptome, for each treatment pair as indicated. (D) Number of differentially expressed genes (DEGs) between cells classified to each indicated cell state across treatment conditions in bone marrow and spleen. E + I: Epo + IL-17A. DEGs are identified by a rank-sum test with BH correction, and |log2(fold-change)| > 0.25 and p_adj < 0.05. (E) The top 20 DEGs across conditions and cell populations by magnitude of fold-change. Bar height shows log2 fold change; colors indicate lineages: dendritic (DC), erythroid (E), granulocyte-monocyte (GM), lymphoid (Ly), multipotent progenitors (MPP), miscellaneous progenitors (Misc. Prog.). Solid bars represent spleen (SP), hashed bars represent bone marrow (BM). (F) Venn diagram of genes found to be significantly regulated in a generalized linear model (GLM) that accounts for separate effects of tissue-of-origin, cytokine treatment (Epo or IL-17A), and cytokine interactions (Epo and IL-17A). The overlap of Epo regulated genes and Epo-IL-17A interaction-specific genes show a large proportion of the Epo and IL-17A interaction genes are regulated by Epo alone. See S6 Fig for GLM definition and underlying results. Data for panels C–E are in S10 Table.

https://doi.org/10.1371/journal.pbio.3003462.g004

Analysis of cell density changes along the erythroid trajectory showed that both Epo alone and the combined Epo + IL-17A treatments led to increased cell frequency of EEP (functionally BFU-E, see S3A Fig for nomenclature) and CEP-1/CEP-2 (functionally early and late CFU-E, Fig 4C), consistent with our flow-cytometry results (Fig 2). Further, differential gene expression analysis showed that most differentially expressed genes (DEGs) in response to either Epo or the combined treatment were seen in the same cell subsets (EEP and CEP, Fig 4D and S6 Table). Of note, in contrast to Epo or the combined treatments, there was little transcriptional response to IL-17A alone. Further, when compared with vehicle, the majority of DEGs in response to the combined Epo + IL17A treatment were similar to the DEGs in response to Epo alone, but their expression was consistently higher in the former, by up to 2-fold (Fig 4E). Of the top DEGs (Fig 4E and 4F), most were previously reported as Epo-responsive genes, including Podxl, Cda, Cpd, Gdf3, Gab2, and Arhgdig [5,40,41]. Of note, unlike erythroid progenitors, the immature B cells cluster shows increased cell density that is the result of stimulation with IL-17A alone and is not affected by the presence of Epo (Fig 4C).

We undertook three additional analyses of the scRNA-seq dataset, all of which similarly indicated that there was little response to IL-17A alone, and that the combined Epo + IL-17A treatment induced virtually the same set of genes as Epo alone, but to higher levels. First, we used a generalized linear model (GLM) to determine the contribution to gene expression of tissue type (spleen or bone-marrow) and of each of the cytokine treatment combinations (Figs 4F, S6B, and S6C; S7 Table). We found 192 genes that were induced as a result of the combined Epo + IL17-A treatment; significantly, of these, 179 genes were also induced in response to Epo alone; none were unique to either spleen or bone marrow. Second, we calculated an Epo-response score using genes previously shown to respond to high Epo levels [5] (S5 Table). This expression score was increased in our experiments in response to Epo alone, and further increased to even higher levels in response to the combined Epo + IL17A treatment, specifically in erythroid progenitors (EEP and CEP) in both spleen and bone marrow (Fig 5A and 5B). Third, we carried out an unbiased gene-set enrichment analysis for canonical and non-canonical Epo [42] and IL-17A [43] signaling pathway response genes (S5 Table). This suggested that STAT gene targets (both Stat3 and Stat5), which are canonical EpoR-activated pathways, are differentially enriched in response to the combined Epo + IL-17A treatment, compared with Epo alone (Fig 5C and 5D).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. IL-17A amplifies the transcriptional response to Epo in erythroid progenitors.

(A, B) Epo response scores for different cytokine treatments, evaluated from the scRNA-seq data (Fig 4) across erythroid pseudotime in spleen (A) and bone marrow (B). The pseudotime span representing erythroid progenitors (EEP and CEP) is marked. Epo response genes as in S5 Table. The observed increased expression score in response to the combined Epo + IL-17A treatment is not due to global changes in mRNA composition, as seen from expression-matched random gene set showing no change in response to the combined treatment (S6D Fig). Detailed description of Epo response score calculation can be found in Methods. (C) Gene set enrichment analysis of transcription factor targets from differentially expressed genes in scRNA-seq data (Fig 4), for cells classified as erythroid progenitors, comparing treatment with Epo + IL-17A to Epo. Line indicates a significance threshold with 5% BH-corrected p-values, calculated using Fisher’s exact test. Colors distinguish spleen (SP) and bone marrow (BM) results. (D) Volcano plots showing differentially expressed genes (gray points), and differentially-expressed pSTAT3 target genes (yellow points) in erythroid progenitors from spleen and bone marrow in Epo + IL-17A vs. Epo conditions. (E) Independent evaluation of gene expression for selected Epo-responsive gene transcripts (Podxl, Cda, Arhgdig, Cpd) measured by qPCR in sorted CFU-E/CEP cells (n = 4 mice per treatment, pooled from 4 independent experiments). Error bars represent the standard error. *p < 0.05, **p < 0.01, ns: not significant, one-tailed Wilcoxon rank-sum test. (F) Podxl cell-surface protein expression measured by flow cytometry in erythroid progenitors for each of the four treatments. Data pooled from 4 independent experiments. For early progenitors, data is mean of n = 3 mice per treatment; for ProE, n = 5 mice per treatment. Statistical significance tested with a one-tailed Wilcoxon signed-rank test across paired FACS subsets, show a statistically significant difference between treatment with Epo vs. Epo+IL-17A (p_adj = 0.0039). Data for panels A–E are in S10 Table. Data for panel F are in S11 Table.

https://doi.org/10.1371/journal.pbio.3003462.g005

To validate the scRNA-seq analysis, we carried out RT-qPCR using independently-sorted CFU-E from bone marrow and spleen (4 mice per treatment), and we additionally measured cell-surface Podxl by flow cytometry. Of the four Epo-responsive genes selected for qPCR analysis (Podxl, Cda, Arhgdig, and Cpd), three were significantly elevated in CFU-E following Epo + IL-17A treatment compared with Epo alone (Fig 5E), and additionally, Podxl surface protein expression was similarly increased in BFU-E, CFU-E, and ProE (Fig 5F). Taken together, multiple lines of evidence suggest that IL-17A amplifies Epo-dependent gene expression. IL-17A does not induce an Epo-independent, unique set of genes in erythroid progenitors, whether administered alone or in combination with Epo.

The accelerated response to hypoxia by IL-17A is associated with faster cycling

The increase in the number of erythroid progenitors in response to the combined Epo + IL-17A treatment, as compared with Epo alone, cannot be attributed to changes in cell survival (S4D Fig). To date, there is little direct data on the cell cycle in CFU-E and BFU-E in vivo during the response to Epo or hypoxic stress. To investigate a potential effect on the cell cycle, we first used our scRNA-seq data to infer progenitors’ cell cycle phase following each treatment (see Methods). We found an increased number of erythroid progenitors in G2 or M following the combined treatment with Epo + IL-17A, compared with all other treatments (Fig 6A). As a complementary approach, we calculated an activity score for genes canonically expressed in G2 or M. The G2/M score increased in erythroid progenitors in response to the combined Epo + IL-17A treatment compared with Epo alone, in both bone marrow and spleen (Fig 6B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. IL-17A enhances hypoxia-induced cell cycle shortening in erythroid progenitors.

(A, B) The fraction of cells in G2/M phase (A), and expression of canonical G2/M-enriched genes (B), corresponding to each treatment. Fractions and gene set scores are inferred from scRNA-seq profiles in each of erythroid progenitor states (EEP, CEP-1, CEP-2, ProE) in bone marrow and spleen (Fig 4). (C) Experimental design for evaluating changes in erythroid progenitor cell cycle duration in response to hypoxia with or without IL-17A pre-treatment, using mice expressing the H2B-FT transgene, which undergoes blue(B)-to-red(R) maturation. The ratio of red to total (red and blue) H2B-FT fluorescence indicates cell cycle length, with a shorter cell cycle corresponding to a lower R/(R + B) ratio. (D) Median cell cycle length in the indicated bone marrow erythroid subsets, measured in n = 3 or 4 mice for each treatment. P_adj-value is Wilcoxon signed-rank test (BH corrected), comparing erythroid subsets in hypoxia injected with either IL-17A or vehicle. (E) Distribution of cell cycle lengths in early erythroid subsets in hypoxia, injected with either IL-17A or vehicle. Data pooled from 3 mice for each treatment. An average of 1,300 cells (range: 328–5,200) per erythroid subset per treatment were analyzed. Data for panels A,B,D,E are in S10 Table.

https://doi.org/10.1371/journal.pbio.3003462.g006

These results indicated possible increased cycling in erythroid progenitors specifically in response to the combined Epo + IL-17A treatment compared with Epo alone, without a corresponding response to IL-17A treatment alone. To investigate this further, we examined how IL-17A impacts cycling of erythroid progenitors during hypoxic stress and in normoxia. We used mice transgenic for a live cell reporter of cell cycle duration, a chimeric histone H2B fused to a fluorescent timer protein (H2B-FT, Fig 6C) [13,44]. H2B-FT fluoresces blue when first synthesized but matures over ~1.5 hours into a red fluorescent protein. The ratio of red to total fluorescence reflects cell-cycle duration [44]. We treated H2B-FT transgenic mice with either IL-17A or PBS for 48 hours. After the initial 24 hours, half the mice from each group were transferred from normoxia to hypoxia (10% oxygen, n = 3 or 4 mice per group, Fig 6C–6E). Using flow cytometry, we found that IL-17A by itself appeared to slightly increase cell cycle duration in erythroid progenitors, while hypoxia resulted in a clear shortening of the cycle in all erythroid progenitor and precursor subsets. The combined treatment of IL-17A and hypoxia resulted in further cell cycle shortening, most pronounced in BFU-E and CFU-E (Fig 6D and 6E; p_adj = 0.006, Wilcoxon signed-rank test comparing all erythroid subsets in hypoxia treatment with the combined hypoxia + IL-17A treatment). We conclude that hypoxia promotes faster cycling in early (BFU-e and CFU-E) erythroid progenitors, an effect that was not previously documented; and that IL-17A further enhances this effect, thereby accelerating the erythropoietic response to hypoxia.

A mathematical model shows advantages of IL-17A-mediated synergism through the sensitization of erythroid progenitors to Epo

Our data show that the stimulatory effect of IL-17A on erythropoiesis is entirely dependent on the presence of elevated Epo. Further, scRNA-seq analysis shows that there is no IL-17A-unique transcriptional response in erythroid progenitors, whether administered alone or in combination with Epo. Instead, IL-17A sensitizes these cells to Epo, amplifying the Epo-induced transcriptional response. A key observation is the finding that IL-17A specifically acts to amplify Epo signaling in early progenitors, as evidenced by the expression of Il17ra in these cells; IL-17A has little effect on the Epo transcriptional response in ETD, consistent with the lack of IL-17A/Epo synergism in ProE and Ery A/B/C survival (S4D Fig). Therefore, IL-17A alters, or ‘tunes’, the regulatory responses of early erythropoiesis to Epo, without impacting Epo regulation of ETD. What is the functional significance of this mode of synergism between IL-17A and Epo?

To explore this question, we developed a simplified mathematical model of erythropoiesis (Figs 7A, S7A, and S7B; S1 Text). Here, (i) MPPs give rise to transit-amplifying erythroid progenitors (BFU-E and CFU-Es); these give rise to ProE precursors that ultimately differentiate into mature RBCs; (ii) circulating RBCs carry oxygen from lung to tissue; (iii) tissue pO₂ represses Epo production; (iv) Epo levels determine ProE survival, an established mechanism through which Epo regulates erythropoietic rate [10–12,29,45]; and (v) expansion of erythroid progenitors may be regulated through either Epo-dependent and/or Epo-independent pathways. In agreement with older work, BFU-E and CFU-E do not require Epo for survival and are relatively insensitive to Epo in the absence of stress (that is, in healthy individuals in normoxia) [2,11,46].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. A dynamical model predicts optimal tuning mechanisms for the erythroid feedback circuit.

(A) Schematic of a simplified dynamical systems model of erythropoiesis. Full model description is in S7 Fig and S1 Text. Mechanisms for tuning by IL-17A (blue arrows) can be summarized into two classes: Model 1 (Epo-independent tuning of progenitor number) and Model 2 (tuning Epo sensitivity of progenitors). (B) Illustration of performance tuning. Two environments A and B show different likelihoods of increased cell demand (left panel), and accordingly different optimal cost/reaction-speed trade-offs (right panel). Blue arrow: tuning control circuits on the Pareto front alters reaction speed at the lowest possible cost. (C) Examples of the dynamical model numerical solutions for pO₂ blood concentration, following gradual hypoxic onset. Two different configurations of the system (Parameter Sets 1,2) show different recovery times (defined as the root integral time squared error () of the pO₂ levels from target). (D) The constitutive cost of ProE over-production during normoxia for the same parameter sets as in (C). The over-production burden results from high levels of apoptosis of ProE (shown schematically), and is defined as 1/(ProE survival). The faster response in panel C for parameter set 2 requires a higher constitutive burden. (E) Tuning individual control parameters from a baseline state identifies variation in their contribution to recovery speed (t_ITSE)⁻¹ after pO₂ perturbation and the over-production cost [defined in (D)]. Parameters corresponding to models 1 and 2 (panel A) are denoted by triangles and circles, respectively. (F) Random sampling of control parameters over physiological ranges reveals a Pareto front in the response-speed/cost trade-off. The heatmaps shows the values of two parameters representative of Model 1 (Epo-independent proliferation, ) and Model 2 (Epo-dependent proliferation EC₅₀, ). Color bar units and additional parameter values are given in S7E Fig. (G) Computational predictions for the expansion of erythroid progenitors in each of the two tuning models following Epo + IL-17A treatment as compared to multiplicative fold change of Epo and IL-17A alone. The dashed line indicates multiplicative synergy. Model 1 is implemented by modulating while model 2 is implemented by modulating (this figure) or (S7F Fig). Refer to S1 Text for modeling parameters used. (H) Measured fold changes in cell populations from spleen and bone marrow comparing [Epo + IL-17A]/[Vehicle] treatment to the expected multiplicative fold change ([IL-17A]/[Vehicle] × [Epo]/[Vehicle]). Data points above the dashed line indicate above-multiplicative synergy. Inset shows zoomed view of lower fold changes in spleen. The same experimental data as in Fig 2F. Data for panels C–H are in S10 Table.

https://doi.org/10.1371/journal.pbio.3003462.g007

Through Epo-mediated negative feedback, the rate of red blood cell production—the erythropoietic rate—can be dynamically adjusted to meet changes in tissue oxygen tension. The performance characteristics of many feedback control systems can be altered, or “tuned”, so that they are optimized for specific goals, such as maximizing the speed or accuracy with which systems sense and react to their environment, minimizing resting-state fluctuations or ongoing costs in energy and resources [47–49]. In an optimized system, improving one objective may be at the expense of another (a pareto front, Fig 7B) [50,51]. In erythropoiesis, there is a clear conflict between optimizing for the system’s readiness to respond quickly to stress and maintaining low ongoing costs during steady-state basal erythropoiesis in health. Indeed, the vital importance of a fast erythropoietic response to hypoxic stress explains the costly, continuous over-production of ProE during steady state erythropoiesis, with most ProE undergoing apoptosis. Rescue of ProE from apoptosis by stress-induced Epo survival signals is a much faster mechanism of increasing ProE number than generating more of these cells through lengthy cell division and differentiation from earlier progenitors [10–12,29,45].

The above considerations suggest that the level of ProE over-production represents a balance, or compromise, between its cost during the steady-state, and the speed of response to stress. A high level of ProE over-production optimizes for a speedy response to stress, while entailing high cost during the basal steady state; conversely, low ProE over-production saves on ongoing costs but compromises the speed of response to stress (Fig 7B–7D). What then is the correct level of ProE over-production for the system? The answer may depend on the likelihood of having to mount a stress response (Fig 7B). Prior to any change in tissue oxygen tension, there may nevertheless be signals that indicate increased risk of impending hypoxic stress (a shift from environment A to environment B, Fig 7B). Under these changed circumstances, tuning the system by further optimizing for response speed may be worth the cost.

Based on our data, we propose that IL-17A is an example of a signal that anticipates increased hypoxia risk, and in turn, tunes the erythropoietic system so as to increase ProE overproduction, thereby allowing for an accelerated stress response should it be required. This suggestion is consistent with reports that IL-17A increases early in experimental pneumococcal pneumonia, and is an early marker of likely disease severity and development of acute respiratory distress syndrome in Covid-19 patients [52–54]. Further, levels of IL-17A in patients hospitalized with viral infections, including SARS-CoV2 or MERS, increase 3-–20-fold over baseline levels [54–57] and can exceed the EC₅₀ for the erythroid stimulatory effect of IL-17A in vitro (2 ng/mL) (S8 Table).

To test how different tuning strategies might alter cost-performance trade-offs, we used numerical simulations of normoxia and of the response to a drop in oxygen tension in alveolar air (as may occur during respiratory distress). Recovery from the latter requires increased RBC production, which increases the number of circulating RBCs, thereby compensating for reduced oxygen loading of individual RBC (Fig 7C, hypoxic stress dynamics in S7C Fig). We investigated potential tuning mechanisms by using alternative sets of model parameters, and then, for each parameter set, determining the rate of response to hypoxia (quantified by the integral-time-squared error or ITSE, Fig 7C), and the rate of ProE over-production during normoxia (Fig 7D). Significantly, the model shows that ProE overproduction is essential for a fast erythropoietic response to sudden hypoxic stress (S7D Fig). It thus recapitulates the experimental observation that loss of ProE overproduction, by genetic elimination of the apoptotic mechanisms that preserve it, delays production of new RBC in response to hypoxia [12].

We proceeded to scan all model parameters by varying them one at a time (Fig 7E). This approach revealed two principal tuning mechanisms by which IL-17A might increase the number of early progenitors. First, it might do so directly and independently of Epo, by either increasing early progenitor proliferation or the rate with which they are produced from MPPs (Fig 7A, parameters controlling Model 1, and S7A and S7B Fig). Alternatively, IL-17A might increase the number of early progenitors by sensitizing their response to proliferative signals generated by Epo (in turn, by either reducing Epo’s EC₅₀ or the maximal Epo proliferative response; Fig 7A, Model 2 and S7A and S7B Fig).

Strikingly, analysis of the response-speed/cost trade-off for single parameters suggested, however, that Models 1 and 2 are not equivalent (Fig 7E). Varying parameters that contribute to sensitization of progenitors to Epo (Model 2) provided gains in response speed at a smaller associated ProE over-production cost, compared with parameters that increased progenitor number independently of Epo (Model 1). To evaluate whether these differences hold for all model configurations, we carried out 10⁵ additional simulations, each randomly sampling model parameter values, and evaluated the two performance goals in each case. This analysis revealed a clear pareto front in the response-speed/cost trade-off (Figs 7F and S7E). Significantly, parameters associated with Model 2 (Epo sensitization) tuned the feedback circuit along a pareto front where the speed of response to stress increased with the lowest additional burden of ProE overproduction, while the parameters associated with Model 1 (Epo-independent progenitor proliferation) did not. Thus, within our model, tuning the sensitivity of early progenitors to Epo offers the lowest possible progenitor over-production burden for a given speed of hypoxic response.

IL-17A-mediated sensitization of progenitors to Epo results in their super-multiplicative expansion, in agreement with model predictions

In addition to differing in the response-speed/cost trade-offs, Models 1 and 2 also differ in expected changes to progenitor pool size. Model 1 (tuning through Epo-independent action) predicts that the combined Epo + IL-17A treatment would result in multiplicative synergy, with the increase in progenitor number being the product of the response to each cytokine alone (Fig 7G, on-diagonal). By contrast, Model 2 (tuning through sensitization of progenitors to Epo) predicts that expansion of erythroid progenitors in response to the combined Epo + IL-17A treatment would be “super-multiplicative”, that is, consistently larger than expected from the product of their independent actions (Figs 7G and S7F, above the diagonal). Analysis of our experimental results in Fig 2 shows that by 72 hours of the combined treatment, CFU-E, ProE, and other progenitor and erythroblast populations have expanded in a super-multiplicative way, consistent with model 2 predictions (Fig 7H).

Ethics statement

All work involving mice was approved by the University of Massachusetts Chan Medical School Institutional Animal Care and Use Committee (IACUC) protocol PROTO202200017 (Socolovsky laboratory).

In vivo animal studies

Mice treated with IL-17 and Epo in vivo were either BALB/cJ, Il17ra^f/f/Vav-iCre (obtained by crossing Il17ra^f/f (B6.Cg-Il17ra^tm2.1Koll/J) with Vav-iCre (B6.Cg-Il17ra^tm2.1Koll/J), both obtained from the Jackson Laboratory) or Vav-iCre. Mice treated with IL-17 and hypoxia were either BALB/cJ, or mice transgenic for H2B-FT (B6;129-Gt(ROSA)26Sor^{tm1(rtTA*M2)Jae} Hprt1^{tm2(tetO-mediumFT*)Sguo}/Mmjax, obtain from the Gao laboratory, Yale school of medicine). All mice were 8–12 weeks old at the time of experiments. Littermates of the same age and sex were randomly assigned to treatment groups. Both male and female mice were used in experiments.

CFU-E colony formation assays

Adult BALB/cJ bone marrow mouse bone-marrow cells were plated at 500,000 cells/ml in methylcellulose (Iscove’s Modified Dulbecco’s Medium, 6% Bovine serum albumin (BSA), 5% fetal calf serum (FCS), 1 mg/ml human transferrin) with added Epo and IL-17 cytokines at the indicated concentrations. Colonies were counted at 72 hours after staining with diaminobenzidine to visualize hemoglobinized cells.

Cytokine treatment in vivo

Mice were injected with cytokines subcutaneously, in a total volume of 4 microliters/g body weight. Recombinant IL-17A (made in Chinese Hamster Ovary (CHO) cells) was freshly resuspended in PBS up to 12 hours prior to use and injected at 200 ng/g body weight every 12 hours. Epo (Procrit) was injected at 0.25 IU/g every 24 hours. Vehicle- treated mice were injected with an equal volume of PBS.

Tissue harvesting

Femur, tibiae, and spleen were harvested following euthanasia and submerged in SB5 buffer (PBS supplemented with 0.2% (w/v) bovine serum albumin (BSA), 0.08% (w/v) glucose, and 5 mM of Ethylenediaminetetraacetic acid (EDTA)). Tissue harvest methods are described in [30].

Hypoxia model of erythropoietic stress

Mice were placed for up to 48 hours in a hypoxic environment using the BioSpherix A chamber (BioSpherix). Hypoxia was achieved by displacing oxygen with nitrogen at normal atmospheric pressure. Temperature, humidity, and carbon dioxide readings were monitored.

Flow cytometry

All antibody panels included either Fc block or rabbit IgG (Jackson ImmunoResearch Laboratories, Item# 011-000-003). All cells were post-labeled with DAPI (Thermofisher Scientific, Item # D1306) to exclude dead cells. Reticulocytes were measured by flow-cytometric analysis of blood labeled with the Draq5 (Cell Signaling Technology, Item # 4084L) and with CD71 antibody (Biolegend, Clone RI7217). Erythroid precursors (ProE, EryA/B/C) were identified in spleen and bone marrow samples that were labeled with lineage markers (Gr1(BV786, biotin,FITC, BD Biosciences, RB6-8C5 Clone; AF700, Biolegend, RB6-8C5), Mac1 (biotin, FITC, BD Biosciences, M1/70 Clone; AF700, PE, Biolegend, M1/70 Clone), CD4 (BV510, AF700, Biolegend, RM4-5 Clone; biotin, FITC, BD Biosciences, RM4-5 Clone), CD8a (biotin, FITC, BD Biosciences, 53-6.7 Clone; AF700, Biolegend 53-6.7 Clone), CD19 (biotin, FITC, BD Biosciences, 1D3 Clone; AF700, Biolegend, 1D3/CD19 Clone; PeCy7. Invitrogen, 1D3 Clone), F4/80 (AF700, FITC, Biolegend, BM8 Clone; biotin, BD Biosciences, T45-2342 Clone), CD71 (BV421, BV510, PeCy7, Biolegend, RI217 Clone) and Ter119 (APC, biotin, BUV395, BD Biosciences, Ter119 Clone; FITC, PeCy5, Biolegend, Ter119 Clone) antibodies [4]. Early hematopoietic and erythroid progenitors (BFU-E/P2, early CFU-E/P1-medium, late CFU-E/P1-high, basophil/mast cell progenitors/P3, megakaryocytic progenitors/P4, EBMP/P5) were identified by labeling with antibodies directed at Kit (APC-Cy7, Biolegend, 2B8 Clone; AF488, Biolegend, ACK2 Clone), lineage markers, Ter119 (APC, Biotin, BUV395, BD Biosciences, Ter119 Clone; FITC, PeCy5, Biolegend, Ter119 Clone), CD71(BV421, BV510, PeCy7, Biolegend, RI217 Clone), CD55 (AF647, Biolegend, RIKO-3 Clone), CD49f (BV421, AF488, Biolegend, GoH3 clone), CD105 (AF488, PE, Biolegend, MJ7/18 Clone), CD150 (BV650, Biolegend, TC15-12F12.2 Clone), CD41(BV605, Biolegend, MWReg30 Clone) (the “10 color panel”) [5,30]. The Podxl (PE, Biolegend, 10B9 Clone) antibody was added to the 10-color stain panel in the indicated experiments. Early hematopoietic progenitors CMP, MEP, and GMP were identified using antibodies directed at Kit (APC Cy7, Biolegend, 2B8 Clone; AF488, Biolegend, ACK2 Clone), Sca1 (PerCP-Cy5.5, Biolegend, E13-161.7 Clone), lineage markers, CD34 (BV421, Biolegend, MEC14.7 Clone) CD16/32 (PeCy7, BD Biosciences, 2.4G2 Clone). Non-erythroid hematopoietic cells (‘lympho-myeloid’ subsets) were identified using antibodies directed at CD19 (FITC or biotin, BD Biosciences, 1D3 Clone; AF700, Biolegend, 1D3/CD19 Clone; PeCy7, Invitrogen, 1D3 Clone), CD4 (BV510, AF700, Biolegend, RM4-5 Clone; biotin, FITC, BD Biosciences, RM4-5 Clone), CD8a (bioitin, FITC, BD Biosciences, 53-6.7 Clone; AF700, Biolegend, 53-6.7 Clone), Mac1 (biotin, FITC, BD Biosciences, M1/70 Clone; AF700, PE, Biolegend, M1/70 Clone) and Gr1 (BV786, biotin, FITC, BD Biosciences, RB6-8C5 Clone; AF700, Biolegend, RB6-8C5). All antibody labeling was done on freshly harvested cells at 4 °C. Cells were analyzed on a Cytek Aurora cytometer with 5 Lasers. Data was analyzed using FlowJo 10.10.0.

Flow sorting

For isolating BFU-E, CFU-E, or Kit⁺: harvested bone-marrow and spleen cells were enriched for Kit⁺ early progenitors using MojoSort Streptavidin Nanobead (Biolegend, Item #480016) bound to biotinylated anti-CD117 antibody. Enriched cells were then labeled with the “10 color panel” [5,30] antibodies and sorted on a BD FACSFusion with 5 Lasers. For isolating ProE, harvested spleen and bone-marrow were labeled with CD71 (BV421, BV510, PeCy7, Biolegend, RI217 Clone), Ter119 (APC, biotin, BUV395, BD Biosciences, Ter119 Clone; FITC, PeCy5, Biolegend, Ter119 Clone), and lineage markers and sorted or on a BD FACSFusion.

qPCR assay for deletion efficiency of Il17ra

Genomic DNA was extracted from whole bone marrow and spleen, or from flow-sorted ProE and Kit+ cells. For each genomic DNA sample, qPCR was performed using unique primers to exons 2 or exon 3 within the ‘floxed’ regions, and to either exon 7 or exon 5 outside the “floxed” region. Deletion efficiency was calculated from ΔΔCT between the exons that are external and internal to the “floxed” region.

IL-17A serum measurements

Murine IL-17A from blood serum was measured using the Mouse IL-17A ProQuantum Immunoassay Kit (Invitrogen Item# A46737) following the manufacturer’s protocol. The CFX96 Touch Real-Time PCR Detection System was used for the readout.

Dynamical systems modeling of tuning in erythropoietic differentiation

Definition of the dynamical systems model shown in Figs 7 and S7 is provided in the file S1 Text, including details on generation of all modeling figure panels. Numerical simulations were performed as described in this supplement using Python (v3.10.13) with SciPy (v1.11.4), with code available at https://zenodo.org/records/17373656.

ScRNA-seq of bone marrow and spleen cells

scRNAseq data analysis

Condition-dependent embedding density mapping.

To identify regions of enrichment or depletion of cell states in our data, we calculated embedding densities for each treatment (Fig 4C). A kNN graph (k = 200) was constructed using the PC space defined in the dimensionality reduction section. The adjacency matrix from this graph was used to estimate a probability density function for each condition.

Let represent the number of neighbors for cell i that belong to treatment j in a neighborhood of size k. To control for sampling biases, we applied a two-step normalization process. We first normalized column-wise to obtain . This adjusts for differences in the total number of cells per treatment, representing the neighborhood composition as a normalized probability. Next, we performed row normalization to convert the treatment-dependent neighborhood compositions into probability densities for each cell, ensuring that the contributions from all treatment within a cell’s neighborhood sum to 1. Each value, , represents the probability that a neighbor of cell comes from treatment .

To quantify treatment-specific enrichment relative to the control (PBS or Epo), we computed the fold change between the probability densities of the perturbed treatment and the control:

where is a pseudocount to avoid divisions by zero.

The calculated enrichment densities per cell, , were plotted on the UMAP embedding to visualize the global differences across treatments.

Generalized linear model for formal assessment of cytokine and organ contributions.

To determine the overlap of IL-17A gene expression response with Epo response while controlling for biological differences between organs (spleen versus bone marrow), we performed GLM on the scRNAseq data (S7 Table). The GLM was fit separately for each cell annotation: EEP, and two subsets of committed erythroid progenitors (CEP-1 and CEP-2). Genes expressed in fewer than 10 cells were excluded. The GLM was applied to the log(1 + CP10K) gene expression values.

A design matrix was constructed using binary indicator variables to encode metadata variables for treatment conditions and organ origin. For treatments, cells treated with IL-17 were assigned a value of 1 for the IL17 variable and 0 otherwise, similarly for EPO with the EPO variable. Organ origin was encoded with 1 representing spleen and 0 representing bone marrow (reference organ) in the Organ variable. The model has the form:

with a gaussian noise. To identify genes whose expression is regulated by specific factors and interactions, we used a Likelihood Ratio Test comparing to nested (“null”) models defined in S6B Fig. For each nested model, we calculated p-values for each gene from a chi-squared test on the LRT test statistic:

where are the log-likelihoods of the full and nested models, respectively. The p-values were then adjusted to control for the FDR using the BH procedure. Correction was performed separately for each cell state. All statistical analyses were conducted using Python (version 3.9.19), using pandas (version 1.4.3) for data manipulation, numpy (version 1.23.0) and scipy (version 1.11.0) for numerical computations, and scanpy (version 1.9.8) for handling single-cell data. The statsmodels package (version 0.14.3) was used for fitting GLMs, using sub-routine GLM.

Epo response score and cell cycle phase response score.

Gene sets were defined from previously published work and are given in S5 Table (worksheets “Gene Sets Used in Paper” and “Gene Set Gene List”). The Epo response gene set (Fig 5A and 5B) was obtained from DEGs from Epo-stimulated progenitors in Tusi and colleagues [2], and the cell cycle G2/M phase gene set (Fig 6A and 6B) was selected from Whitfield and colleagues [83]. For each of these two gene sets, scores (Figs 5A, 5B, 6A, and 6B) were calculated using the scanpy.tl.score_genes function, providing gene expression after pre-processing to log(1 + CP10K) (see Data preprocessing above), and the following parameter values: ctrl_size=N where N is is equal to the size of the test gene set; gene_pool=all_non_set_genes, where all_non_set_genes is the set of all genes excluding the test gene set.

To visualize treatment effects (e.g., IL-17A, Epo, or Epo + IL-17A), scores were centered by subtracting the vehicle (PBS) group mean within each cell state and organ. Because the data were log-transformed, subtraction was performed directly. Standard error was propagated using the formula for independent errors:

RNA isolation and RT-qPCR of sorted CFU-Es (gate P1)

Due to the high level of ambient RNA present in FBS, sorted cells were washed twice in large volumes (4.5 mL) of ice-cold staining buffer to minimize RNA contamination. After washing, cells were pelleted by centrifugation, and the supernatant was removed. Cell pellets were flash-frozen and stored at −80 °C until further analysis. Total RNA was extracted from frozen pellets using the RNeasy Micro Kit according to the manufacturer’s instructions. Whole-transcriptome pre-amplification was carried out prior to qPCR as follows: cDNA synthesis was performed using Maxima H Minus Reverse Transcriptase in the presence of a poly-T oligo and template-switching oligo. Whole-transcriptome cDNA amplification was performed using KAPA HiFi HotStart Ready Mix and cDNA amplification primers following the manufacturer’s protocol. Subsequently, qPCR was carried out on the amplified cDNA using KAPA SYBR Fast at a final reaction volume of 10 µL, containing 2 µL of cDNA template, 5 µL of SYBR Green master mix, and 200 nM each of forward and reverse primers for the genes of interest. The primer sequences for each gene as well as the sample prep primers are listed in S9 Table, worksheet “qPCR Primers for Epo Response”. Measurements were performed in either 96-well or 384-well qPCR plates. For 96-well plates, reactions were run on a CFX96 Touch Real-Time PCR Detection System. For 384-well plates, reactions were run on a QuantStudio 7 Flex Real-Time PCR System. All reactions were performed in technical triplicates.

Statistical tests and statistical significance

Statistical tests are indicated in the figure, and text of each experiment, and S2, S4, S6, and S7 Tables. All tests that are part of multiple hypotheses were corrected for false discovery using the BH procedure or Benjamini–Yekoutieli approach as stated in text, with adjusted p-values given as p_adj. Tests considered significant at a 5% FDR (p_adj < 0.05), or p < 0.05 for isolated tests.