EKLF/KLF1 expression defines a unique macrophage subset during mouse erythropoiesis

Erythroblastic islands are a specialized niche that contain a central macrophage surrounded by erythroid cells at various stages of maturation. However, identifying the precise genetic and transcriptional control mechanisms in the island macrophage remains difficult due to macrophage heterogeneity. Using unbiased global sequencing and directed genetic approaches focused on early mammalian development, we find that fetal liver macrophages exhibit a unique expression signature that differentiates them from erythroid and adult macrophage cells. The importance of erythroid Krüppel-like factor (EKLF)/KLF1 in this identity is shown by expression analyses in EKLF-/- and in EKLF-marked macrophage cells. Single-cell sequence analysis simplifies heterogeneity and identifies clusters of genes important for EKLF-dependent macrophage function and novel cell surface biomarkers. Remarkably, this singular set of macrophage island cells appears transiently during embryogenesis. Together, these studies provide a detailed perspective on the importance of EKLF in the establishment of the dynamic gene expression network within erythroblastic islands in the developing embryo and provide the means for their efficient isolation.


INTRODUCTION
it directly activates Icam4 in the erythroid compartment, and activates Vcam1 in the macrophage compartment (Xue et al., 2014). Together, Icam4 and Vcam1 enable a two-pronged adhesive intercellular interaction to occur with their respective integrin partners on the opposite cell type. In the absence of EKLF, these interactions decrease and the integrity of the island is compromised, contributing to the abundance of nucleated, unprocessed cells seen in circulation (Gnanapragasam et al., 2016). In addition, loss of Dnase2 expression in the macrophage yields a cell engorged with undigested nuclei that triggers IFNß induction (Kawane et al., 2001;Manchinu et al., 2018;Nagata, 2007;Porcu et al., 2011;Yoshida et al., 2005).
Independent evidence for EKLF expression in erythroblastic island macrophage has been attained recently by two sets of studies. One study analyzed EpoR+F4/80+ macrophage, which are present in erythroblastic islands and are negative for Ter119, showing that these cells are highly enriched for EKLF (Li et al., 2019). In the second, a pure population of macrophages (Lopez-Yrigoyen et al., 2018) derived from a human iPSC line carrying an inducible KLF1-ER T2 transgene (Yang et al., 2017) was used to demonstrate that activation of KLF1 in these macrophages altered them to an island-like phenotype as assessed by an increase in expression of erythroblastic island-associated genes and cell surface markers, by an increase in phagocytic activity, and by an increased ability to support the maturation and enucleation of umbilical cord blood derived cells (Lopez-Yrigoyen et al., 2019).
The strongest evidence for a specific macrophage subtype in the erythroblastic island comes from the mouse, where F4/80 antigen and Forssman glycosphingolipid expression, but not Mac1 expression, are enriched in these cells (reviewed in (Manwani and Bieker, 2008)). Island macrophages are also larger than peritoneal macrophage and exhibit a high level of phagocytic activity. Although molecular expression differences between macrophage subsets has been observed (Ginhoux et al., 2016;Hom et al., 2015;Lavin et al., 2014;Seu et al., 2017), this has not been addressed in the context of early erythroblastic island development in the fetal liver. Given the compelling observations implicating EKLF in island macrophage biology, we characterized the molecular expression of the F4/80+ island macrophages in the developing mouse fetal liver, determined the EKLF dependent gene expression program in island macrophages using two independent approaches, and then established its role in specifying a unique cellular identity for this cell type by a single-cell analysis approach.

Global gene expression in E13.5 fetal liver (FL) macrophages reflect both erythroid and macrophage properties
We dissected E13.5 fetal livers (FL) and FACS sorted F4/80+ cells to obtain a pure population of fetal liver macrophages (supplementary Fig. S1). Approximately 9% of the total cells in a wild type fetal liver are F4/80+ (supplementary Fig. S1A). The sorted singlets were monitored after cytospin to determine whether they were free of contaminating erythroid cells (supplementary Fig. S1B). We found that >95% of the sorted F4/80+ population are single cells and free of any attached or engulfed erythroid cells or nuclei (supplementary Fig. S1C). We then used this pure population of F4/80+ FL macrophages to determine their global gene expression profile using RNA-Seq of biological triplicates.
We compared the global gene expression of E13.5 FL F4/80+ macrophages with two sets of gene expression data. One was from primary long-term cultures of extensively self-renewing erythroblasts (ESREs) isolated from fetal liver that can be differentiated to form mature erythroid cells (England et al., 2011;Gnanapragasam et al., 2016). The second was from adult spleen F4/80+ macrophage (Lavin et al., 2014), which is also an in vivo site of erythroblastic islands (Chow et al., 2013;Jacobsen et al., 2015;Ramos et al., 2013). Hierarchical clustering of the gene expression profile from these cell types show that the FL macrophages cluster closer to differentiating ESREs than splenic red pulp macrophages (Fig. 1A), suggesting that the FL macrophages have an early erythroid-like gene expression profile rather than a mature macrophage-like profile. Yet at the same time, we find using principle component analysis (PCA) that these cell types cluster separately, indicating that each has a unique identity (Fig. 1B). Further, we find that for a list of macrophage and erythroid markers (supplementary Table 1 (Murray and Wynn, 2011;Ng and Wood, 2014)), FL macrophages have intermediate expression of both sets of markers compared to ESREs or spleen macrophages (Fig. 1C). Together, these data suggest that FL F4/80+ macrophages essentially have dual characteristics of erythroid and macrophage-like cell populations in terms of marker expression, but still form their own unique subset.
Cell-type specific expression of a subset of genes in FL macrophages provides them with a distinct cellular identity Since our PCA analysis showed that FL macrophages have unique characteristics compared to ESREs and spleen macrophages, we performed k-means clustering of the RNA-Seq datasets of the 3 cell-types (Fig.  1D). We find a cluster that contains a set of 1291 genes were almost exclusively expressed in FL macrophages ( Fig. 1D -indicated by bracket; supplementary Table 2). Neither ESREs nor spleen macrophages have a similar set of cell-type specific gene expression as evident from the lack of clusters showing genes only expressed in these cell types (Fig. 1D). This again suggests that FL macrophages may have a distinct cellular identity, and likely possess unique functions compared to other macrophage types. We selected a set of 304 genes that were only expressed in FL macrophages and not in ESREs or spleen macrophages, and refer to them as "signature genes" (Fig. 1E, supplementary Table 2).
To determine whether signature genes are a random subset of genes or if they indeed have biological significance with respect to FL macrophage function, we performed GO analysis and filtered the results down to the unique GO terms using Revigo (Supek et al., 2011) (supplementary Table 3). We find that the signature genes are involved in 4 major biological processes -circulatory system development, tube development (vasculature development), locomotion and motility, negative regulation of blood coagulation, and cell adhesion (supplementary Table 3). Of these, cell adhesion between erythroblast island macrophages and developing erythroblasts during erythropoiesis is known to be an important function of a subset of FL macrophages (Xue et al., 2014). The additional GO categories point to novel biological or developmental roles for FL macrophages.

Loss of EKLF leads to significantly altered gene expression in F4/80+ FL macrophages
As a prelude to analyzing the effects of EKLF on F4/80+ macrophage, we directly verified EKLF protein expression and find that it is expressed in the F4/80+ macrophage as judged by immunofluorescence ( Fig.  2A). Consistent with our previous data, not all F4/80+ cells are EKLF+, and vice versa ( Fig. 2A,B). Additional support for macrophage specificity of EKLF expression comes from a published RNA-seq analyses of an extensive series of staged, sorted cells in the fetal liver (Mass et al., 2016). Mature macrophage cells (MAC; ckit-/CD45+/F480+/AA4.1-/CD11b+) do not exhibit EKLF expression in FL at E10.25; however, EKLF expression 6 hrs later in the FL is apparent (E10.5) and robust by E12.5 where it remains high until E18.5, dropping off considerably until it is not detectable at postnatal stages in the liver (Fig. 2C). As a positive control, Adgre1 (F4/80) is expressed in all samples (Fig. 2D). As a negative control, EKLF is not expressed in any other tissue macrophage cell in the same study (all samples from skin, brain, kidney, lung; (Mass et al., 2016)).
As a result, we used FACS-sorted F4/80+ FL macrophage from an EKLF-/-mouse and compared its gene expression with wild type FL F4/80+ macrophage by RNA-Seq to determine which genes are affected by the loss of EKLF. We observe that there are about half as many F4/80+ FL macrophages in EKLF-/-FL as in WT suggesting a vital role for EKLF in FL macrophage development ( Fig. 3A; compare to Fig. S1A). Using k-means clustering of the RNA-Seq data, we find the predominant effect is that genes are downregulated in the EKLF-/-FL macrophages (Fig. 3B). This is consistent with the role of EKLF as a transcriptional activator (Miller and Bieker, 1993). We performed differential gene expression analysis using DESeq2 and find that a set of 1210 genes are significantly downregulated in the EKLF-/-FL macrophages (Fig. 3C, supplementary Table 4). Using Revigo analysis we find that among other, many of the downregulated genes are involved in cell-cell adhesion (Table 1) (complete GO results are available in  supplementary table 5).

EKLF-expressing F4/80+ FL cells are a functionally distinct population from EKLF-F4/80+ cells based on their gene expression program
In our previous study, we had used a mouse strain derived from embryonic stem (ES) cells that contain a single copy of the EKLF promoter directly upstream of a GFP reporter (pEKLF/GFP) to address whether EKLF might be expressed in both the erythroid cell and macrophage (Lohmann and Bieker, 2008). This promoter/enhancer construct is sufficient to generate tissue specific and developmentally correct expression in vitro and in vivo (Chen et al., 1998;Lohmann et al., 2015;Xue et al., 2004;Zhou et al., 2010); thus GFP onset faithfully mirrors EKLF onset (Lohmann and Bieker, 2008). Using this surrogate marker, we had found that ~36% of F4/80+ macrophage singlet cells express EKLF (Xue et al., 2014).
Presently, we used FACS to isolate both F4/80+GFP+ (EKLF+) and F4/80+GFP-(EKLF-) subsets and assayed gene expression using RNA-Seq. Principal component (PCA) (Fig. 4A) and correlation analysis (supplementary Fig. S2A) show that the two populations have widely distinct gene expression profiles. Differential expression analysis shows that 2330 genes are enriched in F4/80+EKLF/GFP+ (supplementary Figure S2B, supplementary Table 6), with EKLF and Vcam1 among the enriched mRNAs consistent with prior work (Xue et al., 2014) ( Fig. 4B,C). In addition, we find that Epor mRNA is also enriched in F4/80+EKLF/GFP+ (Fig. 4B,D). Since Epor+/F4/80+ macrophages form erythroblast islands in bone marrow (Li et al., 2019), our data indicates that the same is true for EKLF+F4/80+ FL macrophages. When we analyze the functional categories of genes significantly enriched in each of the subsets (supplementary Figure S2B), we find that the EKLF/GFP+ F4/80+ subset is enriched for genes involved in heme synthesis, iron transport and homeostasis, and myeloid/erythroid differentiation (Table 2), functions consistent with those performed by erythroblast island macrophages. In contrast, the genes enriched in EKLF-F4/80+ macrophages are mostly involved in innate and cellular immune responses (Table 3) indicating that these are inherently distinct from the EKLF-expressing macrophages in mouse fetal liver.

EKLF specifies expression of a substantial number of genes including important transcription factors in FL macrophages
Both the above datasets provide us with unique information. One (Fig. 3) identifies EKLF-dependent macrophage genes, but does not distinguish between EKLF-expressing and EKLF-deficient macrophages in a genetically unaltered state. The second dataset (Fig. 4) identifies genes with enriched expression in F4/80+ cells where EKLF is also expressed, but does not identify EKLF-dependent genes. By comparing the datasets, we can determine which genes have enriched expression in EKLF-expressing macrophages and are also significantly downregulated in EKLF-/-, and therefore truly EKLF-dependent (Fig. 5A, red box). Overlapping these two independent datasets is an extremely powerful way to parse down the potential direct/indirect genes whose expression are dependent on the presence of EKLF. We find that 504 genes are EKLF-dependent in F4/80+EKLF+ macrophages, a highly significant number given the size of the datasets (Fig. 5B, supplementary Figure S3A, supplementary Table 7).
To determine whether these genes may be under EKLF transcription control, we used Centrimo (MEME suite) to analyze the promoters of these 504 genes for TF motifs that are differentially enriched over a background set comprising promoter sequences of the rest of the transcriptome (supplementary Table 8). Indeed, we find that Klf1 motifs are overrepresented in these promoters, consistent with the idea that they are EKLF-dependent (Fig. 5C). In addition, we find that the motifs of transcription factors Klf3, E2f1, E2f4 and Sp4 are significantly enriched (Fig. 5D) and these TFs are also among the 504 EKLF-dependent genes (Fig.  5E). This strongly suggests that EKLF together with Klf3, E2f1, E2f4, and Sp4 may constitute a transcriptional network regulating the distinct gene expression program of FL island macrophages. E2f2 is also EKLF-dependent in F4/80+ macrophages (Fig. 5E), but its motif is not significantly enriched (Supplementary Figure S3B, E-value=0.17), suggesting that E2f2 may not be a critical part of the EKLF transcription network in island macrophages.
The overlap of the datasets (Figs. 5A,B) suggest that EKLF may regulate the expression of a significant number of other transcription factors in FL macrophages including Foxo3, Ikzf1, MafK, Nr3c1; cell-cycle E2f factors; and other members of the Klf family (Supplementary Figure S3C). This will ultimately be verified by a search of consensus target sequences in putative target genes and by EKLF ChIP. Thus, along with the known transcriptional role of EKLF in erythroid cells, our data is consistent with a global regulatory role for EKLF in proliferation and development of FL island macrophages.
We used immunofluorescence to directly demonstrate that Adra2b protein is expressed in erythroblastic islands (Supplementary Figure S4D). The localization of Adra2b at the surface of the central macrophage cell readily distinguishes it from the more diffuse staining exhibited by F4/80.

Resolving the cellular heterogeneity in F4/80+ FL macrophages
One critical issue is that fetal liver macrophages are a heterogeneous population of cells, a notion readily apparent from the published literature (Lee et al., 2018;Seu et al., 2017), and from our own observation that not all F4/80+ cells express EKLF (Fig. 4). To segregate fetal liver F4/80+ subpopulations and to illuminate the role of EKLF in this process, we performed single cell RNA-Seq on purified F4/80+ fetal liver cells. We used a magnetic bead purification strategy in the presence of Icam4/av inhibitor peptide (Xue et al., 2014) to isolate and maintain healthy F4/80+ cells for single cell barcoding and library preparation using the Chromium V3 platform (see Methods). Using flow cytometry, we find that about 83% of our purified population is F4/80+ after two rounds of selection (Supplementary Figure S5).
Single-cell RNA-Seq confirmed the cellular heterogeneity in the F4/80+ population, with 13 separate clusters of cells after unsupervised dimensionality reduction using the Seurat package (Butler et al., 2018;Stuart et al., 2019) (Fig. 6A). F4/80 mRNA (encoded by the Adgre1 gene) is present in all the clusters, although some clusters have higher levels (Fig. 6B). Additional macrophage markers such as Marco and Vcam1 mRNAs are also present in all clusters, whereas the macrophage transcription factor PU.1 (encoded by Spic) is enriched in clusters 0,1,2, and 8 ( Fig. 6C). Differential enrichment analysis reveals the mRNAs that are enriched in each cluster (Fig. 6D, Supplementary Table 10), and we find certain genes with almost exclusive expression in a particular cluster that serve as markers for that cluster (Supplementary Figure S6).
It is apparent from these analyses that Clusters 0 and 1 have a high overlap in cluster markers ( Fig.  6D), and due to the high expression of macrophage-specific genes ( Fig. 6B,C), these clusters likely are comprised of macrophages. This is also confirmed by GO analysis of the top 100 markers for these clusters (Supplementary Table 10). Further, GO analysis of markers for clusters 2 and 3 yields terms compatible with activated macrophage functions (Supplementary Table 10), and indeed these clusters express genes correlated with activated macrophages such as Csf1r, Dnase2a, and Il4ra (Supplementary Figure S7A). In contrast, GO analysis of the top enriched genes for clusters 4,5,7 and 8 relate to erythro-myeloid characteristics and heme metabolism (Supplementary Table 11), with highly enriched markers for these clusters being glycophorin A, a-synuclein, and a-spectrin (Supplementary Figure S7B). A search for the terminal erythroid marker Ter119 (Ly76) yields no results in our single cell sequencing dataset, indicating that perhaps its mRNA is undetectable and that our F4/80+ purification is largely devoid of terminally differentiating erythroid cells. To further support the heterogeneity of expression in these population, in contrast we find that the mRNA for the constitutively active gene, Gapdh, is uniformly highly expressed in all clusters (Supplementary Figure S7C) whereas CD71 (Tfrc) mRNA was expressed at moderate levels in most clusters (Supplementary Figure S7D).

Cellular heterogeneity in EKLF+ F4/80+ FL macrophages, and an improved strategy to isolate this population
Our earlier observations from the pEKLF/GFP mice indicated that about 36% of the F4/80+ FL cells express EKLF (Xue et al., 2014). EKLF expression is detected exclusively in clusters 4,5, and 7 (Fig. 7A), and these clusters comprise about 23% of the cells in our dataset. We also find that most of the EKLF+ cells express Epor (Fig. 7B), consistent with our earlier observations as well as others (Li et al., 2019). To further test our previous observations that Adra2b expression correlates with EKLF expression and is found in erythroblast island macrophages (Supplementary Figure S4), we looked for Adra2b expression in single cells. We find specific Adra2b enrichment in cluster 4, thus correlating with some EKLF+ as well as Epor+ cells, albeit the remaining EKLF-expressing clusters 5 and 7 have little Adra2b expression (Fig. 7C). This indicates high amounts of heterogeneity even within EKLF+ F4/80+ macrophages, and suggests that Adra2b alone as a marker is not sufficient to enable efficient isolation of EKLF+ F4/80+ cells.
Upon staining E13.5 fetal liver cells with both F4/80 and Adducin2, or F4/80 and Spectrin b antibodies, we find that the majority (~88%) of the Add2+ or Sptb+ cells are F4/80-(and presumably erythroid). However, about 25% of all F4/80+ cells are Add2+ or Sptb+ in each case (Fig. 9A) aligning with our single cell RNA-Seq observations (Fig. 8A), and strongly suggesting that Add2 and Sptb are markers for F4/80+ EKLF+ fetal liver macrophages. We repeated the F4/80+ purification and stained the purified F4/80+ cells for Add2 or Sptb to find that in each case about 24% of the F4/80+ cells are Add2+ or Sptb+ (Fig. 9B), a proportion resembling the 23% of cells in clusters 4,5 and 7 where these mRNAs are expressed. This indicates a high correlation between the Add2 and Sptb protein and mRNA expression in F4/80+ cells. To test the possibility that any Add2 and Sptb expression seen in F4/80+ cells was due to residual erythroid contamination in our F4/80+ population, we performed Imagestream analysis. Using the pEKLF/GFP mouse, we stained for F4/80 and Add2 and we find a number of single cells expressing F4/80 and Add2 that are also EKLF/GFP+ (Fig. 9C). This not only confirms that the Add2 signal is coming from single cells, it also demonstrates visually that Add2 expression in a subset of F4/80+ macrophages correlate with EKLF expression in those macrophages.
Finally, since earlier studies from our group (Xue et al. 2014, Li et al. 2019 showed that EKLF expression is enriched in macrophages forming erythroblast islands, we isolated erythroblast islands and tested for Add2 and Sptb protein expression by immunofluorescence. We find high Sptb and Add2 staining in the central macrophage as well as few surrounding erythroid cells (Fig. 9D, E) indicating that these markers are expressed in erythroblast island macrophages. Thus, Add2 or Sptb can be used as reliable markers to isolate F4/80+ EKLF+ fetal liver macrophage population for further characterization of their unique properties.

Identification of a novel cell type in fetal liver macrophage
Although there is overlap among the cell populations, we have shown that E13.5 murine fetal liver F4/80+ macrophage exhibit a distinct expression pattern when compared to adult spleen F4/80+ macrophage, one that is also divergent from that of fetal liver erythroid cells, thus providing them with a discrete cellular identity. Our data suggests the existence of a unique macrophage cell type with novel markers that defines erythroblastic island-associated macrophage. This is perhaps not surprising, as there is extensive macrophage heterogeneity (Lee et al., 2018;Paulson, 2019;Seu et al., 2017), and it has been long noted that island macrophage may have a distinctive surface marker expression (Manwani and Bieker, 2008).
The unique expression signature exhibited by these cells includes over 300 genes that are functionally involved in positive regulation of developmental processes, particularly cell movement, localization, and adhesion. Our data suggests that establishing a macrophage cell dedicated to maintaining such a unique expression profile makes developmental sense given its role in efficiently aiding the huge demand for red blood cells during early development, specifically within the expanding fetal liver site (Chasis and Mohandas, 2008;Hom et al., 2015;Klei et al., 2017;Manwani and Bieker, 2008;Yeo et al., 2019).

Transient nature of a singular, EKLF-dependent fetal liver macrophage population that coincides with the onset of definitive erythropoiesis during mouse embryonic development
The idea of a dedicated island macrophage cell is further supported by the overlap in the single cell seq and the developmental RNA-Seq expression datasets. These show there is a specific onset of many of the markers of interest that coincide with the peak of EKLF expression in macrophage at E12.5, at the same time as definitive erythropoiesis is occuring in the mouse fetal liver. Strikingly, expression of many of these also dissipate coordinately at later embryonic stages. This may follow from either transient EKLF expression in the macrophage or the transient presence of a population of EKLF-expressing macrophage. Such dynamic regulation has been observed with IL7Ra (Leung et al., 2019), but the remarkable coherence of the erythroblastic island macrophage subset in clusters 4,5,7 suggests the existence of a cross-regulatory mechanism that leads to establishment of a network of genes critical for proper island niche function. Consistent with this, KLF binding motifs are enriched in active macrophage genes (Gosselin et al., 2014;Lavin et al., 2014) and correlate with binding by other macrophage factors such as CJUN and P65 (Link et al., 2018). Our comparative analysis of EKLF-/-and EKLF/GFP+ strongly supports the idea, postulated previously from other studies (Li et al., 2019;Porcu et al., 2011;Xue et al., 2014), that EKLF is a central player in establishing this network at the right time and place in development. Given our studies, the cause of the embryonic lethality in the absence of EKLF could be a combination of impaired erythropoiesis due to the loss of EKLF in developing erythroid progenitors as well as impaired island macrophage function supporting definitive erythropoiesis.

EKLF regulation of island macrophage signature genes
By combining both EKLF-/-and EKLF/GFP+ RNA-Seq data, and then further parsed by the single cell seq data, we find that loss of EKLF expression alters expression of many macrophage genes. We also find that the EKLF-expressing macrophages are functionally different from those not expressing EKLF, with high enrichment of genes performing functions consistent with erythroblast islands. Thus, the subset that are specific to the F4/80+ macrophage and whose expression is EKLF-dependent provides a novel expression signature that identifies targets that may be unique to the erythroblastic island. We have identified three in particular, Adra2b, Add2, and Sptb, that are enriched in EKLF WT macrophage and in the erythroblastic island. As a result, we suggest that these are additional novel markers that, in conjunction with F4/80, provide a further specification to island-associated macrophage identity. Heterogeneity remains an issue; however, from our single cell seq data, it is likely that combining select markers, in particular F4/80+, Add2+, and Sptb+, will distinguish a discrete subpopulation that is highly enriched for island-associated macrophage.
Identification and molecular knowledge of unique island macrophage expression and receptors may be functionally relevant to studies that utilize these cells to help expand in vitro erythropoiesis more efficiently (Hom et al., 2015;Rhodes et al., 2008). These could be used in combination with cytokines known to enhance island macrophage such as erythropoietin (Li et al., 2019), dexamethasone (Falchi et al., 2015;Heideveld et al., 2018), or the KLF1-stimulated combo of ANGPTL7/IL33/SERPINB2 (Lopez-Yrigoyen et al., 2019). Efficient growth and maintenance become important when designing strategies to improve macrophage responses in the context of myelodysplastic syndromes (Buesche et al., 2016) or in the anemia of inflammation (Hom et al., 2015).

Resolution of macrophage heterogeneity
Not surprisingly, we find that the fetal liver F480+ population is heterogeneous, with our single cell analysis suggesting 13 different clusters. Within this mixture we discovered a subset of clusters that express EKLF and its network of genes important for island macrophage. It is of interest that this subset does not express CD11b (Itgam) consistent with studies suggesting it is not an island macrophage marker (Seu et al., 2017;Tay et al., 2020;Ulyanova et al., 2016). Of additional interest, the granulocyte Ly6G marker did not appear in any of our clusters, consistent with an efficient removal of granulocytes during our enrichment procedure.
In this context, it is perhaps surprising that other markers historically suggested to be critical for island function such as Vcam1 are expressed at lower levels in the EKLF clusters than in others. Three explanations can be suggested. First, EKLF+/Vcam1+ cells may be the relevant functional subset of total Vcam1-expressing cells, a different subset of which may have a separate, non-EKLF-dependent function (e.g. homing (Li et al., 2018)). Second, we are not suggesting that EKLF-expressing clusters are the sole source of macrophage islands; there may be others that arise following pathologic conditions (e.g., ßthalassemia or polycythemia vera (Chow et al., 2013;Ramos et al., 2013)), or when comparing steady-state vs stress/anemia (Paulson et al., 2020). Third, erythroblastic islands are also found in bone marrow and spleen, and these arise within a significantly different niche than what we have focused on here during prenatal development. Such directive effects of the environment on macrophage identity have been noted before (Gosselin et al., 2014;Lavin et al., 2014). With respect to our present observations, given the importance of neural signaling in the bone marrow (Mendez-Ferrer et al., 2020), it is possible that a molecule such as Adra2b may be more highly expressed and play a more important role in bone marrow macrophage than in fetal liver macrophage.

Human island macrophage
Collectively, our study shows that EKLF plays a critical role within the specific subset of unique macrophage cells that are transiently required for proper establishment of erythroblastic islands in the developing embryo. Of relevance to human biology (May and Forrester, 2020), although the positive effects of EKLF expression on island macrophage function have been previously noted (Lopez-Yrigoyen et al., 2019), it is also relevant that a recent single cell analysis of human fetal liver hematopoiesis shows that EKLF and many of its target genes identified in the present study are also expressed in the "Vcam1+ erythroblastic island macrophage" cluster (Popescu et al., 2019).

MATERIALS AND METHODS Cell isolation
Fetal livers were dissected from embryonic day E13.5 embryos, and mechanically dispersed into single cells for FACS or RNA isolation. EKLF mouse heterozygotes were as described (Perkins et al., 1995). Photos were taken with a Nikon Microphot-FX fluorescence microscope equipped with a Q-Imaging camera or with a Zeiss Axio Observer Z1 equipped with a Hamamatsu C11440 camera. For single cell sequencing, wild type E13.5 fetal liver cells were isolated from 2 littermate embryos from 1 donor mother, stained with anti-F4/80-PE antibody (eBiosciences #12-4801-80) and isolated using an EasySep™ mouse PE positive selection kit that uses a magnetic bead-based purification strategy (Cell Signaling Technologies #17656) and in the presence of 2mM Icam4/av inhibitor peptide (Xue et al., 2014) to eliminate macrophage-erythroid interactions. The cells were selected by repeating the magnetic bead binding step to increase purity.

Flow Cytometry
Suspended cells from fetal livers were stained for FACS with the following antibodies: anti-mouse F4/80-PE (eBiosciences #12-4801-80), anti-Adra2b (Alomone Labs #AAR-021), anti-Adducinb (Santa Cruz # sc-376063), and anti-Spectrinb1 (Santa Cruz # sc-374309). For anti-Adra2b staining we used an Alexa 647 conjugated Donkey anti-rabbit secondary antibody (Life Technologies). For Adducinb and Spectrinb1 staining, primary unconjugated antibodies were conjugated to AlexaFluor 647 using a primary antibody conjugation kit (Invitrogen # Z11235). Flow Cytometry data was analyzed by FCS Express software and gates were drawn based on unstained and single color compensation controls from the same samples, using the same dyes and within the same experiment.

Imagestream analysis
Cells from intact E13.5 fetal livers were isolated from the pEKLF/GFP mouse and stained with the same antibodies for F4/80 and Adducinb as above, except the primary unconjugated antibody was labeled with a Texas Red labeling kit (Abcam #ab195225). Data was acquired using a Luminex Amnis Imagestream MkII Imaging Flow Cytometer and analyzed using the Amnis Ideas Software.

RNA Isolation and RNA-Seq
FACS sorted cells were directly suspended in Trizol and total RNA was extracted (Rio et al., 2010). RIN values for all EKLF+/+ and EKLF-/-samples were between 9.1-9.8. Poly-A library preparations of biological triplicate samples were analyzed by 100 nt single reads on an Illumina HiSeq 2500 or Illumina Novaseq, 60-90 million reads per sample. For F4/80+EKLF/GFP+ population, the low cell numbers led us to use an Agilent RNA Nanoprep kit (#400753) for isolating reasonably good-quality RNA (RIN~7). RNA-seq data is being submitted to the Gene Expression Omnibus.

Single cell RNA-Sequencing
Libraries were generated from purified F4/80-PE+ using Chromium single cell 3' reagent kit V3 (10X Genomics) to generate cDNA and barcoded indexes for 25,000 individual cells. Paired-end sequencing was performed using a Novaseq instrument.
Single-cell sequencing reads were aligned to the mouse transcriptome build GRCm38.p6vM24 using the software Alevin(Srivastava et al., 2019) and subsequent analysis was performed using the Seurat package (R-based) with built-in functions for plotting, clustering, PCA and U-MAP analysis. After filtering, 3066 cells were retained and for each cell and 4000 variable genes were considered for analysis.
Motif analysis was performed using the Centrimo program (www.meme-suite.org). Promoter sequences from -300 to +100 were extracted using a specific Perl script of Homer for the target EKLFdependent gene set, and the promoters of the rest of the coding genes in the genome were used as background. GO analysis (go.princeton.edu) was performed using GO::TermFinder (Boyle et al., 2004) and GO terms were distilled using Revigo (Jiang and Conrad similarity).