Zebrafish were raised and maintained as described (Westerfield, 2000). The following strains were used: the AB strain, the gfi1aa ^smu10 mutant (Wu et al., 2021), the gfi1b ^smu11 mutant, and the gfi1ab ^smu12 mutant. All zebrafish studies were approved by the South China University of Technology Animal Advisory Committee.

For the gfi1b ^smu11 mutant and the gfi1ab ^smu12 mutant, the gRNA (gfi1b: 5′- gga​gga​aac​tct​gcc​agc​tg-3′, gfi1ab: 5′- ggt​act​cgg​ggt​gtg​aaa​tc-3′) was co-injected with Cas9 protein (NEB, MA, United States; M0646M) into one-cell stage embryos, the gRNAs were synthesized as described (Chang et al., 2013). The raising and screening of mutants were performed as previously described (Chang et al., 2013; (Liu et al., 2014). The genotyping primers were listed in Supplementary Table S1.

Probes synthesis and WISH were carried out as described (Thisse and Thisse, 2008). The following probes were synthesized: gata1, alas2, scl, gata2a, fli1, etv2, sox7, and flk1. Embryos for immunofluorescence were fixed with 4% paraformaldehyde at 23 hpf and dehydrated by methanol. Then the embryos were permeabilized by acetone and stained with GFP antibody (Abcam, Cambridge, UK; ab6658).

For Tg (hsp70:gfi1aa-eGFP) transgenic zebrafish, the embryos injected with pTol-hsp70-eGFP construct and transposase mRNA (Wu et al., 2021) were raised to adult, then the stable transgenic lines were screened as previously described (Westerfield, 2000). To overexpress gfi1aa, 12 hpf embryos were heat shocked for 2 h at 39°C, then the GFP + embryos were picked out for subsequent experiments.

The gfi1aa ^smu10 mutant, gfi1b ^smu11 mutant, and gfi1ab ^smu12 mutant were generated from gfi1aa ^smu10/+ , gfi1b ^smu11/+ , and gfi1ab ^smu12/+ intercrossed embryos by genotyping respectively. The gfi1aa ^smu10 gfi1b ^smu11 mutant, gfi1aa ^smu10 gfi1ab ^smu12 mutant, and the gfi1aa ^smu10 gfi1b ^smu11 gfi1ab ^smu12 mutant were generated from gfi1aa ^smu10/+ gfi1b ^smu11 , gfi1aa ^smu10/+ gfi1ab ^smu12 , and gfi1aa ^smu10 gfi1b ^smu11 gfi1ab ^smu12/+ intercrossed embryos by genotyping respectively. Then, RNA from gfi1-related single, double, and triple mutants as well as WT (wild type siblings) embryos was extracted with TRIzol reagent (Invitrogen, CA, United States; 15596026). Sequencing libraries were generated using the NEBNext® UltraTM RNA Library Prep Kit for Illumina® RNA (NEB; E7770) according to the manufacturer’s instructions.

For RNA-seq data, the sequencing reads were mapped to Ensemble zebrafish reference genome (GRCz11) using STAR alignment software (Dobin et al., 2013). The differential gene expression analysis was performed by DESeq2 (Love et al., 2014). For GO enrichment analysis, the Metascape website (https://metascape.org/gp) (Zhou et al., 2019) was used.

Gfi1aa-GFP ChIP assay was performed as previously described (Wu et al., 2021). In detail, ∼250 WT embryos injected with the hsp-gfi1aa-eGFP plasmid or hsp-eGFP plasmid were heat-shocked and collected at 15 hpf, then the samples were performed by cross-linking, sonication, antibody binding, washing, reverse-cross linking, and ChIP DNA extraction. The ChIP DNA was assessed by qPCR with a LightCycler 96 system (Roche). The comparable WT group and gfi1aa^smu10 mutant group were respectively intercrossed for H3K4me1 ChIP. About 200 embryos of each group were collected at 15 hpf and ChIP DNA was extracted as above. The etv2 ChIP-qPCR primers are used as previously described (Takeuchi et al., 2015), and sox7 ChIP-qPCR primer is listed in Supplementary Table S1.

For the transient GFP reporter assay, pTol-etv2-eGFP and pTol-sox7-eGFP plasmids were constructed for GFP expression under the control of etv2 or sox7 regulatory regions. For the pTol-etv2-eGFP plasmid, the 3.4 kb etv2 promoter (Veldman and Lin, 2012), containing etv2 up-1 to intron-2 region, was cloned by PCR (Primers are listed in Supplementary Table S1) from genomic DNA and inserted into the pTol vector to drive GFP. For the pTol-sox7-eGFP plasmid, the 0.7 kb promoter (containing the Gfi1aa binding peak) was cloned and constructed as above. Then, 100 ng/μL of the construct was injected into the WT control and gfi1aa^smu10 mutant embryos.

MOs for etv2 (5′-cac​tga​gtc​ctt​att​tca​cta​tat​c-3′) (Sumanas and Lin, 2006), lsd1 (5′-gtt​att​cac​acc​ttg​ttg​aga​ttt​c-3′) (Takeuchi et al., 2015), and sox7 (5′-acg​cac​tta​tca​gag​ccg​cca​tgt​g-3′) (Cermenati et al., 2008) were synthesized by Gene Tools and dissolved in water. One-cell stage embryos were collected and injected. For double knockdown, the final concentration of 0.005 pmol etv2 MO and 0.5 pmol sox7 MO were used.

GraphPad Prism 7.0 was used for analysis of experimental data. The Fisher’s exact test was used to compare the difference between two categorical variables. The Unpaired t-test was used to compare the mean difference of two independent groups. The p-value less than 0.05 was considered statistically significant.

To determine the relationship of three Gfi1(s) to primitive hematopoiesis, we utilized a gfi1aa ^smu10 zebrafish mutant (Wu et al., 2021)) and generated gfi1b ^smu11 and gfi1ab ^smu12 zebrafish mutants with CRISPR/Cas9 technology (Supplementary Figure one). Similar to the gfi1aa ^smu10 mutant (Wu et al., 2021), gfi1b ^smu11 and gfi1ab ^smu12 mutants, with a 58-nt insertion (Supplementary Figure S1A) and a 1-nt deletion (Supplementary Figure S1B), respectively, were predicted to disrupt C2H2 type zinc finger domains. To identify the respective roles of Gfi1 members in primitive erythropoiesis, we compared erythroid marker, gata1, expression by WISH in each mutant. We found the expression of gata1 was decreased in gfi1aa ^smu10 mutant embryos compared to their siblings, while no apparent difference in the gfi1b ^smu11 mutant was found compared to siblings (Figure 1A), which is consistent with previously described gfi1aa ^qmc551 and gfi1b ^qmc554 mutants (Moore et al., 2018). We also monitored the phenotype of gfi1ab ^smu12 mutants and found gata1 expression was no altered (Figure 1A), suggesting that loss of gfi1ab does not affect primitive erythropoiesis.

To further identify the relationships among the three gfi1 members, we performed RNA-seq on wild-type (WT), gfi1aa ^smu10 , gfi1b ^smu11 , gfi1ab ^smu12 single mutant, gfi1aa ^smu10 gfi1b ^smu11 , gfi1aa ^smu10 gfi1ab ^smu12 , gfi1b ^smu11 gfi1ab ^smu12 double mutant and gfi1aa ^smu10 gfi1b ^smu11 gfi1ab ^smu12 triple mutant (hereafter referred to as gfi1aa ^−/− , gfi1b ^−/− , gfi1ab ^−/− , gfi1aa/1b ^DM , gfi1aa/1ab ^DM , gfi1b/1ab ^DM , and gfi1 ^TM ). As shown in the RNA-seq heatmap, we found that erythroid markers (alas2, hbae3, hbbe3, gata1a, sptb, trf1a, and epb41b) were decreased in gfi1aa related mutants (gfi1aa ^−/− , gfi1aa/1b ^DM , gfi1aa/1ab ^DM and gfi1 ^TM ) compared to WT and gfi1aa unrelated mutants (gfi1b ^−/− , gfi1ab ^−/− and gfi1b/1ab ^DM ) (Figure 1B). For validation, we further performed gata1 WISH on these mutants. Consistent with the RNA-seq data, the expression of gata1 was not altered in WT and gfi1aa unrelated mutants (Figures 1C,D). The expression of gata1 was decreased in gfi1aa mutants and gfi1aa/1ab ^DM , further decreased in gfi1aa/1b ^DM and the most decreased in gfi1 ^TM (Figures 1C,D). We then explored the genetic interplay among gfi1s and found gfi1b was decreased in gfi1aa-related mutants whereas gfi1ab was ectopic increased in gfi1aa-related mutants (Supplementary Figure S2A,B), suggesting gfi1aa dominates the expression of gfi1b and gfi1ab. These data indicate that Gfi1aa plays a predominant role in promoting primitive erythropoiesis, and that Gfi1ab, together with Gfi1b, play synergistic roles in the process.

Gfi1aa and Gfi1b control primitive erythroblast differentiation by inhibition of endothelial programs (Moore et al., 2018), but the regulatory mechanisms and the key downstream factors are largely unknown. We speculated that Gfi1aa target genes probably exist in the upregulated genes of gfi1aa ^−/− mutant RNA-seq. Through Gene Ontology (GO) enrichment analysis of upregulated genes, we found vasculature development to be the most enriched GO term (Supplementary Figure 3A). Representative endothelial markers (including sox7, flt4, cdh5, clec14a, etv2, and egfl7 (Kaipainen et al., 1995; (Parker et al., 2004; (Sumanas et al., 2005; (Pham et al., 2007; (Cermenati et al., 2008)) were all upregulated in gfi1aa ^−/− mutant RNA-seq (Supplementary Figure S3B). By comparison of the differential expression of the endothelial markers among all gfi1 mutants, we found representative genes were specifically upregulated in all gfi1aa-related mutants (Figure 2A), and particularly upregulated in gfi1 ^TM . These data suggest that Gfi1aa, rather than Gfi1b or Gfi1ab, plays a predominant role in the inhibition of endothelial programs during hemangioblast differentiation into primitive erythrocytes.

As Gfi1(s) function as transcription repressors, it is important to know which genes are directly targeted by Gfi1(s). By reanalyzing our previously performed Gfi1aa-eGFP ChIP-seq data (Wu et al., 2021), we found 12,524 genes bound by Gfi1aa with analyzing the peaks located 2 kb upstream and 2 kb downstream from the transcription start site (TSS) (Figure 2B). When RNA-seq upregulated genes of the gfi1aa ^−/− mutant were combined with the Gfi1aa ChIP targeted genes, we identified 378 candidates that may be directly targeted and transcriptionally suppressed by Gfi1aa (Figure 2B). As expected, the GO term analysis for the 378 candidate targets showed that the vasculature development pathway was highly enriched (Figure 2C). 29 endothelial associated genes were found to be involved in the pathway (Figure 2D). We then compared the differential expression of these genes among all gfi1 mutants and found sox7, flt4, egfl7, cdh5, etv2 were upregulated in gfi1aa-related mutants (Supplementary Figure S4A).

As transcription factors are thought to be critical for cell fate determination, we speculated that some transcription factors may be responsible for Gfi1aa involvement in primitive erythropoiesis. Etv2 and Sox7, two hemangioblast markers, were both highly expressed in mesodermal precursors but downregulated in differentiated hematopoietic cells (Gandillet et al., 2009; (Costa et al., 2012; (Veldman and Lin, 2012; (Sumanas and Choi, 2016). Previous studies showed that overexpression of either one promoted endothelial specification (Kataoka et al., 2011; (Costa et al., 2012). Moreover, etv2 and sox7 genes were highly bound by Gfi1aa-eGFP and their mRNAs were upregulated in gfi1aa-related mutants (Figures 2D,E, Supplementary Figure S4A). Therefore, we speculate that Gfi1aa may directly target and suppress etv2 and sox7 to promote hemangioblast differentiation into primitive erythrocytes by preventing the endothelial specification program.

To test the hypothesis, we first validated our digital data. For validation of ChIP-seq results, we performed a ChIP-PCR assay using the pTol2-hsp-gfi1aa-eGFP construct to assess whether Gfi1aa could bind to etv2 and sox7 regulatory regions (Figure 3A). Previous data showed that three etv2 regulator regions (up1, -110 ∼ -35bp and intron-2) recapitulated etv2 expression (Veldman and Lin, 2012). ChIP PCR results showed that Gfi1aa could bind to these etv2 regulator regions (up1, -110 ∼ -35bp, intron-2) compared to the gene body control region (exon-8) (Figures 3B,C), which is consistent with the ChIP-seq data (Figure 2E). Moreover, ChIP PCR also showed an enrichment of Gfi1aa on sox7 regulatory region (-520 ∼ 180bp) (Figures 3D,E). These data suggest that the regulatory regions of etv2 and sox7 were directly bound by Gfi1aa.

As etv2 and sox7 are the master regulators of hematopoietic/endothelial cell differentiation, we examined whether etv2 and sox7 were the specific downstream target genes of Gfi1aa. We detected a series of hemangioblast markers—scl, gata2, and fli1, as well as etv2 and sox7—at the beginning of primitive hematopoiesis. The results showed that etv2 and sox7 expression were markedly increased in gfi1aa ^−/− mutants compared to siblings, while expression of scl, gata2, and fli1 was not altered (Supplementary Figure S5A). The expression of etv2 and sox7 by qPCR also showed a similar increase in gfi1aa ^−/− mutants compared to WT (Supplementary Figure S5B). The WISH and qPCR results verified the RNA-seq results that etv2 and sox7 are upregulated in gfi1aa ^−/− mutants.

We further performed reporter assays to determine whether Gfi1aa could repress etv2 and sox7 transcription in vivo. We generated pTol-etv2-eGFP and pTol-sox7-eGFP reporter constructs and injected each construct into gfi1aa ^+/- intercross embryos to monitor whether GFP expression was affected by Gfi1aa (Figure 3F). The reporter assays showed that both etv2-eGFP and sox7-eGFP expression were increased in gfi1aa ^−/− mutants compared to their respective WT control (Figures 3G,H), suggesting a transcriptional repressive role for Gfi1aa in etv2 and sox7 regulatory regions.

The above data demonstrated that Gfi1aa targets the regulatory regions of etv2 and sox7 and suppresses their transcription.

We were eager to know whether downregulation of sox7 rescued the blood deficiency of the gfi1aa ^−/− mutant. We injected sox7 MO into gfi1aa ^−/− mutants and found that alas2 ⁺ erythroid cell reduction and flk1 ⁺ endothelial cell augmentation within the intermediate cell mass (ICM) region could be partially restored (Supplementary Figures S6A–D). It has been reported that etv2 MO can also partially rescue gfi1aa mutant primitive hematopoietic defects (Moore et al., 2018). These data suggest that Gfi1aa targets not only etv2 but also sox7 to promote primitive erythrocyte differentiation from the hemangioblast.

Given the fact that either etv2 or sox7 partially rescued the primitive erythrocytes of the gfi1aa mutant, we speculated that sox7 might cooperate with etv2 for Gfi1aa regulated primitive erythropoiesis. To test this hypothesis, we knocked down both genes in gfi1aa ^−/− mutants to see if the hemangioblast differentiation defect could be further rescued. As a high dosage of etv2 MO could cause severe vasculature defects of developing embryos (Sumanas and Lin, 2006), the cooperative effect on endothelial cells between etv2 MO and sox7 MO would be masked. Owing to this, we decreased etv2 MO concentration and found 0.01 pmol etv2 MO was enough to partially rescue the erythroid defect in gfi1aa mutant but not affect the vasculature which concentration was comparable to sox7 MO (Supplementary Figures S7A–D). We therefore utilized the low dosage etv2 MO to involve in the double knockdown. Results showed that alas2 ⁺ erythroid cell reduction and flk1+ endothelial cells augmentation in gfi1aa ^−/− mutants could be almost completely restored (Figures 4A–D). These data suggest that the two transcription factors, sox7 and etv2, act cooperatively downstream of Gfi1aa during hemangioblast differentiation.

As lsd1-deficient zebrafish (Takeuchi et al., 2015) phenocopied gfi1aa ^−/− mutants during primitive hematopoiesis and Gfi1aa could interact with Lsd1 in zebrafish (Wu et al., 2021), we speculated that Gfi1aa regulated hemangioblast differentiation into primitive erythrocytes was dependent upon Lsd1. We first inhibited lsd1 to assess Gfi1aa repression of etv2 and sox7, and found that the repression was indeed dependent on lsd1. Inhibited etv2 and sox7 expression levels in gfi1aa-overexpressing (gfi1aa-OE) embryos were rescued by downregulating lsd1 (Figures 5A,B). This suggests that Gfi1aa requires Lsd1 to function as a transcriptional repressor. Furthermore, gfi1aa-OE rescued decreased alas2 and increased flk1 in gfi1aa ^−/− mutants, but downregulation of lsd1 in gfi1aa-OE gfi1aa ^−/− mutants showed similar expression patterns to gfi1aa ^−/− mutants so that counteracted the restoration by gfi1aa-OE (Figures 5C–F), suggesting that Gfi1aa requires Lsd1 to function in promotion of hemangioblast differentiation into the primitive erythroid lineage.

Lsd1 is a histone demethylase that has been shown to repress etv2 by alteration of associated H3K4 methylation during zebrafish primitive hematopoiesis (Takeuchi et al., 2015). Therefore, H3K4 methylation of etv2 and sox7 in gfi1aa ^−/− was assessed. The results showed H3K4me1 levels (primed and active enhancers marker (Heintzman et al., 2007; (Mercer et al., 2011)) to be upregulated in the regulatory regions of the two genes in gfi1aa ^−/− mutants (Figures 5G,H), suggesting that Gfi1aa and Lsd1 downregulate etv2 and sox7 by suppressing their H3K4me1 levels.

The above data demonstrate Gfi1aa to depend on Lsd1 to repress downstream etv2 and sox7 by altering H3K4 methylation during primitive hemangioblast differentiation.