Zebrafish Hif3α modulates erythropoiesis via regulation of gata1 to facilitate hypoxia tolerance

O₂ is indispensable for the survival of aerobic organisms (Aragonés et al., 2009; Majmundar et al., 2010; Semenza, 2014). Organisms have evolved sophisticated cellular sensors that respond to O₂ gradients (Bigham and Lee, 2014; Prabhakar and Semenza, 2015, 2016). Hypoxia is condition characterized by low ambient O₂, triggering acute and chronic organismal responses and inducing the expression of numerous genes (Semenza, 2014; Aragonés et al., 2009; Greer et al., 2012; Prabhakar and Semenza, 2012; Semenza, 2012; Shen and Kaelin, 2013). Hif1α (Hif1ab) and Hif2α (Epas1b) are regulators of the cellular response to O₂ (Majmundar et al., 2010; Semenza, 2012, 2014)_. Under normoxia, Phd1, Phd2, and Phd3 (Egln3) use O₂ and 2-oxoglutarate as substrates for the hydroxylation of Hif1α and Hif2α. Hydroxylated Hif1α and Hif2α are bound by VHL protein. VHL recruits a ubiquitin ligase complex that targets Hif1α and Hif2α for proteasomal degradation. Hypoxia inhibits Phd enzymatic activity, preventing the Phds from hydroxylating Hif1α and Hif2α. This results in Hifα protein stabilization and the induction of transcriptional activity (Bishop and Ratcliffe, 2015; Semenza, 2014).

Hif3α (Hif1al) is another Hifα protein (Augstein et al., 2011; Duan, 2016; Ravenna et al., 2016). Different from Hif1α and Hif2α, Hif3α comprises a transactivation domain (TAD), a leucine zipper domain (LZIP), and an LXXLL motif (Gu et al., 1998; Zhang et al., 2012). Therefore, Hif3α may be functionally distinguishable from Hif1α and Hif2α. Mammalian HIF3a genes use different promoters, different transcription initiation sites and alternative splicing to transcribe a large number of mRNA variants (Duan, 2016). Some of the short HIF3α isoforms lack TADs (Hara et al., 2001), and others have weak or absent transcriptional activity (Gu et al., 1998; Pasanen et al., 2010). Moreover, the overexpression of some HIF3α isoforms suppresses HIF1α- and/or HIF2α-induced reporter activity in cells (Maynard et al., 2003; Makino et al., 2001, 2007; Yamashita et al., 2008). Thus, it has been widely accepted that HIF3α acts as a negative regulator of HIF1α and HIF2α, even though, to date, only partial variants of mammalian HIF3α transcripts have been investigated, mostly via overexpression in cell culture systems with artificial reporter constructs (Duan, 2016; Ravenna et al., 2016). However, multiple lines of evidence support that the full length of HIF3α acts as a transcription factor (Duan, 2016; Heikkilä et al., 2011; Zhang et al., 2014; Zhou et al., 2018).

Interestingly, in zebrafish, only two isoforms of Hif3α (Hif3α/Hif3α1 and Hif3α2) have been identified (Zhang et al., 2016, 2012, 2014; Makino et al., 2001). Zebrafish Hif3α is a hypoxia-induced transcription factor that activates gene expression distinct from HIF1α (Zhang et al., 2014), whereas Hif3α2 is an oxygen-insensitive nuclear protein that inhibits canonical Wnt signaling by binding to β-catenin and destabilizing the nuclear β-catenin complex (Zhang et al., 2016).

To date, the roles of Hif3α in hypoxia signaling, and the mechanisms underlying these roles, are almost entirely unclear. Here, we knocked out hif3a in zebrafish and found that the resulting mutants exhibited increased sensitivity to hypoxia and reduced erythropoiesis. Our mechanistic studies indicated that Hif3α acted as a transcription factor and directly regulated gata1 expression.

Zebrafish carry two isoforms of hif3a: hif3a/hif3a1 (herein referred to as hif3a) and hif3a2 (Fig. S1A,B; Zhang et al., 2016, 2012, 2014). We designed two gRNAs to disrupt the expression of this gene (Fig. S1A). Two mutants in the hif3a gene – hif3a^{ihb20180620/ihb20180620} (http://zfin.org/ZDB-ALT-180620-1), herein designated M1, and hif3a^{ihb20180621/ihb20180621} (http://zfin.org/ZDB-ALT-180620-2), herein designated M2 – were screened (Fig. S1A-C). The mutant hif3a encoded two truncated peptides (Fig. S1B). hif3a mRNA expression was largely downregulated in the two mutants compared with the wild type (WT; Fig. S1D). An anti-Hif3α antibody had been developed and confirmed to recognize zebrafish Hif3α protein specifically (Zhang et al., 2012). Using western blot analysis, Hif3α protein could not be detected in the mutant (Fig. S1E). Overall, hif3a^−/− zebrafish were identical to their WT siblings (hif3a^+/+) under normal conditions. Of note, the predicted truncated peptide of M1 contains the basic helix-loop-helix (bHLH) domain and that of M2 contains bHLH-PAS-PAC-ODD domains. The bHLH domain is important for DNA binding and dimerization with Hif1β. The PAS-A/B and PAC domains are also involved in Hif1β for dimerization (Semenza, 2014). To determine whether M1 and/or M2 mutant proteins may act in a dominant-negative manner, we examined overexpression of the predicted truncated peptides of M1 and M2 on a hypoxia response element (HRE)-luciferase reporter activity. As shown in Fig. S2A-C, overexpression of the predicted truncated peptides of M1 and M2 had no effect on the transcriptional activity of hif1ab, hif2ab and hif3a in epithelioma papulosum cyprini (EPC) cells. In the following experiments, we primarily used mutant M1 (hif3a^{ihb20180620/ihb20180620}) for phenotype analysis, and confirmed the observed M1 phenotypes in M2 (hif3a^{ihb20180621/ihb20180621}) to exclude off-targeting effects.

Given that Hif3α has been identified as an oxygen-dependent factor (Zhang et al., 2014), we aimed to determine whether disruption of hif3a impacted zebrafish hypoxia tolerance (Cai et al., 2018). In this study, after exposing hif3a^+/+ and hif3a^−/− larvae to 2% O₂ simultaneously for 12 h, more hif3a^−/− larvae were dead than hif3a^+/+ larvae (Fig. 1A,B). Under normoxia (21% O₂), no significant differences were detected between hif3a^+/+ and hif3a^−/− larvae (Fig. 1A,D).

Subsequently, we measured the hypoxia tolerance of adult zebrafish [3 months post fertilization (mpf)]. When hif3a^+/+ and hif3a^−/− adults with similar body weights (0.32±0.02 g; mean±s.d.) were subjected to hypoxia (5% O_2, adjusted before experimentation) simultaneously for 30 min, there were no obvious differences in behavior (Movie 1). However, as the duration of hypoxia increased, two hif3a^−/− zebrafish appeared dead or near dead, whereas three hif3a^+/+zebrafish remained active (Movie 2).

We then tested hif3a^+/+ and hif3a^−/− adults (6 mpf), with similar body weights (0.65±0.02 g), which were subjected to hypoxia (5% O_2, adjusted before experimentation). After 30 min, no significant difference in behaviors was observed between the hif3a^+/+ and hif3a^−/− (Fig. 1C). However, hif3a^−/− began to die after 46 min of hypoxia. After 50 min of hypoxia, all hif3a^−/− zebrafish were dead, and all hif3a^+/+ zebrafish were still alive (Fig. 1C). Therefore, hif3a^−/− zebrafish were more sensitive to hypoxia than hif3a^+/+ zebrafish.

We investigated whether the difference in hypoxia tolerance exhibited between hif3a^+/+ and hif3a^−/− zebrafish was due to hif3a^−/− zebrafish having higher oxygen consumption. Unexpectedly, in fact, the oxygen consumption rate of the hif3a^+/+ was even higher than that of the hif3a^−/− (Fig. 1E), indicating that the oxygen consumption is not the cause. In order to validate the fact that the dissolved O₂ in water of the flasks is actually correlated with the O₂ concentration previously adjusted in the hypoxia workstation, we measured the dissolved O₂ in water with an LDO101 probe at different time points when the flasks were put into the hypoxia workstation set at 5% O₂ and 2% O₂ respectively (Fig. S2D,E). As expected, the dissolved O₂ in the water in the 2% O₂ workstation decreased faster than that in the 5% O₂ workstation, suggesting a precise correlation (Fig. S2D,E). Thus, our data suggested that disruption of hif3a attenuated hypoxia tolerance in zebrafish.

When we routinely examined the hif3a^+/+ and hif3a^−/− larvae under a dissection microscope, we noticed that the hif3a^−/− larvae always had fewer blood cells compared with hif3a^+/+ larvae. Given the importance of red blood cells for hypoxia tolerance (Bigham and Lee, 2014; Lee and Percy, 2011; Lorenzo et al., 2014; Sun et al., 2017), we measured the red blood cells of hif3a^+/+ and hif3a^−/− embryos using o-Dianisidine staining. At 36 h post fertilization (hpf), there were fewer o-Dianisidine-positive cells in the hif3a^−/− embryos than in the hif3a^+/+ embryos (Fig. 2A,B). Gata1 is an erythroid-specific transcription factor that is essential for erythropoiesis, and Tg(gata1:eGFP) zebrafish are widely used for monitoring living red blood cells (de Jong and Zon, 2005; Ferreira et al., 2005; Long et al., 1997; Lyons et al., 2002). To validate our observed phenotype, we mated Tg(gata1:eGFP) zebrafish with hif3a^−/−, generating Tg(gata1:eGFP)/hif3a^+/+ and Tg(gata1:eGFP)/hif3a^−/−. From 24-48 hpf, we observed fewer gata1-positive cells in the hif3a^−/− than in their WT siblings (Fig. 2C,D). These data suggest that knockout of hif3a disrupts erythropoiesis in zebrafish.

Reduced hypoxia tolerance was not only exhibited by the hif3a^−/− larvae (Fig. 1A,B,D), but also by the hif3a^−/− adults (Fig. 1C; Movie 2). Thus, we examined erythrocyte numbers in adult. As it is difficult to measure total erythrocytes in each adult, we used relative erythrocyte number (the number of cells counted in a given blood volume) to compare hif3a^+/+ and hif3a^−/− adults. Consistently, hif3a^−/− had fewer erythrocytes than hif3a^+/+ (Fig. 2E).

To determine whether the defective erythropoiesis displayed by the hif3a^−/− was associated with erythroid maturation, we analyzed the morphology of isolated red blood cells using May-Grunwald-Giemsa staining (Fig. 2F,G; De La Garza et al., 2016). hif3a^−/− had a higher percentage of proerythroblasts and a lower percentage of mature erythroid precursors at 2 days post fertilization (dpf) than in the WT (Fig. 2F). The relative level of proerythroblasts decreased in the hif3a^−/− at 5 dpf, but remained higher than the level in their WT siblings (Fig. 2G). These data suggested that the deletion of hif3a might impede erythroid cell maturation, resulting in fewer mature red blood cells in hif3a^−/− embryos.

To determine whether loss of one copy of hif3a can affect red blood cells and survival rate, we also compared the red blood cells among hif3a^+/+, hif3a^+/− and hif3a^−/− embryos using o-Dianisidine staining. No significant difference was detected between hif3a^+/+and hif3a^+/− (Fig. S2F). In agreement with this, under hypoxia, the death curve was similar between hif3a^+/+and hif3a^+/− (Fig. S2G).

To figure out the mechanisms of hif3a on erythropoiesis, we examined the expression of hematopoietic markers using whole mount in situ hybridization. scl (tal1) and lmo2 are two primitive progenitor cell marker genes in zebrafish hematopoiesis (de Jong and Zon, 2005). At the 10-somite stage, no significant difference was detected in expression levels of scl and lmo2 between hif3a^+/+ and hif3a^−/− (Fig. S3A). MyoD staining (the somatic mesoderm marker) at the 14-somite stage indicated that overall embryogenesis was not influenced by disruption of hif3a (Fig. S3B). However, at 24 hpf, gata1 expression was dramatically reduced in hif3a^−/− embryos compared with hif3a^+/+ embryos. Consistently, the expression levels of alas2 (a key enzyme for heme biosynthesis) and band3 (slc4a1a; an erythroid-specific cytoskeletal protein) were reduced in hif3a^−/− at 24 hpf (Fig. 3A; Brownlie et al., 1998; Paw et al., 2003). In addition, the expression levels of hbae1, hbae3 and hbbe1 (three erythrocyte-specific hemoglobin genes), were reduced in hif3a^−/− at 48 hpf compared with hif3a^+/+ (Fig. 3B). The downregulation of gata1 expression in hif3a^−/− at 24 hpf was confirmed with quantitative RT-PCR assays (qRT-PCR) (Fig. S4A). The decreased expression levels of alas2, band3, hbae1, hbae3 and hbbe1 in hif3a^−/− compared with hif3a^+/+ were also confirmed by qRT-PCR assays (Fig. S4A,B).

Murine models suggest that Runx1 and c-myb are important factors for adult erythropoiesis (Ferreira et al., 2005). Based on the erythrocyte reduction we observed in adult hif3a^−/−, we sought to determine whether runx1 and c-myb were also downregulated in adult hif3a^−/−. Surprisingly, runx1 and c-myb (myb) were upregulated, not downregulated, in the kidneys of adult hif3a^−/− compared with hif3a^+/+ (Fig. S4C). These results suggest that hif3a might not induce runx1 and c-myb expression, and that the decreased erythrocytes in adult hif3a^−/− might not be because of the effects of runx1 and c-myb.

The glycoprotein hormone erythropoietin (Epo), which is induced by Hifα, regulates red blood cell mass, connecting the hypoxia signaling pathway with erythropoiesis (Lee and Percy, 2011). To determine whether hif3a modulates adult erythropoiesis by regulating epo (epoa), similar to the behavior of hif1a and hif2a, we measured epo expression in adult zebrafish kidneys. Unexpectedly, epo expression was upregulated, not downregulated, in hif3a^−/− compared to hif3a^+/+ (Fig. S4C). To further determine whether the modulation of erythropoiesis by hif3a is indeed independent of Epo, we examined the effect of micro-injection of epo mRNA on erythropoiesis in hif3a^−/− embryos. Of note, micro-injection of epo mRNA could not rescue the defects of erythropoiesis in hif3a^−/− embryos (Fig. S5). These data suggested that hif3a might not modulate erythropoiesis by directly regulating epo expression.

To further confirm that erythropoiesis defects of hif3a^−/− were specifically due to silencing of hif3a, we microinjected synthesized hif3a mRNA into hif3a^−/− embryos at the one-cell stage. Expression of microinjected hif3a mRNA was confirmed (Fig. S6A). We then examined red blood cells using o-Dianisidine staining, and quantified marker gene expression using whole-mount in situ hybridization (WISH) and qRT-PCR assays. At 36 hpf, embryos microinjected with hif3a mRNA had more red blood cells than embryos microinjected with GFP-mRNA (Fig. 3C). Consistently, the expression levels of gata1, alas2, band3, hbae1, hbae3 and hbbe1 were higher in the hif3a^−/− embryos microinjected with hif3a mRNA compared with the hif3a^−/− embryos microinjected with GFP-mRNA (Fig. 3D,E; Fig. S6B,C).

These data suggest that the disruption of zebrafish hif3a abrogated the expression of hematopoietic marker genes, resulted in defects of erythropoiesis; and that gata1 might be the downstream effector mediating the function of hif3a in erythropoiesis.

Zebrafish have two waves of hematopoiesis, primitive hematopoiesis (embryonic hematopoiesis) and definitive hematopoiesis (adult hematopoiesis) (de Jong and Zon, 2005; Paik and Zon, 2010). gata1 is crucial for both primitive and definitive erythropoiesis (Ferreira et al., 2005). In this study, we found that gata1 was downregulated in hif3α^−/−, which correlated well with the reduction of erythrocytes in hif3α^−/−. Therefore, gata1 might be the main target by which hif3α mediates erythropoiesis.

Although the function of mammalian HIF3α is debatable due to the complexity of the splicing isoforms, zebrafish Hif3α serves as an oxygen-dependent transcription factor (Zhang et al., 2016). We observed that erythroid cell maturation was retarded and gata1 expression was reduced during erythropoiesis in hif3a^−/− zebrafish. Therefore, we attempted to determine whether zebrafish Hif3α acted as a transcription factor to regulate gata1 expression. Initially, we examined expression patterns of hif3a and gata1 in adult zebrafish tissues (3 mpf) as well as at different developmental stages. hif3a was highly expressed in kidney, and gata1 was highly expressed in spleen and kidney (Fig. 4A,B), indicating a correlation expression pattern between hif3a and gata1 in tissues. Intriguingly, during development hif3a expression reached its highest level from 12-16 hpf, gata1 expression started to increase from 12 hpf and reached its highest level at 16 hpf (Fig. 4C,D), further implying an intrinsic connection between hif3a and gata1 expression.

Subsequently, we examined whether Hif3α had transcriptional activity using an artificial luciferase reporter assay system in embryos (Zhou et al., 2009). Hif3α indeed had transcriptional activity (Fig. 4E). Subsequently, we prepared a series of deletion and mutation constructs for the zebrafish gata1 promoter luciferase reporter (Fig. 4F). Overexpression of hif3a significantly activated the gata1 promoter luciferase constructs, −1380-+1580, −890-+1580, −406-+1580 and −164-+1580 in EPC cells (Fig. 4G). However, when a potential HRE (GCGTG) located at −105-−101 was mutated (GAAAG) (Fig. 4F), the promoter luciferase reporter (−406-+1580/HRE mutant) was not activated by overexpression of hif3a in EPC cells (Fig. 4H). Further chromatin-immunoprecipitation (ChIP) assays using anti-Hif3α antibody (Zhang et al., 2012) indicated that Hif3α could bind to the gata1 promoter-containing HRE site (Fig. 4I).

In addition to hif3a, another splicing alternative isoform is known in zebrafish: hif3a2 (Duan, 2016; Zhang et al., 2016). Disruption of hif3a at two loci also generated two novel peptides (M1 and M2) (Fig. S2B). To determine whether these three proteins affected gata1 induction, we performed promoter assays. Interestingly, overexpression of these three proteins did not activate the gata1 promoter (Fig. S7A-C). These findings not only suggested that hif3a plays a specific role for gata1 induction, but also indicated that the knockout of hif3a at two loci completely disrupted hif3a function in zebrafish.

In the mutant M2 (hif3a^{ihb20180621/ihb20180621}), we confirmed that the expression levels of gata1, alas2, hbae1 and hbbe1 were reduced compared with WT siblings (hif3a^+/+) (Fig. S8A,B). Thus, our results suggested that zebrafish hif3a directly activated gata1 expression by recognizing the HRE site located in the promoter of gata1.

Whether HIF3α acts as a dominant negative transcriptional regulator of HIF1α and/or HIF2α, or acts as a transcription factor in response to hypoxia, is largely dependent upon the variant and the biological model, particularly in mammals (Heikkilä et al., 2011; Makino et al., 2007; Maynard et al., 2005). However, in zebrafish, Hif3α binds to the promoter sequences of several genes, and induces the expression of these genes under hypoxic conditions (Zhang et al., 2014). Here, we provide additional evidence supporting that zebrafish Hif3α serves as a transcription factor to induce gata1 expression.

As reported previously, Hif3α is degraded during normoxia in zebrafish (Zhang et al., 2014). However, in this study we observed that defects of erythropoiesis in hif3a^−/− zebrafish were steady-state. We sought to determine whether Hif3α protein stability was also steady-state from embryos to adult tissues. Using western blot analysis, we confirmed that Hif3α protein was stable from embryos to adult tissues (Figs S1E and S9).

In addition, we noticed that disruption of hif3a enhanced expression of hif1ab and hif2ab, suggesting some redundant functions between hif3a and hif1a/hif2a in zebrafish (Fig. S10A,B). Consistent with this notion, the hif1a downstream targets glut1 and pdk1, and the hif2a downstream targets pou5f1 and pai1 were increased in hif3a^−/− larvae (Fig. S10C-F).

Given the well-known role of hif1a in regulating erythropoiesis (Semenza, 2009), we intended to determine whether microinjection of hif1ab mRNA could rescue the defects of erythropoiesis in hif3a^−/− embryos. Based on o-Dianisidine staining of embryos, microinjection of hif1ab mRNA could partially restore the defects of erythropoiesis in hif3a^−/− embryos (Fig. S10G-I). Furthermore, we found that the red blood cell numbers were partially recovered and their maturation was obviously fixed after microinjection of hif1ab mRNA (Fig. S10J-M), which seemed to rely on gata1 upregulation because gata1 expression was indeed increased (Fig. S10N,O).

In this study, we noticed that disruption of hif3a could cause redundant upregulation of hif1ab. It appeared that microinjection of hif1ab mRNA could induce gata1 upregulation, resulting in partially rescuing defects of erythropoiesis in hif3a^−/− embryos. However, disruption of hif3a in zebrafish eventually caused defects of erythropoiesis. Therefore, the direct upregulation of gata1 by Hif3α might account for a main mechanism of Hif3α in modulating erythropoiesis of zebrafish.

Given the well-known role of Phd enzymes and VHL proteins in regulating HIF activity and the similarity between HIF1α, HIF2α and HIF3α, we sought to determine whether zebrafish phd2a, phd2b, phd3 and vhl have effects on hif3a activity. We performed promoter assays and western blot analysis. Co-expression of phd2a, phd2b, phd3 and vhl decreased the activity of HRE luciferase reporter and gata1 promoter reporter induced by Hif3α (Fig. S11A,B). As expected, co-expression of phd2a, phd2b, phd3 and vhl also caused Hif3α protein degradation (Fig. S11C). These data suggest that Hif3α might behave similar to Hif1α and Hif2α in the hypoxia signaling pathway.

We used CRISPR/Cas9 to knock out hif3a in zebrafish (Danio rerio). First, hif3a sgRNA was designed using the CRISPR design tool (http://crispr.mit.edu). The zebrafish codon Optimized Cas9 plasmid (provided by Dr Bo Zhang, Peking University, China) was digested using XbaI, then purified and transcribed using the T7 mMessage mMachine Kit (Ambion). We used a PUC9 gRNA vector (provided by Dr Bo Zhang, Peking University, China) to amplify the hif3a sgRNA template. The primers used to amplify hif3a sgRNA were: forward primer 1 (mutant 1), 5′-GTAATACGACTCACTATAGGACAAAGCTGCCATCATGAGTTTTAGAGCTAGAAATAGC-3′; forward primer 2 (mutant 2), 5′-GTAATACGACTCACTATAGGTGGTGTTATTTCACTCTGGTTTTAGAGCTAGAAATAGC-3′; and the reverse primer 5′-AAAAGCACCGACTCGGTGCC-3′. SgRNA was synthesized using the Transcript Aid T7 High Yield Transcription Kit (Fermentas).

We injected zebrafish embryos at the one-cell stage (generated as described above) with 1 ng Cas9 RNA and 0.15 ng sgRNA per embryo. The mutations were initially detected using a heteroduplex mobility assay (HMA) as previously described (Cai et al., 2018). Briefly, a short fragment that included the target site was amplified from genomic DNA and two-step PCR was carried out as follows: 95°C for 2 min, and 40 cycles of 95°C for 10 s, 55°C for 30 s and 72°C for 30 s. PCR amplicons were electrophoresed on 15% polyacrylamide gels for 30 min. If the HMA results were positive, the remaining embryos were raised to adulthood as the F0 generation, and were then backcrossed with WT zebrafish (strain AB) to generate the F1 generation. F1s were genotyped with HMAs. Genotype was confirmed by sequencing target sites. Heterozygous F1s were back-crossed with WT zebrafish (strain AB; disallowing offspring-parent matings) to generate the F2 generation. F2 adults carrying the target mutation were inter-crossed to generate F3 offspring. The F3 generation contained WT (+/+), heterozygous (+/−) and homozygous (−/−) individuals. The primers used to identify mutants were: forward primer 1 (mutant 1), 5′-AGTTTGGAGCAGCGGAAG-3′; reverse primer 1 (mutant 1), 5′-AGCATTAGGACATTATGCAGGT-3′; forward primer 2 (mutant 2), 5′-CGAAAGGACAGTCAGAGGTAGA-3′; and reverse primer 2 (mutant 2), 5′-ACCGTTTCCTAGAATTACTGGTTAG-3′. The two novel mutants were named following zebrafish nomenclature guidelines, hif3a^{ihb20180620/ihb20180620} (http://zfin.org/ZDB-ALT-180620-1) and hif3a^{ihb20180621/ihb20180621} (http://zfin.org/ZDB-ALT-180620-2).

Zebrafish strain AB, as well as the transgenic line Tg(gata1:EGFP) (provided by Tingxi Liu, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, China) were raised, maintained and staged according to standard protocols. EPC cells originally obtained from the American Type Culture Collection were cultured in M199 medium supplemented with 10% fetal bovine serum, maintained at 28°C in a humidified incubator containing 5% CO₂. EPC cells were transfected with the constructed plasmids using VigoFect (Vigorous Biotechnology) following the manufacturer's instruction. pTK-Renilla (Promega) was used as an internal control. After transfection, the luciferase activity was measured using the dual-luciferase reporter assay kit following the manufacturer's instruction (Promega).

The Ruskinn Invivo2 I-400 workstation was used for hypoxia treatment of zebrafish (larvae and adults). The O₂ concentration was adjusted to the appropriate value (2% for larvae and 5% for adults) before experimentation. In our previous studies, we noted that the body weight of adult zebrafish significantly affected hypoxia tolerance (Cai et al., 2018). Therefore, we selected adult zebrafish (at 3 and 6 mpf) with similar body weights (0.32±0.02 g; 0.65±0.02 g) for the hypoxia tolerance tests.

For the hypoxia treatments of zebrafish larvae, hif3a-null and WT zebrafish were placed into a 10 cm cell culture dish filled with 30 ml of water. The oxygen concentration in the Ruskinn INVIVO2 I-400 workstation was adjusted to 2% ahead of time. Each experiment was repeated three times.

WISH was performed as described previously (Hu et al., 2014). Probes for scl, lmo2, gata1, myoD, alas2, band3, hbae1, hbae3 and hbbe1 were amplified from the cDNA pool using the primers.

O-Dianisidine staining for hemoglobin was performed as previously described (Hu et al., 2014). Live embryos were soaked in o-Dianisidine staining solution (0.6 mg/ml o-Dianisidine, 166ul 3 M ammonium acetate, 20 ml absolute ethyl alcohol, 30 ml ddH₂O) for 15 min in the dark.

The software IpWin32 was used for quantifying the erythrocyte numbers, GFP-positive cell numbers and gene expression levels in WISH staining. The cell numbers and gene expression levels were measured from the field of view with a same square, and different larvae were chosen for counting (n=3).

EPC cells were seeded in 24-well plates and transfected with the indicated plasmids together with zebrafish gata1 promoter luciferase reporters and pTK-Renilla as an internal control. Luciferase activity was measured 20-24 h after transfection using the Dual-luciferase Reporter Assay System (Promega). For embryos, the plasmids were injected into the embryos at the one-cell stage. About 30 embryos were harvested at 10 hpf and homogenized in Passive Lysis Buffer (Promega). Each experiment was conducted in triplicate and repeated at least three times.

The embryos (2 dpf and 5 dpf) were placed in 1× PBS dropped on glass slides. The blood cells were released by puncturing the pericardial sac and upper yolk sac of embryos with fine forceps. The slides were air dried at room temperature before staining. The blood cells were stained with May–Grunwald-Giemsa solution 1 (100 μl; ServiceBio) for 5 min, briefly rinsed in purified water, and then stained with May–Grunwald-Giemsa solution 2 (200 μl) for 10 min and briefly rinsed with purified water. Once the slides were dry, a drop of neutral resin was added. Subsequently, the slides were covered with slips and dried overnight. The stained blood cells were visualized and photographed under a 100× oil-immersion lens.

Adult zebrafish (n=3 for hif3a^+/+ and hif3a^−/−; 6 mpf; body weight=0.63±0.01 g) were skin-dried carefully using filter paper and dissected near to the heart region using an eye scissor. Approximately 15 μl blood was collected from the beating heart using a syringe infiltrated with heparin in advance. Subsequently, 1 μl blood was mixed with 99 μl phosphate-buffered saline (PBS) in 1.5 ml EP tube. Then 10 μl diluted blood was added into a hemocytometer for counting the erythrocyte number. The erythrocytes in 10 chambers (1mm×1 mm) were counted under an inverted microscope (BX53, Olympus) for each zebrafish. Each zebrafish was counted three times in a randomly selected different field. Simultaneously, the blood cell pictures were photographed for reference.

Total RNAs were extracted from embryos or kidneys using the TRIzol reagent (Invitrogen), and the first-strand cDNA synthesis kit (Fermentas) was used to synthesize cDNA. qRT-PCR assays were performed using MonAmp™ SYBR^® Green qPCR Mix (high Rox) (Monad Bio.). The primers used for RT-PCR were: hif3a, 5′-GCTGGATGGCTTGTCTGATGG-3′ and 5′-CCCTCATGAGAGCTGCTGTG-3′; gata-1, 5′-GAGACTGACCTACTGCCATCG-3′ and 5′-TCCCAGAATTGACTGAGATGAG-3′; alas2, 5′-GCAAAATGGCCTTCTCCCTC-3′ and 5′-TCAAACCTGAGGTGTCTTGG-3′; band3, 5′-GTGATGGTTGGTGTCTCAAT-3′ and 5′-TAGTTGGCACACGGGTGACA-3′; hbae1, 5′-CTCTCTCCAGGATGTTGATT-3′ and 5′-GGGACAGAATCTTGAAATTG-3′; hbae3, 5′-CTCTTTCCAGGACTTTGTTC-3′ and 5′-GGTTGATGATCTTGAAGTTT-3′; hbbe1, 5′-ATGGTTGCTGCCCACGGTAA-3′ and 5′-CAGCCAAAAGCCTGAAGTTG-3′; β-actin, 5′-TACAATGAGCTCCGTGTTGC-3′ and 5′-ACATACAATGGCAGGGGTGTT-3′; runx, 5′-GGGACGCCAAATACGAACCT-3′ and 5′-GCAGGACGGAGCAGAGGAAG-3′; c-myb, 5′-AGTTACTTCCGGGAAGAACCG-3′ and 5′-AGAGCAAGTGGAAATGGCACC-3′; epo, 5′-GTGCCTCTCACTGAGTTCTTGGAAG-3′ and 5′-CTCGTTCAGCATGTGTAAGCCTGAC-3′. β-actin was used as internal controls. Applied Biosystems Step One was used for data collection.

We measured zebrafish oxygen consumption in 250 ml flasks (n=12), each containing 250 ml water. The oxygen concentration in water was measured using an LDO101 probe (HQ30d, HACH). A total of 12 adult zebrafish with similar weight (n=6 for hif3a-null and WT) were used for measurement. We placed each hif3a-null or WT zebrafish in an individual flask, and then tightly sealed the flasks with plastic film. After 4 h, we measured the oxygen concentration in each flask (n=6) with the LDO101 probe. After 8 h, we measured the oxygen concentration in the remaining flasks (n=6) using the LDO101 probe.

We performed ChIP assays using an Enzymatic Chromatin IP Kit (9002s) (Cell Signaling Technology) following the manufacturer's protocol. Briefly, 2000 embryos were harvested at 16 hpf and sonicated. Then, the protein A/G agarose beads (30 μl) (Santa Cruz Biotechnology) were added to each sample and the mixtures were rotated at 4°C for 1 h. Subsequently, the supernatants were incubated with anti-Hif3α antibody (provided by Dr Cunming Duan, University of Michigan, USA) (Zhang et al., 2012) or rabbit IgG (control) (Santa Cruz Biotechnology) and rotated at 4°C overnight. The primers for amplifying the promoter region of gata1 were: 5′-GTCTATAAGGTCATATAGGC-3′and 5′-CTTCAGTCTTTGGGAACTAG-3′. The primers for amplifying β-actin were: 5′-ATCATGTTCGAGACCTTCAA-3′ and 5′-TAGCTCTTCTCCAGGGAGGA-3′.

We used Image-Pro Plus software to analyze digital images for counting erythrocyte numbers and quantifying RNA levels. For counting erythrocyte numbers: briefly, a standard color parameter of one cell was set and the rectangular area of interest was used to select region, then, the information object definition parameter was chosen for measuring the signal. For quantifying RNA levels of in situ hybridization staining, the RNA signal measured from in situ hybridization staining of one control zebrafish was set as ‘1’ initially, the RNA levels in other zebrafish were calculated after being compared with the signal value of control zebrafish.

The zebrafish larvae (2 dpf) were placed in 10 μl 1×PBS dropped on glass slides. The blood cells were released by puncturing the pericardial sac and upper yolk sac of embryos with fine forceps, and then mixed with 90 μl PBS in 1.5 ml EP tube. We added 10 μl diluted blood into a hemocytometer for counting the erythrocyte number. The erythrocytes in four chambers (1 mm×1 mm) were counted under an inverted microscope (BX53, Olympus) for each zebrafish (n=5 larvae). Each zebrafish was counted three times in a randomly selected different field. Simultaneously, the blood cell pictures were photographed for reference.

GraphPad Prism 7 software was used for all statistical analysis. Differences between experimental and control groups were determined using unpaired two-tailed Student's t-test (where two groups of data were compared). P values less than 0.05 were considered statistically significant. For animal survival analysis, the Kaplan–Meier method was adopted to generate graphs, and the survival curves were analyzed by log-rank analysis.

We are grateful to Drs Cunming Duan and Ling Lu for providing anti-Hif3α antibody. We also thank Drs Peter Ratcliffe, Navdeep Chandel, Bo Zhang and Jingwei Xiong for the generous gift of reagents.

The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.185116.reviewer-comments.pdf