Bidirectional crosstalk between Hypoxia-Inducible Factor and glucocorticoid signalling in zebrafish larvae.

In the last decades in vitro studies highlighted the potential for crosstalk between Hypoxia-Inducible Factor-(HIF) and glucocorticoid-(GC) signalling pathways. However, how this interplay precisely occurs in vivo is still debated. Here, we use zebrafish larvae (Danio rerio) to elucidate how and to what degree hypoxic signalling affects the endogenous glucocorticoid pathway and vice versa, in vivo. Firstly, our results demonstrate that in the presence of upregulated HIF signalling, both glucocorticoid receptor (Gr) responsiveness and endogenous cortisol levels are repressed in 5 days post fertilisation larvae. In addition, despite HIF activity being low at normoxia, our data show that it already impedes both glucocorticoid activity and levels. Secondly, we further analysed the in vivo contribution of glucocorticoids to HIF activity. Interestingly, our results show that both glucocorticoid receptor (GR) and mineralocorticoid receptor (MR) play a key role in enhancing it. Finally, we found indications that glucocorticoids promote HIF signalling via multiple routes. Cumulatively, our findings allowed us to suggest a model for how this crosstalk occurs in vivo.

Introduction Glucocorticoids (GC) constitute a well-characterized class of lipophilic steroid hormones produced by the adrenal glands in humans and by the interrenal tissue in teleosts. The circadian production of glucocorticoids in teleosts is regulated by the hypothalamus-pituitary-interrenal (HPI) axis, which is the equivalent of the mammalian hypothalamus-pituitary-adrenal (HPA) axis. Both are central to stress adaptation [1][2][3][4]. Interestingly, both in humans and teleosts cortisol is the primary glucocorticoid and regulates a plethora of physiological processes including glucose homeostasis, inflammation, intermediary metabolism and stress response [5]. In particular, cortisol can exert these functions via direct binding both to the glucocorticoid receptor (Gr) and to the mineralocorticoid receptor (Mr), which bind cortisol with different affinities. [3,6]. Together they act as a transcription factor, which can function either in a genomic or in non-genomic way [5,[7][8][9].
Hypoxia-inducible factor (HIF) transcription factors are key regulators of the cellular response to hypoxia, which coordinate a metabolic shift from aerobic to anaerobic metabolism in the presence of low oxygen availability in order to assure homeostasis [10]. In mammals there are at least three isoforms of HIF-α (HIF-1α, HIF-2α and HIF-3α) and two main isoforms of HIF-1β (ARNT1 and ARNT2). [11]. Interestingly, due to a genome duplication event, there are two paralogs for each of the three Hif-α isoforms (Hif-1αa, Hif-1αb, Hif-2αa, Hif-2αb, Hif-3αa and Hif-3αb) in zebrafish. Among these, Hif-1αb is thought to be the key zebrafish homologue in the hypoxic response [12]. With respect to HIF-1β (ARNT) paralogues, the expression of two genes encoding Arnt1 and Arnt2 proteins has been described in zebrafish [13][14][15][16].
Whilst ARNT is constitutively expressed, the cytoplasmic HIF-α subunits are primarily regulated post-translationally via the PHD3-VHL-E3-ubiquitin ligase protein degradation complex. This is believed to occur in order to allow a rapid response to decreasing oxygen levels [12,[17][18][19]. Indeed, hypoxia, is a common pathophysiological condition [20,21] to which cells must promptly respond in order to avert metabolic shutdown and subsequent death [12]. In the presence of normal oxygen levels, a set of prolyl hydroxylases (PHD1, 2 and 3) use the available molecular oxygen directly to hydroxylate HIF-α subunit. Hydroxylated HIF-α is then recognised by the Von Hippel Lindau (VHL) protein, which acts as the substrate recognition part of a E3-ubiquitin ligase complex. This leads to HIF-α proteasomal degradation to avoid HIF pathway activation under normoxic conditions. On the other hand, low O 2 levels impair the activity of the PHD enzymes leading to HIF-α stabilisation and subsequent translocation in the nucleus. Here, together with the HIF-β subunit, HIF-α forms a functional transcription complex, which drives the hypoxic response [22]. Although the HIF response is aimed to restore tissue oxygenation and perfusion, it can sometimes be maladaptive and can contribute to a variety of pathological conditions including inflammation, tissue ischemia, stroke and growth of solid tumours [23]. Finally, it is important to note for this study that HIF signalling is able to regulate its own activation via negative feedback, by inducing the expression of PHD genes, in particular prolyl hydroxylase 3 (PHD3) [24,25].
The presence of a crosstalk between glucocorticoids and hypoxia dependent signalling pathways has been reported in several in vitro studies [26][27][28][29][30]. Moreover, synthetic glucocorticoids (ie. betamethasone and dexamethasone), which are analogous to naturally occurring steroid hormones, have been extensively used for decades as anti-inflammatory drugs for treating pathological conditions which are linked to hypoxia (i.e. asthma, rheumatoid arthritis, ischemic injury, etc.) [31][32][33]. However, due to the presence of adverse effects [34] and glucocorticoid resistance [35,36], their use has been limited. Therefore, extending the research on how precisely this interplay occurs in vivo, may have a wide physiological significance in health and disease.
The first evidence of interaction between HIF and GR was provided by Kodama et al. 2003 [26], who discovered that ligand-dependent activation of glucocorticoid receptor enhances hypoxia-dependent gene expression and hypoxia response element (HRE) activity in HeLa cells. Leonard et al. 2005 [27] subsequently revealed that GR is transcriptionally upregulated by hypoxia in human renal proximal tubular epithelial cells. Furthermore, the hypoxic upregulation of GR was confirmed by Zhang et al 2015 [29]. In contrast, a dexamethasone-mediated inhibition of HIF-1α target genes expression in hypoxic HEPG2 cells was demonstrated by Wagner et al. 2008 [28]. In addition to that, they showed retention of HIF-1α in the cytoplasm, suggesting a blockage in nuclear import. Finally, Gaber et al., 2011 [37] indicated the presence of dexamethasone-induced suppression of HIF-1α protein expression, which resulted in reduced HIF-1 target gene expression.
From these in vitro results it has become clear that the HIF-GC crosstalk is complex and may depend on cell type. In the present study, we have used the zebrafish (Danio rerio) as an in vivo model organism to study how and to what degree hypoxic signalling affects the endogenous glucocorticoids' response and vice versa. The use of whole animals allows us to show how these signals interact at a more global level than in cell culture, where interactions between different tissues and cell types are not easily modelled. The zebrafish offers an excellent genetic vertebrate model system for endocrine studies, and similar to humans, they are diurnal and use cortisol as the main glucocorticoid hormone [38]. Importantly, unlike other teleosts, zebrafish have only a single glucocorticoid (zGr) and mineralocorticoid receptor (Mr) (zMr) isoform [3]. Moreover, zGr shares high structural and functional similarities to its human equivalent, making zebrafish a reliable model for studying glucocorticoids activity in vivo [39][40][41]. Additionally, zebrafish share all the components of the human HIF signalling pathway and it has been proved to be a very informative and genetically tractable organism for studying hypoxia and HIF pathway both in physiological and pathophysiological conditions [12,25,42].
In our previous work, we identified new activators of the HIF pathway, e.g. betamethasone, a synthetic glucocorticoid receptor agonist [43]. Counterintuitively, GR loss of function was shown by Facchinello and colleagues to hamper the transcriptional activity linked to immuneresponse (i.e of cytokines Il1β, Il8 and Il6 and of the metalloproteinase Mmp-13) [5]. Finally, glucocorticoid receptor has been also found to synergistically activate proinflammatory genes by interacting with other signalling pathways [40,[44][45][46].
In the present study, we utilised both a genetic and pharmacological approach to alter these two pathways during the first 120 hours post fertilisation of zebrafish embryos. In particular, we took advantage of two different mutant lines we have generated (hif1β sh544 (arnt1) and gr sh543 (nr3c1) respectively), coupled to an already existing vhl hu2117/+ ;phd3:eGFP i144/i144 hypoxia reporter line [25], to study the effect of HIF activity on GC signalling and vice-versa, via a "gain-of-function/loss-of-function" approach. Phenotypic and molecular analyses of these mutants have been accompanied by optical and fluorescence microscope imaging. Importantly, we not only confirm that betamethasone is able to increase the expression of phd3:eGFP, a marker of HIF activation in our zebrafish HIF-reporter line, but we also show that BME-driven HIF response requires Hif1β/Arnt1 action to occur.
Furthermore, our results also demonstrate that both Gr and Mr loss of function are able to partially rescue vhl phenotype, allowing us to confirm the importance of glucocorticoids in assuring high HIF signalling levels. This finding may have wider significance in health and disease, as so far it is proven difficult to downregulate HIF signalling.
Our results also demonstrate that in the presence of upregulated HIF pathway (by mutating vhl), both the glucocorticoid receptor activity and the endogenous cortisol levels are repressed in 5 dpf larvae, whereas when the HIF pathway is suppressed (by mutating hif1β) they are significantly increased. Finally, qPCR analysis on GC target genes, in situ hybridisation on the expression of steroidogenic genes and cortisol quantification on the aforementioned mutant lines confirmed our hypothesis.
Taken together, these results allow us to deepen the knowledge of how the crosstalk between HIF and glucocorticoid pathway occurs in vivo and to underscore a new model of interaction between these two major signalling pathways.

Generating arnt1 and arnt1;vhl knockout in zebrafish
To study the interplay between HIF and GC signalling in vivo, using a genetic approach, we required a Hif1β/Arnt1 mutant line (in a phd3:eGFP;vhl +/background) to enable the downregulation of the HIF pathway. Hif-1β (hypoxia-inducible factor 1 beta, Arnt1) is a basic helixloop-helix-PAS protein which translocates from the cytosol to the nucleus after ligand binding to Hif-α subunits, after the stabilization of the latter in the cytoplasm. It represents the most downstream protein in the HIF pathway and for this reason it is the most suitable target.
Furthermore, since homozygous vhl mutants are lethal by 8-10dpf [47], we analysed the efficacy of arnt1 mutation in rescuing vhl phenotype. To this end, we attempted to raise arnt1 -/-;vhl -/after day 5 post fertilization. Notably, double mutants were able to survive beyond 15 dpf, but failed to grow and thrive when compared to their wild-type siblings, which led us to euthanise them due to health concerns at 26 dpf (S1B Fig). Of note, arnt1 homozygotes, in a vhl /+ or wt background, were morphologically indistinct and adults were viable and fertile. In contrast, the previously published arnt2 -/zebrafish larvae were embryonic lethal around 216 hpf [14].

Arnt1 and Arnt2 are mutually required for HIF signalling in zebrafish
As arnt1;vhl double mutants still activate the phd3:eGFP HIF reporter, we examined the importance of Arnt2 isoform in the HIF pathway. Phenotypic analysis was carried out on 5 dpf Arnt2 CRISPANTs, created both in a vhl +/and arnt1 +/-;vhl +/background, according to the protocol of Wu et al., 2018 [48]. By analysing the expression of the phd3:eGFP transgene, we observed that arnt2 CRISPR injected vhl mutants were characterized by a significant downregulation of phd3:eGFP-related brightness at the level of the head (equals to 53%, P<0.0001), in the liver (equals to 54%, P<0.0001) and in the rest of the body (equals to 46%, P<0.0001), compared to uninjected vhl mutant larvae (Fig 1H' compared to 1H, white asterisks; Fig 1K).
Furthermore, when both arnt1 and arnt2 isoforms were simultaneously knocked-out (Fig  1I'), the downregulation was even stronger at the level of the head (equals to 74%, P<0.0001), the liver (equals to 86%, P<0.0001) and in the rest of the body (equals to 83%, P<0.0001) ( Fig  1I' compared to 1H; Fig 1K). Of note, phd3:eGFP-related brightness in these mutants was still slightly higher than wildtype, (not shown; these levels are undetectable). Overall, these data show that Arnt1, even if not fundamental for survival, is the main isoform in the zebrafish liver required for HIF signalling, whereas Arnt2 is more expressed in the developing central nervous system (CNS), as reported by Hill et., al 2009. Of note, since both isoforms can form a functional complex with Hif-α isoforms and appear to function in the same organs, this allows us to confirm that they have partially overlapping functions in vivo and to show that they synergistically contribute to the HIF response.

Modulation of HIF signalling affects GR signalling
To investigate the interaction between HIF and glucocorticoid signalling, we quantified the expression of four potential glucocorticoid target genes from mammalian studies (fkbp5, il6st, pck1 and lipca) both in a HIF upregulated (vhl -/-), and downregulated scenario (arnt1 -/-) via RTqPCR analysis on 5 dpf larvae. We confirmed that in zebrafish larvae, fkbp5 is the most sensitive and well-established readout of Gr activity [5,49,50], whilst the other aforementioned genes do not directly take part in the GC-GR negative feedback loop. Therefore, we focused this analysis on fkbp5.
To further examine the effect of HIF signalling on glucocorticoid responsiveness, we also performed betamethasone (BME) treatment [30 μM] on the aforementioned mutant lines, followed by RTqPCR analysis. Of note, BME was able to increase fkbp5 expression in vhl siblings and was only able to mildly do that in vhl mutants. Indeed, its induction levels appeared not only lower in BME treated vhl mutants (fold change = 2.1) than in BME treated siblings (fold change = 7, P = 0.0286), but also its expression was not significantly different from DMSO treated wild-types (Fig 2A). In contrast, when the HIF pathway was suppressed (arnt1 -/-), BME treatment was able to further upregulate the expression of fkbp5 (fold change = 107,5; P = 0.0031), compared to DMSO treated arnt1 mutants (Fig 2A').
Collectively, we speculate that the upregulated HIF levels are able to repress the glucocorticoid receptor activity and can blunt its responsiveness to an exogenous GR agonist (BME treatment). On the other hand, importantly, although HIF activity is expected to be low in wildtype larvae in a normoxic environment, its function is also detectable with respect to suppression of GR activity. Indeed, if arnt1 gene is knocked-out (arnt1 -/-) an increased GR sensitivity is observed (Fig 2A'). To further test whether this had repercussions on steroidogenesis and/or cortisol levels, we analysed them both in a HIF upregulated (vhl -/-) and downregulated scenario (arnt1 -/-).

HIF signalling acts as negative regulator of steroidogenesis
To investigate the relationship between HIF signalling and steroidogenesis, we initially performed in situ hybridization on larvae obtained from the arnt1 +/mutant line, using both proopiomelanocortin (pomca) and Cytochrome P450 family 17 polypeptide 2 (cyp17a2) as probes. Expression of pomca, at the level of the anterior part of the pituitary gland, is a well-established readout of GR function in zebrafish larvae. Pomca is negatively regulated by increased blood cortisol levels via GC-GR signalling, as part of the HPI axis feedback loop [4,51]. Previous comparison test ( � P < 0.05; �� P < 0.01; ��� P <0.001; ���� P < 0.0001). B-C'. Representative pictures of WISH performed on DMSO and BME [30 μM] treated arnt1 mutant line, at 5 dpf, using pomca as probe. arnt1 wt DMSO treated (n = 30/30 larvae) showed normal pomca expression; arnt1 wt BME treated (n = 29/30 larvae) showed downregulated pomca expression. In contrast, arnt1 -/-DMSO treated (n = 28/ 30) and arnt1 -/-BME treated (n = 30/30) larvae showed downregulated pomca expression. Chi-square test ( ���� P < 0.0001). Scale bar 50 μm. D-E'. Representative pictures of WISH performed on DMSO and BME [30 μM] treated vhl mutant line, at 5 dpf, using pomca as probe. DMSO treated vhl siblings (n = 26/28) showed normal pomca expression; BME treated vhl siblings (n = 28/30) showed downregulated pomca expression. In contrast, vhl -/-DMSO (n = 28/29) and BME (n = 28/28) treated larvae showed downregulated pomca expression. Chi-square test ( ���� P < 0.0001). Scale bar 50 μm. F-F'. Steroid quantification results showed a significantly reduced cortisol concentration (P value <0;0028) in vhl mutants (92.7 fg/larva, in triplicate), compared to vhl siblings (321 fg/larva, in triplicate) at 5 dpf (F). Moreover, a significantly increased cortisol concentration (P value <0;0001) was measured in arnt1 mutants (487.5 fg/larva, in triplicate), compared to arnt1 wild-types (325 fg/larva, in triplicate) at 5 dpf (F'); unpaired t-test ( �� P < 0.01; ��� P <0.001). G. DMSO: Speculative scheme of how the putative HIF-GC crosstalk occurs in wildtypes and how it is affected both in arnt1 -/and in vhl -/larvae at 5 dpf. In wildtype scenario HIF signalling helps the GC-GR negative feedback to protect the body from an uncontrolled stress response. In particular, we speculate that HIF transcriptional activity is able to inhibit pomca expression when cortisol levels arise over a certain threshold in order to maintain both HIF and GC basal levels. However, in arnt1 -/scenario, the HIF-mediated negative feedback is compromised by the lack of a functional Arnt1. This triggers an initial uncontrolled pomca expression, which increases cortisol levels and subsequently downregulate pomca expression itself. Vice versa, in vhl -/scenario, the HIF-mediated negative feedback can exert a stronger inhibition of pomca due to the presence of upregulated HIF signalling. This results both in downregulated cortisol levels and in a suppressed GR responsiveness. However, the presence of alternative mechanisms cannot be completely excluded (i.e HIF might interact more directly with GC/GR to impair its function). Finally, the combination high cortisol/low pomca is very rare and this combination may change over the course of development. G. BME: Speculative scheme of how the putative HIF-GC crosstalk occurs in wildtypes and how it is affected both in arnt1 -/and in vhl -/larvae at 5 dpf, after BME[30μM] treatment. In all the cases, because of betamethasone acts downstream of the HPI axis, by binding directly to Gr, it is able to upregulate glucocorticoid target genes expression. Consequently, since GC are able to stimulate HIF signalling, as expected, we observed an increase phd3:eGFP-related brightness both in wildtypes and in vhl -/-. However, the fact that we did not observed any HIF upregulation both in arnt1 -/and in arnt1 -/-;vhl -/-, highlighted the fact that the BMEinduced HIF signalling activation is an Arnt1 dependent mechanism.
https://doi.org/10.1371/journal.pgen.1008757.g002 work also suggested that HIF promotes POMC activity in the mouse hypothalamic region [52]. On the other hand, Cyp17a2 is an enzyme involved in steroid hormone biosynthesis at the level of the interrenal gland, which is activated upon ACTH stimulation [53][54][55].
We found that 5 dpf arnt1 -/larvae, which were characterized by an upregulated GC responsiveness, showed upregulated cyp17a2 expression (S3C-S3C' Fig) coupled to downregulated pomca (Fig 2C). As expected, arnt1 siblings showed normally expressed cyp17a2 (S3A-S3A' Fig) and pomca (Fig 2B), which were observed to be downregulated only as a consequence of BME treatment (S3B-S3B' and S2B' Figs). Therefore, we speculate that in the absence of arnt1 (HIF suppressed scenario), pomca downregulation is most likely to occur as a consequence of GC-GR induced negative feedback loop, triggered by putative high cortisol levels (Fig 2A and 2G, DMSO, arnt1 mutant).
We subsequently examined both pomca and cyp17a2 expression in the opposite -HIF upregulated-scenario, by performing WISH analysis on the vhl mutant line. Interestingly, 5 dpf vhl -/larvae, which were characterized by a downregulated GR activity, displayed downregulated cyp17a2 expression (S3G-S3G' Fig), coupled to downregulated pomca expression ( Fig  2E). On the other hand, vhl siblings showed normally expressed pomca (Fig 2D), which was observed to be downregulated after BME treatment, as expected (Fig 2D'). Consequently, we speculate that in the absence of vhl (HIF upregulated scenario), pomca downregulation is most likely to occur as a consequence of HIF-mediated downregulation of pomca expression. (Fig  2G, DMSO, vhl mutant).
Cumulatively, if this is true, we predicted to observe reduced levels of endogenous cortisol in vhl -/larvae and normal or even increased levels in arnt1 -/larvae at 5 dpf.
Taken together, these data confirmed our hypothesis and showed for the first time that HIF signalling can act as negative regulator both of GR transcriptional activity and of steroidogenesis. Indeed, if only GR transcriptional activity was blocked by HIF, cortisol levels would be expected to be high in vhl mutants. This is because by blocking GR (i.e as occurs in gr -/-), the GC-GR mediated negative feedback cannot occur, making larvae hypercortisolemic [3,5]. Interestingly, since vhl -/larvae are characterized both by downregulated cortisol levels and GR transcriptional activity, this strongly suggests that HIF signalling can act both at the hypothalamic level (to inhibit pomca expression) and intracellularly to block GR transcriptional activity itself.

Generating gr and gr;vhl knockout in zebrafish
Conversely, to investigate the role of glucocorticoids on the HIF response, we created a novel glucocorticoid receptor (gr, nr3c1) mutant line and we crossed it with the vhl hu2117/+ ;phd3: eGFP i144/i144 hypoxia reporter line (this line will be called gr +/-;vhl +/hereafter). We created this line because the existing gr s357 allele may still have some activity via non-genomic pathways or tethering, promoting HIF activation upon GC treatment [4,43,51]. Of note, gr mutants are hypercortisolemic [3,5]. This is due to the inability of glucocorticoids to bind to a functional receptor (GR). As a result, they fail to provide negative feedback and are not able to shut down GC biosynthesis [3,5]. We generated an 11 bp deletion at the level of gr exon 3, which is predicted to truncate the DNA binding domain, lacks the C-terminal ligand binding domain and is predicted to be a true null (Fig 3A). The homozygous gr/nr3c1 mutants, characterized during the first 5dpf, were morphologically similar to control siblings and adult fish were viable and fertile, as predicted [5].
To confirm loss-of-function, we initially subjected larvae to a visual background adaptation (VBA) test, as it is linked to impaired glucocorticoid biosynthesis and action [4,56]. Larvae derived from gr +/incross were VBA tested and sorted according to melanophore size at 5 dpf. PCR-based genotyping on negative VBA-response sorted samples revealed that most larvae were homozygous for the gr allele, whereas positive VBA-response samples were always gr siblings.
Furthermore, WISH analysis performed on 5 dpf DMSO and BME treated gr +/incross derived larvae using pomca as probe, showed the presence of upregulated pomca expression in DMSO treated gr -/at the level of the anterior part of the pituitary gland (Fig 3C), compared to wild-type siblings (Fig 3B). Of note, BME treatment was not able to downregulate pomca levels in gr -/- (Fig 3C'), as it occurs in BME treated siblings (Fig 3B') via GC-GR mediated negative feedback loop, due to the absence of a functional gr allele. Finally, the loss of function was also determined in 5 dpf gr mutants by the strong downregulation of fkbp5 mRNA levels quantified via RTqPCR, both in the presence (fold change = 0.01; P<0.0001) and in the absence of BME treatment (DMSO treated, fold change = 0.01; P<0.0001), compared to DMSO treated wildtypes (Fig 3D).

gr mutation partially rescues vhl phenotype
We next analyzed the effect of gr loss of function on vhl phenotype. Phenotypic analysis carried out on 5dpf larvae, derived from gr +/-;vhl +/incross, revealed that nr3c1 mutation was able to cause an efficient, but not complete rescue of vhl phenotype, in a way which resembled arnt1 mutation (Fig 3F' and 3G').
Rescue was also apparent by morphology. Indeed, even if gr -/-;vhl -/showed reduced yolk usage, they displayed a reduction in ectopic vessel formation at the level of the dorsal tailfin, no pericardial edema, and developed air-filled swim bladders (Fig 3G and 3E). Moreover, whilst vhl mutants are inevitably deceased by 10 dpf [47], we were able to raise all selected double mutants beyond 15 dpf, but then (similarly to arnt1 -/-;vhl -/-) they failed to grow and thrive when compared to their siblings. This led us to euthanise them due to health concerns at 21 dpf (S4B Fig). Together, these data indicate for the first time, in our in vivo animal model, that GR function is essential for HIF signalling in zebrafish larvae, particularly at the level of the head and the liver.

gr loss of function can further reduce HIF signaling in arnt1;vhl double mutants
The similarity of gr and arnt1 mutations could mean they work in a single linear "pathway". If true, mutation of both genes should not lead to a further attenuation of the reporter expression. To test this, we bred the gr mutant line with the arnt1;vhl double mutant line and we crossed gr +/-;arnt1 +/-;vhl +/triple carriers. Phenotypic analysis carried out on 5 dpf phd3:eGFP positive larvae (n = 488) showed a small class of larvae with an even more rescued phenotype and a stronger downregulation of phd3:eGFP related brightness compared both to arnt1 -/-; vhl -/- (Fig 4B-4B') and gr -/-;vhl -/double mutants (Fig 4C-4C'). Of note, 7 putative very weak GFP + larvae were selected and genotypic analysis confirmed that 5 out of 7 were indeed gr -/-; arnt1 -/-;vhl -/-. In particular, these triple mutants showed a 54% downregulation at the level of the head, a 71% downregulation in the liver and a 72% downregulation in the tail region, in terms of phd3:eGFP-related brightness compared to vhl -/- (Figs 4D-4D' and S5). Thus, these data suggest that glucocorticoids are likely to act on both Arnt1 and Arnt2 mediated HIF signalling pathway.
These data, in accordance with our hypothesis, suggest that in gr -/-;vhl -/mutants the upregulation of pomca, triggered by the absence of functional Gr (and of the GC-Gr mediated negative feedback), cannot be inhibited with the same efficiency by HIF activity at the hypothalamic level. In gr -/-;vhl -/mutants, we speculate that the upregulated endogenous cortisol interacts with Mr to stimulate the HIF pathway, resulting in a mildly upregulated phd3:EGFP expression, in-between the levels seen in vhl mutants and wild-type larvae (Fig 3J and 3I). To test this assumption, we set up to block the mr gene in a gr -/-;vhl -/background, in order to check the importance of Mr in the HIF signaling pathway.

Both Gr and Mr are directly required in the HIF signalling pathway
Cortisol has high affinity both for Gr and Mr and they have been recently shown to be differentially involved in the regulation of stress axis activation and function in zebrafish [3]. Therefore, we analysed the role of Mr on the HIF signaling pathway. To achieve this, we knockedout mr in gr +/-;vhl +/-;phd3:eGFP incross-derived embryos, using CRISPant technology [48,59]. Interestingly, phenotypic analysis performed on 5 dpf injected and uninjected larvae revealed that mr CRISPR injected vhl mutants were characterized by a significant downregulation of phd3:eGFP-related brightness at the level of the head (equals to 49%, P<0.0001), in the liver (equals to 56%, P<0.0001) and in the rest of the body (equals to 47%, P<0.0001), compared to vhl -/mutant uninjected larvae (Fig 6D compared to 6A). Moreover, when both gr and mr were knocked-out, the downregulation was even stronger at the level of the head (equals to 62%, P<0.0001), in the liver (equals to 77%, P<0.0001) and in the rest of the body (equals to 63%, P<0.0001) compared to vhl -/mutant uninjected larvae (Fig 6E compared to 6A). Of note, mr injection in vhl -/larvae was more efficient in downregulation of phd3:eGFP expression compared to uninjected gr -/-;vhl -/larvae at the level of the head (equals to 31%, P = 0.0087) (Fig 6D compared to 6B).
To test the reliability of CRISPant method, we chose to knock-out a gene (which was not involved in the HIF pathway) into vhl +/incross derived embryos, to test whether it was able to affect HIF signalling. Laminin, beta 1b (lamb1b), which codes for an extracellular matrix glycoprotein, was injected as CRISPR-injection control in vhl +/incross derived embryos at 1 cell stage. Genotypic analysis carried out on these larvae confirmed that these guides were effective. Finally, quantification of phd3:eGFP-related brightness performed on 5 dpf injected and uninjected vhl -/larvae, showed no significant differences between the two groups (S6A and S6C  Fig). Overall, these data corroborated the efficiency of the CRISPant method and, at the same time, confirmed that both glucocorticoid and mineralocorticoid receptor play a pivotal role in the HIF signalling in vivo.

Discussion
Both HIF and glucocorticoid mediated transcriptional responses play a pivotal role in tissue homeostasis, glucose metabolism and in the regulation of cellular responses to various forms of stress and inflammation [60][61][62]. Previous in vitro studies highlighted the potential for crosstalk between HIF and glucocorticoid pathways, however there are still conflicting data on how this interaction occurs in vivo and there is no information on Mr contribution to HIF signalling. In this regard, we have presented a novel in vivo study using zebrafish larvae, focusing on the crosstalk between these two pathways. In contrast to in vitro cell culture studies, a whole animal study allows us to consider the interactions that occur between various tissues and provide novel insights. To this end, we generated arnt1 and gr null mutants to downregulate HIF and GR signalling respectively, as a basis for a genetic analysis of this crosstalk.
As a prelude to this, we had to establish the relative importance of arnt1 and arnt2 in the overall HIF response. To achieve this, a discriminative test was devised to place them in a vhl mutant background, where HIF signaling is strongly upregulated [25,42]. Phenotypic analysis performed on 5 dpf arnt1 -/-;vhl -/larvae showed reduced phd3:eGFP related brightness, normal yolk usage, properly developed and air-filled swim bladder as well as by the absence of pericardial oedema and excessive caudal vasculature. However, beyond 5 days, these double mutants exhibited only partial recovery from the vhl phenotype. Indeed, they developed well till 15 dpf, but subsequently failed to grow and thrive when compared to their siblings. In addition, arnt1 homozygous mutants were found to be viable and fertile, in contrast to both homozygous vhl and arnt2 mutants, which are embryonic lethal by 8-10 dpf [14,47].
Even though Arnt1 is not fundamental for survival, we found that it is required in the liver and in organs outside the central nervous system for HIF−α function. Conversely, using CRIS-Pant technology [48,59], we established that Arnt2 is mainly required in the developing central nervous system (CNS), as also reported by Hill et al. in 2009 [14]. However, the similarities observed in terms of phd3:eGFP-induced brightness in both arnt1 -/-;vhl -/and arnt2 CRISPR injected vhl mutants, suggest there is no strong functional separation. Therefore, both Arnt2 and Arnt1 have partially overlapping functions in vivo and both contribute to the HIF response. , which showed to be upregulated after BME treatment (33/35 larvae). On the other hand, DMSO treated vhl -/showed upregulated ldha expression (32/35 larvae), which was further upregulated after BME treatment (34/35 larvae). (black arrowhead: head and liver) Chi-square test ( ���� P < 0.0001). Scale bar 200 μm. C-D'. Representative pictures of WISH performed on DMSO (C-D) and BME [30 μM] (C'-D') treated vhl +/incross derived larvae, at 5 dpf, using phd3 (egln3) as probe. As expected, vhl siblings DMSO treated (n = 30/30 larvae) showed basal phd3 expression, which was mildly increased after BME treatment (n = 27/30 larvae). Vhl -/-DMSO treated (n = 28/30 larvae) showed upregulated phd3 expression, which was further increased after BME treatment (n = 26/30 larvae). (black arrowhead: head and liver) Chi-square test ( ���� P < 0.0001). Scale bar 200 μm. The effect of HIF signalling on the glucocorticoid pathway We next investigated the effect of HIF signalling and glucocorticoid responsiveness, by performing RTqPCR analysis on 5 dpf larvae. Collectively, we show that strong activation of HIF signalling (in vhl -/-) is able to blunt glucocorticoid receptor transcriptional regulation as judged by fkbp5 expression, whereas arnt1 loss of function derepressed it. As our experiments are done at normal atmospheric oxygen levels, we conclude from the latter result that normoxic HIF activity nevertheless suffices to attenuate GR transcriptional regulation.
We checked whether HIF signalling affects steroidogenesis. To this end, we quantified the expression of steroidogenesis-related genes (pomca and cyp17a2) both in vhl -/and in arnt1 -/larvae, via whole-mount in situ hybridization. Surprisingly, both lines showed downregulation of pomca expression. However, arnt1 -/larvae showed upregulated cyp17a2 expression, whereas vhl -/larvae, were characterized by downregulated cyp17a2.
Considering our results with GR-target fkbp5 in these mutants, we assume that in an arnt1 knock-out scenario, pomca downregulation occurs as a consequence of the GC/GR-mediated negative feedback loop aimed to control cortisol biosynthesis. This is also consistent with a significant upregulated basal cortisol levels quantified in these mutants. Vice versa, when HIF signalling is upregulated (in vhl mutants) we speculate that pomca and cyp17a2 downregulation may occur via HIF-mediated activity, leading to the observed low cortisol levels coupled to suppressed GR activity.
Indeed, glucocorticoids regulate a plethora of physiological processes, act on nearly every tissue and organ in the body to maintain homeostasis and are characterized by a potent antiinflammatory and immunosuppressive actions. For these reasons, their secretion must be finely controlled by the HPA/I axis [63]. As previous work in our laboratory showed that glucocorticoids also act as HIF activators [25,43], we infer that HIF can in turn control GC levels by acting on pomca. This would enable HIF signalling not only to control its own levels, but also to assure homeostasis. Finally, since HIF signalling is a master regulator of cellular pro-inflammatory responses to hypoxia [64][65][66], which would counteract the anti-inflammatory glucocorticoid activity, we speculate that the simultaneous expression of both upregulated HIF and GC pathway would be detrimental to homeostasis.
Our data would also be in accordance with a previous study showing that hypoxia exposure resulted in downregulation of steroidogenic genes (StAR, cyp11c1, hmgcr, hsd17b2, cyp19a, cyp19b) in 72 hpf larvae, whereas zHIF-α loss of function triggered the upregulation specifically of StAR, cyp11b2 and cyp17a1 [67].
Cumulatively, if this is true, we predicted to observe reduced levels of endogenous glucocorticoids in vhl -/and normal or even increased levels in arnt1 -/-. Importantly, the fact that cortisol levels were lowered in vhl mutants and were upregulated in arnt1 mutants is consistent with our hypothesis.
As a consequence of the above considerations, the HIF-mediated pomca negative regulation seems to be a logic homeostatic interaction: Increased HIF reduces GR activity, which in turn should lead to less HIF signalling.

The effect of glucocorticoids on the HIF signaling pathway
To further investigate the role of glucocorticoids on the HIF signalling, we initially analyzed the effect of gr loss of function on vhl phenotype. Surprisingly, we observed that gr mutation was able to cause an efficient, but not complete rescue of the vhl phenotype. Notably, gr -/-;vhl -/survived much longer than vhl -/-(> = 21 dpf compared to max. 10 dpf), but then similar to arnt1 -/-;vhl -/-, they failed to grow and thrive when compared to siblings. Our previous work [43] established that activation of GR signalling negatively regulates VHL protein in human liver cells. Our current genetic analysis shows that in zebrafish larvae, there must be an additional point of interaction between these two pathways, as we observed further activation of our HIF reporter after GC treatment even in the absence of VHL. Cumulatively, we showed for the first time in an in vivo animal model that Gr is fundamental to allow high HIF signalling levels.
We next analysed the effect of betamethasone treatment in arnt1 -/-. Although BME activated the GR target fkbp5, as expected, it failed to activate HIF signaling [43]. This was unexpected and would be best explained by assuming that a Gr-BME complex would preferentially interact with a HIFα/ARNT1 complex but not a HIFα/ARNT2 complex. Whether this holds up in mammalian cells would be interesting to address.

Evaluation of mineralocorticoid receptor contribution to HIF signalling
Recent work published by Faught and Vijayan, 2018 showed that both Gr and Mr are involved in the regulation of zebrafish stress axis activation and function [3]. Nothing is known about mineralocorticoid receptor contribution to HIF signalling. Therefore, we tested the effect of mr knock-out in gr +/-;vhl +/-;phd3:eGFP incrossed derived embryos. Interestingly, in mr injectedvhl -/we observed a significant reduction of phd3:eGFP-related brightness, compared to uninjected vhl -/larvae. Moreover, a further reduction of phd3:eGFP expression was found at the level of the head in mr injectedvhl -/compared to gr -/-;vhl -/larvae. Finally, the additional removal of mr in a gr -/-;vhl -/background reduced the hypoxia reporter expression even further. Therefore, we were able to show that both the glucocorticoid receptor and mineralocorticoid receptor play a pivotal role in promoting HIF signaling in zebrafish. In contrast to mammals, teleosts lack aldosterone and cortisol is the primary glucocorticoid hormone that can interact both with Gr and Mr to assure a correct HPI axis activity [68,69]. Of note, Mr was shown to not have a role in rapid non-genomic behaviors that required HPI axis signaling in zebrafish [70]. However, our hypothesis is consistent with Faught and Vijayan, 2018 elegant work, showing that both Gr and Mr signalling are involved in the GC negative feedback regulation. Importantly, this outcome may have a wider significance in health and disease. This is because so far, HIF signalling, which plays a key role in tumour growth, is proven difficult to downregulate. In this regard, our study suggests that modulation of Gr and Mr might be a potential avenue. In conclusion, although Mr contribution to HIF response in other organisms remains unclear, our work suggests that research into its function is warranted.

Conclusion
Our present study stresses the importance of the glucocorticoid pathway in driving HIF signalling. In addition, we uncovered a negative regulatory role played by HIF in regulating both GR responsiveness and steroidogenesis as demonstrated via RTqPCR and steroid hormone quantification. We also identified a mineralocorticoid receptor contribution to HIF-GC crosstalk. Finally, we presented novel gr +/-;vhl +/-, arnt1 +/-;vhl +/and arnt1 +/-;gr +/-;vhl +/zebrafish mutant lines which helped to better understand how the interplay between HIF and glucocorticoids occurs in vivo. For these reasons, we believe that this work could pave the way for further in vivo analysis to precisely identify the extensive crosstalk behind these two major signalling pathways.

Ethics statement
Zebrafish (Danio rerio) lines were raised and maintained under standard conditions (14 hours of light and 10 hours of dark cycle, at 28˚C) in the Aquaria facility of the University of Sheffield. Zebrafish embryos used for experiments were reared in E3 medium (5 mM NaCl, 0.17 mM KCl, 0.33 mM MgCl2, 0.33 mM CaCl2, pH 7.2) with or without methylene blue (Sigma-Aldrich) and staged according to standard methods [71] for up to 5,2 days post fertilisation (dpf) in accordance with UK Home Office legislation. Our studies conform with the UK Home Office guidelines (ASPA), Licence No. PC9C3D4CB and PB2866ED0. Ethics approval was obtained from the University of Sheffield Ethics committee AWERB.

Generation of gr (nr3c1) and hif1β (arnt1) null zebrafish lines
Both nr3c1 mutant line (gr sh543/+ ) and arnt1 mutant line (hif1β sh544/+ ) were generated using the CRISPR/Cas9-based mutagenesis method. A gene-specific guide RNA (sgRNA) sequence was identified using the CHOPCHOP website [72,73]. To design both gr and arnt1 sgRNA, an 18 nucleotides sequence upstream to a selected PAM site (gr sh543 : CCAGCTGACGATGTGG CAG; hif1β sh544 : TCGGTGCTGGTGTTTCCAG) was inserted into a scaffold sequence [74], containing a promoter for the T7 Polymerase. The sgRNA was amplified via PCR, purified from agarose gel and in vitro transcribed using MEGAshortscript T7 kit (Ambion). 1 nl of CRISPR mixture containing 2,4 μg/μl of gRNA and 0.5 μl Cas9 protein (NEB) was injected in one-cell stage embryos and raised for 24 hours. Wild-type (wt), strain AB embryos were used to generate the gr mutant line, whereas vhl hu2117/+ ;phd3:eGFP i144/+ incross-derived embryos were used to create the hif1β mutant line. Efficiency was verified via whole-embryo PCR-based genotyping, by a diagnostic restriction digest. Injected embryos were raised to adulthood. Embryos collected from transmitting G0 founders crossed with WT(AB) fish were raised and genotyped to confirm germline transmission of the mutation (F1 generation). Heterozygous mutants, carrying the same mutation, were selected and crossed to obtain homozygous mutant embryos (F2 generation).

Embryos harvesting, drug treatment and fixation for WISH
Embryos intended for whole-mount in situ hybridisation were treated with 16,8 μl of 1-phenyl 2-thiourea (PTU, stock concentration 75mg/ml) diluted into 35 ml E3 medium to inhibit melanogenesis, according to Karlsson et al., 2001 [77]. GR agonist treatment was performed on batches of 15 embryos each, at 4 dpf, treated in 6-well plates, with 30 μM Betamethasone 17,21-dipropanoate (BME) and with 1% DMSO (Sigma-Aldrich), as control, for 24 hours [4]. Inside the 6-well plates, embryos were incubated in 3 ml total volume of E3 medium, without methylene blue. Afterwards, up to 30 embryos at 5 dpf were collected in 1,5 ml Eppendorf tubes and anaesthetized using Tricaine Solution (MS-222, Sigma Aldrich) prior to fixation in 1 ml 4% PFA solution overnight, at 4˚C. Embryos were then washed twice for 10 minutes in PBST and post-fixed in 1 ml 100% MeOH. Finally, samples were stored at -20˚C.

gr sh543 mutants sorting by visual background adaptation (VBA)
Visual background adaptation (VBA) is a glucocorticoid receptor-dependent neuroendocrine response which causes zebrafish melanocytes to shrink when exposed to bright illumination [78,79]. To identify gr sh543 mutants from siblings and to confirm the absence of a functional VBA response, 5dpf larvae were exposed to 30 minutes darkness and then transferred onto a white background under bright, whole-field illumination, using a 30W fluorescent lamp mounted 50 cm above the dish [80,81].

Cortisol extraction and quantification
Three biological replicates of 150 larvae at 5 dpf each of hif1β sh544 mutants, hif1β sh544 siblings, vhl hu2117 mutants and vhl hu2117 siblings, respectively, were used for steroid hormone extraction and quantification. vhl -/larvae were sorted among siblings at 4 dpf according to both their phenotype and phd3:eGFP-related brightness. Because of the lack of visible phenotype, arnt1 -/larvae where derived from arnt1 -/fish incrossed, whereas siblings were from arnt1 +/fish crossed with arnt1 +/+ ones. Cortisol quantification was carried out according to the protocol published by Eachus et al., 2017 [55], based on the use of an Acquity UPLC System (Waters, Milford, CT) coupled to a Xevo TQ-S tandem mass spectrometer (Waters).

RNA isolation, cDNA synthesis and qPCR analysis
Transcript abundance of target genes was measured by quantitative real-time PCR (RTqPCR). Three biological replicates of 10 larvae at 4 dpf each, were treated for 24 hours with 30 μM Betamethasone 17,21-dipropanoate and with 1% DMSO, used as control, prior to RNA isolation. Total RNA was extracted from pools of 10 larvae at 5dpf with TRIzol reagent (Invitrogen by Thermo Fisher Scientific, 15596026). RNA extracted was quantified using a Nanodrop ND-1000 spectrophotometer. cDNA was then synthesized from 1μg RNA template through reverse transcription using Protoscript II First Strand cDNA Synthesis Kit (New England Biolabs), as recommended by manufacturer's instructions. All RTqPCR reactions were performed in triplicate using TaqMan probes in combination with CFX96 Touch Real-Time PCR Detection System (BioRad), paired with CFX Maestro Analysis Software.
Expression levels for each gene were normalized to eef1a1 (Dr03432748_m1) and/or rps29 (Dr03152131_m1) and fold change values were generated relative to wild-type DMSO treated control levels, according to ΔΔCT method [82]. All data were expressed as fold change mean ± s.e.m and P � 0.05 was considered statistically significant.

Quantifying phd3:eGFP-related brightness
Images were acquired using Leica Application Suite version 4.9, which allowed the capture both of bright-field and GFP fluorescent images. To quantify the phd3:eGFP-related brightness of live embryos derived from each incrossed mutant line used in this project, Fiji (Image J) software v.2.0.0 was used. Images were converted into a grey scale 8-bit format and subsequently analysed by the software, by summing the grey values of all the pixels in the selected area, divided by the number of pixels. By default, since values equal to 0 are assigned to black and values equal to 255 to white, the quantified mean grey values are proportional to the intensity of the eGFP-related brightness expressed in the embryos. In particular, head, liver and tail (from the anus to the caudal peduncle) related brightness were selected and measured in all the mutant lines used in this study (S1D Fig). Genotyping post phenotypic analysis on phd3: eGFP sorted larvae confirmed the genotype-phenotype correlation.

Statistical analysis
GraphPad Prism version 8.0 for MacOS (GraphPad Software, La Jolla, California, USA, www. graphpad.com) was used to perform statistical analysis on all the samples analysed. Unpaired t tests were used to test for significant differences between two sample groups (i.e cortisol quantification). One-way ANOVA was used for assessing mean grey values data quantification, whereas two-way ANOVA was used to evaluate qPCR data. As post-hoc correction tests, Sidak's method for multiple comparisons was used on normally distributed populations following one-way ANOVA, while Dunnett's correction was used for comparing every mean to a control mean, on normally distributed populations following two-way ANOVA.