Oxygen-sensing mechanisms in development and tissue repair

Before the onset of vasculogenesis, normal embryonic tissues are under developmental hypoxia, with oxygen partial pressure (pO₂) averaging ∼23 mm Hg (∼3% of sea level air) (Fathollahipour et al., 2018). Although the development of the cardiovascular systems brings in additional oxygen, the upper limit of the embryonic oxygenation level is dictated by placental pO₂ at 60 mm Hg (7.9%), which is substantially lower than the 10-13.1% found in adult arterial blood (Rodesch et al., 1992). In addition, rapid tissue growth is energetically demanding and consumes oxygen at high rates. For these reasons, embryonic tissues remain more hypoxic throughout development relative to adult tissues (Rieger et al., 1997; Siggaard-Andersen and Huch, 1995).

Hypoxia in normal embryos serves as a physiological signal that regulates diverse developmental processes. A key mechanism mediating hypoxia signaling is oxygen-dependent degradation of HIFα proteins. Under normoxia, prolyl hydroxylase domain (PHD) proteins split O₂ to hydroxylate specific proline residues in hypoxia inducible factor (HIF) α proteins, labeling them for polyubiquitylation and proteasomal degradation (Fig. 1A). Under hypoxia, hydroxylation is blocked; HIFα proteins translocate into nuclei where they heterodimerize with HIFβ proteins to activate the expression of their target genes (Fig. 1B) (Bruick and McKnight, 2001; Epstein et al., 2001; Hu et al., 2006; Ivan et al., 2002; Maxwell et al., 1999; Wang et al., 1995). For their discovery of the oxygen-sensing mechanism, William Kaelin Jr, Sir Peter Ratcliffe and Gregg Semenza were awarded the 2019 Nobel Prize in Physiology or Medicine (NobelPrize.Org, 2019).

In this Primer, we provide an overview of molecular and cellular mechanisms of oxygen sensing and hypoxia signaling, and summarize loss-of-function phenotypes in development, mostly in mice. We also highlight clinical relevance, focusing on angiogenesis and tissue ischemia (lack of perfusion). A discussion of other pathological processes in which hypoxia is also a component, such as tumorigenesis, diabetes and inflammation, is beyond the scope of this Primer. However, interested readers may consult several recent reviews on these subjects (Choueiri and Kaelin, 2020; Hayes et al., 2020; Semenza, 2019; Stothers et al., 2018).

The first PHD family member to be discovered was the EGLN1 protein encoded by the egg laying 9 (egl-9) gene in C. elegans (Epstein et al., 2001). In mammals, the PHD family has three members: EGLN1 (PHD2), EGLN2 (PHD1) and EGLN3 (PHD3, also known as SM-20 in rats) (Bruick and McKnight, 2001; Epstein et al., 2001; Ivan et al., 2002). Structural features of PHD proteins are shown schematically in Fig. 2A. Notably, PHD3 has a shortened sequence, containing only the catalytic domain, whereas PHD2 is unique in that it contains a MYND domain, which mediates specific protein interaction (Barth et al., 2009; Shao et al., 2009; Song et al., 2013). PHD1, PHD2 and PHD3 are all soluble enzymes: PHD1 is exclusively located in the nucleus; PHD2 is predominantly cytoplasmic, but also shuttles into the nucleus; PHD3 is evenly distributed between both the nucleus and the cytoplasm (Metzen et al., 2003; Steinhoff et al., 2009). Functionally, PHDs are iron (Fe²⁺) and 2-oxoglutarate (2-OG)-dependent dioxygenases, which split O₂ to hydroxylate their substrates (Epstein et al., 2001). In addition, PHDs have various oxygen-independent functions (Box 1) and substrates (Box 2).

Box 1. Oxygen-independent functions of PHDs

Besides hydroxylating HIFα and non-HIFα substrates, there have been several reports that PHD1 or PHD3 also stabilize certain proteins by hydroxylase-independent mechanisms. For example, PHD3 stabilizes occludin and inhibitor of nuclear factor kappa-B kinase subunit γ (IKKγ) by physically blocking ubiquitylation of these proteins (Chen et al., 2015; Fu and Taubman, 2013). One of these proteins, occluding (Ocln), is developmentally important, with Ocln^−/− mice exhibiting growth retardation (Saitou et al., 2000).

Fig. 2.

View largeDownload slide

Regulation of prolyl hydroxylase domain protein (PHD) expression, stability and activity. (A) Linear structure of prolyl hydroxylases. MYND in PHD2 interacts with p23 and FKBP38; EF hand (EF) domain is believed to bind to Ca²⁺. (B) Induction of PHD expression and degradation of PHD1 and PHD3 under hypoxia. PHD2 and PHD3 are more strongly expressed in response to hypoxia than PHD1 (indicated by arrow thickness). Siah1a/2 (a ubiquitin ligase) polyubiquitylates PHD1 and PHD3 (but not PHD2) under hypoxia. PHD1 and PHD3 are then degraded in proteasomes. In hypoxia, oxygen-dependent prolyl hydroxylation is inefficient, but is not completely blocked (blue box), leading to the slow degradation of HIF1α and HIF2α (indicated as HIFα). (C) Regulation of PHD activities by modulating co-factor availability and interaction. Reduced redox state (e⁻) may facilitate PHD activation by promoting the regeneration of Fe²⁺ from Fe³⁺, whereas reactive oxygen species (ROS) do the reverse. Several Krebs cycle intermediates (fumerate, succinate and isocitrate – not included in the figure) are structural analogs of 2-OG and competitively inhibit PHDs. (D) A complex containing HSP90, HIF1α and co-chaperonin proteins (p23 and possibly also FKBP38) can interact with the MYND domain in PHD2. As a result, PHD2 is brought into close proximity of HIF1α, enhancing prolyl hydroxylation efficiency of HIF1α and leading to its degradation. aa, amino acids; P4H-TM, prolyl 4-hydroxylase transmembrane; TM, transmembrane domain.

Fig. 2.

View largeDownload slide

Regulation of prolyl hydroxylase domain protein (PHD) expression, stability and activity. (A) Linear structure of prolyl hydroxylases. MYND in PHD2 interacts with p23 and FKBP38; EF hand (EF) domain is believed to bind to Ca²⁺. (B) Induction of PHD expression and degradation of PHD1 and PHD3 under hypoxia. PHD2 and PHD3 are more strongly expressed in response to hypoxia than PHD1 (indicated by arrow thickness). Siah1a/2 (a ubiquitin ligase) polyubiquitylates PHD1 and PHD3 (but not PHD2) under hypoxia. PHD1 and PHD3 are then degraded in proteasomes. In hypoxia, oxygen-dependent prolyl hydroxylation is inefficient, but is not completely blocked (blue box), leading to the slow degradation of HIF1α and HIF2α (indicated as HIFα). (C) Regulation of PHD activities by modulating co-factor availability and interaction. Reduced redox state (e⁻) may facilitate PHD activation by promoting the regeneration of Fe²⁺ from Fe³⁺, whereas reactive oxygen species (ROS) do the reverse. Several Krebs cycle intermediates (fumerate, succinate and isocitrate – not included in the figure) are structural analogs of 2-OG and competitively inhibit PHDs. (D) A complex containing HSP90, HIF1α and co-chaperonin proteins (p23 and possibly also FKBP38) can interact with the MYND domain in PHD2. As a result, PHD2 is brought into close proximity of HIF1α, enhancing prolyl hydroxylation efficiency of HIF1α and leading to its degradation. aa, amino acids; P4H-TM, prolyl 4-hydroxylase transmembrane; TM, transmembrane domain.

Prolyl 4-hydroxylase transmembrane (P4H-TM, PHD4 or EGLN4) is anchored in the endoplasmic reticulum (ER) membrane with its catalytic domain protruding into the ER lumen, and has an EF domain (Fig. 2A), which may potentially bind Ca²⁺ (Koivunen et al., 2007; Myllykoski et al., 2021). The P4H-TM protein sequence is more closely related to collagen hydroxylase than to PHDs, except that the substrate-binding region is more conserved with PHDs (Koivunen et al., 2007). Although P4H-TM has been shown to carry out prolyl hydroxylation in both HIF1α and HIF2α (Koivunen et al., 2007), it is unclear how it overcomes the topological barrier imposed by the ER membrane, or whether there are physiologically relevant levels of HIFα proteins within ER lumens. Finally, factor inhibiting HIF (FIH) is an oxygen-dependent asparagine hydroxylase but its domain organization has not been clearly defined (Koivunen et al., 2004; Mahon et al., 2001). FIH mostly hydroxylates a C-terminal asparaginyl residue on HIF1α and, to a much less extent, HIF2α (Koivunen et al., 2004).

Among the different PHD family members, PHD2 is most broadly expressed and plays a major role in oxygen sensing (Berra et al., 2003). The expression of PHDs, especially PHD2 and PHD3, is upregulated by HIF1α (Fig. 2B), presumably as a feedback mechanism to prevent uncontrolled HIFα accumulation under hypoxia (Bagnall et al., 2014; Fujita et al., 2012). On the other hand, the levels of PHD1 and PHD3, but not PHD2, are further fine-tuned by Siah1a/2-mediated ubiquitylation and proteasomal degradation under hypoxia (Fig. 2B) (Nakayama et al., 2004). It should be noted that in cells under physiological hypoxia, oxygen-dependent prolyl hydroxylation is still an ongoing process (Fig. 2B), although at a significantly slower pace than under normoxia. Once hydroxylated, degradation proceeds independently of oxygen concentration.

As mentioned above, the catalytic activity of PHDs depends on not only oxygen, but also their essential co-factors Fe²⁺ and 2-OG (Fig. 2C) (Epstein et al., 2001; Gerald et al., 2004; Hewitson et al., 2007). The availability of Fe²⁺ is significantly influenced by cellular redox states. Reactive oxygen species (ROS) lead to the oxidation of Fe²⁺ to Fe³⁺, the latter of which strongly suppresses PHD activities (Briggs et al., 2016; Epstein et al., 2001; Gerald et al., 2004). Several Krebs cycle intermediates, including fumarate, succinate and isocitrate, are structural analogs to 2-OG. In wild-type cells, these intermediates are mostly confined to mitochondria, but in cells with dysfunctional mitochondria, they can accumulate at high levels and leak into the cytoplasm to competitively inhibit PHDs (Hewitson et al., 2007).

Prolyl hydroxylation efficiency of HIF1α is enhanced by recruiting PHD2 to HSP90/HIF1α complex (Fig. 2D), via the interaction between MYND domain in PHD2 and co-chaperonin proteins of HSP90, including p23 and possibly also FKBP38 (Barth et al., 2009; Shao et al., 2009; Song et al., 2013). As a result, PHD2 is brought into close proximity of HIF1α. Furthermore, interaction between PHD2 and a protein called OS-9, also thought to work by bringing PHD2 closer to HIF1α, can also enhance hydroxylation, but specific domains involved in the interaction have not been characterized (Baek et al., 2005).

As depicted in Fig. 3, both HIFα and β subunits are members of the bHLH-PAS family of transcription factors. The N-terminal basic-helix-loop-helix (bHLH) motif (for DNA binding) and central PAS domain (mediating protein-protein interaction) are well conserved among different subunits, but the length and location of catalytic domains display substantial variability, except between HIF1α and HIF2α (Rieger et al., 1997; Tian et al., 1997; Wang et al., 1995).

HIF3α exists in three isoforms due to alternative splicing (Huang et al., 2013; Makino et al., 2001; Yamashita et al., 2008). HIF3α isoforms have very weak or no transcriptional activity, due to the lack of C-TAD or the entire catalytic domain, depending on the isoform (Huang et al., 2013; Makino et al., 2001; Yamashita et al., 2008). Therefore, HIF3α isoforms, especially inhibitory PAS domain protein (IPAS), play inhibitory roles by trapping HIF1β and competing for binding to hypoxia response elements (HREs) (Huang et al., 2013; Makino et al., 2001). In addition, IPAS also directly binds to HIF1α to inhibit its activity (Huang et al., 2013; Makino et al., 2001).

HIF1β, also known as ARNT or ARNT1 [aryl hydrocarbon receptor (AhR) nuclear translocator], heterodimerizes with HIF1α and HIF2α proteins to form HIF1 and HIF2 heterodimers, respectively (Motta et al., 2018; Wang et al., 1995; Wu et al., 2015). ARNT also heterodimerizes with AhR to mediate xenobiotic detoxification response, a HIF-independent function (Reisz-Porszasz et al., 1994).

ARNT2 is similar to ARNT in structure, although its expression pattern is much more restricted (Hirose et al., 1996; Keith et al., 2001; Mandl et al., 2016). Within the HIF family, studies on ARNT2 are mostly limited to its interaction with HIF1α (Hirose et al., 1996; Keith et al., 2001; Mandl et al., 2016). However, ARNT2 (but not ARNT) is known to heterodimerize with SIM1, an oxygen-independent member of the bHLH-PAS family of transcription factors involved in hypothalamus development (Michaud et al., 2000).

In mouse embryos, HIF1α is broadly expressed, whereas HIF2α has a more restricted expression pattern (Jain et al., 1998). HIF2α was first thought to be specifically expressed in the vascular endothelium (hence, also named endothelial PAS protein 1 or EPAS1) (Tian et al., 1997), but was later found to be expressed in additional cell types, such as kidney interstitial fibroblasts, neural crest cells and retinal astrocytes (Mowat et al., 2010; Niklasson et al., 2021; Paliege et al., 2010; Tian et al., 1998). In adult mice and cultured cells, the expression of both HIF1α and HIF2α is induced by growth factors and cytokines, independently of oxygenation conditions, but it is unclear whether similar induction also occurs in embryos (Dutta et al., 2008; Feldhoff et al., 2017; Wang et al., 2018).

HIF3α expression varies greatly among different isoforms. NEPAS expression is strong in multiple embryonic tissues but undetectable in adults (Yamashita et al., 2008). The expression of the prototypical HIF3α isoform is weak in embryos, but is moderately increased in adult tissues (Yamashita et al., 2008). IPAS is undetectable in embryos but is strongly expressed in Purkinje neurons and cornea epithelium in adult mice (Makino et al., 2001; Yamashita et al., 2008).

ARNT is broadly expressed in mouse embryos. In contrast, ARNT2 is strongly expressed only in developing neural and kidney tissues, but its expression is weak in most other embryonic tissues (Jain et al., 1998).

HIF1α and HIF2α are both hydroxylated by PHDs at two proline residues in the conserved LXXLAP motif, each located within a separate oxygen-dependent degradation (ODD) domain (Fig. 3) (Hirsila et al., 2003). Hydroxylation occurs efficiently under normoxia, and hydroxylated HIFα proteins are recruited into E3 ubiquitin ligase complexes by pVHL (von Hippel Lindau protein), polyubiquitylated, and then routed to proteasomes for degradation (Fig. 1A) (Epstein et al., 2001; Ivan et al., 2002; Maxwell et al., 1999; Min et al., 2002). The entire process is very rapid; the half-life of HIF1α protein is only 5-8 min in cells cultured under normoxia (Berra et al., 2001). In a hypoxic tissue environment, such as in developing embryos, the intracellular oxygen concentration is below K_m values (Hirsila et al., 2003), thus allowing much of HIFα proteins to escape hydroxylation and accumulate at relatively high levels.

In addition to oxygen-dependent degradation, HIFα stability is further fine-tuned by oxygen-independent mechanisms (Fig. 4A). HIF1α is methylated by lysine-specific methyltransferase SET7/9, but demethylated by LSD1, a lysine-specific demethylase (Kim et al., 2016; Liu et al., 2015). Lysine methylated HIF1α is subject to E3 ligase-dependent polyubiquitylation and proteasomal degradation (Kim et al., 2016). Oxygen-independent HIF1α ubiquitylation in hypoxia is suppressed by two mechanisms, including interaction with the molecular chaperone HSP90 (Liu et al., 2007) and phosphorylation by protein kinase A (PKA) (Bullen et al., 2016).

HIFα is also modified by acetylation and SUMOylation under hypoxia, leading to stabilization or degradation by the same modification in a context-dependent manner (Bilton et al., 2005; Carbia-Nagashima et al., 2007; Cheng et al., 2007; Jeong et al., 2002). In addition, HIF1α has been shown to be stabilized or destabilized by p53 in different studies (Madan et al., 2019; Ravi et al., 2000; Sano et al., 2007), but p53 itself is consistently stabilized by HIF1α (An et al., 1998; Chen et al., 2003). Although HIF2α also interacts with p53, the effect on HIF2α is unknown; however, HIF2α suppresses p53 expression and/or activity (Bertout et al., 2009; Das et al., 2012; Roberts et al., 2009).

Although the stability of HIFα is mostly regulated by prolyl hydroxylation, the assembly of HIFα into transcriptionally active HIF complexes is regulated by PHD-independent mechanisms (e.g. Fig. 4B). Specifically, HIF1α phosphorylation by p42/p44 MAPK indirectly promotes heterodimerization by promoting HIF1α nuclear localization (Mylonis et al., 2006). On the other hand, HIF3α, especially its IPAS isoform, inhibits HIF heterodimerization by trapping HIF1β (Makino et al., 2001; Yamashita et al., 2008). HIF activity also depends on the recruitment of p300/CBP, which is a major co-activator for HIFα (Ema et al., 1999; Kasper et al., 2005). Oxygen-dependent (but PHD-independent) HIF1α asparaginyl hydroxylation, mediated by FIH, inhibits the interaction between HIF1α and p300/CBP, thus suppressing HIF1α transcriptional activity (Koivunen et al., 2004; Mahon et al., 2001; Chen et al., 2014; Kasper et al., 2005; Lando et al., 2002).

HIFα prolyl hydroxylation by PHDs plays a major role in oxygen sensing. However, a variety of other regulatory mechanisms are also highly relevant. Both PHD expression and activity are subject to regulation. PHD genes themselves are HIF targets, a feedback loop exists to prevent excessive HIFα accumulation. The hydroxylase activity of PHDs is regulated by a variety of conditions that influence the intracellular levels and distribution of different co-factors of PHDs. At the functional level, PHDs not only hydroxylate HIFα proteins, but also several other proteins (Box 2). Conversely, HIFα proteins are not only hydroxylated by PHDs, but also by P4H-TM and FIH. Furthermore, HIFα stability and activity are also regulated by several mechanisms that are independent of oxygen and hydroxylases.

HIF1 and HIF2 activate the expression of their target genes by interacting with HRE or HRE-like elements located in the enhancer sequences (Coulet et al., 2003; Kappel et al., 1999). Some of the target genes are common to HIF1 and HIF2, such as VEGF-A and Glut1, but others are preferably activated by either of them (Chen et al., 2001; Downes et al., 2018; Forsythe et al., 1996; Hu et al., 2003). Genes encoding several glycolytic enzymes, such as 6-phosphofructo-2-kinase (PFKFB3) (Obach et al., 2004), phosphoglycerate kinase 1 (PGK1) (Li et al., 1996) and pyruvate dehydrogenase kinase 3 (PDK3) (Lu et al., 2008) are HIF1-specific target genes. Conversely, endothelial expression of several angiogenesis regulators, including eNOS (Coulet et al., 2003), Flk1 (VEGFR2) (Elvert et al., 2003; Kappel et al., 1999), Flt1 (VEGFR1) (Dutta et al., 2008) and Tie2 (also known as Tek) (Jiang et al., 2019), are upregulated by HIF2α. HIF2α also activates the expression of the pluripotency genes Nanog, Pou5f1 (encoding Oct4), Sox2 and Erk1 (also known as Mapk3) in stem or progenitor cells, such as neural stem cells, myoblast progenitors and mesenchymal stem cells (Gustafsson et al., 2005; Zhu et al., 2016).

In vivo regulation of erythropoietin (EPO) expression represents an interesting example where the expression patterns of HIF1α and HIF2α play a decisive role. EPO is strongly upregulated by HIF1α in vitro (Wang and Semenza, 1995). However, in vivo studies have demonstrated that EPO expression is dependent on HIF2α in mouse liver, kidney and cortical astrocytes, whereas HIF1α is dispensable (Chavez et al., 2006; Kapitsinou et al., 2010; Rankin et al., 2007). There is evidence that HIF2α, but not HIF1α, is expressed in the cortex interstitial cells, the main cell type responsible for EPO production (Paliege et al., 2010). However, HIF1α overexpression in brain tissues activates local EPO expression, conferring neural protection in the cerebral ischemia-reperfusion model (Li et al., 2020).

Interestingly, HIF2α promotes translation, in addition to transcription, under hypoxia (Uniacke et al., 2012). In hypoxic cells, the normal translational initiation mechanism is suppressed. However, HIF2α interacts with polysomes to facilitate the translation of mRNAs important for cellular functions under hypoxia (Uniacke et al., 2012).

Besides HIFβ and p300/CBP, HIFα proteins also directly bind to a number of other transcription factors. In a mouse myogenic cell line, HIF1α interacts with Notch intracellular domain (NICD), leading to the upregulation of NICD target genes that are important for the maintenance of an undifferentiated state (Gustafsson et al., 2005). HIF2α-Ets interaction enhances the expression of Flt1 and Flk1 in vitro and/or in mice (Dutta et al., 2008; Elvert et al., 2003). HIF1α stabilizes p53, activating the expression of p53-dependent genes (An et al., 1998; Carmeliet et al., 1998; Zhou et al., 2015). In Hif1α^−/− mouse embryonic stem cells, p53 protein levels are significantly reduced and apoptosis is blocked under hypoxia (Carmeliet et al., 1998).

HIF may regulate cell events in a cell-autonomous manner, in which HIF-activated gene products exert their functions in the same cells in which they are produced. An example of this type of regulation is metabolic reprogramming. HIF may also regulate cell functions in a non-autonomous manner. This may be best illustrated by VEGFA expression, which leads to endothelial cell differentiation, proliferation and migration. Here, we focus on cell-autonomous events.

Reprogramming of cellular metabolism by HIFα has been most extensively studied in cancer cells (reviewed by Elzakra and Kim, 2021; Semenza, 2011). More pertinent to this Primer is metabolic reprogramming in stem cells for tissue regeneration. In this regard, HIF1α promotes glycolysis in several types of stem cell, such as mouse embryonic stem cells (Carmeliet et al., 1998), neural stem cells (Lange et al., 2016), hematopoietic stem cells (Kocabas et al., 2015) and human mesenchymal stem cells (Palomaki et al., 2013). In diabetic mice, HIF1α overexpression confers protection against myocardial injury by promoting glycolysis and angiogenesis (Xue et al., 2010).

A major role of HIF2α in metabolism is to promote oxidative phosphorylation in mitochondria (Oktay et al., 2007; Scortegagna et al., 2003). Mechanistically, HIF2α activates the expression of SOD2 (mitochondrial superoxide dismutase) and frataxin, a chaperonin protein that enhances activity of aconitase in the Krebs cycle. Although HIF2α has not been found to upregulate glycolysis in most studies, one publication shows HIF2α-dependent upregulation of glycolysis during iPSC induction from human fibroblasts (Mathieu et al., 2014). At present, there is no explanation for this exception. However, we speculate that in most cells there might be a factor(s) that prevents HIF2α from interacting with the genes encoding glycolytic enzymes. Occasionally, the expression or function of such a factor(s) might be lost for different reasons, such as mutation or culture conditions, thus allowing HIF2α to activate the expression of glycolytic enzymes.

HIF1α and HIF2α also regulate various other cellular events. HIF1α promotes autophagy (Cosin-Roger et al., 2017), alveolar epithelial cell differentiation (Saini et al., 2008) and apoptosis in mice (Carmeliet et al., 1998), but suppresses ER stress in rat nucleus pulposus cells and proliferation of mouse embryonic mesenchymal cells (Carmeliet et al., 1998; Novais et al., 2021). HIF2α is required for neural crest cell migration in chick embryos (Niklasson et al., 2021), EPO synthesis in kidney interstitial fibroblasts in rats (Paliege et al., 2010), iron transport in intestinal epithelial cells in a humanized mouse model of sickle cell disease (Das et al., 2015), suppression of lipid accumulation in adipocytes (Feng et al., 2018) and production of catecholamine by the organ of Zuckerkandl in mouse embryos (Tian et al., 1998). These observations indicate that, despite biochemical similarities between HIF1α and HIF2α, their biological functions are often distinct. To some extent, the differential functions probably reflect differences in their expression profiles.

HIF1α and HIF2α share overlapping but distinct sets of target genes. Activation of target gene expression in vivo depends on not only molecular specificity of HIF1α and HIF2α, but also on their expression profiles. In addition, interaction between HIFα and other transcription factors may also influence the expression of target genes. HIFα target genes regulate various cellular functions, such as metabolism, autophagy and apoptosis.

Here, we discuss major phenotypes related to embryonic and neonatal development in mice. Adult phenotypes are also briefly mentioned if they share common elements with development, such as tissue growth and organization. A more complete list of phenotypes is included in Table 1. However, studies strictly focused on specialized topics, such as tumorigenesis and erythrocytosis, are not discussed; however, interested readers may consult several reviews on these subjects (Chan and Giaccia, 2010; Haase, 2017).

Phd2^−/− (Egln1^−/−) embryos die at mid-gestation stages, but Phd1^−/− (Egln2^−/−), Phd3^−/− (Egln3^−/−) and Phd1^−/−/Phd3^−/− are viable (Takeda et al., 2006). In Phd2^−/− mice, HIF1α and HIF2α protein levels are significantly elevated in both the placenta and embryo proper, except in the heart. However, the myocardium thickness and trabeculation are dramatically reduced in Phd2^−/− embryos, potentially secondary to poor nutrition resulting from defective placental development (Takeda et al., 2006). PHD2-deficient placentas display reduced labyrinthine branching morphogenesis, abnormal migration of spongiotrophoblasts and giant cells, and reduced angiogenesis. The placental defects in Phd2^−/− embryos partially resemble those in Vhl^−/− mouse embryos, which also display vascular deficiency in the poorly developed labyrinth (Gnarra et al., 1997). The exact reason for reduced angiogenesis is unknown, but poor labyrinthine development and cellular defects in the placenta could have some negative impacts.

PHD2 deficiency in retinal astrocytes decelerates their differentiation from proliferative immature states towards non-proliferative mature states, leading to prolonged expansion of the astrocyte population (Duan and Fong, 2019). Increased abundance of immature astrocytes blocks retinal vascular pruning in neonatal mice (Duan and Fong, 2019).

Neural-specific Phd2 deletion promotes brain vascularization due to HIF1α stabilization and VEGFA overexpression (Nasyrov et al., 2019). In kidneys, PHD2 and PHD3 double deficiency in stromal cells suppresses kidney development, manifested as reduced number of glomeruli and smaller kidney size (Kobayashi et al., 2017). In the lung, endothelial and hematopoietic PHD2 deficiency leads to obliterative lung vascular remodeling and pulmonary hypertension (Dai et al., 2016). Mechanistically, upregulated HIF2α in PHD2-deficient endothelium promotes CXCL12 expression, which in turn stimulates the proliferation of associated vascular smooth muscle cells (Dai et al., 2016).

FIH knockdown in zebrafish by morpholino oligonucleotides (MO) promotes angiogenesis of intersomitic vessels (So et al., 2014). However, Fih^−/− mice and zebrafish both develop normally. In adults, Fih^−/− mice display elevated metabolism, whereas Fih^−/− mice and zebrafish both have improved tolerance to hypoxia (Cai et al., 2018; Zhang et al., 2010). Discrepancies between MO and gene targeting studies are most likely due to MO off target effects (Kok et al., 2015).

Most mutant phenotypes discussed below are associated with loss-of-function mutations in mice, but some data are based on knock-in studies. A more complete list of phenotypes is presented in Table 2.

Consistent with its broad expression pattern and important roles in a variety of cellular processes, HIF1α is required in multiple embryonic tissues. Hif1a^−/− embryos are grossly abnormal, displaying a variety of developmental defects, including poor angiogenesis, failure of neural tube closure, branchial arch hypoplasia, pericardial effusion, brain tissue deficiency and cardiac hyperplasia. Hif1a^−/− embryos suffer from growth retardation and die in utero between E9.5 and E10.5 (Carmeliet et al., 1998; Iyer et al., 1998; Ryan et al., 1998).

Hif2a^−/− embryonic defects are milder, in part due to the more restricted expression pattern of HIF2α. In some background strains, Hif2a^−/− embryos exhibit vascular disorganization, hemorrhage (Peng et al., 2000), bradycardia and die in utero at mid-gestational stages (Peng et al., 2000; Tian et al., 1998). In other cases, Hif2a^−/− mice are born live, but exhibit growth retardation, reduced body weight and a high mortality rate at juvenile ages (Peng et al., 2000; Scortegagna et al., 2003). Mechanistically, SOD2 expression and aconitase activity are both reduced in Hif2a^−/− mice, leading to oxidative stress in the mitochondria and compromised activity of the Krebs cycle (Oktay et al., 2007; Scortegagna et al., 2003). Hif2a^−/− mice delivered by Cesarean section die within a few hours due to respiratory distress syndrome (RDS). The development of RDS is due to surfactant deficiency, which in turn arises from diminished VEGFA expression in epithelial cells during alveolarization (Compernolle et al., 2002).

HIF1α and HIF2α have both overlapping and unique roles in development. Hif1a^−/−/Hif2a^−/− embryos die before E8.5, substantially earlier than either Hif1a^−/− or Hif2a^−/− embryos (Cowden Dahl et al., 2005). In addition, Hif1a^−/−/Hif2a^−/− embryos completely lack placental vasculogenesis. Conversely, substituting HIF1α with HIF2α by HIF2α cDNA knock-in at the Hif1a locus leads to elevated Oct4 (encoded by Pou5f1) expression and embryonic death before E7.5 (Covello et al., 2006).

In Hif3a^−/− (also known as NEPAS^−/−) neonatal mice, HIF2α activity is upregulated, presumably due to the relief of HIF1β sequestration. Elevated HIF2α activity promotes the expression of its target gene endothelin 1 (Edn1), which leads to incomplete lung alveolarization (Yamashita et al., 2008).

Arnt^−/− (Hif1b^−/−) embryos essentially phenocopy the developmental defects of Hif1a^−/− mice and die by E10.5, which is slightly later than embryonic lethality at E9.5-E10.5 in Hif1a^−/− mice (Maltepe et al., 1997). The phenotypic similarity between Hif1a^−/− and Arnt^−/− embryos is consistent with HIF1α and ARNT functioning as a heterodimer. Arnt2^−/− mice die at perinatal stages due to hypothalamus defects, but Arnt^−/−/Arnt2^−/− embryos die before E8.5 (Keith et al., 2001). The earlier embryonic lethality in Arnt^−/−/Arnt2^−/− is suggestive of functional redundancy between ARNT and ARNT2.

Owing to embryonic lethality in germline knockout mice, tissue-specific approaches have been taken to determine HIFα functions. Cardiomyocyte-specific deletion of Hif1a in mice causes failure in heart-tube looping (Krishnan et al., 2008), while Hif1a knockout in the second heart field results in decreased sympathetic innervation in the heart (Bohuslavova et al., 2019). Specific loss of Hif-1a in neural cells or lung epithelium causes defective brain development (Tomita et al., 2003) or RDS-like phenotypes (Saini et al., 2015), respectively.

Neuro-specific disruption of HIF2α leads to altered expression of genes involved in neurogenesis. In these mice, migration of HIF2α-deficient neural stem cells is defective, leading to pyramidal neuron deficiency and cognitive abnormalities (Kleszka et al., 2020). Endothelial cell-specific HIF2α deficiency reduces Tie2 expression in airway endothelial cells (ECs) and hence causes apoptosis (Jiang et al., 2019). In the same mice, pericyte coverage is also reduced, presumably secondary to EC apoptosis. As pericytes normally provide a major source of angiopoietin 1, a Tie2 ligand, their deficiency further compromises Tie 2 signaling in airway ECs (Jiang et al., 2019). In neonatal retinas, astrocyte-specific Hif2a deficiency greatly accelerates the differentiation of proliferative astrocyte progenitors (proangiogenic) into non-proliferative mature astrocytes (non-angiogenic) (Duan et al., 2014b; Perelli et al., 2021). As a result, the pool of proangiogenic astrocyte progenitors is rapidly depleted and retinal angiogenesis is severely suppressed (Duan et al., 2014b; Perelli et al., 2021).

Phd1^−/− mice exhibit defective placentation and heart development, and die in mid-gestational stages, whereas Phd1^−/−, Phd3^−/− and Phd1^−/−/Phd3^−/− are viable. Cre-mediated Phd2 knockout has also revealed important roles in several neonatal tissues. HIF1α and HIF2α play both overlapping and unique roles in development. While Hif1a^−/− embryonic defects are more severe and widespread than in Hif2a^−/− mice, Hif1a^−/−/Hif2a^−/− embryos display the most severe defects, suggesting partially overlapping functions. On the other hand, HIF1α cannot be replaced by HIF2α, suggesting that at least some roles are unique. Tissue-specific gene-targeting studies also revealed developmental defects specifically associated with HIF1α or HIF2α deficiency. ARNT and ANRT2 also display partially overlapping roles, with ARNT being the major isoform required for development.

Among different human diseases, tissue ischemia shares many molecular and cellular features related to embryonic development. In both cases, cells are under hypoxia and their growth/repair depends on angiogenesis. As tissue repair is generally promoted by HIFα, but suppressed by HIFα deficiency, discussion in this section is focused on approaches that enhance HIFα activity, such as tissue-specific PHD knockout, delivery of PHD inhibitors and delivery of HIFα overexpression vectors. More information is included in Table 1 for PHD-related studies and in Table 2 for HIFα-related studies.

In several models, tissue-specific PHD2 deficiency is beneficial for the survival and repair of ischemic tissues. Cardiomyocyte-specific Phd2 disruption in mice protects mice against acute myocardial ischemic injury (Holscher et al., 2011). The specific mechanism underlying protection has not yet been investigated, but HIF1α protein level is increased in these mice, suggesting a role of the canonical PHD2-HIF1α axis in this model (Holscher et al., 2011).

Keratinocyte-specific PHD2 deficiency promotes dermal angiogenesis moderately, but loss of additional PHD proteins yields much more robust effects (Takeda et al., 2014). Molecularly, dermal levels of HIF1α and VEGFA are increased in a dose-dependent manner – most strongly upregulated in mice deficient in all three PHDs in their skin tissues (Takeda et al., 2014). In a wound-healing model, dermal PHD2 deletion promotes angiogenesis and wound healing (Zimmermann et al., 2014). In the mouse model of hindlimb ischemia, shRNA-mediated knockdown of all three PHDs promotes vascularization by HIF-dependent mechanisms (Loinard et al., 2009).

PHD deficiency may also confer protective effects by HIF-independent mechanisms. In Phd1^−/− mice, brain tissues are better protected against ischemia injury than wild-type mice, mostly due to NFĸB-dependent metabolic reprogramming and redox balancing (Quaegebeur et al., 2016). In the mouse model of hindlimb ischemia, Phd2 haploinsufficiency accelerates arteriogenesis through NFĸB-dependent macrophage repolarization (to a M2 phenotype), which is proangiogenic (Takeda et al., 2011). Molecularly, the partial loss of PHD2 promotes NFκB assembly by reducing prolyl hydroxylation of IKKβ, a non-HIF substrate protein (Cummins et al., 2006), promoting NFĸB assembly.

Interestingly, PHD2 deficiency in one tissue may also protect or accelerate repair in a remote tissue. PHD2 deficiency in keratinocytes or hepatocytes promotes arteriogenesis in ischemic hindlimb muscles (Takeda et al., 2014). Similarly, skeletal muscle-specific Phd2 disruption protects the heart against ischemia-reperfusion (I/R) injury (Olenchock et al., 2016). PHD2 deficiency in skeletal muscles leads to the accumulation of 2-OG, which enters the circulation and promotes the secretion of kynurenic acid (KYNA) from the liver.

Several human diseases are linked to abnormalities in oxygen sensing. For example, heterozygous loss-of-function mutations in PHD2 and gain-of-function mutations in HIF2A are both associated with erythrocytosis: a high concentration of red blood cells (Percy et al., 2012, 2006). In the VHL locus, thousands of heterozygous germline and somatic mutations have been identified, many of which underlie the development of erythrocytosis and/or a variety of malignant tumors (Kaelin, 2007).

In mice, similar pathological disorders may develop due to defective oxygen sensing. For example, Cre-dependent global Phd2 deletion in postnatal or adult mice leads to severe erythrocytosis, cardiac congestion and death within several months of Phd2 deletion (Minamishima et al., 2008; Takeda et al., 2008). Hepatocyte-specific deficiency of all three PHDs or VHL also leads to robust erythrocytosis and major defects in the liver, including severe steatosis (fatty liver) and tissue loss in PHD-deficient mice (Duan et al., 2014a), and vascular tumors in VHL deficient mice (Haase et al., 2001). These findings call for caution when targeting the oxygen-sensing mechanisms for clinical benefits.

The therapeutic potential of small molecule PHD hydroxylase inhibitors are being actively tested (Semenza, 2019). One study loaded a PHD hydroxylase inhibitor (1,4-dihydrophenonthrolin-4-one-3-carboxylic acid, ‘DPCA’) into a hydrogel nanofiber construct and then implanted it into wounded sites, where it upregulated HIF1α protein levels and improved tissue repair (Cheng et al., 2019). Some of the small molecule-based studies have advanced to early phase clinical trials; Daprodustat, another PHD hydroxylase inhibitor, has been tested in a Phase I trial. The inhibitor was found to be safe when applied topically, with very minimal systemic absorption (Olson et al., 2019). As systemic PHD deficiency could lead to undesirable consequences, being able to deliver PHD hydroxylase inhibitors locally is of practical importance.

Transgenic mice overexpressing HIF-1α in specific tissues display improved protection and repair of ischemic tissues, whereas tissue specific HIFα deficiency has the opposite effect. Cardiomyocyte-specific HIF1α overexpression prevents cardiac remodeling in streptozotocin-induced diabetic mice, in part by promoting the glycolytic pathway under diabetic conditions (Xue et al., 2010). In a rabbit model of corticosteroid-induced osteonecrosis, HIF1α overexpression in bone marrow cells promotes bone tissue regeneration, along with increased VEGFA expression and bone angiogenesis (Ding et al., 2013). Keratinocyte-specific HIF1α overexpression leads to hypervascularization of dermal tissues with non-leaky blood vessels (Elson et al., 2001). In contrast to overexpression, tissue-specific HIFα deficiency interferes with repair of ischemic tissues (Koeppen et al., 2018; Sarkar et al., 2012; Skuli et al., 2012).

In addition to transgenic overexpression, vector-based approaches have also been taken to achieve HIFα overexpression by localized delivery. In the mouse model of hindlimb ischemia, intramuscular injection of adenoviral vector overexpressing HIF-1α accelerated the recovery of hindlimb perfusion and function, in part by promoting the recruitment of circulating angiogenic cells (Ibuki et al., 2020). In a rat model of Alzheimer-like neurodegeneration, intracerebral injection of adeno-associated viral vector expressing HIF1α attenuated hippocampal neuronal apoptosis (Chai et al., 2014). In a rat stroke model, brain infarct volume was reduced by the transplantation of bone marrow-derived stem cells engineered to overexpress HIF1α (Lv et al., 2017). Finally, angiogenesis in wounded skin tissue is promoted by local injection of a plasmid vector overexpressing HIF2α (Chen et al., 2010).

Some of the tissue repair-related studies have moved on to clinical trials. For example, a Phase I trial has demonstrated the safety of intramuscular injection of an adenoviral vector expressing HIF1α lacking prolyl hydroxylation sites (Rajagopalan et al., 2007), although success in more advanced clinical trials remains to be demonstrated. Although not the focus of this Primer, HIF inhibitors are actively studied in both animal models and clinical trials, often for the purpose of suppressing neovascularization. Interested readers may consult a recent review focused on this important topic (Semenza, 2019).

In preclinical models for angiogenesis and tissue repair, HIFα, especially HIF1α, levels are boosted by different approaches, including Phd gene targeting, pharmacological inhibition of PHD activities, and HIFα overexpression. These studies overwhelmingly indicate that tissue-specific or localized upregulation HIFα levels is often beneficial for the protection and repair of ischemic tissues. On the other hand, systemic or excessive HIFα stabilization can be harmful, as shown in mice globally deficient for PHD2 or liver-specific loss of all three PHDs. Overall, pros and cons of targeting the oxygen-sensing mechanisms are context dependent, and each specific approach needs careful evaluation on a case by case basis.

Physiological hypoxia in embryonic tissues stabilizes HIFα proteins, allowing them to activate the transcription of developmentally important target genes. Embryonic phenotypes in Hif1a^−/− and Hif1b^−/− mice provide unequivocal evidence that HIF1 is broadly required in many cellular functions during embryogenesis. Although the role of HIF2α is more restricted during early embryogenesis, it is required in multiple processes in postnatal mice. PHD family knockout studies have provided evidence that oxygen-sensing mechanisms operate in vivo, essentially as predicted by in vitro molecular studies, involving both the canonical PHD-HIF axis and non-HIF PHD substrates. Furthermore, the embryonic lethality of Hif1a^−/−, Hif1b^−/−, Hif2a^−/− and Phd2^−/− mice suggests that embryogenesis requires an optimal level of HIFα activity. Embryonic development is perturbed by HIF levels above or below this level in either direction. On the other hand, adult mice are relatively tolerant to locally elevated HIFα levels in several types of tissues, such as myocardium, brain, skeletal muscles and dermal tissues, and can even benefit from such conditions during the repair process of ischemic tissues. Nonetheless, systemic PHD2 deficiency or tissue-specific deficiency of multiple PHD proteins may have deleterious consequences. Collectively, these studies have demonstrated the proof of principle that oxygen-sensing mechanisms may provide novel therapeutic targets to accelerate the repair of ischemia-injured tissues, whereas safety and efficacy-related issues need to be evaluated for each specific approach.