Evolution of a novel regulatory mechanism of hypoxia inducible factor in hypoxia-tolerant electric fishes

Abstract

Hypoxia is a significant source of metabolic stress that activates many cellular pathways involved in cellular differentiation, proliferation, and cell death. Hypoxia is also a major component in many human diseases and a known driver of many cancers. Despite the challenges posed by hypoxia, there are animals that display impressive capacity to withstand lethal levels of hypoxia for prolonged periods of time and thus offer a gateway to a more comprehensive understanding of the hypoxic response in vertebrates. The weakly electric fish genus Brachyhypopomus inhabits some of the most challenging aquatic ecosystems in the world, with some species experiencing seasonal anoxia, thus providing a unique system to study the cellular and molecular mechanisms of hypoxia tolerance. In this study, we use closely related species of Brachyhypopomus that display a range of hypoxia tolerances to probe for the underlying molecular mechanisms via hypoxia inducible factors (HIFs)—transcription factors known to coordinate the cellular response to hypoxia in vertebrates. We find that HIF1⍺ from hypoxia tolerant Brachyhypopomus species displays higher transactivation in response to hypoxia than that of intolerant species, when overexpressed in live cells. Moreover, we identified two SUMO-interacting motifs near the oxygen-dependent degradation and transactivation domains of the HIF1⍺ protein that appear to boost transactivation of HIF1, regardless of the genetic background. Together with computational analyses of selection, this shows that evolution of HIF1⍺ are likely to underlie adaptations to hypoxia tolerance in Brachyhypopomus electric fishes, with changes in two SUMO-interacting motifs facilitating the mechanism of this tolerance.

Oxygen is crucial for animal cells since it drives cellular metabolism and helps to generate the majority of cellular ATP through aerobic respiration. The lack of sufficient oxygen supply (hypoxia) triggers cells to switch to anaerobic means of metabolism, relying on glycolysis for energy production and reducing reliance on aerobic metabolic processes such as the tricarboxylic acid cycle and oxidative phosphorylation in the mitochondria (1). Fishes are an ideal system for investigating hypoxia, as many lineages inhabit environments that are repeatedly impacted by this stressor (2, 3, 4), but also many species are readily adapted to these harsh environments (5). Oxygen is far less abundant in water than air, and its diffusion rate in water is estimated to be 1/10,000 that of air (6). Consequently, aquatic organisms are far more vulnerable to hypoxia than terrestrial animals. Hypoxic conditions can be particularly challenging in tropical habitats like the Amazon, where high water temperatures reduce the dissolved oxygen content of water while increasing the metabolic rates of fish and other ectotherms. In many Amazonian habitats, dissolved oxygen also varies dramatically both seasonally and on a diel basis (7). Hypoxia has therefore likely shaped the evolutionary trajectory of Amazonian fishes. We are interested in the metabolic mechanisms underlying the adaptations to varying hypoxia levels, which allows certain species a high degree of metabolic plasticity and the capacity to occupy some habitats, such as highly-productive whitewater floodplains that are uninhabitable to hypoxia-intolerant species (8). During hypoxia, glycolytic enzymes are upregulated (9), while mitochondrial enzymes are downregulated (10), to reduce oxygen consumption while maintaining ATP supply through glycolysis. This cellular response occurs, at least in part, through hypoxia inducible factors (HIFs) (11). HIFs act as molecular oxygen sensors, regulating the expression of hypoxia response genes in response to fluctuations in oxygen levels (12). As such, they are central to the physiological and molecular processes underlying hypoxia tolerance.

HIFs coordinate the cellular response to hypoxia at the molecular level (12, 13). Working as heterodimer, HIF consists of a constitutively expressed beta subunit (named HIF1β or ARNT) and an oxygen sensitive alpha subunit (14). In mammals, there are three known HIF⍺: HIF1⍺, HIF2⍺, and HIF3⍺ genes (15, 16). The function of HIF1⍺ (17, 18, 19, 20) and, to a lesser extent, HIF2⍺ (21, 22, 23, 24) have been well-characterized. However, the roles of these transcription factors in hypoxia tolerance are poorly understood. In zebrafish, hypoxia tolerance is mediated largely by HIF1⍺, and loss of the genes encoding this factor leads to significant decrease in hypoxia tolerance (25). During normoxia, HIF1⍺ is hydroxylated at two residues (P402 and P564 in human) by prolylyl-hydroxylase domain proteins, which allows the von Hippel-Lindau tumor suppressor (VHL) E3 ubiquitin ligase to target it for proteasomal degradation (26, 27). Of these two proline residues, P564 is thought to be hydroxylated first, which then facilitates the hydroxylation of P402 (28). During hypoxia, the lack of oxygen molecules and Fe²⁺ co-factors for prolylyl-hydroxylase domain hydroxylation reduces HIF1⍺ proteolysis. However, hypoxia does not cause substantial increase in the expression levels of HIF1⍺ or HIF1β (29), suggesting that posttranslation modifications are the main regulatory mechanisms of the HIF heterodimer.

In addition to being a target of oxygen-dependent regulation, HIF1⍺ is also a target of other posttranslation modifications, which can further modulate its function. Hydroxylation of an asparagine residue near the C-terminal end (N803 in human) by factor inhibiting HIF1 gene diminishes its transactivation by weakening the interaction between HIF1⍺ and the transcription machinery (30). Acetylation of K532 is thought to enhance the interaction of HIF1⍺ with VHL, thereby increasing the rate of HIF1⍺ ubiquitination and degradation (31). Phosphorylation of HIF1⍺ by p42 and p44 mitogen-activated protein kinases has been shown to enhance its transcriptional activity (32). SUMOylation of HIF1⍺ by PIASy has also been reported (33), although it remains unclear if this modification can positively or negatively impact its stability or transactivation (34, 35). The effect of SUMOylation can either be direct—by affecting transcription factor function, blocking interaction with partners or hindering recognition of targets—or indirect, through interactions with other motifs within the same protein or on other proteins. This motif is called the SUMO-interaction motif (SIM). Traditionally, SUMOylation sites were thought to be located near ψ–K–X–E motif (with ψ being a hydrophobic amino acid) and SIM were believed to comprise a hydrophobic core of residues, followed by a highly acidic core of amino acids (36, 37). However, recent research has shown that only 40% of SUMOylation sites and SIMs adhere to this canonical consensus (38), leading to the development of new programs to facilitate the identification of novel SUMOylation and SIM residues.

The weakly electric fish genus Brachyhypopomus (Order: Gymnotiformes) contains 29 species of small (mostly <200 mm total length) species distributed from southern Costa Rica to Argentina (39). Some Amazonian species of Brachyhypopomus are regularly exposed to varying levels of seasonal and diel hypoxia (40). Like other electric fishes, Brachyhypopomus generate energetically expensive electric signals to navigate in poorly-lit environments (41, 42) and to communicate with congeners (43). They are nocturnal and are therefore most active when dissolved oxygen levels reach their lowest level in the diel cycle, as a result of aquatic plants and phytoplanktons switching from net oxygen production to net oxygen consumption. Hypoxia tolerance has consequently evolved on several occasions in this genus, in tandem with the occupation of permanently or intermittently hypoxic habitats (44). Whitewater floodplain inhabiting Brachyhypopomus species display physiological adaptations to survive prolonged hypoxia or complete anoxia (45), while other species occupy permanently normoxic habitats. This makes the genus an ideal model system to study molecular changes associated with hypoxia tolerance, without the phylogenetic noise that typically arise in the studies of distantly related species. While the ecological habitats, life histories, physiological adaptations, and phylogenetic interrelationships of Amazonian Brachyhypopomus species have been documented (40, 44, 45, 46, 47), the molecular mechanisms of variation in hypoxia tolerance in this group remain unknown.

Previous work has demonstrated that transactivation of HIF1⍺ is elevated in hypoxia-tolerant species (48) and proposed that transcriptional regulation of hypoxia-response element could be a mechanism of hypoxia tolerance in fishes. HIF1⍺ is therefore a promising candidate for studies of hypoxia tolerance in Brachyhypopomus. In this study, we focus on two congeners, B. bennetti and B. benjamini, which share 90 to 95% sequence identity at the phylogenetic marker loci cytochrome b and recombination activating gene 2, respectively. Despite such high genetic similarity, B. bennetti is restricted to macrophytes rafts in whitewater floodplains of the Amazon where dissolved oxygen (DO) ranges between 0 to 6.3 mgl⁻¹ year round but are severely hypoxic or completely anoxic (0–0.5 mgl⁻¹ at night) during the high water period when there is a very high biological oxygen demand from organic material in flooded forests. In contrast, B. benjamini is restricted to blackwater terra firme rainforest streams, where DO ranges are permanently normoxic: between 1.8 to 6.0 mgl⁻¹ at night (39, 40). The fish model organism, Danio rerio, is also most commonly found in normoxic habitats, between 3.92 to 8.92 mgl⁻¹ DO, although it is reported to be tolerant to very brief exposure to 0.8 mgl⁻¹ (49).

When experimentally challenged with declining oxygen concentrations, B. bennetti remained active in anoxia for the 6 h time course of the experiment (46), while other electric fish species from normoxic habitats quickly showed signs of agitation as soon as oxygen levels declined below approximately 0.5 mgl⁻¹. In addition, B. bennetti displays distinct morphological adaptations to anoxia, such as aerial gill respiration and larger gill lamellae sizes, compared to normoxia-restricted species like B. benjamini (45, 50). We henceforth make a distinction between ‘hypoxia tolerance’ in B. bennetti and ‘hypoxia intolerance’ in B. benjamini, although we also include other hypoxia tolerant species as B. beebei and B. walteri and intolerant species B. sullivani.

The early ancestors of Brachyhypopomus species experienced a whole-genome duplication event that occurred ∼300 mya at the base of the teleost lineage (51), as well as two earlier rounds of duplication events at the base of vertebrates (52). These ancestral genome duplication events provided genetic material for divergence and raw material for the evolution of novel biological functions (52), such as hypoxia tolerance. Recent studies have suggested the loss of one of each teleost-specific paralogous pairs of genes in most teleost species although both paralogs were retained in Cypriniformes (53), an order of fishes that contains some of the most hypoxia (and anoxia) tolerant species (6). A limitation of the studies of evolution of HIF⍺ genes in fishes in the context of hypoxia tolerance has been the limited scope of taxonomic sampling within fishes, which assumed that HIF⍺ evolved similarly across all teleost fishes. A recent examination of synteny in published genomes of fishes, spanning both teleost and non-teleost species, reached the conclusion that some fishes retain three HIF⍺ after the loss of at least one paralog of the teleost-specific duplicates (54). Exceptions include zebrafish, which appears to have retained six HIF⍺ genes, and Salmoniformes, which retain seven genes (54) as a result of an additional salmonid-specific whole genome duplication event (55, 56). Nonetheless, this study failed to explain how the patterns of evolution correlate with hypoxia tolerance.

To better understand the possible adaptive role of these transcription factors, particularly in hypoxia-tolerant and hypoxia-intolerant species, here we examine the molecular evolution of HIF⍺ genes in electric fishes. We focused on the hypoxia-tolerant species, B. bennetti, known to inhabit one of the most hypoxic ecosystems on earth and the hypoxia-intolerant congener, B. benjamini, which is exclusively found in permanently well-oxygenated waters. We specifically examined the molecular mechanisms of the differences of hypoxia tolerance between these species, by additionally including two species that are seasonally exposed to hypoxia (B. beebei and B. walteri), and one that occurs in perennially normoxic terra firme streams (B. sullivani). We find that transactivation of hypoxia-tolerant hif1ab is significantly higher than that of intolerant, but the steady-state levels, as proxy for protein stability of hif1ab alone does not explain the difference in transactivation, which suggests that amino acid differences in these proteins play a role in the elevated transactivation. Our study also shows that hypoxia-tolerant hif1ab causes higher aerobic and anaerobic metabolism in vitro, compared to WT human cells, but during chemically induced hypoxia, only aerobic metabolism is repressed, unlikeWT human cells where both systems are suppressed. Taken together, our study highlights the role of HIF1⍺ in hypoxia tolerance in the natural system and its role in regulating the balance between aerobic and anaerobic processes.

Results

Evidence of six HIF⍺ genes in some fishes but only three in other teleost fishes and tetrapods

We present the most comprehensive analysis of HIF⍺ genes in vertebrates, spanning tetrapods, nonteleost, and teleost fishes (Table S1). Our dataset is comprised of 675 sequences, including 592 amino acid sites: 565 parsimony informative sites and two invariant sites. Our analyses suggest that the two rounds of whole genome duplication events at the base of vertebrates likely generated four HIF⍺ genes (Fig. 1A). This is supported by the presence of four HIF⍺ genes in Holostei, the reedfish, and the Elephant shark, suggesting that these genes evolved prior to the teleost-specific whole genome duplication event. This was followed by a series of gene losses in different groups, including the loss of HIF4⍺ in tetrapods and most fish groups, but its retention in Otocephala, Salmoniformes, and Osteoglossiformes (Fig. 1A). We find weak support for the relationships between the HIF⍺ genes, illustrated by low bootstrap supports, largely due to the short branches at the base of each of the gene clades; however, we find strong bootstrap support for monophyly of each gene (Fig. 1A). Moreover, our results show that one of the teleost-specific genome duplication paralogs of HIF3⍺ is lost entirely in teleost fishes (Fig. 1A). For the other two genes, HIF1⍺ and HIF2⍺, one paralog of each gene appears to be lost in Euteleosts (hif1ab and epas1a), but both paralogs are retained in Otocephala and Osteoglossiformes.

Figure 1.

Molecular evolution of HIFα genes in vertebrates.A, maximum likelihood phylogeny of amino acid sequences of all vertebrate species and using invertebrate and placozoan species as outgroups. Only bootstrap values <100% are indicated. Inset, zoomed-in phylogeny showing Otocephalan hif1ab relationships and random-sites M8 analyses of different datasets (top right). Middle right, site-wise dN/dS estimates of the Brachyhypopomus dataset showing sites with positive selection, compared to that of Cypriniformes (bottom right). Domains are as follows: bHLH, basic helix-loop-helix; PAS1/2, Per-Arnt-Sim domains 1 and 2; PAC, PAS-associated COOH-terminal domain; ODDD, oxygen-dependent degradation domain; VHLRS, von Hippel-Lindau recognition site; NTAD, N-terminal transactivation domain; CTAD, C-terminal transactivation domain. B, branch-sites model analyses of the full dataset with Brachyhypopomus genes assigned different evolutionary rates from the rest of the dataset. Similarly, branch-sites analyses were performed with the Otocephalan hif1ab + epas1b and hif1al for reference. Gray bars indicate that the likelihood ratio test was not statistically significant. C, branch-sites analyses of teleost-fish–specific gene datasets to test the evolutionary rates of Brachyhypopomus, Cypriniformes, and Brachyhypopomus + Cypriniformes gene paralogs in comparison to other fishes in the background. Faded out bars indicate lack of statistical significance. For reference, gray dashed lines show dN/dS = 1. HIF, hypoxia inducible factor; SIM, SUMO-interaction motif.

Based on these findings, we suggest that within teleost fishes, teleost-specific whole genome duplication event resulted in four pairs of genes, followed by loss of a paralog of each gene and the complete loss of HIF4⍺ in most species. An exception to this scenario is Otocephala (inclusive of Ostariophysi and Clupeiformes in our dataset), where only a single paralog of each of HIF3⍺ and HIF4⍺ appears to be lost. We therefore identified (1): three HIF⍺ genes in tetrapods (2), three HIF⍺ genes in most teleost (Euteleostei) species (3), four genes in Holostei, and (4) six gene copies in Otocephala. In Salmoniformes, salmonid-specific whole genome duplication event resulted in at least seven HIF⍺ genes, as predicted by previous studies. We provide a summary table highlighting the naming conventions of these genes in fishes, and their homologs in mammals (Table 1).

Table 1.

Gene nomenclature in Danio rerio and their homologs in tetrapods

Gene name	Previously designated homolog	Proposed homolog
hif1aa	HIF1A	HIF1A
hif1ab	HIF1A	HIF1A
epas1a	HIF2A	HIF2A
epas1b	HIF2A	HIF2A
hif1al	HIF3A	HIF3A
hif1al2	HIF3A	HIF4A

Strong evidence of positive selection in hif1ab in Brachyhypopomus

Clade model analyses of the full dataset estimates the dN/dS rates for the divergent site class, largely comprised of functionally divergent codons between genes, suggesting purifying selection in all genes. All ω estimates for the vertebrate genes were above that of the invertebrate outgroups (Fig. S1). However, we found that the divergent site class of HIF2A has the lowest estimated ω of all genes suggesting it is the most conserved, followed by HIF1A and HIF3A, while HIF4A shows the highest ω, although all ω estimates fell well below 1 (Fig. S1 and Table S2). The elevated rates of evolution of HIF4A could be expected given how widely it has been lost in vertebrate species.

In our full dataset, branch-sites model analyses show significantly high evolutionary rates in Brachyhypopomus hif1ab clade, when assigned different rate from the remainder of the dataset (Fig. 1B, and Tables S3 and S4). The dN/dS of hif1ab were much higher than analyses where we labeled other Brachyhypopomus gene clades in the foreground (Fig. 1B). We tested this further by using the smaller datasets of teleost-specific genes, where we find positive selection in hif1ab, epas1a, and hif1al in Brachyhypopomus, while positive selection in only hif1ab in Cypriniformes and the combined Brachyhypopomus + Cypriniformes analysis (Fig. 1C and Tables S5-S8). Notably, the Brachyhypopomus hif1ab partition shows the highest dN/dS estimates of all analyses conducted.

We further confirmed the estimated positive selection in hif1ab using random sites model tests by curating a dataset of hif1ab only from the Otocephala, Ostariophysi, Cypriniformes, Gymnotiformes, and Brachyhypopomus groups. These analyses allow estimation of the evolutionary rates of these groups completely independently of each other, and our data show that Brachyhypopomus dataset displays the highest evolutionary rates of the groups tested, providing additional evidence of potential strong adaptive evolution of hif1ab in this group, even stronger than that of Cypriniformes—a group that includes strongly hypoxia-tolerant species. The site-wise ω estimates of the two hypoxia-adapted groups, Brachyhypopomus and Cypriniformes, show a different pattern of positively selected sites in these two groups (Fig. 1A inset; Tables S9-S13 and Fig. S2). These results are highly indicative of the different mechanisms of adaptation of these two groups.

When looking at expression levels of these genes in the teleost fishes of Gymnotiformes and Cypriniformes (representing two groups with known hypoxia-tolerant species), we find that HIF1⍺ and HIF3⍺ are the most highly expressed in most tissues and across developmental stages. For HIF3⍺ (hif1al gene in fishes), this is expected as the paralog of this gene seems to have been lost, but for HIF1⍺, it appears that hif1ab paralog, in particular, is more abundantly expressed than hif1aa (Fig. S4).

Hif1ab transactivation recapitulates organismal hypoxia tolerance in vitro

We used a luciferase reporter assay to measure the ratio of firefly luciferase, driven by PGK-1 hypoxia-response element, to constitutively expressed renilla luciferase to quantify HIF1⍺ transactivation (Fig. 2A). The transactivation of transiently expressed hif1ab from hypoxia-tolerant species, B. beebei, B. bennetti, and B. walteri, show a 1.5 to 2.7 times higher median relative luciferase signal than hypoxia-intolerant B. benjamini hif1ab and 2.9 to 5.1 times higher median luciferase signal than heterologous overexpressed human HIF1⍺, during hypoxia (Fig. 2B). When compared to the endogenous response, hypoxia-tolerant hif1ab median luciferase signal is 6.2 to 10.8 times higher during hypoxia, while only 4-times higher than hypoxia-intolerant hif1ab. Notably, hypoxia-tolerant hif1ab transactivation was also higher than that of intolerant species and human during normoxia, suggesting higher basal activity (Fig. 2B). The immunoblot shows more protein after 24 h of 1% O₂ than ambient conditions, as expected; however, the pattern of protein stability alone does not seem to explain the difference in transactivation (Fig. 2C). For example, B. bennetti and B. benjamini show relatively similar protein amount during normoxia and hypoxia; however, their transactivation is very different.

Figure 2.

HIF1⍺ of hypoxia tolerant species has higher transactivation capacity, despite high disorder in oxygen-dependent degradation domains.A, schematic of the luciferase assay approach to express extrinsic proteins and measurement of transactivation of HIF1⍺. B, transactivation of HIF1⍺ from hypoxia tolerant and intolerant electric fishes, in comparison to plasmid-expressed human control and endogenous (empty plasmid) response. Samples were grown in ambient (normoxia) oxygen levels or 1% O₂ (hypoxia) for 24 h prior to luciferase assay. Statistical significance between means of trials are shown: ns, p > 0.05; ∗p ≤ 0.05; ∗∗p ≤ 0.01; ∗∗∗p ≤ 0.001; ∗∗∗∗p ≤ 0.00001. C, immunoblot of 50 μg total protein from cells transiently expressing fish or control HIF1⍺, grown at ambient O₂ or 1% O₂ for 24 h before lysis. D, top, protein schematic showing the main domains of HIF1⍺ and the regulatory motifs in von Hippel-Lindau recognition sites and c-terminal transactivation domains; middle, amino acid conservation across all vertebrates colored according to domain; bottom, disorder probability as predicted by ODiNPred colored by respective domains. Domains are as follows: bHLH, basic helix-loop-helix; PAS1/2, Per-Arnt-Sim domains 1 and 2; PAC, PAS-associated COOH-terminal domain; ODD, oxygen-dependent degradation domain; NTAD, N-terminal transactivation domain; CTAD, C-terminal transactivation domain. E, immunoblot of 50 μg total protein from cells transiently expressing B. benjamini WT or mutant hif1ab grown at ambient, A, or 1% O₂, H, for 24 h before lysis. HIF, hypoxia inducible factor; SIM, SUMO-interaction motif.

Previous studies have postulated the order by which the VHL recognition sites are hydroxylated, and the P564 was determined to be hydroxylated before P402 (28). We tested this finding in our system, by mutating the homologous Pro residues in the hypoxia-intolerant B. benjamini hif1ab. Loss of either single hydroxylation residues do not result in a statistically significant increase in transactivation, while the double mutant does (Fig. 2D). Loss of P564 causes more equivocal protein levels between hypoxia and normoxia treatments in vitro, suggesting that P564 is more crucial for protein regulation, while the double mutant shows the highest protein stability (Fig. 2E). While the double mutant causes a significant increase in transactivation compared to WT B. benjamini trials, it does not fully account for the difference between hypoxia-tolerant and hypoxia-intolerant transactivation.

SIMs strongly impact transactivation independent of genetic background

SIM1 and 3 were identified in hypoxia-tolerant B. bennetti hif1ab, while SIM2 was identified in the hypoxia-intolerant B. benjamini sequence (Fig. 3A). Single SIM1, 2, and 3 mutants alone decrease transactivation of hif1ab, although only the SIM3 single mutant effect was statistically significant in hypoxia-tolerant B. bennetti (Fig. 3B). All double and triple mutants cause a significant reduction of hif1ab transactivation in the B. bennetti background, both in ambient O₂ and hypoxia conditions (Fig. 3B). The triple B. bennetti mutant shows median luciferase ratio during normoxia and hypoxia comparable to that of the hypoxia-intolerant B. benjamini hif1ab.

Figure 3.

SUMO-interacting motifs impact transactivation of HIF1a.A, schematic of HIF1a protein showing three different SUMO-interacting Motifs (SIM) identified in fish transcription factors but not in human; SIM1 and 3 in hypoxia tolerant species, SIM2 identified in intolerant species. Hypoxia tolerant and intolerant motif amino acid consensus is highlighted with sequence logo. B, luciferase reporter assay of plasmid-expressed mutants compared to WT. In hypoxia tolerant background, B. bennetti (left), hypoxia intolerant SIM motifs were introduced, while in hypoxia intolerant backgrounds, B. benjamini (top right) or H. sapiens (bottom right), hypoxia tolerant SIM modifications were introduced. Nucleotide substitutions of the mutants are summarized in Table S14. In (B), cells were grown in ambient O₂ condition (normoxia) or 1% O₂ (hypoxia) for 24 h before cell lysis and luciferase assay. Statistical significance between means of trials are shown: ns, p > 0.05; ∗p ≤ 0.05; ∗∗p ≤ 0.01; ∗∗∗p ≤ 0.001; ∗∗∗∗p ≤ 0.00001. HIF, hypoxia inducible factor; SIM, SUMO-interaction motif.

On the other hand, only SIM1 mutant in the B. benjamini hif1ab genetic background shows a statistically significant increase in transactivation (Fig. 3C), approaching levels observed in the hypoxia-tolerant B. bennetti hif1ab (Fig. S6). In the human genetic background, single and double mutants of SIM1 and SIM3 caused a statistically significant increase in transactivation (Fig. 3D). These enhanced rates, however, did not match the rates of the fish hif1ab, suggesting potential roles of epistatic interactions and other amino acid changes between the fish and the human genetic backgrounds, and their role in the adaptive evolution of hif1ab. The SIM mutations do not, however, seem to affect HIF1A protein levels in-vitro (Fig. S11).

We wondered if SIMs are likely to occur naturally in non-fish sequences, so we surveyed all genetic variants of human HIF1A on Ensembl for missense mutations around the SIM regions. Intriguingly, we identified a missense mutation at the second amino acid of SIM1; E365G (Fig. S7). This mutation is predicted to be deleterious and pathogenic, according to PolyPhen (57) and SIFT (58) scores (Fig. S7). We also identified A821T, a single-nucleotide polymorphism found in SIM3 and similarly expected to have pathogenic effects (Fig. S7). We experimentally tested these two mutants and found that E365G causes a significant increase in HIF1A transactivation in response to hypoxia and A821T causes an increase in HIF1A stability in response to hypoxia (Fig. S8). These results support our finding that SIMs enhance HIF1A function in vitro.

Stably expressed hypoxia-tolerant hif1ab shows higher transactivation than hypoxia-intolerant

We used CRISPR-Cas9 to knock-in fish hif1ab to knockout the endogenous human HIF1⍺ and generate cells stably expressing fish transcription factors. We used two sgRNA flanking the entire HIF1⍺ locus and supplied donor plasmid with hypoxia tolerant, intolerant, or human (control) HIF1⍺ and tagBFP marker for screening positive clones (Fig. 4A). The stably transfected cells reassert the pattern of transactivation we observe with the plasmid-expressed HIF1⍺; hypoxia-tolerant hif1ab shows higher transactivation than hypoxia intolerant or human control (Fig. 4B), although the magnitude of response is significantly lower than that of the plasmid-expressed response. The hypoxia-intolerant hif1ab transactivation, however, was not significantly higher than that of the human controls (Fig. 4B). We used confocal microscopy to confirm success of our genomic integration (Fig. 4C) by looking for expression of tagBFP marker in our donor cassettes. Finally, the immunoblots of the transgenic cell lines show an increase in protein levels in cells exposed to hypoxia compared to those grown at normoxia, indicative of proper protein degradation in the transgenic cell lines (Fig. 4D). We also note that in the trials using anti-HA mAb, we detect a higher protein band in hypoxia lanes, in addition to the expected ∼100 kDa band of HIF1⍺, which could be that of HIF1⍺-HIF1β dimer. We do not see this band with the anti-HIF1A pAb (Fig. 4D). These secondary bands are likely HIF1⍺-HIF1β dimers, which are likely not detected by the pAb.

Figure 4.

Knock-in to knockout approach to generate stably transfected fish transgenic cell lines.A, schematic showing the guide RNA cut sites at the HIF1A locus (ch14:61,695,401–61,748,259) and the design of donor plasmid for B. bennetti, B. benjamini, and human control. Small arrows show position of genotyping primers spanning the 3’ cut site. B, luciferase reporter assay of stable cell lines expressing hypoxia tolerant HIF1⍺ (B. bennetti), intolerant (B. benjamini), human controls (hemi-/homozygous and heterozygous), and WT cell lines. Statistical significance between means of trials are shown: ns, p > 0.05; ∗p ≤ 0.05; ∗∗p ≤ 0.01; ∗∗∗p ≤ 0.001; ∗∗∗∗p ≤ 0.00001. C, confocal microscopy images of stably transfected cells expressing exogenous HIF1A with tagBFP fluorescent marker. Nuclei were labeled with NucRed marker (red) and mitochondrial membrane potential shown with tetramethylrhodamine, methyl ester (TMRM) probe. Scale bar indicates 10 μm for all images. D, Western blot of 100 μg total protein extracted from stably transfected cells after exposure to 24 h of 21% or 1% oxygen for all the species. Top, membrane was blotted with anti-HIF1A pAb, then mild stripping used to remove antibody before blotting again with anti-HA mAb (bottom). HIF, hypoxia inducible factor.

Hypoxia-tolerant transgenic cells show a highly energetic metabolic profile

The measured respiratory rates of our transgenic cells lines show highest basal oxygen consumption rates (OCRs) in cells expressing B. bennetti hif1ab. Upon inhibition of ATP synthase using oligomycin, OCR drops significantly to similar levels as the hypoxia-intolerant cell lines. Similar pattern is seen also in the CoCl₂ treatments, although the basal rates are slightly lower (Fig. 5A). For the extracellular acidification rates (ECAR), a measure of glycolytic rate and lactate production, the cells expressing hypoxia-tolerant hif1ab show comparable ECAR rates as the hypoxia-intolerant hif1ab in normal media. However, following 24 h of hif1ab stabilization with CoCl₂, cells expressing B. bennetti hif1ab show an elevated ECAR rates compared to B. benjamini hif1ab (Fig. 5A). As a result, cells expressing hypoxia-intolerant hif1ab show a decline in overall ATP production (both aerobic and anaerobic) in response to chemically induced hypoxia. In contrast, cells expressing hypoxia-tolerant hif1ab show no change in overall ATP production, but a decrease in aerobic ATP production and a compensation through anaerobic ATP production (Fig. S9). Taken together, our Seahorse aerobic and anaerobic metabolic rate assay suggests that cells stably transfected with hypoxia-tolerant B. bennetti HIF1⍺ have elevated rates of both aerobic and anaerobic metabolism in normal media (>300% that of WT cells). When media is supplemented with CoCl₂ to chemically stabilize HIF1⍺, we see a modest decrease in the rate of aerobic but not anaerobic metabolism. On the contrary, cells expressing hypoxia-intolerant B. benjamini show higher glycolytic profile than aerobic. Upon stabilization of HIF1⍺, both glycolytic and aerobic rates decrease (Fig. 5B). These results suggest that hypoxia-tolerant HIF1⍺ mediates higher activity of aerobic and glycolytic metabolism during normoxia, but during a hypoxic insult, only aerobic metabolism is reduced while glycolytic activity is maintained. This could allow maintenance of ATP supply during prolonged hypoxia.

Figure 5.

Hypoxia tolerant stable cell lines display more energetic metabolic profile than other cell lines.A, top, mean oxygen consumption rate (OCR) across replicates of stably-transfected cell lines with normal D-MEM media (right) or CoCl₂-supplemented media (left) using Seahorse ATP Rate Assay. A, bottom, mean extracellular acidification rate (ECAR) of the replicates of stably-transfected cell lines with normal D-MEM media (left) or CoCl₂-supplemented media (right) using XF Seahorse ATP Rate Assay. A, data were normalized to blank well controls and to total protein concentration per well (in μg/ml). B, energy map of different cell lines basal metabolism showing mean and SEM of oxygen consumption rate (OCR) and extracellular acidification rates (ECAR) in response to normal media (left) or media + CoCl₂ (right). DMEM, Dulbecco Minimum Essential Medium.

Discussion

We used computational and experimental approaches to identify the adaptive mechanisms of hypoxia tolerance in a group of Amazonian electric fishes. We compiled the largest dataset of HIFα genes assembled to date, spanning all vertebrate groups, to investigate signatures of selection associated with hypoxia tolerance. We focused on a group of electric fishes, the genus Brachyhypopomus, due to the largest diversity of adaptability to hypoxia within its 29 species. Our analyses show strong evidence of positive selection in hif1ab, a teleost-specific paralog of HIF1α, in Brachyhypopomus. For comparison, we additionally identified positive selection in Cypriniformes, another group that includes hypoxia-tolerant species, further supporting the idea that hif1ab is a hotspot for hypoxia adaptive evolution. What makes Brachyhypopomus a unique study system for hypoxia tolerance is that all species inhabit environments that would be otherwise uninhabitable (<2 mgl⁻¹), due to high water temperature and associated increase in metabolic and oxygen demands, including those we denote as hypoxia-intolerant. Our experimental data suggest that hif1ab from hypoxia-tolerant species exhibits stronger transactivation than intolerant species, owing to SIMs we identified in the oxygen-dependent degradation and C-terminal transactivation domains. This exaggerated nuclear factor transactivation causes a reduction in aerobic metabolism but an increase in anaerobic metabolism in response to hypoxia.

Most extant teleost fishes possess three HIFα genes, similar to that found in mammals, but that Otocephalan species (and possibly others not covered in this work) possess six copies of HIFα genes. These six genes include both paralogs of HIF1A, HIF2A, one copy of HIF3A, and one copy of HIF4A. In species with additional whole genome duplication events, for example, Salmoniformes, there are at least seven copies. For HIF1, it appears that hif1aa is the retained copy in most teleosts and potentially the ancestral copy (based on synteny), but in Otocephala, we see a shift to reliance on the duplicate copy hif1ab. This is supported by (1) more ubiquitous expression levels of hif1ab across tissues in otocephalan species and (2) loss of oxygen-dependent regulation domain in hif1aa in the same group (Fig. S3). For HIF2A, most teleost species retain the epas1b copy, while Otocephala retain both epas1a and epas1b; however, epas1b remains to be the more abundantly expressed copy. For the other two genes, most teleost fishes retain only hif1al paralog, which is homologous to HIF3A of tetrapods, and otocephalan and salmonid species additionally retain also hif1al2. These findings are supported by a recent study (54).

Multiple lines of evidence from our molecular evolutionary analyses suggest that HIF1α is likely associated with the evolution of hypoxia tolerance in fishes. Our gene-specific datasets of teleost fishes show the highest dN/dS rates in hif1ab in two groups that retain both copies of HIF1α, Brachyhypopomus and Cypriniformes. While there is evidence for elevated evolutionary rates in both groups, the mechanism(s) of the adaptation is likely to be different in these groups, shown by a lower evolutionary rate estimate for their combined partition as well as the incongruent patterns of sites under positive selection in the site-wise dN/dS analysis. Additionally, we find strong evidence of positive selection in hif1ab using random-sites model of the small datasets of hypoxia-tolerant groups and the branch-sites model analyses of the complete dataset with the Brachyhypopomus hif1ab branches labeled in the foreground. These results highlight that HIF1α is a hotspot for adaptive evolution associated with hypoxia tolerance in fishes and that Brachyhypopomus shows strong evidence of genetic adaptation at this genetic locus. Previously knockout of HIF1α paralogs in zebrafish, individually or in concert, caused a decrease in hypoxia tolerance as implied by reduced time to loss of equilibrium during a hypoxia insult (25). Interestingly, knockout of hif1ab, but not hif1aa or hif1aa + hif1ab, causes the largest drop in routine O₂ consumption (25). Findings from this earlier study emphasize the importance of hif1ab gene copy in the regulation of oxygen consumption in fishes.

In general, we find an elevated transactivation of hypoxia response genes (measured through PGK-1 hypoxia response element reporter) in vitro in the presence of hypoxia-tolerant HIF1A. An interesting deviation from this trend is B. sullivani, a species that, like B. benjamini, is restricted to normoxic terra firme rainforest streams (39). However, we find that its sequence contains hypoxia-tolerant SIM3 motif, and its transactivation in vitro is more resemblant of hypoxia-tolerant levels. One plausible explanation for this is that B. sullivani, but not B. benjamini, occasionally enters the edges of relatively poor oxygenated ephemeral swamps adjacent to streams. These swamps form when terra firme streams are flooded by heavy rain and are characterized by DO < 1.5 mgl⁻¹ (40, 59). Our data suggests that B. sullivani has some elements of the transcriptional mechanisms to respond to hypoxia in a manner similar to hypoxia-tolerant species at the molecular level, as shown by the elevated transactivation and presence of SIM3. However, it remains a challenge to infer organismal response without conducting whole-animal experiments to confirm the hypoxia tolerance capacity of B. sullivani. More conspicuously, the B. sullivani hif1ab sequence has a 40 amino acid indel in the oxygen-dependent degradation domain of ∼22 amino acids downstream of the SIM1 motif. This indel was not found in any other species by means of BLAST of the entire Genbank database. It remains unclear how, if at all, this indel could affect hif1ab function, as this region of the protein is characterized by high disorder probability (Fig. S5). The functional consequences of this large indel could be the subject of future studies.

The results of our in vitro assays for B. bennetti and B. walteri are highly congruent with an experimental study of the whole organism response to hypoxia in Brachyhypopomus and other electric fish (46). Both of these species (labeled Brachyhypopomus sp. nov 2 and 3, respectively) show resilience to anoxia for the entire period of the trial in the lab. Both our results and the whole organism study suggest a strong correlation between the molecular mechanism we present with our luciferase assay and the organismal responses to hypoxia, which allow us to postulate a link between the organismal and cellular adaptation to hypoxia. Higher HIF1α transactivation means that the magnitude of cellular response to hypoxia is higher than that of hypoxia intolerant. Our Seahorse ATP rate assay suggests that the elevated hif1ab transactivation causes cells to maintain consistent anaerobic metabolism during a hypoxia mimicry while only depressing aerobic metabolism in hypoxia-tolerant cells, which is in disagreement with intolerant hif1ab cells where both processes are depressed. It also seems that cells expressing B. bennetti hif1ab appear to be more energetic (i.e., higher aerobic and anaerobic metabolism) than other cells, during normoxia. In Amazon floodplain systems where B. bennetti occur, a prolonged period of normoxia occurs during the dry season when levels of biological oxygen demand decline in floodplain waters, but as water levels rise, the normoxia is interrupted by sudden periods of hypoxia until the waters are eventually permanently hypoxic during the night (45). Using next-generation sequencing, further studies should investigate the metabolic genes and networks downstream of HIF1α that are upregulated and downregulated by hypoxia-tolerant hif1ab activity, compared to that of intolerant and WT human cell lines.

Our study is the first to implicate SIMs in the regulation and enhancement of HIF1α transactivation and particularly in the context of hypoxia tolerance. SIMs are motifs within proteins that have been shown to noncovalently interact with SUMO modifications on proteins (36, 60), and this interaction has been reported to affect function of nuclear factors (61). SUMO–SIM interaction can be within the same protein or on different proteins, and previous work has shown the involvement of this interaction in modulating protein activity, localization, stability (61), or impacting protein–protein interactions (62), respectively. For example, SIMs in the transactivation domain of human c-Myb factor boost transactivation of this hematopoetic transcription factor and increase its proliferative capacity (37). In our study, we find two SIMs in the oxygen-dependent degradation domain, part of the protein with characteristics of intrinsically disordered region. Disordered regions generally show a high degree of sequence divergence and large number of indels, making them generally unsuitable for molecular evolutionary analyses, despite harboring valuable functional information (63). Despite the lack of structure, we found striking conservation of VEEVV motif (SIM1) in all hypoxia-tolerant species. Introducing this motif alone to the hypoxia-intolerant hif1ab or human HIF1A backgrounds enhance their transactivation ability. Due to the proximity of this motif to the PAC domain (18 amino acids upstream of SIM1) involved in dimerization of HIF1α and HIF1β, we hypothesize that this SIM could be impacting transactivation through modulating the binding affinity of these proteins by interacting with SUMO on HIF1α itself or on HIF1β. Both HIF1α and HIF1β are known to be SUMOylated (34, 64, 65). Less surprisingly, SIM3 in the C-terminal transactivation domain seems to also enhance transactivation of HIF1α, although this observation is only statistically significantly in the human background. This SIM, however, could be involved in the interaction between HIF1α and factor inhibiting HIF1, given that SIM3 is 14 amino acids downstream of N803, the site of hydroxylation in HIF1α. Alternatively, SIM3 could be involved in the interaction with the transcription machinery. Further biophysical models and analyses are required to test these hypotheses. We also identified one SIM in hypoxia-intolerant hif1ab, SIM2, which we suspected would not enhance HIF1α transactivation or perhaps repress it. Introducing this SIM seems to only cause a statistically significant reduction in the transactivation of hypoxia-tolerant hif1ab when introduced in conjunction with SIM1 or SIM1+SIM3. Due to the proximity of this SIM to N-terminal VHL recognition site (25 amino acid upstream of this SIM), we hypothesize that SIM2 affects HIF1α interaction with its regulatory machinery.

Taken together, our results show that adaptive evolution in HIF1α, particularly hif1ab in fishes, has facilitated hypoxia tolerance through SUMO-interaction mechanisms. At the center of the mechanism of this adaptation, two SIM motifs, in the ODDD and CTAD domains, likely interact with SUMO on modified residues of either HIF1α itself or HIF1β. This SUMO–SIM interaction is likely to either cause better folding of structured domains of HIF1α or higher affinity binding to HIF1β, respectively (Fig. S10). Future studies should investigate these two scenarios as possible mechanisms of the role of the SUMO–SIM interaction in hypoxia response. Additionally, investigation of the expression levels of hypoxia-response genes in transgenic hypoxia-tolerant cell lines could shed light on the genes involved in the highly energetic metabolic profiles.

Studies of adaptation to detrimental environmental conditions, such as hypoxia, are challenging because responses to such environmental challenge involves many cellular and systems processes. For example, some of the physiological adaptations reported include (but not limited to) the following: excretion of ethanol as a by-product of anaerobic metabolism in carps and goldfish (5), structural remodeling of gill morphology and surface area (66, 67), alteration of cardiac output (68), elevated hemoglobin production or hemoglobin-oxygen affinity (69). While these physiological processes garner much of the attention of hypoxia tolerance studies, cellular and molecular mechanisms of hypoxia tolerance are often overlooked. The reductionist approach of interrogating molecular players involved in some of these processes in vitro offers a powerful means for elucidating their specific roles in the adaptive processes. To capture the complexity inherent in metabolic adaptations, our study integrates data across ecological, evolutionary, organismal, and molecular dimensions, building on a study system that has previously been investigated both in the field and at a whole-organism level. With the expanded use of genome editing in fishes, future studies might investigate our findings in vivo for a more complete organismal and ecological insights. Our study is the first to leverage closely-related species with varying degrees of hypoxia tolerance for an in-depth examination of the molecular mechanisms. We anticipate that our work will serve as a foundation for further investigations into the roles of SUMOylation and SIM interaction in the fine-tuning of the hypoxic response, as well as their implications for evolution and human diseases.

Experimental procedures

Sequence evolution analyses

We used Ensembl database to collect coding sequences of HIF⍺ genes to ensure consistent taxonomic sampling across genes. Using BLAST program, we queried the Ensembl database using zebrafish hif1aa, hif1ab, epas1a, epas1b, hif1al, and hif1al2 for orthologous sequences from all available vertebrate and invertebrate genomes. Resulting sequences were annotated with complete species and common names using custom-written bash script. All resulting sequences were combined with a small dataset of sequences that were curated by local BLAST search of zebrafish coding sequences in Ostariophysi transcriptomes (70, 71, 72) or sequenced by RT-PCR or RNAseq from frozen tissues in this study (Table S1; see below for RT-PCR). For transcriptome-sequenced samples (B. beebei, B. bennetti, B. sullivani, and B. verdii), RNA was extracted from field-collected, RNAlater-preserved frozen electric organ or skeletal muscle tissues using RNeasy Mini kit (Qiagen) following manufacturer’s recommendations, and the library prep and sequencing was performed at The Centre for Applied Genomics at the Sickkids hospital. Trimmomatic v0.35 (73) was used to remove low quality ends of the reads, then Trinity v2.5.2 (74, 75) was used to de novo assemble the transcriptomes for each species separately, using the recommended paired-end reads settings, and using in-silico normalization for B. bennetti. Local BLAST searches were then used to collect coding sequences of HIF⍺ genes from these transcriptomes. For instances where Brachyhypopomus sequences were not completely recovered from their assembled transcriptomes using BLAST (B. beebei hif1aa and epas1a; B. bennetti epas1b and hif1al2; B. sullivani hif1aa, hif1ab, epas1a, epas1b, hif1al and hif1al2), we used a custom pipeline to assemble these HIF⍺ genes from raw RNAseq reads, by aligning the reads to another Brachyhypopomus reference sequence using BWA algorithm v0.7.17 (76), then generating a consensus sequence of the aligned reads using Samtools v1.16.1.

Duplicate (i.e., sharing the same accession number) and incomplete sequences were removed from the combined dataset. Our final dataset, therefore, consists of 675 sequences. Sequences were first trimmed at the 5’ end of the correct reading frame, then aligned using translation alignment by first translating nucleotide sequences to amino acids, followed by alignment of amino acids using MAFFT algorithm (77) with BLOSUM45 scoring matrix, and finally the amino acid alignment was reverse translated back to nucleotide sequences post alignment, as implemented in Geneious Prime (78). Gaps that were found in >50% of the sequences were removed, and the full amino acid alignment (675 sequences) was used to generate a maximum likelihood phylogeny in IQ-Tree (79) using JTT+F+G4 (according to Bayesian Information Criterion model test) substitution model with 1000 replicates of SH-like approximation likelihood (80) and 1000 bootstrap replicates.

Using the full dataset, we used random-sites and clade model D in the algorithm codeml as implemented in the package PAML (81, 82) to estimate the rate of non-synonymous (dN) to synonymous (dS) substitutions (ratio also referred to as ω). Clade model D allows a site class with divergent site rates where the estimation of dN/dS in the designated clades is computed independently. We used the computed dN/dS (or ω) in the divergent site-class as a proxy for the rates of evolution of the different genes, with the assumption that divergent codons between these genes would fall into this site class. We used the random-sites (M0, M1, M2, M3, M7) analyses to evaluate the overall evolutionary rates in the entire dataset and as a null model to the clade model analyses. We additionally used branch-sites models to test if Brachyhypopomus genes have experienced elevated evolutionary rates indicative of adaptive evolution, compared to other taxa in our dataset. We used branch-sites models to analyze the full dataset with different Brachyhypopomus genes in the foreground and the remainder of the 675 sequences in the background. We tested hif1ab alone and a combination of genes that appear abundant in the transcriptomes of Otocephala (hif1ab, epas1b, and hif1al), a combination of genes that have low expression levels (hif1aa, epas1a, and hif1al2), or all genes combined into a single foreground partition. We also tested a combination of Brachyhypopomus branches of hif1ab, hif1al, and hif1al2. Branch-sites model results were compared to a null model that restricts the dN/dS in the foreground partition to 1, and likelihood ratio test was used in all analyses to compare the alternative models to their null models and computed as twice the difference between the likelihood scores of the alternative and null models. Using the likelihood ratio test and the difference of parameters between the alternative and null models, p-value was computed using a chi-distribution (83).

From the full dataset, we also generated a teleost fishes HIF1A, HIF2A, HIF3A, and HIF4A datasets, which were used to further conduct separate random-sites and branch-sites analyses to test for adaptive evolution of the teleost-specific paralogous copies of each of the genes in Brachyhypopomus, in reference to Cypriniformes, another group of hypoxia-tolerant fishes. These datasets also included basal fishes that predate the teleost-specific whole genome duplication event, including the spotted gar, the reedfish, and elephant shark. For these subset datasets, gaps that were present in most sequences were additionally removed. We labeled clades corresponding to (1) Brachyhypopomus, (2)Cypriniformes, and (3) Brachyhypopomus + Cypriniformes combined. We used branch-sites model in codeml to estimate the rates of evolution of specific groups of the dataset compared to the background. Branch-sites model results were compared a null model with dN/dS fixed at 1.0 in the foreground to determine the statistical significance of the branch-sites model. Finally, we generated small datasets of Brachyhypopomus, Gymnotiformes, Cypriniformes, Ostariophysi, and Otocephala sequences from our full datasets and used random-sites analyses (models M0, M1, M2, M3, M7, and M8) on these datasets to estimate evolutionary rates of these groups independent of each other. Model M8 was statistically significant in all datasets, and we used the dN/dS estimate of the positively selected site classes of these analyses compared between groups.

We compiled expression levels of HIF⍺ genes from the electric eel (72), Electrophorus sp., and zebrafish (zfin.org), D. rerio. We scaled the expression levels across tissue (or developmental stage) to identify gene(s) that are most abundant at the tissue or developmental stage. SIMs were predicted in the amino acid sequences of B. bennetti, B. benjamini, and Homo sapiens HIF1⍺ using GPS-SUMO (38). Motifs that were shared across all sequences were ignored; only those unique to one sequence were considered. Three candidate motifs were identified, one was only present in hypoxia-tolerant B. bennetti (SIM1), another was only present in the hypoxia-intolerant B. benjamini (SIM2), and a third was identified in C-terminal transactivation domain in all sequences, but the hypoxia-tolerant sequence possessed His substitution while the other two sequences showed Arg (SIM3). These three motifs were further investigated experimentally by mutating residues at homologous residues of SIM motifs to the opposite amino acid identity (i.e., hypoxia-tolerant residues into intolerant background and vice-versa).

Reverse transcription and cloning

Total RNA was extracted from muscle or electric organ tissues collected in the field in the Amazon and preserved in RNAlater, using RNeasy mini kit (Qiagen) following manufacturer’s instructions and quantified using Qubit (ThermoFischer). First-strand cDNA was synthesized using SuperScript III kit and random hexamers with 2.0 μl of total RNA. Briefly, primers were first annealed to templates at 65 °C for 5 min, then reverse transcriptase enzyme was added and reverse transcription proceeded for 10 min at 25 °C, 50 min at 50 °C, and 5 min at 85 °C. Full HIF1α was then amplified from cDNA library using gene-specific degenerate primers (HIF1ab_Brachy_Fd: 5′-ATGGATACTGGKGTYGYCACGGAAAAGAAAAGGGTG-3′ and HIF1ab_Gym_Rd: 5′-TCARTTAACTTGGTCSAGAGC-3′) and proof-reading Pfu DNA polymerase (Biobasic). PCR product was then run on a 1% agarose gel, and the band equivalent to ∼2300 to 2400 bp was excised and purified using QIAquick gel extraction kit (Qiagen). Inserts were then blunted using blunting enzyme (Thermo Fisher Scientific), before ligation to pJET1.2, followed by transformation of DH5α Escherichia coli. Clones were then screened using colony-PCR, and positive colonies were further cultured overnight, and plasmids were purified using GeneJET plasmid Miniprep kit (Thermo Fisher Scientific). Purified plasmids were then sanger-sequenced at the Centre for Applied Genomics using multiple primers that span the entire transcript with at least 100 bp overlap.

HIF1α inserts were amplified from pJET plasmids using primers that incorporate a BamHI restriction enzyme site (GTCAGTGGATCCGCG) upstream and NotI site (GCGATGCGGCCGC) downstream of the reading frame. Inserts with restriction enzyme sites were ligated into pJET, used to transform E. coli as previously described, and purified. The enzymes BamHI and NotI were then used to clone HIF1α inserts into p1D4-hrGFPII. Upon validation of correct insertion using sanger-sequencing, expression plasmids were mass-produced and purified using PureLink Plasmid Maxi kit (Invitrogen) and quantified using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific). Alternatively, HIF1α inserts were amplified from pJET backbone using primers with overlapping fragments to regions up- and down-stream of the BamHI and NotI sites in the p1D4-hrGFPII plasmid, and cloning was performed using Gibson assembly (84). Isothermal (Gibson) assembly was performed for general cloning when restriction-enzyme cloning was not possible due to internal cut sites in templates, using equimolar amounts of DNA fragments with at least 21 bp of homologous ends, in thermocycler at 50 °C for 1 h, then directly used to transform chemically competent E. coli.

Transient expression and harvesting

Adherent human embryonic kidney cells (GripTite 293 MSR; Thermo Fisher Scientific) were cultured in Dulbecco’s Minimum Essential Media (D-MEM; Invitrogen Cat: 11995073) supplemented with 10% fetal bovine serum (Invitrogen Cat: 12484028), penicillin-streptomycin, 0.1 mM non-essential amino acids (Invitrogen Cat: 11140050) at 37 °C under 5% CO₂. Cells at ∼90 to 100% confluency (0.2 × 10⁶ cells per well) were transfected using Lipofectamine 3000 transfection reagent (Invitrogen Cat: L3000015) following manufacturer’s recommendations. GFP expression was used to confirm success of transfection 24- and 48-h posttransfection. For hypoxia treatments, cells were moved into a tri-gas incubator 24 h posttransfection and incubated at 1% O₂ for 6 to 24 h prior to lysing or fixation.

Luciferase-reporter assay

Cells were grown in a 24-well plate and transfected with 1 μg of HIF1α expression construct, equimolar of reporter construct HRE-Luciferase (85) with 3× HRE from phosphoglycerate kinase 1 (gift from Navdeep Chandel; Addgene plasmid # 26731) driving the expression of firefly luciferase, and 1/50 M equivalent of Renilla luciferase reporter construct pGL4.75 (Promega) as a transfection control. Two sets of transfections were done in four replicates. At 24 h post transfection, media was changed, and one set of plates was incubated in at 37 °C, 5% CO₂, and ambient O₂ and another set was incubated in a tri-gas incubator maintained at 37 °C, 5% CO₂, and 1% O₂ to induce hypoxia. At 48 h posttransfection, media were removed from all wells, and cells were washed with warm PBS (1× PBS) prior to being lysed with 100 μl/well of 1× passive lysis buffer (Promega) in the wells. Cells were incubated at room temperature for 10 min with gentle horizontal shaking and then frozen at −80 °C for at least 1 h to facilitate cell lysis. Cell lysates (40 μl) were assayed for luciferase activity using the dual-luciferase assay (Promega) in Luminoskan Ascent Microplate luminometer (Thermo Fisher Scientific).

Western blotting

Cell lysates were quantified using Qubit Protein Assay (Thermo Fisher Scientific; Cat: Q33212), and 40 to 100 μg total proteins were mixed with equal volumes of 2× protein loading dye and boiled at 95 °C for 10 min before loading on Novex WedgeWell 4 to 20% Tris-Glycine, 1.0 mm pre-cast gels (Thermo Fisher Cat: XP04202BOX), or in-house made gels. After SDS-PAGE, proteins were transferred to polyvinylidene difluoride membrane (Bio-Rad) and visualized using Ponceau staining to confirm success of transfer. Blots were then blocked in 5% w/v skim milk for 1 h at room temperature on rocking platform and then incubated overnight at 4 °C with gentle agitation in 1:1000 HIF1α Rabbit pAb (Novus Biologicals Cat: NB100-479) and 1:1000 α-tubulin Mouse mAb (CST Cat: DM1A) diluted in 5% w/v skim milk. For stably transfected cells, 1:1000 HA-tag Rabbit mAb (CST Cat: C29F4) and 1:1000 α-tubulin Mouse mAb were used. After washes of primary antibodies, blots were incubated for 1 h at room temperature with gentle rocking in 1:10,000 Goat anti-Mouse (LI-COR Cat: 925-68020) and 1:10,000 Goat anti-Rabbit (LI-COR Cat: 926-32211) IRdye secondary antibodies. Blots were visualized on LI-COR Odyssey CLx imager.

Site-directed mutagenesis

Amino acid substitutions/changes were introduced using Q5 mutagenesis kit (New England Biolabs Cat: E0554S). In short, in 25.0 μl reaction, <25.0 ng of template plasmids were amplified using 0.5 μM forward primers harboring the desired changes and 0.5 μM reverse primers that perfectly match the template sequence. After 25-cycle amplification, parental plasmids were digested using KLD enzyme mix (New England Biolabs), then transformed in chemically competent E. coli as previously described. A summary of the nucleotide substitutions introduced are summarized in Table S14. Colonies were cultured overnight, and plasmid was purified using GeneJet Plasmid Mini kit before sanger sequencing at The Centre for Applied Genomics.

Genome editing, DNA extraction, genotyping

We used a knock-in to knock-out approach to integrate donor plasmids carrying B. bennetti (hypoxia tolerant), B. benjamini (hypoxia intolerant), and Homo sapiens (control) with tagBFP marker at the endogenous HIF1α locus. Two double-stranded cuts were made at ch14:61695771 and ch14:61747525 to remove the entire coding region and introns of endogenous HIF1α in HEK293H-MSR cells, following previously published protocol (86). In short, cells were grown in 100 mm plates up to 80 to 100% confluency, then transfected with 5 μg of each 5’ and 3’ sgRNA (cloned in empty gRNA cloning vector) and 10 μg of each of donor plasmid and human-optimized Cas9 (hCas9) using Neon electroporation kit (ThermoFisher) with 1100V, 20 msec duration, and 2 pulses. The gRNA cloning vector and hCas9 were a gift from George Church (Addgene plasmid # 41824 and 41815, respectively) (87). Cells were then plated and left to recover at 37 °C, 5% CO₂ incubator. After 1 week, cells were first bulk sorted using flow cytometer at Temerty Faculty of Medicine Flow Cytometry Facility based on tagBFP brightness, and cells were left to recover for another week. After recovery, cells were sorted again into single cells in a 96-well plate based on tagBFP fluorescence and left to grow until ∼70% confluent for passaging.

DNA was extracted from cell clones using DNeasy Blood and Tissue extraction kit following manufacturer’s recommendations (Qiagen). Genotyping PCR was conducted using primer pairs that span the 3’ sgRNA cut-site (summary in Fig. 4A). Amplification was conducted in 25.0 μl reactions containing 0.5 μM forward and reverse primers, 0.4 μM dNTPs, 10 ng template DNA, and 0.25 μl Q5 DNA polymerase (New England Biolabs). Amplification reaction included 30 s denaturation at 98 °C; 35 cycles of denaturation at 98 °C for 30 s, 59 °C annealing for 20 s, and extension at 72 °C for 90 s; and final extension step at 72 °C for 120 s. Successful amplification was confirmed using 1% agarose gel electrophoresis.

Real-time ATP rate assay

Transgenic cell lines were seeded into Seahorse Xfe96 well plate at both 0.5 × 10⁴ and 1.0 × 10⁴ cells per well and two sets of four replicate wells per seeding density and cell line. To one set, full D-MEM media was used (10% fetal bovine serum and 1× non-essential amino acid supplement) while the other set received full D-MEM media with 250 μM CoCl₂ (Sigma-Aldrich Cat: 15862). Seeded cells were left at room temperature for 1 h before being incubating at 37 °C, 5% CO₂, 21% O₂ overnight. On the day of the assay, cells were washed with Seahorse D-MEM media (pH 7.4, 10 mM XF glucose, 1 mM XF pyruvate, 2 mM glutamine) once, then incubated in Seahorse media at 37 °C in a non-CO₂ incubator for 1 h. Cells were washed again with Seahorse D-MEM media before the assay was run. ATP rate assay was completed at the Sickkids Proteomics, Analytics, Robotics & Chemical Biology Centre. Assay included 20 min equilibration, basal rate measurement step for 18 min, three cycles of 1.5 μM oligomycin treatment for 18 min, and three cycles of 0.5 μM rotenone and 0.5 μM antimycin A treatment for 18 min. Seahorse data was normalized to background readings of empty wells (A01, H01, A12, and H12) and total protein concentration per well, measured using Pierce Bradford Assay (Thermo Fisher) kit and Spectra Max (Molecular Devices) plate reader. ATP production rate is then calculated from OCR and ECAR using pre-established formulas (88) and relying on ATP phosphorylation per oxygen atom reduced, calculated in the Seahorse Analytics software (Agilent Technologies).

Data availability

Newly sequenced HIF1α sequences are deposited on Genbank accession PP408221 to PP408246.

Supporting information

This article contains supporting information.

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.

Reviewed by members of the JBC Editorial Board. Edited by Clare E. Bryant