Several EGLN assays have been described which include the determination of enzyme activity by measure of cosubstrates (dioxygen and 2-oxoglutarate) consumption/CO₂ production and the detection of HIF hydroxylation using mass spectrometry or capture of VHL 32 33 . In addition, strategies to determine EGLN activity within cells have also been developed. These include the detection of reporter proteins fused to ODD 34 and the use of antibodies raised against hydroxyproline-containing ODDs 35 . The major limitation of the in vitro assays is that they are performed in a non physiological environment. On the other hand, the cell-based assays have to deal with the endogenous HIF machinery. To circumvent these problems we sought to reconstitute the HIF regulatory machinery in yeast cells, where expressed proteins are in a cellular environment but where no endogenous proteins related to the HIF pathway are expressed, therefore eliminating possible sources of interference. To this end, we cloned a fragment of human HIF1α (residues 554–576) containing P564 in frame with yeast Gal4 activation domain (AD-P564) (figure 1A). This fragment is sufficient for EGLN recognition 23 and VHL binding 36 . In addition, we cloned a VHL fragment encoding for its β-domain (residues 63 to 157) in frame with the yeast Gal4 DNA binding domain (DB-VHL) (figure 1A). The VHL β-domain is involved in HIF1α binding, while the α-domain, missing in this construct, interacts with the elongins B/C that are required to promote HIFα ubiquitination. Then we tested the interaction between these fusion proteins as their ability to support the growth of a yeast strain conditionally auxotrophic for histidine and adenine (figure 1B). As shown in figure 1C, yeast cells transformed with AD-P564 in combination with DB-VHL failed to grow in plates lacking histidine, suggesting lack of significant interaction between the two fusion proteins. This result was expected since no EGLN orthologs are present in yeast and VHL binding to HIF1α is strictly dependent on proline hydroxylation 8 14 . In addition, these data indicates that residual binding of VHL to non-hydroxylated AD-P564 is below detection limit in our assay conditions. Importantly, the expression of EGLN3 (construct E3/DB-VHL, figure 1A), together with the two fusion proteins, is sufficient to promote direct binding of DB-VHL to AD-P564, as indicated by the growth of transformed yeast in restrictive plates (figure 1C). The effect was not restricted to EGLN3 since the expression of EGLN1 also triggered the interaction between the fusion proteins (figure 1D). To further confirm that DB-VHL/AD-P564 interaction was mediated by the hydroxylation of AD-P564 by EGLNs, we repeated the experiment using a mutant HIF peptide in which P564 was replaced by an alanine residue (AD-mutP564A, figure 1A). As shown in figure 1D, DB-VHL was unable to interact with AD-mutP564A regardless of the presence of either EGLN. Moreover, in spite of relaxed target requirements 23 37 , it has been described that EGLN isoforms display a specific preference for the two target sequences in HIF1α 17 21 22 23 . Therefore, we next wanted to determine whether in this system the EGLNs retained their native substrate specificity. Figure 1D shows that EGLN3 efficiently hydroxylates the peptide containing the sequence surrounding proline 564 (AD-P564), but it has no detectable activity toward a peptide comprising residues 392–414 derived from HIF1α (AD-P402). In contrast, EGLN1 was able to hydroxylate both sequences allowing their binding to VHL (figure 1D). Interestingly, the activity of EGLN1 toward AD-P402 was slightly lower than its activity using AD-P564 substrate as evidenced by the different growth of yeast transformed with each sequence in the high stringency plates (figure 1D). These results are in perfect agreement with previous reports describing the activity of the EGLN isoforms against different substrates 23 37 38 39 . Finally, further confirmation that EGLNs retained their substrate specificity was obtained through the analysis of the effect of L574A and L562R mutations (Additional file 1).

All together, these results indicate that this system faithfully reconstitutes the regulation of HIFα proteins by the sequential action of EGLN enzymes and VHL binding. In addition, data show that the β-domain of VHL is sufficient for target recognition and binding.

The system described above relies on EGLN activity for DB-VHL/AD-P564 interaction, thus we reasoned that EGLN inhibitors should interfere with yeast growth. However, in cell-based assays, the inhibitory effects of drugs have to be carefully controlled for unspecific side effects on cell physiology or on the assay system itself. In order to have a control for the drug specificity and rule out potential side effects on yeast growth, we generated a set of constructs analogous to those described above but based on a protein machinery unrelated to the HIF pathway. In this control system, yeast growth is dependent on the interaction between an immunoreceptor tyrosine-based activation motif (ITAM), derived from the type Iγ IgE receptor (FcεRIγ), and the Src-Homology domain 2 (SH2) from the Syk protein. Importantly, the binding of Syk-SH2 tandem domains to the FcεRIγ-ITAM is dependent on the tyrosine phosphorylation of the latter 40 . The vectors generated for this system (figure 2A) encode for a Syk-derived tandem SH2 domains fused to the activation domain of Gal4 (AD-SH2) and for an ITAM from FcεRIγ fused to the Gal4 DNA binding domain (DB-ITAM). Finally, the cytoplasmic tyrosine kinase Lck was cloned into the pBridge vector to induce DB-ITAM phosphorylation and trigger its binding to AD-SH2 (figure 2B). Since yeast do not contain orthologues of the proteins of this system, AD-SH2 did not significantly interact with DB-ITAM unless the tyrosine kinase Lck was co-expressed in the assay (figure 2C), as previously reported 40 . Thus, the Lck and EGLN-based assays are analogous in that both have three components: an enzyme and two fusion proteins whose interaction allows yeast growth in restrictive media. Additionally, in both systems the interaction between the fusion proteins, and thus yeast growth, requires a modification of one of these proteins by the enzyme (compare figures 1B and 2B).

Next we tested the ability of these combined systems to detect molecules that interfere with HIF regulation. For this purpose we selected two previously characterized EGLN inhibitors, dimethyloxaloylglycine (DMOG) 41 and 6-chlor-3-hydroxychinolin-2-carbonic acid-N-carboxymethyl-amide (S956711) 31 . In order to facilitate the quantification of the effects of inhibitors on these systems, we set up the experiments in liquid media so that yeast growth could readily be determined by the optical density of the cell culture. Since the growth rate of yeast transformed with different constructs in each culture media varies, we adjusted the initial concentration of each yeast strain for each media so that all cultures achieved logarithmic growth in overlapping time windows (Additional file 2). As shown in figure 3A, treatment with increasing doses of S956711 had a strong effect on the growth of yeast expressing AD-HIF and DB-VHL/EGLN3 in restrictive media. In contrast, when these cells were cultured in control media, where no interaction between fusion proteins was required for growth, S956711 had only a minor effect when used at 100 μM (figure 3A). Thus, the effect of S956711 is not due to an unspecific effect on yeast viability/growth. To further confirm the specificity of the effect upon the HIF system, we tested the effect of S956711 on yeast transformed with AD-SH2 and DB-ITAM/Lck. As shown in figure 3, S956711 had no significant effect in this control system regardless of the culture media. These results rule out that the inhibition of the EGLN system could be due to side effects of S956711 on transcription/translation of the three-hybrid elements. The specific inhibitory effect of S956711 was highly reproducible and unspecific effects were undetectable at doses below 100 μM (figure 3B). Similarly to the results observed for S956711, treatment with DMOG also resulted in a dose-dependent inhibition of the interaction between AD-HIF and DB-VHL (figure 4A). As shown for S956711, DMOG inhibition was specific for the HIF regulatory machinery, since no effect was observed on the binding of AD-SH2 to DB-ITAM (figures 4A and 4B).

DMOG is a 2-oxoglutarate analogue that probably inhibits EGLN activity by competition with the reaction cosubstrate 2-oxoglutarate 42 . Thus, DMOG should affect EGLN catalitic activity (substrate hydroxylation), but not substrate binding. With the purpose of testing this hypothesis, we investigated the effect of DMOG on EGLN3 binding to HIF using a two hybrid-based assay 23 . In this assay, yeast growth in restrictive media is dependent on the direct interaction between a Gal-4 DNA binding domain-EGLN3 (DB-E3) and AD-HIF fusion proteins (figures 5A and 5B). The interaction between DB-E3 and AD-HIF is specific since a single mutation on the target proline in HIF sequence abolishes yeast growth (figure 5C) 23 . Importantly, this assay allows the direct comparison of DMOG effect on EGLN3 ability to bind HIF and act upon it. As shown in figure 5D, DMOG treatment had no effect on the direct binding of EGLN3 to its substrate, as demonstrated by the lack of effect on the growth of yeast expressing DB-EGLN3 and AD-HIF (E3/HIF system). However, at the same doses it clearly affected the hydroxylation of AD-HIF by EGLN3 (figure 5D, VHL/HIF system). This result indicates that, as expected, DMOG does not affect EGLN binding to substrate. Therefore, the effect of this drug on the reconstituted HIF system (figure 4) is likely due to its effect on EGLN activity rather than the EGLN3/HIF interaction. It should be noted however that these results do not exclude a potential effect of DMOG on VHL/hydroxylated-HIF interaction.

In addition to S956711 and DMOG, we also tested more general inhibitors, such as deferoxamine and cobalt chloride, that have been widely used to activate HIF 8 14 . EGLNs contain a tightly bound Fe⁺² atom in the catalitic site that is required as a cofactor in the hydroxylation reaction. Consequently, iron sequestration by the chelating agent deferoxamine results in EGLN inhibition 8 14 . On the other hand, the mechanism of EGLN inhibition by cobalt chloride is controversial 43 44 . It might act by substitution of the iron from the enzyme catalytic site and/or interfere with ascorbate transport 44 , which is required to prevent the oxidation of the catalytic site of the enzyme. In our assays, cobalt chloride had an important toxic effect on yeast and showed little specific inhibition of the EGLN system (data not shown), thus we did not study it in further detail. On the other hand, we were unable to detect any significant effect of deferoxamine on yeast growth at concentrations up to 1 mM (data not shown). A possible explanation for the lack of effect of deferoxamine is that treatment might induce an adaptive response in yeast that increases their iron uptake 45 . In spite of their wide use as hypoxia mimetics, a recent report describes that cobalt and deferoxamine are ineffective inhibitors of EGLNs in vitro 43 , which is in agreement with the lack of significant effect of these drugs in our system.

All together these results indicate that the combination of the control (Lck-dependent) and experimental (EGLN-dependent) systems constitutes an efficient assay to identify small molecules that have a specific effect on EGLN activity, while revealing those that have unspecific or toxic effects. Importantly, the effect of S956711 and DMOG in our system was achieved at doses equivalent to those required for the inhibition of EGLNs in mammalian cells 31 41 .

The S. cerevisiae AH109 strain, yeast growth media reagents YPD (yeast extract/peptone/dextrose), SD (synthetic defined) amino acids, X-gal (5-bromo-4-chloroindol-3-yl β-D-galactopyranoside) and plasmids (pGAD and pBridge) were from BD Biosciences, (Palo Alto, CA).

Constructs expressing HIF1α(554–576) and hSYK(11–263) as Gal4AD fusion proteins were generated by PCR amplification (figure 5 and reference 23 ) of the indicated coding region from human cDNA and cloned into the EcoRI/BamHI or EcoRI sites of pGAD-T7 plasmid (BD Biosciences) respectively. The Gal4DBD fusion proteins were generated by PCR amplification (table 1) of the indicated coding regions from the cloned full length sequence (VHL) or from human cDNA (FceRIγ). VHL (encoding for residues 63–157) and FceRIγ (encoding for residues 45–86) amplicons were then cloned into the XmaI/SalI or EcoRI sites respectively. Finally, the coding sequences for EGLN3 and LCK were PCR-amplified (table 1) and cloned into the BglII or NotI/BglII sites of pBRIDGE-VHL(63–157) or pBRIDGE-FceRIγ (45–86) constructs respectively. LCK was amplified from human cDNA and EGLN-3 was amplified from its cloned cDNA, which was generously provided by Steven L. McKnight. For the two hybrid assays, the constructs expressing Gal4DBD fused to EGLN-3 were generated by cloning its coding sequence into the XmaI/SalI sites of the pBRIDGE plasmid (figure 5 and reference 23 ).

In Table 2 are summed up all the constructs used in this study.

An equal number of fresh (up to two weeks old) yeast colonies were transferred to 6 ml of minimal stringency liquid media and grown overnight at 250 rpm and 30°C. Yeast cells were subsequently pelleted, thoroughly washed with maximal stringency medium and resuspended in 1 ml of the appropriate liquid culture media at a fixed concentration as determined in pilot experiments (Additional file 2). Cultures were then grown at 250 rpm and 30°C for the indicated times. Growth of cell cultures was measured as the increase in 620 nm absorbance of 150 μl culture aliquots taken at initial (t = 0) and the indicated times. For the drug assays, drugs were added to the culture media at t = 0.

Experimental data were analyzed with the Prism™ GraphPad (version 4.01) software. Data were analyzed by the analysis of variance test (ANOVA) followed by the Tukey's multiple comparison test.