Asparagine hydroxylation is a reversible post-translational modification

Amino acid hydroxylation is a common post-translational modification, which generally regulates protein interactions or adds a functional group that can be further modified. Such hydroxylation is currently considered irreversible, necessitating the degradation and re-synthesis of the entire protein to reset the modification. Here we present evidence that the cellular machinery can reverse FIH-mediated asparagine hydroxylation on intact proteins. These data suggest that asparagine hydroxylation is a flexible and dynamic post-translational modification akin to modifications involved in regulating signalling networks, such as phosphorylation, methylation and ubiquitylation.


Introduction
Post-translational modifications (PTMs) are chemical alterations of amino acids or proteins which increase the complexity of the proteome and allow the cell to modify protein function in a dynamic or sustained manner (1)(2)(3)(4). Some of these modifications are irreversible, such as the cleavage of a polypeptide chain resulting in altered activity of the protein. Nevertheless, most PTMs are reversible and comprise the dynamic addition or elimination of functional groups that range from large oligosaccharide chains to a few atoms, such as glycosylation, phosphorylation, acetylation, methylation, or carboxylation.
The reversibility of these PTMs is frequently achieved by the action of pairs of enzyme classes with opposing functions, one of which catalyses the forward reaction and another the reverse reaction (5)(6)(7). In cases where one of the reactions is thermodynamically unfavourable, the reaction may not be straightforwardly reversible and therefore includes intermediate products. An example would be the formation and dissolution of cysteine bonds, where the formation of the bond can include the oxidation of the sulphur of cysteine to sulfenic acid, which then forms a disulphide bond by reacting with another cysteine (8,9) Oxidation or more precisely hydroxylation of proteins on residues other than cysteine was recognised as a PTM in the 1960s, when the enzymatic hydroxylation of proline and lysine was identified as taking place during collagen synthesis (10,11). Subsequently, the molecular mechanism of hydroxylation catalysed by the evolutionarily conserved family of the 2-Oxoglutarate dependent dioxygenases (2OG-ox) was described (12,13). The 2OG-ox are a family of proteins composed of over 60 enzymes of which the so-called HIF-hydroxylases form a subgroup consisting of four hydroxylases: 3 prolyl hydroxylases (PHD1,2 and 3) and an asparaginyl hydroxylase "factor inhibiting HIF" (FIH). HIFhydroxylases act as sensors of the oxygen levels within the cells. Under normoxic conditions prolyl hydroxylases (PHD1, 2 and 3) catalyse the specific hydroxylation of two proline residues in the alpha subunit of HIF1, the master regulator of the hypoxic response. Once hydroxylated, HIF1α is bound by the von Hippel-Lindau ubiquitin ligase (VHL) complex, which promotes the ubiquitination and rapid proteasomal degradation of HIF1α, resulting in the ablation of the protein under normoxic conditions (14)(15)(16)(17)(18). FIH, the other component of the subfamily, catalyses an asparagine hydroxylation in the Cterminus of HIF1α. Upon hydroxylation of this residue, the interaction with co-factors required for the formation of the active transcription factor complex is impeded, resulting in the downregulation of HIF-driven transcription in the presence of oxygen (19,20).
Over the past few years, it has been argued that HIF1α is not the only protein that is hydroxylated by HIF-hydroxylases and several additional substrates, particularly of FIH, have been postulated and validated (21). In a similar manner to that observed in the context of HIFα, hydroxylation of these other substrates by FIH alters the physicochemical properties of the hydroxylated domains. These changes can induce or destroy proteinprotein interactions that ultimately control substrate activity, folding or localisation (22)(23)(24)(25)(26).
Presently, it has been suggested that the hydroxylation mediated by HIF-hydroxylases is an irreversible process, and that the hydroxylation can only be reset by the degradation and new synthesis of a non-hydroxylated protein (27)(28)(29). A mass spectrometric study monitoring two FIH-mediated hydroxylation sites of Rabankyrin-5 substantiated this, as no evidence of dehydroxylation was found under the investigated conditions (30).
Nevertheless, some authors have proposed the existence of a dehydroxylation mechanism (25,31), but this notion is only founded on the assumption that PTMs should be generally reversible. Some data have emerged that implicitly suggest that asparagine hydroxylation may be a reversible PTM after all. First, in contrast to proline hydroxylation of HIF, FIH-mediated asparagine hydroxylation does not lead to the rapid decrease of protein half-life. This suggests that FIH-substrates are longer-lived proteins and that the cell would therefore benefit from a mechanism of resetting the level of hydroxylation in a more dynamic, non-destructive manner (21). One such example is transient receptor potential vanilloid 3 (TRPV3), an ion channel that is hydroxylated by FIH. Hydroxylation on an asparagine residue reduces TRPV3-mediated current, whereas hypoxia, FIH inhibition or mutation of the asparagine residue potentiates it without affecting protein stability. Intriguingly, the increases in current through the channel are observable in less than one hour of hypoxia or FIH inhibition. This rapid response indicates that there has to be a very rapid turnover of the protein or that the hydroxylation can be reversed without destruction and re-synthesis of TRPV3 (25). Moreover, indirect evidence from mathematical modelling indicates that signalling networks require reversibility of asparagine hydroxylation. Nguyen et al published a comprehensive ODE-based mathematical model of the immediate HIF network (32). Intriguingly, the mathematical model assumes irreversibility of both proline hydroxylations but includes an undefined reaction that leads to the dehydroxylation of the N-terminal asparagine residue (Fig. 1A).
The reversibility of the asparagine hydroxylation was required for the model to reproduce the experimental data, which showed the transient induction of HIF protein levels and transcriptional output. Upon removal of the unspecified dehydroxylation reaction, the DMOG was obtained from Cayman Chemical (71210), DFO and cycloheximide were purchased from SIGMA (D9533-C4859). For immunoprecipitation anti-Flag-M2 beads (Sigma Aldrich), anti-myc beads (Cell signalling-9B11) and anti-GFP (GFP-Trap Magnetic Agarose-Chromotek) were used.
Total lysates were fractionated by SDS-PAGE and transferred onto nitrocellulose filters.
XIC were generated with a width of (646.6980-646.7140)/(652.0300-653.0440) for the for non-hydroxylated/hydroxylated peptide. Peptide elution times were calibrated using hydroxylated/non-hydroxylated standards. Calculation of hydroxylation occupancy: To calculate the relative molar ionisation efficiency we devised a method that assumes a peptide exists predominantly in two molecular states, hydroxylated and unmodified. By determining the relative abundance of either species with respect to a reference intensity we estimated the relative ionisation efficiency for either unmodified or hydroxylated peptides. To determine the relative ionisation efficiencies of unmodified peptides, we generated a list of TNKS2/HIF1a/TRPV3 peptides, respectively, detectable in light and heavy labelled samples, excluded M-containing peptides (and N-containing peptides in the case of TNKS2) and summed up the intensities in the heavy channel of latter timepoint. We then divided the intensity of the unmodified containing heavy peptide by this sum to generate the relative intensity of the non-hydroxylated peptide over a reference intensity. If the hydroxylation appeared not to be stochiometric, such as in TRPV3, we included an additional step. We calculated the ratios of the non-hydroxylated peptide in heavy conditions. This gave us the ratio of the relative ionisation efficiency of the unmodified peptide. We repeated this step for the light labelled, unmodified peptide at time point 0.
The difference in the relative ionisation efficiency of the unmodified peptides allows calculating the difference of occupancy between the DMOG treated sample and the control allowing us to calculate the % of occupancy of the unmodified peptide. Knowing this occupancy then allows to calculate the occupancy of the hydroxylated peptide and the relative ionisation intensity by multiplying the occupancy with the ratio of the light labelled hydroxylated peptide over the sum of peptides. These calculations generated a ratio of molecular ionisation efficiencies of unmodified/hydroxylated peptide. We used this to transform the ratio modified/unmodified as calculate by MaxQuant into molar ratios.
The molar ratios were converted into estimated occupancies by dividing the ratio by itself +1; using the Perseus software (36). In cases where we were unable to calculate the ratio of molar ionisation efficiency, we estimated it to be 1.

Experimental Design and Statistical Rationale Statistical analysis:
Overall, 150 biological replicate samples were analysed. 8  Samples sequences for mass spectrometry analysis were randomised using the excel =RAND() command. Statistical tests were performed using Graphpad Prism 8 (multiple t tests), distribution was assumed to be normal. Protein intensities and hydroxylation occupancies are shown with error bars representing standard error of mean (SEM).

Database Search Parameters and Acceptance Criteria for Identifications
The mass spectrometry raw data was analysed by the MaxQuant and Andromeda (1.6.10.43) software package (37) using the pre-selected conditions for analysis (specific proteases, 2 missed cleavages, 7 amino acids minimum length). Protease was set to trypsin or trypsin+GluC, for the TNKS2/TRPV3 or HIF1a pulldowns, respectively. In silico modelling: To in silico predict the effects that putative dehydroxylation may have on dynamic behaviour of the HIF-1α signalling pathway, we modified a well-calibrated mathematical model of the HIF pathway which we developed previously (32), by removing the steps representing asparagine dehydroxylation in this model (Fig.1A). This was done by setting the kinetic parameters describing the rate of these reactions to null in the model's ordinary differential equations. Simulations of the intact and the adjusted models under various conditions (i.e. hydroxylase inhibition by DMOG and JNK, and 1% and 3% hypoxia, Fig. 1B-E) show that removal of the asparagine dehydroxylation steps failed to reproduce the experimental patterns of HIF-1α expression.

Results
There are several analytical techniques that can be applied to the quantitation of changes in PTMs, such as the use of modification-specific antibodies or the monitoring of shifts in the apparent molecular weight in PAGE (38,39). Unfortunately, neither of the abovementioned methods is universal and for this reason, we decided to use quantitative mass spectrometry (qMS) to investigate whether hydroxylation is a reversible process. qMS has been shown to be superior to western blotting in terms of relative quantification and is a method widely used for monitoring changes in the hydroxylation status of proteins (14,22,23,30,39). Initially, we planned to monitor the dynamics of asparagine hydroxylation on the well-described FIH-substrate Tankyrase 2 (TNKS2).
Tankyrases are members of the poly(ADP-ribose)polymerase (PARP) protein super family, which participate in the regulation of the degradation complex in the Wnt/β-catenin signalling pathway (40)(41)(42). TNKS2 has been characterized as a FIH substrate that contains the FIH consensus motif [LXXXXXV/IN] in several ankyrin repeat domains (ARD) (21,43). As such, TNKS2 contains multiple FIH-hydroxylation sites, and furthermore hydroxylation on TNKS2 by FIH has not been reported to have any effect on its stability.
Multiple hydroxylations would allow us to monitor several sites in the same experiment, thus improving our chances of detecting if one or more hydroxylation(s) are reduced over time with statistical significance.
As an experimental approach we adapted the pulsed stable isotope labelling with amino acids in cell culture (SILAC), that has been successfully used to measure protein and PTM turnover (44,45). HEK 293T cells were transfected with a V5-tagged FIH and a Flag-tagged TNKS2. 24 hours after transfection the medium was replaced with SILAC medium (containing 13

TNKS2 protein turnover is not altered by hydroxylation
To determine whether SILAC media and DMOG affected TNKS2 protein stability, we first validated by western blotting that the expression of TNKS2 or FIH was not affected by either treatment at any time point (Fig. 2C). Furthermore, to estimate TNKS2 protein turnover we blocked protein synthesis with cycloheximide, a rapidly acting eukaryotic protein synthesis inhibitor (47,48) commonly used for protein turnover studies (49). We initially monitored how CHX inhibited protein synthesis by western blotting and observed a reduction in the levels of Flag-TNKS2 and V5-FIH, with an approximate half-life of 6 hours for TNKS2 (Fig. 2D). We next determined how TNKS2 hydroxylation levels change dynamically upon SILAC and DMOG pulsing. We immunoprecipitated Flag TNKS2, digested the protein on-beads, analysed the peptides by LC-MS/MS and quantified TNKS2 expression using MaxQuant. This confirmed our initial observation that DMOG did not affect the stability of TNKs when compared to the DMSO control (Fig. 2E) and also allowed us to monitor the incorporation of heavy SILAC, by detection of the increase in heavy-labelled TNKS2 (Fig. 2F). Based on the labelling data, we calculated the TNKS2 protein half-life to be around 7 hours. This matched the Western-blot-based estimation, suggesting that newly synthesised TNKS2 protein is predominantly heavy-labelled upon pulsing with heavy amino acids and that DMOG treatment does not affect the expression and stability of exogenously expressed TNKS2.

TNKS2 N-hydroxylations are reversible
Having confirmed this, we began using this set-up to determine if hydroxylated proteins could be dehydroxylated in cells. Initially, we analysed the ion chromatogram of the lightlabelled TNKS2 peptides containing hydroxylated and non-hydroxylated asparagine 586, and detected a reduction in the intensity of the hydroxylated peptide associated with DMOG treatment, whilst at the same time we detected an increase of the absolute levels of the unmodified peptide (Fig. 2G), suggesting that hydroxylated residue can be reverted.
To increase certainty, we increased the number of experimental repeats to six and analysed several hydroxylation sites in an automated manner. We selected the peptides that we identified with a localization-specificity for hydroxylated asparagine of greater than 0.8 and where both the hydroxylated and non-hydroxylates isoform were detected in at least four of the six replicates. The hydroxylation occupancy of the peptide was estimated by first correcting the intensity ratio of the modified over unmodified peptide by the ratio of molar ionisation efficiencies (Table 1, Experimental Procedures). Subsequently we transformed the ratio of the hydroxylated over non-hydroxylated peptide into occupancies Using this automated analysis method, we monitored how the occupancy of several lightlabelled hydroxylated peptides changed over time (Fig. 2H, Fig. S2, Table S1). Six sites showed a statistical significant decrease in hydroxylation occupancy relative to a negative control. The occupancy of some sites was only altered by a few percentage points, whereas others by up to 50%, suggesting a gradient of reversibility.
To rule out the possibility that the pulsing with heavy amino acid did not comprehensively abrogate protein synthesis of light-labelled protein, we repeated the experiment using CHX as an additional control. We combined CHX with a heavy SILAC and DMOG/DMSO pulse in order to abrogate TNKS2 synthesis during the time course of DMOG treatment.
Heavy-SILAC pulsing further allowed us to determine how efficiently CHX abrogated protein synthesis. We initially monitored how CHX inhibited protein synthesis by LC-MS and observed a near complete abrogation of heavy-labelled TNKS2 production when compared to a control (Fig. 3A). Overall, TNKS2 synthesis was inhibited by over 98% in the presence of CHX and can therefore be considered residual. Once this was confirmed, we immunoprecipitated Flag TNKS2 from HEK293T cells post pulse and analysed the hydroxylated by LC-MS/MS as above. When monitoring how the relative occupancy of the light-labelled hydroxylation changed upon pulsing with DMOG, we were able to observe a significant reduction of several asparagine residues (Fig. 3B-D experiments. An alternative explanation is that the dehyxdroxylase or an essential cofactor is rapidly degraded upon CHX treatment. Should the enzyme/s catalysing the dehydroxylation reaction have a short half-life, this would result in partial dehydroxylation during the initial time points of the experiment, followed by a plateau. This is a profile not dissimilar to what we have observed (Fig. 3B-D).
To rule out a non-specific side-effect of DMOG treatment, we utilised a second, structurally and functionally unrelated inhibitor, deferroxamine (DFO), which inhibits FIH by chelating the iron in the active centre (50,51). Firstly, as we did previously for DMOG, we checked by western blot and mass spectrometry that DFO did not affect the stability of the protein (Fig. 3E) and the incorporation of heavy SILAC (Fig. 3F, 3G). Secondly, we quantified the occupancy for two previously characterised sites. As previously, we observed a reduction in the hydroxylated/non-hydroxylated ratio upon DFO treatment in the Light SILAC samples (Fig. 3H, 3I, Table S3).

Rapid reversal of TRPV3 N-hydroxylation
To determine whether dehydroxylation takes place on other substrates we decided to investigate the reversibility of a hydroxylation site on TRPV3. As mentioned in the introduction, TRPV3 is hydroxylated by FIH on asparagine 242 (25) and the hydroxylation regulates TRPV3 activity without affecting its expression. Interestingly, TRPV3 activity responds rapidly to hydroxylase inhibition, suggesting that dehydroxylation could be taking place. Replicating our experimental setup in HEK 293T cells and over-expressing myc-tagged TRPV3, we initially confirmed that inhibition of FIH by DMOG and incubation with SILAC media did not affect TRPV3 protein expression. By western blotting we confirmed previous observations that TRPV3 protein stability was not affected by hydroxylase inhibition (Fig. 4A). We then immunoprecipitated myc-TRPV3, digested with trypsin and analysed resulting peptides by mass spectrometry. Overall intensity of the light-labelled TRPV3 was altered over 2 hours and we detected less than 20% incorporation of the heavy label (Fig. 4B), suggesting that TRPV3 is a protein with a long half-life. In addition, we readily detected the reported hydroxylation of N242. We could further determine that for light-labelled N242 the hydroxylation occupancy was rapidly reduced over time upon DMOG treatment (Fig. 4C, Fig. S3A, Table S4). Together, these data suggest that the hydroxylation of N242 in TRPV3 is reversible.

HIF N803 hydroxylation is reversible
Finally, we checked if our finding could be extrapolated to the best-studied substrate of FIH, the alpha subunit of HIF1α (19,20). HEK293T were transfected with a GFP tagged HIF1α plasmid in order to overexpress the protein. As previously, we checked if the total levels of GFP HIF1α were affected by DMOG or heavy SILAC treatments. By western blotting we surprisingly detected that HIF1α expression appeared to be stable under normoxic conditions and that DMOG only marginally, if at all, increased HIF1α protein levels (Fig. 5A). We repeated the same experiment and using mass spectrometry we observed a reduction in the levels of light-labelled HIF1α, at the same time that we detected an induction of the heavy-labelled GFP HIF1α (Fig. 5B). Finally, we studied the oxidation levels of N803 (the asparagine residue hydroxylated by FIH) and we observed that the ratio of oxidation of N803 was rapidly reduced upon DMOG treatment (Fig. 5C,   Fig. S3B, Table S5). As we expressed full-length HIF1α we decided to additionally monitor the hydroxylation of both reported proline sites. Whereas we identified the unmodified peptides with ease and over 50 MS/MS, we were not able to detect a single MS/MS identifying the 402 and 564 proline hydroxylation sites with confidence. We have therefore concluded that overexpressing HIF1α overwhelms the capacity of the endogenous PHD1-3 enzymes to hydroxylate the proline residues stoichiometrically, delivering an explanation as to why we can detect HIF protein expression in normoxia. Overall, these results match the data that we obtained for TNKS2 and TRPV3, suggesting the presence of an asparagine dehydroxylation reaction. To determine whether other modifications can occur on the hydroxylated peptide we repeated the data analysis including the dependentpeptide matching option, which allows for the unbiased identification of modified derivatives of "base" peptides. The algorithm detected several modifications localised to N803, one of which was the elimination of two hydrogens (Fig. S4). We could only detect this modification on light-labelled peptides, suggesting that presence of DMOG reduces the amount of the precursor, indicating that didehydrogenated N803 is derived from hydroxylated N803. This would be consistent with an intermediate product in the potential dehydroxylation reaction where water is eliminated from the hydroxylated side-chain.
Nevertheless, this is speculation as the abundance of didehydrogenated peptide is miniscule preventing us from reliably quantifying it across the samples.
To determine if we could observe asparagine dehydroxylation in a cell-free reaction, we generated hydroxylated protein by incubating the recombinant C-terminal transactivation domain of HIF2α (CAD) with recombinant FIH. To prevent re-hydroxylation by FIH, hydroxylated CAD was re-purified and subsequently incubated with cellular lysates derived from FIH-knockout HeLa cells (35). After 3 hours the CAD was affinity purified, digested and analysed using mass spectrometry. The ratio of non-hydroxylated peptide over hydroxylated peptide was approximately double that of the control sample that was immediately purified upon mixing with the lysate (Fig. 5D-E). Importantly, we observed an increase in non-hydroxylated peptide. Although the overall amount is small, the ratio increased to less than 1/100 of the hydroxylated intensity, this is not unexpected given the supraphysiological levels of hydroxylated peptide that were incubated with the HeLa cellular lysates.
Taking together, the data obtained from three different FIH substrates supports the hypothesis that asparagine hydroxylation is a post-translational modification that can be reversed within cells, with additional evidence that the hydroxylation can be reversed in vitro.

Discussion
Protein functions can be switched on and off by distinct PTMs, such as phosphorylation, glycosylation and others (1)(2)(3)(4), which allows cells and organisms to respond dynamically to changes in the environment. The postulated reversibility of hydroxylation could explain how acute hypoxia and reoxygenation elicits rapid responses irrespective of protein degradation. This may be especially important in tumours as cyclical/intermittent hypoxia and protein hydroxylation are cancer hallmarks (52,53).
We have not yet identified the enzyme/s responsible for the postulated reaction. We initially took a candidate approach and knocked-down proteins that were commonly binding to several N-hydroxylated proteins. However, none of these perturbations had an effect on the base-line hydroxylation levels, which suggests that the best option would be to devise a genome-wide screen and using N-hydroxylated-dependent protein-protein interactions or antibodies as a readout. Identification of the enzyme responsible would also allow design of a tailored in vitro assay that could reveal parameters and molecular mechanisms of the reaction, the precise nature of which we can currently only speculate on. Unexpectedly, the experiments we conducted in presence of CHX may have narrowed the search area. It is attractive to speculate that the dehydroxylase is a shortlived protein, as this would enable rapid and dynamic regulation of its activity. Rapid degradation would also bestow the system with the ability to adapt to the cellular needs dynamically. It is also plausible that  Table 1 Gene