Quantitative MS‐Based Proteomics: Comparing the MCF‐7 Cellular Response to Hypoxia and a 2‐Oxoglutarate Analogue

Abstract The hypoxia‐inducible factors (HIFs) are key transcription factors in determining cellular responses involving alterations in protein levels in response to limited oxygen availability in animal cells. 2‐Oxoglutarate‐dependent oxygenases play key roles in regulating levels of HIF and its transcriptional activity. We describe MS‐based proteomics studies in which we compared the results of subjecting human breast cancer MCF‐7 cells to hypoxia or treating them with a cell‐penetrating derivative (dimethyl N‐oxalylglycine; DMOG) of the stable 2OG analogue N‐oxalylglycine. The proteomic results are consistent with reported transcriptomic analyses and support the proposed key roles of 2OG‐dependent HIF prolyl‐ and asparaginyl‐hydroxylases in the hypoxic response. Differences between the data sets for hypoxia and DMOG might reflect context‐dependent effects or HIF‐independent effects of DMOG.


Introduction
In response to limited oxygen availability,e ukaryotic cells adapt by alteringt he expression of multiple genes in ac ontext-dependent manner. [1] The a,b-hypoxia-inducible factors (HIFs) are key transcriptionf actorsf or the hypoxic response in animal cells. [2] Low oxygen concentrationsr educe the activity of the HIF prolyl hydroxylase domain-containingp roteins (PHD1-3) and factor inhibiting HIF (FIH), which are human 2oxoglutarate (2OG)-dependent oxygenases that suppress HIF's transcriptional activity by post-translational modification of HIF-a subunits. [3][4][5][6] As ar esult,t ranscriptionally active HIF levels increase, leading to the context-dependent upregulation of HIF targetg enes. The hypoxic response is also proposed to be important in many tumours,w hich are often hypoxic due to poor vascularisation. [7][8][9] Hypoxic tumour cells have been found to have increased metastatic potential and resistance to radiotherapy and chemotherapy. [10][11][12] An improved characterisation of gene regulation in response to hypoxia is of interest to further understand the fundamental biology of the hypoxic response, whichi nt urn might help to inform the development of drugs aimed at suppressing or enhancing levels of HIF. [13][14][15][16][17] Genome-wide transcription-profiling studies have been conductedt oc ompareh ypoxica nd normoxic cells using RNA sequencing and microarray analyses. [18][19][20][21] Thereare many mechanisms downstream of transcription that can affect protein concentrationsincells (such as splice variants,transcriptional regulation, post-translational modificationsa nd protein degradation), and it is well documented that protein abundance does not alwayscorrelate with RNA expression levels. [22][23][24][25][26] Therefore, studies to quantify changes in protein levels are important for ad etailed understanding of the cellular hypoxic response.
Several studies have suggested that there is ar esponse at the translational level to hypoxia, in addition to the transcriptional response mediated by HIF. [27][28][29][30][31] For example, hypoxia has been reported to suppressc ap-mediated translation by sequestering the translational initiation factor eIF4E. [23][24][25]27] An alternative translational initiation complex containing the hypoxically upregulated HIF-2a isoform has been identified, which selectively rescuesc ap-dependent protein synthesis in hypoxic cells. [30] It has also been proposed that, in addition to the HIF hydroxylases, other 2OG-oxygenases might play roles in regulating the hypoxic response, including at the translational level. [32] OGFOD1c atalyses hydroxylationo faproliner esidue in aribosomal protein (RPS23), and hasbeen reported to be associated with changes in translational accuracy. [32,33] It is antici-The hypoxia-inducible factors (HIFs) are key transcription factors in determining cellular responses involving alterations in protein levels in response to limited oxygen availability in animal cells. 2-Oxoglutarate-dependent oxygenases play key roles in regulatingl evels of HIF and itst ranscriptional activity. We describeM S-based proteomics studies in which we compared the results of subjecting human breast cancer MCF-7 cells to hypoxia or treating them with ac ell-penetrating deriv-ative (dimethyl N-oxalylglycine;D MOG) of the stable 2OG analogue N-oxalylglycine.T he proteomic results are consistent with reported transcriptomica nalyses ands upport the proposed key roles of 2OG-dependent HIF prolyl-and asparaginylhydroxylases in the hypoxic response.D ifferences between the data sets for hypoxia and DMOG might reflect context-dependent effects or HIF-independent effects of DMOG. pated that comparisons of globalR NA expression levels with protein levels in response to hypoxia might help reveal the mechanisms, other than transcriptional regulation, that constitute the hypoxic response.
The discovery that the human HIF-mediated hypoxic response is signalled for by the reduced activity of 2OG-oxygenases PHD1-3 andF IH has inspired the developmento fp harmacological methods for stimulatingt he HIF response. [14] Dimethyloxalylglycine (DMOG)i sac ell-permeable ester derivative of the near isosteric 2OG analogue N-oxalylglycine (NOG), which inhibits (to different extents) the HIF prolyl and asparaginyl hydroxylases along with other 2OG-oxygenases. [3,5,34,35] In ag lobal RNA analysis of DMOG-a nd hypoxia-treated cells, a good, but imperfect, overall correlation wasf ound between DMOG-a nd hypoxia-induced gene regulation. [18,19] However, some genes were found to be regulated by hypoxia, but not apparently by DMOG;i th as been suggested that this might be due to non-2OG-dependent oxygenase hypoxicr egulation. Comparison of protein responses to hypoxiaa nd DMOG could help to identify these regulation mechanisms, and highlight post-transcriptional differences between DMOG and hypoxic regulation. DMOG and other PHD inhibitors are also of interest for the treatment of ischemicd iseases,b ut the effects of the molecules beyond simulating the hypoxic response, for example, through the inhibition of other 2OG-oxygenases,a re still not established. [14] Comprehensive transcriptomic and proteomic studies on the effects of 2OG-oxygenases inhibitors will be of value to inform the design and use of therapeutics targeting the HIF pathway while minimising toxicity.S uch considerations would be particularly important in the context of the longterm use of HIF modulators for the treatment of chronic diseases (such as anaemia), the current target indications of PHD inhibitors in clinicaluse and development.
To date, the effects of hypoxia and DMOG on protein levels have been studied on smalls elections of proteins using western blotting, 2D gel electrophoresis and relatively low sensitivity mass spectrometry( MS)-based proteomics (< 1000p roteins quantified). [36][37][38][39][40] Recent developments in MS-based proteomic techniques allow the simultaneous quantification of thousands of proteins;h owever,t hesem ethods have not yet been applied to the hypoxic system.W ed escribe ag lobalM S-based proteomics study of the hypoxicr esponse in human breast cancer MCF-7 cells comparing the resultsw ith DMOG treatment. Proteinl evels in cells treated with normoxia, hypoxia, or DMOG were compared using triplex dimethyl labelling and LC-MS/MS analysis.

Results
Humanb reast cancer MCF-7 cells were cultured 1) under normoxia,2 )under hypoxia (0.5 %O 2 )o r3 )with inhibitor (1 mm DMOG)f or 16 h, in triplicate, before harvesting ( Figure 1). The nine cell pelletsw ere lysed in 8 m urea solution, then digested by treatment with dithiothreitol (DTT), iodoacetamide, then Lys-C and trypsin accordingt or eportedp rotocols. [41,42] Triplex dimethyl labellingo fs amples as light (normoxia), medium (hypoxia) and heavy (DMOG treated) was carried out "on column", as described. [43] Completion of labelling was confirmedb yL C-MS/MS analysis( 1h gradient, collision induced dissociation (CID)),w hich indicated > 99 %l abelling of peptides in all samples. The nine samples were divided into two, and the remaining steps were carried out in technical duplicates.L ight, mediuma nd heavy samples were mixed in ar atio of 1:1:1 based on base peak intensities in LC-MS/MS ion chromatogramst oa fford six light/medium/heavy mixed samples.T he six samples were fractionated by using an SCX-cartridge into seven fractions, as described in the Supporting Information. The resulting 6 7f ractions were analysed by using an Orbitrap Elite HybridI on Trap-Orbitrap mass spectrometer with a 4h LC gradient. Fragmentation of peptides was either by CID or electron transfer dissociation (ETD) according to ad ata-dependentd ecision tree (DDDT)). Analysis of the MS data was performed by using MaxQuant. The 42 LC-MS/MS runs were annotated in MaxQuant as six experiments,w ith seven fractions per experiment. Subsequent downstream analysisw as performed by using Perseus.
Analysis of the MS data using MaxQuant identified 38 048 peptidesa nd 4560 proteins. Proteins for which ar atio was de- Figure 1. An overviewo ft he protocol for the global comparison of proteinl evels in cells subjected to normoxia, hypoxia( 0.5 %O 2 ,16h), or treatedwith DMOG (1 mm,1 6h). Cellswerelysed, digested and dimethyl isotope labelled as light (normoxia), medium (hypoxia) or heavy (DMOG). Labelled digests were mixed and separated into sevenf ractions by using an SCX cartridge. Fractions were analysed by LC-MS/MS, and proteinr atios were calculated based on peptide ratios in the massspectra( MaxQuant). termined in less than three experiments were removed, to afford 3366, 3348 and 3365 proteins in the hypoxia/normoxia (M/L), DMOG/normoxia (H/L) andD MOG/hypoxia (H/M) datasets, respectively.A nalysis of the hypoxia/normoxiad ata set will be described first;t his represents the response to hypoxia, with the subsequent description of the DMOG/normoxia and DMOG/hypoxia data sets, reflecting the response to DMOG as compared to hypoxia.

Analysis of proteinlevels in hypoxia versus normoxia
Data from the 3366 quantified proteins were visualised using a volcanop lot of significance (Àlog[p value])v ersus ratio change (log[hypoxia/normoxia]) ( Figure 2). For the proteins quantified, the distribution of M/L ratios was found to be relatively narrow,w ith very few proteins changed by more than twofold (log 2 [ratio] > AE 1; Figure 2). The two proteins with the largest ratio change were NDRG1 (N-myc downstream-regulated gene 1p rotein;1 3-foldu pregulated in hypoxia) and BNIP3L (BCL2/adenovirus E1B 19 kDa protein-interacting protein 3; sixfold upregulated in hypoxia). These are both known HIF target genes andh ave previously been shown by western blotting analysist ob eu pregulatedi nh ypoxia. [44,45] Up-and downregulated proteins were selected by applicationo fa"one-sample test" to the six replicates (performed in Perseus, s 0 = 0.02, Benjamini-Hochberg FDR < 0.05). This analysisi dentified 163 upregulated and 154 downregulated proteins ( Figure 2, red points, Ta bles S6 and S7 in the Supporting Information).

Functionalannotation of proteins regulated by hypoxia
The potential functions of proteins that were identified as regulated by hypoxia were studied by analysiso ft heir gene ontology biological pathway( GO-BP) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) terms. Over-represented GO-BPa nd KEGG terms in the up-and downregulated protein sets were compared with the total quantified protein population by using the Fisher test (implemented in Perseus, Benjamini-Hochberg FDR > 0.02). Consistentw ith prior transcriptomic analyses, [46,47] amongst the upregulated proteins, the KEGG pathways "glycolysis" and "fructose and mannose metabolism" were found to be significantly enriched by nine-and six-fold, respectively (Tables 1a nd S1). The most over-represented GO biological pathway terms also related to glycolysis and catabolic processes.T his observation is consistentw ith prior work [46,47] and highlights an enhancement of glycolytic pathways in hypoxicc onditions as the cell shifts toward less efficient nonoxidativef ormso fc arbon metabolism and ATPp roduction, that is, anaerobic glycolysis. [48,49] Examples of upregulated glycolytic enzymes include fructose biphosphate aldolases Aa nd C( ALDOA and ALDOC), phosphoglycerate kinase (PGK1) and phosphoglycerate mutase 1( PGAM1). Lactate dehydrogenase A( LDHA) is also upregulated to divert pyruvate from oxidative breakdown in the mitochondria. [49] Among the downregulated proteins,t he KEGG term "ribosome"w as significantly over-represented, possibly reflecting a reduction in protein synthesis under hypoxicc onditions to conserve ATP( Ta bles2and S2);t his is consistent with the pro- posal that translation is generally suppressed in hypoxia. [50] Indeed, 26 of the 154 significantly downregulated proteins were ribosomal (e.g.,R PL24, RPL29,R PL36A,R PL37A,R PS25). The most significantly over-represented GO biological pathway term related to all stages of translation and the transport of proteins to the endoplasmic reticulum.

Comparison of hypoxia protein regulation with RNA-sequencing results
An analogousm RNA sequencing study to our proteomic study has been conducted to compare the transcriptional response of MCF-7 cells treated under the same hypoxic and normoxic conditions. [18] The mRNA levels of 13 351 genes were studied; 565 and 166 transcripts were observed to be significantly upand downregulated, respectively,a fter additional filtering to exclude low-expressed transcripts under normoxic condition (fragments per kilobase of transcripts per million mapped reads,F PKM ! 2). The MS-based proteomics study reported here quantified protein levels for 72 of the 565 transcripts that were upregulated and 38 of the 166 transcripts that were downregulated at the mRNA level (Figure 3, red points). There are correlations among the 72 genes found to be upregulated at the mRNA level, with 68 %o ft hem also being upregulated at the protein level (Àlog[p value] > 2, Figure 3A). There was poorer agreement for the 38 genes that weref ound to be downregulated at the mRNA level, with only 26 %also downre-  Table 3. Genes found to be significantly upregulated in am icroarray analysis of RNA levels (FDR < 0.05) [19] that werea lso quantifiedi no ur LC-MS/MSa nalyses.
Names Proteinratio RNA ratio [19] NamesP roteinratio RNA ratio [19] OGDH À1.  Figure 3B). The most striking difference in the regulation at the mRNA and protein levels was forA TP1B1 and RPL7, which we found to be significantly upregulated at the mRNA level but significantly downregulated at the protein level, thus suggesting possible regulation of them at the translational or protein levels. [18]

Analysis of proteinlevels in DMOG versusnormoxia
Previous RNA studies have indicated that the broad-spectrum 2OG-oxygenase inhibitor DMOG( which also inhibits other enzyme types, e.g.,i socitrate dehydrogenase) [57] is an effective mimic of hypoxiai na nimal cells. [19] An umber of transcripts, however,w eref ound to be regulated by hypoxia, but not by DMOG;t his might result from non-2OG-oxygenase-mediated hypoxicr egulation (or reflect relativelyp oor inhibition of a 2OG-oxygenase by DMOG). [19,58] The efficacy of DMOG in mimicking hypoxia at the protein level was studied to further investigate non-2OG-oxygenase-dependent regulatory pathways. Comparison of the histogramso fr atios obtained in the hypoxia/normoxia (M/L) and DMOG/normoxia (H/L) datas ets indicated that, under the tested conditions, proteins were more perturbed in hypoxia than after treatment with DMOG ( Figure 4, standard deviation:M /L 0.28, H/L 0.21). This could be due to factors including reduced induction of HIF by DMOG (1 mm)t han by hypoxia (0.5 %O 2 ), or due to mechanisms of hypoxicr egulation that are not induced by DMOG,o racombination of these. Previously,i mmunoblotting studies indicated that HIF levels induced by DMOG were lower than those induced by hypoxia under the tested conditions. [18] Proteins significantly up-andd ownregulated in response to DMOG were identified by analysiso ft he DMOG versus normoxia (H/L) data set by using the statistical "one sample test" in Perseus( s 0 = 0.01, Benjamini-Hochberg FDR < 0.05). Of the 3348 proteins that were quantified in at least three repeats, 40 were found to be upregulateda nd 39 were downregulated ( Figure S1, Tables S8 and S9). This was markedlyf ewer than were found to be regulated in the hypoxia data set.
Analysis of GO and KEGG terms in the upregulatedp roteins (Perseus, Fisher test with significantly regulated categorical column, Benjamini-Hochberg FDR < 0.02) afforded very similar results to those obtained from the hypoxiadata set with glycolytic processes being over-represented ( Table 5). Analysis of the downregulated proteins by this methodd id not identifya ny over-represented terms;t his probably reflects the relatively small number of proteins that were annotateda ss ignificantly downregulated. Al ess stringent analysisw as conducted to identify GO terms that correlatedw ith high Àlog 10 [p value] among the downregulated proteins (Perseus, Fisher test with Àlog 10 [p value] numerical column, threshold Àlog 10 [p value] > 2.5, Benjamini-Hochberg FDR < 0.02). This analysis provided very similarr esultst ot he analysiso ft he proteins downregulated in hypoxia, with GO and KEGG terms involving ribosome and translational processes over-represented (Table6).

Comparison of regulation by hypoxiaand DMOG
Substantial overlap was found for proteins that were significantly regulated by DMOG or hypoxia, with 75 %( 30/40) of the DMOG-upregulated and 49 %( 19/39) of the DMOG-downregulated genes also up-/downregulated in hypoxia. The extent of correlation between the two conditions was analysed by direct comparison of the ratio changes. The ratios of proteins that were found to be significantly up-and downregulated by hypoxia and/orD MOGw ere plotted in as catterp lot, with a lower significance threshold {Àlog 10 [p value] > 2.5} ( Figure 5, significantly regulated by:h ypoxia = red, DMOG = blue, both = green).M ost proteinsw ere found to be regulated in the same direction by either DMOG or hypoxia (the upper-right and lower-left quadrants), and many of these found to be significant in both data sets (green points). The proteins that were found to be regulated in opposite directions were only significant in one data set (blue and red points), witht he exception of CLIC3 (chloride intracellular channel protein 3), which was found to be statistically significantly upregulated by hypoxia and downregulated by DMOG (Àlog 10 [p value] > 2.5).

Identification of proteins that are differentially regulated by DMOG or hypoxia
Although most proteins were regulated in the same direction in response to DMOG and hypoxia (relative to normoxia), some proteins weref ound to be differentially regulated. To further investigate this, the H/M ratios werea nalysed to identify significantlyd ifferentially regulated proteins (Perseus, "one sample test", s 0 = 0.01, Benjamini-Hochberg FDR < 0.05, FigureS2, Tables S10 and S11). This analysis found 36 proteins to be significantly more abundant in the DMOG-treated cells and 67 proteins to be significantly more abundant in the hypoxia-treated cells.
Analysis of GO and KEGG terms in the two differentially regulated protein lists by using Perseus did not identify any significantly over-represented pathways.Al ess stringent analysis with p value correlation (Perseus, Fisher test based on the numerical column Àlog[p value],t hreshold Àlog[p value] > 2.5) also did not identify any significant terms among the proteins more abundant in hypoxia. Notably,t he proteins more abundant in DMOG-treated cells, however,h ad over-represented GO and KEGG terms for the ribosome and translational processes (Table 7).

Discussion
Our MS-based proteomics study identified 4560 proteins and quantified differences in the abundance of 3366 proteins under hypoxicv ersus normoxic conditions in MCF-7 human breastc ancerc ells. Of the quantified proteins, 163 and 154 were significantly up-and downregulated by hypoxia, respectively.I ts hould be noted that these lists are not exhaustive, as only approximately aq uarter of the predicted proteomew as quantified, and many proteins that were not quantified might be regulated by hypoxia, including regulatory proteins (e.g., transcription factors), which are often present at low abundance. The majority of protein levels were found to change by less than af actor of 2; this is consistentw ith the ratio changes observedi no ther global proteomic studies of hypoxia by using MS or 2D gel electrophoresis. [36][37][38] Immunoblotting studies have reported that some proteins undergo much larger fold changes in hypoxia,f or example, HIF,P HD3 and CA9; [19] however, these proteins were not detected in any of the MS analyses reported here, possibly because they are too low in abundance even after hypoxia-induced upregulation. Over-represented GO and KEGG terms for the proteins found to be upregulated by hypoxiai nclude those related to glycolytic pathways, an observation that is consistent with the well-established cellular enhancement of glycolytic processes in response to al ow-energy environments uch as hypoxia. [59] Notably,a mongst the significantly downregulated proteins,G O and KEGG terms relatingt ot ranslation and the ribosome were over-represented.T his probably reflects ar eduction in protein synthesis during hypoxiai no rder to preserve energy resources (ATP). [60,61] Many of the upregulation assignments were consistent with previousl iterature studies on mRNA and protein levels in response to hypoxia. Several proteins found to be significantly regulated by hypoxiah ad not previously been reported to be, and these assignments could provide ag ood startingp oint for biological studies to develop ad etailed understanding of the hypoxicr esponse at the protein level.
The hypoxia-induced protein regulation was compared with two studies on the mRNA response to hypoxia, one using microarraya nalysis [19] and the other using RNA sequencing. [18] In both comparisons, there was good agreement for the upregulated assignments,b ut very little overlap in the downregulated assignments. Poor correlation of RNA and protein regulation in response to stimuli, particularly amongd ownregulated genes, has been reported in other studies. [52][53][54][55][56] This raises the question of to what extent the downregulation of ribosomal proteins under hypoxic conditions is mediated by protein translation or degradation mechanisms rather than by the general reductionint ranscription that occursi nh ypoxia.
The most striking difference in regulation at the RNA and protein levelsw as for ATP1B1 (sodium/potassium-transporting ATPase subunit beta-1) and RPL7 (ribosomal protein L7), which were assigned as being significantly upregulated at the RNA level but significantly downregulateda tt he protein level. RPL7 had also been found to be downregulated at the protein level by hypoxia in Caenorhabditise legans, [62] and ATP1B1 has been reported to be downregulated by hypoxia at the protein and RNA levels in human retinal pigmented epithelial cells. [63] Further studies are required to validate the changes in protein level observed here;i fc orrect, these observations could offer insight into the post-transcriptional mechanismso fh ypoxic regulation. ATP1B1 is aN a + /K + -transporting ATPase that is required for maintaining an ormalp olarised epithelial phenotype. [63,64] Decreased ATPase function is associated with epithelial-to-mesenchymalt ransition, which is essential to numerous developmental processes, and in the initiation of metastasis in cancer progression. [65,66] Hypoxic regulation of ATP1B1 could have implications in cancerp athogenesis.
Protein regulation in response to DMOG was found generally to resemble the response to hypoxia, with substantial overlap between proteins found to be significantly up-/downregulated in the two data sets. The log[ratio] changes in response to hypoxiaw ere generally larger in magnitude than for DMOG under the tested conditions. Of course,t he degree of concord between the data sets might vary with changes in the extent of hypoxia or DMOG concentration,o ri nd ifferent cellular contexts. Immunoblotting studies of the same treatment conditions have indicated that the induction of HIF by DMOG (1 mm)w as less pronounced than in hypoxia (0.5 %O 2 ), which could cause the observed smaller upregulation of HIF target genes by DMOG. [13] Additionally,t here could be alternative non-2OG-oxygenase-mediated responses to hypoxia that are not activated by DMOG. Some proteins were apparently increased by DMOG treatment but not by hypoxia, potentially reflecting 2OG-oxygenase-mediated responses not linked to HIF (e.g.,i nhibition of JmjC KDMs or ribosome-modifying 2OGoxygenases). However,o ur focused observations should not be taken as evidencet hat the regulation of these proteins is HIF independent, as expression of HIFt arget genes is likely to be limitedb yo ther factors in ac ontext-dependentm anner.N ote also that DMOG inhibits enzymes other than 2OG-oxygenases, for example, isocitrate dehydrogenase, [57] inhibition of which will affect 2OG levels.A nalysis of the proteins that were most differentially regulated by hypoxiaa nd DMOG suggests that DMOG is effective at mimickingt he enhancement of glycolytic processes by hypoxia, but possibly less effective in suppressing translational processes. Thish as also been observed at the RNA level in microarray studies. [19] The differences between them RNA-a nd protein-level regulation implied by the combined studies highlight am inimal requirementf or both proteomic and transcriptomic studies to ensurea na dequate understanding of the hypoxiae ffects on gene expression. It should be noted that globalM Sa nd RNA studies are affected by noise, and any observed differential regulation should be confirmed by alternative techniques including detailed cellular and biochemical studies (in different cell types) before being considered to be validated. It is anticipated that the data sets anda nalyses reported here could support the initiation of further studies to elucidate the mechanisms and effects of the hypoxicresponse in different contexts.
In addition, application of this MS-based proteomics approach to the study of selective 2OG-oxygenase inhibitors (e.g.,s elective inhibitors of PHD1-3, FIH or OGFOD1) could help further elucidatet he role of particular 2OG-oxygenases in mediating the hypoxic response.