Identification of proteins and cellular pathways targeted by 2-nitroimidazole hypoxic cytotoxins

Tumour hypoxia negatively impacts therapy outcomes and continues to be a major unsolved clinical problem. Nitroimidazoles are hypoxia selective compounds that become entrapped in hypoxic cells by forming drug-protein adducts. They are widely used as hypoxia diagnostics and have also shown promise as hypoxia-directed therapeutics. However, little is known about the protein targets of nitroimidazoles and the resulting effects of their modification on cancer cells. Here, we report the synthesis and applications of azidoazomycin arabinofuranoside (N3-AZA), a novel click-chemistry compatible 2-nitroimidazole, designed to facilitate (a) the LC-MS/MS-based proteomic analysis of 2-nitroimidazole targeted proteins in FaDu head and neck cancer cells, and (b) rapid and efficient labelling of hypoxic cells and tissues. Bioinformatic analysis revealed that many of the 62 target proteins we identified participate in key canonical pathways including glycolysis and HIF1A signaling that play critical roles in the cellular response to hypoxia. Critical cellular proteins such as the glycolytic enzyme glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and the detoxification enzyme glutathione S-transferase P (GSTP1) appeared as top hits, and N3-AZA adduct formation significantly reduced their enzymatic activities only under hypoxia. Therefore, GAPDH, GSTP1 and other proteins reported here may represent candidate targets to further enhance the potential for nitroimidazole-based cancer therapeutics.


Introduction
Rapid growth of cancer cells, coupled with chaotic tumour vasculature, creates regions within the tumour where cells are deprived of nutrients and oxygen [1]. This phenomenon, clinically termed "tumour hypoxia", has been reported in almost all solid malignancies. Cells residing in these hypoxic niches up-regulate a myriad of pro-survival adaptive responses including impaired drug delivery [2], oncogene activation [3], abnormal cellular metabolism [4] and deregulated DNA damage checkpoint signalling [5]. Furthermore, low O 2 tension stimulates metastasis, invasion and relapse [6], and sustains growth and maintenance of tumour initiating cells [7]. Most importantly, these adaptations make hypoxic cancer cells resistant to radiation, chemotherapy and immunotherapy, resulting in poor patient prognosis [8,9]. While this rationalizes the need for hypoxia directed therapy, it has yet to become a part of routine clinical practice.
Nitroimidazoles (NIs) are a class of hypoxia activated prodrugs that are bioreductively activated in an inverse O 2 -dependent manner. The process involves reduction by nitroreductases that generates nitroradical anions; this intermediate is re-oxidized in the presence of O 2 , allowing the drug to freely diffuse out of well-oxygenated cells. Under hypoxia, the nitro-radical anion undergoes reductive decay producing a reactive hydroxylamine, which covalently binds to cellular macromolecules, leading to hypoxia-selective accumulation of the drug [10]. This unique property makes NIs ideal for targeting hypoxic tumour cells, both as cytotoxins and radiosensitizers. However, despite showing early promise, several NIs failed in clinical trials as hypoxia directed cancer therapeutics [11,12]. These disappointing outcomes can be attributed, in part, to our limited understanding of the precise molecular mechanism of NIs. Indeed, the protein targets of NIs in a human cancer context have never been explored, making it difficult to comprehend their effects on target proteins. Here, we outline a strategy to label and isolate NI target proteins using a novel click chemistry compatible 2-NI agent, azidoazomycin arabinofuranoside (N 3 -AZA). The proteins were then characterized using liquid chromatography coupled mass spectrometry (LC-MS/MS). Bioinformatics analysis revealed that many of these proteins play critical roles in the cellular response to hypoxia.

Materials and methods
General synthesis methods: Solvents used in reactions were purified before use by successive passage through columns of alumina and copper under an argon atmosphere. All reagents were purchased from commercial sources and were used without further purification unless noted otherwise. All reactions were carried out under a positive pressure argon atmosphere and monitored by thin-layer chromatography (TLC) on Silica Gel G-25 UV254 (0.25 mm) unless stated otherwise. TLC spots were detected under UV light and/or by charring with a solution of anisaldehyde in EtOH, acetic acid and H 2 SO 4 . Column chromatography was performed on Silica Gel 60 (40-60 mm). Organic solutions were concentrated under vacuum at <50 • C. 1 H and 13 C NMR spectra were recorded at 400 and 101 MHz, respectively. 1 H and 13 C NMR chemical shifts are referenced to CD 3 OD (d = 3.35 and 4.78 for 1H, 48.9 for 13C). 1 H NMR data are reported as though they are first order and the peak assignments were made on the basis of 2D-NMR ( 1 H-1 H COSY and HMQC) experiments. ESI-MS spectra were obtained on samples suspended in CH 3 OH with added NaCl. Synthesis of N 3 -AZA was explored from two precursors, namely 1α-D-(5-O-tosyl-arabinofuranosyl)-2-nitroimidazole (Ts-AZA) [17] [See supporting materials] and 1-α-D-(5-deoxy-5-iodo-arabinofuranosyl) -2-nitroimidazole (IAZA) [18]. The IAZA route described here showed better yield and accommodated the increase in the reaction temperature to 100 • C without producing detectable side products. Briefly, NaN 3 (0.33 g, 10.2 mmol) was added to a solution of IAZA (1.8 g, 5.1 mmol) in DMF (15 mL) and the reaction mixture was heated at 100 • C for 2 h. After cooling to room temperature, the reaction mixture was concentrated under vacuum and quenched with H 2 O (20 mL). The product was extracted into EtOAc (3 × 10 mL) and the organic layer was dried over anhydrous Na 2 SO 4 , filtered and the solvent was removed under reduced pressure. The crude product was purified using a short column chromatography μL of methanol and the optical density (OD) was measured at 584 nm using a FLUOstar OPTIMA microplate reader (BMG Labtech, Ortenberg, Germany). Percent cell survival was calculated by subtracting OD of blank wells and then normalizing DMSO controls to 100%. The first point on each curve represents 0 μM drug.

Click chemistry reaction cocktail preparation:
The Click-IT reaction cocktail (Molecular Probes, Eugene, OR) was made up according to the manufacturer's directions using a 1:500 dilution of 10 mg/mL biotin-alkyne (Cat. #C37B0, Lumiprobe, Hallandale Beach, FL) for proteomic studies or with a 1:5000 dilution of the 2 mg/mL Alexafluor 488-, 555-or 594-conjugated alkyne stock (A10267, A20013, A10275, Molecular Probes) for imaging. Reaction cocktail was used within 10 min of preparation. All reactions were carried out at room temperature in the dark.

Isolation of N 3 -AZA-protein adducts: FaDu cells treated with 100
μM N 3 -AZA (or vehicle control, 0.02% DMSO) for 24 h under normoxia or hypoxia (<0.1% O 2 ) were harvested directly in RIPA buffer supplemented with protease inhibitor. Click chemistry was performed on cell lysates at room temperature for 30 min using a biotin-labelled alkyne. Streptavidin-mutein beads were resuspended and processed as per the manufacturer's guidelines (Cat. # Roche 03708152001, Sigma-Aldrich). Clicked lysates (without equilibration buffer) were loaded on 1% BSA blocked streptavidin-mutein beads and put on a rocker overnight at 4 • C. Afterwards, the supernatant was removed, the beads were washed several times, and the bound biotinylated protein fraction was either eluted from the beads using free biotin (Sigma-Aldrich) or by boiling the beads in SDS-PAGE loading dye. LC-MS/MS: Clicked crude extracts and eluates were run on SDS-PAGE, stained with Coomassie G-250 and processed for in-gel trypsin digestion. Briefly, excised gel bands were de-stained twice in 100 mM ammonium bicarbonate/acetonitrile (50:50 v/v). The samples were then reduced in 10 mM 2-mercaptoethanol in 100 mM sodium bicarbonate and alkylated by treatment with 55 mM iodoacetamide in 100 mM bicarbonate. After dehydration, enough trypsin (6 ng/μL, Promega sequencing grade, Madison, WI) was added to just cover the gel pieces and the digestion proceeded overnight (~16 h) at room temperature. Tryptic peptides were first extracted from the gel using 97% water/2% acetonitrile/1% formic acid, followed by a second extraction using 50% of the first extraction buffer with 50% acetonitrile. Fractions containing tryptic peptides were resolved and ionized using a nanoflow HPLC (Easy-nLC II, Thermo Scientific, Waltham, MA) coupled to an LTQ XL-Orbitrap hybrid mass spectrometer (Thermo Scientific). Nanoflow chromatography and electrospray ionization were accomplished using a PicμoFrit fused silica capillary column (ProteoPepII, C18) with 100 μm inner diameter (300 Å, 5 μm, New Objective, Woburn, MA). Peptide mixtures were injected onto the column at a flow rate of 3000 nL/min and resolved at 500 nL/min using a 60 min linear gradient from 0 to 45% v/v aqueous acetonitrile in 0.2% v/v formic acid. The mass spectrometer was operated in data-dependent acquisition mode, recording highaccuracy and high-resolution survey Orbitrap spectra using external mass calibration, with a resolution of 30,000 and m/z range of 400-2000. The fourteen most intense multiply charged ions were sequentially fragmented by using collision induced dissociation, and spectra of their fragments were recorded in the linear ion trap; after two fragmentations, all precursors selected for dissociation were dynamically excluded for 60 s. Data was processed using Proteome Discoverer 1.4 (Thermo Scientific) and a human proteome database (UniProt) was searched using SEQUEST (Thermo Scientific). Search parameters included a precursor mass tolerance of 10 ppm and a fragment mass tolerance of 0.8 Da. Peptides were searched with carbamidomethyl cysteine as a static modification, and oxidized methionine and deamidated glutamine and asparagine as dynamic modifications. Protein IDs were manually analyzed, and unreviewed UniProtKB/TrEMBL IDs were substituted with respective reviewed UniProtKB/Swiss-Prot IDs for downstream data analysis.
Bioinformatics analysis: Proteins identified through LC-MS/MS were plotted as a Venn diagram using the FunRich network analysis tool [20]. An enrichment profile was generated by comparing the relative ion intensities of individual proteins in lysates and eluates from N 3 -AZA treated hypoxic cells, and plotted using GraphPad Prism V7. Canonical protein sequences of putative N 3 -AZA target proteins were retrieved from the UniProt database and analyzed for cysteine (Cys) and proline (Pro) content. Functional categorization of target proteins was performed using gene ontology by PANTHER classification system (biological processes, molecular activity, and cellular components) [21]. The proteomic data set was submitted into Ingenuity Pathway Analysis (IPA) for core analysis [22]. IPA calculates a distinct p-value (p-value of overlap calculated using Fisher's Exact Test) which depicts the association between the protein molecules in our dataset to the literature-reported molecules in IPA knowledgebase. The identified canonical pathways were ranked based on P value. Upstream regulatory analysis was performed to identify the N 3 -AZA target proteins regulated by common regulators. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) activity assay: The activity of cellular GAPDH enzyme was quantified using a GAPDH activity assay kit (Abcam, ab204732). Briefly, FaDu cells grown on 60-mm glass petri dishes were treated with either N 3 -AZA (100 μM or 250 μM) or vehicle control (0.02% DMSO) for 24 h under normoxia or hypoxia (<0.1% O 2 ). Afterwards, attached cells were harvested by trypsinization, counted, pelleted by centrifugation and dissolved in GAPDH assay buffer (1 × 10 6 cells per 0.5 mL buffer); 25 μL cell lysate was used to measure GAPDH enzymatic activity using 96 well plates according to the manufacturer's protocol. Absorbance was measured at 450 nm in kinetic mode with a FLUOstar OPTIMA microplate reader; a total of 30 readings were recorded over 48.3 min. GAPDH activity was calculated according to the manufacturer's guidelines.
Tumour generation and drug treatments: Female BALB/c nude mice (Charles River Laboratories, Wilmington, MA) aged 6 weeks were used and cared for with protocols approved by the Cross Cancer Institute Animal Care Committee (protocol #AC14208) in accord with the Canadian Council on Animal Care guidelines. We subcutaneously injected 6 × 10 6 FaDu cells in 0.1 mL of sterile 0.9% NaCl into the upper right flank to form palpable tumours. The tumours were measured daily until they reached desired volume, when the mice were divided into 4 groups and injected intraperitoneally (i.p.) with 0.15 mL 0.9% NaCl containing (i) 80 mg/kg N 3 -AZA, (ii) 80 mg/kg pimonidazole, (iii) total 80 mg/kg N 3 -AZA: pimonidazole in a 1:1 ratio and (iv) 0.9% NaCl control. We also used a primary mouse model of HPV-negative squamous cell carcinoma of the oral cavity, developed by Dr. Yvonne Mowery [23]. These animal studies were performed in accordance with protocols approved by the Duke University Institutional Animal Care and Use Committee (IACUC, protocol #A032-19-02) and adhered to the NIH Guide for the Care and Use of Laboratory Animals. In brief, tumours were generated by breeding KRT5-CreER T2 mice (tamoxifen-inducible Cre in basal epithelial cells) with Trp53 fl/fl ; Rb fl/fl or p53 fl/fl ; Ink4a/Arf fl/fl mice, and applying 4-hydroxytamoxifen and the carcinogen, benzo[a]pyrene, topically to oral mucosa of these genetically engineered mice. Mice containing varying sizes of tumours were administered i.p. with 80 mg/kg N 3 -AZA: pimonidazole in a 1:1 ratio. 2 h after injection, mice were sacrificed, tumours and organs (brain, lung, liver, kidney and cervical lymph nodes) removed and frozen in optimal cutting temperature compound (O.C.T, Fisher Healthcare 4585). Serial sections (7-10 μm each) from tumours and organs were taken, dried and stored at -80 • C.
Tissue staining: Frozen sections were air dried for 25 min, fixed in acetone for 5 min precooled to -20 • C, air dried 25 min and blocked in PBS with 1% BSA for 20 min. Sections were incubated with primary antibodies for small vessel endothelium (rat anti-CD31, 1:50 dilution; 550274, BD Pharmingen) and hypoxia (rabbit anti-pimonidazole, 1:100 dilution; Hypoxyprobe) for 2 h at room temperature, followed by three 5-min washes in PBS. The secondary antibody (Alexafluor 647 goat antirabbit, 1:1000 dilution; Molecular Probes), with 4',6-diamidino-2-phenylindole (DAPI) to stain DNA, was incubated for 30 min at 37 • C in the dark, followed by three 5-min washes with PBS. The Click-iT reaction cocktail (containing Alexafluor 555-conjugated alkyne) was incubated with the sections for 30 min. Sections were washed 5 times with PBS for 5 min and mounted using Mowiol 4-88 (Millipore, Etobicoke, ON) mounting media. Slides were stored at 4 • C until imaged. Images were obtained by tile scan with a Leica SP8 STED microscope.
Hypoxia fraction calculation: An estimate of hypoxic tumour fraction was calculated with MATLAB R2018b (MATLAB and Image Processing Toolbox Release 2018b, The MathWorks, Inc., Natick, Massachusetts, United States) using a modified protocol described previously [24]. To account for background staining on the click channel, a manual threshold was determined from a negative control (click staining performed on tumour section from saline injected mice) and was applied for the N 3 -AZA stained slides. The Otsu threshold was chosen for DAPI [25]. Images were converted to binary according to their respective threshold and the hypoxic fraction estimate was calculated using following formula:

Hypoxic fraction
( HF ) = Area of pixels stained with N3 − AZA Area of pixels stained with DAPI × 100 Data processing and statistics: Immunoblots and microscopic images were processed with Adobe Photoshop to adjust for brightness and contrast, to add scale bars, and for rearrangement and labelling. Gel scans were quantified using Image Studio lite v5.2 (LI-COR) and normalized to vehicle treated normoxia controls. Microscopic images were processed with IMARIS software (Bitplane, Zürich, Switzerland) to quantify area and channel intensity of cell, nucleus and cytoplasm. GraphPad Prism V7 (GraphPad Software, La Jolla, CA) was used to generate graphs and to perform statistical analysis. Non-liner regression analysis was performed on crystal violet staining data to obtain IC 50 values. Graphs display the mean with standard error of the mean (S.E. M.). Statistical analysis was performed using 2-tailed unpaired t-test, with p < 0.05 considered statistically significant. GAPDH activity was analyzed using Dunnett's 2-way ANOVA. Asterisks depict statistically significant differences: ns (not significant), * (P ≤ 0.05), ** (P ≤ 0.01), *** (P < 0.001), **** (P < 0.0001). Fig. 1A depicts the synthesis of the target compound N 3 -AZA from two precursors namely Ts-AZA [17] and IAZA [18]. Since the latter route provided better scalability and no detectable side products, the reaction was carried out on a multigram scale with a yield of 93%. Atomic numbering and NMR analysis of N 3 -AZA is shown in Fig. S1. N 3 -AZA dissolves in DMSO and remains active, even in solution, for months when stored at -20 • C. N 3 -AZA is selectively cytotoxic to hypoxic cells: The hypoxia selectivity of N 3 -AZA was first validated to determine doses that would limit the utility of the compound. A crystal violet staining (CVS) assay was used to determine the cytotoxicity of the drug under normoxia and hypoxia [26]. In all four human cancer cell lines tested, N 3 -AZA was more toxic under hypoxia than normoxia (Fig. 1B-E). Cytotoxicity of N 3 -AZA in FaDu cells was directly compared to that of pimonidazole. For N 3 -AZA, hypoxic FaDu cells had an IC 50 value > 5 fold lower than under normoxia (normoxic and hypoxic IC 50 values for N 3 -AZA were ~1463 μM and ~271 μM, respectively). Importantly, the compound showed almost no toxicity at concentrations ≤100 μM, even under hypoxia ( Fig. 1B). A similar toxicity profile was seen with pimonidazole, with an IC 50 of ~179 μM under hypoxia and ~3035 μM under normoxia ( Fig. S8).

Synthesis of N 3 -AZA:
Identification of proteins that interact with 2-NI under hypoxia: In hypoxic cells, activated (bioreduced) 2-NIs bind to cellular proteins and form drug-protein adducts. To identify these protein targets, normoxic and hypoxic FaDu cells were treated with N 3 -AZA (or 0.02% DMSO vehicle), lysed, and click chemistry was performed on protein lysates with a biotin labelled alkyne. "Clicked" lysates were processed for western blotting and probed using horse radish peroxidaseconjugated streptavidin (streptavidin-HRP), which showed that "clicked" proteins were present only in N 3 -AZA treated hypoxic cells. These proteins were then successfully isolated with streptavidin-mutein beads (Fig. 1F). Immunoblotting confirmed low background binding of non-biotinylated proteins to the column matrix (Fig. 1G). The recovered proteins were identified by mass spectrometry, and bioinformatic analysis was carried out for functional characterization. From the combined proteomic dataset of eluates obtained from three independent experiments, a total of 65 proteins were identified with a peptidespectrum match (PSM) score of ≥2 PSM, of which 5 were in "Normoxia + DMSO" eluate (Table S1), 13 were in "Normoxia + N 3 -AZA" eluate (Table S2), 10 were in "Hypoxia + DMSO" eluate (Table S3) and 62 were in "Hypoxia + N 3 -AZA" eluate (Table 1). Overall, 48 proteins were exclusively identified in the "Hypoxia + N 3 -AZA" eluate ( Fig. 2A). All proteins identified in the eluates of cells treated with N 3 -AZA (normoxic and hypoxic) are shown with their average PSM scores (Fig. 2B). Enrichment analysis between crude extracts and eluates from N 3 -AZA-treated hypoxic cells revealed that mostly high abundance proteins reacted with the drug. However, several highly abundant proteins also escaped labelling by N 3 -AZA (e.g., acyl-CoA dehydrogenase family member 9, ACAD9). Likewise, several relatively low-abundant proteins, such as ferritin heavy chain, desmoplakin and calpastatin were readily modified by N 3 -AZA (Fig. 2C). Activated 2-NI compounds are known to attack cellular nucleophiles to form adducts. Cysteine (Cys) has often been proposed as the site of attack on proteins [27]. Hence, N 3 -AZA target protein sequences were analyzed to assess any association with their Cys content. Surprisingly, 5 of these proteins did not have any Cys residues whereas 34 proteins had 1-5 Cys, 17 had 6-10 Cys, 3 had 11-15 Cys, and only 3 proteins had >20 Cys residues (Fig. S2A). This prompted analysis of the dataset for the next most nucleophilic amino acid, proline (Pro) [28]. Interestingly, all but one (tropomyosin alpha-3 chain, TPM3) of the N 3 -AZA target proteins contained proline residues, with almost 60% of them containing >15 Pro Western blotting showed that the signal for Streptavidin-HRP is only present in drug treated hypoxic samples. Drug bound proteins could successfully be isolated using streptavidin-mutein beads, with no significant background binding. Representative immunoblots are displayed from three independent experiments. residues (Fig. S2B). TPM3, on the other hand, contains 1 Cys residue.
The 62 putative target proteins of N 3 -AZA were categorized using PANTHER classification system (cellular component, molecular function and biological process) (Fig. S2C). Cellular component analysis showed that N 3 -AZA target proteins are mostly cytoplasmic (44 out of 62 proteins; ~71%), followed by 13 nuclear proteins (~21%), 4 plasma membrane proteins and 1 with an annotated extracellular matrix localization. According to molecular function, most of the proteins belong to the subcategories of binding (33), catalytic activity (24) and structural activity (12). Biological process-related proteins mainly belong to the subcategories of cellular process, metabolic process, localization, biological regulation and response to stimulus. IPA identified literature reported canonical pathways these proteins are involved in, and the top 10 pathways (based on the P value) are listed ( Table 2). The dataset was also analyzed to assess if these target proteins are under the regulation of common upstream regulators. Two clusters, pertinent to hypoxic response, are reported here: (i) cluster 1 regulated by heat shock transcription factor 1, HSF1 (8 proteins) (Fig. 2D) and (ii) cluster 2 regulated by hypoxia inducible factor 1 subunit alpha, HIF1A (8 proteins) (Fig. 2E). N 3 -AZA treatment did not affect the levels and localization of target proteins, but reduced GAPDH and GSTP1 enzymatic activity under hypoxia: To assess the effects of N 3 -AZA treatment on its target proteins, the following proteins were studied: actin, tubulin, heat shock protein 90 (Hsp-90), glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and glutathione S-transferase P (GSTP1); HIF1A was used as an indicator of successful hypoxia induction. Hypoxic cells treated with 250 μM N 3 -AZA appeared more compact, with fewer projections, however no significant difference was observed in target protein levels nor their cellular localization following N 3 -AZA treatment (Fig. 3A and B;   (Fig. 3C). Total GST activity was determined using an assay that measures the conjugation of 1-chloro-2,4-dinitrobenzene (CDNB) with reduced glutathione, resulting in an increase in absorbance at 340 nm. The rate of increase is directly proportional to the GST activity in the sample. While hypoxic exposure markedly increased GST activity in vehicle treated cells, in agreement with others [29], N 3 -AZA treated hypoxic cells showed a significant dose-dependent reduction in GST activity (Fig. 3D). We estimate a minimum of 40 ± 13% GAPDH and 57 ± 5% GSTP1 protein was modified by cell exposure to 100 μM N 3 -AZA (Figs. S4-5).

Sub-cellular localization of N 3 -AZA:
Click chemistry between the azido moiety of N 3 -AZA and a fluorophore-conjugated alkyne enabled us to image the sub-cellular localization of N 3 -AZA adducts formed under hypoxia using fluorescence microscopy. Fluorescent staining was observed only in O 2 -starved cells when treated with N 3 -AZA for 6 h (Fig. 4A-D); incubation duration was determined from a separate time course experiment (Fig. S6A). Furthermore, the compound was entrapped in FaDu cells in an inverse oxygen-dependent manner, with the highest retention seen in cells cultured under <0.1% O 2 . Normoxic cells showed minimal to almost no click staining. Staining of N 3 -AZA under hypoxia increased in a dose dependent manner, with 100 μM concentration demonstrating the most vivid differences (Fig. 4D, Fig. S6B). N 3 -AZA-treated hypoxic cells showed strong nuclear staining (Fig. S6C) with a nucleolar uptake pattern; the latter was confirmed by immunostaining for nucleolin (Fig. 4E, Movie 1). A similar hypoxia selective N 3 -AZA staining pattern was confirmed in additional cancer cell lines (Fig. S6D).
Supplementary video related to this article can be found at https:// doi.org/10.1016/j.redox.2021.101905 To assess the metabolic stability of N 3 -AZA-protein adducts upon reoxygenation, FaDu cells were treated with 100 μM N 3 -AZA for 6 h under hypoxia and then reoxygenated, fixed at the indicated time points and processed for N 3 -AZA click staining; N 3 -AZA was present in the medium throughout the reoxygenation period. N 3 -AZA click fluorescent signal could be detected up to 24 h following reoxygenation, suggesting that N 3 -AZA modification is relatively stable. Although the intensity weakened by about 40%, the decrease in intensity appeared to be a consequence of cell doubling (data not shown). Minimal background was seen in their normoxic counterparts even after prolonged incubation with the compound (Fig. S7).
To confirm the hypoxia selectivity of N 3 -AZA click staining, cells treated simultaneously with pimonidazole and N 3 -AZA were processed for pimonidazole immunostaining followed by N 3 -AZA click chemistry. Both N 3 -AZA click staining and pimonidazole immunostaining were detected only in drug treated hypoxic cells, with minimal signal in normoxic samples (Fig. 5A). Both staining strategies were equally efficient at differentiating between normoxic and hypoxic cells, however, N 3 -AZA click staining proved to be simpler and faster compared to pimonidazole immunostaining and produced better signal to background ratio (Fig. 5B). Interestingly, although both compounds contain 2-NI moieties, they showed perceptibly different cellular staining patterns. N 3 -AZA generated a strong nucleolar signal in hypoxic cells (Movie 1), whereas pimonidazole staining remained excluded from nucleoli (Movie 2).
Supplementary video related to this article can be found at https:// doi.org/10.1016/j.redox.2021.101905 N 3 -AZA click chemistry for hypoxia mapping in tumours: Since N 3 -AZA click chemistry could efficiently stain hypoxic cells, its capacity to detect hypoxic regions in tumour tissue was examined in comparison to pimonidazole. Initially, a FaDu cell subcutaneous xenograft model was used, which has previously been shown to have a hypoxic fraction of 11-12% [30]. Regions of tumour hypoxia defined by the click chemistry reaction were present in areas bordering necrosis and were independent of the vasculature (Fig. 5C). Pimonidazole immunostaining generated a similar staining pattern (Fig. 5D). Pimonidazole and N 3 -AZA staining co-localized to areas bordering necrotic regions as expected (Fig. 5E). These findings were then validated in mouse primary head and neck tumours (Fig. 5F). For both tumour models, a low level of non-specific staining of tumour sections by the secondary antibodies used for pimonidazole and CD31 was observed; N 3 -AZA click staining provided a clearer background. We further quantified the hypoxic fraction in tumours with different volumes to determine if there is a relationship between tumour size and hypoxia. Our data (Fig. S10A) did not indicate a size dependence, which implies that small tumours may retain a hypoxic fraction that is resistant to therapy. Importantly, significant click staining was not seen in any of the harvested organs, which suggests that N 3 -AZA preferentially accumulates in hypoxic tumour niches (Figs. S10B-G).

Discussion
The tumour microenvironment has garnered considerable attention in recent times for its role in therapy outcome [31]. Hypoxia, a key feature of the tumour microenvironment, offers both a challenge and a vulnerability that can be exploited for therapy. However, very few strategies have yielded successful results when it comes to targeting hypoxic tumours. Despite several setbacks, NIs still hold significant value as hypoxia directed therapeutics. For example, nimorazole (a 5-NI), in combination with radiotherapy, showed promise in treating advanced head and neck cancer patients [32]. Interestingly, meta analysis of NI clinical trials showed an overall improvement in patient outcome, particularly when stratified depending on tumour oxygenation status [33]. Therefore, an in depth understanding of NI molecular mechanism is crucial to better guide future development of NI based cancer therapeutics.
NIs have long been known to bind to nucleic acids and proteins in hypoxic cells [34]. While early NI studies almost exclusively concentrated on their potential to induce DNA damage [35][36][37], NI protein targets remained relatively unexplored. A handful of proteomic studies do exist, albeit mostly limited to reaction of 5-NIs in protozoans [38][39][40]. Recently, a click chemistry based method to identify candidate 5-NI drug targets in Giardia lamblia using a modified metronidazole (metronidazole alkyne) and an azido-biotin was outlined [41]. The current study shows that N 3 -AZA click chemistry with a biotin-alkyne is compatible with mammalian cell-based assays to efficiently biotinylate proteins that interact with activated 2-NIs under hypoxia (Fig. 1G). These proteins were then isolated, identified by mass spectrometry and characterized for functional significance with respect to molecular targeting of hypoxia. We report a total of 62 protein targets of N 3 -AZA, 48 of which were not present in our control samples ( Fig. 2A, Table 1). For proteins that appeared ubiquitously across the panel, their abundance was generally multiple-fold higher in the N 3 -AZA treated hypoxic sample (based on PSM scores, Fig. 2B). To assess if N 3 -AZA is binding to a subset of proteins, a relative enrichment profile was generated by comparing crude extracts and eluates from N 3 -AZA treated hypoxic cells (Fig. 2C). Overall, N 3 -AZA-protein interaction appeared to be dictated by the abundance of the protein, with a few exceptions; several highly abundant proteins (such as ACAD9) did not react with N 3 -AZA, while proteins like desmoplakin and calpastatin, despite being relatively low in abundance, appeared as positive hits. This suggests that there might be some degree of selectivity for NI-protein interaction. In contrast to previous reports proposing thiol groups as a major site for NI attack [27], no clear association was found between the Cys content of a protein and its likelihood of binding to N 3 -AZA, which is consistent with the findings described by Ref. [41]. For example, despite having 11 Cys residues, ACAD9 escaped N 3 -AZA binding while 5 of the N 3 -AZA target proteins (histone H1.2, prohibitin, ATP synthase subunit beta, histone H4 and activated RNA polymerase II transcriptional coactivator p15) lack any Cys residues (Fig. S2A). While this observation could be a result of Cys residues being buried within the protein structure or forming cysteine-cysteine disulfide bonds, it could also imply that other nucleophilic amino acids, such as Pro, may serve as the site of adduct formation for NI compounds. Indeed, the current analysis showed that most of the N 3 -AZA target proteins are enriched for Pro residues (Fig. S2B). Unfortunately, without the exact chemical formula of the N 3 -AZA-amino acid derivatives we were unable to identify potential conjugates by MS. Analysis of the proteomic data was performed using several bioinformatics platforms. N 3 -AZA target proteins appeared to be predominantly cytoplasmic (~71%) with only ~21% being nuclear proteins (Fig. S2C). Upstream regulatory analysis with IPA identified two regulatory clusters with direct implications to hypoxic response. Cluster 1 (regulated by HSF1) has 8 downstream targets, 5 of which are molecular chaperones directly involved in adaptations to low O 2 levels and protect against reoxygenation-induced oxidative stress (HSP90AA1, HSP90AB1, HSPA1A, HSPA8 and HSPD1) (Fig. 2D) [42,43]. Of particular interest to us was cluster 2, consisting of 8 target proteins of N 3 -AZA, which are under the regulation of HIF1A. HIF1A, known as the master regulator of hypoxia, plays a key role in cellular response to hypoxic insults [44]. Hypoxia induced stabilization of HIF1A results in upregulation of pro-survival genes, including those involved in glucose metabolism. 7 out of the 8 N 3 -AZA target proteins in cluster 2 are implicated in carbohydrate metabolism (Fig. 2E). Whether or not binding of N 3 -AZA (and other 2-nitoimidazole compounds) to these downstream targets can counteract the actions of HIF1A remains to be explored.
Interestingly, from our list of 62 target proteins (Table 1), canonical pathway analysis identified glycolysis I as one of the top hits with a P value of 8.05E-09; five of the N 3 -AZA target proteins belong to this pathway (ALDOA, ENO1, GAPDH, PKM and TPI1) and are regulated by HIF1A (Table 2, Fig. 2E). It is important to note that reprogramming of glucose metabolism through upregulation of glycolysis is a crucial hallmark of tumourigenesis. Despite being less efficient than the mitochondrial respiration system, cancer cells readily utilize aerobic glycolysis for ATP production, mostly to counteract mitochondrial defects or to compensate for the defective mitochondrial oxidative phosphorylation under hypoxia. Increased glycolysis also provides an acidic microenvironment through elevated production of lactate, and thus facilitates malignant progression [45]. GAPDH was identified as the top target for N 3 -AZA with an average PSM score of 18.3. It is a Cys-directed glycolytic enzyme that catalyzes the reversible conversion of glyceraldehyde-3-phosphate (GAP) to 1,3-bisphosphoglycerate (BPG). GAPDH transcription has been reported to be upregulated in hypoxic cells in a HIF1A dependent manner [46,47], however, such regulation appears to be cell-type specific [48]. In addition to its role in glycolysis, GAPDH has also been implicated in diverse physiological and pathophysiological processes [45,49,50]. This makes GAPDH an interesting target to follow-up on. Three aspects of N 3 -AZA treatment on GAPDH were explored: total protein level, cellular distribution and enzymatic activity. Total GAPDH protein levels remained unaffected in response to N 3 -AZA treatment, both under normoxia and hypoxia. (Fig. 3A). NIs (such as PA-824) can induce nitrosative stress by acting as an NO donor, which can lead to oxidization/S-nitrosylation of GAPDH and its translocation to the nucleus [51]. Immunocytochemical analysis showed that N 3 -AZA treatment did not alter GAPDH cellular distribution (Fig. 3B). The highly conserved active site of GAPDH contains a Cys residue, which is readily acetylated during the reversible oxidative phosphorylation of GAP [52], is essential for its role in oxidative stress response [53][54][55], and could potentially serve as a binding site to 2-NIs. Importantly, ornidazole (a 5-NI) has been reported to inhibit GAPDH activity in mouse spermatocytes [56], protozoa and anaerobic bacteria [57], and depletion of GAPDH has also been reported to induce cytostatic effects on human cancer cell lines [58]. This prompted us to explore the effects of N 3 -AZA treatment on GAPDH enzymatic activity. GAPDH activity did not differ significantly between the normoxic-treatment groups, whereas upon hypoxic incubation, its enzymatic activity was reduced by >30% in cells treated with 250 μM N 3 -AZA as compared to vehicle control (Fig. 3C).
Another key protein target identified in our analysis was GST family protein GSTP1 (Table 1), which is the predominant non-hepatic isozyme that eliminates toxic by-products of oxidative stress and xenobiotics by catalyzing their conjugation with glutathione (GSH) [59]. GSTP1 has also been implicated in the mitogen-activated protein (MAP) kinase pathway through protein: protein interactions with c-Jun N-terminal kinase 1 (JNK1) and apoptosis signal-regulating kinase (ASK1), which are activated in response to cellular stress [60]. While GSTs can prevent tumour formation by protecting cellular proteins and DNA from endogenous reactive compounds, their detoxification properties may aid in multi-drug resistance in tumour cells. Indeed, tumours often express high levels of GST (particularly GSTP1), and GSTs are frequently overexpressed in multidrug-resistant cells [61]. Additionally, hypoxic exposure is often accompanied by an increase in GST activity to cope with the altered intracellular redox potential and the generation of increased cellular reactive oxygen species (ROS) [29]. Interestingly, inhibition of GSTP1 reduces cell viability and induces vulnerability to both intracellular oxidative stress and hypoxia/reoxygenation stress [62]. Nitroimidazoles, such as etanidazole and misonidazole, have previously been reported to inhibit GST activity [63,64]. Our analysis is in agreement with these findings, and show a significant reduction in GST specific activity only in N 3 -AZA treated hypoxic cells. Compared to vehicle treated cells, we observed >50% reduction in hypoxic cells treated with 100 μM N 3 -AZA and >75% reduction in hypoxic cells treated with 250 μM N 3 -AZA (Fig. 3D). The inhibitory effect of the drug does not seem to be due to proteolysis since no significant change in GSTP1 protein levels was observed in response to drug treatment or hypoxic exposure (Fig. 3A). While our data does not preclude toxicity arising from DNA damage [35], given the dynamic roles of GAPDH and GSTP1 in the hypoxic tumour microenvironment, such an effect of N 3 -AZA on their activities may partially explain the observed hypoxic cytotoxicity of the drug.
The list of NI-targeted proteins (Table 1) includes additional proteins that demand further attention, for example, peroxiredoxin-1 (Prdx1). Similar to other members of this protein family, Prdx1 active site contains two Cys residues that are essential for its catalytic function and could serve as potential binding sites for thiol-reacting NIs [65]. It is therefore tempting to postulate that binding of N 3 -AZA to Prdx1 through its catalytic Cys residues may exert a negative impact on its enzymatic activity. Prdx1 plays a key role in cellular detoxification, functions as a chaperone, has anti-apoptotic property, is upregulated in cancer and contributes to radioresistance [66,67]. Interestingly, Prdx1 is upregulated under hypoxia possibly to counteract the detrimental effects of hypoxia/reoxygenation mediated accumulation of reactive oxygen species and to provide aggressive survival advantage for malignant progression [66,68]. Hence, N 3 -AZA binding and subsequent reduction in enzymatic activity of Prdx1 could provide an additional mechanism for its hypoxic cytotoxicity. Further studies will be required to test this hypothesis. Interestingly, a related protein, thioredoxin reductase (TrxR), has previously been reported as a common target for 5-NIs across three microorganisms [38][39][40].
A unique feature of N 3 -AZA click chemistry is its versatility since it can also be used as a hypoxia marker. N 3-AZA shares structural homology with 18 F-FAZA, which itself has proven to be superior in comparison to other 2-NI hypoxia radiotracers [69]. 2-NIs are widely used for immunohistological hypoxia diagnosis, aided by the development of antibodies against pimonidazole-and EF-5-protein adducts [15,16]. N 3 -AZA click chemistry offers a simpler alternative by exploiting the Cu (I) catalyzed click chemistry instead to visualize cellular hypoxia. While the 2-NI moiety of N 3 -AZA ensures that it is selectively forming drug-protein adducts in hypoxic cells, the azido group can be exploited to fluorescently label these adducts using a fluorescently tagged alkyne molecule in the presence of a Cu(I) catalyst. A similar strategy for fluorescent labelling of hypoxic cells using an alkyne conjugated NI, SN33267, has previously been reported in a patent and a MSc. thesis [70,71]. N 3 -AZA click fluorescence is extremely selective for hypoxic cells, with minimal background staining in their normoxic counterparts. It significantly reduces the staining duration (30 min versus 2.5 h for pimonidazole) while still maintaining high fidelity. An inverse relationship between O 2 levels and N 3 -AZA click staining intensity is reported ( Fig. 4A-D, S4B). N 3 -AZA demonstrated preferential uptake by the nucleus (Fig. S6C) and concentrated in nucleoli (Fig. 4E, Movie 1). While this does not necessarily agree with the proteomic analysis (N 3 -AZA target proteins being predominantly cytoplasmic, Fig. S2C), some of the nuclear N 3 -AZA click staining may be attributed to the formation of drug-nucleic acid adducts. N 3 -AZA-nucleic acid binding was not investigated here, however, previous studies reported direct binding of activated NIs to DNA [34]. N 3 -AZA and pimonidazole showed similar cytotoxicity patterns in FaDu cells (preferential toxicity under hypoxia) (Fig. 1B & S8). Both pimonidazole immunostaining and N 3 -AZA click chemistry efficiently labelled hypoxic cells, however, the latter was significantly faster and produced a clearer background ( Fig. 5A and B). The efficiency of N 3 -AZA and pimonidazole as hypoxia probes was also compared in vivo, using two mouse models of head and neck cancer; (i) FaDu cell based subcutaneous tumour model (Fig. 5C-E) and (ii) a Cre-LoxP-based primary mouse model of HPV-negative squamous cell carcinoma (Fig. 5F). In tumour sections, N 3 -AZA-and pimonidazole-stained areas overlapped, suggesting that N 3 -AZA click chemistry is staining the same regions that are considered hypoxic by pimonidazole immunostaining (Fig. 5E). As expected, both N 3 -AZA and pimonidazole stained regions were mostly mutually exclusive with vasculature (CD31 positive regions). N 3 -AZA click chemistry following pimonidazole immunostaining did not affect the latter (Fig. 5D). Also, no significant click staining was detected in organs (brain, lung, liver, kidney and cervical lymph nodes) harvested from tumour-bearing mice administered with N 3 -AZA, indicating that N 3 -AZA is preferentially accumulating in the hypoxic regions of tumours (Figs. S10B-G). An important clinical question is to what degree does tumour size dictate hypoxia levels? Previous data obtained by computerized Eppendorf pO 2 histography indicated that hypoxia is independent of tumour size [72]. Our data with N 3 -AZA click staining performed on tumours ranging in volume between ~150 and 350 mm 3 also indicated no clear association between tumour size and hypoxia levels (Fig. S10A).
In summary, N 3 -AZA click chemistry provided a robust chemical strategy to identify potential protein targets of 2-NI compounds in a human head and neck tumour model. N 3 -AZA binding to the glycolytic enzyme GAPDH and detoxification enzyme GSTP1 showed a profound effect on their enzymatic activities only under hypoxia, which creates an interesting avenue for not only further mechanistic exploration of NI compounds in general, but also suggests their potential role in targeted hypoxia therapy. In addition, the modularity of the N 3 -AZA click reaction allowed for its application as an antibody independent histological hypoxia marker that offers a faster but equally efficient alternative to pimonidazole immunostaining.