Abstract 2-Aminoethanethiol dioxygenase (ADO) is a thiol dioxygenase that sulfinylates cysteamine and amino-terminal cysteines in polypeptides. The pathophysiological roles of ADO remain largely unknown. Here, we demonstrate that ADO expression represents a vulnerability in cancer cells, as ADO depletion led to loss of proliferative capacity and survival in cancer cells and reduced xenograft growth. In contrast, generation of the ADO knockout mouse revealed high tolerance for ADO depletion in adult tissues. To understand the mechanism underlying ADO’s essentiality in cancer cells, we characterized the cell proteome and metabolome following depletion of ADO. This revealed that ADO depletion leads to toxic levels of polyamines which can be driven by ADO’s substrate cysteamine. Polyamine accumulation in turn stimulated expression of proline dehydrogenase (PRODH) which resulted in mitochondrial hyperactivity and ROS production, culminating in cell toxicity. This work identifies ADO as a unique vulnerability in cancer cells, due to its essential role in maintenance of redox homeostasis through restraining polyamine levels and proline catabolism. __________________________________________________________________ ADO supports cancer cell growth by restraining polyamines, PRODH expression, and mitochondrial redox metabolism. INTRODUCTION Deregulation of cellular energetics is one the fundamental hallmarks of cancer ([60]1, [61]2). Cancer cells rewire metabolic pathways to meet the energetic, anabolic, and antioxidant demands imposed by aberrant proliferation. These needs are in part nurtured by high uptake of cysteine in conjunction with transsulfuration to supply cysteine as a precursor for glutathione (GSH), acetyl–coenzyme A (CoA), protein synthesis, and reduced inorganic sulfur and sulfate ([62]3). Conversely, cysteine catabolism through cysteine dioxygenase (CDO) is thought to represent a liability in non–small cell lung cancer cells due to the depletion of cysteine pools required for GSH synthesis ([63]4). Accordingly, the CDO locus is epigenetically silenced in lung, breast, esophagus, bladder, and stomach cancer ([64]5). CDO catalyzes the oxidation of cysteine to cysteine sulfinic acid (Cys-SO[2]H) which is then decarboxylated to produce hypotaurine ([65]Fig. 1A) ([66]6). However, an alternative route from cysteine to hypotaurine is provided through 2-aminoethanethiol dioxygenase (ADO), whose functional roles in (patho)physiology remain largely unexplored. Fig. 1. Loss of ADO inhibits cancer cell proliferation, cell survival, and tumor growth. [67]Fig. 1. [68]Open in a new tab (A) Illustration of cysteine and cysteamine metabolism through CDO and ADO. Figure was created with ChemDraw (PerkinElmer). (B) Illustration of the protein modifying role of ADO. Figure was created with BioRender. (C) Overall survival analysis using TCGA and HPA data in three different cancer types in patients with high (red) and low (black) levels of ADO tumor expression. Liver hepatocellular carcinoma: high (n = 98), low (n = 266); pancreatic adenocarcinoma: high (n = 45), low (n = 131); cervical squamous cell carcinoma and endocervical adenocarcinoma: high (n = 18), low (n = 273). (D) Western blot from cells with siRNA (si) against nontargeted control (NTC) or ADO. UT, untreated. Vinculin: loading control. (E) Cell confluency as a function of time after siRNA transfection with nontargeted control (siNTC) or ADO (siADO) measured by automated live cell imaging for the indicated cell lines. Data points represent the average of three technical replicates ± SD. Data are representative of three independent replicates. (F) Clonogenic surviving fraction following siRNA transfection with nontargeted control (siNTC) or ADO (siADO). All cells were plated 3 days after siRNA transfection except HuH-6 (5 days). Survival was normalized to the plating efficiency (PE) of NTC. Data points represent independent experiments. Bars represent the average ± SEM. ****P < 0.0001, ***P < 0.0001, **P < 0.01, and *P < 0.05. (G) Western blot of ADO protein in HeLa cells following CRISPR-Cas9–mediated KO. Vinculin: loading control. (H) Engraftment time (days) of mice injected subcutaneously with HeLa cells with ADO wild type (WT) and ADO knockout (KO). n = 10 in each group. (I) Engraftment time (days) of mice injected intramuscularly with HeLa cells with ADO WT and ADO KO. n = 5 in each group. The canonical enzymatic function of ADO is to catalyze the dioxygenation of cysteamine to hypotaurine which is further converted to taurine ([69]Fig. 1A) ([70]7). ADO has recently emerged as a remarkably versatile enzyme with additional ability to accommodate N-terminal cysteines in polypeptides in its catalytic site, oxidizing the sulfhydryl group to sulfinic acid ([71]Fig. 1B) ([72]8). Substrate promiscuity is supported by proline-rich loops flanking the active site that allow for structural flexibility ([73]9). The N-terminal Cys-sulfinic acid represents a N-degron that directs proteins for proteasomal degradation. A handful of proteins have been identified as endogenous substrates of ADO [Interleukin 32 (IL-32), regulator of G protein signaling 4/5 (RGS4/5), and Acyl-CoA dehydrogenase family member 10 (ACAD10)] ([74]8, [75]10, [76]11), but the consequences of their ADO-dependent regulation is unclear. A recent study correlated high expression of ADO with high grade and poor survival in glioblastoma multiforme (GBM) ([77]12). ADO overexpression was shown to drive GBM cell proliferation as well as expression of SOX2, OCT4, and nuclear factor κB ([78]12). ADO also emerged as a hit in a CRISPR-based screen to identify genes whose loss reduces cancer cell sensitivity to tumor necrosis factor (TNF)–mediated killing by cytotoxic lymphocytes ([79]13). How ADO might affect these phenotypes was not elucidated. Here, we took a comprehensive approach to assess the role of ADO in cancer cell biology through its potential to alter both the metabolome and proteome directly. We also assessed the physiological importance of ADO by generating ADO knockout (KO) mice. We demonstrate that ADO represents a vulnerability in cancer cells from multiple origins due to its unexpected role in restraining proline metabolism and maintaining redox homeostasis but is dispensable in adult murine tissues. As such, this study reveals the mechanism underlying ADO essentiality in cancer cells and suggests that ADO may represent a therapeutic target associated with a viable therapeutic window. RESULTS Depletion of ADO inhibits cancer cell proliferation and survival To probe for a potential role of ADO in cancer, we first explored how expression levels of ADO relate to patient survival. Analysis of The Cancer Genome Atlas (TCGA) data demonstrated that high ADO expression is associated with poor overall survival in liver, cervix, and pancreatic cancer ([80]Fig. 1C). No associations were found for breast, prostate, colon, lung, myeloid, and acute lymphoblastic leukemia. ADO expression provided the strongest prognostication in hepatocellular carcinoma where analysis of ADO expression as a continuous variable resulted in a hazard ratio of 1.2 (P < 0.001). Given this association, we assessed whether ADO plays a functional role in promoting cancer cell proliferation, migration, and survival. To determine the importance of ADO for proliferation, we depleted ADO with small interfering RNA (siRNA) and monitored cell confluence using an automated live-cell imaging system in one hepatocellular (SNU-449), one hepatoblastoma (HuH-6), one cervical adenocarcinoma (HeLa), one cervical squamous carcinoma (SiHa), and two pancreatic ductal adenocarcinoma (PANC-1 and Capan-2) cancer cell lines. siRNA-mediated knockdown of ADO attenuated proliferation in all six cancer cell lines ([81]Fig. 1, D and E). To assess long-term survival, we plated single cells for colony formation 3 days after siRNA transfection. In line with the proliferation data, transient depletion of ADO reduced cell survival severely in liver cancer cells and moderately in cervical and pancreas cancer cells ([82]Fig. 1F). Similar results were achieved using a SNU-449 model with doxycycline (dox)–inducible short hairpin RNA (shRNA) targeting ADO (fig. S1, A to C). Using a wound healing assay, we did not observe notable changes to migration when ADO was depleted (fig. S2A). Given the hypoxic nature of the tumor microenvironment and that ADO is an oxygen-sensitive enzyme, we repeated the same assays in hypoxia (0.2% O[2]). Proliferation and survival phenotypes after depletion of ADO in hypoxia were like those in normoxia (fig. S2, B and C), but hypoxia revealed a significant dependency on ADO for migration in HeLa and SiHa cell lines (fig. S2D). To test whether ADO is also important for tumor growth in vivo, we subcutaneously or intramuscularly implanted CRISPR-Cas9–mediated ADO KO cells engineered from HeLa cells ([83]Fig. 1G) in immunocompromised mice. We observed a significant delay in engraftment when cells lacked ADO in both models ([84]Fig. 1, H and I), thereby prolonging survival of the mice (fig. S2E). Together, high ADO expression is associated with poor survival of patients with liver, cervix, and pancreas cancer, and ADO plays a functional role in promoting cancer cell survival and proliferation, both in vitro and in vivo. ADO is dispensable for postnatal murine physiology To investigate the physiological essentiality of ADO, we attained ADO^+/+ [ADO wild type (WT)], ADO^+/− [ADO heterozygous (HET)], and ADO^−/− [ADO knockout (KO)] mice in the C57Bl/6J genetic background. ADO consists of one exon, and the entire coding region of exon 1 was deleted using CRISPR-Cas9 gene editing (fig. S3A). DNA sequencing confirmed the complete loss of the open reading frame of the ADO gene due to a 1029-bp deletion (fig. S3B). To further confirm the loss of ADO expression in tissues, we first examined tissue expression of ADO in ADO WT mice. Liver showed the highest level of expression across a panel of tissues ([85]Fig. 2A), and ADO was undetectable on both mRNA and protein level in ADO KO mice ([86]Fig. 2B). Fig. 2. Creation and characterization of ADO KO mice. [87]Fig. 2. [88]Open in a new tab (A) Fold tissue ADO mRNA expression in WT mice relative to L32 expression and normalized to spleen expression in males. (B) Left: Fold liver ADO mRNA expression in ADO WT and KO mice relative to L32 expression and normalized to WT in six animals (three females and three males). Right: Western blot of ADO protein levels in ADO WT, ADO HET, and ADO KO liver tissue lysates. Number represents unique animal. Vinculin: loading control. HET, heterozygous. (C) Kaplan-Meier curve of survival of bred females. Thirty-nine ADO WT, 46 HET, and 27 KO mice. (D) Bodyweight of female and male ADO WT, HET, or KO mice at 25 weeks old. (E) Kidney weight divided by whole animal weight in ADO WT, HET, and KO mice. (F) Serum albumin (ALB)/creatinine (CRE) ratio in blood from ADO WT, HET, or KO mice. (G) Relative abundance of taurine levels in serum from ADO WT, HET, or KO mice. Each dot represents an animal, bars represent average, and error bars represent SD. **P < 0.01 and *P < 0.05. ADO KO offspring were underrepresented from HET × HET and HET × KO crosses, at 13% (n = 53) and 39% (n = 132), respectively, with the latter being statistically significant (P < 0.05). The weight of pups was not affected by their ADO genotype (fig. S3C). Following weaning, nonbreeding HET and KO mice were healthy with no clinical complications (fig. S3D). In a small subset of breeding females, a diverse range of partum complications (e.g., birth canal obstruction, rectal prolapse, and swollen genital) led to a moderately increased mortality rate in ADO HET and ADO KO females compared to ADO WT females ([89]Fig. 2C). In addition, adult male KO mice were slightly smaller than their WT counterparts ([90]Fig. 2D). We observed a decrease in kidney size in female and male ADO HET and ADO KO mice when compared to ADO WT ([91]Fig. 2E), with no statistically significant changes in heart, lungs, thymus, intestines, pancreas, liver, spleen, brain, or reproductive organs. Evaluation of kidney function by the serum albumin/creatinine ratio (ACR) revealed a statistically significant decrease in ACR levels in male mice devoid of ADO, while HET females had high levels ([92]Fig. 2F). These differences were largely driven by changes in creatinine. Histopathological kidney analysis revealed no glomerular, interstitial, or vascular injury and no observations of interstitial inflammation, edema, fibrosis, mineralization, or hyalinization. Multifocal pyknotic nuclei indicative of apoptosis was identified in a small number of tubules in only females (fig. S3E). Reproductive capacity was measured in 47 breeding pairs. No difference between genotypes was observed for weight gain of plugged females (fig. S3F), and litter size was not significantly different in ADO WT and KO mice (fig. S3G). To determine whether loss of ADO leads to dysregulation of its known canonical metabolites, we performed multiplexed ultra-high pressure liquid chromatography–mass spectrometry analysis (UHPLC-MS) on blood serum samples from both female and male ADO WT, ADO HET, and ADO KO mice (data file S1). Cysteamine and hypotaurine, the direct metabolic substrate and product of ADO, respectively, were not detected. However, taurine was reduced by 25% in serum from ADO KO mice compared to ADO WT or HET mice ([93]Fig. 2G), which was driven by males only. No difference in taurine levels was detected in the livers of ADO KO versus WT mice. Together, our data demonstrate that ADO plays a moderate role in embryonic development and a minor role in (post-)partum females. Observed minor changes in kidney size, histopathology, and function, as well as a reduction in circulating taurine levels, remain subclinical. This suggests that a severe loss of proliferative capacity and viability upon loss of ADO represents a largely cancer-specific vulnerability. Since most patients with cancer are not undergoing pregnancy, ADO may represent a therapeutic cancer target. ADO has limited direct impact on the proteome To investigate how ADO promotes these cancer cell phenotypes, we first focused on the role of ADO in regulating protein stability. We generated a list of putative direct oxidation targets of ADO through searching UniProt for proteins that contain an N-terminal cysteine. Our criteria included proteins that contain a cysteine in the first position, the second position after the cleaved methionine, and/or cysteine as the first amino acid after cleavage of the signal peptide, transit peptide, or other propeptide. This analysis identified 2224 unique proteins or cleavage products of which 1698 were major histocompatibility complex class I antigens (data file S2). To assess the impact of ADO on the proteome, we performed liquid chromatography–tandem mass spectrometry (LC-MS/MS)–based proteomics on HeLa, SNU-449, and HuH-6 cell lines 5 days after siRNA transfection. Differential protein expression analysis confirmed ADO depletion across the three independent experiments ([94]Fig. 3A). On average, proteomic analysis of the whole cell lysates identified 8193 proteins (data file S3). Among the previously validated ADO targets, only IL32 and ACAD10 were detected by LC-MS/MS (data file S3), but neither was enriched by ADO knockdown ([95]Fig. 3B). Across all samples, 74 putative ADO substrates were detected. Of these, asparagine synthase (ASNS) was the only candidate that was enriched in all three cell lines with at least two being statistically significant (q < 0.05) ([96]Fig. 3B). ASNS transfers ammonia from glutamine to aspartate to form glutamate and asparagine. We confirmed accumulation of ASNS in ADO KO cells using Western blot ([97]Fig. 3C). However, accumulation of protein was accompanied by an increase in mRNA ([98]Fig. 3D), suggesting that ASNS is regulated at a transcriptional level. ASNS was also previously ruled out as a bona fide direct ADO substrate due to poor ability of ADO to oxidize the N-terminal ASNS peptide in solution and no ASNS accumulation upon ADO KO in cellulo ([99]8). We therefore regard ASNS as an unlikely direct ADO substrate and conjure that transcript up-regulation may be a response to cell stress. Nevertheless, we wondered whether the up-regulation of ASNS was functionally important for cellular survival upon ADO loss. Although cells were sensitive to losing ASNS alone, we observed a partial rescue of toxicity in cell survival when ASNS was codepleted with ADO ([100]Fig. 3E). Together, our data provide no evidence for a substantial role for ADO in directly regulating the proteome in these cells but reveal maladaptive ASNS induction upon ADO loss. To further understand the reason for the essentiality of ADO, we turned to its canonical role as a metabolic enzyme. Fig. 3. ADO has limited direct impact on the proteome. [101]Fig. 3. [102]Open in a new tab (A) Western blot of ADO 5 days following siRNA transfection (si) with nontargeted control (NTC) or ADO in SNU-449, HeLa, and HuH-6 cells. Vinculin: loading control. (B) Volcano plots representing P value adjusted for false discovery rate (Padj.) versus fold change in protein expression in proteomic data from indicated cell lines after siRNA-mediated knockdown of ADO for 5 days. Blue: Proteins with an N-terminal cysteine, putative ADO target. Those that are significantly changed upon ADO siRNA are named. Orange: Validated ADO substrates. Cutoff values for 1.5 fold change and Padj. value of 0.05 are indicated. (C) Western blot of ADO and asparagine synthetase ASNS protein levels in SNU-449 ADO WT and ADO KO cell lines. Vinculin: loading control. (D) Fold change of ASNS transcript levels after siRNA-mediated knockdown of ADO (siADO) or nontargeted control (siNTC) for 5 days in SNU-449 and HeLa cells. (E) Clonogenic surviving fraction in SNU-499 ADO WT or KO cells following siRNA transfection with nontargeted control (siNTC) or ASNS (siASNS). Cells were plated 3 days after siRNA transfection. Bars represents the average ± SEM, and data points represent independent experiments. ****P < 0.0001, ***P < 0.001, and *P < 0.05. Metabolomics reveals enrichment of polyamines in ADO-depleted conditions We addressed the role of ADO in regulating the cancer cell metabolome using high-throughput liquid chromatography–mass spectrometry–based metabolomics. To capture both very early and perhaps transient changes as well as slower responses, we analyzed samples from both 3 and 5 days after siRNA transfection in three cell lines. We focused particularly on central energy and redox metabolites, identifying 168 unique metabolites from these pathways present in all samples (data file S4). Unsupervised principal components analysis revealed distinct cell line–specific metabolic phenotypes (fig. S4). Hypotaurine and taurine levels were not significantly different in ADO-depleted cells (fig. S5). This suggests that flux through ADO provides a minor contribution to the (hypo)taurine pool in these cells. ADO’s substrate cysteamine was not detected, but we observed accumulation of its oxidized derivative cysteamine disulfide upon ADO depletion in SNU-499 and HeLa cells ([103]Fig. 4A). To test whether cysteamine enrichment itself could cause toxicity, we exposed cells to cysteamine, and this was sufficient to cause cell death ([104]Fig. 4B). Fig. 4. ADO drives polyamine metabolism. [105]Fig. 4. [106]Open in a new tab (A) Fold change in cellular cysteamine disulfide levels following siRNA-mediated knockdown targeting nothing (siNTC) or ADO (siADO). (B) Clonogenic surviving fraction of SNU-449, HuH-6, and HeLa cells treated with 0 or 5 μM cysteamine for 24, 24, and 12 hours, respectively. (C) Volcano plots representing significance versus fold change of metabolites from HuH-6, SNU-449, and HeLa cells 5, 3, and 3 days, respectively, after siRNA transfection, siADO versus siNTC. Cutoff values for 1.5 fold change and Padj. value of 0.05 are indicated. (D) Quantification of total polyamines in cells after cysteamine treatment as in (B). (E) Clonogenic surviving fraction after polyamine treatment: SNU-449: 2 μM putrescine and spermidine for 24 hours. HeLa: 2 μM putrescine and spermidine for 12 hours. Huh-6: 2 μM putrescine, spermidine, and spermine for 24 hours. (F) Cell confluency as a function of time after cells were treated with polyamines as in (E). Data points represent the average of three technical replicates ± SD. Data are representative of three independent experiments. (G) Left: Clonogenic surviving fraction in SNU-449 ADO WT (white bars) and ADO KO (red bars) cells treated with cysteamine (5 μM) and DENSpm (10 μM) for 24 hours, normalized to the PE of ADO WT. Right: Images of the colonies. (H) Clonogenic surviving fraction of SNU-449 ADO WT and ADO KO cells transfected with siRNA targeting nothing (siNTC) or PAOX (siPAOX). Unless otherwise noted, data points represent three independent experiments, and bars represent ± SEM. ns, not significant. ****P < 0.0001, *** P <0.001, **P < 0.01, and *P < 0.05. PA, polyamines. Beyond ADO’s canonical substrate and products, its depletion resulted in significant deregulation of 92 metabolites across cell lines and time points. Across all cell lines, the polyamine putrescine was consistently the most enriched metabolite, while its nearest intermediates ornithine and spermidine were also commonly enriched ([107]Fig. 4C and fig. S6). We wondered whether accumulation of cysteamine may be contributing to the dysregulation of polyamines in cells when ADO is lost. To this end, we used a commercially available assay which quantifies the total cellular polyamine pool (putrescine + spermidine + spermine). We observed a significant accumulation of polyamines across three cell lines when cells were treated with cysteamine, phenocopying loss of ADO ([108]Fig. 4D). To assess whether polyamine accumulation itself was toxic, we treated cells with exogenously supplied polyamines. For each cell line, we composed a treatment with individual polyamine species that reflected the fold increase upon ADO loss detected by mass spectrometry, and we confirmed that this resulted in the expected increases in total polyamine levels (fig. S7A). This substantially reduced cell survival ([109]Fig. 4E) and proliferation ([110]Fig. 4F). Conversely, we treated cells with the polyamine analog N^1,N^11-diethylnorspermine (DENSpm), which is known to cause overexpression of N^1-acetyltransferase (SSAT), an enzyme responsible for polyamine acetylation and depletion by cell export (fig. S6) ([111]14–[112]16). We validated that DENSpm prevented polyamine enrichment in cells with loss of ADO (fig. S7B), and it promoted rescue of cell survival when cells were treated with cysteamine or when ADO was knocked out ([113]Fig. 4G) or reduced by siRNA (fig. S7C). Codepletion of ADO with polyamine oxidase (PAOX), which is required for spermidine and putrescine synthesis from spermine, fully rescued ADO loss–induced cell toxicity in both siRNA and CRISPR-cas9 models ([114]Fig. 4H and fig. S7D). Together, these results suggest that toxicity in cancer cells upon loss of ADO is due to the lack of processing of its substrate cysteamine, which in turn drives accumulation of toxic levels of polyamines. We next sought to identify the underlying mechanism of why polyamine accumulation causes cell death. ADO depletion leads to loss of redox homeostasis To determine how ADO depletion might drive polyamine-induced cell death, we performed an unbiased pathway analysis of the metabolomic dataset. This identified that loss of ADO results in dysregulation of GSH metabolism across all cell lines and siRNA time points ([115]Fig. 5A). Consistent with the kinetics of ADO depletion (fig. S8), enrichment of metabolites in this pathway manifested most strongly in SNU-449 and HeLa cells 3 days after siRNA transfection and in HuH-6 cells 5 days after siRNA transfection ([116]Fig. 5B). The enrichment was partially transient in SNU-449 and HeLa cells, possibly reflecting adaptation over time. Deregulation of the GSH pathway included enrichment of GSH and GSH disulfide (GSSG) as well as GSH precursors cysteine, glutamate, and gamma-l-glutamyl-cysteine. In addition, there was enrichment of the redox pairs (dehydro)ascorbate and NADP(H) and polyamines ([117]Fig. 5B). Given that overall abundance of GSH, (dehydro)ascorbate, and NADP(H) was increased upon ADO knockdown, we postulated that loss of ADO may cause oxidative stress leading to an adaptive response. In line with this postulation, the GSH-synthesizing enzymes glutamate-cysteine ligase catalytic and modifier subunit (GCLC and GCLM) were transcriptionally up-regulated upon ADO depletion (fig. S9). Thus, we anticipated that the observation of cell death following the loss of ADO may be a consequence of oxidative stress. Fig. 5. Depletion of ADO drives formation of ROS. [118]Fig. 5. [119]Open in a new tab (A) Visualization of overlap among metabolic pathways that are significantly deregulated in SNU-449, HuH-6, and HeLa cells following siRNA-mediated ADO knockdown for 3 (D3) and 5 (D5) days. (B) Heatmap of levels of all metabolites in the GSH metabolism pathway (as defined by MetaboAnalyst 5.0) with siRNA targeting nothing (siNTC) and ADO (siADO), 3 (gray) and 5 (blue) days after transfection. Metabolite levels were normalized to the median of NTC and scaled using z scores. Each column represents an independent replicate. (C) Top: Flow cytometry histograms of intracellular ROS measured by oxidized CellROX after transfection with siRNA targeting nothing (siNTC) (black) or ADO (siADO) (red). Data are representative of three independent experiments. Bottom: Quantification of relative fluorescence intensity of intracellular ROS measured by oxidized CellROX. (D) Quantification of relative fluorescence intensity of intracellular ROS from cells (top) transfected with siRNA targeting nothing (siNTC) or ADO (siADO) or (bottom) ADO WT and KO cells with or without 10 mM N-acetyl cysteine (NAC) for 48 hours. (E) Clonogenic surviving fraction with or without 10 mM NAC for 48 hours (top) in cells treated with siRNA targeting nothing (siNTC) or ADO (siADO) and (bottom) in ADO WT and KO cells, with images of colonies. Data points represent independent experiments. Bars represent the average ± SEM. ****P < 0.0001, ***P < 0.001, **P < 0.01, and *P < 0.05. To test whether loss of ADO may cause oxidative stress, we assessed overall intracellular reactive oxygen species (ROS) levels by flow cytometry. Significantly higher levels of ROS were detected in ADO-depleted conditions in all three cell lines ([120]Fig. 5C). To test whether high ROS levels were responsible for the cell death upon ADO depletion, we supplemented cells with the ROS scavenger N-acetylcysteine (NAC). We observed a significant decrease in ROS levels upon NAC supplementation through flow cytometry ([121]Fig. 5D and fig. S10). Addition of NAC also resulted in a rescue of clonogenic survival in all three cell lines using either ADO siRNA or KO models ([122]Fig. 5E), confirming the role of ROS in driving cell death upon loss of ADO. ADO depletion causes deregulation of mitochondrial activity Given that mitochondria represent a substantial source of ROS, we sought to determine whether loss of ADO causes changes in mitochondrial function. We used a mitochondria-specific fluorescent oxidation probe to confirm that depletion of ADO increased mitochondrial superoxide (mtSuperoxide) content ([123]Fig. 6A). To test whether this could be driven by cysteamine or polyamines, we exposed ADO replete cells to these metabolites. Both caused an increase in mtSuperoxide ([124]Fig. 6B). Furthermore, polyamine depletion by DENSpm could prevent the increase in mtSuperoxide induced by either cysteamine or knocking out ADO. This suggests that the mtSuperoxide stimulation by cysteamine occurs via polyamine enrichment. In line with this, depletion of PAOX prevented the increase in mtSuperoxide levels in ADO KO cells ([125]Fig. 6B), again suggesting that polyamine interconversion is required to drive ROS. Fig. 6. Loss of ADO affects mitochondrial activity. [126]Fig. 6. [127]Open in a new tab (A) Fold change in relative fluorescence of MitoSOX representing mtSuperoxide in cells after transfection with siRNA targeting nothing (siNTC) or ADO (siADO). (B) Fold change in relative fluorescence of MitoSOX in ADO WT (white bars) and ADO KO (red bars) cells treated with 5 μM cysteamine for 24 hours, 2 μM putrescine and spermidine (PA) for 24 hours, 10 μM DENSpm for 24 hours, or PAOX siRNA (siPAOX) for 3 days. (C to E) Left: Oxygen consumption rate (OCR) determined by Seahorse analysis in (C) and (D) cells following transfection with siRNA targeting nothing (siNTC) or ADO (siADO) or (E) cells expressing dox-inducible shRNA targeting nothing (shNTC) or ADO (shADO) following 3 days of dox. Error bars indicate ± SD. Middle: Basal OCR. Right: Max OCR. (F) Max OCR determined by Seahorse analysis in ADO WT and ADO KO cells following treatment with 5 μM cysteamine (Cys) for 24 hours, 10 μM DENSpm for 24 hours, and 2 μM putrescine and spermidine (PA) for 24 hours. (G) Fold change in relative fluorescence of MitoTracker in cells with siRNA targeting nothing (siNTC) or ADO (siADO). (H) Quantification of ATP in cells with siRNA targeting nothing (siNTC) or ADO. Unless otherwise noted, data points represent independent experiments, and bar/lines represent the average ± SEM. ****P < 0.0001, ***P < 0.001, **P < 0.01, and *P < 0.05. Using the Seahorse mitochondrial stress test, we found that depletion of ADO by siRNA produced consistent trends toward increased basal mitochondrial respiration rate (SNU-449, P = 0.05; HuH-6, P = 0.08) and significant increases in maximum mitochondrial respiration rates ([128]Fig. 6, C and D). This was even more evident in the SNU-449 model with dox-inducible knockdown of ADO with shRNA ([129]Fig. 6E). These changes observed in siRNA models were manifested at time points when ADO was first significantly depleted and then reversible (fig. S11). Similar to mtSuperoxide production, an increase in maximum mitochondrial respiration rate could be recapitulated by exposure to cysteamine and reversed with the addition of DENSpm to deplete polyamines ([130]Fig. 6F). Increase in maximum respiration rate in ADO KO cells could also be prevented by polyamine depletion ([131]Fig. 6F). Changes in extracellular acidification rates (ECARs) that can reflect glycolytic activity were not consistent between cell lines (fig. S12). Higher basal and max respiration rates could potentially be driven by higher mitochondrial mass. A significant increase in mitochondrial content of SNU-449 and HuH-6 cells was observed when ADO was lost ([132]Fig. 6G). Increased respiration was also accompanied by a significant elevation of adenosine 5′-triphosphate (ATP) levels following ADO depletion ([133]Fig. 6H). Together, these data show that ADO restrains mitochondrial activity, leading to high respiration rates, ATP levels, and ROS production in its absence. These mitochondrial phenotypes can be driven by cysteamine and rescued by DENSpm, suggesting that they originate from polyamine accumulation as a result of impaired ability to metabolize cysteamine. ADO affects proline and arginine metabolism To determine how ADO depletion might affect mitochondrial function and ROS production, we probed deeper into the metabolomic dataset. We noticed that in addition to GSH metabolism, proline/arginine metabolism (PAM) was deregulated in all ADO-depleted samples, albeit identified as “PAM” in some samples and “arginine biosynthesis” in others. While several metabolites overlap with those in GSH metabolism (glutamate, ornithine, putrescine, spermidine, and spermine), the PAM pathway also includes amino acids and their derivatives like 2-oxoglutarate (i.e., α- ketoglutarate), proline, glutamine, and S-adenosyl-l-methionine (SAM) ([134]Fig. 7A). Arginine and proline metabolism are linked through l-pyrroline-5-carboxylate which interconverts to proline through either pyrroline-5-carboxylate reductase 1 (PYCR) or proline dehydrogenase (PRODH) (fig. S6) ([135]17, [136]18). PRODH serves as a key enzyme that drives proline catabolism while reducing flavin adenine dinucleotide (FAD) to FADH[2] which can fuel mitochondrial complex II to support ATP production ([137]19). Overexpression of PRODH has also been linked to production of mitochondrial ROS and cell death ([138]20–[139]22). We therefore hypothesized that PRODH is driving mitochondrial activity and ROS-mediated toxicity upon ADO depletion. Fig. 7. PRODH overexpression drives ROS and cell death. [140]Fig. 7. [141]Open in a new tab (A) Heatmap of metabolites in the proline and arginine metabolism pathway with siRNA targeting nothing (siNTC) or ADO (siADO), normalized to the median of NTC and scaled using z scores. Columns represent independent replicates. (B) Western blot from ADO WT/KO cells or cells treated with siRNA (si) targeting nothing (NTC) or ADO. (C) Western blot. OE, ADO overexpression. (D) [^13C[5],^15N]-l-proline in media of SNU-449 cells after isotope addition and siRNA targeting nothing (NTC) or ADO. Zero hours is directly before isotope addition. (E) As in (D) for cells. (F) CellROX intensity from (left) cells with siRNA targeting nothing (siNTC) or ADO (siADO) and (right) ADO WT/KO cells, with l-tetrahydro-2-furoic acid (L-THFA) (10 mM) for 12 hours. (G) Clonogenic survival with THFA (10 mM) for 12 hours in (left) cells treated with siRNA targeting nothing (siNTC) or ADO (siADO) or (right) ADO WT/KO. (H) Western blot from cells with siRNA (si) targeting either nothing (NTC) or PRODH (PRO). (I) Clonogenic survival of SNU-449 cells with siRNA targeting nothing (siNTC) or PRODH (siPRODH). (J) ATP after transfection with siRNA targeting nothing (siNTC) or PRODH (siPRODH). (K) Polyamines after transfection with siRNA targeting nothing (siNTC) or PRODH (siPRODH). (L) Western blot from cells treated with 5 μM cysteamine for 12 (SNU-449, HeLa) or 24 (HuH-6) hours, with quantification. (M) Western blot from cells with 10 μM DENSpm (DEN) for 24 hours. HuH-6/HeLa cells were transfected with siRNA (si) targeting nothing (NTC) or ADO (ADO). (N) Western blot from cells with siRNA (si) targeting nothing (NTC/N), PRODH (PRO/P), or ASNS (ASNS/A), with quantification. Data points represent independent experiments. Bars represent average ± SEM. ****P < 0.0001, ***P < 0.001, **P < 0.01, and *P < 0.05. ADO depletion leads to PRODH-mediated ROS formation ADO knockdown did not change PYCR levels but resulted in a clear accumulation of PRODH ([142]Fig. 7B), which could be reversed by reintroduction of ADO ([143]Fig. 7C). To determine whether accumulation of PRODH could drive increased proline catabolism, we incubated cells with [^13C[5],^15N]-l-proline and traced downstream metabolites by LC-MS for 48 hours, corresponding to 3 to 5 days after siRNA transfection. Labeled proline in the media was unchanged in ADO replete and deplete conditions ([144]Fig. 7D), but intracellular labeled proline was substantially lower in ADO knockdown cells, suggesting increased proline catabolism ([145]Fig. 7E and fig. S13A). This is consistent with metabolomic data in [146]Fig. 7A, demonstrating that initial proline enrichment at day 3 after siRNA transfection is followed by proline depletion 48 hours later. The only other labeled downstream metabolites detected were glutamate and reduced GSH, but the labeled fractions were very low (0.28 and 0.05%, respectively), consistent with rapid consumption and divergence of downstream metabolites (fig. S13, B to E). This suggests that intracellular pools of GSH and glutamate are predominantly supplied by other sources than proline catabolism, presumably maintaining high levels to meet the redox buffer needs of the cells. To test whether high PRODH activity is responsible for increased ROS upon ADO depletion, we used l-tetrahydro-2-furoic acid (L-THFA), a first-generation competitive inhibitor of PRODH ([147]23). L-THFA substantially mitigated the elevation in ROS observed after ADO depletion ([148]Fig. 7F). In line with this, L-THFA also partially rescued clonogenic survival when ADO expression was disrupted using either siRNA or CRISPR-Cas9 ([149]Fig. 7G). We confirmed the importance of PRODH expression through a genetic approach, where partial knockdown of PRODH ([150]Fig. 7H and fig. S14) partially rescued cell survival after ADO knockdown ([151]Fig. 7I). Since FADH[2] from PRODH can stimulate ATP production, we interrogated the role of PRODH in regulating cellular ATP levels after ADO depletion. The increase in total ATP content following ADO loss was reversible upon concomitant loss of PRODH ([152]Fig. 7J), demonstrating that ADO restrains mitochondrial activity through suppressing PRODH expression. Next, we wanted to elucidate whether PRODH accumulation is upstream or downstream of metabolic changes induced by ADO loss. Knocking down PRODH did not prevent the polyamine enrichment observed in ADO KO cells ([153]Fig. 7K). In contrast, exposing cells to ADO’s substrate cysteamine resulted in the accumulation of PRODH ([154]Fig. 7L). Furthermore, chemically induced stimulation of polyamine export by DENSpm exposure led to decreased PRODH expression levels when ADO was depleted ([155]Fig. 7M). These results place cysteamine and polyamine increases upon ADO loss upstream of PRODH induction. Last, we were curious about the potential link between PRODH expression and ASNS. In ADO replete cells, PRODH depletion led to a reduction in ASNS, while ASNS depletion led to a reduction in PRODH ([156]Fig. 7N). ASNS depletion also prevented PRODH up-regulation upon ADO loss, and there was a similar but nonsignificant trend for PRODH depletion to prevent ASNS up-regulation. As such, these proteins appear to be coregulated with interdependence. Together, these data demonstrate that loss of ADO leads to polyamine accumulation which stimulates PRODH overexpression to promote proline catabolism and drive mitochondrial activity which results in toxic ROS production leading to cell death ([157]Fig. 8). These effects also result in maladaptive ASNS induction which is coregulated with PRODH. This work suggests that clearing cysteamine is a critical role of ADO in cancer cells to maintain redox homeostasis. Fig. 8. ADO governs PRODH levels, mitochondrial activity, and ROS. [158]Fig. 8. [159]Open in a new tab Model illustrating consequences of ADO loss in cancer cells. Loss of ADO leads to overexpression of PRODH through polyamine accumulation, fueling mitochondrial activity and ROS production, leading to cell death. ASNS is coregulated with PRODH. Figure was created with BioRender. DISCUSSION In this study, we identified ADO expression as a negative prognostic factor in multiple cancer sites and established a functional relationship between ADO expression and cancer cell proliferation, survival, and xenograft growth. Our characterization of ADO-deficient mice revealed that loss of ADO does not significantly compromise postnatal development and physiology in nonbreeding animals, suggesting that ADO is dispensable for cysteine catabolism under normal conditions. However, low creatinine levels in ADO HET females warrants future studies of their muscle mass, while high creatinine levels in ADO KO males may indicate compromised liver or kidney function, which could be exacerbated under stress. Nevertheless, these subclinical phenotypes induced by the loss of ADO contrast the display of early postnatal mortality and growth delay in CDO KO mice ([160]24). This may be driven by a more severe reduction in taurine levels upon loss of CDO versus ADO. Taurine was reduced by 98% in the livers and 93% in the plasma of CDO KO mice ([161]24), while here we found no difference in the livers of ADO KO mice, and a 25% taurine reduction in the plasma. This suggest that CDO is the major contributor to taurine production in the liver and that ADO may contribute proportionally more to circulating taurine levels from other tissues. The discrepancy between the two KO mouse models suggests that the two dioxygenases play distinct role in mammalian physiology. This is exacerbated in the cancer setting, where cancer cell prosperity is supported by epigenetic silencing of CDO but requires expression of ADO. Our data suggest that ADO expression represents a unique cancer-specific vulnerability. With ADO being a promiscuous enzyme operating both on metabolite and protein substrates, we sought to define the relative contribution of these activities on ADO-dependent phenotypes. We found no support for ADO as a major direct contributor to the proteome but discovered that the metabolic role of ADO is key to maintaining redox homeostasis in human cancer cells. The known protein substrates of ADO RGS4, RGS5, and ACAD10 are also responsive directly to oxidative stress, independent of ADO. Their N-terminal cysteine can be oxidized directly by nitric oxide (NO) or other ROS, leading to sulfinylation and proteasomal degradation ([162]10, [163]11, [164]25). This effect is paralleled in plants, where the group VII ethylene response factors (ERF-VII) transcription factors that are PCO substrates are also sensitive to non-enzymatic oxidation by NO ([165]26). It is possible that the high ROS upon ADO loss in our human cancer cells caused non-enzymatic modification of cysteine-carrying ADO protein substrates, precluding identification of validated and putative ADO targets in the proteomic screen. If so, this implies that ADO can only be important as a direct regulator of the proteome in cells that are not prone to high ROS from loss of ADO due to alternative metabolic wiring or lower PRODH induction compared to the cells studied here. ADO has multifaceted roles where overall outcomes of its depletion may depend on cell context. Here, we demonstrate that loss of ADO, or exposure of cells to ADO’s substrate cysteamine, leads to accumulation of polyamines. It is therefore likely that the polyamine accumulation upon ADO loss is a consequence of the lack of cysteamine catabolism. However, since we were unable to reliably measure cysteamine abundance, we cannot directly attribute phenotypes to the accumulation of cysteamine, redirection of cysteamine to alternative metabolic processing, or other potential mechanisms. Concentrations of cysteamine chosen for cell exposures that phenocopied ADO loss were based on those commonly used in literature and in the low range of what is found in the plasma of patients treated with cysteamine for cystinosis ([166]27–[167]29). The mechanisms underlying how loss of ADO or cysteamine exposure drives polyamine accumulation, as well as how polyamine accumulation drives PRODH and ASNS expression, remain open and interesting questions. The route from cysteamine to spermine goes via CoA synthesis, cysteine, and the transsulfuration pathway (fig. S6). SAM was enriched in SNU-449 and HuH-6 cells upon ADO depletion ([168]Fig. 7A), while there was a trend toward enrichment of cysteine in HeLa ADO KO cells and SNU-449 and HuH-6 ADO siRNA cells compared to controls, supporting that polyamine enrichment could arise from this route. The finding that DENSpm exposure or depletion of PAOX both rescue cell toxicity upon ADO loss is also in line with polyamine accumulation arising from spermine rather than de novo putrescine synthesis via the urea cycle. Our data suggest that enrichment of proline (and other metabolites) initially upon ADO loss ([169]Fig. 7A) leads to up-regulation of PRODH protein expression, which stimulates proline flux and normalized levels ([170]Fig. 7, A, D, and E). PRODH expression is induced transcriptionally by p53 and peroxisome proliferator-activated receptor gamma ([171]30, [172]31) and negatively regulated at a posttranscriptional level by miR-23b ([173]32). We observed no induction of PRODH mRNA upon ADO depletion, warranting examination of whether miR-23b abundance is responsive to ADO expression and functionally linked to PRODH accumulation. An alternative and enticing possibility is that PRODH translation may be stimulated by eIF5A hypusination (eIF5A^H). Spermidine is a precursor for hypusine, and eIF5A^H has been shown to stimulate the translation of some transcripts that code for proteins with a mitochondria localization signal (MTS) ([174]16). Given that PRODH has a MTS, it may belong to this group of transcripts particularly sensitive to eIF5A^H. ASNS is transcriptionally stimulated by activating transcription factor 4 (ATF4), which might be activated upon ADO depletion in response to amino acid imbalance or ROS ([175]33, [176]34). However, we did not find evidence for induction of other canonical ATF4 targets such as C/EBP homologous protein (CHOP), TRIB, and ULK1. In line with our data, an unbiased metabolomic analysis on patient-derived glioma stem cells with high ASNS expression revealed an increase in metabolites such as GSH and GSSG and minimal activation of CHOP ([177]33). Curiously, our data show that ASNS induction in this context is maladaptive. This could potentially be by contributing glutamate for proline synthesis, exacerbating flux through PRODH. The PAM pathway has been widely explored in the context of redox biology, where production of ROS is mediated through the activity of PYCR and PRODH ([178]35–[179]37). Here, depletion of ADO led to PRODH accumulation, which was responsible for driving ROS production. Although our data are consistent with the mitochondrial phenotypes being promoted by increased proline catabolism, they do not rule out potential other noncanonical roles for PRODH. The role of PRODH in cancer appears multifaceted. On one hand, PRODH was coined a mitochondrial tumor suppressor due to its role in mediating cell death through FADH[2]-stimulated production of ROS in cells under nutrient-limiting conditions ([180]20, [181]22, [182]23, [183]35). In contrast, overexpression of PRODH under hypoxia had a protective effect against ROS by activating autophagy ([184]38). This suggests that the consequences of ADO-dependent PRODH accumulation may be cancer and environment dependent. Although ADO is a bona fide oxygen sensor given its direct dependence on molecular oxygen for enzymatic activity ([185]8), its functional role in governing a hypoxia response remains elusive. While activation of ERF-VII leads to induction of hypoxia-protective genes in plants ([186]39), no specific roles of ADO targets in hypoxia have been demonstrated. Here, we found that ADO is important for cell proliferation under 0.2% O[2], which is somewhat unexpected given its reported high Michaelis constant (K[M]) for O[2] of 500 μM ([187]8). This might suggest that the effective enzyme efficiency is higher in our cells ([188]40) or that ADO has additional unknown non-enzymatic functions. Our in vivo work here was limited to the study of tumor growth in an immune-compromised setting. It will be important in the future to assess the importance of ADO for carcinogenesis and tumor growth in immune competent models, especially given links between ADO and the interaction between cancer cells and cytotoxic T cells ([189]13). Work presented here reveals an unexpected role for ADO in guarding redox homeostasis in cancer cells through its governance of polyamine and proline metabolism. As such, we expose ADO dependency as a vulnerability in different cancer cell types. With previous work describing ADO as an oxygen sensor and sensitizer of TNF-mediated cell killing, ADO has emerged as a functionally multifaceted protein which may be a manifestation of its enzymatic promiscuity. MATERIALS AND METHODS Reagents All reagents used in the study can be found in table S1. Patient survival analysis The gene expression data for the following conditions were obtained from TCGA consortium ([190]https://cancer.gov/tcga): liver cancer (TCGA-LIHC), cervical cancer (TCGA-CESC), pancreatic cancer (TCGA-PAAD), breast cancer (TCGA-BRCA), prostate cancer (TCGA-PRAD), colon cancer (TCGA-COAD), and myeloid and acute lymphoblastic leukemia (TCGA-AML). These data were obtained using TCGAbiolinks package in R. Survival data for all datasets except for TARGET-ALL-P2 were obtained from the Human Protein Atlas (HPA) database, since they were more recent than the data downloaded from TCGA using the TCGAbiolinks. Survival data for TARGET-ALL-P2 were unavailable from the HPA and, therefore, were obtained from the TCGA database. Patients were classified into two groups using fragments per kilo base of transcript per million values for ADO (ENSG00000181915). Kaplan-Meier curves were created, and the survival outcomes of the two groups were compared using Wald tests. Maximally separated Kaplan-Meier analysis, as reflected by the P value, provided the ADO expression threshold. Kaplan-Meier plots were created using the autoplot function from the package ggfortify in R. Cell culture SNU-449 cells [American Type Culture Collection (ATCC), CRL-2234] were cultured in RPMI 1640 (Invitrogen). HuH-6 cells (Riken, RCB1367) were cultured in low glucose Dulbecco’s modified Eagle’s medium (DMEM) (Invitrogen). CAPAN-2 cells (ATCC, HTB-80) were cultured in McCoy 5A (Invitrogen). PANC-1 (ATCC, CRL-1469), HeLa (ATCC, CCL-2), and human embryonic kidney 293T (ATCC, CRL-3216) cells were cultured in high-glucose DMEM (Invitrogen). SiHa cells (ATCC, HTB-35) were cultured in alpha minimum essential medium (Invitrogen). All media were supplemented with 10% fetal bovine serum (Gibco), 5% penicillin-streptomycin (Invitrogen), and 1% amphotericin B (Gibco). Cells were cultured under 5% CO[2] in air at 37°C. All cell lines were tested for short tandem repeats to verify cell line of use. Cells were exposed to hypoxia (0.2% O[2]) in H45 hypoxystation (Don Whitley Scientific). All media and solutions were equilibrated to the desired oxygen concentration before use. CRISPR-Cas9–mediated gene editing of HeLa and SNU-449 cells PX458 [pSpCas9(BB)-2A-GFP] plasmid encoding Cas9 nontargeting or ADO-targeting guide RNA (gRNA) were purchased from Addgene and GenScript, respectively. The single guide RNA sequences can be found in table S2. Cancer cells were transiently transfected with 2.5 μg of the PX458 vector using X-tremeGENE 9 DNA Transfection Reagent Polymer reagent (Roche) following the manufacturer’s instruction. Green fluorescent protein–positive cells were sorted 48 hours later using a MoFlo Astrios Cell Sorter (Beckman Coulter) at the Princess Margaret Flow Cytometry Facility. Cells were single-cloned by dilution in flat 96-well plates (Corning) and expanded individually. Clones were screened for ADO expression by Western blot. Editing of the gene was further validated by sequencing the amplified genomic DNA region. Genomic DNA was extracted using the Genomic DNA Isolation Kit per the manufacturer’s protocol (Norgen). Genomic target regions were amplified using Platinum SuperFi II PCR Master Mix according to the manufacturer’s protocol using 35 cycles, denaturation at 98°C, and annealing temperature at 60°C. The polymerase chain reaction (PCR) products were run on a 1.5% agarose gel. Then, the PCR products were submitted for DNA sequencing. Validation sequences can be found in table S2. Guides 1 and 3 yielded in successful clones for HeLa and SNU-449 cells, respectively. Cells transfected with the nontargeting control vector are referred to as ADO WT, and cells transfected with the ADO-targeting vector are referred to as ADO KO. Generation of SNU-449 cells expressing dox-inducible shRNA Tetracycline/doxycycline (dox)-inducible ADO shRNA oligos (TRCN0000163983, Broad Institute) were annealed and inserted in the pLKO.1 Tet-On plasmid, a gift from D. Wiederschain (Addgene plasmid #21915) ([191]41). Briefly, the Age I–HF and Eco RI–HF restriction sites were used to digest the plasmid, and the open vector DNA fragment was purified on E-Gel CloneWell II Agarose Gel (0.8%; Thermo Fisher Scientific) and extracted using a Qiaquick Gel Extraction kit (Qiagen). shADO oligos were cloned into it, and the ligation product was used to transform competent Stbl3 Escherichia coli. To determine the presence of shRNA-containing plasmids, sequencing was performed using the primer 5′-ggcagggatattcaccattatcgtttcaga-3′. SNU-449 cells were transfected with packaging plasmid psPAX2 (Addgene #12260), envelope plasmid pMD2.G (Addgene #12259), and pLKO-Tet-On shRNA using FuGENE 6 transfection reagent (Thermo Fisher Scientific). The viral supernatant was collected, and the titer was calculated by the QuickTiter Lentivirus Kit (Cell Biolabs Inc.). Virus was added to SNU-449 cells with polybrene (8 μg/ml) at a multiplicity of infection of 0.5 and selected with puromycin (5 μg/ml) after overnight incubation. Single cells were seeded and propagated for stable clone creation. For subsequent experiments, dox (2 μg/ml) was added at the time of cell seeding. siRNA transfection ADO siRNA or nontargeted control (NTC) siRNA was transfected using Lipofectamine RNAiMAX at a final concentration of 30 pM (SNU-449, HeLa, SiHa, PANC1, CAPAN2 and HuH-6) according to the manufacturer’s protocol. Four microliters of RNAiMAX lipofectamine reagent was used in either a 6-cm dish (VWR) or a six-well plate (Corning). Unless otherwise noted, assays were performed 3 days after siRNA transfection in SNU-499, HeLa, SiHa, PANC1 and CAPAN2 cells and 5 days after siRNA transfection in HuH-6 cells. Immunoblotting Cells were washed twice with phosphate-buffered saline (PBS) and then lysed in radioimmunoprecipitation assay (RIPA) buffer containing complete protease and phosphatase inhibitor cocktail (Pierce) on ice. Samples were then centrifuged at 14,000 rpm for 15 min at 4°C. Then, the supernatant was collected and measured for its total protein concentration by bicinchoninic protein assay (Thermo Fisher Scientific) using an OMEGA plate reader (BMG Labtech) at 562 nm. Samples were then prepared with Bolt loading buffer (Invitrogen) and dithiothreitol (DTT) and boiled at 95°C for 5 min. Proteins were resolved on SDS–polyacrylamide gel electrophoresis gels and transferred to a polyvinylidene difluoride membrane (Immobilon-P, Millipore) and blocked in Licor blocking buffer. Membranes were incubated with primary antibodies overnight at 4°C and secondary antibodies for 1 hour. All antibodies used can be found in table S1. Cell proliferation, migration, and clonogenicity Cells were monitored for confluency through the IncuCyte ZOOM Kinetic Imaging System (Essen BioScience) in multiwell plates. All cell lines were grown in a monolayer, and confluence was confirmed to be proportionate to the cell number. Eight hundred–micrometer scratch was made with the Wound Maker (Essen BioScience) to monitor cell migration. Clonogenic survival was monitored through seeding single cells in a six-well plate following appropriate treatments and incubating for 14 days. Crystal violet was used to stain the colonies. Plating efficiency (PE) was calculated as the number of colonies (>50 cells) divided by the number of cells seeded. Calculation of the surviving fraction was done by dividing PE (treated) by PE (control). Xenograft establishment and growth CRISPR-Cas9–mediated ADO KO HeLa cells (2 × 10^6) were injected subcutaneously in the flank of female NSG:NOD.Cg-Prkdc II2rg mice (the Jackson Laboratory) and intramuscularly in NOD-SCID mice (the Jackson Laboratory). Engraftment was defined as detection of a 150-mm^3 tumor as measured by a caliper. Tumors were measured two to three times a week by a caliper until they exceeded 1500 mm^3, upon which the mice were euthanized. Overall survival analysis was performed by fitting a Cox proportional hazards regression model in GraphPad Prism 5 (GraphPad Software Inc., San Diego, CA, USA). P values were obtained using the likelihood ratio test. Generation of ADO KO mice ADO KO mice were created by Shanghai Model Organisms (Shanghai, China). CRISPR-Cas9 mRNA and two gRNA were microinjected into fertilized eggs of C57BL/6J mice. Twenty-seven F0 mice were produced. Confirmation of the mutant mice was done through PCR analysis. F0 mice were crossed to generate nine positive offspring carrying the KO allele. These were then transferred to the Princess Margaret Cancer Center Animal Quarantine Facility and independently validated by sequencing the WT and KO alleles: gRNA1, GCCCCGCGCGCCCCGCTGGCCGG, gRNA2, AGCAGGGCAGGGTACATCTTTGG. Animal procedures All animal experiments were performed according to procedures approved by the Princess Margaret Cancer Centre Animal Care Committee in accordance with the Ontario Cancer Institute Animal Use Protocol number 6449. Mice were housed in a pathogen-free barrier facility under standard temperature, humidity and light conditions mandated by the Princess Margaret Cancer Centre Animal Research Centre. All animals were on a standard rodent diet and water ad libitum. Animals were genotyped at, on average, postnatal day 7, weaned at day 21, and monitored daily. Mice were genotyped for ADO through Transnetyx (Cordova, TN, USA) using the following primers: forward primer (Fp), CGCCCGACCATTCCATAA; reverse primer (Rp), TCCAGTGGGCCGTGTTC. Tissue RNA extraction and quantification Tissue from ADO WT and ADO KO mice was preserved in liquid nitrogen. Tissues were homogenized using TissueRuptor (Qiagen) for 10 s, and RNA was then extracted by using the Animal Tissue RNA Purification Kit (Norgen Biotech, 25700) as per the manufacturer’s protocol. RNA concentration was determined using NanoDrop (Thermo Fisher Scientific). Reverse transcription was performed using qScript cDNA SuperMix (QuantaBio) according to the manufacturer’s instruction. Quantitative PCR analysis was performed using Perfecta SYBR Green Supermix (QuantaBio). The following primers were used: L32 Fp, ATGGCTCCTTCGTTGCTGC; L32 Rp, CTGGACGGCTAATGCTGGT; ADO Fp, CGGGACTGCCACTATTACCG; ADO Rp, GGCAGGAGCTTCAAGGTAGG. Tissue lysate preparation Whole cell lysates of left lobe liver samples were prepared by cutting 20 mg of flash frozen liver on dry ice. Samples were homogenized in RIPA buffer containing complete protease and phosphatase inhibitor cocktail (Pierce) for 10 s with Qiagen Tissue Ruptor at max speed on ice. They were placed on HuLa Mixer (Thermo Fisher Scientific) for 2 hours in the cold room on 10-s rotate and 5-s shake cycle and then sonicated twice with Bioruptor at 0°C for 3 min of 30-s on/off at medium intensity. Samples were put on ice for 3 min to cool in between sonication steps and centrifuged for supernatant collection. Clinical chemistry Blood samples were collected by cardiac puncture. Samples were submitted to the Centre for Phenogenomics (Mount Sinai Health System, Toronto, ON, Canada) for measurement of albumin and creatinine on a Beckman AU480 Biochemistry Analyzer. Histopathology analysis Tissues isolated from ADO WT, ADO HET, and ADO KO mice were fixed in 10% buffered formalin and embedded in paraffin. Sections were stained by hematoxylin and eosin. Stained sections were analyzed at the histopathology core at the Centre for Phenogenomics (Mount Sinai Health System, Toronto, ON, Canada) by a board-certified veterinary pathologist. Protein substrate candidate analysis Using [192]https://uniprot.org/uniprot/?query=approach, amino acids in the positions of interest were extracted for all proteins available in the UniProt database. For signal and transit peptides, first and second amino acids were extracted (if available), while for chain, peptide, and propeptide, first amino acid was extracted for each listed segment (if available). Records with position data containing only one position instead of a range (for example, “45” instead of “45 to 60”) were excluded from further analysis. Records with position data containing “?,” ”>,” or “<” were excluded from further analysis as well. Further selection was done in the presence of N-terminal cysteine after cleavage of the signal (in transit peptide, chains, peptides, and propeptides), yielding the candidate list in data file S2. Whole cell proteomics Cell pellets were lysed by repeated freeze-thaw cycles in lysis buffer [50 mM Hepes (pH 8) and 1% SDS]. Samples were sonicated on a probe-less ultrasonic sonicator for five 10-s cycles at 10 W per tube (Hielscher VialTweeter) to shear genomic DNA. Disulfide bonds were reduced with 5 mM DTT, followed by 30-min incubation at 60°C. Free sulfhydryl groups were alkylated by incubating samples in 25 mM iodoacetamide in the dark for 30 min at room temperature. An additional 5 mM of DTT was added to quench the alkylation reaction, and samples were incubated at room temperature for 5 min. The magnetic bead–based SP3 protocol was used to capture proteins before digestion ([193]42). Briefly, magnetic beads were added to proteins in a 10:1 (w/w) ratio. Absolute ethanol was added to bring the ethanol concentration to 70%. Samples were shaken at room temperature for 5 min at 1000 rpm, and the supernatant was discarded. The beads were rinsed two times with 80% ethanol and discarded. Proteins were digested in 100 mM ammonium bicarbonate containing 2 μg of trypsin/Lys-C enzyme mix (Promega) at 37°C overnight. Peptides were desalted using C18-based solid phase extraction and then lyophilized in a SpeedVac vacuum concentrator. Peptides were solubilized in mass spectrometry–grade water with 0.1% formic acid. LC-MS/MS data were acquired as previously described with modifications ([194]43). Raw data were searched in MaxQuant (version 1.6.3.3) using UniProt protein sequence database containing human protein sequences from UniProt (complete human proteome; 2019-09) ([195]44). Searches were performed with a maximum of two missed cleavages and carbamidomethylation of cysteine as a fixed modification. Variable modifications were set as oxidation at methionine and acetylation (N-term). The false discovery rate for the target-decoy search was set to 1% for protein and peptide levels. Intensity-based absolute quantification (iBAQ), label-free quantitation (LFQ), and match between runs (matching and alignment time windows set as 0.7 and 20 min, respectively) were enabled. The proteinGroups.txt file was used for subsequent analysis. Proteins matching decoy and contaminant sequences were removed, and proteins identified with two or more unique peptides were carried forward. LFQ intensities were used for protein quantitation ([196]44). For proteins with missing LFQ values, median-adjusted iBAQ values were used as replacement ([197]45). Protein intensities (data file S3) were log[2]-transformed for further analysis. RNA sequencing Sequencing library was prepared using NuGEN Encore Complete strand-specific RNA library preparation kits as per the manufacturer’s protocol. The library size was determined using the Bioanalyzer DNA 1000 chip. Concentration was determined using a Qubit fluorometer and the Quant-iT dsDNA BR Assay Kit (Invitrogen). Samples were loaded on an Illumina HiSeq 2000 sequencer and sequencing using 2 × 100 paired-end reads. The reads were trimmed on the basis of quality and adapters using TrimGalore 0.6.6. Next, the reads were mapped to the human genome GRCh38 using STAR short-read aligner (v.02021) ([198]46), and genes were quantified using HTSeq count ([199]47). Genes, showing low levels of expression, were removed using filterByExpr function in EdgeR package ([200]48) in R, and counts for the remaining genes were normalized with trimmed mean of M-values normalization method ([201]49). Next, normalized values were transferred using voom transformation ([202]50) and used for differential expression analysis. Polyamine quantification Cells were trypsinized and resuspended in media. Total polyamine measurements (putrescine, spermine, and spermidine) were determined using the Total Polyamine Assay Kit (MAK349-1KT, Sigma-Aldrich) following manufacturer’s protocol, incubated at 37°C for 30 min, and read using an OMEGA plate reader (BMG Labtech) using end-point fluorescence (λex = 535 nm/λem = 587 nm). Global metabolomics Metabolites from serum or cell pellets were extracted at 1 × 10^6/ml in ice-cold 5:3:2 methanol (MeOH):acetonitrile:water (v/v/v) and vortexed for 30 min at 4°C. Supernatants were clarified by centrifugation and analyzed using Thermo Vanquish UHPLC coupled to a Thermo Q Exactive mass spectrometer. Global metabolomic analysis was performed as described previously ([203]51). A mass/charge ratio range of 65 to 950 was used for data acquisition. Annotations, integrations, and quality control of peaks were completed as previously described ([204]51). The obtained values [files S1 (serum) and S4 (cells)] were log[2]-transformed, and then for each cell line, the groups were compared using LIMMA package in R. Pathway enrichment analysis was completed through MetaboAnalyst 5.0, using a cutoff of raw P < 0.05 and q < 0.08 for significant regulation. Venn diagram to illustrate common pathways have been created using jvenn ([205]52). Overall ROS quantification Cells were trypsinized and washed once in PBS. CellROX Deep Red (Thermo Fisher Scientific) was used to stain as per the manufacturer’s protocol. Flow cytometry was performed at the Flow Cytometry Core at the Hospital for Sick Children. Analysis was performed using FlowJo software version 10 (FlowJo LLC). Mitochondrial ROS quantification Cells were trypsinized and washed once in PBS. Cells were resuspended in media containing 5 μM MitoSOX Red (Invitrogen) and incubated at 36°C for 30 min. Fluorescence was read on an OMEGA microplate reader (BMG Labtech) at 580 nm. Relative fluorescence unit was normalized to cell number through a Cyquant NF cell proliferation assay kit (C7026, Thermo Fisher Scientific). Seahorse XF analysis Oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) were measured using Seahorse XFe96 Extracellular Flux Bioanalyzer (Agilent). Cells (10^4) were seeded and cultured for 24 hours. Oligomycin, trifluoromethoxy carbonylcyanide phenylhydrazone, and mixture of antimycin and rotenone were injected to final concentrations of 2 to 3, 0.05 to 0.2, 1, and 0.5 μM, respectively. OCR and ECAR were normalized to cell number through a Cyquant NF cell proliferation assay kit (C7026, Thermo Fisher Scientific). Mitochondria quantification Treated cells were trypsinized and washed twice in PBS. Cells were resuspended in media without fetal bovine serum containing 200 nM MitoTracker Green (Invitrogen) and incubated at 36°C for 45 min in a 96-well microplate. Signal was read on an OMEGA microplate reader (BMG Labtech) at 480 nm. Mitochondria content was normalized to cell number through a Cyquant NF cell proliferation assay kit (C7026, Thermo Fisher Scientific). ATP quantification Cells were trypsinized and washed once in PBS. A total of 1 × 10^6 cells were collected, and intracellular ATP concentration was analyzed using the EnzyLight ATP Assay Kit as per the manufacturer’s protocol (BioAssay Systems, Hayward, CA, USA). Overexpression of ADO in SNU-449 ADO KO cell line SNU-449 ADO KO cells were transiently transfected with pCMV6-ADO-Myc-DDK to overexpress ADO (Origene). The manufacturer’s protocol was followed with a slight modification: 4 μg of plasmid was used with 3 μl of lipofectamine 2000 for 48 hours in a six-well plate (Corning). Quantitative intracellular proline tracing To quantify intracellular [^13C[5],^15N]-l-proline tracing and GSH and glutamate quantification, SNU-449 cells were plated and transfected with ADO siRNA for 3 days. Following the 3-day transfection period, cells were washed and was fed [^13C[5],^15N]-l-proline containing (20 mg/liter) medium for 0, 8, and 48 hours. After the indicated time points, medium was aspirated, and the cells were quickly washed with ice-cold PBS. Metabolites were extracted from frozen cell pellets at 2 million cells/ml by vigorous vortexing in the presence of ice-cold 5:3:2 MeOH:Acetonitrile:water (v/v/v) for 30 min at 4°C. Metabolites were extracted from media (20 μl) using 480 μl of the aforementioned extraction solution with identical agitation as for cells. After 30 min, supernatants were clarified by centrifugation (10 min, 12,000g, 4°C) and then analyzed (15 μl per injection) by UHPLC-MS (Vanquish and Orbitrap Exploris 120, Thermo Fisher Scientific). Metabolites were resolved on a Kinetex C18 column (2.1 mm by 150 mm, 1.7 μm) using a 5-min gradient method exactly as previously described ([206]51). Following data acquisition, .raw files were converted to .mzXML using RawConverter, then metabolites were assigned, and peaks were integrated using Maven (Princeton University) in conjunction with the Kyoto Encyclopedia of Genes and Genomes database and an in-house standard library. ^13C, ^15N isotopic enrichment was visualized using GraphPad Prism 9.0. ^13C[2], ^15N, and ^15N[2] metabolite peak areas were corrected for natural abundance. Quality control was assessed by using technical replicates run at beginning, end, and middle of each sequence as previously described ([207]53). Data display and statistical analysis For data showing fold changes, individual data points normalized by the average of the control group (WT, NTC siRNA, or untreated) are shown. Unless specified otherwise, data were analyzed using GraphPad Prism 5 (GraphPad Software Inc., San Diego, CA, USA) with unpaired, one-tailed Student’s t test, or analysis of variance (ANOVA). To assess whether the observed distribution of genotypes significantly deviated from the expected Mendelian ratio, a chi-square (χ^2) test was conducted using the formula: [MATH: χ2=∑ (Oi−Ei)2Ei :MATH] . Whereas O[i] is the observed count for each genotype, E[i] is the expected count, and the sum is taken over all genotypes. The degrees of freedom (df = 1) for the chi-square test (χ^2 = 6.8) were determined as the number of genotypes minus one. To test the differences in survival across the observed genotypes, pair-wise comparison test was performed. Acknowledgments