Targeting Mitochondrial Proline Dehydrogenase with a Suicide Inhibitor to Exploit Synthetic Lethal Interactions with p53 Upregulation and Glutaminase Inhibition

From prokaryotes to the highest eukaryotes, proline is catabolized by a unique and structurally conserved flavoprotein, proline dehydrogenase (PRODH; refs. 1–3). In eukaryotes, PRODH associates with the inner mitochondrial membrane and catalyzes the first and rate-limiting catabolic step to yield the intermediate metabolite, pyrroline-5-carboxylate (P5C), with transfer of two electrons to the electron transport chain where they produce either ATP or reactive oxygen species (ROS), as illustrated in Fig. 1A. Proline's high metabolic potential of approximately 30 ATP equivalents/mol is exploited by many insects like flies, whose flight is fueled by proline oxidation and prevented by mutational inactivation of PRODH (4, 5).

PRODH's potential importance in cancer first emerged when it was uncovered by Polyak and colleagues as one of the most strongly upregulated genes by the tumor suppressing protein, p53, although its identity was unknown at that time, so it was simply referred to as “p53-induced gene 6” (6). PRODH's functional importance in tumor mitochondria grew out of later studies spearheaded by Phang and colleagues stemming from their interest in cancer cell stress responses to nutrient deprivation and hypoxia, as well as the importance of exogenous proline derived from the breakdown of extracellular collagen in sustaining intracellular ATP by anaplerosis (1, 7–10). While the Phang laboratory's earlier studies indicated that the capacity of PRODH to generate ROS and induce apoptosis qualified it as a tumor suppressor and provided rationale for its induction by p53 (1, 7), their later studies emphasized the importance of PRODH in sustaining ATP production and cancer cell survival particularly under microenvironmental stress conditions (8–10). Our subsequent studies comparing proline oxidation between insect cells and cancer cell lines suggested that while PRODH can induce mitochondrial ROS production, it likely does so largely via the electron transport chain [e.g., complex I, α-ketoglutarate (KG) dehydrogenase complex] from anaplerotic glutamate and α-KG production, and not as a direct enzymatic product of inner membrane PRODH activity (11). Today we appreciate that the cancer-specific role of mitochondrial PRODH must be understood within the full proline-P5C cycle, wherein the balance between proline biosynthesis (from glutamate and ornithine via P5C reduction by PYCR1) and proline catabolism (via PRODH) can differentially contribute to cancer cell growth, death, or senescence in various ways (12).

With recognition that a key hallmark of cancer is its reprogramming of normal cell metabolism (13), there has been a resurgence of research into the metabolic pathways commonly altered during oncogenesis along with academic and pharmaceutical discovery efforts to develop small-molecule metabolic inhibitors as novel cancer therapeutics (14, 15). In this regard, considerable focus has been given to cancer's rewiring of glutamine metabolism to meet its own cellular energy and biosynthetic needs under microenvironmental stress conditions, most commonly accomplished by activation of the MYC oncogene, which transcriptionally upregulates glutaminase (GLS1; ref. 16). This focus led to the development of a promising and well tolerated new GLS1 inhibitor, CB-839 (Calithera), now advancing through clinical trials against breast cancer and other human malignancies (17, 18). However, along with such advances, came the growing appreciation of the major challenge faced with development of new metabolic inhibitors for cancer therapy, the adaptive ability of cancer cells to utilize alternative metabolic reactions to bypass any well-blocked enzyme (15, 16). Thus, it may seem surprising that during development of GLS1 inhibitors little if any attention was paid to targeting PRODH, arguably an equally important anaplerotic source of mitochondrial glutamate and α-KG production as GLS1. This inattention may have been due in part to the initial functional classification of PRODH as a mitochondrial tumor suppressing protein (1, 7, 9).

Ironically, biochemical searches for PRODH inhibitors predated by at least three decades all efforts to target GLS1 or any other oncogenic metabolic pathways. Instead, these biochemical searches were motivated by the desire to eradicate disease-carrying mosquitoes (4), but were severely limited by lack of purified PRODH enzyme and structural information about the target protein. Not until decades later were the first competitive and irreversible PRODH inhibitors proposed (19), followed by the first derived crystallographic structures of prokaryotic versions of PRODH (2), including one bound to a first-generation competitive inhibitor, L-THFA (20). As a result, it was only very recently that investigators began exploring the use of a PRODH competitive inhibitor like L-THFA as a new cancer therapeutic (21). Concurrently, we independently developed evidence supporting the hypothesis that PRODH represents a unique mitochondrial cancer target and explored structural models of human PRODH to evaluate a potentially more effective irreversible suicide inhibitor. We now describe our chemical biology studies targeting cancer mitochondria with either a second-generation competitive PRODH inhibitor, S-5-oxo-2-tetrahydrofurancarboxylic acid (S-5-oxo), or with an irreversible suicide inhibitor, N-propargylglycine (N-PPG), unique in its ability to induce selective mitochondrial decay of PRODH at doses that can be safely administered in vivo.

Cell lines were all obtained from ATCC and immediately frozen (liquid nitrogen) as cell stocks; before each experiment a stock vial was thawed and serially passaged (no more than 10 generations, without further authentication) under 5% CO₂ and 37°C culture conditions as recommended. Routine Mycoplasma checks were performed using the MycoAlert Detection Kit from Lonza. Antibodies used in this study included the anti-PARP/cPARP (46D11) rabbit monoclonal from Cell Signaling Technology; β-actin (C4), PRODH (A-11), NDUFS1 (E-8), and Rieske FeS IgG (A-5) mouse monoclonals from Santa Cruz Biotechnology; horseradish peroxidase (HRP)-conjugated goat anti-mouse secondary from Bio-Rad Laboratories, Inc.; α-tubulin mouse monoclonal (T9026) from MilliporeSigma; PRODH and NDUFS1 rabbit polyclonals from ProteinTech; Alexa 488 goat anti-mouse and 594 goat anti-rabbit secondaries from Life Technologies (Thermo Fisher Scientific); and mouse anti-ATP-synthase α and anti-cytochrome C from BD Biosciences Transduction Laboratories.

CB-839 was provided by Calithera Biosciences, Inc, and MI-63 from Sanofi US; nutlin-3 was purchased from Sigma Aldrich/MilliporeSigma, and the S and R stereoisomers of 5-oxo were purchased from MilliporeSigma. DL-propargylglycine (DL-PPG) was purchased from Cayman Chemical Co. N-PPG was synthesized as follows: 5 g of propargylamine (MilliporeSigma) in 5 mL of DEPC (diethyl pyrocarbonate) water was slowly mixed over ice with 2 g of iodoacetic acid until fully dissolved. The 10.5 mL reaction mix protected from light was incubated overnight at room temperature and then concentrated (Savant SPD131DDA SpeedVac Concentrator, Thermo Fisher Scientific) down to 4.5 mL, aliquoted into eppendorf tubes and further dried down to a brown viscous liquid. To separate N-PPG from unreacted propargylamine in the pooled mix, 2.5 mL of ethanol was added followed by acetone-ethanol (30.5 mL-3.25 mL) to a final volume of 40 mL, resulting in a light beige N-PPG precipitate that was dried, extensively rewashed, and its purity confirmed by HPLC-mass spectrometry and NMR spectroscopy. The newly synthesized compound matched the spectroscopic data for N-PPG as reported previously (22) and with >90% purity: ¹H-NMR (500 MHz, deuterium oxide) δ 3.94 (d, J = 2.6 Hz, 2H), 3.71 (s, 2H), 2.96 (t, J = 2.6 Hz, 1H).

The three-dimensional (3-D) structure of mammalian PRODH remains experimentally undetermined; however, crystal structures of bacterial PRODH-like catalytic domains bound to either a proline analogue or mechanism-based inhibitor have become available over the past decade (20, 23, 24). To estimate at atomic level resolution the 3-D structure of human PRODH bound to either the competitive inhibitor S-5-oxo or the suicide inhibitor N-PPG, we used homology-based structural modeling to align the known primary target sequence of human PRODH with the known 3-D crystal template structure of a prokaryotic PRODH-like catalytic domain (PutA), employing the well-validated MODELLER (version 9.19) computational modeling program (25, 26). Specifically, we used the primary amino acid sequence of human PRODH as the target (20-516; GenBank ADD24775.1), the 3-D coordinates of the crystallography resolved 86-610 portion of the E. coli PutA dehydrogenase domain complexed to L-THFA (RCSB 1TIW) as the structural template (23), and our homology alignment was based on multiple PutA dehydrogenase sequences including Bdellovibrio bacteriovorus (27) and Sinorhizobium meliloti (28). The structurally determined position of L-THFA within PutA (1TIW) served as anchor for our homology placement of S-5-oxo and N-PPG structures within the human PRODH catalytic pocket.

To quantify cell viability, the CellTiter-Glo Cell Luminescence-based Viability Assay (Promega) was used according to the manufacturer's protocol and as described previously (29). Luminescence was measured using a Fluoroskan Ascent FL Luminometer (Thermo Fisher Scientific) to obtain triplicate sample 24-, 48-, and 72-hour timepoint determinations after treatment, normalized to that of control cells, and graphically displayed as mean (±SD) percent control values. Coefficient of drug interaction (CDI) values were determined as previously described by us and others (29, 30), with CDI = %viability(drugA + drugB)/%viability(drugA) × %viability(drugB).

The ON-TARGET plus SMART pool of four different siRNA oligonucleotides to PRODH (J-009543-05-0002, J-009543-06-0002, J-009543-07-0002, and J-009543-08-0002) and a nonspecific control siRNA (D-001810-01-05) were all obtained from Dharmacon (GE Dharmacon). Pooled and single oligos were transfected into replicate cell cultures using Lipofectamine 2000 (Thermo Fisher Scientific) as we have reported previously (31, 32).

Normalized, log₂-scaled transcriptome profiles from 51 different breast cancer cell lines, and their corresponding phenotypic assignment into triple negative, HER2⁺, or Luminal (ER or PR⁺, HER2⁻) subtypes, were obtained and analyzed for single gene (PRODH and GLS1) expression levels compared across breast cancer subtypes as reported previously (33, 34). Full genome transcript levels (RNAseq) on 817 primary human breast tumors, originally acquired and analyzed by The Cancer Genome Atlas (TCGA) research network (35, 36), were obtained from their public data deposit site (https://synapse.org), interrogated for PRODH and GLS1 mRNA levels (normalized and log₂-scaled transcript per million values), and graphed as box-whisker plots for each of the five different TCGA intrinsic breast cancer subtypes (basal-like, n = 136; HER2 enriched, n = 65; Luminal-A, n = 415; Luminal-B, n = 176; and normal-like, n = 25). Correlations between PRODH and GLS1 expression were tested by Pearson linear regression (Rp).

ZR-75-1 cells were plated on glass bottom microwell dishes from MatTek Corporation (P35G-1.5-14-C) in RPMI media and incubated with 5 mmol/L N-PPG or S-5-oxo for 24 hours; 2.5 μL of MitoTracker Green FM Molecular Probes (Cell Signaling Technology) added to 2 mL of charcoal-stripped media (Gibco, Thermo Fisher Scientific), and 1.5 mL of this mix then added to each well for 30 minutes after which two drops of NucBlue Live Stain Ready Probes Reagent (Invitrogen by Thermo Fisher Scientific) was added to DAPI counterstain cell nuclei. Immunofluorescent cell imaging was performed as previously described on cells plated and then treated on 4-well glass slides (Lab-TekII, MilliporeSigma) using a Zeiss LSM 780 Confocal Microscope (Zeiss) equipped with constant temperature/CO₂-regulated enclosure, under 63× oil immersion (31). Plated and treated cells were cross-linked (4% paraformaldehyde) and blocked (10% IGEPAL CA-630 NP 40 in PBS/DEPC-water with 5% goat serum), probed with the indicated primary (overnight) and secondary (90 minutes) goat anti-mouse or anti-rabbit antibodies, and then counterstained with ProLong Gold antifade reagent with DAPI (Molecular Probes by Life Technologies, Thermo Fisher Scientific) prior to imaging.

Snap frozen organs and xenografted tumors were first pulverized under liquid nitrogen and then sonicated in low salt buffer (10 mmol/L Tris pH 7.5, 50 mmol/L NaCl, 0.4% NP40 with protease and phosphatase inhibitors) and extracted as described for cell lines. Cells harvested at approximately 70% confluency were washed with ice-cold Dulbecco phosphate-buffered saline and then harvested in RIPA buffer [50 mmol/L Tris-HCl (pH 8.0), 150 mmol/L NaCl, 1% triton X-100, 0.5% sodium deoxycholate, 0.1% SDS] containing complete Mini, EDTA-free Protease Inhibitor Cocktail tablets (Roche) and PhosSTOP Phosphatase Inhibitor Cocktail tablets, (Roche); lysates were spun at 14,000 rpm for 5 minutes and supernatants collected. Intact mitochondria were isolated from the resected and pulverized mouse tissues or freshly harvested ZR-75-1 cells (2–4 500-cm² dishes at ∼90% confluence) as described previously (11); after centrifugation (10,000 × g for 10 minutes) the mitochondrial pellets were resuspended in STE (250 mmol/L sucrose, 5 mmol/L Tris-HCl, 2 mmol/L EGTA, pH 7.4) on ice, for either PRODH enzymatic assay or immunoblotting. Prior to immunoblotting, protein content was determined by Bradford Coomassie Assay Kit (Pierce) and then diluted into 2× Laemmli sample buffer. Immunoblotting was performed as described previously (29, 31, 32) using polyvinylidene fluoride membranes blocked with 5% nonfat milk in TBST (TBS with 0.1% tween-20) incubated with primary and then secondary antibodies conjugated to HRP; the resulting immunoblot signals were scanned for densitometry.

PRODH enzymatic activity was assessed on isolated ZR-75-1 mitochondria by two different bioassays, as we have described previously (11): (i) spectrophotometric detection of o-aminobenzaldehyde (oAB) reacting with the proline generated P5C, and (ii) time-dependent fluorescence spectrometry monitoring mitochondrial NADH levels as a function of substrate addition and inhibitor treatment. The first assay detects formation of the PRODH-specific enzymatic product as P5C-oAB, and the second assay allows for the time-dependent generation of mitochondrial NADH immediately upon addition of proline, pyruvate, or malate relative to control (no added substrate) or treatment condition. By comparing proline's NADH generating capacity with that of another substrate such as pyruvate or malate, this more sensitive NADH-generating assay enables the indirect bioassay of PRODH-specific inhibitors relative to those also affecting other FAD-containing oxidases and comparing the potency or stereospecificity of different enzyme inhibiting candidates. To measure mitochondrial NADH formation, solutions of 40 mmol/L proline and 40 mmol/L malate were first prepared, with or without the inhibitor to be tested, in DEPC-water (GeneMate) with 4 mg/mL rotenone (MilliporeSigma) in KHE buffer (120 mmol/L KCl, 3 mmol/L HEPES, 5 mmol/L KH₂PO₄ pH 7.2). Freshly isolated ZR-75-1 mitochondria (0.15 mg resuspended in 30 μL) were mixed with 165 μL of 4 mg/mL rotenone in KHE buffer (to final volume of 195 μL), and the mix aliquoted into a 96-well plate. Into replicate wells, 5 μL of 40 mmol/L solution of proline and/or malate was added followed by 5 μL vehicle or enzyme inhibitor, and the time-dependent accumulation of NADH was monitored every 8 seconds for up to 30 minutes on a PHERAStar FS fluorescent microplate reader (BMG LABTECH GmbH) with λ_excitation = 340 nm and λ_emission = 460 nm.

Flies: because their intracellular energy for flight and geotaxis is dependent on high rates of mitochondrial proline catabolism (5, 11), Drosophila melanogaster were chosen as a first model to assess the bioavailability and organismal effects of orally administered PPG. Raised on standard laboratory food until 7 days of age, male Oregon Red flies (Fly Base, Oregon-R) were then starved for 2 hours to synchronize their feeding and transferred to vials (20/vial) containing Whatman paper with 500 μL of 5% w/v sucrose in water ± 5 mmol/L N-PPG (and supplemented daily with 100 μL standard fly food). Fly activity and survival were monitored twice daily; and geotaxis was measured once daily by tapping flies to the bottom of the vial. For comparison with N-PPG–treated wild-type flies, the same measurements were performed on slgA-mutant flies lacking functional PRODH (5).

Xenografted mice: mouse studies reported here (PTC1797 and PTC1854) were all conducted under Institutional Animal Care and Use Committee approval (AN092211 and AN142193) at the UCSF Cancer Center's Preclinical Therapeutics Core (PTC). MCF7mutER knockin sublines were chosen for xenografting into 6-week old NCR nu/nu athymic female mice (Taconic Farms), as these generate subcutaneous MCF7 tumors that grow without the need for supplemental estradiol administration as we have described previously (32). Untreated mice from these earlier studies were used to assess the in vivo bioavailability and compare the distant tissue pharmacodynamic effects of intravenous, oral, and intraperitoneal administration of N-PPG. In PTC1797, 10 different tumor-implanted mice (#) were given either saline vehicle (#970) or N-PPG at 50 mg/kg × 3 (every other day) by either oral gavage (#965, #966, and #967), intravenous (#962, #925, and #968), or intraperitoneal (#963, #964, and #969) treatments. In PTC1854, 3 different mice were treated with nine daily oral doses of either saline (#3883) or 50 mg/kg N-PPG (#3840 and #3880), beginning 58 days after tumor implantation. The tolerance of these mice to either vehicle or N-PPG treatments was assessed by daily monitoring of animal health and activity and twice weekly measurement of animal body weight. In both of these mouse studies no animal lost ≥10% of body weight, and the overall health and activity of all N-PPG–treated mice appeared identical to that of controls. In PTC1854, tumor growth was measured by calipers along the largest (length) and smallest (width) axes, and tumor volumes were calculated by the following formula: tumor volume = [(length) × (width) × (width)]/2. At the conclusion of each mouse study, all animals were euthanized, kidneys (in PTC1797) and tumors (in PTC1854) were resected and snap frozen into liquid nitrogen within 2 hours of the final N-PPG treatment for subsequent mitochondrial extraction or whole-tissue protein extraction.

Technical and biological replicates were performed as described in figure legends, and data analyzed by GraphPad Prism 7 or Bioconductor R (www.bioconductor.org) software programs. Graphical plots show single, median, or mean (±SD) values. All replicate measures were statistically compared by ANOVA F-test (e.g., box-whisker plots) or two-tailed Student t test; and correlations were determined by Rp. Significant differences were determined as *, P < 0.05 or **, P < 0.01.

While cancer cells catabolize proline to produce anaplerotic carbon for energy (ATP) production and biosynthesis, the immediate product of proline catabolism by PRODH is the imino acid, pyrroline-5-carboxylate (P5C), in rapid equilibrium with glutamate semi-aldehyde that is further oxidized by P5C dehydrogenase to glutamate (Fig. 1A). As we have previously shown (11), mitochondrial P5C production can be directly detected as a colored adduct when reacted with oAB, forming P5C-oAB (Supplementary Fig. S1A); however, as we have also shown (11) a more sensitive albeit indirect measure of mitochondrial PRODH activity is to monitor the time-dependent accumulation of NADH upon mitochondrial exposure to proline in the presence of rotenone to prevent complex 1 reoxidation of NADH (Fig. 1A and B). By substituting for the substrate proline, this assay can detect and compare the oxidative formation of NADH by different mitochondrial enzymes (Supplementary Fig. S1B); and by adding proline and either pyruvate or malate sequentially after preexposure to a PRODH-specific inhibitor, this assay can be used to either compare active (L-THFA and 5-oxo) versus inactive (furoic and furan) structural analogues as competitive inhibitors or to test their stereospecificity (5 mmol/L S-5-oxo of R-5-oxo) as competitive inhibitors (Supplementary Fig. S1C). By comparing equimolar exposures to different inhibitors, we used this assay to show that R-5-oxo is inactive, that S-5-oxo is significantly more active than L-THFA, that the mechanism-based PRODH inhibitor N-PPG appears to be more potent than the competitive PRODH inhibitor S-5-oxo (Supplementary Fig. S1D), and that a series of propargyl structural analogues, including the isomer DL-PPG, cannot significantly inhibit PRODH enzymatic activity (Supplementary Fig. S2). Furthermore, when a 30 minutes mitochondrial washout step was inserted between inhibitor pretreatment and sequential substrate additions, washout completely eliminated the PRODH inhibitory effect of 5 mmol/L S-5-oxo while retaining the inhibitory effect of N-PPG, demonstrating the specific and irreversible inhibition of PRODH by N-PPG in the absence of any other inhibitory effect on mitochondrial NADH production by either malate or pyruvate oxidation (Fig. 1B; Supplementary Fig. S1B).

To better understand the chemical and biological basis underlying and differentiating PRODH inhibition by either S-5-oxo or N-PPG, and given that there is as yet no experimentally determined 3-D atomic level structure for mammalian PRODH, we employed a well-validated homology-based computational modeling program, MODELLER (25, 26), to produce a 3-D structure of human PRODH bound to these inhibitors by aligning the known primary amino acid sequence of human PRODH with the experimentally determined 3-D structural coordinates of the PutA dehydrogenase domain from E. coli (23). From our S-5-oxo–bound PRODH model (Fig. 1C), we observed its probable hydrogen bond formation with tyrosine (Y548) and arginine (R564 and R563) to stabilize this competitive inhibitor within the PRODH catalytic pocket. In contrast, we compared the expected prereactive configuration of N-PPG within PRODH with its presumed pocket stabilized covalent adduction to the N(5) of FAD (Fig. 1D). This postreactive N-PPG–PRODH structure is likely similar to the reported crystal structure of E. coli PutA irreversibly inactivated by N-PPG that showed bicovalent linkage of N-PPG with PutA FAD N(5) and the ϵ-amino of a PutA pocket lysine, K329 (37). This specific active site K residue is evolutionarily conserved and corresponds to K234 in human PRODH (Fig. 1D). Assuming a conserved mechanism of inactivation, we predict that the reaction of human PRODH with N-PPG results in a three-carbon covalent link between the epsilon nitrogen of K234 and the N5 atom of the reduced FAD (Fig. 1D). Because the covalently modified FAD is irreversibly locked into a reduced electronic state, the enzyme is unable to redox cycle and is therefore inactive. Notably, this type of bicovalent pocket reaction differs from a similar FAD-facilitated inhibition of human mitochondrial monoamine oxidase (MAO)-B where the PPG-like N-propagyl-1(R)-aminoindan (rasagiline) is covalently bound to FAD N(5) but not also to a pocket residue (38). Of likely functional significance, inactivation of human PRODH by N-PPG is predicted to cause major distortion of the active site, including butterfly bending of the FAD isoalloxazine, crankshaft rotation of the FAD ribityl chain into the reduced conformation, and movement of the α8 helix into an open conformation. All of these structural distortions are incompatible with binding by the substrate proline.

Confocal imaging of PRODH expression in ZR-75-1 cells treated over 72 hours with either vehicle, S-5-oxo, or N-PPG (5 mmol/L) showed the exclusive mitochondrial localization of PRODH as expected but revealed early (within 24 hours) decay of PRODH protein only upon N-PPG exposure (Fig. 2A). This microscopic observation was confirmed by Western blotting, which also showed that the N-PPG isomer, DL-PPG, does not induce PRODH decay (Fig. 2B), and further demonstrated the progressive and selective 24–72 hours loss of PRODH protein (normalized to β-actin and untreated control ratio) in the absence of any concurrent decline in either mitochondrial NADH-ubiquinone oxidoreductase NDUFS1 or cytosolic methylenetetrahydrofolate reductase MTHFR flavoprotein (Fig. 2C). The mitochondrial chaperonin, HSP60, is responsible for folding more than 300 mitochondrial matrix proteins and physically interacting with the mitochondrial protease, LONP, to maintain mitochondrial proteostasis as a proposed mediator of the mitochondrial unfolded protein response (UPR^mt; refs. 39–42). As such, we observed an induction of total cell and mitochondrial-localized HSP60 coincident with the initiation of PRODH decay during the first 24 hours of N-PPG exposure, that was not apparent following S-5-oxo exposure (Fig. 2D). While the preservation of NDUFS1 expression largely ruled out N-PPG–induced mitophagy (43), we employed confocal imaging to revisit the question of mitophagy versus UPR^mt induction by visually assessing mitochondrial membrane integrity and looking for membrane recruitment of the PINK1/PARKIN complex (known to recruit the mitophagic machinery) relative to mitochondrial import of the GRP-75 chaperone (known to be associated with UPR^mt) during the initial period (24 hours) of N-PPG–induced PRODH decay. In keeping with our Western blot data as described, confocal images of control and N-PPG–treated ZR-75-1 cell cultures showed early loss of mitochondrial PRODH expression without loss of mitochondrial membrane structure as assessed by the outer membrane protein TOM20 (Fig. 3A and B), no evident membrane recruitment of PINK1/PARKIN complex despite mitochondrial import of GRP-75 (Supplementary Fig. S3), and no decline in the extramitochondrial FAD-containing enzyme MTHFR (Fig. 3C and D). The right top and bottom panels (Fig. 3C and D) show a similar but independent experiment as that shown in the left panels (Fig. 3A and B), except that control and N-PPG–treated ZR-75-1 cells were stained for PRODH using different primary and fluorochrome-conjugated secondary antibodies to control for potential differences in primary immunoreactivity and nonspecific secondary antibody effects. Merged imaging confirmed the different subcellular localizations of PRODH and MTHFR, as well as the colocalization of mitochondrial PRODH and TOM20.

Because SlgA-null–mutant flies that lack systemic PRODH activity are known to be viable, fertile, and with normal lifespan, yet possess a characteristic and easily recognizable Sluggish-A phenotype (5), we fed fruit flies millimolar concentrations of N-PPG in sucrose for several days to assess its oral bioavailability and systemic activity. Within 48 hours the orally fed N-PPG (5 mmol/L) treated flies exhibited clear loss of geotaxis relative to control fed flies (Supplementary Video S1A), perfectly phenocopying the Sluggish-A phenotype (Supplementary Video S1B) without any loss in fly vitality, indicating its efficient oral bioavailability and systemic biological activity. On the basis of these fly results, studies of N-PPG bioavailability and tolerance were then undertaken in nude mice implanted with MCF7mutER xenografts capable of tumorigenic growth in the absence of exogenous estradiol supplementation (32). Untreated xenografted mice from two recently reported studies (PTC1797 and PTC1854) were used to assess the in vivo bioavailability and distant tissue pharmacodynamic effects of either intravenously, orally, or intraperitoneally administered N-PPG at a dose (50 mg/kg) estimated to achieve up to 5 mmol/L systemic blood concentrations (although direct assays of N-PPG were not available). In PTC1797, 10 different control and N-PPG–treated mice (#) were given either vehicle alone (#970) or N-PPG at 50mg/kg × 3 (every other day) by either oral gavage (#965, #966, and #967), intravenous (#962, #925, and #968) or intraperitoneal (#963, #964, and #969) administration. In a follow-up study, PTC1854, 3 different mice were treated with nine daily oral doses of either saline (#3883) or 50 mg/kg N-PPG (#3840 and #3880), beginning 58 days after tumor implantation. In both these murine studies no mouse lost ≥10% of body weight, and the overall health and activity of all N-PPG–treated mice appeared identical to that of vehicle-treated controls. On the basis of Genotype-Tissue Expression (https://commonfund.nih.gov/gtex) and Human Protein Atlas (https://www.proteinatlas.org/cell) databases indicating that human kidneys (especially tubule cells) express high PRODH mRNA and protein levels relative to many other normal organs, mouse kidneys were excised and evaluated at the conclusion of PTC1797 to look for biological evidence of N-PPG–induced PRODH decay. The total number of study mice was insufficient to statistically compare N-PPG effects based on different routes of administration; however, Fig. 4A demonstrates that following three (every other day) doses of N-PPG relative to vehicle treatment, the various routes of N-PPG administration (intravenous, oral, and intraperitoneal) were each able to affect at least 60% reduction in mouse kidney PRODH protein levels (normalized to ATP synthase). In the more limited PTC1854 study, N-PPG given orally over 9 days was evaluated for its potential tumor bioavailability and pharmacodynamic impact. As shown in Fig. 4B, despite the xenografted tumors' lack of normal vasculature and the relative resistance of the parental MCF7 breast cancer line's in vitro growth to continuous millimolar exposure of N-PPG, the in vivo–treated xenografts showed modest evidence of N-PPG–induced decay in tumor cell mitochondrial PRODH expression (∼20%–50% reduction, normalized to ATP synthase) with slight impact on the tumors' volumetric growth during that treatment period (Fig. 4C).

Because PRODH provides a bypass carbon source for glutamate-addicted and GLS1-overexpressing cancers (16), particularly those proving sensitive to clinically emergent GLS1 inhibitors like CB-839 (17, 18), we investigated the cancer cell relationship between PRODH and GLS1 expression using publically available transcriptome data from 51 different human breast cancer cell lines (34), and 817 primary breast tumors molecularly characterized and classified by the TCGA according to their intrinsic subtypes (36). In both, the cell line and primary tumor transcriptome datasets, PRODH mRNA expression varied significantly by intrinsic breast cancer subtype, with luminal and HER2⁺ tumors showing the highest and basal/triple-negative tumors showing the lowest levels of PRODH mRNA expression (Fig. 5A and B). Irrespective of breast cancer subtype, PRODH and GLS1 mRNA levels showed significant inverse correlations most pronounced among the cell lines (Rp = −0.52; P = 9.02E-5) but also apparent among the primary breast tumors (Rp = −0.12; P = 0.0006), and consistent with the idea that one or the other of these GLS1 and PRODH pathways are needed to feed breast cancer's anaplerotic addiction to glutamate (16). Comparing PRODH inhibitors, S-5-oxo and N-PPG, for their ability to at least additively enhance the antitumor activity of the GLS1 inhibitor, CB-839, we demonstrated this to be true against MCF7 breast cancer cells in culture (Fig. 5C) but not against the nonmalignant breast epithelial cell line, MCF10A, whose growth in culture remained unaffected by PRODH and/or GLS1 inhibition (Fig. 5D).

The variable expression of PRODH among human breast and other cancers is primarily due to its positive transcriptional control by the tumor-suppressing protein p53 (1, 6), as well as some negative control by MYC upregulation (44), consistent with the fact that most basal/triple-negative (and MYC upregulated) breast cancers possess mutationally inactivated p53, whereas most luminal (ER/PR⁺) and HER2⁺ breast cancers express wild-type and functionally active p53 protein (Fig. 5A, left). The luminal breast cancer lines MCF7, DU4474, and ZR-75-1, as well as the nonmalignant breast epithelial cells, MCF10A, express varying basal levels of p53 and PRODH, and both of these proteins are variably upregulated within 24 hours of exposure to an MDM2 antagonist like, nutlin-3 or MI-63 (Fig. 6A). As we and others have previously shown (45, 46), the posttranslational induction of p53 by such MDM2 antagonists produces either cancer cell senescence or apoptosis, the latter marked by an increase in PUMA and cleaved (c) PARP protein evident only in p53 wild-type–expressing cancer cells but not in more minimally PRODH-expressing nonmalignant cells like, MCF10A cells (Fig. 6A). Following siRNA knockdown of PRODH in luminal breast cancer cell lines, cPARP is induced and this apoptosis indicator is further enhanced by 24–48 hours pretreatment of siRNA-treated cells with a p53-upregulating MDM2 antagonist (Fig. 6B). Because both GLS1 inhibition and MDM2 antagonism each produced anticancer effects that were at least additive in combination with PRODH loss of function, we compared these treatments when given to DU-4475 or ZR-75-1 cells in combination with either the competitive (S-5-oxo) or suicide (N-PPG) PRODH inhibitors. As seen in Fig. 6C and numerically demonstrated by calculating a CDI for each two- and three-drug combination, significant synergy (CDI < 0.70; P < 0.05) was observed in both cancer cell lines and for nearly all two-drug (PRODH inhibitor + MDM2 antagonist or GLS1 inhibitor) and three-drug (PRODH inhibitor + MDM2 antagonist + GLS1 inhibitor) combinations, as might be expected for “synthetically lethal” interactions between these pathway-targeted therapeutics. When comparing the two types of PRODH inhibitors, the two- and three-drug combinations that included N-PPG nearly always appeared more cytotoxic than those involving S-5-oxo; however, because these differences did not reach statistical significance, more extensive in vitro and in vivo comparisons are needed to confirm whether N-PPG is indeed a more effective anticancer agent than S-5-oxo.

The quest for PRODH inhibitors goes back to more than 40 years when this was considered a possible approach to eradicate tsetse flies and prevent African trypanosomiasis (4). While the first mechanism-based PRODH inhibitor was proposed in 1993 (19), further progress in this effort was stymied until a decade later when Lee and colleagues generated the first crystal structure of a prokaryotic bifunctional precursor of PRODH known as PutA (2). The following year, Zhang and colleagues reported the first crystal structure of a bacterial PRODH complexed with the competitive inhibitor L-THFA (23). Subsequent crystal structures of other bacterial PRODHs showed that the binding mode of L-THFA is highly conserved (20, 28). Together, these structural studies employing prokaryotic forms of PRODH provided the foundation for our computer modeling of human PRODH bound to either the S-stereoisomer of 5-oxo, newly identified in this study as a second-generation competitive PRODH inhibitor more potent than L-THFA, or to the irreversible suicide inhibitor, N-PPG. Tanner and colleagues showed that N-PPG is a mechanism-based covalent inactivator of four different bacterial PRODHs, which have pairwise sequence identity as low as 27% (24, 27, 37). In each case, the structure of the inactivated enzyme shows a covalent link between the FAD N5 atom and a conserved active site lysine (K) residue. Homology modeling strongly suggests that human and bacterial PRODHs share a conserved active site structure, including the K residue implicated in the N-PPG inactivation mechanism (ref. 12; Fig. 1C and D).

With more recent elucidation of additional PRODH structures recognizing its evolutionarily conserved higher order (βα)₈-barrel structure (27, 28) predicted to form within residues 121–579 of human PRODH, along with the use of protein structure modeling algorithms it now appears that the bicovalent N-PPG inhibitory mechanism known to occur in bacterial PutAs may similarly occur in human PRODH. Unlike other propargylic (PPG-like) analogues like rasagiline, an FDA-approved selective and irreversible inhibitor of the FAD-containing MAO-B currently used to treat Parkinson disease (38), our model predicts that N-PPG bicovalently binds and irreversibly inhibits human PRODH by tethering FAD to a pocket K residue. Using our MODELLER-generated human PRODH structure, K234 appears to be the most likely candidate for pocket adduction by N-PPG, in agreement with a recent SWISS-MODEL homology solution showing proximity of this K234 residue to a pocket bound L-THFA (12). However, our model also suggests that another lysine at position K326, which is part of a conserved KLS motif similar to that found in E. coli PutA, however unlikely based on bacterial studies could become an alternative candidate for involvement in N-PPG–PRODH adduct formation, pointing to the critical need for an experimentally determined 3-D structure of human PRODH following renewed efforts to overcome the current challenges limiting bacterial production of recombinant human PRODH (47). In conjunction with our model suggesting that N-PPG covalently binds and distorts the human PRODH catalytic pocket, we also demonstrate that its activity as a suicide inhibitor in living cells is associated with induction of intramitochondrial PRODH protein decay, likely mediated by a UPR^mt mechanism and not similarly activated by a competitive PRODH inhibitor like S-5-oxo or by the N-PPG isomer, DL-PPG, which also fails to inhibit PRODH activity.

Mass spectrometry and NMR confirmed the ≥90% purity of our newly synthesized N-PPG, but given the ubiquity of flavoproteins like PRODH, our earliest concern focused on the specificity of N-PPG as a PRODH inhibitor. This concern was largely mitigated by prior knowledge that FAD association with the TIM barrel structure of PRODH occurs in only one other known instance (23), rendering PRODH's catalytic site a relatively unique molecular target not readily susceptible to other PPG-like structural analogues. Nonetheless, rather than assaying the inhibitory effect of N-PPG on PRODH by simply measuring mitochondrial P5C generation (the first product of proline catabolism by PRODH), we chose to monitor the time-dependent generation of mitochondrial NADH by oxidation of individually introduced substrates including proline, malate, and pyruvate, enabling us to show that N-PPG inhibits PRODH specifically and irreversibly, without affecting other mitochondrial NADH-generating enzymes. Following our initial and surprising observation that N-PPG, but not S-5-oxo or DL-PPG, treatment of cultured cancer cells produces an early (within 24 hours) decline in PRODH protein levels, we employed both confocal microscopy and Western blotting to confirm the specificity of this protein-degrading effect on PRODH not seen in other mitochondrial or extramitochondrial FAD-containing enzymes, and occurring without loss of total cell mitochondria, their outer membrane constituent TOM20 or its recruitment of PINK1, thereby excluding N-PPG induced mitophagy (43).

Understanding that UPR^mt is often experimentally activated by introducing a structural distortion in a single intramitochondrial protein (41), and given our model's suggestion that bicovalent adduction of N-PPG to FAD and a nearby residue like K234 produces structural distortion of the PRODH catalytic pocket, we probed for and observed N-PPG–specific induction of two mitochondrial chaperone proteins, HSP60 and GRP-75, coincident with PRODH decay and consistent with mechanistic induction of UPR^mt (40–42). Of interest, targeting mitochondrial HSP60 has very recently been shown to be a promising new anticancer strategy with little, if any, toxic impact on normal cells (42), suggesting that the combination of PRODH-targeted N-PPG therapy with an HSP60-targeted agent is worth exploring.

Although an orally bioavailable propargylic analogue that inhibits and covalently reacts with its FAD-containing enzymatic target has been successfully developed and approved for medical use (38), the in vivo administration of N-PPG has never been reported. In fact, only very recently has the first in vivo administration of a PRODH competitive inhibitor, L-THFA, been reported (21). Giving daily intraperitoneal injections of up to 60 mg/kg of L-THFA into mice bearing small orthotopic implants of murine breast cancer cells, these investigators observed excellent host tolerance to this competitive PRODH inhibitor and, after 16–18 days of sequential treatment, showed that it can reduce pulmonary metastasis formation by 50% without any significant impact on primary tumor growth (21). We explored in vivo administration of N-PPG first in flies, where oral consumption of 5 mmol/L N-PPG (in sucrose) led to loss of geotaxis within 48 hours, phenocopying the SlgA-mutant fly strain lacking all PRODH activity; and this N-PPG–induced loss of Drosophila flight muscular energy occurred without detectable loss in fly vitality or fertility. Using nude mice xenografted with engineered human breast cancer cells capable of estrogen-independent in vivo tumor growth, whose parental cell line (MCF7) is only modestly growth inhibited in vitro by 5 mmol/L N-PPG, we observed excellent host tolerance to repeated 50 mg/kg administration of N-PPG given by either oral, intraperitoneal, or intravenous routes of administration. Although not powered to statistically compare N-PPG systemic bioavailability between these different routes of administration or to comprehensively assess N-PPG effects on xenografted tumor growth, these mouse studies clearly demonstrated that N-PPG at this dose and schedule via all routes of administration could reduce PRODH protein levels in both normal tissue (kidneys) and xenografted tumors. The ability to read out a systemic pharmacodynamic effect from N-PPG as tissue/cell loss of mitochondrial PRODH protein expression, apart from signifying a functionally different cellular consequence induced by this mechanistically unique inhibitor, offers a novel future advantage in monitoring tissue exposure to this suicide PRODH inhibitor, unlike the more complex metabolomic measures required to monitor tissue or tumor responses to a competitive PRODH inhibitor, like L-THFA (21). Given that both normal and tumor tissues react similarly to N-PPG with degradation of mitochondrial PRODH protein, likely occurring by a common UPR^mt mechanism, and appreciating that inhibition of PRODH function is deleterious to malignant but not normal mammalian cells (21), it is noteworthy that UPR^mt induction is now considered a promising new cancer treatment strategy in itself (39) that is not only nontoxic to normal cells but, in fact, increasingly recognized as beneficial to normal growth and development by enhancing cell vitality and potentially increasing organismal longevity (41).

As with other anticancer strategies, inhibiting PRODH is not to be envisioned as a monotherapy, especially given that metabolic reprograming as a hallmark of cancer is highly adaptive and subject to, as well as capable of, inducing epigenetic shifts in cancer cell phenotypes associated with altered responsiveness to a wide variety of anticancer agents (13–15). One proven approach to identifying more effective anticancer strategies now being considered in the context of oncometabolic targets is the identification of synthetic lethality based treatment combinations (15). In this study, we demonstrate strong biological interrelationships underlying cancer cell mitochondrial addiction to glutamate, manifest as either PRODH or GLS1 dependence, and between PRODH transcriptional stimulation and MDM2 antagonists that upregulate p53 wild-type. Common to the preclinical studies of all three of these emerging cancer targets (MDM2, GLS1, and PRODH) is that their individual inhibition in normal or nonmalignant host tissues is not cytotoxic and is commonly well-tolerated (16, 21, 46), although their combinatorial tolerance in vivo remains to be assessed. We now provide experimental rationale compelling future in vivo evaluation of synthetic lethality based two- and three-drug combinations involving a PRODH inhibitor (e.g., S-5-oxo or N-PPG), a GLS1 inhibitor (e.g., CB-839), and a p53-upregulating MDM2 antagonist (e.g., MI-63 or nutlin-3). In particular, based on our chemical biology modeling and because of its novel pharmacodynamic advantage in terms of inducing measurable reductions in distant tissue/tumor mitochondrial PRODH expression, the suicide inhibitor N-PPG may now be considered the preferred drug of choice in pursuing future in vivo studies focusing on the anticancer utility of PRODH inhibition. Therefore, given its pharmacodynamic advantage, as well as it's seemingly greater potency over competitive inhibitors like L-THFA or S-5-oxo, the suicide inhibitor N-PPG should be advanced further into preclinical studies designed to exploit and evaluate its potential synthetic lethal interactions with p53 upregulation and inhibition of GLS1.

No potential conflicts of interest were disclosed.

Conception and design: G.K. Scott, S. Khateeb, C.C. Benz

Development of methodology: G.K. Scott, S. Khateeb, S. Mahoney, C.C. Benz

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G.K. Scott, B.C. Becker, M.B. Jensen, B. Hann, B.J. Cowen, S.D. Pegan, C.C. Benz

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G.K. Scott, C. Yau, B.C. Becker, S. Khateeb, M.B. Jensen, B.J. Cowen, S.D. Pegan, C.C. Benz

Writing, review, and/or revision of the manuscript: B.C. Becker, S. Khateeb, M.B. Jensen, B.J. Cowen, S.D. Pegan, C.C. Benz

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Khateeb, C.C. Benz

Study supervision: G.K. Scott, S. Khateeb, C.C. Benz

We appreciate the technical contributions from Buck Institute scientists Katya Frazer, Daniel Rothschild, and Mauricio Ortega. In addition, we wish to sincerely thank one of the anonymous expert reviewers of this article for specific suggestions that substantially enhanced the accuracy and presentation of our study results. This work was supported by Elizabeth MA Stevens memorial funding (to C.C. Benz) and Alfred Benzon Fellowship and NIA-1K99AG056680 (to M.B. Jensen).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.