Selective Vulnerability to Pyrimidine Starvation in Hematologic Malignancies Revealed by AG-636, a Novel Clinical-Stage Inhibitor of Dihydroorotate Dehydrogenase

Rapidly proliferating cells reprogram metabolism to support increased biosynthetic demands, a feature that can expose targetable vulnerabilities for therapeutic intervention. Agents targeting nucleotide metabolism have been widely adopted for treatment of proliferative disorders such as cancer and autoimmunity. Cellular demands for nucleotides are elevated in the setting of T-cell activation, during which pyrimidine ribonucleotide pools can expand up to 8-fold (1), a process requiring stimulation of the de novo pyrimidine biosynthesis pathway. In this setting, nucleoside salvage pathways are likely insufficient to satisfy elevated nucleotide demands. Inhibitors targeting dihydroorotate dehydrogenase (DHODH), a key enzyme in de novo pyrimidine biosynthesis, are currently in clinical use for the treatment of rheumatoid arthritis (leflunomide) and multiple sclerosis (teriflunomide), with the presumed mechanism of action involving antiproliferative effects on effector lymphocytes (2–4).

Brequinar, a more potent DHODH inhibitor (5), was pursued as an anticancer strategy in the 1990s, prompted by preclinical activity in several solid tumor types (6). Clinical studies were discontinued, however, owing to lack of evidence of efficacy in phase II studies encompassing diverse solid tumor types (7). Questions remain as to whether this lack of clinical efficacy was due to suboptimal dosing regimens resulting in a lack of sustained DHODH inhibition, or to a failure to enrich for patients with tumor types most likely to respond.

There has been a recent resurgence in interest in targeting DHODH in the oncology setting, with four different molecules currently in phase I/II clinical trials in acute myeloid leukemia (AML). AML was suggested as a target indication for DHODH inhibitors in a study by Sykes and colleagues, in which an unbiased screen for compounds that can overcome blockade of leukemia cell differentiation uncovered molecules targeting DHODH as top hits (8). In vivo validation studies with brequinar in systemic mouse leukemia models demonstrated reduced leukemic cell burden and leukemia-initiating cells, and significantly improved survival. Subsequent studies by other groups have proposed specific contexts in solid tumor types that may elicit enhanced dependence on DHODH. In particular, metabolic rewiring in response to chemotherapy-induced DNA damage, PTEN loss, or KRAS mutation was suggested to render tumors particularly dependent on de novo pyrimidine synthesis and to thereby sensitize to DHODH inhibition (9–11). Most recently, a study by Cao and colleagues indicated selective sensitivity of cell lines of hematologic origin compared with solid tumor cell lines to PTC299, a molecule initially developed as an inhibitor of stress-induced VEGFA protein synthesis and later identified as a DHODH inhibitor (12).

Heterogeneity in metabolic dependencies across cancers exists (13), and identifying such tumor subtype–specific metabolic vulnerabilities could lead to novel treatment strategies that are genotype agnostic. We performed a chemical biology screen utilizing compounds targeting a diverse set of metabolic enzymes and identified potent and selective activity of a novel DHODH inhibitor, AG-636, in a distinct subset of tumor cell lines, with an enrichment for sensitivity in cell lines of hematologic origin. Given several recent publications describing activity of DHODH inhibitors in leukemia, here we focus on our work characterizing AG-636 activity in lymphoma models, though we found that leukemia models were comparably sensitive both in vitro and in vivo. Lymphoma tumor cells exhibited broad dependence on DHODH, regardless of subtype. Further characterization of AG-636 confirmed on-target cellular activity and translation of selectivity profile from the in vitro to the in vivo setting. DHODH inhibitor–sensitive cell lines did not exhibit an impaired ability to salvage extracellular uridine, and differential effects of AG-636 on growth and viability occurred even in the setting of depleted extracellular uridine. A CRISPR depletion screen and proteomics analyses were carried out to evaluate in an unbiased manner factors contributing to response to DHODH inhibition and pointed to DNA-damage response and DNA double-strand break (DSB) repair pathways as likely adaptive mechanisms. In support, cotreatment with checkpoint kinase 1 (CHK1) or DNA-dependent protein kinase (DNA-PK) inhibitors resulted in synergistic effects with AG-636. Taken together, our findings demonstrate a particular vulnerability of cancer cells of hematologic origin to inhibition of de novo pyrimidine biosynthesis, and point to adaptive mechanisms to supply pyrimidine pools and/or to cope with nucleotide stress/DNA damage in cancer cells of solid tumor origin.

AG-636 was synthesized as described in the Supplementary Methods. Brequinar was purchased from Sigma.

A sample for cocrystallization was prepared by incubating purified DHODH (27 mg/mL) with 1 mmol/L AG-636, 1 mmol/L FMN, and 2 mmol/L orotate on ice for 1 hour. The random microseed matrix-screening (14) was performed at 20°C by the sitting-drop vapor-diffusion method using seeds of lysozyme crystals for the initial screening. Crystals were grown by the sitting-drop vapor-diffusion method by mixing 200 nL of co-complex sample with 80 nL of precipitant solution (0.1 M sodium acetate pH 4.8, 2.2 M ammonium sulfate, 40 mmol/L N,N-dimethylundecylamine-N-oxide, 20 mmol/L N, N-dimethyldecylamine-N-oxide), and 20 nL of seeding solution equilibrated against reservoir solution (0.1 M sodium acetate pH 4.8, 2.5 M ammonium sulfate, 30% glycerol) and incubation at 20°C. Rectangular crystals of size 20–50 μmol/L were harvested in ∼10 days, cryoprotected in reservoir solution, and flash frozen in liquid nitrogen.

All diffraction data were collected at the Shanghai synchrotron radiation facility beamline 17U1 using Eiger 16M detector at 100 K and processed and scaled with HKL2000 (15) to a resolution limit of 1.97 Å. Initial phases were derived by performing molecular replacement using existing DHODH crystal structure (3zws.pdb) and were refined by manual model building and iterative refinement using COOT (16), PHENIX (17), and REFMAC (18). The Ramachandran analysis for this structure showed 96.08% favored, 3.02% allowed, and 0.90% outliers. The data collection and structure refinement statistics are summarized in Supplementary Table S4. All figures representing structures were prepared with PyMOL (https://www.pymol.org). Atomic coordinates, and experimental structure factors have been deposited at the RCSB Protein Data Bank, accession code 6VND.

Cell lines were purchased from ATCC or DSMZ and cultured in vendor-recommended media or Gibco advanced RPMI-1640 supplemented with 10% dialyzed FBS, 2 mmol/L glutamine, and 25 mmol/L HEPES where indicated. Cell lines were authenticated by 16-marker STR profiling and verified as mycoplasma-free by PCR-based methods performed by IDEXX BioAnalytics.

Cell growth assays were performed either using CellTiter-Glo (CTG) as a readout of viable cells or by real-time imaging of cell confluence. For CTG assays, cells were plated in 96-well tissue culture plates (Corning) at either 10,000 cells/well for cell lines of hematologic origin or 2,000 cells/well for solid tumor–derived cell lines. Cells were incubated with compound or DMSO control for 96 hours. CTG readings were performed at time 0- (T0) and at time 96-hour (T96) using a Molecular Devices SpectraMAX Paradigm plate reader. Relative growth rates (μ/μ_max) were calculated using the CTG T0 and T96 ATP measurements according to the following formula:

where T is the signal measure for the drug-treated arm at T96 and V is the DMSO vehicle-treated control measure at T96. V₀ is the vehicle-treated control measure at T0. A value of 1 indicates no growth inhibition, 0 indicates complete growth inhibition, and a value <0 indicates cell death.

For growth assays based on confluence, cells were plated as above and allowed to settle onto the bottom of the plate overnight prior to addition of compound or DMSO control. Confluence was measured every 2 hours using the IncuCyte Live-Cell Analysis System (Sartorius). To correlate growth with uridine depletion, culture medium was collected from duplicate plates, and uridine concentration was measured by liquid chromatography (LC) mass spectrometry (MS).

Cells were plated as described above and treated in a dose response matrix using nine doses of prexasertib in a 3-fold serial dilution and five doses of AG-636 in a 3-fold serial dilution in advanced RPMI with 10% dialyzed FBS. CTG readings were performed at T0 and T48. Relative growth rates (μ/μmax) were calculated as above. The combination index (CI) was calculated as previously described (19).

Metabolites from cells or tumors were extracted in 80/20 methanol (MeOH)/H₂O and analyzed by LC-MS. Details are provided in Supplementary Methods.

Animal studies were carried out under Institutional Animal Care and Use Committee–approved protocols (#2015–002 and #2018–004), and institutional guidelines for the proper and humane use of animals were followed. AG-636 was formulated in 0.5% methylcellulose in water. Transgenic female 6–8-week-old CB17/Icr-Prkdc^scid/IcrIcoCrl (CB17 SCID) mice (Charles River Laboratories Stock #236) were inoculated with 7 × 10⁶ OCILY19 or 5 × 10⁶ Z138 cells in Matrigel (1:1). Female 6–8-week-old BALB/cAnNCrl (BALB/c) mice (Charles River Laboratories Stock #028) were inoculated with 5 × 10⁶ HCT116 cells or 1 × 10⁷ A549 cells. Details are provided in Supplementary Methods.

The concentrations of AG-636, DHO, and uridine in plasma and tumor tissue were determined using nonvalidated LC with tandem MS methods. Details are provided in Supplementary Methods.

A VSV-G pseudotype lentiviral plasmid library containing single guide RNAs (sgRNAs) was generated using the Cellecta Three-Module Human Genome-Wide CRISPR construct library with the pRSG16-U6-sg-HTS6C-UbiC-TagRFP-2A-Puro plasmid. A lentiviral plasmid encoding stable Cas9 from Cellecta (pR-CMV-Cas9–2A-Blast) was introduced into A549 cells. The sgRNA library was introduced into the Cas9-expressing cells at a multiplicity of infection of 0.3 and selected under antibiotics for 48 hours. A sample of cells was collected, and then 1 μmol/L AG-636 or 0.05% DMSO was added. Cell culture medium was changed every 2 days until the cells completed 12 doublings. Genomic DNA was extracted using phenol:chloroform followed by ethanol precipitation. sgRNAs were amplified by PCR using primers specific for the regions of the plasmid flanking the sgRNA (FwdU6-1: 5′-CAAGGCTGTTAGAGAGATAATTGG-3′ R2: 5′-CGACAACAACGCAGAAATTTTGAAT-3′). The PCR product was sequenced using the Hiseq4000 sequencing platform with 80 to 100 million single-end 50 base pair reads.

The constant fixed sequence immediately before and after the 20-nucleotide CRISPR sgRNA sequence was first removed before read mapping using AWK command in Unix. To determine the read count value for each sgRNA, the trimmed reads were mapped to a custom sgRNA reference library using Bowtie2 (version 2.1.0) with option -L 20 -N 0 -k 1. These criteria retained only the perfectly matched reads for downstream analysis. The raw read count from each of three separate Cellecta Human Genome-Wide CRISPR modules was used to calculate gene-level statistics using Mageck (version 0.5.4) and DrugZ (version 1.1.0.2) with default options. The gene-level statistics from each module were merged together after Mageck and DrugZ processing.

Cells were cultured at 1 × 10⁶/mL (OCILY19) or 5 × 10⁵/mL (A549) in standard culture media containing 1 μmol/L AG-636 or DMSO for 24 (OCILY19) or 28 (A549) hours. Cells were collected, washed in PBS, and snap frozen. Samples were processed and analyzed as described (20). Details are provided in Supplementary Methods.

Cells were plated at 1 × 10⁶/mL (lymphoma lines) or 3 × 10⁵/mL (solid tumor lines) in advanced RPMI containing 10% dialyzed FBS, 2 mmol/L glutamine, and 25 mmol/L HEPES. Compounds were added at the indicated doses and incubation times. Protein was extracted in RIPA extraction buffer containing phosphatase and protease inhibitors, and 30 μg of total protein was electrophoresed on 4% to 12% bis–tris NuPAGE gels (Thermo Fisher). Proteins were transferred to a nitrocellulose membrane using the iBlot 2 dry transfer system, blocked in Odyssey TBS blocking buffer (LI-COR), and incubated with the indicated antibodies overnight (Cell Signaling Technologies) followed by a 1-hour incubation with the appropriate secondary antibody (LI-COR) and imaged on the LI-COR Odyssey CLx system.

To identify selective metabolic vulnerabilities within tumor subtypes, a panel of 395 cell lines was screened with a diverse array of compounds targeting metabolic enzymes. AG-636 (Fig. 1A), a DHODH inhibitor, stood out as having strong growth-inhibitory activity in a distinct subset of cell lines. Of the 395 cell lines screened, 28 (∼7%) scored as sensitive, exhibiting at least 75% growth inhibition (GI) and a GI₅₀ (half-maximal growth-inhibitory concentration) <1.5 μmol/L (Fig. 1B). Sensitivity was enriched (P < 0.001) in cell lines of hematologic origin, while solid tumor lines tended to be largely insensitive (Fig. 1B; Supplementary Fig. S1A). The area under the curve (AUC) of GI versus AG-636 concentration was calculated as an additional metric to compare degree of AG-636 response across cell lines of different tissues of origin. Cell lines with the strongest responses (high AUC values) tended to be of hematologic origin, though some cell lines of solid tumor origin also exhibited comparably high sensitivity (Fig. 1C). Publicly available data from a genome-wide CRISPR screen across 342 cell lines spanning 27 cell lineages (21) similarly revealed a particularly high dependency on DHODH in cancer types of hematologic lineage (Supplementary Fig. S2). Due to a moderate correlation observed between in vitro growth rate and response to AG-636 (Supplementary Fig. S1B), we aimed to utilize cell lines with comparable growth rates in our mechanistic studies evaluating differential sensitivity to AG-636.

A more focused cell line panel consisting of 49 B-cell lymphoma lines derived from different clinical subtypes [e.g., activated B-cell (ABC) and germinal center B-cell (GCB) diffuse large B-cell lymphoma (DLBCL), double-hit lymphoma, and mantle cell lymphoma] was examined for DHODH dependence using brequinar. The vast majority (40/49, 82%) of the B-cell lymphoma cell lines tested were sensitive [concentration of drug that resulted in a 50% reduction in growth rate (GR₅₀) <1.5 μmol/L], regardless of subtype (Fig. 1D).

To understand the molecular mechanism of DHODH inhibition by AG-636, we determined the high-resolution (1.97 Å) crystal structure of its quaternary complex with human DHODH, flavin mononucleotide (FMN), and orotic acid (DHODH•FMN•ORO•AG-636). AG-636 binds in the ubiquinone binding pocket adjacent to FMN and orotic acid and is encapsulated by the two N-terminal helices (Fig. 2A). The benzotriazole group of AG-636 stacks above H55 and hosts hydrogen bond interactions with Y355 (Fig. 2B). The nearly coplanar conformation of the carboxylic acid group enables salt bridge interactions with R135, a hydrogen bond directly to Q46, and interactions with T359 via a bridging water molecule. The methyl–biphenyl group binds in an alternative conformation, extending into the hydrophobic channel with apolar interactions to amino acids in the pocket (Y37, M42, L45, A58, F61, T62, L66, L67, P363, and L358). Unique polar interactions observed for AG-636 contribute to the enhanced inhibition potency compared with teriflunomide, which binds in the same pocket (22).

AG-636 is a potent [half-maximal inhibitory concentration (IC₅₀) = 17 nmol/L], reversible DHODH inhibitor exhibiting rapid-equilibrium (fast-on/fast-off) kinetic behavior (Supplementary Fig. S3A and S3B), and an uncompetitive mode of inhibition with respect to the substrate dihydroorotate (DHO; Supplementary Fig. S3C). Results are consistent with a ping-pong catalytic mechanism of DHODH and reports of a similar class of DHODH inhibitors (23) and also consistent with the X-ray crystallographic data showing that AG-636 binds in the ubiquinone binding site.

The cellular effects of AG-636 were evaluated by broad metabolic profiling on a panel of six cell lines treated with or without 1 μmol/L AG-636 for 20 hours. Consistent with perturbation of the de novo pyrimidine biosynthesis pathway at the level of DHODH, pool sizes of pathway metabolites upstream of DHODH, ureidosuccinic acid and DHO, were elevated up to >1,000-fold (P < 10⁻¹⁵). Less dramatic, albeit significant, increases were also observed in the levels of aspartate (2.3 × ↑; P < 10⁻¹⁵), a substrate required for the second step in the pathway (Fig. 2C and D; Supplementary Table S1). Metabolites with the most significantly decreased levels largely consisted of pyrimidine nucleotides downstream of DHODH (e.g., UDP, UTP, and CTP). In contrast, levels of purine nucleotides were not similarly decreased at this timepoint (Fig. 2D and E; Supplementary Table S1), consistent with on-target pathway inhibition by AG-636.

The relationship between AG-636 concentration, DHODH inhibition (as monitored by DHO increase), and tumor growth inhibition (TGI) was assessed in the OCILY19 DLBCL tumor xenograft model. Complete tumor stasis was observed (102% TGI) at a dose of 100 mg/kg twice daily (b.i.d.), with minimal coincident body-weight loss (Fig. 3A; Supplementary Fig. S4A). Reduced levels of antitumor activity and less DHO accumulation (Fig. 3A and B) were observed at the 30 and 10 mg/kg b.i.d. dose levels (42% and 27% TGI, respectively), consistent with the pharmacokinetic data (Supplementary Fig. S4B), suggesting a clear association between AG-636 exposure, target engagement, and TGI. Notably, even stronger antitumor activity was achieved with AG-636 (100 mg/kg b.i.d.) in an ibrutinib-resistant (24) xenograft model of mantle cell lymphoma, Z138, with complete tumor regression occurring in all mice (Fig. 3C). Strong TGI was also observed in a TP53-mutant mantle cell lymphoma model (25), JEKO1 (Supplementary Fig. S5).

To evaluate whether the selective sensitivity to inhibition of DHODH in vitro extended to the in vivo setting, two solid tumor xenograft models, A549 (lung) and HCT116 (colon), were examined for AG-636 antitumor activity. In contrast to the strong response observed in the lymphoma models, comparatively poor TGI was achieved with AG-636 in these models (27% in A549, 40% in HCT116; Fig. 3D and E), consistent with the in vitro sensitivity data (Fig. 3F). The reduced antitumor activity of AG-636 in the A549 and HCT116 models compared with the lymphoma models could not be attributed to reduced tumor AG-636 exposure (Fig. 3G). Evidence of target engagement as monitored by tumor DHO levels was observed in the treated HCT116 and A549 tumors, though absolute levels were 2 to 4 times lower than in the lymphoma tumor models (Fig. 3G; Supplementary Fig. S5B). In addition, tumor uridine concentrations were not decreased (A549) or decreased to a lesser degree (19%↓ in HCT116) by AG-636 in the solid tumor xenograft models compared with a 39% to 44% decrease in the lymphoma models (Fig. 3G). As these analyses were performed at the end of study and on tumor samples derived from mice treated with AG-636 for varying lengths of time, a separate study was performed in which tumor samples were harvested from mice following administration of two doses of AG-636 or vehicle control. Consistent with the previous results, metabolic analysis of these tumor samples showed lower total DHO accumulation and less pyrimidine (UMP and CMP) depletion in the two solid tumor models compared with the hematologic models (Fig. 4A and B). In contrast, purine nucleotide pools (AMP and GMP) remained unchanged by AG-636 treatment in all models (Fig. 4C).

To determine whether the reduced accumulation of DHO and lack of effect on tumor pyrimidines by AG-636 in the colon and lung tumor models resulted from incomplete DHODH inhibition or reduced dependency on DHODH, we performed an additional study comparing effects of 100 versus 600 mg/kg AG-636 in HCT116 tumors. Despite the approximately dose-proportional increase in AG-636 concentration, DHO levels were not further increased by the 600 mg/kg AG-636 dose (Supplementary Fig. S6A–S6C), suggesting that maximal DHODH inhibition was likely achieved at the 100 mg/kg dose level.

As cells can synthesize pyrimidine nucleotides from salvage pathways in addition to de novo, we evaluated whether differential uridine salvage capabilities could in part underlie the varied response to DHODH inhibition. First, we confirmed the ability of supplemented uridine to rescue the metabolic and growth effects incurred by AG-636. Supplementation with 100 μmol/L uridine restored the levels of pathway metabolites downstream of DHODH, as shown for UDP-glucose (Supplementary Fig. S7A), and completely rescued the growth-inhibitory effects of AG-636 (Supplementary Fig. S7B). Lower concentrations of uridine (5 and 25 μmol/L) resulted in intermediate effects on growth inhibition by AG-636. The AG-636–induced increase in DHO levels was also blunted by uridine supplementation (24 × ↓with 100 μmol/L; Supplementary Fig. S7A), likely due to negative feedback by elevated levels of the pathway product UTP on activity of carbamoyl phosphate synthetase II, the enzyme responsible for catalyzing the first step of the de novo pyrimidine biosynthesis pathway (26).

We next monitored uridine uptake as a function of cell growth with or without AG-636 (10 μmol/L). Cell growth was measured in real time by IncuCyte imaging and quantification of confluence. The AG-636–sensitive lymphoma (Z138, OCILY19, and JEKO1) and insensitive solid tumor (A549 and HCT116) lines effectively took up uridine, though Z138 cells preferentially did so in the setting of inhibited DHODH (Fig. 5A). In the other four cell lines, uridine uptake was unaffected by AG-636 treatment, suggesting that in the presence of sufficient extracellular uridine, the salvage pathway predominates and is sufficient for satisfying pyrimidine demand. Culture of cells with isotopically labeled uridine (± AG-636) and tracing of label incorporation into UMP confirmed uridine uptake as a readout of functional uridine salvage pathway activity (Fig. 5B). Notably, impairment of cell growth by AG-636 did not occur until uridine was depleted (or near depleted) from the medium (Fig. 5A). Although the two solid tumor lines continued to grow, albeit at a slower rate, after extracellular uridine was depleted, the confluence of all three lymphoma cell lines immediately halted (Z138) or declined (OCILY19 and JEKO1), indicative of cell death. Given that the lymphoma lines were proficient for extracellular uridine salvage, these results implicate other mechanisms driving the differential growth-inhibitory effects of AG-636.

To understand this further, we next evaluated for differential potency of AG-636 in inhibiting DHODH and de novo pyrimidine biosynthesis across cell lines. OCILY19, Z138, JEKO1, HCT116, and A549 cells were labeled with ¹⁵N(amide)-glutamine, and the concentration of ¹⁵N-UMP was measured. Experiments were performed under conditions in which extracellular uridine was depleted to ensure maximal usage of the de novo biosynthesis pathway. Although a greater degree of pathway inhibition occurred in the lymphoma cell lines at lower AG-636 concentrations, >95% pathway inhibition was achieved in all cell lines examined at AG-636 concentrations above approximately 1 μmol/L (Fig. 5C). Thus, at 10 μmol/L AG-636 (the concentration used in the above IncuCyte experiment), the de novo pyrimidine biosynthesis pathway is comparably inhibited in the sensitive lymphoma and insensitive solid tumor lines. To determine if de novo pyrimidine biosynthesis pathway inhibition in the cell lines used above led to a comparable impact on total intracellular pyrimidine pools, UMP pools were assessed at two different timepoints following 10 μmol/L AG-636 treatment in the absence of extracellular uridine. At 4 hours posttreatment, UMP pools were reduced by >80% in all cell lines, with the exception of the most AG-636–insensitive cell line, A549, in which a less pronounced drop of ∼60% was observed (Fig. 5D). By 20 hours, UMP pools were reduced by >80% in all five lines. The delayed kinetics of UMP pool drop in A549 cells by a concentration of AG-636 that maximally inhibits de novo pyrimidine biosynthesis, and in the absence of extracellular uridine, hints at additional mechanisms in place to adapt to blockade of pyrimidine supply. We next took unbiased approaches to identify such potential adaptive mechanisms.

A genome-wide CRISPR depletion screen was performed in A549 cells with or without AG-636 to identify specific genes and pathways limiting response to AG-636. Uridine was provided in the medium at a concentration of 5 μmol/L, resulting in alternate periods of uridine availability and depletion over the course of the culture period. Two of the strongest hits were genes involved in uridine uptake (SLC29A1) and its phosphorylation to UMP (UCK2; Supplementary Table S2), consistent with the ability of extracellular uridine salvage to rescue the effects of AG-636. The top depleted hit was the mTOR negative regulator TSC1. Of note, Wyant and colleagues recently described a role for the mTOR pathway in the negative regulation of ribophagy, a process wherein degradation of ribosomes can result in increased free nucleoside pools (27). Consistent with loss of mTOR-negative regulation sensitizing to DHODH loss, pharmacologic inhibition of mTOR antagonized response to AG-636 (Supplementary Fig. S8).

Gene ontology (GO) analysis of CRISPR screen hits with a P value < 0.05 was performed to identify pathways or processes impacting AG-636 sensitivity. DNA DSB repair/DNA-damage response pathways stood out as enriched GO terms (Fig. 6A). Gene targets involved in multiple different repair pathways, including single-strand annealing (OGG1 and BLM), nonhomologous end joining (RIF1, MAD2L2, and SMCHD1), and homologous recombination (GEN1, RECQL5, and SAMHD1) were among the significantly depleted hits (Fig. 6B; Supplementary Table S2).

Involvement of DNA-repair pathways in response to AG-636 was also implicated by proteomic analysis of A549 cells treated ± AG-636. Cells were treated for 28 hours under culture conditions that resulted in depletion of uridine from the medium within 16 hours (Supplementary Fig. S9); thus, the cells experienced a period of uridine availability followed by a period of uridine depletion (comparable to the conditions in the CRISPR screen). GO analysis of proteins with at least a 1.5-fold increase in expression in the presence of AG-636 compared with vehicle only (DMSO) identified DNA-damage response and repair pathways as upregulated in response to AG-636 (Supplementary Fig. S10). Upregulated proteins included enzymes directly involved in DNA repair (BRCA1, POLB, UNG, LIG1, and DDB2) as well as sensors and mediators of DNA repair (CLSPN, TIGAR, and APBB1; Supplementary Table S3). Consistent with previous reports, p53 was the most upregulated protein in response to AG-636 (28). Proteomic analysis of the OCILY19 lymphoma cell line treated with AG-636 showed some overlap in upregulated DNA-damage response–related proteins (e.g., DDB2 and LIG1), though DNA-repair pathways were not enriched by GO analysis (Supplementary Fig. S11; Supplementary Table S3).

We next evaluated activation of the DNA-damage response by AG-636 treatment using γH2AX. A strong induction of γH2AX was observed following AG-636 treatment in the sensitive Z138 and OCILY19 cells as early as 18 hours posttreatment. In contrast, a more modest induction was observed in HCT116 cells and a longer time of AG-636 treatment was required to induce γH2AX in A549 cells (Fig. 7A). To examine the effect of activation of a DNA-damage response on AG-636 sensitivity, cells were treated with the selective CHK1 inhibitor prexasertib (LY2606368) in combination with AG-636. Combination treatment decreased cell viability compared with either treatment alone in the solid tumor models and resulted in clear synergy in A549 cells (CI = 0.18; Fig. 7B). In HCT116 cells, the combination also resulted in increased γH2AX compared with treatment with either agent alone (Fig. 7C). Although prexasertib had limited impact on response of the highly sensitive Z138 and OCILY19 cells to AG-636 (Fig. 7B), in an expanded panel of AML (n = 3) and DLBCL (n = 8) cell lines, prexasertib demonstrated evidence of synergy in nine of the 11 lines (Supplementary Fig. S12). An additional CHK1 inhibitor (SCH90076) and two DNA-PK inhibitors (M3814 and NU7441) also resulted in synergistic effects in combination with AG-636 in this panel (Supplementary Fig. S12). Taken together, these results point to induction of DNA damage as a key component of response to AG-636 and to potential combination strategies aimed at cotargeting DNA-damage response pathways.

Genetically driven tumor dependencies have resulted in the development of transformative cancer medicines, a prime example of which is imatinib for patients with BCR-ABL–driven chronic myelogenous leukemia. The genetic complexity and heterogeneity of most cancers, however, have limited the long-term efficacy of single agents targeting a specific genotype (29). In DLBCL, tumors have been reported to have a median of 17 different genetic alterations (30). In addition to tumor genotype–directed therapies, lineage-based dependencies present opportunities for therapeutic intervention (31–33). Genome-wide RNAi and CRISPR-based gene-silencing screens have been powerful tools to identify such lineage-based vulnerabilities (21, 34). Heterogeneity in metabolic, in addition to genetic, dependencies exists across cancers and may be linked to tumor retention of metabolic features inherent to their normal tissue of origin (13). A recent chemical biology screen identified dependency on the cholesterol biosynthetic enzyme SQLE selectively in tumor cells of neuroendocrine lineage (20). As part of the same screen, we herein report the identification of a heme lineage dependence on DHODH, using a novel and specific inhibitor, AG-636.

The dependence on DHODH in cell lines of hematologic origin appeared to be lineage-based as the dependency broadly spanned across lymphoma cell lines representing various clinical subtypes and did not correlate with particular genotypes. Similarly, other reports describe DHODH inhibitor sensitivity in leukemia models representing diverse subtypes (8, 35). Consistent with such prior work, we also demonstrated AG-636 in vitro and in vivo antitumor activity in a range of leukemia models and opted to focus on activity in lymphoma models here. A role for DHODH in cancer cells of hematologic origin is not surprising, given the literature documenting the importance of pyrimidine biosynthesis for lymphocyte proliferation in response to strong activation signals (1, 36, 37) and the current use of DHODH inhibitors as immunomodulatory agents. It remains to be determined how administration of a DHODH inhibitor in the oncology setting will affect the antitumor immune response; however, optimized intermittent dosing strategies may enable antitumor efficacy while minimizing toxicity to normal immune cells and other tissues (8). Although we found a moderate correlation between proliferation rate and sensitivity to AG-636 across cell lines, differential sensitivity could not be explained solely by proliferative rate as the cell lines of solid tumor origin exhibiting poor sensitivity to AG-636 that were used in our mechanistic studies had similarly fast growth rates as the sensitive heme lines used for comparison.

Given the ability of cells to generate pyrimidines via salvage pathways in addition to via de novo synthesis, a key question was whether AG-636–sensitive cells had an impaired ability to salvage extracellular uridine. In our studies comparing uridine uptake rates and conversion to UMP, we did not identify substantial differences between sensitive and insensitive lines in these parameters. In contrast, a recent study demonstrating activity of the PTC299 DHODH inhibitor in hematologic cancer cells concluded that PTC299-sensitive cells tended to have reduced uridine salvage activity compared with insensitive lines, as evidenced by having a lower fraction of total UMP derived from uridine salvage relative to de novo synthesis (12). A caveat of these findings is that the proliferative rate of the cell lines utilized was not controlled for and the percentage of UMP derived from de novo synthesis versus salvage was assessed at just a single timepoint. Of note, the cell lines included in the PTC299-sensitive group are lines characterized as having shorter doubling times compared with the lines included in the insensitive group. Among the PTC299-sensitive lines was the fast-growing A549 cell line, which we characterized as AG-636–insensitive. The use of different metrics for DHODH inhibitor sensitivity calls likely underlies the discrepancy in categorization of A549 cells between these two studies.

Although AG-636–sensitive and –insensitive cells were not distinguished by differential abilities to utilize the de novo pyrimidine biosynthesis or extracellular uridine salvage pathways in our in vitro studies, in vivo data indicated a greater impact of DHODH inhibition on total pyrimidine pools in sensitive lines. In vitro, the difference between sensitive and insensitive cells with regard to impact of AG-636 treatment on pyrimidine pools was less clear and was transitory, potentially owing to the absence of extracellular uridine in these assays and hence the lack of a key salvage pathway to compensate for DHODH inhibition. In the in vivo setting, homeostatic plasma uridine concentrations range from ∼2 to 4 μmol/L, though it remains unclear how levels may fluctuate in the tumor microenvironment. Regardless, the ability of insensitive solid tumor lines to continue growth for a period of time following inhibition of DHODH in the setting of depleted extracellular uridine suggests the presence of compensatory adaptive pathways in these cells. One candidate pathway of interest is ribophagy (the autophagic breakdown of ribosomes). Ribosomes contain approximately 80% of the total RNA in cells, and their degradation can serve as a source of free nucleosides during starvation conditions (27). The identification of the mTOR negative regulator TSC1 as the top hit in sensitizing A549 cells to AG-636 (PTEN and TSC2 were also identified as weaker hits) was of note given the reported role for mTOR in regulating ribophagy. Although mTOR inhibition blunted response to AG-636, follow-up studies were unable to validate a role for knockout of TSC1 (mTOR activation) in sensitizing A549 cells to AG-636 in vitro.

The possibility remains that a key component of differential in vitro sensitivity to DHODH inhibition lies in the response downstream of depleted pyrimidine pools. The combined results of the CRISPR depletion screen and proteomics analysis implicate activation of DNA-damage response and repair pathways as adaptive mechanisms to blockade of de novo pyrimidine biosynthesis and the resultant nucleotide imbalance. Prior studies have demonstrated a role for CHK1 and p53 pathway activation in response to pyrimidine depletion (38). Consistent with this, the selective CHK1 inhibitor prexasertib exerted synergistic effects in reducing cell viability in combination with AG-636 both in AG-636–insensitive solid tumor lines as well as in a panel of cancer cells of hematologic origin. Our findings implicating the DNA-damage response as potentially limiting to DHODH inhibitor response are supported by other recent reports. In one study, the CHK1 inhibitor PF477736 was demonstrated to sensitize transformed cells to teriflunomide (39). In another study using an isogenic model system, teriflunomide induced DNA damage to a greater degree in the PTEN-null setting, reportedly owing to reduced ATR checkpoint activation and CHK1 phosphorylation (Ser145) in the absence of PTEN (10). Along a similar theme, an increase in the activity of the de novo pyrimidine biosynthesis pathway was suggested by Brown and colleagues as an adaptive mechanism to supply nucleotides needed for DNA repair in response to genotoxic chemotherapy (doxorubicin; ref. 9). Collectively, these studies bolster our findings pointing to the combination of AG-636 with agents that enhance DNA damage (e.g., inhibitors of the DNA-damage response or direct DNA-damaging agents) as a potential strategy to maximize antitumor activity.

Approximately 60% of DLBCL patients can be cured by the standard-of-care regimen, R-CHOP. However, patients who are refractory or relapse have a poor prognosis (40). DLBCL subtypes based on gene-expression profiling have been identified with prognostic value, with the ABC subtype tending to have worse outcomes compared with GCB (41). High-grade B-cell lymphomas with translocations involving MYC and BCL2 and/or BCL6 (double- or triple-hit lymphoma) are rarely cured by R-CHOP and more intensive chemotherapy regimens (R-EPOCH) are required (42). More complex methods of genetic characterization (e.g., exome/transcriptome sequencing and DNA copy-number analysis) have recently uncovered additional DLBCL subtypes associated with variable outcomes following R-CHOP and suggest potential subtype-specific therapeutic strategies (30, 43). To date, there have been no approved targeted therapies for particular DLBCL subtypes. CAR-T cell therapy and polatuzumab vedotin (combined with bendamustine and rituximab) are new promising therapies with activity in the relapsed/refractory population yielding complete responses in ∼40% to 50% of patients (44–46), yet even these treatments ultimately fail the majority of patients. Mantle cell lymphoma, a rarer non-Hodgkin lymphoma subtype, is generally incurable and characterized by multiple relapses. Chemotherapy-free combination therapy approaches including BTK inhibitors (e.g., ibrutinib) as a backbone can deliver strong responses and are moving into earlier lines of therapy (47); however, primary and acquired resistance remains a hurdle to prolonged efficacy (48). In all, there remains an unmet need for novel therapies that can broadly target the relapsed/refractory B-cell lymphoma population. Our demonstration of complete in vivo tumor regression by a single agent, AG-636, in an ibrutinib-resistant mantle cell lymphoma model and broad activity across B-cell lymphoma subtypes (including double-hit and ABC DLBCL) lends promise for DHODH inhibition as a novel treatment strategy for relapsed/refractory lymphomas.

AG-636 is a novel and potent small-molecule inhibitor of DHODH with favorable properties for oral administration in humans. We confirmed the specificity of the mechanism of action of AG-636 as a DHODH inhibitor via the crystal structure of AG-636 with DHODH as well as through our findings of the tight relationship between uridine availability and response to AG-636. Our preclinical data with AG-636 demonstrating a vulnerability of tumor cells of hematologic origin to inhibition of DHODH supported initiation of a phase I study of AG-636 in relapsed/refractory lymphoma (NCT03834584). Together with ongoing clinical studies with other DHODH inhibitors in AML, the potential for DHODH inhibition as a lineage-based therapeutic strategy will begin to be elucidated.

G. McDonald, V. Chubukov, and K. Truskowski report being an employee and stockholder in Agios Pharmaceuticals. S. Choe reports other from Agios (employee) during the conduct of the study. A.K. Padyana reports being an employee of Agios Pharmaceuticals and has ownership interest in stock. K. Nellore and H. Subramanya report a patent 20200078339 issued. S.S. Rao and K.S. Reddy report a patent for US10080740B2 issued. No potential conflicts of interest were disclosed by the other authors.

G. McDonald: Conceptualization, data curation, formal analysis, investigation, writing–original draft, project administration, writing–review, and editing. V. Chubukov: Data curation, formal analysis, investigation, writing–original draft, writing–review, and editing. J. Coco: Investigation, writing–review, and editing. K. Truskowski: Investigation, writing–review, and editing. R. Narayanaswamy: Investigation, writing–review, and editing. S. Choe: Data curation, investigation, writing–review, and editing. M. Steadman: Investigation, writing–review, and editing. E. Artin: Formal analysis, investigation, writing–review, and editing. A.K. Padyana: Formal analysis, investigation, writing–original draft, writing–review, and editing. L. Jin: Investigation, writing–review, and editing. S. Ronseaux: Investigation, writing–review, and editing. C. Locuson: Investigation, writing–review, and editing. Z.-P. Fan: Investigation, writing–review, and editing. T. Erdmann: Investigation, writing–review, and editing. A. Mann: Investigation, writing–review, and editing. S. Hayes: Investigation, writing–review, and editing. M. Fletcher: Data curation, investigation, writing–review, and editing. K. Nellore: Investigation, writing–review, and editing. S.S. Rao: Investigation, writing–review, and editing. H. Subramanya: Investigation, writing–review, and editing. K.S. Reddy: Investigation, writing–review, and editing. S.K. Panigrahi: Investigation, writing–review, and editing. T. Anthony: Investigation, writing–review, and editing. S. Gopinath: Investigation, writing–review, and editing. Z. Sui: Investigation, writing–original draft, writing–review, and editing. N. Nagaraja: Investigation, writing–review, and editing. L. Dang: Investigation, writing–review, and editing. G. Lenz: Supervision, writing–review, and editing. J. Hurov: Supervision, writing–review, and editing. S.A. Biller: Supervision, writing–review, and editing. J. Murtie: Supervision, writing–review, and editing. K.M. Marks: Supervision, writing–review, and editing. D.B. Ulanet: Conceptualization, data curation, formal analysis, supervision, investigation, writing–original draft, project administration.

This study was funded by Agios Pharmaceuticals, Inc. We would like to thank Stuart Murray of Agios for design and oversight of the cell line panel screen and Shirley Louise-May, PhD, CMPP, and Christine Ingleby, PhD, CMPP, of Excel Medical Affairs for providing editorial and submission assistance, supported by Agios.

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.