Loss of UXS1 Selectively Depletes Pyrimidines and Induces Replication Stress in KEAP1-Mutant Lung Cancer

UXS1 loss in KEAP1-mutant cells causes pyrimidine nucleotide depletion, DNA replication stress induction, and ultimately cell-cycle exit that results in tumor stasis, highlighting UXS1 as a potential therapeutic target in KEAP1-mutant tumors.

Abstract

Kelch-like ECH-associated protein 1 (KEAP1) is the third most commonly mutated gene in non–small cell lung cancer and is associated with poor prognosis. In this study, we investigated synthetic lethal interaction genes in KEAP1-mutated cancer cells and identified a dependency on UDP–xylose synthase 1 (UXS1), which converts UDP–glucuronic acid (UDP-GlcA) to UDP–xylose in the proteoglycan synthetic pathway. UDP–glucose dehydrogenase (UGDH), a transcriptional target of NRF2 that converts UDP–glucose to UDP-GlcA, was highly expressed in KEAP1-mutant tumors. Upon UXS1 knockdown, depletion of UDP–xylose occurred in both KEAP1-mutant and wild-type cells, whereas UDP-GlcA accumulated to a greater extent in the KEAP1-mutant setting. The resulting shortage of available UDP and other pyrimidines slowed S-phase progression and stalled DNA replication fork marks, causing cells to undergo prolonged cell-cycle exit or apoptosis. Dependency on UXS1 was rescued by knocking out UGDH to prevent UDP-GlcA accumulation and UDP depletion. DNA replication stress in UXS1-depleted cells sensitized them to clinical cell-cycle checkpoint inhibitors. Furthermore, CRISPR screening experiments identified genes that modulate UXS1 dependency. Whereas the liver had the highest normal tissue expression of UGDH, UXS1 knockout in the liver did not result in hepatotoxicity. Taken together, these data demonstrate that UXS1 is a selective dependency in KEAP1-mutant tumors, and loss of UXS1 creates additional therapeutically exploitable vulnerabilities in KEAP1-mutant tumors.

Significance:

UXS1 loss in KEAP1-mutant cells causes pyrimidine nucleotide depletion, DNA replication stress induction, and ultimately cell-cycle exit that results in tumor stasis, highlighting UXS1 as a potential therapeutic target in KEAP1-mutant tumors.

See related commentary by Yasseen and DeNicola, p. 4582

Graphical Abstract

Introduction

Lung cancer is the leading cause of cancer-related mortality globally, accounting for 1.8 million deaths annually (1). Non–small cell lung cancer (NSCLC) comprises 85% of all lung cancer cases, of which, about 20% are caused by loss-of-function mutations in the protein Kelch-like ECH-associated protein 1 (KEAP1; refs. 2, 3). There is currently no clinically approved treatment specific for KEAP1-mutant tumors. Because KEAP1 mutations are typically mutually exclusive with mutations in EGFR (4), patients with KEAP1-mutant cancers often do not benefit from the significant advancements made in EGFR-targeted therapies during the past couple of decades.

KEAP1 is a redox-sensitive adapter protein for the E3 ubiquitin ligase, CUL3, and it binds and promotes the proteasomal degradation of the transcription factor nuclear factor erythroid 2-related factor 2 (NRF2; ref. 5). Under oxidative stress, oxidation of key sensor cysteine residues on the KEAP1 protein cause a conformational change, resulting in the release and nuclear translocation of NRF2 (5). NRF2 is a master regulator of more than 200 cytoprotective genes involved in redox, xenobiotic, and metabolic stress responses as well as several other cellular pathways, such as survival and proliferation, autophagy, and DNA repair (5–7). Notably, most loss-of-function KEAP1 mutations are within the Kelch domains that result in disruption of its interaction with NRF2 (6), leading to constitutive activation of NRF2 and its target genes. In turn, this provides cancer cells with an increased oxidative and metabolic fitness. Indeed, KEAP1 mutations are associated with chemotherapy, radiotherapy and targeted therapy resistance (3, 7, 8), leading to poor treatment responses and an overall adverse outcome. Moreover, recent studies have uncovered that KEAP1 mutations are associated with poorer response to immunotherapy (9, 10), underscoring a refractory oncology indication with unmet clinical need.

To identify potential therapeutic targets for KEAP1-mutated tumors, we interrogated the Cancer Dependency Map (DepMap) database (11) and identified UDP–glucuronate decarboxylase/UDP–xylose synthase 1 (UXS1) as a selective dependency in KEAP1-mutant cancer cell lines. UXS1 is an enzyme that decarboxylates UDP–glucuronic acid (UDP-GlcA) to UDP–xylose, which is required for proteoglycan synthesis (12, 13). UDP-GlcA is synthesized by UDP–glucose dehydrogenase (UGDH), which is a NRF2 target gene (14), and thus is highly expressed in KEAP1-mutant cancer cells. UDP–xylose is a negative feedback regulator of UGDH (15), and hence UXS1 loss leads to an uninhibited activity of UGDH leading to a constitutive synthesis of UDP-GlcA in UGDH-overexpressing cells. High UGDH expression has been linked to increased metastasis and worse prognosis in several types of cancers (16). In addition to proteoglycan synthesis, UDP-GlcA is involved in phase II metabolism in which it serves as a substrate for UDP–glucuronosyltransferases (UGT) that catalyze the glucuronidation of various endogenous and xenobiotic compounds to prepare them for excretion (17).

In this study, we sought to understand the mechanism of selective dependency on UXS1 in KEAP1-mutant/UGDH-high cancer cells. In this study, we show that the primary cellular consequence of UXS1 loss in KEAP1-mutant cells is hyperaccumulation of UDP-GlcA, pyrimidine nucleotide loss, DNA replication stress, and ultimately cell-cycle exit, which results in the observed tumor stasis. We anticipate that our findings will shed light on UXS1 as a novel therapeutic target in KEAP1-mutant tumors.

Materials and Methods

Cell culture and cell line generation

The KEAP1-mutant cell lines A549 (CCL-185, RRID: CVCL_0023), H460 (HTB-177, RRID: CVCL_0459), H2122 (CRL-5985, RRID: CVCL_1531), H2023 (CRL-5912, RRID: CVCL_1515), H1944 (CRL-5907, RRID: CVCL_1508), and H1792 (CRL-5895, RRID: CVCL_1495) were purchased from the ATCC. The KEAP1 wild-type (WT) cell lines H1299 (CRL-5803, RRID: CVCL_0060), Calu6 (HTB-56, RRID: CVCL_0236), and Chago-K1 (HTB-168, RRID: CVCL_1121) as well as HEK293 cells (RRID: CVCL_0045) were purchased from the ATCC. All lines were maintained at 37°C with 5% CO₂ in RPMI 1640 (Gibco) supplemented with 10% FBS and 1× Glutamax. All cell lines are routinely tested for Mycoplasma every month. All cell lines were kept at low passage after purchase from the ATCC; therefore, additional authentication was not performed.

Low-passage A549, H460, H2122, H2023, H1944, H1792, and H1299 cells were transduced with the validated doxycycline (dox)-inducible shUXS1 construct or analogous constructs for inducing shSLC33A1, shNFE2L2, or a nontargeting control. Cells were selected, and stable expression was continually maintained by treatment with 100 μg/mL hygromycin B (Fisher Scientific). Induction was also confirmed in each experiment by induction of Turbo-RFP.

Short hairpin RNA–mediated gene knockdown

Dox-inducible shUXS1 construct or analogous constructs for inducing shSLC33A1, shNFE2L2, or a nontargeting control were packaged in HEK293 cells with third-generation lentivirus packaging plasmids using Lipofectamine 2000 transfection reagent (Thermo Fisher Scientific, cat. #11668027) according to the manufacturer’s instruction. Cell lines were transduced with virus harvest 48 hours after transfection in the presence of 6 µg/mL polybrene and a multiplicity of infection (MOI) of 0.3. Transduced cells were selected with 250 µg/mL of hygromycin for 7 days. Single clones with validated gene knockdown were pooled and maintained in 100 µg/mL hygromycin. Dox (0.5 µg/mL) was used to induce shRNA expression.

Low-passage H2122 and H2023 cell lines already expressing the inducible shUXS1 construct were further transduced with H2B-iRFP670 (nuclear marker) and mVenus-Geminin [1-110] (S/G₂ phase reporter) lentivirus. Cells stably expressing both sensors were isolated by two rounds of FACS using a Bigfoot Cell Sorter (Thermo Fisher Scientific).

Knockdown by siRNA treatment

siRNA transfections were performed using the DharmaFECT 1 reagent (Dharmacon) according to the manufacturer’s instructions. Transfection mixtures were added to cells for the indicated times before downstream viability or imaging assays. Oligonucleotides used in this study are indicated in the Key Resources Table.

CRISPR/Cas9-mediated single-gene knockout

Guide RNA sequences were obtained from the Avana CRISPR library (18), and single-guide RNAs (sgRNA) were cloned into lentiCRISPRv2 vector (RRID: Addgene 98292) and packaged in HEK293 cells as described above. Cells expressing Cas9 were transduced with virus harvested 48 hours after transfection in the presence of 6 µg/mL polybrene and a MOI of 0.3. Transduced cells were selected with 2 µg/mL of puromycin for 48 hours before using cells for experiment.

CRISPR/Cas9 KO whole-genome screen

shNTC or shUXS1 expressing H460 and H2122 cells were transduced with Avana whole-genome CRISPR KO library (6 guides per gene; ref. 18) at a MOI of 0.3 in the presence of 6 µg/mL polybrene for 48 hours followed by puromycin (2 µg/mL) selection for 5 days. Cell number for transduction was determined such that a library coverage of 500x will be maintained after transduction with a MOI of 0.3 and puromycin selection. Cells were harvest at time 0 and after 3, 7, and 11 days of dox treatment. Genomic DNA was extracted from cells using Macherey Nagel Nucleospin Blood XL Kit (item number 740950.50) according to the manufacturer’s instructions. sgRNAs were amplified and sequenced using the NovaSeq 6000 sequencer (Illumina). sgRNA count data were used to determine relative guide abundance in shUXS1 vs shNTC cells using method outlined in the Model-based analysis of genome-wide CRISPR/Cas9 knockout (KO; MAGeCK; ref. 19) article. CRISPR screen raw FASTQ data as well as count files are available on Gene Expression Omnibus (GEO) under accession number GSE304915.

RNA extraction, RT-PCR, and RNA sequencing

RNA extraction was done using the RNeasy kit (Qiagen, cat. #74104) according to manufacturer’s instructions. cDNA was synthesized using 2 µg RNA using Maxima H Minus cDNA synthesis master mix (Thermo Fisher Scientific, cat. #M1662).

RT-PCR was done using target-specific TaqMan probes and QuantStudio RT-PCR system (Thermo Fisher Scientific).

RNA libraries for RNA sequencing (RNA-seq) were prepared using NEBNext Ultra II Directional RNA Library Prep Kit with polyA enrichment (New England Bio Labs, cat. #E7760L) and sequenced on the NovaSeq 6000 (Illumina). Sequencing reads were aligned to the human genome using STAR aligner (RRID: SCR_005622; ref. 20), and transcript counts for aligned reads were generated using featureCounts (21). Differential gene expression analyses were done using DESeq2 (22). Pathway analyses were done using Metascape (23). RNA-seq raw FASTQ data as well as count files are available on GEO under accession number GSE305124.

Metabolomics extraction

5e5 cells were washed with room temperature PBS followed by quenching in 40:40:20 MeCN:MeOH:H₂O containing internal standards (D4-L-tyrosine, 15N4-L-arginine, and D5-benzoic acid; Sigma-Aldrich) maintained at −0°C. Cells were scraped, transferred to Eppendorf tubes, and then centrifuged at 16,000 × g at 4°C for 15 minutes. Supernatants were diluted 10x in MeCN:MeOH:H₂O (40:40:20, v/v/v) for targeted metabolomics analysis.

For untargeted metabolomics, supernatants were dried down under nitrogen at 4°C and resuspended in 100 μL of water containing 1 μg/mL of d-lysine, d-phenylalanine, and 250 ng/mL of d-succinate (Sigma-Aldrich) to analyze in negative ionization mode. For positive ionization mode, resuspension in water was diluted 1:4, v/v with MeCN.

Mouse livers were freeze-clamped and stored in −80°C until processing. Approximately 30 mg of liver was weighted on dry ice and homogenized at cryo-temperature using the CryoMill (Thomas Scientific). Metabolites were extracted by adding 1 mL/30 mg of 40:40:20, MeCN:MeOH:H₂O containing internal standards (D4-L-tyrosine, 15N4-L-arginine, D5-benzoic acid, D5-L-phenylalanine, D4-L-lysine, and D5-succinate; Sigma-Aldrich) maintained at −20°C. One mL of homogenized mixture was transferred to a new glass vial. The mixture was incubated on ice for 10 minutes and then centrifuged at 16,000 × g at 4°C for 15 minutes. Supernatants were collected for untargeted metabolomics analysis.

Targeted liquid chromatography–mass spectrometry metabolomics for in vitro assay

Five µL of metabolomics extracts were analyzed on an Agilent 1,290 Bio HPLC coupled to a Sciex 7,500 triple-quadrupole mass spectrometer, targeting 215 endogenous polar compounds. Chromatographic separation was achieved on a SeQuant ZIC-pHILIC 5 µm 150 × 2.1 mm high-performance liquid chromatography column, as previously described (24). Briefly, mobile phase A was 20 mmol/L ammonium carbonate brought to pH 9.2 using ammonium hydroxide, with 1 µmol/L medronic acid. Mobile phase B was 95% acetonitrile and 5% mobile phase A. The gradient ran as follows at 150 µL/min flowrate: 0 minute 84.2 %B, 2.5 minutes 76.9 %B, 5 minutes 68.4 %B, 7.5 minutes 60 %B, 10 minutes 52.6 %B, 15 minutes 36.8 %B, 20 minutes 21.1 %B, 22 minutes 15.8 %B, 22.5 minutes 84.2 %B, and 30 minutes 84.2 %B. Mass spectrometry acquisition occurred in scheduled MRM mode with 2 transitions per compound whenever possible. Dwell time was set to a minimum of 4 ms and a maximum of 250 ms, with a target cycle time of 1 second. Source conditions were as follows: ion source gas 1 60 psi, ion source gas 2 60 psi, curtain gas 46 psi, CAD gas 8, and source temperature 350°C. Polarity switching was utilized with a 1,600 V spray voltage for positive mode, 1,900 V for negative, and a pause time of 4 ms.

Data files were converted to mzML format using ProteoWizard (RRID: SCR_012056; ref. 25) msConvert, and all peaks were manually reviewed using MAVEN2 (26) software. All reported compounds were confirmed by retention time and a secondary transition whenever possible. Intensities were reported as the log₂ value of smoothed peak area. For peaks with large tailing in which the full peak width may have exceeded the length of the MRM window (e.g., citrate, histidine, etc.) the topmost 3 points of each smoothed peak were instead used for quantification. All peaks were compared with the median signal of external blank injections collected alongside samples, and any peak below 1.5× the signal of the blanks’ median intensity for its metabolite was removed. For statistical testing, values below the limit of detection were imputed around a mean value of 2× the median blank intensity with a SD of 0.15. Sample-to-sample intensity differences were corrected for by performing a probabilistic quotient normalization. The median of each feature is calculated, the quotient between the feature median and sample value is calculated for all samples, and all values for a sample are divided by its median quotient values were controlled for the FDR using the q value R package from the Storey Lab (27). Significant changes were reported for q values <0.01. For absolute quantitation, an external calibration curve containing all quantified metabolites was prepared in 40:40:20 MeCN:MeOH:H₂O and run alongside experimental samples. Quantified values were normalized to the median bicinchoninic acid assay value of replicates and reported as nmol/mg of protein.

Absolute quantification of pyrimidine and UDP-GlcA

External calibration curves were prepared in 40:40:20 MeCN:MeOH:H₂O for the quantification of uridine, nucleotides (UMP, UDP, UTP, CMP, CDP, and CTP), and UDP-GlcA. Calibration levels were selected based on anticipated endogenous concentrations: low-abundance analytes ranged from 200 ppt to 100 ppb; mid-abundance metabolites from 2 ppb to 1 ppm; and UDP-GlcA from 20 ppb to 10 ppm. Each curve comprised nine concentration levels. To evaluate potential matrix effects, parallel calibration curves were generated in a representative sample matrix. Comparison of signal intensities and curve slopes between matrix-based and solvent-based preparations indicated no significant matrix effects. Analyte concentrations in study samples were calculated using the external calibration curves, with some samples requiring up to a 10-fold dilution to fall within the quantitation range.

Untargeted liquid chromatography–mass spectrometry metabolomics for in vivo assay

Five µL metabolomics samples were analyzed in both positive and negative ESI-liquid chromatography–mass spectrometry methods on Vanquish Ultra-high Performance Liquids coupled to Q-Exactive Plus mass spectrometers (Thermo Fisher Scientific). Metabolites were separated using a SeQuant ZIC-pHILIC column (5 μm, 200 Å, 150 × 2.1 mm) as described above in the targeted metabolomics assay.

Data was acquired using data-dependent acquisition mode with the following parameters: resolution = 70,000, AGC target = 3.00 × 10 maximum IT (ms) = 100, and scan range = 70 to 1,050. The MS2 parameters were as follows: resolution = 17,500, AGC target = 1.00 × 105, maximum IT (ms) = 50, loop count = 6, isolation window (m/z) = 1, (N)CE = 20, 40, 80; underfill ratio = 1.00%, Apex trigger(s) = 3 to 10, and dynamic exclusion(s) = 25. For negative mode, (N)CE = 20, 50, 100.

Negative ionization analysis of the cell extract was analyzed using a reverse-phase ion-pairing chromatograph with an Agilent Extend C18 RRHD column, 1.8 μm particle size, 80 Å, 2.1 × 150 mm. LC-MS/MS analysis was acquired using Vanquish Ultra-high Performance Liquids coupled to Q-Exactive Plus mass spectrometers (Thermo Fisher Scientific). Mobile phase A was 10 mmol/L tributylamine and 15 mmol/L acetic acid in 97:3 water:methanol, pH 4.95; mobile phase B was methanol. The flow rate was 200 μL/minutes, and the gradient was t = −4, 0% B; t = 0, 0% B; t = 5, 20% B; t = 7.5, 20% B; t = 13, 55% B; t = 15, 95% B; t = 18.5, 95% B; t = 19, 0% B; and t = 22, 0% B. The mass spectrometer was operated in negative ion mode using data-dependent acquisition mode with the following parameters: resolution = 70,000, AGC target = 1.00E + 06, maximum IT (ms) = 100, scan range = 70 to 1,050. The MS2 parameters were as follows: resolution = 17,500, AGC target = 1.00E + 05, maximum IT (ms) = 50, loop count = 6, isolation window (m/z) = 1, (N)CE = 20, 50, 100; underfill ratio = 1.0 0%, Apex trigger(s) = 3 to 12, and dynamic exclusion(s) = 20. The injection volume was 5 µL.

Raw files were converted to the mzML format using ProteoWizard v3 (25). Compound identifications were detected and grouped using the OpenCLaM R package (https://github.com/calico/open_clam) and manually inspected using the MAVEN2 peak analysis program (https://github.com/eugenemel/maven; ref. 26) with the criteria of a precursor ion tolerance of 10 ppm and a product ion tolerance of 20 ppm, comparing fragmentation and retention time to an in-house library generated from authentic standards for metabolomics. The sum of the raw scan intensity of the smoothed maximum, as well as the scan immediately preceding and following the smoothed maximum scan, was taken as the peak quantity, hereafter referred to as “peak area top.” Peak areas on top of each feature underwent log₂ transformation, and peaks below the limit of detection were assigned a value of 12. P values were controlled for the FDR using the qvalue R package from the Storey Lab (27). Significant changes were reported for q values <0.01.

Mouse studies

AbbVie is committed to ensuring the humane care and use of laboratory animals in the company’s research and development programs. Our programs exceed regulatory agency standards, and we are committed to the internationally accepted principles of the 3Rs (refinement, reduction, replacement). All animal studies were reviewed and approved by AbbVie’s Institutional Animal Care and Use Committee or Oversight Body (in accordance with national regulations). Animal studies were conducted under an Association for Assessment and Accreditation of Laboratory Animal Care International–accredited program, in which veterinary care and oversight was provided to ensure optimal animal care. Female C.B-17 SCID mice were purchased from Charles River Laboratories. Animals were housed in a light cycle-controlled room with HEPA filtration at 22°C. Animals had nesting and climbing cage topper as the primary environmental enrichment.

Dox-inducible shNTC or shUXS1 expressing H2122 (2e⁶ cells/mouse) or A549 cells (5e⁶ cells/mouse) were subcutaneously injected in C.B-17 SCID mice, and the inclusion criterion was 6 to 8 weeks old female mice. The inoculation volume (0.1 mL) comprised a 50:50 mixture of cells in growth media and Matrigel (BD Biosciences). Electronic calipers were used to measure the length and width of each tumor 2 to 3 times per week. Tumor volume was estimated by applying the following equation: volume = length × width²/2. When tumors reached approximately 200 mm³, mice were randomized and distributed into cages containing standard rodent chow or Envigo (TD.01306) dox-containing rodent chow as the only food source. All xenograft trials were conducted using 8 to 10 mice per group, and all mice were ear-tagged and monitored individually throughout the studies. A separate cohort of mice (n = 3) treated the same as above were euthanized 7 days after the start of doxfood treatment to confirm UXS1 knockdown by RT-PCR and measure metabolites, including UDP-GlcA levels, in the tumor tissue. Once study started, the exclusion criteria were tumor sizes more than 1.5 cm or morbidities caused by fight wounds; there were no attritions. Blinding was not done throughout the study.

In vivo gene editing in the liver

Lipid nanoparticles (LNP) encapsulating gRNAs were used for in vivo gene editing. gRNAs targeting EGFP (control), Ugt1a1, Uxs1, and Ugdh were synthesized with end chemical modification as described previously described (28) by GenScript. Three gRNAs targeting the same gene were mixed at equal molar ratio and co-encapsulated into LNPs. For Uxs1 and Ugdh double KO, three gUxs1 and three gUgdh were mixed at equal molar ratio and co-encapsulated into LNPs. LNPs used in this study contain SM102 as ionizable lipids and were formulated by GenScript.

Mice expressing Cas9-Egfp (Jax 026179), age between 8 to 12 weeks, were used in this study. LNP gRNAs were dosed via the retro-orbital vein or tail vein at 1 mpk (for single gene) or 2 mpk (for Uxs1 and Ugdh double-KO) per animal. Each animal received LNPs every other day, for a total of three doses. Mice were monitored closely for potential adverse effects after LNP injections. To assess liver toxicity following gene editing, blood was collected at various time points after LNP injection from the retro-orbital vein. To assess gene editing efficiency and to perform liver metabolic/proteomic analysis, mice were euthanized at end-time points, and their livers were immediately freeze-clamped in liquid nitrogen.

Statistical analysis for metabolomics

Analysis of the metabolomics data was performed using the clamanR package (https://github.com/calico/claman) in R. A linear model using the R lm function was fit to the data for each identified compound using the following model: y_ik = α + β₁g_i + β₂s_k, in which y_ik is the median polished normalized log₂ (peak area top) of the annotated metabolite, α is an arbitrary intercept, i represents the gene targeted by the gRNA, and β₁ is expected to be the difference between gene editing targets; k represents sex, and β₂ is the expected difference between sexes. P values were controlled for the FDR on a term-by-term basis using the qvalue R package from the Storey Lab (27). Significant changes were reported for q values <0.01.

Clinical chemistry to measure liver enzymes

Blood collected was processed to serum and analyzed on the Abbott ARCHITECT c16000 Clinical Chemistry Analyzer for the following parameters in order of priority: alanine transaminase, aspartate transferase, total bilirubin (TBIL), alkaline phosphatase, and gamma-glutamyl transferase. For samples with TBIL values ≥1.000, a reflex direct bilirubin was also measured.

Cell viability assay

Cells were seeded into 96-well plates (Corning, cat. #3916) 24 hours prior to treatments. Cells were treated with indicated compound(s) for the indicated durations throughout the article and were immediately assessed for viability at each point of observation by proceeding with the CellTiter-Glo protocol, as specified by the manufacturer (Promega). In brief, CellTiter-Glo reagent was added to each well in a volume equal to the volume of cell culture medium present (100 µL of reagent solution added to 100 µL of media in each well). Plates were then incubated for 15 minutes at room temperature, and luminescent signal was immediately recorded using an EnVision plate reader (PerkinElmer/Revvity).

For cell growth experiments, cells were counted before and after treatment to determine growth. Trypan blue was used to determine viability using the ViaCell cell counter (Beckman Coulter).

Flow cytometry apoptosis assay

Cells were seeded in 12-well plates at 10⁵ cells/well 24 hours prior to drug treatments. Wells were treated in triplicate. After treatment, nonadherent cells were first harvested by pipetting, adherent cells were harvested by trypsinization, and these two populations were then combined. Cell suspensions were centrifuged and resuspended in calcium-rich binding buffer provided by the apoptosis staining kit. Cell suspensions were stained with both Annexin V-FITC (1:100) and 7-Aminoactinomycin D (1:100). Single-cell fluorescent signals were acquired on a BD LSRFortessa flow cytometer. By convention, cells were gated in FlowJo (RRID: SCR_008520) to remove debris and doublets, as shown in gating scheme. Annexin V-FITC and 7-Aminoactinomycin D values were plotted as a bivariate scatter and quadrant gating was applied to all plots to reach final apoptotic population percentages.

Senescence-associated β-galactosidase staining

Cells were seeded in 96-well plates 24 hours prior to treatments, as indicated throughout. After treatments were complete, growth media were removed, and fixative from the senescence-associated β-galactosidase (SA-β-gal) kit was added to the cells for 15 minutes at room temperature. The plate was then rinsed twice with 1X PBS. PBS was then removed, and 0.1 mL of SA-β-gal solution (as defined by the kit protocol) was added to each well. Plates were wrapped in parafilm and incubated at 37°C for 24 hours in a dry incubator with no CO₂. While the staining solution was still on the plate, cells were imaged using a Keyence digital microscope. Color brightfield images were captured at 20×, and the percent of positive cells was manually scored for triplicate sites.

Western blotting

Cell suspensions were pelleted and lysed in 1% SDS. Lysates were then sonicated and heated at 95°C for 5 minutes. Proteins were separated by 4% to 20% Criterion TGX Precast Gel (Bio-Rad) for 1 hour at 150 V and transferred to a polyvinylidene difluoride membrane using a Trans-Blot Turbo Midi PVDF Transfer System (Bio-Rad). The membrane was incubated in blocking buffer (LI-COR) at room temperature for 2 hours before overnight incubation at 4°C with primary antibodies (listed on Key Resources table). The membrane was then washed for 5 minutes with PBS supplemented with 0.1% Tween-20 three times and then incubated with anti-rabbit or anti-mouse secondary antibodies (cat. #926-32210, RRID: AB_621842, cat. #926-68073, RRID: AB_10954442) for 1 hour at room temperature. The corresponding signals were detected on an Odyssey CLx Imager (LI-COR) equipped with Image Studio, and images were quantified in Fiji (RRID: SCR_002285).

EdU incorporation

To identify cells actively synthesizing DNA, cells were pulsed with 10 µmol/L of the synthetic nucleotide 5-ethynyl 2´-deoxyuridine (EdU; alkyne-conjugated) at 37°C for 30 minutes before fixation with 4% paraformaldehyde. The EdU was visualized by click chemistry with azide-linked fluorescent dyes as described in the manufacturer’s protocol (Thermo Fisher Scientific). Cells were then washed twice with PBS and blocked with 3% BSA for 1 hour at room temperature to prepare for further immunostaining.

Time-lapse microscopy and single-cell tracking

Cells were seeded on a 96-well plate (PhenoPlate, PerkinElmer/Revvity) 24 hours before the start of drug treatment. Cells were then treated with dox (0.5 μg/mL) for 96 hours to induce shRNA expression, and then the imaging period began using a PerkinElmer OperaPhenix. Movie images were taken with appropriate filter sets (for mVenus, turbo-RFP, and iRFP670) at a frequency of 15 minutes between frames for a total imaging duration of 40 hours. Cells were maintained in a humidified incubation chamber at 37°C with 5% CO₂. Cells were imaged in phenol red–free full-growth media. Movies were then analyzed using the built-in image analysis available in Harmony software (PerkinElmer/Revvity), in which H2B-iRFP670 signal was used to detect and segment nuclei in each frame, and then automated cell tracking was captured across all movie frames. Cells with complete tracks across the imaging period were then used to extract mVenus-Geminin signal intensity as a measure of S/G₂ cell-cycle phase length for individual cells, and these dynamics were then plotted in MATLAB (RRID: SCR_001622). Turbo-RFP expression was also captured and used as an indicator of successful shRNA induction.

Immunofluorescence imaging

Cells were seeded on a 96-well plate (PhenoPlate, PerkinElmer/Revvity) 24 hours before the start of drug treatment. After indicated treatments and times throughout, cells were fixed with 4% paraformaldehyde for 10 minutes. For immunofluorescence, a standard protocol was used: Cells were permeabilized with 0.1% Triton X-100 at 4°C for 15 minutes, blocked in 3% BSA solution for 1 hour at room temperature, incubated with primary antibodies overnight at 4°C, and incubated with secondary antibodies for 2 hours at room temperature. Where applicable, cells were stained with Hoechst 33342 (1:10,000) for 10 minutes to visualize DNA content. Immunofluorescence imaging was performed on an Opera Phenix high-content screening system (PerkinElmer/Revvity) with a 40×/1.1 NA water objective, acquired and analyzed via Harmony software to segment nuclei (using Hoechst) and extract mean nuclear intensities (for p21, phospho-Rb, gH2AX, pCDK1, and pCDK2) or nuclear foci (for FANCD2). GM130 was used to segment the Golgi compartment adjacent to each cell’s nucleus and then record the Golgi area value in pixels. The extracted measurements were then plotted as violin plots, ksdensity plots, or density scatter plots in MATLAB.

FISH to visualize mRNA molecules

Cells were seeded on a 96-well plate (PhenoPlate, PerkinElmer/Revvity) 24 hours before the start of siRNA treatment. Following 48 hours after transfection, cells were fixed with 4% paraformaldehyde and processed for RNA-FISH analysis according to the manufacturer’s protocol (ViewRNA ISH Cell Assay Kit, Thermo Fisher Scientific). In summary, mRNA probes were hybridized at 40°C for 3 hours, followed by standard amplification and fluorescent labeling steps also at 40°C. The UGDH probe (type 6) used in this study is listed in the Key Resources Table. FISH images were taken on an OperaPhenix high-content screening system (PerkinElmer/Revvity) with a 40×/1.0 NA water objective, and puncta detection was performed using Harmony image analysis software.

Delivery of UXS1^R236H mutant

Dox-inducible WT or UXS1^R236H variant gblocks were synthesized by GenScript and cloned into PSHUSH vector. Packaging into lentivirus was done in HEK293 cells as described above. Lentivirus transduction was done as described above.

DepMap analyses

CRISPR gene effects (e.g., UXS1) and expression profiles (e.g., UGDH) across all DepMap (RRID: SCR_017655) database cell lines were accessed via the public database portal and exported for downstream plotting using Excel (RRID: SCR_017294) or custom R scripts (https://depmap.org/portal/). Cell lines were further stratified by tissue origin (e.g., lung) or by KEAP1 mutational status (allele frequency >80%) for downstream plotting.

Pathway analyses

After differential analyses were performed for each dataset, significance cutoffs were made and displayed on the corresponding RNA-seq volcano plots. These significantly upregulated and downregulated hits, along with fold change and P value information for each hit, were used as inputs for Gene Ontology pathway analyses in the SRplot portal (https://www.bioinformatics.com.cn/en), in which significantly upregulated and downregulated biological process gene sets were identified and displayed as bubble plots and cnet plots that link the nested genes.

Top UXS1 codependencies and CRISPR screen hits were used as inputs for other pathway analyses throughout, which were performed in Metascape (RRID: SCR_016620; https://metascape.org/gp/index.html#/main/step1).

Dose–response curve fitting

Dose–response curve fits of viability fraction after treatments were calculated using GraphPad Prism (v10; RRID: SCR_002798. All cell line dose–response curves were fit using the following standard inhibitory Hill function:

$$f\left(c\right)\mathrm{=}f\left(\infty \right)\mathrm{+}\frac{f\left(0\right)\mathrm{-}f\left(\infty \right)}{1\mathrm{+}{\left(\frac{c}{{IC}_{50}}\right)}^n},$$

in which f(c) is the fraction of cells in the population at drug concentration c; f(0) and f(∞) are the fractions at no drug and at the maximal tested drug concentration, respectively; IC₅₀ is the half-maximal inhibitory concentration of the drug; and n is the Hill coefficient. f (c), f (0), and f (∞) were obtained from experiments; and the values of IC₅₀ and n were obtained from fitting.

Statistics

For bar charts and line plots used throughout the study, the bar or line represents the mean value and error bars denote the SD unless otherwise specified. For violin plots used throughout the study, thick lines represent the median values, and thin bars above and below each median represent the IQRs of the distribution. The full distributions are displayed by the full range of the violin shape, with the width along the violin corresponding with the value frequency. Data plotted throughout are representative of multiple independent experiments. Statistical tests were performed using GraphPad Prism and MATLAB, and test type was chosen on the basis of sample size: n = 3 to 20, unpaired t test (parametric); n > 20, permutation test (500 permutations performed per comparison). The permutation test function was accessed via MathWorks File Exchange (https://github.com/lrkrol/permutationTest). Significance levels are reported as P values ≤ 0.05 (*), 0.01 (**), 0.001 (***), and 0.0001 (****). Throughout, “ns” denotes no statistical significance (P > 0.05).

Results

KEAP1-mutant NSCLC cells are sensitive to UXS1 loss in a UGDH expression–dependent manner

We sought to identify selective dependency genes in KEAP1-mutant cancer cell lines in the DepMap loss-of-function CRISPR/Cas9 screen dataset (https://depmap.org/portal). The top scoring dependency genes included NFE2L2 (Nrf2), SLC333A1, SUCO, UXS1, and TAPT1 (Fig. 1A; Supplementary Fig. S1A). We validated some of these dependency hits by knocking down the genes individually in KEAP1-mutant NSCLC cells (Fig. 1B; Supplementary Fig. S1B–S1D). The dependency on NRF2 has been noted previously (29) and was expected as it is a direct ubiquitination target of KEAP1. SLC33A1, an endoplasmic reticulum (ER) transmembrane protein involved in Acetyl-CoA transport, was also expected, as it has previously been shown to be a dependency in KEAP1-mutant tumors (30).

Figure 1.

KEAP1-mutant NSCLC cells are sensitive to UXS1 loss in a UGDH expression–dependent manner. A, Dependency on the indicated genes (data from DepMap) of all cell lines harboring either KEAP1-WT or KEAP1-mutant status. B, Effects of UXS1 knockdown in KEAP1-mutant and -WT cell lines. C, Time-course of cell count following CRISPR gene KOs in a KEAP1-mutant cell line. PLK1 KO is used as a positive control for loss of proliferation. D and E, Xenograft tumor growth of H2122 and A549 cell lines expressing dox-inducible shRNA against UXS1 or a nontargeting control (NTC). F, Gene dependency data of UXS1 KO versus UGDH expression for all cell lines (data from DepMap). G, UGDH expression of KEAP1-WT or -mutant lung lines, color coded for UXS1 gene effect. H, Western blot of UGDH expression in the indicated cell lines. β-Actin is shown as a loading control, and quantitation of UGDH band intensities is shown below. I, KEAP1-mutant cells are sensitive to UXS1 knockdown, which is lost if UGDH is knocked out. P values were generated by unpaired t test. J, KEAP1-WT cells are sensitized to UXS1 loss by transient overexpression (OE) of UGDH. P values were generated by unpaired t test. K, Diagram showing the metabolic pathways involving UGDH and UXS1. **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001; ns, nonsignificant. TPM, transcripts per million.

We chose to focus further work on UXS1 as it is a novel understudied target that showed a strongly selective dependency in KEAP1-mutant cancer cell lines. We used dox-inducible shRNAs targeting UXS1 (shUXS1) or a nontargeting control (shNTC; Supplementary Fig. S1B) to confirm the effect of UXS1 knockdown in NSCLC cells. All KEAP1-mutant NSCLC cells we tested (H460, H2122, H2023, H1792, and H1944; Fig. 1B and C; Supplementary Fig. S1C–S1F) displayed a significant loss of viability and proliferation upon induction of UXS1 knockdown with dox treatment, whereas KEAP1-WT NSCLC cells (H1299 and Chago-K1) were completely unaffected by UXS1 loss (Fig. 1B; Supplementary Fig. S1F).

To test whether the observed in vitro dependency can be recapitulated within the in vivo tumor microenvironment, we subcutaneously inoculated H2122 or A549 cells expressing a dox-inducible shNTC or shUXS1 in C.B-17 SCID mice. Once tumors reached an average size of 200 mm³, mice were randomized and fed either regular chow or dox chow to induce UXS1 knockdown (Supplementary Fig. S1G). Whereas mice bearing shNTC tumors or those fed regular chow showed tumor growth over time, mice bearing UXS1 knockdown tumors displayed tumor stasis upon dox treatment (Fig. 1D and E).

UGDH expression is necessary and sufficient for dependency on UXS1

Upon further probing of the DepMap dataset, we found that UGDH expression is the top inversely correlated gene with UXS1 gene effect (Chronos score; Fig. 1F and G). UGDH is a NRF2 target gene (14) and thus is highly expressed in KEAP1/Nrf2-mutant cells relative to KEAP1/Nrf2 WT cells (Fig. 1G). We confirmed this by Western blotting for UGDH and staining for UGDH mRNA using RNA FISH. KEAP1-mutant cells showed a significantly higher expression of UGDH protein and mRNA relative to KEAP1-WT cells (Fig. 1H; Supplementary Fig. S1H). Furthermore, siRNA-mediated knockdown of Nrf2 in both KEAP1-mutant and -WT cells significantly decreased UGDH mRNA levels (Supplementary Fig. S1H).

Next, we asked whether UGDH expression is necessary and sufficient for dependency on UXS1. To test this, we first used CRISPR/Cas9 to knock out UGDH in a KEAP1-mutant cell line (H1944), followed by siRNA-mediated knockdown of UXS1. UGDH KO completely rescued dependency on UXS1 (Fig. 1I). Conversely, overexpressing UGDH in KEAP1-WT cells (Calu6) sensitized them to UXS1 loss (Fig. 1J). Collectively, we identified and validated UXS1 as a selective dependency in KEAP1-mutant NSCLC cell lines, driven by NRF2-dependent overexpression of UGDH (Fig. 1K).

UXS1 loss results in hyperaccumulation of UDP-GlcA and depletion of pyrimidine nucleotides

NRF2-driven high UGDH expression in KEAP1-mutant cells leads to increased UDP-GlcA synthesis that requires a functional UXS1 to convert it to UDP–xylose, and UDP–xylose has been shown to be a negative feed-back regulator of UGDH (15). Hence, we hypothesized that loss of UXS1 leads to a depletion of UDP–xylose as well as an increased synthesis and accumulation of UDP-GlcA. Indeed, KEAP1-mutant cells expressing dox-inducible shUXS1 displayed a time-dependent decrease in UDP–xylose and increase in UDP-GlcA (Fig. 2A). We also confirmed that UXS1 knockdown induced a significant UDP-GlcA accumulation in tumors after shUXS1 induction (Supplementary Fig. S1G). Additionally, reintroducing a previously described (31) catalytically dead form of the enzyme (UXS1^R236H) was unable to reverse the viability loss and UDP-GlcA buildup caused by UXS1 knockdown, whereas addition of UXS1^WT resulted in a complete rescue (Supplementary Fig. S2A and S2B), confirming that loss in enzymatic UXS1 activity is indeed responsible for the synthetic lethality observed in KEAP1-mutant cells.

Figure 2.

Loss of UXS1 enzymatic activity results in hyperaccumulation of UDP-GlcA and a loss of pyrimidine nucleotides. A, Time-course metabolic profiling of UDP–xylose and UDP-GlcA in A549 KEAP1-mutant cells expressing shNTC versus shUXS1. B, Metabolic profiling of UDP–xylose and UDP-GlcA in H460 and H2122 KEAP1-mutant cells after shUXS1 induction or UGDH KO. C and D, Targeted metabolomic profiling of H2122 and H460 KEAP1-mutant lung cancer cells after shUXS1 induction. As expected UDP-GlcA is the most increased metabolite. Notably, pyrimidine nucleotides (bold font) are significantly depleted. Right, metabolomic profiling after UGDH KO in the same cell lines, in which UXS1 loss no longer causes significant metabolic changes. E, Time-course metabolomics of A549 KEAP1-mutant cells showing the temporal dynamics of pyrimidine, but not purine, nucleotide loss. F, Rescue of pyrimidine nucleotides by uridine (100 µmol/L) or cytidine (100 µmol/L) supplementation after 6 days of shUXS1 induction in H2122 cells. G, Diagram of the metabolic process blocked by targeting UXS1, resulting in insufficient UDP recycling. *, P ≤ 0.05; ***, P ≤ 0.001; ****, P ≤ 0.0001; ns, nonsignificant.

As expected, UGDH KO led to a depletion of both UDP–xylose and UDP-GlcA (Fig. 2B; Supplementary Fig. S2C). Given that UGDH KO alone does not cause a viability loss (Fig. 1I), it can be reasoned that UXS1 dependency is not solely due to the depletion of UDP–xylose or disruption of proteoglycan synthesis. Because UXS1 resides in the Golgi, we performed high-content immunofluorescence imaging of the matrix protein GM130, finding that a six-day induction of shUXS1 caused no significant changes in Golgi morphology or area (Supplementary Fig. S2D). Together, these data indicate that interruption of glycoprotein processing is not the primary cause of UXS1 dependency and that additional mechanism(s) are at play.

Next, we performed metabolomics to understand the acute and global metabolic impact of UXS1 loss. Interestingly, we uncovered that UXS1 knockdown led to a significant depletion of pyrimidine nucleotides (UDP, UTP, CDP, and CTP), but not purine nucleotides, that is rescued by UGDH KO (Fig. 2C–E). Consistent with this, the dependency profile of UXS1 is more similar to and clusters with genes involved in nucleotide metabolism rather than those involved in proteoglycan synthesis (Supplementary Fig. S2E), suggesting that cell lines more sensitive to alterations in nucleotide metabolism are also sensitive to UXS1 loss.

We hypothesized that the constitutively high UDP-GlcA production, in the context of UXS1 loss, consumes the cell’s UDP pool and sequesters it in the ER/Golgi with no means of recycling UDP back to the cytosol. We reasoned that supplementing cells with uridine or cytidine could promote the synthesis of pyrimidine nucleotides via the salvage pathway and provide UXS1 knockdown cells with a cellular UDP pool. Indeed, H2122 treated with uridine showed rescue of all the pyrimidine nucleotides depleted by UXS1 knockdown, whereas cells treated with cytidine rescued the cytidine nucleotides but not the uridine nucleotides (Fig. 2F).

To determine whether UXS1 loss–induced increase in UDP-GlcA is large enough to trap a meaningful amount of cellular pyrimidines, we quantified the absolute levels of UDP-GlcA as well as pyrimidines (Table 1). Using the absolute quantification of the metabolites, we performed an ad hoc calculation of the molar contribution of pyrimidines to UDP-GlcA synthesis in NTC vs. shUXS1 samples. The ratio of UDP-GlcA to pyrimidines increased from 0.4 in NTC to 20.8 and 73.3 in shUXS1 cells exhibiting a 52-fold and 183-fold increase in H2122 and H460 cells, respectively (Table 1). This suggests that the demand for pyrimidines required for the increased synthesis of UDP-GlcA upon UXS1 loss far exceeds the available pyrimidine pool in the cell. Interestingly, although supplementation of cells with uridine rescues UTP levels, the ratio of UDP-GlcA to pyrimidines is unchanged because uridine supplementation further increases UDP-GlcA synthesis. Together, our data suggest that the substantial rerouting of pyrimidines toward UDP-GlcA upon UXS1 loss is sufficient to deplete the pool of pyrimidines normally available for nucleic acid synthesis, including DNA and RNA.

Table 1.

Absolute quantification of UDP-GlcA and pyrimidines (nmol/mg of protein) in NTC and shUXS1 H2122 and H460 cells, average of triplicate values shown.

Cell line	Condition	Uridine	UMP	UDP	UTP	CMP	CDP	CTP	Sum pyrimidines	UDPGA	UDP-GlcA/pyrimidines
H2122	NTC-dox	0.002	0.070	0.343	7.693	0.015	0.067	1.516	9.705	4.204	0.4
H2122	NTC-dox + uridine	0.012	0.087	0.517	13.640	0.023	0.083	2.090	16.451	4.704	0.3
H2122	shUXS1-dox	0.011	0.164	0.128	1.667	0.144	0.048	0.694	2.857	58.803	20.8
H2122	shUXS1-dox + uridine	0.076	0.089	0.296	6.199	0.006	0.066	1.248	7.980	154.341	19.6
H460	NTC-dox	0.002	0.049	0.264	6.075	0.023	0.071	1.756	8.240	3.546	0.4
H460	NTC-dox + uridine	0.004	0.079	0.485	11.867	0.024	0.112	3.041	15.611	4.870	0.3
H460	shUXS1-dox	0.013	0.070	0.044	0.599	0.058	0.025	0.465	1.273	91.319	73.3
H460	shUXS1-dox + uridine	0.023	0.073	0.055	0.747	0.065	0.030	0.556	1.549	156.139	102.8

NOTE: The ratio of UDP-GlcA to the sum of all pyrimidines reflects the significant increase in flux of pyrimidines toward UDP-GlcA synthesis.

An alternative pathway for UDP-GlcA utilization, and thus UDP recycling, is the glucuronidation pathway mediated by UGTs. UXS1-depleted cells treated with 4-methylumbelliferone, a UGT substrate that can be glucuronidated, displayed negligible reduction in UDP-GlcA, suggesting a continued hyperproduction of UDP-GlcA by UGDH despite it being used for glucuronidation (Supplementary Fig. S2F). Together, these metabolomic data indicate that UGDH^high cells rely on UXS1 activity to metabolize UDP-GlcA and recycle UDP to maintain pools of pyrimidine nucleotides, which cannot be sustained upon UXS1 ablation (Fig. 2G).

UXS1 loss in KEAP1-mutant cancer cells causes DNA replication stress

We next turned to RNA-seq to determine how the transcriptome responds to UXS1 loss. We evaluated two KEAP1-mutant cell lines, H2122 and A549, after shUXS1 induction and performed differential expression analyses relative to the matched shNTC conditions. UXS1 was the most significantly downregulated transcript in both cell lines, validating this approach (Supplementary Fig. S3A). After generating differential expression statistics, significantly downregulated or upregulated genes were independently subjected to pathway enrichment analysis (Supplementary Fig. S3B; Supplementary Table S1). The most downregulated processes included many gene sets related to DNA replication and mitotic division for H2122, along with many gene sets involved in mRNA processing and translational initiation in A549 (Fig. 3A, left).

Figure 3.

UXS1 loss in KEAP1-mutant cancer cells causes DNA replication stress. A, Pathway analysis of significantly altered genes from RNA-seq data in KEAP1-mutant lines after indicated times of shUXS1 induction. Left, downregulated pathways show significant loss of cell-cycle signaling and translational processes. Right, upregulated pathways show increased DNA damage signaling and p53-driven apoptotic pathways. B, DNA synthesis rate measured by EdU incorporation in KEAP1-mutant or -WT cells after 6 or 12 days of shRNA induction. C, Introduction of a live-cell biosensor for Geminin abundance, a metric of S and G₂ cell-cycle phases. Time-lapse imaging of this sensor after 4 days of dox pretreatment (shUXS1 induction) shows that KEAP1-mutant cells spend significantly longer durations in S/G₂. D, Immunofluorescence imaging of FANCD2, a protein that forms distinct nuclear foci at sites of stalled replication forks. Images and quantification demonstrate that shUXS1 causes a time-dependent significant increase in these DNA replication stress lesions, which can be rescued by supplementation of free pyrimidine nucleosides uridine (100 µmol/L) or cytidine (100 µmol/L). Turbo-RFP images are shown as a positive control for shRNA induction. *, P ≤ 0.05; **, P ≤ 0.01; ****, P ≤ 0.0001; ns, nonsignificant.

Because nucleotide imbalance, shown by our metabolomic profiling, is a known driver of DNA replication stress (32–35), and because genes involved in DNA replication are affected by UXS1 knockdown, we sought to determine whether UXS1 loss affects cell-cycle progression. To do so, we first measured the rate at which cells can incorporate the synthetic nucleotide EdU. After 6 or 12 days of shUXS1 induction, the percent of cells found in the S-phase (EdU-positive) was reduced in shUXS1 KEAP1-mutant cell lines H460 and H2122 but not in the KEAP1-WT line H1299 or in any of the shNTC conditions (Fig. 3B), indicating a reduced DNA synthesis rate (36, 37). As another way to observe S-phase progression, we next generated cells to express both the inducible shUXS1 construct as well as fluorescently tagged Geminin, an abundance readout that marks S and G₂ phases of the cell cycle (Fig. 3C, top left schematic; ref. 38). To measure S-phase progression after UXS1 loss in real time, we used time-lapse microscopy after 4 days of shUXS1 induction and measured the Geminin dynamics. In KEAP1-mutant shUXS1 cells H2023 and H2122, the duration of S–G₂ was significantly lengthened compared with the control condition (Fig. 3C, bottom and right), confirming that UXS1 loss in this genetic background causes prolonged cell-cycle completion.

Among the most enriched pathways for the upregulated genes after UXS1 loss, DNA damage checkpoints and apoptotic responses were strongly represented in both cell lines tested (Fig. 3A, right; Supplementary Fig. S3B), suggesting that UXS1 loss leads to DNA lesions that affect proliferation and survival. The Fanconi Anemia complementation group of proteins constitutes a conserved pathway for responding to situations of DNA replication stress, particularly those involved in nucleotide depletion (39, 40). Activation of the FA pathway is facilitated by ATR kinase activity (41, 42), which allows for formation and monoubiquitination of the FANCD2/FANCI heterodimer that binds DNA lesions, particularly stalled replication forks, and recruits repair elements (43). To test whether these lesions are present in cells sensitive to UXS1 loss, we immunostained for FANCD2 and performed high-resolution confocal imaging. In KEAP1-mutant H2122 cells, shUXS1 induction caused a significant time-dependent increase in nuclear FANCD2 foci (Fig. 3D), a hallmark of FA pathway activity (37, 44). Because we were able to rescue pyrimidine nucleotide levels with uridine supplementation (Fig. 2F), we hypothesized that we could prevent these FANCD2 lesions with such treatments, and indeed, uridine or cytidine nucleoside supplementation rescued this distinct DNA replication stress response phenotype (Fig. 3D).

DNA replication stress resulting from UXS1 loss leads to states of cell-cycle exit and apoptosis

DNA replication stress and resulting DNA damage are known triggers for cell-cycle exit and apoptosis, and we sought to quantify how cells move through these different fates (Fig. 4A) following UXS1 loss. To do so, we first performed analysis of exposed phosphatidyl serine and DNA content by flow cytometry, as a measure of early and late apoptosis, revealing varying degrees of apoptotic induction across KEAP1-mutant lines and no increased cell death in the WT cell line (Fig. 4B; Supplementary Fig. S4A and S4B). Because viable cells remained in all conditions, we next sought to quantify the extent of quiescence induction (cell-cycle exit) in these surviving cells. To measure cell-cycle (45) exit directly, we examined phospho-Rb status, as Rb is only phosphorylated as cells are currently committed to a cell cycle (46, 47). Upon measuring phospho-Rb in single cells, we observed that the fraction of cells that exited from the cell cycle (phospho-Rb–negative) significantly increased over time in the KEAP1-mutant setting (Supplementary Fig. S4C–S4E). Using our apoptosis and cell-cycle measurements together across 12-day treatment timescales for all cell lines, we were able to compile fate maps for each line, showing progressive sensitivities to UXS1 loss in KEAP1-mutant lines (Fig. 4C).

Figure 4.

DNA replication stress resulting from UXS1 loss leads to states of cell-cycle exit and apoptosis. A, Illustration of movement through cell states related to cell-cycle status and cell death. B, Quantitation of apoptotic cell states in the indicated cell lines after 0, 6, or 12 days of shUXS1 induction, measured by flow cytometry. Different states and corresponding quadrant gating are shown in Supplementary Fig. S4. C, Fate mapping after UXS1 loss across the indicated cell lines and treatment durations of dox. Pie charts are constructed by first bifurcating populations into viable or dead, using the flow cytometry apoptosis data. The viable cells are then stratified further by phospho-Rb status, as quantified by immunofluorescence imaging in Supplementary Fig. S4. D, Percentage of cells staining positive for SA-β-gal after the indicated treatments and times. Generated P value is from an unpaired t test. Uridine was used at 100 µmol/L. Palbociclib (1 µmol/L) was used as a positive control for generating senescent cells. E, Immunofluorescence imaging quantification of p21 protein after shUXS1 induction alone or in combination with uridine supplementation (100 µmol/L). F, Imaging and automated nuclei counting of H2122 cells after 7 days of shUXS1 induction alone or in combination with uridine (100 µmol/L) or cytidine (100 µmol/L) supplementation. **, P ≤ 0.01; ns, nonsignificant.

To understand if the cell-cycle exit observed after UXS1 loss is a long-arrested state, we turned to staining for SA-β-gal in our most sensitive KEAP1-mutant cell line, H2122. After 7 days of shUXS1 induction, we see that 20% of H2122 cells are positive for this senescence marker, whereas no increase is seen in the H1299 control WT line (Fig. 4D; Supplementary Fig. S4F). Additionally, we see that this increase in senescent cells can be partially rescued with uridine supplementation (Fig. 4D; Supplementary Fig. S4F). Because p21 (CDKN1A), another strong driver and indicator of quiescence and senescence (46, 47), was among the most significantly upregulated genes seen in the RNA-seq results (Figures S3A-B), we next evaluated p21 protein expression by immunofluorescence imaging. After shUXS1 induction, p21 protein expression was increased, and we could reverse this effect with the addition of uridine (Fig. 4E). RNA-seq results also confirmed partial but significant rescue in CDKN1A and other DNA damage signaling transcripts (Supplementary Fig. S4G). Finally, we attempted to rescue this shUXS1-driven proliferation defect by nucleoside supplementation and found that cell counts could indeed be partially rescued (Fig. 4F).

Together, these results show that prolonged states cell-cycle arrest after UXS1 loss are concomitant with elevated p21, a marker of the p53 response. The loss in cells is attributed to a combination of apoptosis and proliferative withdrawal, which can only be partially rescued by supplementing with uridine, ultimately indicating that UXS1 loss drives other deleterious changes in cells that cannot be acutely prevented with this supplementation approach.

CRISPR screen identifies sensitizers as well as suppressors of UXS1 dependency

To gain a deeper understanding of the mechanism of dependency on UXS1, we performed a whole-genome loss-of-function CRISPR screen in both KEAP1-mutant and -WT cell lines expressing dox-inducible shNTC and shUXS1 (Supplementary Fig. S5A). Briefly, cells were first transduced with a previously established whole-genome CRISPR KO library (18), followed by induction of shNTC or shUXS1 for 11 days. After DNA extraction, sgRNA abundance was determined by deep sequencing. Relative guide abundance in shUXS1 vs. shNTC cells was used to determine enrichments (resistance hits) and depletions (sensitivity hits; Supplementary Fig. S5A). Several common hits were observed that significantly overlap between both KEAP1-mutant cell lines tested (H2122 and H460) that were not observed in the KEAP1-WT cell line H1299 (Fig. 5A and B; Supplementary Fig. S5B–S5D; Supplementary Table S2). As expected, UGDH was one of the top resistance hits that suppresses UXS1 dependency in both KEAP1-mutant cell lines (Fig. 5A and B), confirming our previous results.

Figure 5.

CRISPR KO screening reveals additional effectors of sensitivity to UXS1 loss. A, Summary volcano plot of CRISPR KO screen in H460 and H2122 cells, plotted as the shUXS1 condition relative to shNTC induction after 11-day treatments. Positive and negative significance cutoff values for each axis are marked by the shaded boxes. B, Venn diagram of resistance gene hits for each cell line, in which the overlapping region between the two cell lines is displayed as an expanded table. UGDH marks an expected hit, and SLC35E3 and TP53 are highlighted as hits of interest based on their mechanistic role in nucleotide sugar transport and proliferation control, respectively. C, Validation experiments after KO of TGDS or SLC35E3 in H2122 cells and assessment of relative viability after 7 days of shUXS1 induction. Generated P values by unpaired t test. D, Validation experiments after KO of TGDS or SLC35E3 in H2122 cells and assessment of UDP-GlcA and UDP abundance after 4 days of shUXS1 induction. Generated P value summaries are from unpaired t tests. E, DepMap KO data of cell lines harboring either no TP53 mutations or at least one TP53 hotspot mutation. All cell lines are shown on the left, and UGDH^high lines are shown on the right. UGDH^high lines are defined as those with UGDH expression of log₂ (TPM+1) > 6 in the 24Q2 Public dataset. TPM, transcripts per million. F and G, Pathway analysis obtained through Metascape, in which the inputs were the CRISPR KO screen hits that conferred sensitivity, from A. Top 15 pathway hits are displayed. Common hits relate to cell-cycle progression, highlighted in red text. *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001; ns, nonsignificant.

SLC35E3 is another top resistance hit in both H460 and H2122 cells, which we validated by a viability assay using two different sgRNAs targeting SLC35E3 (Fig. 5C). The function of SLC35E3 is not well known, although one study has showed that it is involved in transporting UDP-GlcA to the ER lumen (48). We performed metabolomics to assess how loss of SLC35E3 affects the metabolic consequences of UXS1 knockdown. We found that SLC35E3 depletion blunted the accumulation of UDP-GlcA induced by UXS1 loss (Fig. 5D; Supplementary Fig. S5E) and completely rescued the depletion of UDP (Fig. 5D; Supplementary Fig. S5E), consistent with the rescue of cell death. Although we do not fully understand how SLC35E3 KO decreases the buildup of UDP-GlcA upon UXS1 loss, one plausible hypothesis is that keeping UDP-GlcA in the same compartment as UGDH (cytosol) by blocking UDP-GlcA transport to the ER lumen has a negative feedback effect on UGDH and overall UDP-GlcA production. Indeed, UGDH has been shown to be sensitive to substrate induced enzyme hysteresis (49, 50).

Thymidine diphosphate glucose-4,5-dehydratase (TGDS) is one of the top genes showing increased sensitivity to UXS1 loss (Fig. 5A; Supplementary Fig. S5D), which we validated by viability assays (Fig. 5C). Whereas TGDS is an enzyme with unknown function in humans (51, 52), it is the closest paralogue of UXS1, and they both belong to short chain dehydrogenase/reductase family of proteins (51, 52). We hypothesized that TGDS could partially compensate for the loss of UXS1 by metabolizing UDP-GlcA and increasing UDP levels, and hence a loss of TGDS synergizes with UXS1 by further increasing UDP-GlcA levels and exacerbating the depletion of UDP. Indeed, metabolomics showed modest but significant increases in UDP-GlcA and further decreases in UDP levels upon TGDS KO in the context of UXS1 loss (Fig. 5D; Supplementary Fig. S5E).

KO of TP53 also resulted in resistance to UXS1 loss (Fig. 5A and B), presumably due to the acute loss of its growth-suppressive functions in the face of DNA damage (53). In addition to TP53 itself, components of RNA polymerase II and mediator complex that have been shown to mediate the transcriptional activity of TP53 (54) were among the top resistance hits (Fig. 5A and B). This supports the model that pyrimidine depletion induces DNA replication stress and leads to activation of the p53–p21 axis and results in cell-cycle arrest and cell death. Hence, acute loss of TP53 allows cells to persist in the face of nucleotide stress. In contrast to these findings pertaining to acute loss of TP53, analysis of the DepMap dataset showed no significant difference in UXS1 dependency between TP53-mutant and -WT cell lines, suggesting different cellular responses to UXS1 loss in cells with acute TP53 loss vs. those adapted to TP53 loss (Fig. 5E).

To determine which functional processes when deregulated confer greater sensitivity to UXS1 loss, we ran a pathway analysis of the top negative hits in H460 and H2122, finding that cell-cycle progression and mitotic genes were significantly represented (Fig. 5F and G), suggesting that UXS1 loss sensitizes cells to cell-cycle perturbations. Taken together, these screening results extend our understanding of UXS1 dependency in KEAP1-mutant cells, opening new avenues of research to exploit this target.

UXS1 loss sensitizes KEAP1-mutant cells to cell-cycle checkpoint kinase inhibitors

Insufficient resources for DNA synthesis will lead to stalled replication forks and activation of the ATR kinase and downstream signaling through checkpoint kinase 1 (Fig. 6A; ref. 55). Checkpoint kinase 1 then phosphorylates and activates PKMYT1 and WEE1 kinases (55), which go on to place inhibitory phosphorylations on core cell-cycle kinases CDK2 and CDK1 (56, 57), ultimately resulting in slowed cell-cycle progression allowing DNA replication to recover before mitotic entry. Because slowed cell-cycle dynamics and DNA replication stress were observed following shUXS1 induction (Fig. 3A–D), we sought to test the activity of this stress signaling axis in this context. To do so, we first measured phospho-CDK2 (Y15) and phospho-CDK1 (Y15), the consequences of either WEE1 or PKMYT1 kinase activity. UXS1 depletion in KEAP1-mutant H2122 cells caused a 2-fold increase over baseline in phospho-CDK2 and a 4-fold increase in phospho-CDK1, respectively (Fig. 6B). Treatment with a WEE1 inhibitor reduced these inhibitory phospho-species, confirming the dependency on this kinase activity (Fig. 6B).

Figure 6.

UXS1 loss sensitizes KEAP1-mutant cells to cell-cycle checkpoint kinase inhibitors. A, Illustrated mechanism of DNA replication stress signaling after UXS1 loss. B, Single-cell quantification of fluorescent imaging for EdU versus pCDK2 or pCDK1, inhibitory phosphorylations made by checkpoint kinases. Cells cycling through shUXS1 induction experience increases in these replication stress marks. WEE1 inhibitor (adavoserib, 250 nmol/L) was used as a positive control to reduce these phosphospecies. C and D, Imaging and single-cell quantification of γH2AX after UXS1 loss, alone or in combination with PKMYT1i (lunresertib, 500 nmol/L) or WEE1i (250 nmol/L). E, Flow cytometric measurement of apoptotic cells (Annexin V–positive) after indicated treatments in H2023 cells. Generated P value summaries are from unpaired t tests. F and G, Dose–response experiments combined with dox-inducible shUXS1. UXS1 loss offers more complete response to these kinase inhibitors and to 5-fluorouracil (5-FU) at a later time point (ATR inhibitor used was ceralasertib). H, Time-course combinations across multiple KEAP1-mutant or -WT cells. Responder lines experience cooperative killing from the combinations tested. *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001.

After confirming checkpoint kinase activity after UXS1 loss, we hypothesized that inhibiting this stress signaling axis would force these cells through early mitoses without sufficient completion of DNA replication, ultimately leading to increased cell death. To first characterize this, we cotreated H2122 cells to induce shUXS1 and inhibit WEE1 or PKMYT1 and measured DNA lesions in single cells by way of γH2AX imaging. Whereas UXS1 loss caused a significant increase in γH2AX staining in cycling cells (EdU-positive), additional inhibition of WEE1 or PKMYT1 deepened the response to a more significant level compared with any of the monotherapy conditions (Fig. 6C and D). WEE1 inhibition combined with UXS1 loss also resulted in cooperative induction of apoptosis after 6-day treatments (Fig. 6E). These results indicate that this combination indeed pushes DNA damage into a more intolerable range for KEAP1-mutant cells.

To next assess how the DNA damage induced by these combinations affect viability phenotypes, we performed dose–response analysis for WEE1 and PKMYT1 kinase inhibitors combined with a time-course of shUXS1 induction in three KEAP1-mutant cell lines, H460, H2122, and H2023 (Fig. 6F and G; Supplementary Fig. S6A and S6B). In H460 cells, this combination caused cooperative viability loss, with the strongest response demonstrated by the largest drop in IC₅₀ value at the 12-day treatment time point (Fig. 6F and G). In H2122 and H2023 cells, which are more sensitive to UXS1 loss alone, WEE1 or PKMYT1 were also able to further decrease cell viability (Supplementary Fig. S6A and S6B). We also tested the combination of shUXS1 with two other clinical compounds: an inhibitor of ATR, which is upstream of WEE1 and PKMYT1 (Fig. 6A), or 5-fluorouracil, a nucleotide analog that will cause DNA synthesis aberrations when incorporated. In both additional drug contexts, we indeed identified cooperativity in cell killing across all three cell lines tested (Fig. 6F and G; Supplementary Fig. S6A and S6B), a result of exploiting the DNA replication stress phenotype caused by UXS1 ablation. Finally, we assessed the effects of these drug combinations on proliferation over time in these three KEAP1-mutant lines and the KEAP1-WT line H1299. To do so, we selected the lowest dose for each clinical molecule that caused significant cooperativity in the dose–response studies. Relative to the shUXS1 alone or drug monotherapy groups, the combinations all caused cooperativity in growth inhibition for all the KEAP1-mutant cell lines, and no additive effects in the KEAP1-WT cells (Fig. 6H). Together, these results demonstrate that targeting UXS1 in vulnerable cells elicits a DNA replication stress phenotype that can be further leveraged for tumor killing by cotreating with targeted therapies proximal to DNA synthesis.

UXS1 loss is well tolerated in normal mouse liver tissue

The expression of UGDH is generally significantly higher in tumors as compared with normal tissues (58), making UXS1 a potentially viable clinical target. Because the liver has the highest UGDH expression as compared with other normal tissues (58), we asked whether depleting UXS1 in the liver would be associated with any adverse effects. To this end, we used a LNP formulation to deliver sgRNAs targeting Uxs1, Ugdh, Uxs1 plus Ugdh in combination, Ugt1a1, and GFP (positive control) to the livers of C57BL/6 mice expressing Cas9-P2A-EGFP (Fig. 7A). Ugt1a1 loss in the liver is known to cause an increase in serum bilirubin levels, and thus Ugt1a1 KO was used as a positive control for confirming and tracking successful LNP delivery and gene KO in the liver of live mice. LNPs encapsulating the sgRNAs were intravenously injected for a total of 3 times every other day (Fig. 7A). Blood was drawn every week to monitor TBIL, aspartate aminotransferase, alanine transaminase, alkaline phosphatase, and readouts associated with adverse liver effects.

Figure 7.

UXS1 loss is well tolerated by normal mouse livers, a tissue with high UGDH expression. A, Schematic of LNP delivery for mouse liver assessments. IV, intravenous. B, Western blots of liver tissue lysates after targeted KOs via LNP delivery. EGFP is used as a delivery control. Ugt1a1 KO is used as a positive control for loss of glucuronidation in downstream assays. C, Body weight measurements over the duration of the in vivo LNP experiment. Mean and SEM of six mice plotted per time point. D, Measurement of liver enzymes at indicated times of the in vivo LNP experiment. Mean and SEM of six mice plotted per time point. E and F, Metabolite measurements for each tumor sample, presented as boxplots showing the full range and individual samples. Generated P values by unpaired t test. Bilirubin buildup is measured as a metabolite of hepatotoxicity, seen in the positive control Ugt1a1 KO but not in Uxs1 KO; P values are generated by unpaired t test. ***, P ≤ 0.001; ****, P ≤ 0.0001; ns, nonsignificant. ALKP, alkaline phosphatase; ALT, alanine transaminase; AST, aspartate transferase.

To confirm guide delivery and efficiency, we immunoblotted for each protein and observed an almost complete KO of EGFP, Ugdh, and Ugt1a1 (Fig. 7B). Uxs1 KO was confirmed by analysis of genomic DNA for frequency of insertions and deletions near CRISPR PAM sites (determined by the Inference of CRISPR Edits, ICE, algorithm; Supplementary Fig. S7A; ref. 59). All mice showed no change in liver enzymes or body weight until the end of the experiment (Fig. 7C and D). Mice injected with LNPs encapsulating Ugt1a1 sgRNAs showed an increase in TBIL levels starting on day 7 and further increased in the following weeks (Fig. 7D and E), which was accompanied by yellow serum; this indicated that LNP delivery and KO was successful within the first week of injections.

Mice were euthanized after 6 weeks and livers were harvested. Metabolomics analysis of the liver tissues demonstrated a consistent pattern with our in vitro observations. Uxs1-KO livers displayed significantly higher UDP-GlcA levels as compared with EGFP and Ugt1a1-KO controls, which were completely rescued by Ugdh-KO (Fig. 7E; Supplementary Fig. S7B). On the other hand, UDP–xylose was significantly decreased in both Uxs1 and Ugdh-KO conditions (Fig. 7E; Supplementary Fig. S7B). Similar to Ugdh-high cells in vitro, Uxs1-KO liver tissues exhibited a significant decrease in pyrimidine phosphates (UDP, UTP, CDP, and CTP; Fig. 7F; Supplementary Fig. S7B). The lack of liver toxicity despite the depletion of pyrimidine phosphates in Uxs1-KO liver as compared with NSCLC cells is likely because liver tissue has a slow rate of cellular turnover in which close to 99% of nontransformed hepatocytes are in the G0 (quiescent) state (60). Together, our data reinforce UXS1 as a UGDH^high cancer-selective target with minimal impact on the liver (61).

Discussion

The concept of synthetic lethality to treat tumors was proposed in 1997 by Hartwell and colleagues (62) and reduced to practice in 2005 by the Ashworth and Helleday laboratories (63), which showed that PARP inhibitors were effective at killing cells with mutations in BRCA1 or BRCA2. Cells defective in these DNA damage repair pathways have further been shown to be preferentially sensitive to PARG, POLQ, USP1, and RAD51/52 inhibitors (64). In these cases, compromised DNA repair by homologous recombination mediated by BRCA1 and BRCA2 renders the cells more dependent on alternative repair mechanisms. This has been termed contextual synthetic lethality. Another example of this is the dependency on the Werner helicase WRN to resolve complex DNA structures that evolve in mismatch repair–deficient colorectal, ovarian, and endometrial tumors (65–67).

A previous study screened isogenic Kras^{G12D/+;p53−/−} murine lung adenocarcinoma cell lines with or without KEAP1 deletion to identify druggable KEAP1-mutant dependencies (30). The ER membrane acetyl-CoA transporter SLC33A1 was a top hit in this approach, and the high score of SLC33A1 in our analyses is consistent with this (Supplementary Fig. S1A). Loss of SLC33A1, as well as TAPT or SUCO, resulted in exacerbation of an ER stress pathway, resulting in a loss of cell viability. In contrast to the mechanism underlying the sensitivity of KEAP1-mutant cells to these genes, UXS1 dependency is generated by a collateral increased level of UGDH because of its transcriptional activation by NRF2. This renders these cells susceptible to massive increases in UDP-GlcA levels upon loss of UXS1, resulting in depletion of UDP levels needed for effective DNA and RNA synthesis.

A recent study also uncovered the requirement for UXS1 activity in cells with high UGDH expression (58). This group also showed large increases in UDP-GlcA following KO of UXS1 in these cells, which resulted in aberrant Golgi morphology and altered glycosylation of several cell-surface proteins, including EGFR. Although we did not observe as dramatic effects on Golgi content following knockdown of UXS1 in our experiments (Supplementary Fig. S2D), this pathway could be important in contributing to some of the cytotoxic effects of UXS1 depletion. Due to the metabolic bottleneck that we have identified, we conclude that the depletion of pyrimidine phosphates plays a more proximal and acute role in this synthetic lethal process.

Disruption of metabolic pathways can cause critical genomic or translational resource deficits, which has been a strategy to prevent tumor growth, albeit with challenges to execute in a tumor-selective manner. Certain therapies such as hydroxyurea have been developed and leveraged to limit nucleotide production (39, 68), and imbalances in these resources have significant impact on proliferation (33). Analogously, inhibition of amino acid biosynthesis has also been explored to cause translational stalling and prevent tumor growth, but these approaches have been challenging to implement due to such therapies having poor tumor selectivity, eliciting immune responses in humans, or being subject to resistance (69). Recent work examining the removal of individual amino acids has shown that a subset of amino acids (e.g., methionine, lysine, etc.) have significant control over cell-cycle activity when depleted alone (70), suggesting that biosynthetic pathways for specific amino acids could be a focus for future metabolic cancer therapies, akin to our findings of selective pyrimidine depletion shown in this study.

The mechanism of action of classical chemotherapy drugs is through interference with DNA and RNA synthesis, inducing severe genomic stress that leads to cell death. However, as mentioned above, most chemotherapies are not selective and are associated with broad toxicity to normal cells. Our data suggest that targeting UXS1 leads to a selective DNA replication stress only in KEAP1-mutant/UGDH-high cancer cells, essentially acting as a selective chemotherapy. This has a significant clinical implication as cancer cells have generally higher levels of UGDH expression relative to normal cells. Furthermore, Doshi and colleagues (58) showed chemoresistant cancer cells increase UGDH expression levels, ultimately sensitizing them to UXS1 loss.

KEAP1-mutant cancer subtypes come at a high incidence and often become resistant to the current standard of care treatments, presenting a critical unmet need in the field. Additionally, methylation of KEAP1 promoter has been found to be prevalent in breast cancers with no somatic KEAP1 gene mutations (71), resulting in high UGDH expression. NFE2L2 is also mutated in squamous lung cancers and head and neck tumors (72, 73), resulting in high UGDH expression (30). A recent group also identified UXS1 dependency in biliary tract cancers (74), demonstrating that this target may have broad applicability beyond the KEAP1-mutant setting. Our findings highlight UXS1 as an effective and potentially safe drug target, alone or in combination with targeted cell-cycle therapies, for the treatment of KEAP1/NRF2-defective or UGDH-overexpressing cancers.

Supplementary Material

Table S1

Significantly upregulated and downregulated genes in RNA-seq data

Table S2

Significantly enriched or depleted guides from CRISPR knockout screen

Table S3

Primers for CRISPR screen

Table S4

Guide information for LNP studies

Figure S1

KEAP1-mutant NSCLC cells are sensitive to UXS1 loss in a UGDH expression-dependent manner

Figure S2

Loss of UXS1 enzymatic activity results in hyper-accumulation of UDP-GlcA and a loss of pyrimidine nucleotides

Figure S3

UXS1 loss in KEAP1-mutant cancer cells causes DNA replication stress

Figure S4

DNA replication stress resulting from UXS1 loss leads to states of cell-cycle exit and apoptosis

Figure S5

CRISPR knockout screening reveals additional effectors of sensitivity to UXS1 loss

Figure S6

UXS1 loss sensitizes KEAP1-mutant cells to cell-cycle checkpoint kinase inhibitors

Figure S7

UXS1 loss is well tolerated by normal mouse livers, a tissue with high UGDH expression

Key Resources Table

Critical reagents used with corresponding identifiers

Data Availability

Further information and requests for resources and reagents should be directed to the lead contact, David Stokoe (dhstokoe@calicolabs.com). Cell lines and constructs used in this work are available from the lead contact upon request after a Material Transfer Agreement has been approved by Calico Life Sciences. Significantly altered genes from bulk RNA-seq analyses are listed in Supplementary Table S1. Significant gene summaries from the CRISPR screen are listed in Supplementary Table S2, and corresponding primers used are listed in Supplementary Table S3. Encapsulated gRNAs in the LNP delivery studies are listed in Supplementary Table S4. RNA-seq and CRISPR screen raw FASTQ data as well as count files are available on GEO under accession numbers GSE305124 and GSE304915, respectively. Any additional information required to reanalyze the data reported in this article is available from the lead contact upon request.

Authors’ Disclosures

A. Boudreau reports he was an employee at Calico Life Sciences at the time he contributed to the work and is a current employee of Antares Therapeutics. J. Gajda reports personal fees from AbbVie outside the submitted work. C. Grant reports other support from AbbVie during the conduct of the study and outside the submitted work; in addition, AbbVie is collaborating with Calico for activities described in this publication. W.R. Buck reports being an employee of AbbVie and may own Abbvie stock. J.A. Hickson reports personal fees from AbbVie outside the submitted work. B. Shotwell reports personal fees from AbbVie outside the submitted work. No disclosures were reported by the other authors.

Authors’ Contributions

M.T. Gebru: Conceptualization, formal analysis, investigation, writing–original draft, writing–review and editing. T.E. Hoffman: Conceptualization, formal analysis, investigation, writing–original draft, writing–review and editing. A. Boudreau: Conceptualization, formal analysis, investigation. N. Vu: Formal analysis, investigation. F. Han: Formal analysis, investigation. P. Wang: Formal analysis, investigation. E. Folk: Formal analysis, investigation. X. Wu: Investigation, methodology. J. Newton: Conceptualization. K.D. Foster-Duke: Formal analysis, investigation. J. Gajda: Formal analysis, investigation. S. Villa: Investigation. N. Yang: Data curation. C.M. Sandoval: Methodology. C. Grant: Formal analysis, investigation. S.E. Wildeboer: Investigation. W.R. Buck: Investigation. J.A. Hickson: Formal analysis, supervision. A.J. Firestone: Formal analysis, visualization, methodology. M. Kort: Resources. B.D. Bennett: Conceptualization, resources, formal analysis, supervision, investigation, methodology. J.B. Shotwell: Conceptualization, resources. N.S. Wilson: Conceptualization, resources, supervision. D. Stokoe: Conceptualization, resources, formal analysis, supervision, writing–original draft, writing–review and editing.