NAPRT Silencing in FH-Deficient Renal Cell Carcinoma Confers Therapeutic Vulnerabilities via NAD⁺ Depletion

Hereditary leiomyomatosis and renal cell carcinoma (HLRCC) is a rare inherited cancer syndrome characterized by germline mutation and subsequent loss of heterozygosity of the fumarate hydratase (FH) gene. HLRCC leads to the development of FH-deficient RCCs, which are extremely aggressive with a high propensity to metastasize even at a small size (1). Currently, the only curative approach is early detection and surgical resection of localized disease (1). Although there are some targeted therapies that have shown modest activity, none are curative (2, 3).

A hallmark of FH-deficient RCC is the disruption of the tricarboxylic acid (TCA) cycle through loss of function of FH and accumulation of the oncometabolite, fumarate. Mutations in other TCA cycle enzyme genes, including isocitrate dehydrogenase 1/2 (IDH1/2) and succinate dehydrogenase (SDH), have been reported in diverse cancer types, and oncometabolites, 2-hydroxyglutarate and succinate, contribute to disease progression for these malignancies (4, 5). Oncometabolites competitively inhibit a diverse family of enzymes known as a-ketoglutarate-dependent dioxygenases (6, 7). Notably, this includes the DNA demethylases ten–eleven translocation (TET) methylcytosine dioxygenases, which catalyze the demethylation of 5-methylcytosine. Oncometabolites have been found to induce a DNA hypermethylation phenotype, which has been extensively studied in the context of IDH1/2 mutations as well as FH and SDH-deficient renal cancers (8–14). In oncometabolite-producing cancers, the hypermethylation phenotype affects multiple cancer-related pathways that promote disease progression, such as epithelial to mesenchymal transition and stemness-related factors (9, 13).

Though clinically actionable biomarkers are poorly understood in FH-deficient RCC, genomic and epigenomic analyses of primary tumors identified global DNA hypermethylation patterns in the majority of cases (12, 13). FH-deficient RCCs exhibited similar methylation patterns to CpG island methylator phenotype (CIMP) RCC tumors from The Cancer Genome Atlas (TCGA) papillary RCC cohort (KIRP) and were associated with higher rates of metastasis (12, 13). CIMP tumors are characterized by DNA hypermethylation in regions with numerous CpG sites, or CpG islands, surrounding the promoters of multiple genes. Hypermethylation in FH-deficient RCC was identified in several tumor suppressor gene loci; however, the functional consequences of the DNA hypermethylation phenotype have not been fully elucidated. Further study of genes impacted by hypermethylation will help to devise therapeutic strategies to target FH-deficient RCC. In this study, we identify a targetable vulnerability in NAD⁺ biosynthesis linked to the hypermethylation phenotype. We characterize the downstream consequences of disruption of NAD⁺ biosynthesis and identify potential drug combinations to exploit this vulnerability. Overall, this report proposes a novel biomarker-based therapeutic strategy for targeted treatment of FH-deficient RCC.

The SV40-immortalized Yale University normal kidney 1 (YUNK1) cell line [derived from a nephrectomy specimen as previously described (15)] was maintained in DMEM (Gibco). YUNK1 shFH and shRNA control (shNT) cell lines were generated as previously described (15), using a constitutive short hairpin RNA (shRNA) system targeting FH or a nontargeting shRNA. UOK262 (FH^Q396P/−; RRID:CVCL_1D72) and NCCFH1 (FH^−/−) cells as previously described (16, 17) were maintained in RPMI. UOK262 cells were a gift from Dr. W. Marston Linehan at the NCI, Urologic Oncology Branch. NCCFH1 cells were a gift from Dr. Bin Tean Teh at the National Cancer Centre of Singapore. UOK268 (FH^H192D/−) cells (RRID:CVCL_1D73) as previously described (18) were maintained in DMEM and 1-mmol/L sodium pyruvate. Normal human astrocytes were a gift from Dr. Timothy A. Chan at the Cleveland Clinic and were maintained as previously described (8) in DMEM.

As previously described (19), U87 (ATCC, Cat. # HTB-14), U87 IDH1^R132H/+ (ATCC, Cat. # HTB-14IG), HT1080 IDH1 R132C (ATCC, Cat. # CCL-121, RRID:CVCL_0317), and SW1353 IDH2 R172S (ATCC, Cat. # HTB-94, RRID:CVCL_0543) cells were purchased from ATCC. U87 cells were maintained in DMEM-F12 (Gibco). HT1080 and SW1353 cells were maintained in DMEM. HUCCT1 (Cat. # RCB1960, RRID:CVCL_0324), TFK1 (Cat. # RCB2537, RRID:CVCL_2214), and RBE (Cat. # RCB1292, RRID:CVCL_4896) cells were purchased from the RIKEN Cell Bank and maintained in RPMI. All media was supplemented with 10% FBS (Sigma) and 1% pen/strep (Gibco). All cell lines were tested for mycoplasma contamination when initially thawed for culturing. Details on methods and testing dates are provided (Supplementary Table S1). Cells were collected for experiments 1 to 6 weeks after thawing.

Where short tandem repeat (STR) information for cell line authentication was available, (U87, U87 IDH1^R132H/+, HT1080 IDH1^R132C/+, SW1353 IDH2^R172S/+, TFK1, RBE, and HUCCT1), cell lines were authenticated through the Yale Keck Biotechnology Core with STR genotyping by fragment analysis. For cell lines where STR information was unavailable (UOK262, UOK268, and normal human astrocytes), STR profiles were generated via Yale Core or IDEXX BioAnalytics and will be made available upon request. The STR profile for the NCCFH1 cell line was verified with STR profiling done by the National Cancer Center Singapore (17).

FH mutations in NCCFH1, UOK262, and UOK268 cells were validated by performing PCRs for exon 8, exon 9, and exon 5 respectively using the following primers:

• Exon 8 Fwd: 5′-GTT​GGG​CCT​TGC​TTT​ATT​GT-3′,

• Exon 8 Rev: 5′-ACC​CAA​CTA​CCC​AAT​GTG​GA-3′,

• Exon 9 Fwd: 5′-GTG​CCT​TCA​AAT​GTT​CAT​GC-3′,

• Exon 9 Rev: 5′-GCT​GTT​CTC​AAA​CAC​TGA​TCC​A-3′,

• Exon 5 Fwd: 5′-GAA​GTT​TGT​TTT​TGT​TGC​CTC​TG-3′,

• Exon 5 Rev: 5′-ATT​GGC​CAT​TTG​TAC​CAA​GC-3′.

IDH1/2 mutations in HT1080, RBE, and SW1353 were validated by performing PCRs for exon 4 in the case of both IDH1 and IDH2 mutations using the following primers:

• IDH1 Fwd: 5′-AAT​GAG​CTC​TAT​ATG​CCA​TCA​CTG-3′

• IDH1 Rev: 5′-TTC​ATA​CCT​TGC​TTA​ATG​GGT​GT-3′

• IDH2 Fwd: 5′-TGC​AGT​GGG​ACC​ACT​ATT​ATC-3′

• IDH2 Rev: 5′-TGT​GGC​CTT​GTA​CTG​CAG​AG-3′

PCR was performed using the PrimeSTAR Max DNA Polymerase (Takara Bio) according to the manufacturer’s instructions, followed by Sanger sequencing through the Yale Keck Biotechnology Core (Supplementary Fig. S1A–S1G). The IDH1 mutation in the U87 cell line was validated through western blotting with an IDH1^R132H-specific antibody (Dianova Cat. # DIA-H09, RRID:AB_2335716).

Human tissue procurement and analysis procedures for specimens from patients with HLRCC were approved by the Institutional Review Board at Yale University (IRB #2000027341) and at UCLA (IRBs #18-001988, #20-000790) in accordance with applicable laws, regulations, and guidelines in the United States and other countries, including, but not limited to, U.S. Department of Health and Human Services regulations (45 CFR Part 46), the Belmont Report, World Medical Association Declaration of Helsinki, and Council for International Organizations of Medical Sciences. Informed written consent was obtained from each patient. HLRCC tumor and normal tissue were analyzed via hematoxylin and eosin and IHC staining.

Patient-derived xenograft (PDX) generation was conducted at Champions Oncology through the Champions Clinical Laboratory Improvement Amendments clinical program, and informed written consent was obtained from each individual covering the use of de-identified patient information and tumor material for research purposes. For all PDX models, stock mice were bilaterally implanted with fragments from Champions TumorGraft models. All experiments and procedures were approved by the Institutional Animal Care and Use Committee of Champions Oncology. Tissue microarray of samples from renal cell carcinoma PDX models was generated at Champions Oncology. Additional data for the PDX models are available from Champions Oncology’s database (https://lumin.championsoncology.com).

An anti-NAPRT mouse monoclonal antibody (4A5D7) was developed and validated by Promab for Alphina Therapeutics. NAPRT IHC specificity and staining optimization were performed on cell line models and normal human tissue by NeoGenomics Laboratories using a Leica Bond III. Briefly, epitope retrieval was performed for 25 minutes at 100°C, blocked with Leica Protein Block for 30 minutes, and incubated with primary antibody (0.5 µg/mL) for 30 minutes. 3-3-Diaminobenzidine (DAB) was used for colorimetric detection. Samples were counterstained with hematoxylin. NAPRT signal (DAB intensity) was found to be binary in blinded scoring by two pathologists and was evaluated as either negative or positive in regions confirmed to contain tumor. In PDX TMA cores, negative control mouse antibody (Leica Biosystems Cat. # PA0996, RRID:AB 10554595) was used to evaluate background staining.

FK866 (Selleckchem), GNE-617 (Selleckchem), GMX1778 (Selleckchem), olaparib (Selleckchem), and BMN673 (Selleckchem) were dissolved in DMSO and used as indicated. 5-aza-2-deoxycytidine (Sigma) was solubilized in a 1:1 ratio of acetic acid:water and used as indicated. Nicotinic acid (Sigma) was solubilized in 1-mol/L NaOH and then diluted in complete media immediately prior to treatment alone or in combination with FK866. Methyl methane sulfonate (MMS; Sigma) was diluted in cell media for indicated treatments.

In vitro cellular viability was assessed using a previously described microscopy platform developed by our group (20). Briefly, cells were seeded in triplicate in a 96-well plate 24 hours prior to treatment. Plates were treated with vehicle (0.5% DMSO) or serial dilutions of drug. Following 6 to 8 days of incubation, cells were fixed and stained with 1-µg/mL Hoechst 33342 for 30 minutes. Cells were imaged using a Cytation3 (BioTek) and counted via CellProfiler (https://cellprofiler.org, RRID:SCR_007358). To assess viability, survival curves and IC₅₀ calculations were generated by GraphPad Prism, fitting data to an (inhibitor) versus response variable slope four-parameter nonlinear regression. For two drug synergy assays, synergy was calculated using the software Combenefit (21).

Cells were lysed in RIPA buffer with Protease Inhibitor Cocktail (Roche) and sonicated. For subcellular fractionations, cell membranes were sequentially lysed according to manufacturer instructions (ThermoFisher). Proteins were separated by SDS-PAGE with 4% to 12%, Bis-Tris gels (Invitrogen) and transferred to polyvinylidene difluoride membrane. Membranes were blocked with 5% BSA in 1× Tris Buffered Saline with Tween-20 for 1 hour, washed, and incubated overnight at 4°C with primary antibodies raised against: Vinculin (Cell Signaling Technology, Cat. # 13901, RRID:AB_2728768, 1:1,000), Fumarase (Cell Signaling Technology, Cat. # 4567, RRID:AB_11178522, 1:1,000), NAPRT (Proteintech, Cat. # 66159-1-Ig, RRID:AB_2881555, 1:3,000), NAMPT (Cell Signaling Technology, Cat. # 86634, RRID:AB_2800084, 1:1,000), anti-poly ADP-ribose binding reagent (Millipore Sigma, Cat. # MABE1031, RRID:AB_2665467, 1:10,000), GAPDH (Proteintech, Cat. # 60004-1-Ig, RRID:AB_2107436, 1:5,000), PARP (Cell Signaling Technology, Cat. # 9542, RRID:AB_2160739, 1:1,000), Sirt6 (Cell Signaling Technology, Cat. # 12486, RRID:AB_2636969, 1:1,000), H3 (GeneTex, Cat. # GTX122148, RRID:AB_10633308, 1:1,000), IDH1^R132H (Dianova, Cat. # DIA-H09, RRID:AB_2335716, 1:1,000), and PPM1D (Santa Cruz Biotechnology, Cat. # sc-376257, RRID:AB_10986000). Membranes were then incubated with secondary antibodies of HRP conjugated anti-mouse (Thermo Fisher Scientific, Cat. # 31432, RRID:AB_228302, 1:5,000) or HRP conjugated anti-rabbit (Thermo Fisher Scientific, Cat. # 31462, RRID:AB_228338, 1:5,000) for 2 hours at room temperature. Immunoblots were exposed to Clarity Western ECL substrate (Bio-Rad) and imaged via ChemiDoc (Bio-Rad). In some cases, membranes were stripped with Restore Western Blot Stripping Buffer (Thermo Fisher Scientific) for 5 minutes at room temperature before being blocked and probed with a new primary antibody.

mRNA transcripts were purified from cells using an RNAeasy kit (Qiagen) and subsequently reverse transcribed using a High-Capacity cDNA reverse transcription kit (Applied Biosystems). NAPRT, ACTB, and QPRT gene expression levels were assessed through qPCR with TaqMan fluorescent probes (Applied Biosystems): NAPRT (4351372), ACTB (4331182), and QPRT (4331182) according to the manufacturer’s protocol. Expression level fold change was calculated via ΔΔCt comparison, using ACTB as a reference gene. Quadruplicate qPCR reactions for each sample were run on a StepOnePlus Real Time PCR system (Applied Biosystems).

UOK262 cells were transfected with 2 µg of a NAPRT cDNA plasmid (Genscript, Cat. # OHu28558D) and an FH cDNAplasmid (GeneCopoeia, Cat. # EX-T3095-M68) using Lipofectamine 3000 (Thermo Fisher Scientific). Forty-eight hours after transfection, cells were treated with either 1.44-mg/mL G418 (Geneticin) to select NAPRT-positive cells or 1-µg/mL puromycin (InvivoGen, Cat. # ant-pr-1) to select FH-positive cells. Cells were cultured in antibiotic selection for 2 weeks after which protein overexpression was confirmed through western blotting.

Cells were seeded in chamber slides well and incubated overnight. At the indicated times after treatment, cells were fixed with 4% EM grade PFA (Electron Microscopy Sciences) for 15 minutes. Cells were permeabilized with 0.5% Triton X100 and blocked with 5% goat serum and 0.2% Triton X100 in 1× casein. Primary antibodies raised against γH2AX pS139 (Millipore, Cat. # 05-636, RRID:AB_309864) and 53BP1 (Novus, Cat. # NB100-904, RRID:AB_10002714) were diluted 1:500 in blocking buffer and added to the cells for overnight incubation at 4°C. Cells were then stained with AlexaFluor-conjugated secondary antibodies diluted 1:500 in blocking buffer for 2 hours. After mounting, three fields of view were imaged using a BZ-X800 Keyence microscope. Secondary antibodies used include Alexa Fluor 488 goat anti-mouse (Thermo Fisher Scientific, Cat. # A-11001, RRID:AB_2534069) and Alexa Fluor 647 goat anti-rabbit (Thermo Fisher Scientific, Cat. # A-21245, RRID:AB_2535813). γH2AX and 53BP1 foci were counted in each nucleus in each image using a previously described foci quantification pipeline (22).

Total NAD level analyses were performed in triplicate using the NAD/NADH Quantification kit (Sigma) or NAD/NADH-GloAssay Kit (Promega), as per the manufacturer’s specifications.

Genomic DNA was bisulfite-converted and analyzed for genome-wide methylation patterns using the Illumina Human EPIC Bead Array 850K platform at the Yale Center for Genome Analysis (cell line models) or as previously described (patient samples; ref. 13).

Raw microarray intensity data files were processed with the Python package methylprep to perform quantile normalization for dye-bias correction (23) followed by computation of detection P-values for each array probe via the pOOBAH method (24). Probes with a detection P-value of less than 0.05 were excluded from the analysis. The remaining probes’ intensity data were converted to methylation M-values and β values for downstream analysis.

Clustermaps were generated from β values using the clustermap function from the seaborn package in Python (25). Both rows and columns of the heatmap are clustered via the “complete” linkage method from the scipy package (26). Chromosomal position plots were generated using the heatmap function from the seaborn package, with chromosomal position of probes provided by Illumina array manifest. Gene ontology plots were generated from lists of top 1,500 significantly hypermethylated genes using the clusterProfiler (RRID:SCR_016884; ref. 27) and enrichplot (28) packages in R. The Human Molecular Signatures Database (MSigDB) Hallmark gene sets were used to construct lists of relevant pathways.

Twenty-four hours after plating, cells were treated with the indicated drugs. Cells were harvested at 48 or 96 hours after treatment by trypsinization and fixed dropwise in 70% ice-cold ethanol. Cells were stained in Propidium Iodide/RNase A stain (BD Biosciences). Stained cells were run on a CytoFlexLX and analyzed using FlowJo software (RRID:SCR_008520).

All statistical tests were performed using GraphPad Prism software (RRID:SCR_002798). All experiments were performed in triplicate unless otherwise noted. Tests for normality were performed using the Shapiro–Wilk test for small sample sizes (n < 50) and the Kolmogorov–Smirnov test in cases of n ≥ 50. Student two-tailed t test was used for comparing two groups. One-way ANOVA with Dunnett’s multiple comparison test was used to evaluate experiments involving one variable across multiple groups. Two-way ANOVA with Tukey’s multiple comparison test was used to evaluate two variables across multiple groups. All error bars display standard deviation. NS, not significant (P > 0.05); ^∗, P < 0.05; ^∗∗, P < 0.01; ^∗∗∗, P < 0.001; ^∗∗∗∗, P < 0.0001.

Analysis of significantly methylated genes was performed via custom software in Python. Methylation probes from the Illumina EPIC Bead Array were grouped by annotation to the UCSC gene symbol, with methylation probe annotations provided by Illumina. For each sample pair, the distribution of M-values from probes in each gene was compared via a two-sample Kolmogorov–Smirnov test. P-values were calculated against the null hypothesis of both sample groups being drawn from the same distribution. Differentially methylated genes were defined by a 15% change in β value from baseline and were considered statistically significant for tests resulting in a P-value < 0.05 after controlling for false discovery rate via the Benjamini–Hochberg method (29).

The patient methylation data analyzed in this study were obtained from Gene Expression Omnibus at GSE155207 and analyzed with permission from the authors of the original publication, Sun and colleagues (13). Cell line methylation data generated in this study is available on Gene Expression Omnibus under accession number GSE269619. All non-commercial materials are available upon request. Code Ocean Capsule: https://codeocean.com/capsule/0458306/tree/v1.

To study DNA methylation associated with FH loss, we performed DNA methylation array profiling in FH-deficient RCC cell line models. We previously established and characterized isogenic of YUNK1 parental, shRNA nontargeting control (YUNK shNT), and FH shRNA knockdown cells (YUNK1 shFH; ref. 15). Analysis revealed significantly differential methylation of 453 different genes in the shFH cells compared with shNT cells (Fig. 1A). We find 6,615, 5,372, and 6,758 differentially methylated genes (DMG) in the NCCFH1, UOK262, and UOK268 models of FH-deficient RCC, respectively, when compared with the normal kidney YUNK1 parental cells (Supplementary Fig. S2A–S2C). The larger number of DMGs in these comparisons when considered with the YUNK1 shNT vs. shFH (Fig. 1A) or YUNK1 parental versus shFH (Supplementary Fig. S2D) is notable and may be inherent to the primary tissue niches from which these cell lines were derived. GO pathways analysis of the top 1,500 hypermethylated genes in these cell lines indicates that cancer-related pathways, such as TNFα signaling, epithelial to mesenchymal transition, and hypoxia, are impacted by hypermethylation in FH-deficient cells (Supplementary Fig. S2E–S2H; Supplementary Table S2). From the full list of DMGs, we observed that average gene methylation β values at CpG islands are increased in NCCFH1, UOK262, and UOK268 models compared with the YUNK1 parental cells (Fig. 1C; Supplementary Table S3). Dysregulation of methylation at both CpG islands and the surrounding areas (shores and shelves) is observed in cancer and can impact gene expression. Increases in β values occurred at the surrounding shores in FH-deficient cells compared with YUNK1 parental (Fig. 1C; Supplementary Table S3). Notably, it seems that methylation in these genomic regions varies between cell lineages (Fig. 1C; Supplementary Table S3).

The nicotinate phosphoribosyl transferase (NAPRT) gene was found to be significantly hypermethylated across our FH-deficient isogenic cell line model comparisons (Supplementary Fig. S2I; Supplementary Table S4). We identified NAPRT from the list of common genes as it is clinically relevant via its association with nicotinamide phosphoribosyl transferase (NAMPT) inhibitors and other oncometabolite-producing cancers (30, 31). The NAPRT enzyme is essential to the Preiss–Handler pathway of NAD⁺ biosynthesis. FH-deficient RCC models cluster into two groups by methylation of probes spanning the NAPRT gene (Fig. 1C). Differentially methylated probes are associated with TSS200 and TSS1500 labels, denoting regions 200 and 1,500 base pairs upstream of the transcription start site, suggesting that differential methylation occurs in the promoter. When aligning probes with their chromosomal locations, we find that differential methylation occurs upstream of and within a CpG island present in the NAPRT promoter (Fig. 1D and E). This region is differentially methylated in other models of NAPRT-silenced cancers (30).

Cell lines with high levels of methylation in a discrete region of the promoter (chr8:144660872-chr8:144660395) also exhibited reduced NAPRT mRNA expression (Supplementary Fig. S2J). Cell lines with reduced mRNA expression and hypermethylation in the NAPRT promoter showed complete silencing of NAPRT protein expression (Fig. 1F). The UOK268 cell line does not exhibit hypermethylation across this specific region and does not silence expression. To confirm that the lack of protein expression in two of three RCC models was caused by the DNA methylation, we treated UOK262 and NCCFH1 cell lines with the hypomethylating agent, 5-aza-2-deoxycytidine. When treated with this agent consecutively for 7 days, NAPRT protein was strongly re-expressed, suggesting that DNA methylation at the promoter controls NAPRT protein expression (Fig. 1G and H). When FH was overexpressed in the FH-deficient UOK262 cell line, we observed no restoration of NAPRT expression after 14 passages (Supplementary Fig. S2K). However, previous findings indicate that oncometabolite-driven changes in DNA hypermethylation can take up to 40 passages to revert back to incomplete resolution of oncometabolite-induced epigenomic changes (9).

Analysis of methylation array data on a previously published cohort of patients with FH-deficient HLRCC also shows hypermethylation at 5,712 genes in tumor compared with normal tissue (Fig. 2A; ref. 13). NAPRT-associated probes show increased mean methylation in tumor samples (Fig. 2B) and FH-deficient tumors cluster distinctly from normal tissue by methylation of NAPRT (Fig. 2C). We observe hypermethylation at the NAPRT promoter in the same region identified in cell lines (Supplementary Fig. S3A). Previously published analysis of these samples shows similarities between methylation in FH-RCC and CIMP renal tumors (13). As FH-deficient HLRCC is rare, we identified two FH-deficient cases from the TCGA KIRP dataset showing reduced NAPRT mRNA expression (Supplementary Fig. S3B).

To determine whether NAPRT protein expression is indeed silenced in patient HLRCC tumor samples, we performed NAPRT IHC on 14 FH-deficient HLRCC leiomyomas and renal tumors (Supplementary Table S5). An anti-NAPRT antibody was developed and validated for this purpose (See Methods; Supplementary Fig. S3C). We find that 13 of 14 tumors do not express NAPRT protein (NAPRT negative = 92.8%; Fig. 2D and E; Supplementary Fig. S3D). We also stained a set of RCC PDX models, confirmed samples as FH wild-type, and detected a lack of NAPRT staining in only eight of 23 tissue microarray cores (NAPRT negative = 34.8%; Supplementary Fig. S3E–S3H; Supplementary Table S6). The proportions of NAPRT-positive and negative samples were significantly different between the FH-deficient HLRCC and FH-wild-type RCC datasets (Supplementary Fig. S3F).

Previous studies have identified an inverse correlation between NAPRT expression and nicotinamide phosphoribosyl transferase inhibitor (NAMPTi) sensitivity (32, 33). NAMPT is the rate-limiting enzyme in the NAM salvage pathway and is expressed in our cell lines (Fig. 3A; Supplementary Fig. S4A). NA is converted to nicotinic acid mononucleotide by NAPRT as a precursor to NAD⁺ in cells (34). In the body, NA is typically absorbed from diet or the gut microbiome (35). We find that FH-deficient and NAPRT-silenced YUNK1 shFH cells are sensitive to the NAMPTi, FK866, when co-treated with NA. YUNK1 parental cells, which express NAPRT, are rescued by NA (Fig. 3B). Total NAD⁺ levels in YUNK1 cells treated with FK866 are depleted as expected with NAMPTi treatment, however, NAD⁺ levels are rescued in cells that express NAPRT protein with the addition of NA (Fig. 3C). Similarly, we find that NAPRT-silenced, FH-deficient UOK262 cells and NCCFH1 cells are sensitive to FK866 even in the presence of NA (Fig. 3D and E) whereas NAPRT expressing, FH-deficient UOK268 cell viability is rescued by NA (Fig. 3F). This is consistent in experiments with other NAMPTis, GMX1778 and GNE-617 (Supplementary Fig. S4B–S4D). The increase in UOK268 cell viability is likely due to the rescue of total NAD⁺ in UOK268 cells treated with FK866 and NA versus FK866 alone, whereas UOK262 and NCCFH1 cells do not exhibit increased NAD⁺ with NA co-treatment (Fig. 3G–I).

The inverse correlation between NAPRT expression and NAMPTi sensitivity is also observed to some extent in DepMap cancer cell lines, but in assays without nicotinic acid supplementation (Supplementary Fig. S5A). Cells can also synthesize NAD⁺ through the de novo NAD⁺ synthesis pathway, however, previous reports have shown that RCC cell lines may not express QPRT, one of the enzymes in this pathway (Fig. 3A; refs. 33, 36). QPRT mRNA is significantly downregulated in the UOK268 and NCCFH1 cells and significantly upregulated in the UOK262 cells relative to YUNK1 (Supplementary Fig. S5B). In the presence of NA, these cell lines show sensitivity to NAMPTi, suggesting that QPRT levels are not a determinant for NAMPTi sensitivity in cell culture, consistent with previous reports (37). When querying DepMap for associations between NAD⁺ biosynthesis and consumption pathways (Fig. 3A; Supplementary Fig. S5C–S5H), we identified that there may be an association between QPRT mRNA expression with NAMPT and PARP1 mRNA expression (Supplementary Fig. S5F and S5G; Supplementary Table S7), which warrants further investigation into regulatory crosstalk between NAD⁺ pathways. To confirm that NAPRT status determines NAMPTi sensitivity, we overexpressed NAPRT in the UOK262 cell line (Supplementary Fig. S6A). NAPRT overexpression in the presence of NA led to increased survival in response to NAMPTi (Supplementary Fig. S6B). Together, these results suggest that NAPRT-silenced cells are sensitive to NAMPTis in the presence of NA irrespective of QPRT expression levels.

NAPRT silencing has been reported in other contexts, namely IDH1/2 mutant cancers, which produce the ocometabolite 2HG (30). 2HG induces a DNA hypermethylation phenotype as well (8). We profiled NAPRT expression and methylation in an isogenic U87 IDH1^R132H glioblastoma cell line model (Supplementary Fig. S7A). U87 IDH1^R132H cells are highly methylated at the NAPRT promoter in the same region that differential DNA methylation occurs in our other models (Supplementary Fig. S7B). mRNA expression is significantly downregulated in U87 IDH1^R132H cells (Supplementary Fig. S7C). U87 parental cells express NAPRT protein whereas IDH1^R132H cells do not (Supplementary Fig. S7D). Together, with our results in FH-deficient models, we observe a negative correlation between NAPRT methylation and mRNA expression (Supplementary Fig. S7E). With 5-aza-2-deoxycytidine treatment, U87 IDH1^R132H cells re-express the NAPRT protein (Supplementary Fig. S7F), suggesting that methylation of the NAPRT promoter controls its expression and the mechanism of silencing is generalizable across multiple contexts.

The U87 IDH1^R132H model is also sensitive to FK866 and cannot be rescued by co-treatment with NA (Supplementary Fig. S8A). Although NAD⁺ is depleted in both U87 WT and IDH1^R132H cells with FK866 treatment, there is a difference in the NAD⁺ levels at baseline and with FK866 treatment (Supplementary Fig. S8B) that may lead to viability differences when treated with FK866 alone (Supplementary Fig. S8A). We characterized NAPRT expression, NAMPT expression, and FK866 sensitivity with 10-µmol/L NA co-treatment in other models of IDH WT or IDH1/2 mutant cancers, namely, cholangiocarcinoma and chondrosarcoma models (Supplementary Fig. S8C–S8G). Notably, both IDH WT models, TFK and HUCCT1 cells, express relatively high levels of NAPRT (Supplementary Fig. S8C). The complete silencing of NAPRT expression evident in SW1353 and HT1080 cells leads to sensitivity to FK866 that cannot be rescued by NA (Supplementary Fig. S8C and S8G). The RBE IDH1 mutant model expresses some NAPRT protein and is rescued from FK866 with NA (Supplementary Fig. S8G). Variable NAPRT expression is consistent with TCGA data in IDH1 mutant glioblastoma, cholangiocarcinoma, and sarcomas (Supplementary Fig. S8H–S8J). Our data suggest that complete loss of NAPRT protein expression, which occurs in some but not all IDH1/2 mutant models, confers vulnerability to NAMPTis.

We sought to identify other targeted therapies to combine with NAMPTis against FH-deficient, NAPRT-silenced tumors. Given that NAD⁺ is depleted in NAPRT-silenced cells treated with NAMPTi, we hypothesized that these models would be extremely sensitive to the combination of PARPis and NAMPTis. In general, PARPis mainly target PARP1 and PARP2, proteins that bind to sites of DNA damage and consume NAD⁺ to form poly (ADP-ribose) chains. PAR chains help to recruit DNA damage response proteins (38) and allow for the release of PARP1 from DNA (39). Oncometabolite-producing cancer models have also been shown to be sensitive to PARPis (15, 20) as a consequence of homologous recombination repair suppression via local chromatin signaling (40). To test this combination, we treated cells with increasing concentrations of both the PARPi, olaparib, and the NAMPTi, FK866, in media with 10-µmol/L NA. Drug synergy results in greater than additive cell death of the individual drugs (41). Parental YUNK1 cells are sensitive to olaparib but do not exhibit synergy with the combination of drugs (Fig. 4A; Supplementary Fig. S9A). In contrast, we find that the combination synergizes in YUNK1 shFH cells (Fig. 4A; Supplementary Fig. S9A; ref. 41). Synergy is also observed in NAPRT-silenced UOK262 (Fig. 4B; Supplementary Fig. S9B) and NCCFH1 (Fig. 4C; Supplementary Fig. S9C) models but not in the NAPRT-expressing UOK268 model (Fig. 4D; Supplementary Fig. S9D). NAPRT overexpression in the UOK262 model and the addition of NA leads to abrogation of the observed synergy of FK866 and olaparib in the parental cells (Supplementary Fig. S6C–S6F). Synergy was also observed with the combination of FK866 and olaparib in the presence of 10-μmol/L NA in NAPRT-silenced, IDH1/2 mutant glioma and chondrosarcoma cells, but not NAPRT-expressing IDH1/2 WT cells (Supplementary Fig. S10A–S10F). These results suggest that NAPRT silencing, in multiple cancer types, is required for synergy between FK866 and olaparib.

To understand the consequences of NAMPTi and PARPi treatment, we quantified NAD⁺ levels in cells treated with concentrations of both drugs that previously exhibited synergy. FK866 significantly depleted total NAD⁺ in NAPRT-silenced NCCFH1 cells in a dose-dependent manner (Fig. 5A). We hypothesized that the reduction in NAD⁺ levels would correspond to a reduction in PAR chain formation. We observed a dose-dependent reduction in PAR chains at various molecular weights in NCCFH1 cells treated with FK866 alone (Fig. 5B). FK866 and olaparib in combination reduced PAR significantly when compared with olaparib alone, and showed minor reductions when compared with FK866 alone, though not significant (Fig. 5B). We observed a similar effect of FK866 and olaparib in NAPRT-silenced UOK262 cells in decreasing total NAD⁺ and observed non–significant reductions in PAR chains compared with single drugs, even with NA co-treatment (Fig. 5C and D). Converely, NAD⁺ was not decreased in UOK268 cells and PAR chain synthesis was rescued with NA (Fig. 5E and F). As olaparib inhibits PARPs from consuming NAD⁺, we did observe slight increases in total NAD⁺ with the combination of FK866 and olaparib, though not significant (Fig. 5A, C, and E). In NCCFH1 and UOK262 models, the remaining PAR signal with the combination of FK866 and olaparib seems to be PARP1-associated auto-PARylation (Fig. 5B and D). The activation of NAD⁺ production by NAPRT with NA also seems to abrogate any additional effect of FK866 and olaparib in NAPRT-expressing cells.

PARP is rapidly recruited to sites of DNA damage and synthesizes PAR chains that recruit other repair proteins involved in both single-strand break repair and double-strand break repair. In addition to inhibiting PARP catalytic activity to synthesize PAR chains, PARPis have also been reported to “trap” PARP proteins to DNA, which contributes to their cytotoxicity (42). As PARPis are competitive inhibitors of NAD⁺, we hypothesized that in conditions of low NAD⁺, there may be increased engagement of PARPis to PARP proteins. This, in addition to the observed low levels of PARP1-associated PARylation, may potentiate PARP retention at chromatin by olaparib. We observe an expected increase in PARP1 signal in the chromatin fraction indicating PARP retention at the chromatin in NCCFH1 and UOK268 cells when treated with MMS and BMN673, drugs used to induce single-strand breaks and to retain PARP at the chromatin, respectively (Fig. 6A–D). We also detected more PARP1 signals in the chromatin fraction in cells treated with MMS, FK866, and olaparib compared with MMS with olaparib alone in NCCFH1 cells (Fig. 6A and B). In contrast, we do not observe a significantly increased PARP1 signal in the chromatin fraction of cells treated with MMS, FK866, and olaparib in UOK268 cells, though we do see an expected increase with MMS and BMN673 treatment (Fig. 6C and D). This suggests that selective NAD⁺ depletion in NAPRT-silenced cells by NAMPTis can enhance the PARP retention at the chromatin by PARPis.

In addition to our observation that NAPRT loss can render cells sensitive to depletion of PAR chains and increased PARP retention at chromatin, NAPRT loss has also been previously reported to sensitize cells to DNA damage (43). We hypothesized that NAPRT-silenced, FH-deficient cells are sensitive to the combination of NAMPTis and PARPis because of decreased DNA repair associated with reductions in PAR chain formation. To test this, we performed immunofluorescence to detect γH2AX and 53BP1 foci (Fig. 7A; Supplementary Fig. S11A). γH2AX is an early marker for sites of DNA damage (44). 53BP1 is an essential factor in double-strand break repair and coordinates repair pathway choice (45). We quantified the proportion of “foci positive” cells, in which cells that exhibited foci count above a certain threshold (Fig. 7B–D; Supplementary Fig. S11B–S11D). NA treatment alone did not alter levels of γH2AX or 53BP1 foci (Supplementary Fig. S11E–S11G). As expected, we observed increases in the DNA damage markers in all conditions with olaparib treatment. Treatment with FK866 or olaparib alone increased the percent γH2AX foci positive NCCFH1 cells, which was further increased by the combination of inhibitors at 24 hours posttreatment (Fig. 7B). Nonsignificant increases in γH2AX foci signal were also observed in UOK262 cells with olaparib and the combination; however, FK866 did not seem to induce damage alone (Fig. 7C). In contrast, the UOK268 cells, which express NAPRT, did not exhibit increased γH2AX foci positive cells with the combination of FK866 and olaparib in the presence of NA (Fig. 7D). Though not significant, we also observed similar trends in 53BP1 foci positive cells for FK866 and olaparib-treated cells based on their NAPRT status (Supplementary Fig. S11A–S11D).

Significant increases in DNA damage were also accompanied by cell cycle arrest in G2 at 48 hours, which persisted to 96 hours in NCCFH1 cells, which are extremely sensitive to NAMPTis (Supplementary Fig. S12A–S12D). At 96 hours after treatment, we also observed slight increases in the sub-G1 population of cells, indicating the presence of dead cells (Supplementary Fig. S12C and S12D). These results suggest that unrepaired DNA damage may contribute to the observed synergy of NAMPTis and PARPis in NAPRT-silenced cells. In addition to the depletion of PAR chains and enhanced PARP retention at chromatin, the DNA damage results suggest a potential mechanism where DNA damage contributes to the synergistic relationship of NAMPTis and PARPis in NAPRT-deficient contexts (Fig. 7E).

FH-deficient RCC is a rare but highly aggressive cancer in patients with HLRCC that has an unmet need for rational, targeted therapies. Examining genes impacted by the DNA hypermethylation in both cell lines and patient samples revealed a potential avenue for biomarker-driven treatment of a subset of these tumors with NAMPTis based on NAPRT status. This study also proposes the rational combination of NAMPTis and PARPis to treat NAPRT-silenced cancers and provides data suggesting a mechanism for synergy between these two drugs.

Our methylation analysis, along with other reports (12, 13, 46), paves the way for a better understanding of the consequences of DNA methylation in FH-deficient RCC, and our analysis identifies silencing of NAPRT protein expression as a potential biomarker to identify cases for treatment with NAD⁺ depleting therapies. Although we do not observe NAPRT silencing in all FH-RCC samples, there is a higher prevalence of NAPRT silencing in FH-deficient HLRCC compared with other models of FH wild-type RCC, suggesting an association between FH loss of function and NAPRT expression. Although we do not observe NAPRT re-expression with full-length FH overexpression, previous work indicates that this epigenetic mechanism of gene silencing, also evident in IDH-mutant cells, cannot be fully reversed in at least 25% of loci, even after extended passages (9). Therefore, we do not exclude the possibility that NAPRT gene re-expression could occur over longer culture periods. Factors influencing the penetrance of NAPRT silencing in HLRCC models and tumor tissue also warrant further investigation.

Previous methylation analysis of patient samples found a high concordance between CIMP RCC and FH-deficient RCC (13), suggesting that these and other oncometabolite-producing tumors exhibit CIMP-like patterns of gene silencing (12, 47). Our GO pathway analysis identified multiple cancer-associated pathways that were significantly hypermethylated in FH-deficient RCC cell lines, namely, TNFα signaling via NFKB and epithelial–mesenchymal transition (EMT) pathways, which promote metastasis. EMT pathways were also found to be aberrantly methylated in FH-deficient patient samples (13). Interestingly, NAPRT loss has also independently been associated with EMT (48). These results suggest that dysregulation of EMT-associated genes may be associated with FH-deficient RCC, and further studies are needed to better understand the underpinnings of tumor progression and identify novel therapeutic strategies.

We have previously discovered that NAPRT silencing occurs in diffuse intrinsic pontine glioma and glioblastoma, via different epigenetic-driven mechanisms (33, 49). Our data show that NAPRT silencing can occur in some IDH1/2 mutant contexts, consistent with other reports on glioma (30, 31), but with the added complexity that silencing via methylation may not occur in all models, which has also been studied in other IDH1/2 tumors (10). Given that FH-deficient tumors share a similar methylation phenotype to other CIMP and SDH-deficient tumors, it is possible that NAPRT silencing may be a more generalized finding than previously appreciated (12, 13, 47, 50).

This work, as well as other reports (30, 32, 33, 48), suggests that NAPRT silencing can lead to a synthetic lethal interaction with NAMPTi in a variety of tumor types, warranting the broader study of a biomarker-driven approach to improving specific anti-tumor efficacy when using NAMPTis. When considering NAMPTis as a potential therapeutic for NAPRT-silenced tumors, we recognize that clinical trials of NAMPTis have historically been marred by significant hematologic toxicity, reports of retinal toxicity in preclinical models (51, 52), and modest anti-tumor benefit (53). However, these trials were conducted in solid tumors and hematologic malignancies without patient stratification by biomarker expression (53). Future preclinical work should address the in vivo efficacy of NAMPTis with NA coadministration and evaluate whether considering NAPRT expression can mitigate toxicity.

One mechanism by which PARPis exert cytotoxicity is through competitive inhibition of the NAD⁺ binding site on PARP enzymes, thereby inhibiting their catalytic activity and decreasing PARylation of target proteins. We hypothesize that this, in addition to increased retention of PARP1 at chromatin could potentiate the cytotoxicity of a PARPi, which was observed between FK866 and olaparib. Conversely, in NAPRT-proficient cells, the reduction of PAR chains and subsequent synergy of combined NAMPTi and PARPi was abrogated with NA, which is not present in most cell culture media but is present in mammalian serum and is responsible for NAD⁺ flux in normal mammalian kidney and pancreas (54). Combination NAMPTis and PARPis have previously been tested in other cancer cell line models but have lacked biomarkers to guide the use of this combination (55, 56). The synergistic combination of NAMPTis and PARPis in NAPRT-silenced cases may exhibit potent anti-tumor activity with lower drug doses, thereby limiting toxicity and further widening the therapeutic index. NAMPTis may also be combined with other DNA repair inhibitors or DNA damaging agents, such as temozolomide, which has been shown to sequester NAD⁺ in the form of PAR chains (57). Though the present data on PARPis and NAMPTis points to unrepaired DNA damage as a mechanism of action, there are likely other consequences of modulating NAD⁺ levels in combination with other drugs. This is an especially important avenue for future research in the oncometabolite-producing cancers, which demonstrate the interconnectedness of metabolism, the epigenome, and DNA repair.

In summary, this study presents a biomarker-based therapeutic strategy for the targeted treatment of NAPRT silencing in FH-deficient RCC and deepens the understanding of the interplay between NAD⁺ metabolism and DNA repair in oncometabolite-producing cancers. Our data support the notion that epigenetic dysregulation in cancer can lead to actionable vulnerabilities in areas of unmet need.

K.J. Noronha reports grants from NIH (R01CA21543-05, R.S. Bindra; NRSA F31, K.J. Noronha) and Department of Defense (DOD; W81XWH-22-1-0549, B. Shuch and R.S. Bindra) during the conduct of the study and personal fees from AtlasXomics, Inc., outside the submitted work. K.N. Lucas reports grants from NIH (R01CA21543-05, R.S. Bindra) and DOD (W81XWH-22-1-0549, B. Shuch and R.S. Bindra) during the conduct of the study and other support from York Analytical Laboratories outside the submitted work. S. Friedman reports grants from NIH (R01CA21543-05, R.S. Bindra) and DOD (W81XWH-22-1-0549, B. Shuch and R.S. Bindra) during the conduct of the study. M.A. Murray reports grants from NIH NRSA F31 outside the submitted work. S. Liu reports grants from Yale College First-Year Research Fellowship during the conduct of the study. J. Spurrier reports employment with Alphina Therapeutics. M. Raponi reports a patent for Anti-NAPRT Antibodies and Methods of Use pending to Alphina Therapeutics. J.C. Vasquez reports grants from NIH K08, Robert Wood Johnson Harold Amos Medical Faculty Development Program, Doris Duke Charitable Foundation, and American Cancer Society Institutional Research Grant during the conduct of the study. No disclosures were reported by the other authors.

K.J. Noronha: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. K.N. Lucas: Data curation, formal analysis, validation, investigation, visualization, writing–original draft, writing–review and editing. S. Paradkar: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. J. Edmonds: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. S. Friedman: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. M.A. Murray: Conceptualization, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. S. Liu: Resources, data curation, formal analysis, visualization, writing–original draft, writing–review and editing. D.P. Sajed: Resources, data curation, investigation, methodology. C. Sachs: Resources, investigation, methodology, project administration. J. Spurrier: Resources, validation, investigation, project administration. M. Raponi: Conceptualization, resources, investigation, writing–review and editing. J. Liang: Conceptualization, resources, data curation, methodology, project administration. H. Zeng: Conceptualization, resources, supervision, funding acquisition, project administration. R.K. Sundaram: Resources, supervision, investigation, methodology, project administration. B. Shuch: Conceptualization, resources, supervision, investigation, writing–review and editing. J.C. Vasquez: Conceptualization, resources, supervision, investigation, project administration, writing–review and editing. R.S. Bindra: Conceptualization, resources, supervision, funding acquisition, investigation, methodology, project administration, writing–review and editing.

The authors would like to thank Dr. Collin Heer for his expertise in NAD⁺ metabolism to assist in the conceptualization of the study. This work was supported by the NIH (R01CA21543-05; R.S. Bindra) and DOD (W81XWH-22-1-0549; B. Shuch and R.S. Bindra). K.J. Noronha was supported by F31CA260794. M.A. Murray was supported by F31CA261129. S. Liu was supported by the Yale College First-Year Research Fellowship. J.C. Vasquez is supported in part by the NIH/NCI K08 Career Development award #1-K08 CA258796-01, the Robert Wood Johnson Harold Amos Medical Faculty Development Program, the Fund to Retain Clinical Scientists at Yale, sponsored by the Doris Duke Charitable Foundation award #2015216 and the Yale Center for Clinical Investigation, and by an American Cancer Society Institutional Research Grant, #IRG-21-132-60-IRG. We thank the Yale Center for Genome Analysis and the Yale Keck Biotechnology Core for their assistance. The results shown here are in part based on data from the Cancer Dependency Map (https://depmap.org) and TCGA Research Network (https://www.cancer.gov/tcga).