A non-enzymatic role of Nudix hydrolase 5 in repressing purine de novo synthesis

Abstract

Folate metabolism is intricately linked to purine de novo synthesis through the incorporation of folate-derived one-carbon units into the purine scaffold. By investigating chemical and genetic dependencies caused by mutations in methylenetetrahydrofolate dehydrogenase, cyclohydrolase and formyltetrahydrofolate synthetase 1 (MTHFD1), we discovered a key role for Nudix hydrolase 5 (NUDT5) in regulating purine de novo synthesis. Genetic depletion and selective chemical degradation showed that a scaffolding role, rather than NUDT5 enzymatic activity, was causing this phenotype. NUDT5 interacted with phosphoribosyl pyrophosphate amidotransferase (PPAT), the rate-limiting enzyme of purine de novo synthesis, to repress the pathway in response to increased purine abundance. Through this mechanism, loss of NUDT5 mediates resistance to purine analogs in cancer treatment and prevents adenosine toxicity in MTHFD1 deficiency.

Folate metabolism is essential for providing one-carbon units for the biosynthesis of several metabolites, that include purines, thymidylate, and methionine (1). Consequently, impairment of folate metabolism through genetic mutations or dietary folate deficiency causes varied pathologies including developmental defects and an increased risk of cancer (2–6). Therapeutically, the dependence of proliferating cells on this pathway is exploited by the use of antifolates such as methotrexate and pemetrexed in cancer therapy (7–9).

The folate pathway is compartmentalized between cytoplasm and mitochondria, with additional roles of selected enzymes in the nucleus (10, 11). The usual direction of the folate catalytic cycle of mitochondrial formate production and cytosolic formate utilization can be reversed after mutation of key folate enzymes (12). Under these conditions, or after pharmacological inhibition of folate metabolism, destabilization of the folate scaffold might further contribute to the overall phenotype (13, 14). Folate deficiency can also arise from disruption of downstream pathways by “trapping” of all cellular folates as certain species, e.g., as 5-methyltetrahydrofolate (5-MeTHF) in vitamin B12 deficiency (15, 16) and as 10-formyltetrahydrofolate (10-CHO-THF) after pharmacological inhibition of MTHFD1 (17).

MTHFD1 is a trifunctional enzyme that catalyzes the cytoplasmic interconversion of 10-CHO-THF, methenyltetrahydrofolate (CH⁺-THF), and methylenetetrahydrofolate (CH₂-THF) by its formyltetrahydrofolate synthetase (S) and its combined methylenetetrahydrofolate dehydrogenase and cyclohydrolase (DC) domains (Fig. 1A). Of the MTHFD1 reaction products, CH₂-THF provides one-carbon units for the synthesis of methionine and thymidylate, whereas 10-CHO-THF delivers two of the carbons to the purine scaffold generated in the de novo synthesis pathway. This process might be enhanced by direct interaction of MTHFD1 with the purinosome (18). Genetic depletion of MTHFD1 in yeast and human cells results in purine auxotrophy, presumably by preventing purine de novo synthesis (19–24). In contrast, in cells from patients with MTHFD1 deficiency, purine de novo synthesis is unaffected, but thymidylate and methionine biosynthesis are impaired due to point mutations in the DC domain of the enzyme (25). To date, no comprehensive studies of the metabolic, pharmacologic, and genetic dependencies caused by loss of distinct MTHFD1 enzymatic activities have been reported. To address this gap in our current understanding, we examined distinct features of MTHFD1 domain activities that control the balance between cellular adenosine dependency and toxicity. We found that the antiproliferative effect of adenosine and other purine analogs in this context was dependent on a scaffolding function of NUDT5, revealing a role of this NUDIX family member in de novo purine synthesis.

Fig. 1. MTHFD1 enzymatic functions influence adenosine dependency or toxicity.

(A) Schematic overview of the folate cycle and cellular purine synthesis pathway. Corresponding folate intermediates are generated by the enzymatic functions of the MTHFD1 synthetase (S), cyclohydrolase (C) and dehydrogenase (D) domains. 10-CHO-THF is used for the synthesis of inosine monophosphate (IMP) through the purine de novo synthesis pathway. Alternatively, IMP can be synthesized through the salvage pathway in the presence of exogenous adenosine (Ado). The schematic domain structure of MTHFD1 is shown below with sites of K56R and K386E mutations that ablate the enzymatic reactions of the respective domain. (B) Normalized cell counts of WT, MTHFD1^KO, MTHFD1^K386E, and MTHFD1^K56R cells grown for 72 h in media containing dialyzed serum (DIA) or dialyzed serum supplemented with 50 µM adenosine (ADO). Values are normalized to the respective clone grown in media containing full serum (FULL) (n = 3 biological replicates, mean ± SD, *P < 0.05). (C) Normalized cell counts of WT and MTHFD1^K56R cells at indicated concentrations of adenosine. Cells were cultured in DIA supplemented with respective adenosine concentrations for 72 h. (n = 2 biological replicates, mean ± SD). (D) Normalized cell counts of MTHFD1^K56R cells cultured at indicated media conditions and with supplemented nucleotides and precursors (50 µM) (n = 3 biological replicates, mean ± SD, ****P < 0.0001). Guo – guanosine, Cyd – cytidine, Urd – uridine, Thy – thymidine, AICAR - 5-aminoimidazole-4-carboxamide ribonucleotide, dCTP – deoxycytidine triphosphate, dGTP – deoxyguanosine triphosphate, dATP – deoxyadenosine triphosphate, dUTP – deoxyuridine triphosphate, dTTP – deoxythymidine triphosphate. (E) Representative cell cycle plots for WT and MTHFD1 mutant HAP1 cells in DIA and ADE conditions. (F) Representative images of γH2A.X and RPA2 staining of MTHFD1^K56R cells with adenosine supplementation. (scale bar is 10 µm) (G) Growth assays of fibroblasts derived from MTHFD1 deficiency patients in FULL, DIA and ADE (n = 3 biological replicates, mean ± SD, *P < 0.05). (H) Schematic overview of the genetic folate trap model. (B), (D), (G) - comparison of groups carried out by two-way ANOVA.

Inactivation of distinct MTHFD1 enzymatic activities causes opposing cellular responses to adenosine

We engineered wild-type HAP1 cells (WT) and derived clonal cell lines that either contained a deletion in the MTHFD1 gene (MTHFD1^KO) (11), or were reconstituted with either the wild-type allele (MTHFD1^recon) or catalytically inactive point mutations (24, 26) of the S (MTHFD1^K386E) or DC (MTHFD1^K56R) domains (Fig. 1A). We tested proliferation of these cell lines in medium supplemented with either standard serum or with dialyzed serum depleted of metabolites smaller than 10 kDa. MTHFD1^KO and MTHFD1^K386E cells were sensitive to serum dialysis, whereas the growth of WT, MTHFD1^recon and MTHFD1^K56R cells was unaffected in these conditions (Fig. 1B and fig. S1). To identify compounds that counteracted the detrimental effects of dialyzed serum on the growth of MTHFD1^KO cells, we performed a chemical screen of more than 90,000 structurally diverse small molecules (fig. S1). We realized that mainly adenine-containing compounds, e.g., nicotinamide adenine dinucleotides NADH and NAD⁺, flavin adenine dinucleotide (FAD), S-adenosylmethionine (SAM), adenosine monophosphate (AMP), or the AMP precursor hypoxanthine (Hyp), but not other purine or folate analogues rescued the growth of MTHFD1^KO cells (fig. S1). Similarly, adenosine addition reestablished proliferation of S-domain inactive MTHFD1^K386E cells. However, the addition of adenosine (Ado) to MTHFD1^K56R cells harboring the inactivating mutation in the DC-domain had the opposite phenotype, as in that context addition of adenosine caused an antiproliferative response (Fig. 1B). These effects occurred in a concentration-dependent manner, and MTHFD1^K56R cells became sensitized to adenosine at low micromolar concentrations (IC₅₀ ~6 µM, Fig. 1C). Also, other purines, including inosine, hypoxanthine, and AMP, and to a lower extent guanine, guanosine, and GMP, but not pyrimidines, impaired cell viability (fig. S2A). Moreover, the combinatorial growth deficiencies were not specific to HAP1 cells, as we could phenocopy these effects in other cancer cell lines when cells were grown in adenosine-containing medium together with the MTHFD1 DC-domain inhibitor LY345899 (26) (fig. S2B). Supplementation of thymidine, deoxythymidine triphosphate (dTTP) or 5-aminoimidazole-4-carboxamide ribonucleotide (AICAR) largely reversed these toxic effects in MTHFD1^K56R cells (Fig. 1D).

The addition of adenosine caused strong cell cycle changes. In MTHFD1^KO and MTHFD1^K386E cells, adenosine alleviated the G1-S phase arrest caused by dialyzed serum conditions. In contrast, MTHFD1^K56R cells arrested in G1-S phase after adenosine addition (Fig. 1E, fig. S3A). Increases in foci formation for replication protein A2 (RPA2) and phosphorylated histone variant H2AX (γH2A.X) in these conditions (Fig. 1F, fig. S3B) indicated that replication stress and DNA damage caused this phenotype. These effects occurred at adenosine concentrations orders of magnitude lower than the millimolar concentrations normally needed to induce replication stress through nucleotide imbalance (27). Again, this phenotype was not specific to HAP1 cells, as we observed similar DNA damage effects in fibroblasts from MTHFD1-deficient patients with mutations in the protein’s DC domain (Fig. 1G, fig. S4).

Overall, by engineering a specific mutation that inactivates the DC-domain of MTHFD1, we have established a genetic “folate trapping” model (17). Normally, 10-CHO-THF can be converted to THF by either the folate cycle or by purine de novo synthesis (Fig. 1 and fig. S5A). However, concomitant inhibition of both pathways, i.e., inhibition of the folate cycle through MTHFD1 DC-domain mutation (as also occurring in MTHFD1 deficiency patients), and repression of purine de novo synthesis by adenosine addition, results in the accumulation of cellular folates as 10-CHO-THF (Fig. 1H, fig. S5B). This presumably prevents regeneration of THF (fig. S5C) and thus maintenance of balanced folate and nucleotide abundances, leading to DNA damage and ultimately cell death. We expected that this phenotype might be rescued by genetic and chemical modulators of the folate and purine de novo synthesis pathways.

A genome-wide genetic screen identifies NUDT5 as a modulator of MTHFD1-mediated adenosine response

We performed a genome-wide genetic loss-of-function screen to probe genetic dependencies of the folate-trap conditions. For this, we transduced MTHFD1^K56R cells with a CRISPR/Cas9 knockout library targeting 19,114 human genes with an average of four sgRNAs per gene (28). After selection and growth with adenosine (which is toxic in this context) for two weeks, we analyzed sgRNA abundance as a proxy for beneficial growth effects after genetic depletion of the respective target gene (Fig. 2A).

Fig. 2. A genome-wide knockout screen identifies folate trap modulators.

(A) Schematic overview of the genetic screening strategy. (B) Results of the genome-wide KO screen as schematically shown in (A). Genes are ranked by fold change (FC) with the most significantly enriched hits highlighted in red: HPRT1, NUDT5 and MTHFD1. (C) Model of folate trap rescue when MTHFD1 is depleted in MTHFD1^K56R cells. (D) Normalized viability of MTHFD1^K56RHPRT1^KO and MTHFD1^K56RNUDT5^KO cells cultured in DIA or ADO for 72 h. (n = 3, mean ± SD, two-way ANOVA, ****P < 0.0001). (E) Representative images of γH2A.X and RPA2 staining of MTHFD1^K56R, MTHFD1^K56RHPRT1^KO and MTHFD1^K56RNUDT5^KO cells after being cultured in DIA or ADO for 72 h. (scale bar is 10 µm) (F) Quantification of spots per nucleus from panel (E) (spots per nucleus values are mean ± SEM from six imaged wells).

Individual depletion of each of three genes rescued adenosine-mediated toxicity: MTHFD1, HPRT1 (hypoxanthine phosphoribosyl transferase 1), and NUDT5 (Fig. 2B). MTHFD1 served as a positive control, as a shift from the MTHFD1^K56R mutation to the full knockout was expected to correlate with the transition from adenosine toxicity to adenosine dependency (Fig. 1B and 2C). HPRT1 is a key enzyme in the purine salvage pathway (fig. S5A) (29, 30). We validated the absence of purine-mediated toxicities in MTHFD1^K56RHPRT1^KO double-mutant cells and showed that adenosine did not cause DNA damage or cell cycle changes in these cells (Fig. 2D to F, fig. S6A and B).

For NUDT5, we confirmed that its depletion in the MTHFD1^K56R background (MTHFD1^K56RNUDT5^KO) reestablished cellular proliferation in purine-rich medium and prevented accumulation of aberrant cell cycle and DNA damage signals (Fig. 2D to F, fig. S6A and B). NUDT5 is a hydrolase that functions in the catabolism of ADP-ribose and the synthesis of nuclear adenosine triphosphate (ATP) (31–35). However, to date, a direct association with purine sensing or folate metabolism has not been established (fig. S6C).

NUDT5 is required for the repression of de novo purine biosynthesis in response to adenosine

To assess whether depletion of HPRT1 or NUDT5 impacted adenosine uptake or utilization, we performed isotope tracing metabolomics. We treated WT cells with isotope labeled adenosine containing five heavy ¹⁵N atoms. The ¹⁵N label was incorporated in a wide set of nucleotides, including inosine monophosphate (IMP), AMP, and guanosine monophosphate (GMP) (Fig. 3A). In contrast, labeling was almost absent in HPRT1^KO cells, corroborating the enzyme’s role in nucleotide salvage. The importance of HPRT1 for adenosine salvage in our cell model was further highlighted by AMP containing four heavy nitrogens being the predominant cellular species in WT cells, indicating adenosine processing mainly through inosine and hypoxanthine in HAP1 cells. In contrast, direct conversion of adenosine through adenine phosphoribosyltransferase (APRT) or adenosine kinase (ADK) would have yielded AMP containing five heavy nitrogens. These data indicate that the functional use of exogenous adenosine in intracellular biochemical pathways is required to trigger the folate trap, consistent with the model that cellular modulation of the purine salvage pathway may repress purine de novo synthesis activity (fig. S5 and 6).

Fig. 3. Loss of NUDT5 causes increased de novo purine synthesis in the presence of salvageable adenosine.

(A) Quantification of fractional abundance of heavy isotope labeled metabolites in WT, NUDT5^KO, and HPRT1^KO cells supplemented with 50 µM ¹⁵N₅-labeled adenosine (ADO) for 24 h (n = 2, mean ± SD). (B) Schematic overview of the purine de novo synthesis and salvage pathways indicating labeled precursors used in isotope tracing experiments. WT, NUDT5^KO, MTHFD1^K56R and MTHFD1^K56RNUDT5^KO cells were grown in DIA with 1 mM labeled formate or with 1 mM labeled formate and 50 µM labeled adenosine (ADO) for 24 h. (C) Peak areas of mass spectrometry signals for metabolites AICAR, IMP, AMP and GMP, labeled by detected isotope indicating metabolites derived from purine de novo synthesis in shades of red and those derived from salvaged adenosine in shades of blue. (D) Peak areas corresponding to dA and dG obtained from hydrolyzed DNA, colored as in panel (C). (E) Quantification of fractional abundances of most abundant purine de novo synthesis-derived isotopologs from panel (C) (mean ± SD).

In contrast to HPRT1^KO cells, NUDT5^KO cells showed ¹⁵N incorporation into IMP, AMP and GMP in amounts comparable to those in WT cells, indicating that nucleotide salvage was unaffected by NUDT5 (Fig. 3A). To assess whether NUDT5 impacted purine de novo synthesis, we expanded the isotope tracing setup to concomitantly monitor whether selected metabolites originated from cellular purine salvage or de novo synthesis activity, respectively (Fig. 3B). Specifically, we treated WT, NUDT5^KO, MTHFD1^K56R, and MTHFD1^K56RNUDT5^KO cells with isotope-labeled formate (+1 ¹³C +1 D) that is incorporated into de novo synthesized purines via 10-CHO-THF produced by the MTHFD1 synthetase domain. To assess incorporation of externally added adenosine into various metabolites, we additionally treated cells with isotope labeled adenosine (+5 ¹⁵N). Metabolome analysis by high resolution mass spectrometry enabled us to clearly discriminate metabolites derived from de novo synthesis (containing heavy carbon atoms) from those derived from adenosine salvage (containing heavy nitrogens).

For all measured metabolites (IMP, AMP, GMP), adenosine addition resulted in repression of purine de novo synthesis in both WT and MTHFD1^K56R cells (Fig. 3C). Depletion of NUDT5 prevented this repression, both in WT and in MTHFD1^K56R cells, and resulted in an increased proportion of de novo-derived AMP, GMP and IMP (Fig. 3C). The same effect was also observable in nucleotides derived from hydrolyzed DNA, in which the lower turnover of DNA resulted in a higher fraction of unlabeled nucleotides (Fig. 3D). The fractional abundances of key de novo synthesis-derived isotopologs (Fig. 3E) indicate that loss of NUDT5 – irrespective of MTHFD1 mutational status – enabled cellular purine production through the de novo pathway even when exogenous purines were abundant.

A NUDT5 scaffolding function rather than enzymatic activity is essential for modulating adenosine responses

Given that NUDT5 can be broadly linked to purine metabolism through its ADP-ribose converting activity (31, 33), we investigated whether its enzymatic activity contributes to the repression of de novo purine synthesis repression. We first performed viability assays with the chemical NUDT5 inhibitor TH5427 (36) in MTHFD1^K56R cells. Even at doses more than 1,000x higher than its biochemical IC₅₀ of 29 nM, TH5427 treatment did not rescue the viability of MTHFD1^K56R cells when cultured in high adenosine containing medium (Fig. 4A). To confirm these results with genetic NUDT5 inactivation, we performed proliferation assays in MTHFD1^K56RNUDT5^KO cells reconstituted with either NUDT5^WT or the catalytically inactive NUDT5^E112Q variant (35). Both protein variants resensitized MTHFD1^K56RNUDT5^KO cells to adenosine (Fig. 4A), corroborating our findings with the pharmacological NUDT5 inhibitor. Thus, the enzymatic function of NUDT5 appears to be dispensable for adenosine-mediated toxicity.

Fig. 4. Knockout or chemical degradation, but not enzymatic inhibition, of NUDT5 prevents adenosine-mediated toxicity.

(A) Normalized cell counts of MTHFD1^K56R cells treated for 72 h in FULL, ADO and 10 µM NUDT5 inhibitor TH5427 alone and in combination (n = 3 biological replicates, mean ± SD, two-way ANOVA, ns - not significant) and MTHFD1^K56R, MTHFD1^K56RNUDT5^KO, MTHFD1^K56RNUDT5^KO cells reconstituted with NUDT5^wt or NUDT5^E112Q (catalytically inactive) cultured in FULL or ADO (n = 2 biological replicates, mean ± SD). (B) Chemical structure of the NUDT5 targeting PROTAC dNUDT5. Representative western blot of dose-dependent dNUDT5 effects on NUDT5 protein levels after 20 h treatment in WT cells. (C) Dose-dependent effects of dNUDT5 on the growth of WT and MTHFD1^K56R cells cultured in FULL or DIA for 72 h. (D) Representative images of RPA2-foci formation of MTHFD1^K56R RPA2-RFP (intron tagged) cells at indicated medium conditions and times (RFP channel, scale bar is 25 µm).

As the genetic depletion of the entire NUDT5 protein rescued the toxic phenotype, the physical presence of its protein scaffold rather than its enzymatic activity might have dictated the adenosine-mediated toxicity. To dynamically probe a potential scaffolding role, we developed dNUDT5, a proteolysis-targeting chimera (PROTAC) that induces targeted degradation of NUDT5 (37). We confirmed that dNUDT5 induced selective NUDT5 degradation in a dose-dependent manner with a maximum degradation at 100 nM (Fig. 4B). We treated both WT and MTHFD1^K56R cells with dNUDT5 and performed viability assays in the folate trap conditions. The addition of dNUDT5 did not affect the viability of WT cells, but it prevented adenosine toxicity (Fig. 4C) and accumulation of DNA damage signals (Fig. 4D) in MTHFD1^K56R cells. We observed a dNUDT5 dose-response pattern that perfectly matched that of NUDT5 degradation. The maximum rescue of cell viability occurred at 100 nM, the same concentration that also caused maximum degradation (Fig. 4C). Addition of dNUDT5 also mitigated the toxic effects of adenosine addition in the context of patient-derived MTHFD1-deficient fibroblasts (fig. S7A). Thus, we could phenocopy the genetic ablation of NUDT5 with an acute chemical degrader rather than an inhibitor of its enzymatic activity. The combined chemical and genetic evidence indicate that the NUDT5 scaffold rather than the protein’s enzymatic activity is required to repress purine de novo synthesis when adenosine is abundant.

The direct interaction of NUDT5 with PPAT represses purine de novo synthesis

To mechanistically explore how the non-catalytic function of NUDT5 represses the purine de novo synthesis pathway, we set out to identify interaction partners of NUDT5. We performed interaction proteomics experiments (38) with NUDT5 that is N-terminally tagged with a blue fluorescent protein (BFP) and observed enrichment of three interactors: pyrroline-5-carboxylate reductase 1 and 2 (PYCR1 and 2) and phosphoribosyl pyrophosphate amidotransferase (PPAT) (Fig. 5A). The same factors were also enriched in NUDT5 proximity labeling experiments, in chemical proteomics with a NUDT5 affinity probe (fig. S8A and B), and in unbiased large-scale protein-protein interaction data sets from other cell lines (39). Of these interactors, we focused our attention on PPAT because it catalyzes the rate-limiting step of the purine de novo synthesis pathway. We confirmed that the interaction was reciprocal, as immunoprecipitation of endogenous PPAT resulted in enrichment of NUDT5, irrespective of the cellular background and mutational status of MTHFD1 (Fig. 4B and fig. S8C). Adenosine addition increased the strength of the NUDT5-PPAT interaction (Fig. 5C), and adenosine toxicity in the MTHFD1^K56R context was prevented by overexpression of PPAT (Fig. 5D). This finding underscores the relevance of PPAT repression for the phenotype and highlights the importance of PPAT to NUDT5 stoichiometric ratios for active purine de novo synthesis.

Fig. 5. Adenosine mediated toxicity is mediated by the interaction of NUDT5 with PPAT.

(A) Results of BFP-pulldown in MTHFD1^K56RNUDT5^KO cells reconstituted with BFP-NUDT5. Enriched proteins from BFP-NUDT5 pulldown were normalized to NUDT5^KO control and ranked according to log₂FC. Hits with log₂FC >4 are highlighted in blue. (B) Immunoprecipitation of endogenous PPAT following treatment of WT cells with 100 nM dNUDT5, dNUDT5nc (37) or TH5427 for 20 h. (C) HA-Pulldown of MTHFD1^K56RNUDT5^KO cells transfected with HA₂-NUDT5^wt (representative blot from two independent experiments). Cells were cultured in DIA or ADO for 3 h (in – input, – elution). (D) CellTiterGlo viability assay of MTHFD1^K56R cells that were transduced at a high MOI with lenti-EF1alpha-EGFP or lenti-EF1alpha-PPAT and treated for 2 d in ADO (n = 2, mean ± SD). (E) AlphaFold3 model of the NUDT5-PPAT complex with the interaction site depicted in detail. (F) Results of fluorescence competition assay to test the effects of NUDT5 mutations on adenosine mediated toxicity in MTHFD1^K56R cells. Cells were cultured in DIA (control) or ADO for 72 h. The population of mCherry expressing cells in ADO was normalized to cells in DIA. Dashed lines indicate NUDT5^wt and mCherry control values (values are mean ± SEM from two independent experiments) (G) Pulldown of MTHFD1^K56RNUDT5^KO cells transfected with indicated HA₂-NUDT5 variants (representative blot from two independent experiments). (H) Results of PPAT activity assay together with NUDT5^wt or NUDT5^Y74E (n = 3, mean ± SEM, values are normalized to baseline control). (I) Depiction of the AlphaFold3 NUDT5-PPAT interaction surface highlighting NUDT5 amino acids that maintain PPAT binding when mutated (blue) or that result in lost PPAT binding capacity when mutated (magenta). (J) Proposed model of the NUDT5-PPAT dependent purine de novo synthesis pathway regulation at high and low purine conditions.

To gain insights into the molecular details of the NUDT5-PPAT interaction, we performed structural analyses with AlphaFold3 (40). These results predict a ‘sandwich’ binding mode in which two NUDT5 dimers cross-interact with a PPAT tetramer, thereby potentially locking it in an inactive conformation (Fig. 5E). Gel electrophoresis after protein crosslinking (fig. S9A) and native protein gel electrophoresis (fig. S9B) supported a role of NUDT5 in regulating PPAT oligomerization, a process associated with reduced enzymatic activity (41, 42).

From the AlphaFold model, we identified NUDT5^70-75 as the putative PPAT interaction loop, within which NUDT5^Y74 points towards the PPAT interface and might function as a key residue critical for the interaction (Fig 5E). We confirmed that mutation of residue Y74 did not affect NUDT5 catalytic activity (fig. S9C). To test the functional importance of these residues, we systematically reconstituted MTHFD1^K56RNUDT5^KO cells with mCherry-P2A-NUDT5 constructs that result in the expression of NUDT5 carrying selected point mutations and an mCherry reporter protein with a self-cleaving P2A sequence. We then performed comparative fluorescent competition assays under folate trap and control conditions (Fig. 5F and fig. S9D).

Similar to the reintroduction of NUDT5^WT, inactivating the enzymatic activity of NUDT5 either by abolishing recognition of the adenosine moiety (W28A and W28A/W46A) or catalysis (E112Q) (35) did not rescue adenosine toxicity (Fig. 5F). These data corroborate our previous findings that the NUDT5 scaffold, rather than its enzymatic activity, is the key regulator of the phenotype. Also, altering the NUDT5 posttranslational modification and dimerization states (31) in NUDT5^T45A and NUDT5^T45D mutants as well as introduction of a NUDT5^Y74F mutation did not rescue adenosine toxicity (Fig. 5F and fig. S9D and E). However, MTHFD1^K56R cells reconstituted with the NUDT5^Y74E variant were desensitized to the toxic adenosine conditions and behaved like NUDT5^KO cells (Fig. 5F and fig. S9D and E). Notably, in contrast to NUDT5^WT and NUDT5^Y74F, this mutation resulted in loss of the interaction with PPAT, indicating that the rescue of adenosine-mediated toxicity may act through loss of PPAT binding (Fig. 5G and fig. S9F and G). In biochemical assays with recombinant proteins, addition of NUDT5^WT impaired PPAT activity (Fig. 5H), whereas addition of NUDT5^Y74E caused only minor effects in PPAT activity. To analyze whether naturally occurring NUDT5 mutations exist that impair PPAT binding, we queried the COSMIC database (43). We selected reported NUDT5 patient mutations surrounding the PPAT interaction loop and tested their influence on PPAT binding (Fig. 5I). Conducting assays analogous to the ones described above, we observed loss of binding for all selected mutations which either are directly facing towards the PPAT interface (C76Y, G150V) or had a destabilizing effect on NUDT5 (P66L), whereas the remaining mutations (P158I, P158T, K175M) still retained some residual PPAT binding (Fig. 5G and I). The binding strength correlated with the degree of rescue of adenosine-mediated toxicity in MTHFD1^K56R cells (Fig. 5F and fig. S9E).

The NUDT5-PPAT interaction controls sensitivity to purine analog cancer drugs

Antimetabolites such as purine analogs inhibit key metabolic pathways and thus are widely used in cancer treatment (44). Genetic depletion of HPRT1 and NUDT5 has previously been reported to confer resistance towards the clinically approved anticancer drug 6-thioguanine (6-TG) (28). To test whether this resistance also occurred in vivo, we established a mouse xenograft model and observed that NUDT5 depletion prevented the growth inhibitory effects of 6-TG on A549 lung carcinoma cells (fig. S10).

Testing a wide panel of nucleotide analogs, in addition to 6-TG, we found that NUDT5^KO cells were also resistant to various other purine analogs including 6-thioguanosine, 6-thio-2’-deoxyguanosine, 6-mercaptopurine, fludarabine, and cladribine (Fig. 6A). While loss of HPRT1 may prevent the conversion of 6-TG to toxic downstream analogs through the purine salvage pathway, rescue of toxicity by NUDT5^KO has likewise been attributed to the loss of its enzymatic function and ribose-5-phosphate synthesis, which is an important building block for purine synthesis (28). However, we found that, analogous to adenosine toxicity in the folate trap condition, 6-TG toxicity was similarly dependent on the presence of the NUDT5 scaffold rather than its enzymatic activity. Indeed, NUDT5^Y74E cells showed comparable resistance patterns to NUDT5^KO cells (Fig. 6B and (37)). Also, HPRT1^KO cells exhibited nearly identical resistance patterns, with the notable exception of cladribine and fludarabine to which HPRT1^KO did not confer resistance (fig. S11A).

Fig. 6. NUDT5-PPAT interaction controls sensitivity to purine analog cancer drugs.

(A) Profiling of 422 nucleotide analogs at 3 concentrations (25, 2.5, 0.5 μM) for differential viability effects after 72 h treatment of WT and NUDT5^KO cells. (B) Same as (A), comparing WT and NUDT5^Y74E cells. (C-D) Dose–response curves in WT, NUDT5^KO and HPRT1^KO cells treated for 72 h with azathioprine (C) and 6-mercaptopurine (D). (E) Dose–response curves of WT, NUDT5^KO, HPRT1^KO, NUDT5^KO cells reconstituted with different mutant NUDT5 versions to 72 h 6-thioguanine treatment. (F) Dose response of HEK293T cells transiently transfected with overexpression constructs for GFP or PPAT treated for 48 h with 6-thioguanine (OE – overexpression). (G) 6-Thioguanine dose response in parental, NUDT5^KO and HPRT1^KO A549 cells. C-E, G - n = 3, F - n = 2, mean ± SEM, values are normalized to control condition.

We validated these differences in drug responses in dose response experiments (Fig. 6C to E and S11B and C). The COSMIC-derived NUDT5 cancer mutations yielded 6-TG resistance patterns (Fig. 6E), that correlated with the extent of PPAT binding (Fig. 5G). Confirming the PPAT dependence of these effects, PPAT overexpression also conferred resistance to 6-TG (Fig. 6F). We extended our studies to other cell line models (Fig. 6F and G, S11B and C). Although some compounds showed context-specific effects, the NUDT5 dependence of thiopurine response was conserved in all models tested (Fig. 6A to G, S11B and C).

Discussion

By initially characterizing the pharmacological and genetic dependencies caused by MTHFD1 mutations, we established that MTHFD1^K56R cells constitute a genetic folate trap model in which subphysiological concentrations of adenosine cause toxicity. This model allowed us to systematically identify modulators of the purine de novo synthesis pathways, leading to the discovery of a regulatory role for the NUDIX hydrolase NUDT5 in this context.

We found that a scaffolding rather than an enzymatic role of NUDT5 was essential for repressing purine de novo synthesis. Proteomic profiling revealed that the regulation of this pathway occurred by a direct NUDT5-PPAT interaction, with AlphaFold modeling predicting a tight interaction between two NUDT5 dimers and one PPAT tetramer. Our attempts to experimentally obtain a co-structure of these interacting proteins have been unsuccessful. So far, no structure for human PPAT has been reported, and oxidation sensitivity of PPAT as an iron-sulfur cluster protein (45) may contribute to the challenges in structurally resolving the complex. To validate the AlphaFold model, we therefore used mutagenesis studies and discovered that single point mutations in the predicted interaction surface were sufficient to prevent NUDT5 binding to PPAT.

We and others (46) observed experimentally that NUDT5 stabilizes PPAT oligomers. That stabilization of such higher order complexes presumably inhibits PPAT enzymatic function is consistent with reports that found reduced catalytic turnover in PPAT multimers (41, 42). Indeed, we observed that wildtype NUDT5 but not a PPAT-binding deficient mutant inhibited PPAT activity in a biochemical assay. In addition to direct PPAT inhibition, NUDT5-containing oligomers may also prevent channeling of product to the next enzyme, phosphoribosylglycinamide synthetase (GART), in purine de novo synthesis or interfere with purinosome formation. Accordingly, NUDT5 may be required for adenosine-mediated purinosome disassembly (47). The NUDT5-PPAT interaction appears to be controlled by cellular metabolites, and high millimolar concentrations of PRPP may disrupt the interaction (46). Availability of salvageable adenosine increased the binding of NUDT5 to PPAT. NUDT5^Y74E, but not NUDT5^Y74F, abolished PPAT binding, indicating that the regulatory switch might occur through allosteric rearrangements or posttranslational modification of this residue (48).

Our findings led us to reevaluate the current model of de novo purine synthesis regulation. PPAT is proposed to be regulated through feedback inhibition of AMP and GMP at high µM to mM concentrations (42). However, in our reporter assay conditions, MTHFD1^K56R cells were already sensitized to adenosine concentrations in the low µM range, indicating that the cellular regulation of purine de novo biosynthesis is several orders of magnitude more sensitive than what biochemical data showed for PPAT alone (Fig 5J).

Our findings may have translational applications. Consistent with the presence of MTHFD1 CD-domain mutations, we found adenosine to exert toxicity in cells from MTHFD1 deficiency patients. Such adenosine-dependent effects may contribute to the observed immunodeficiencies. In cancer, NUDT5 modulates the toxicity of the approved drug 6-TG (28, 49). We found that the rescue mechanism extended to other purine analogs and occurred through loss of PPAT binding rather than loss of the catalytic activity of NUDT5 itself. Thus, modulation of purine de novo synthesis may influence the efficiency of purine analogs in anticancer therapy. This may have clinical implications as drug resistance may evolve through loss of HPRT1 enzymatic function, emerging PPAT binding-deficient NUDT5 mutations, or both.

Materials and Methods

Cell culture

HAP1 (KBM7-derived, Horizon Discovery, C859) cells and genetically modified variants (KO and lentiviral transduced cell lines) were cultured in Iscove’s Modified Dulbecco’s Medium (IMDM, Sigma), supplemented with 10% Fetal Bovine Serum (FBS, Sigma) and 1% Penicillin/Streptomycin (P/S, Sigma). MTHFD1^KO cells have been previously characterized (11). Unless otherwise noted, patient-derived MTHFD1-deficient fibroblast and control fibroblast lines were cultured in Minimum Essential Medium alpha media (MEMα, no nucleosides, Gibco), supplemented with 10% (v/v) FBS and 1% (v/v) P/S. These patient-derived fibroblast lines were previously characterized (50). Dialyzed FBS (Gibco) was added to IMDM with 1% (v/v) P/S at a final concentration of 10% (v/v) for indicated experiments (DIA). A549 (ATCC) and HeLa (ATCC) cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM, Sigma) supplemented with 10% FBS (Sigma) and 1% P/S (Sigma). HEK293T (ATCC) cells were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM, Sigma) supplemented with 10% FBS (Sigma), 1% P/S (Sigma) and sodium pyruvate (1 mM final). All cell lines and corresponding cellular assays were incubated at 37°C with 5% CO₂ in a humidified atmosphere. Cells were regularly tested for mycoplasma contamination.

Cloning and transfection

Cloning of target protein constructs for cellular assays

Human MTHFD1 and PPAT cDNA sequences were ordered from Genscript. NUDT5 cDNA sequence was obtained through RT-PCR (NEB, E3006) from isolated HAP1 RNA (Qiagen RNeasy kit).

MTHDF1, NUDT5 and PPAT were subcloned into a lentiviral vector (backbone Addgene #17448 for MTHFD1 and NUDT5, a gift from Eric Campeau & Paul Kaufman, backbone Addgene #216126 for PPAT) using Gibson assembly according to manufacturer’s protocol. Point mutations were introduced by Q5 site-directed mutagenesis kit and site-directed ligation independent mutagenesis (51). MTHFD1^K56R and MTHFD1^K386E were additionally tagged with GFP and NUDT5 with a BFP- or HA₂-tag at the N-terminus. For the genome-wide knock-out screen, the MTHFD1^K56R sequence was cloned into the lentiviral backbone with neomycin resistance (Addgene #17447, a gift from Eric Campeau and Paul Kaufman).

Cloning of intron tagging plasmids

The generic sgRNA targeting plasmid and the mCherry-donor plasmid were generated as previously described (52). Briefly, the pX330 plasmid expression Cas9 and the generic sgRNA targeting the donor plasmid was generated by digesting pU6-(BbsI) CBh-Cas9-T2A-mCherry (Addgene #64324, a gift from Ralf Kuehn) with BbsI followed by ligation with an annealed oligo duplex. mCherry was replaced with a blasticidin resistance (BSD) using Gibson Assembly. The mCherry-donor plasmid containing the coding sequence of mCherry flanked by generic sgRNA targeting sites, splice acceptor and slice donor sites and 20 amino acid linkers was assembled from 4 fragments using Gibson assembly. The DNA fragment with a 25 nucleotide overlap to the pUC19 vector and 32 nucleotides overlap to the N-terminus of mCherry was generated from overlapping oligos and is comprised of a generic sgRNA targeting site that is not present in the human genome followed by a splice acceptor site and a flexible 20 amino acid glycine-serine linker. This fragment is followed by a fragment with the coding sequence of mCherry without a start or stop codon that was generated by PCR. The third fragment has a 27 nucleotide overlap to the C-terminus of mCherry and a 25 nucleotide overlap to the pUC19 vector and was generated from overlapping oligos (Sigma) and comprises a flexible 20 amino acid glycine-serine linker followed by a splice donor site and the generic sgRNA targeting site. The pUC19 vector was linearized by PCR for Gibson Assembly (NEBuilder HiFi DNA Assembly) with the other three fragments. The following RPA2 intron 4 targeting sgRNA sequence was used (5’ to 3’): GTCCCTGCCATCAAGCAGGG.

Cloning of plasmids for KO cell line generation

sgRNAs were designed with the online resource tool provided by Benchling (https://benchling.com, 2023) and purchased as short single-stranded DNA oligos with BbsI sticky ends. Oligonucleotides were phosphorylated, annealed and cloned into BBsI digested pX330 (mCherry selection, Addgene #64324, a gift from Ralf Kuehn) or pX459 (puromycin selection, Addgene #62988, a gift from Feng Zhang) backbone according to published procedures (https://portals.broadinstitute.org/gpp/public/). The following sgRNA sequences were used (5’ to 3’): NUDT5 (exon 4): AATCAGTGAAACGTACAACC, HPRT1 (exon 3): AGCCCCCCTTGAGCACACAG.

Cloning of NUDT5 and EGFP TurboID fusion constructs

Cloning of the fusion constructs were done by Gateway cloning. Briefly, human NUDT5 (NCBI reference NP_054861, residues 1-208) coding DNA sequence (CDS) without stop codon was cloned out from the previously described pNLF-C [CMV/Hygro] NUDT5 vector (53). AttB1 and attB2 sites were added on both ends of the CDS of NUDT5 using primers attB1-NUDT5-F: 5’ GGGGACAAGTTTGTACAAAAAAGCAGGCTCTatggagagccaagaaccaac 3’ and attB2-NUDT5-R: 5’ GGGGACCACTTTGTACAAGAAAGCTGGGTGaaatttcaagaagggcactt 3’. This was used to form the NUDT5 entry clone via BP reaction between the donor vector pDONR221 and the NUDT5 CDS fragment. The C-terminal TurboID fusion of NUDT5 was formed via LR reaction between NUDT5 entry clone and the destination C-terminal TurboID vector.

N-terminal TurboID-EGFP served as control and was created by BP reaction between pDEST-pcDNA5-BirA-FLAG-GFP plasmid (plasmid #V8383; Gingras Lab) (54) and pDONR221 to form the entry clone. Subsequently, TurboID fusion vector was created via LR reaction between the EGFP entry clone and destination N-terminal TurboID vector.

Cloning of NUDT5 for NUDT5 enzyme assay

NUDT5 Y74E, Y74F and E112Q single and double mutants were generated through site-directed mutagenesis of the previously reported NUDT5 wild-type (NCBI reference NP_054861) expression vector (54).

Cloning of NUDT5 and PPAT for protein expression and PPAT activity assay

Codon optimized genes of hPPAT, NUDT5^wt and NUDT5^Y74E were purchased as DNA Strings (Twist Bioscience). PPAT was cloned into the pFastBac1 vector (Thermo Fisher Scientific) and NUDT5 or NUDT5^Y74E was cloned into pBAD vector via restriction cloning.

Transfection

HAP1 cell lines were transfected with Turbofectin 8.0 (Origene) or Polyjet™ (SignaGen Laboratories) according to manufacturer’s instructions. A549 cells were transfected with Polyjet™ according to manufacturer’s instructions. HEK293T cells were transfected with polyethyleneimine (PEI, linear, MW25000, polysciences 23966-1).

Generation of CRISPR-Cas9 genome edited cell lines

KO cell lines

Cells were transiently transfected with pX330 or pX459 plasmid harboring the corresponding sgRNA sequence for the respective protein target. Medium was exchanged 24 h after transfection followed by 24 h of recovery. Cells were enriched by sorting for mCherry⁺ cells using flow cytometry (Sony SH800S) or through puromycin selection (2 µg/ml, 48 h). Cell pools were diluted to single cells and expanded. Individual knockout clones validated by Sanger sequencing and western blot.

Intron tagged cell lines

The three plasmids (intron targeting, donor targeting, mCherry donor) were transfected into MTHFD1^K56R cells. After 48 h, cells were collected and sorted by flow cytometry (Sony SH800S). Cell pools were diluted to single cells and expanded. Colonies were visualized on an Opera Phenix (Perkin Elmer, now Revvity) to determine clones successfully tagged with the mCherry at the intron.

Lentiviral production and transduction

Lentivirus was produced by co-transfecting HEK293T cells with pMD2.G (Addgene plasmid #12259, a gift from Didier Trono), psPAX2 (Addgene plasmid #12260, a gift from Didier Trono) and lentiviral plasmid with PEI. 24 h after transfection, media was exchanged to full IMDM followed by another 48 h of incubation. Lentiviral supernatant was then harvested, filtered (0.45 µm), aliquoted, flash frozen and stored at -80°C until further use. Lentivirus supplemented with 5 µg/ml (genome-wide KO screen) or 8 µg/ml (single mutant cell line generation) were added to respective cells for 24 h, followed by medium exchange, recovery and selection or downstream experiments. Cells were selected either by antibiotic treatment (puromycin, 1µg/ml) or sorted for the expression of the fluorescence marker (Sony SH800S). To determine MOI, lentivirus was titrated at different ratio followed by selection. Fraction of positive cells was plotted against the respective lentivirus concentration. Transduced cell pools were validated by sequencing and western blot.

Genome-wide CRISPR Screens

The genome-wide CRISPR screen was performed as previously described (55). Briefly, MTHFD1^K56R cells were transduced with lentivirus containing the genome-wide ‘Brunello’ knockout library (Addgene #73179, a gift from David Root and John Doench) in medium with 5 µg/ml Polybrene, aiming for an MOI of ~1 and >500 x sgRNA coverage. After 48 h, medium was replaced with fresh medium supplemented with 1µg/ml puromycin. After another 48 h, cells were either collected (control condition) or grown in the following screening conditions: 1) DIA or 2) ADO (DIA with 50 µM adenosine). Medium was replaced every 3-4 days for a total of 14 days. Cells were then collected, and the genomic DNA was extracted (DNeasy Blood and Tissue Kit, Qiagen). Guide RNA sequences were amplified and attached with adapters using primers adapted from ref. (56). Product was gel purified and sequenced (CeMM, Biomedical Sequencing Facility). Sequenced guides were analyzed with PinAPL-Py online tool (57). The gene ranking results are listed in Suppl. Table S2.

Western blot

Unless otherwise stated, cells were lysed in RIPA buffer (50 mM Tris HCl, pH 7.4, 150 mM NaCl, 1% NP40, 0.25% Sodium deoxycholate, 1 mM EDTA, 0.1% SDS) supplemented with 1x protease inhibitor cocktail (Roche, cOmplete^™, Mini, EDTA-free Protease Inhibitor Cocktail). Cells were lysed on ice for 30 min and lysates were cleared by centrifugation for 10 min at maximum speed at 4°C. Protein amounts were determined by BCA assay (ThermoFisher, A32957) prior loading and separation by SDS-PAGE and transfer to PVDF membranes. Membranes were blocked in 5% milk-TBST (1% v/v) for 1 h at r.t. followed by incubation with primary antibody o.n. at 4°C. Membranes were washed three times with TBST and incubated with HRP-conjugated secondary antibodies, diluted 1:10000 in 0.5% milk-TBST for at least 1 h at r.t. After washing three times with TBST, membranes were incubated with Clarity (Max)™ Western ECL substrate (Bio-Rad) and developed with ChemiDoc™ MP imaging system (Bio-Rad). Blots were stripped with twice mild stripping buffer (1.5% glycine (w/v), 0.1% SDS (w/v), 1% Tween-20 (v/v), pH 2.2), followed by washing, blocking and reprobing with target primary and secondary antibody.

The following antibodies were used: rabbit mAb HA-tag (1:5000, Cell Signaling Technologies (C29F4), #3724), rabbit NUDT5 (1:2000, abclonal, A0609 or ab129172, Abcam, 1:1,000), rabbit pAb PPAT (1:2000, abclonal, A6698 or 1:2000, Proteintech, 15401-1-AP), mouse mAb GAPDH (1:5000, scbt (6C5), sc-32233), mouse mAb alpha-Tubulin (1:1000, Abcam, ab7291), β-actin (1:200, scbt, sc-69879) mouse anti-rabbit IgG-HRP (1:5000, scbt, sc-2357), goat anti-mouse IgG-HRP (1:5000, invitrogen, 31430).

Crosslinking Western blot

WT and WT NUDT5^KO cells were cultured in full medium prior to harvesting and washing with PBS. Pellets were resuspended in PBS and incubated with DSG (Santa Cruz Biotechnology, sc-285455) at indicated concentration for 30 min at r.t. Samples were quenched with glycine (200 mM final concentration) for 10 min at r.t. Equal amounts were loaded onto an SDS-PAGE and subjected to western blotting as described above.

Native PAGE

After treatment of cells in respective medium conditions for 24 h, cells were harvested, washed with PBS and pelleted (1000x g, 4°C, 5 min). Cells were lysed in RIPA buffer and lysates were cleared by centrifugation. Protein amounts were determined by BCA assay and equal amounts were mixed with 2x native sample buffer. Samples were loaded onto an 8% native PAGE and run on cold native running buffer. Western blots were performed as described above. Recipes: 2x native sample buffer: 40% glycerol (v/v), 62.5 mM Tris HCl, pH 6.8, 0.01% bromophenol blue (w/v); native running buffer: 25 mM Tris, 192 mM glycine; native separating gel (8%, 1x): 2.6 ml Acrylamide/Bis-acrylamide (30%/0.8% w/v), 7.29 ml 0.375 M Tris HCl (pH 8.8), 100 µl 10% APS (w/v), 10 µl TEMED; native stacking gel: 0.67 ml Acrylamide/Bis-acrylamide (30%/0.8% w/v), 4.275 ml 0.375 M Tris HCl (pH 8.8), 50 µl 10% APS (w/v), 5 µl TEMED.

High-throughput compound screen

MTHFD1^KO cells were cultured in DIA and treated with the CeMM inhouse compound library (89228 chemically diverse compounds). Both cell number, by using Hoechst staining, and CellTiter-Glo (Promega, #G7572), were used as a readout, depending on the stage of the screen. The screening was divided into three parts (1) primary screening, (2) follow-up and (3) validation. During the primary screen, MTHFD1^KO cells were treated with 10 µM of every compound and CellTiter-Glo (Promega) was used to assess their ability to increase cell growth after 48 h. From this primary screening, 17 compounds were selected as hits and rescreened in the follow-up part, in which the MTHFD1^KO cells were treated in a six-point dose response for 72 h to discard any false positives. In the validation part, MTHFD1^KO cells were treated with 10 compounds in an eight-point dose response curve, and cell numbers were counted by staining with Hoechst, imaging and counting nuclei using the Opera Phenix (Perkin Elmer, now Revvity) and Harmony software (Perkin Elmer, now Revvity).

Compounds were transferred on 384-well plates using an acoustic liquid handler (Echo 550, Labcyte/Beckman Coulter) and 1000 cells per well were dispensed on top of the drugs using a dispenser (Multidrop, Thermo Fisher Scientific) for a total of 50 μl/well and incubated for 48 h. Cell viability was measured using CellTiter-Glo assay (Promega) in a multilabel plate reader (EnVision, Perkin Elmer, now Revvity). Signal was then normalized to DMSO and adenosine control wells included on each plate.

Single metabolite supplementation screen

Nucleotide metabolites (adenine, adenosine, guanine, guanosine, cytidine, cytosine, uridine, uracil, thymine, 5-methyluridine) were purchased from Sigma and dissolved in DMSO at a concentration of 50 mM. Deoxynucleotides (dATP, dCTP, dTTP, dGTP) in 100 mM solutions were purchased from Sigma. Folate metabolites (folic acid from Sigma, 5-formyl tetrahydrofolic acid (5-CHO-THF), 5-methyltetrahydrofolic acid (5-MeTHF), 5,10-methenyl tetrahydrofolic acid (CH⁺-THF), 5,10-methylene tetrahydrofolic acid (CH₂-THF), tetrahydrofolic acid (THF), and dihydrofolic acid (DHF) purchased from Schircks Laboratories) were dissolved in DMSO at a concentration of 40 mM.

MTHFD1^KO cells cultured in DIA were treated with a single nucleotide or folic acid metabolite for 72 hours (50 µM). Cells were then stained with Hoechst, imaged, and counted using an Operetta (Perkin Elmer, now Revvity). Similarly, MTHFD1^K56R cells were cultured in DIA and treated with adenosine and in combination with nucleotide metabolites (50 µM final) for 72 h. Cells were then stained with Hoechst, imaged, and counted using an Operetta (Perkin Elmer, now Revvity).

RNA-seq and GO-term enrichment analysis

Patient cells were incubated in FULL, DIA, or ADO for 24 h. RNA was extracted using the RNeasy Mini kit (Qiagen) and sequenced by the Biomedical Sequencing Facility at CeMM using the Illumina HiSeq3000/4000 platform and the 50-bp single-end configuration. Reads were aligned with TopHat (58). Cufflinks (59) was used to assemble aligned RNA-Seq reads into transcripts, estimate their abundances, and test for significant differential expression. Genes were significant if the calculated q-value was less than 0.05. RNAseq data is stored with GEO study accession number GSE201334.

For the GO term enrichment analysis, up-regulated and down-regulated proteins were defined based on log(FC) for each experimental condition. The enrichment analysis for resulting list of proteins was performed using enrichr API (https://maayanlab.cloud/Enrichr/) (60, 61) through ‘enricher’ package in R (https://cran.r-project.org/web/packages/enrichR/index.html) with "GO_Biological_Process_2018" library.

Immunofluorescence staining

Cells were pretreated in different media conditions for 24 hours and then fixed in 4% paraformaldehyde, and permeabilized with Triton-X. The fixed cells were then blocked with 5% BSA, diluted in PBS, for 1 h and then incubated with rabbit anti-γH2A.X antibody (1:500, 9718S, Cell Signaling Technology) or rabbit anti-RPA2 antibody (1:500, HPA026306, Atlas Antibodies) o.n. at 4°C. Cells were washed and incubated with anti-rabbit Alex-Fluor 546 (1:1000, A11010, Thermo Fisher Scientific) or anti-rabbit Alexa Fluor 568 (1:1000, A11036, Thermo Fisher Scientific) and Hoechst 33342 (1 µg/ml, HY-15559A, MedChem Express) for 1 h at r.t. Cells were washed again with PBS prior to imaging. Foci were imaged using Opera Phenix (Perkin Elmer, now Revvity) and quantified using Harmony software (Revvity) and custom CellProfiler pipeline (62).

Cell cycle analysis

Cells were pretreated in indicated media conditions for 24 h, trypsinized and washed twice with ice-cold PBS. After washing, cells were resuspended in 0.5 ml of ice-cold PBS. 2 ml of ice-cold 70% EtOH was then added dropwise to the cells, while gently vortexing. Cells were then stored at -20°C for at least 20 min. After fixation, cells were resuspended in PBS for 5 min and then stained with propidium iodide (Cell Signaling Technology, #4087S) according to manufacturer’s protocol. DNA content was measured with Sony SH800S cell sorter and analyzed with FlowJo™.

Targeted metabolomics and stable isotope tracing

For experiments with ¹⁵N₅-adenosine only, WT, NUDT5^KO and HPRT1^KO cells were seeded and cultured in DIA for 24 h prior treatment with 50 µM isotopically labelled ¹⁵N₅-adenosine (Silantes, 125303601) for 24 h. For each replicate (cell cultured in DIA only as control) 5 million cells were washed with 1x PBS, pelleted, immediately snap-frozen, and stored at -80°C. Cell extraction was performed in 1.5 ml Eppendorf tubes by adding 500 µl of ice-cold 80:20 (v/v) MeOH:H₂O solution to the cell pellet followed by vigorously vortexing. Samples were centrifuged at 10000x g for 10 min at 4 °C before transferring the cell extract supernatant into 1.5 ml HPLC vials. The extraction of the cell pellets was repeated a second time and supernatants of the same samples were combined. Cell extract samples were dried using a nitrogen evaporator. The dried residue was reconstituted in 50 µl water. An aliquot of 10 µl reconstituted sample extract was mixed with 10 µl H₂O in a HPLC vial, vortexed and used for the LC-MS analysis.

A 1290 Infinity II UHPLC system (Agilent Technologies) coupled with a 6470 triple quadrupole mass spectrometer (Agilent Technologies) was used for the LC-MS/MS analysis. The chromatographic separation for samples was carried out on a ZORBAX RRHD Extend-C18, 2.1 x 150 mm, 1.8 µm analytical column (Agilent Technologies). The column was maintained at a temperature of 40 °C and 4 µl of sample was injected per run. The mobile phase A was 3% methanol (v/v), 10 mM tributylamine, 15 mM acetic acid in water and mobile phase B was 10 mM tributylamine, 15 mM acetic acid in methanol. The gradient elution with a flow rate of 0.25 ml/min was performed for a total time of 24 min. Afterwards a back flushing of the column using a 6port/2-position divert valve was carried out for 8 min using acetonitrile, followed by 8 min of column equilibration with 100% mobile phase A.

The triple quadrupole mass spectrometer was operated in an electrospray ionization negative mode, spray voltage 2 kV, gas temperature 150 °C, gas flow 1.3 l/min, nebulizer 45 psi, sheath gas temperature 325 °C, sheath gas flow 12 l/min. The metabolites of interest were detected using a dynamic MRM mode. PeakBot software (vers.0.9.54) was used for data processing (63).

For dual measurements of de novo and salvage derived metabolite flux, cell lines were incubated in the respective media conditions: FULL, DIA, and DIA supplemented with 1 mM isotope labelled sodium formate (Cambridge Isotope Laboratories, CDLM-6203-0.5) or 1 mM isotope labelled sodium formate and 50 µM isotope labelled ¹⁵N₅-adenosine (Silantes, 125303601) for 24 h prior to collection. For each replicate 5 million cells were washed with PBS, pelleted, immediately snap-frozen, and stored at -80°C. Sample extraction and collection was performed as described above.

For measuring nucleotides in DNA, 250000 cells were initially seeded and treated with the respective conditions as described above. Genomic DNA was isolated (Dneasy Blood and TissueKit/RNeasy Mini Kit with RNase-free DNase set, Qiagen), and 1 µg (Qubit) were used for digestion using the Nucleoside Digestion Mix (NEB), o.n. at 37 °C. Samples were snap frozen and stored at -80 °C until measurement.

High mass resolution data were acquired on a Vanquish UHPLC system coupled to an Orbitrap Fusion Lumos Tribrid mass spectrometer and MS1 peak areas quantified using Xcalibur software (Thermo).

Synthesis and characterization of dNUDT5

The synthesis and characterization of dNUDT5 is described in detail in ref (37).

dNUDT5 assays

Viability test

dNUDT5 was spotted (Labcyte Echo 550) at 8 doses (10 µM starting concentration, 10-fold dilution, five replicates) onto a 384-well plate. Cells (1000 cells/50 µl) in corresponding medium condition were seeded on top of compounds and viability measured after 72 h using CellTiter-Glo (Promega) in a multilabel plate reader (EnVision, Perkin Elmer, now Revvity). Signal was then normalized to DMSO and positive killing control (bortezomib, 10 µM) wells included on each plate.

Life cell imaging

MTHFD1^K56R RPA2-RFP intron-tagged cell lines were cultured in DIA with indicated compounds. Cells were imaged at respective timepoints using an Opera Phenix high-content confocal imaging system (Perkin Elmer, now Revvity).

Immunoprecipitation

Co-immunoprecipitation

MTHFD1^K56R cells were cultured in 10 cm dishes in DIA and transfected with HA₂-NUDT5 constructs using Polyjet. After 24 h, media was exchanged, and cells were grown for another 24 h prior harvesting. For adenosine-dependent binding studies, cells were additionally treated with 50 µM adenosine for 3 h. Cells were washed twice with ice-cold PBS, evenly divided in three aliquots, pelleted, flash frozen and stored at -80°C until further use. One aliquot of cell pellets was lysed in TBST (1%) with 0.05% NP40, 1x protease and phosphatase inhibitors (ThermoFisher Scientific, A32963) for 30 min at 4°C followed by clearing of lysate through centrifugation. Equal amounts of lysates were incubated with 2x washed (20 mM Tris HCl, 150 mM NaCl, pH 7.4 - washing buffer) anti-HA magnetic beads (Sigma-Aldrich, SAE0197) for 1 h at 4°C while being rotated. Beads were washed twice with washing buffer and once with water prior elution in Laemmli buffer at 95°C for 10 min. SDS-PAGE and western blot were performed as described above.

BFP-pulldown

MTHFD1^K56RNUDT5^KO cells and cells reconstituted with BFP-NUDT5 (10⁷ cells) were lysed and processed as described above. 20 µl of GFP selector resin (NanoTag Biotechnologies, N0510-L) slurry was washed with twice with lysis buffer according to manufacturer’s protocol and 2.5 mg of lysate was added to the beads. After 1 h incubation at 4°C while being rotated, beads were transferred to a Mini Spin column (BioRad, 7326204) and washed twice with wash buffer (lysis buffer without protease inhibitor) and once with Tris-HCl (100 mM, pH 7.8) according to manufacturer’s protocol. Beads were incubated with lysis buffer supplemented with 4% SDS at 95°C for 2 min. Eluates were captured and subjected to proteomic analysis (CeMM Molecular Discovery Platform).

NUDT5 TurboID

HEK293 FlipIn cells were transfected with NUDT5-TurboID or TurboID-EGFP and pOG44 using Lipofectamin 2000. After 5 days, cells were selected using blasticidin (100 µg/ml) and hygromycin (10 µg/ml) for approximately two weeks. Cells were seeded in 10 cm dishes and incubated o.n. The next day 1.3 µg/ml doxycycline was added, incubated o.n., before changing the medium to DIA, 1.3 µg/ml doxycycline, and 50 µM adenosine. After 24 h incubation, 50 µM biotin was added, incubated for 4 h, before harvesting. Pellets were washed 2x with PBS and then stored at -80 °C. Pellets were lysed with RIPA buffer containing 1 µg/µl benzonase for 30 min on ice. Lysates were cleared by centrifugation at top speed in a tabletop centrifuge at 4°C. Protein concentrations were measured using a detergent compatible protein assay and 3 mg per sample in 850 µl were mixed with 150 µl streptavidin beads and incubated on a wheel at 4 °C o.n. Beads were washed with 3x 1 ml RIPA buffer, 1x 1 ml 1 N KCl, 3x 1 ml PBS, before resuspension in 50 µl 1x Laemmli. Beads were boiled at 95°C for 10 min, eluates collected, and the elution procedure repeated one more time.

PPAT Co-Immunoprecipitation

WT cells were treated with 100 nM TH5427, dNUDT5 or DMSO for 20 h in a 10 cm dish. Cells were washed four times with 2 ml PBS before lysing with 500 µl 0.8% NP-40 based lysis buffer (Tris HCl pH 7.5, 0.8% NP-40, 5% glycerol, 1.5 mM MgCl₂, 100 mM NaCl, 25 mM NaF, 1 mM Na₃VO₄, 1 mM PMSF, 1 mM DTT, 10 µg/ml TPCK, 1 µg/ml Leupeptin, 1 µg/ml Aprotinin, 10 µg/ml soybean trypsin inhibitor). The cell suspension was transferred to a 2 ml reaction tube, incubated on ice for 30 min and then centrifuged for 30 min at 20000x g (4 °C). Cleared lysates were spiked with 1 µg of PPAT (15401-1-AP, Proteintech) or rabbit IgG control antibody (3900S, Cell Signaling Technologies) and incubated o.n. at 4 °C while being rotated. 30 µl magnetic protein G beads slurry (#1003D, ThermoFisher) was incubated with lysate/antibody mix for 15 min at r.t. while being rotated. After 3x washing with 1 ml lysis buffer protein were eluted with 30 µl 2x Laemmli buffer by boiling for 10 min at 70 °C. For proteomic analysis, co-immunoprecipitation was similarly performed with MTHFD1^K56R and MTHFD1^K56RNUDT5^KO cells.

Chemical pulldown

Profiling of the TH5427 proteome wide specificity was performed in triplicates as previously described (64). In brief, NHS-activated sepharose beads (17090601, Cytiva) were derivatized with an amine functionalized TH5427 analog (CBH-003) (53) (manuscript in preparation) at a coupling density of 0.5 mM with 0.75 µL TEA per 50 µl beads in DMSO. Free binding sites were blocked by 2.5 µL ethanolamine per 50 µl beads. Beads were washed with excess of DMSO and 0.8% NP-40 based lysis buffer. WT cell pellets were thawed on ice and lysed with 3x excess lysis buffer containing 1 µl/1 ml benzonase. The cell suspension was drawn through a 21G needle 10x before clearing cell debris by centrifugation for 30 min at 20000 × g (4 °C) in a table-top centrifuge. 20 µM of TH5427 or DMSO was spiked into 1 ml of lysate (5 mg/ml) and incubated for 30 min at 4 °C while being rotated. After mixing cell lysates with 50 µl beads for 2 h on a wheel at 4 °C the suspension was transferred in filter columns, washed 4x 1 ml lysis buffer and subsequently eluted with 2x 50 µl 2x Laemmli buffer by heating for 5 min at 95 °C. Elution fractions were used for gel loading or digested for LC-MS/MS analysis.

Proteomics

Proteomics Sample Preparation

Protein concentrations from BFP-IP samples were determined using BCA Protein Assay to normalize the input for FASP digestion (65). Peptides were cleaned up using SPE columns (ThermoFisher Scientific, San Jose, CA). Digest was confirmed using CPA (ThermoFisher Scientific, San Jose, CA). TMT labeling of peptides was performed according to instructions of the manufacturer. Channels were pooled and adjusted after acquiring a TestMix. Final channel pool was cleaned up using SPE columns (ThermoFisher Scientific, San Jose, CA) and fractionated into six fractions using high-pH off-line fractionation. Odd and even samples were pooled into six fractions and cleaned up using detergent removal kit (ThermoFisher Scientific, San Jose, CA).

2D-RP/RP Liquid Chromatography - Tandem Mass Spectrometry analysis

Mass spectrometry analysis was performed on an Orbitrap Fusion Lumos Tribrid mass spectrometer (ThermoFisher Scientific, San Jose, CA) coupled to a Dionex Ultimate 3000 RSLCnano system (ThermoFisher Scientific, San Jose, CA) via a Nanospray Flex Ion Source (ThermoFisher Scientific, San Jose, CA) interface. Peptides were loaded onto a trap column (PepMap 100 C18, 5 μm, 5 × 0.3 mm, ThermoFisher Scientific, San Jose, CA) at a flow rate of 10 μL/min using 0.1% TFA as loading buffer. After loading, the trap column was switched in-line with a C18 analytical column (2.0 μm particle size, 75μm IDx500mm, catalog number 164942, ThermoFisher Scientific, San Jose, CA). The column temperature was maintained at 50 °C. Mobile phase A consisted of 0.4% formic acid in water, and mobile phase B consisted of 0.4% formic acid in a mixture of 90% acetonitrile and 10% water. Separation was achieved using a three-step gradient over 30 min at a flow rate of 230 nL/min (increase of initial gradient from 6% to 9% solvent B within 30 s, 9% to 30% solvent B within 25.5 min, 30% to 65% solvent B within 4 min. Afterwards a wash and equilibration phase took place). In the liquid junction setup, electrospray ionization was enabled by applying a voltage of 1.8 kV directly to the liquid being sprayed, and a non-coated silica emitter was used.

The mass spectrometer was operated in a data-dependent acquisition mode (DDA). For MS2 acquisition, we collected a 375–1650 m/z survey scan in the Orbitrap at 120 000 resolution (FTMS1), the AGC target was set to “standard” and a maximum injection time (IT) of 50 ms was applied. Precursor ions were filtered by charge state (2-6), dynamic exclusion (60 s with a ±10 ppm window), and monoisotopic precursor selection. Precursor ions for data-dependent MSn (ddMSn) analysis were selected using 20 dependent scans (TopN approach). DDA was focused using an additional targeted mass list generated for proteins of interest. The quadrupole isolation window was set to 0.5 Da and the high-energy collision-induced dissociation (HCD) fragmentation technique was used at a normalized collision energy of 38%. The normalized AGC target was set to 200% with a maximum IT of 55 ms. Xcalibur Version 4.3.73.11 and Tune 3.4.3072.18 were used to operate the instrument.

Data processing and data analysis

Following data acquisition, the acquired raw data files were processed using the Proteome Discoverer v.2.4.1.15 platform, with a tandem mass tag (TMT)18plex quantification method selected. In the processing step, Sequest HT database search engine and the Percolator validation software node were used to remove false positives with a false discovery rate (FDR) set to 1% at the peptide and protein level under stringent conditions. The search queried a full tryptic digestion against the human proteome (Canonical, reviewed, Uniprot) and appended known contaminants with a maximum of two allowable miscleavage sites. Methionine oxidation (+15.994 Da) and protein N-terminal acetylation (+42.011 Da), as well as methionine loss (-131.040 Da) and protein N-terminal acetylation with methionine loss (-89.030 Da) were set as variable modifications, while carbamidomethylation (+57.021 Da) of cysteine residues and TMT18plex labeling of peptide N-termini and lysine residues (+304.207 Da) were set as fixed modifications. Data were searched with mass tolerances of ±10 ppm and ±0.025 Da for the precursor and fragment ions, respectively. Both unique and razor peptides were used for TMT quantification. For reporter quantification a co-isolation threshold of 50% and average reporter mass S/N threshold was set to 10. Correction of isotopic impurities was applied. Data were normalized to total peptide abundance to correct for experimental bias and scaled ‘on all average’. Protein ratios are directly calculated from the grouped protein abundances using an ANOVA hypothesis test. Data from measurements NUDT5_Full and NUDT5^KO_Full were included and list of identified enriched proteins is provided in Suppl. Table S3.

Untargeted global proteomics and chemoproteomics

For determination of the interaction between NUDT5 and PPAT, proteomics samples were digested with trypsin using STrap columns following the manufacturer’s protocol (C02-micro, ProtiFi). Approximately 80% of the elution fraction of the pulldown were spiked with 5% SDS, then reduced by 20 mM dithiothreitol (M02712, Fluorochem), and alkylated by 40 mM iodoacetamide (I1149, Sigma). Phosphoric acid and S-Trap protein binding buffer were added into the sample lysates. Then, the SDS lysate/S-Trap buffer were loaded on a S-Trap column (C02-micro-80, ProtiFi). 1 μg of trypsin (V5111, Promega) was used to digest each sample at 37°C for 20 h. The samples were then eluted and dried by vacuum centrifugation. Peptide pellets were resuspended in mass spectrometry grade water with 2 % acetonitrile (85188, Thermo Scientific) and 0.1 % trifluoroacetic acid (85183, Thermo Scientific). The samples were injected into Orbitrap Fusion Lumos Tribrid Mass Spectrometer (Thermo Scientific). Sample acquisition was performed as previously described (66).

Data analysis

Raw data were searched against the human database (UP000005640, downloaded 08/21) using DIA-NN (v.1.8.1) (67) by enabling FASTA digest for library free search and Deep learning-based spectra, RTs and IMs prediction. Trypsin/P was selected for the in-silico digestion of the provided fasta file, allowing up to 2 missed cleavages. Fixed modifications included N-terminal methionine excision and cysteine carbamidomethylation. Variable modifications were methionine oxidation and N-terminal acetylation, with a maximum of one modification per precursor. Precursor FDR was kept at 1% and MBR enabled. The protein groups matrix was further analyzed using Perseus (2.0.9.0) (68) or R studio (4.3.2). Potential contaminants were filtered against the top 300 most common Homo sapiens contaminants reported in CRAPome (69). Proteins were removed if detected less than 50% across all runs. After column-wise normalization by median subtraction, missing values were imputed by random numbers from a normal distribution (1.8 standard deviation, downshift width 0.3). Volcano plots were generated by calculating fold change and -log10(p-value) from a two-sided Student’s t-test.

AlphaFold modelling

AlphaFold models were generated with AlphaFold3 using the default settings (40). Four NUDT5 (UniprotID: Q9UKK9) and PPAT (UniprotID: Q06203) units were submitted for modelling and resulting structures were analyzed and illustrated using Pymol.

NUDT5 activity assay

Proteins were expressed in Rosetta (DE3), and purified via Ni-NTA affinity chromatography column (HisTrap Crude FF GE Healthcare) and size-exclusion chromatography (16/600 Superdex 75 PG, GE Healthcare) following the previously published procedure (54). Enzymatic activity was performed using the AMP-Glo system (Promega) (54). Briefly, reactions were performed in 384-well plates in a 10 μl reaction volume with 1 nM of NUDT5^wt or mutant protein and 10 μM of ADP-ribose as the substrate. Reactions were incubated for 20 min, and stopped by addition 10 μL of AMP-Glo I. Following incubation with 20 μl of the detection solution for 1 h at room temperature luminescence was recorded in a PHERAstar FSX plate reader. Data were analyzed and plotted using Prism 10 (Graphpad).

PPAT activity assay

Expression and purification of NUDT5^wt and NUDT5^Y74E

Chemically competent E. coli K12 cells were transformed with pBAD H₆-TEV-NUDT5^wt or pBAD H₆-TEV-NUDT5^Y74E. After recovery with 1 ml of SOC medium for 1 h at 37 °C, the cells were cultured o.n. in 50 ml of 2× YT medium supplemented with ampicillin (100 µg/ml) at 37 °C, 200 rpm. The o.n. culture was diluted to an OD₆₀₀ of 0.05 in 200 ml of fresh 2× YT medium supplemented with ampicillin (50 µg/ml) and cultured at 37 °C with shaking (200 rpm) until OD₆₀₀ reached 0.6. Arabinose was added to a final concentration of 0.02 % and protein expression was induced for 3 h at 37 °C. The cells were harvested by centrifugation (4000 × g, 10 min, 4 °C) and resuspended in 15 ml of lysis buffer (20 mM Tris pH 7.5, 300 mM NaCl, 30 mM imidazole, 0.175 mg/ml PMSF). The cell suspension was incubated on ice for 30 min and sonicated with cooling in an ice-water bath. The lysed cells were centrifuged (14,000 × g, 20 min, 4 °C), the cleared lysate added to Ni-Sepharose 6 Fast Flow (Cytiva) (0.5 ml of slurry per 1 l of culture) and the mixture was incubated with agitation for 1 h at 4 °C. After incubation, the mixture was transferred to an empty plastic column and washed with 10 CV (column volumes) of wash buffer (20 mM Tris pH 7.5, 300 mM NaCl, 30 mM imidazole) followed by elution in 300 µl fractions with wash buffer supplemented with 300 mM imidazole. The NUDT5 containing fractions were pooled together and concentrated using Amicon centrifugal filter units with a 10 kDa MWCO (Millipore) followed by size exclusion chromatography (SEC) using a Superdex Increase 75 10/300 (GE Healthcare) with SEC buffer (PBS pH 7.4). Fractions containing NUDT5 were pooled together and concentrated, and protein concentration was calculated from the measured A₂₈₀ absorption (extinction coefficients were calculated with ProtParam (https://web.expasy.org/protparam/)).

Expression and purification of PPAT

PPAT was expressed and purified as previously described in (70) with slight adaptions. Bacmid generation, ExpiSf9 handling, transfection and transduction were performed according to manufacturer’s instructions using the ExpiSf expression system (Thermo Fisher Scientific). In short, pFastBac1-PPAT-H₆ was transformed in MAX Efficiency DH10Bac followed by blue/white selection to identify clones containing PPAT bacmid. PPAT Bacmid was isolated using the PureLink HiPure Plasmid Purification Kit (Thermo Fisher Scientific) and verified by Sanger sequencing. Baculovirus was generated by transfecting ExpiSf cells with PPAT Bacmid using ExpiFectamine and isolated according to manufacturer’s instructions. P0 viral stock was used to infect ExpiSf Enhancer pretreated ExpiSf cells followed by expression at 27 °C, 130 rpm for 60 h. Cells were harvested by centrifugation (1000 × g, 20 min, 4 °C) and resuspended in N₂-purged lysis buffer (25 mM HEPES pH 7.8, 500 mM NaCl, 10 mM MgCl₂, 25 mM Imidazole, 10 % glycerol, 2 mM AMP, 10 mM DTT). The cell suspension was incubated on ice for 10 min and sonicated with cooling in an ice-water bath. The lysed cells were centrifuged (14,000 × g, 30 min, 4 °C), and the supernatant was subjected to Ni-NTA chromatography using a HisTrap FF (Cytiva) with a gradient 0 % - 100 % B (buffer A: 25 mM HEPES pH 7.8, 500 mM NaCl, 10 mM MgCl₂, 25 mM Imidazole, 10 % glycerol, 2 mM AMP, 10 mM DTT, buffer B: 25 mM HEPES pH 7.8, 500 mM NaCl, 10 mM MgCl₂, 400 mM Imidazole, 10 % glycerol, 2 mM AMP, 10 mM DTT). Fractions containing PPAT (identified via SDS-PAGE) were pooled, concentrated and rebuffered into lysis buffer using Amicon centrifugal filter units with a 10 kDa MWCO (Millipore). PPAT was immediately flash frozen in liquid nitrogen and stored at -80 °C until further use.

Activity assay

PPAT activity assay was performed as previously described in (71). In brief, the first half reaction of PPAT reaction was assayed by monitoring the production of glutamate from glutamine, which was further coupled to an NADH fluorescence monitoring assay using glutamate dehydrogenase (GDH). Freshly thawed PPAT was diluted to approx. 100 nM in assay buffer (100 mM K₃PO₄ pH 7.9, 10 mM glutamine, 2.5 mM phosphoribosyl pyrophosphate, 37.5 mM NAD⁺, 115 U/mL GDH (Sigma G2626) with or without 1 µM NUDT5^wt or NUDT5^Y74E and NADH production was monitored in a 96-well plate using a Tecan Spark plate reader (excitation: 340 nm, emission: 460 nm, in triplicates) at 25 °C. Assays without proteins were run in parallel for baseline correction. Data processing was performed using GraphPadPrism (Prism 10). Values of initial lagging phase (0 - 500s) was occluded in the figure representation.

Fluorescence competition assay

MTHFD1^K56RNUDT5^KO cells were transduced with lentiviral mcherry-P2A-NUDT5 constructs and 8 µg/ml polybrene, aiming for an MOI <1. After 24 h, medium was exchanged, and cells were expanded for another 24 h. Equal amounts of cells were then seeded in 6 well plates in DIA. The next day, medium was exchanged to DIA or ADO and cells were cultured for another 72 h. Cells were harvested and subjected to FACS analysis (LSRFortessa™, BD Biosciences). Viable, single cells were included in the analysis (FlowJo™). The population of mCherry positive cells were gated according to mCherry negative control cells, and the ratio of mCherry positive cells was determined by normalizing the values of ADE treated cells to DIA control conditions.

Nucleotide sensitivity screen

Compounds from the Nucleotide library (MedChemExpress) were spotted at three different doses (final: 0.5, 2.5, 25 µM, in triplicates) in 125 nl DMSO onto 384-well plates by acoustic transfer (Labcyte Echo 550). Cells (1000 cells/50 µl, in FULL) were seeded on top of compounds and viability measured after 72 h using CellTiter-Glo (Promega) in a multilabel plate reader (EnVision, Perkin Elmer, now Revvity). Signal was then normalized to DMSO and positive killing control (bortezomib, 10 µM) wells included on each plate. Data were analyzed and visualized with PipelinePilot and TIBCO® Spotfire®.

Nucleobase, nucleoside and nucleotide toxicity assay

MTHFD1^K56R, MTHFD1^K56RNUDT5^KO and MTHFD1^K56RNUDT5^KO cells reconstituted NUDT5 mutant variants were cultured in 96 well plates (1200 cells/100 µl/well, in triplicates) in DIA with or without 50 µM of indicated compound for 72 h. The following cell viability readouts were used: 1) After incubation, medium was exchanged to dialyzed with Hoechst 33343 (1 µg/ml) and after 10 min of incubation nuclei were imaged and counted (Opera Phenix, Revvity). 2) Cell viability was measured using CellTiter-Glo assay (Promega). Signals were normalized to control condition.

Nucleotide analog toxicity assay

Cells were cultured in FULL in 96 well plates (1200 cells/100 µl/well, in triplicates) supplemented with or without indicated concentration of nucleotide analogs. After 72 h incubation cell viability was measured using CellTiter-Glo assay (Promega). Signals were normalized to control condition.

PPAT overexpression and metabolite treatments

MTHFD1^K56R cells were transduced at a high MOI with lenti-EF1alpha-EGFP or lenti-EF1alpha-PPAT. Two days after transduction, cells were seeded in a 96-well plate and treated with different concentrations of adenosine. After 48 h incubation cell viability was measured using CellTiter-Glo assay (Promega). For experiments in HEK293T, cells were transiently transfected in 96-well format and 6-TG was added after 24 h. Cell viability was measured using the CellTiter-Glo assay (Promega) assay 48h after transfection, 24 h after treatment.

Patient-derived MTHFD1-deficient fibroblast and compound treatment

Fibroblast cell lines and four unrelated healthy controls were cultured in Dulbecco’s Modified Eagle Medium (DMEM, high glucose, GlutaMAX™ Supplement, pyruvate, Gibco) supplemented with 10% (v/v) FBS and 1% (v/v) P/S. Fibroblasts were seeded at a density of 1200 cells per well in 96-well optical-bottom plates (Revvity) (100 µl/well) and cultured in Dulbecco’s Modified Eagle Medium (DMEM, high glucose, GlutaMAX™ Supplement, pyruvate, Gibco) supplemented with 10% (v/v) dialyzed FBS and 1% (v/v) P/S. Cells were treated for 72 h with all combinations of five concentrations of adenosine (0–100 µM) and dNUDT5 inhibitor (0–200 nM) prepared in PBS. Proliferation was assessed via nuclear staining using Hoechst 33342 (1 µg/mL, 10 min at r.t), followed by high-content imaging on the Operetta CLS system (Perkin Elmer, now Revvity). Nuclei were segmented and counted using Harmony software (version 4.9) to quantify cell proliferation. Each condition was tested in triplicate across two independent experimental batches.

Mouse xenograft studies

Animal husbandry and the experimental protocol as described followed ethical guidelines and were approved by the Austrian Federal Ministry of Science, Research and Economy (protocol number: 2021-0.746.732). Mice were bred in-house. 2.5 × 10⁶ A549 and A549 NUDT5^KO cells, diluted 1:1 in Matrigel, were transplanted subcutaneously into 8-14 weeks old NOD scid gamma mice. Treatment with 6-TG (1,5 mg/kg body weight, 5 times per week and volume matched 5 % DMSO in PBS as controls) by intraperitoneal injection was started when tumors were established. Tumor volumes were evaluated twice a week by measuring two perpendicular diameters with calipers and calculated using the following equation: (width × width × length)/2.

Statistical Analysis

All statistical tests and plots were performed with GraphPadPrism (Prism 10). Unless otherwise stated, comparisons among three or more conditions were performed ANOVA and Tukey’s HSD to correct for multiple comparisons. For comparisons between two conditions, T-tests were performed with a two-tailed p-value.

Data and materials availability

All modified cell lines generated in this study (HAP1 MTHFD1^K56R, HAP1 MTHFD1^K386E, HAP1 MTHFD1^K56RHPRT1^KO, HAP1 MTHFD1^K56RNUDT5^KO, MTHFD1^K56RNUDT5^Y74E, MTHFD1^K56RNUDT5^Y74F, HAP1 HPRT1^KO, HAP1 NUDT5^KO, HAP1 NUDT5^Y74E, HAP1 NUDT5^Y74F, A549 NUDT5^KO, A549 HPRT1^KO) are available from CeMM under a material transfer agreement with the Institution. RNA-seq data have been deposited in GEO (accession number GSE201334). Data from the genome-wide knockout screen have been deposited in GEO (accession number GSE306775). Proteomics measurements have been deposited in PRIDE (PXD067739 – related to Fig. 5A, PXD067706 – related to Fig. S8).