Mitochondrial NADP(H)–dependent de novo dTMP biosynthesis counteracts the cytotoxicity of PARP inhibitors

Significance

In tumor cells bearing intrinsic or acquired PARP inhibitors (PARPis) resistance, PARPi administration exclusively boosts mitochondrial NADP(H) fluxes for the biosynthesis of mitochondrial dTMP, thereby maintaining functional mitochondria. PARPi-mediated activation of STAT3–FoxO1 axis governs this metabolic rewiring process, which in turn guides PARPi responses across human cancer types, regardless of homologous recombination deficiency status. Our results uncover a fundamental vulnerability that can be leveraged by cotargeting STAT3 and PARP.

Abstract

PARP inhibitors (PARPis), known to elicit mitochondrial protection in nononcological diseases by elevating the cellular NAD⁺ pool, exhibit potent cytotoxicity in selected human cancers. The role of mitochondrial metabolism in PARPi-mediated antitumor therapy remains unexplored. Here, we propose a causal link between mitochondrial NAD⁺ metabolism and PARPi responsiveness. In PARPi-non-responsive tumor cells, PARP inhibition specifically expands mitochondrial NADP(H) [mito-NADP(H)] pool, thereby facilitating de novo mitochondrial dTMP (mito-dTMP) biosynthesis and maintaining mitochondrial dTTP (mito-dTTP) pool to prevent uracil misincorporation into mitochondrial DNA (mtDNA), regardless of homologous recombination (HR) status. Mechanistically, loss of PTPN1 ADPRylation by PARPi abolishes its phosphatase activity toward STAT3, yielding enhanced STAT3 phosphorylation and the subsequent transactivation of FoxO1. FoxO1 modulates transcriptomic signature governing mitochondrial NADPH fluxes to de novo mito-dTMP generation. Our results uncover a fundamental vulnerability that can be leveraged by cotargeting STAT3 and PARP to trigger mitochondrial dysfunction.

PARPis are FDA-approved drugs for treating ovarian, breast, prostate, and pancreatic cancers with HRD, particularly those with BRCA1/2 mutations. However, extending PARPi treatment from BRCA-deficient (BD) to BRCA-proficient (BP) cancers remains a clinical challenge (1–4). Emerging evidence has suggested that BRCA mutations or HRD do not entirely account for the benefits derived from PARPi (5–20). Indeed, the FDA has approved olaparib and niraparib as maintenance therapy for adult patients with recurrent epithelial ovarian, fallopian tube, or primary peritoneal cancer, who are in a complete or partial response to platinum-based chemotherapy, regardless of BRCA1/2 status (21, 22). Supporting this, additional mechanisms, including abrogated ribosome biogenesis and protein translation (23), enhanced replication stress (4, 24), have been reported to underlie PARPi efficacy. Therefore, identifying widespread HRD-independent indicators of PARPi sensitivity will provide insights to prospectively select broader populations of responsive patients, as well as to develop selective/potent agents for combination therapy.

Utilizing NAD⁺ as a substrate (25), PARP family members catalyze the transfer of ADP-ribose to target proteins, a process known as ADPRylation. NAD⁺ and its reduced form NADH [(NAD(H) hereafter] are essential for diverse metabolic processes, including ATP synthesis and energy production in mitochondria (26). Consistently, PARP1 hyperactivation upon DNA damage markedly depletes NAD⁺ and ATP, leading to mitochondrial dysfunction and energy imbalance (27). Meanwhile, PARPis have shown therapeutic potential in nononcological diseases such as neurodegenerative disorders via restoring NAD⁺ to elicit mitochondrial protection (28). The influence of PARPi on mitochondrial homeostasis in the context of antitumor therapy and its potential therapeutic implications remain unknown.

Results

Mitochondrial Homeostasis and NAD⁺ Metabolism Correlate with PARPi Response in HR-Competent Tumor Cells.

To ask whether the paradoxical protective role of PARPi in nononcogenic diseases may hinder its antitumor efficacy, we collected a series of human tumor cell lines derived from ovarian, breast, prostate, and pancreatic cancers (Dataset S1). As expected, a significant number of BP cell lines (23 out of 42) were responsive to different PARPis (SI Appendix, Fig. S1 A and B), with IC₅₀ values of olaparib below 3 μM in colony formation assays. This finding was further confirmed by analyzing olaparib response among BP cell lines provided in the CCLE database (SI Appendix, Fig. S1C). We selected three PARPi-responsive (BP^RP, including A2780, TYKNU, and OVCAR8 cells) and three PARPi-nonresponsive cells (BP^NRP, including CAPAN2, SUM159, and BT20 cells), according to their IC₅₀, for further experiments. Analyzing genetic alterations of key DNA repair pathway genes (29, 30) revealed no significant difference between BP^RP and BP^NRP groups (SI Appendix, Fig. S1D). Moreover, neither group exhibited detectable HR defects based on RAD51 foci formation and HR reporter assay (SI Appendix, Fig. S1 E–H).

We next evaluated mitochondrial homeostasis in BP^RP and BP^NRP cells upon olaparib treatment. Severe mitochondrial disorders, including impaired mtDNA integrity, abnormal mitochondrial morphology, decreases in oxygen consumption rates (OCR), and a reduction in ATP content, were observed in BP^RP but not BP^NRP cells (SI Appendix, Fig. S1 I–L). To confirm these findings in vivo, we applied olaparib treatment to two PDX models generated from BRCA1/2 wild-type (BRCA WT) high-grade serous ovarian cancer (HGSOC) samples directly obtained from patients undergoing surgery without olaparib pretreatment (SI Appendix, Fig. S1 M and N), one was olaparib-responsive (O1-BP-PDX^RP), and the other was olaparib-unresponsive (O2-BP-PDX^NRP). Following olaparib treatment, mitochondrial disorders were only observed in O1-BP-PDX^RP but not in O2-BP-PDX^NRP tumor samples (Fig. 1 A–D). Collectively, our data suggest a tight correlation between the maintenance of mitochondrial function and PARPi unresponsiveness in BP tumors.

Fig. 1.

Mitochondrial homeostasis and NAD⁺ metabolism correlate with PARPi response in HR-competent tumor cells. (A) qPCR analysis of relative mtDNA copy numbers in the indicated PDXs treated with olaparib (50 mg/kg). (B) Representative transmission EM (electron microscopy) images (Left panel) and quantitative results (Right panel) of the indicated PDXs as described in (A) were shown. (C) OCR of the indicated PDXs as described in (A) was determined using a Seahorse Bioflux analyzer. (D) ATP content of the indicated PDXs as described in (A) was determined by luminescence. (E) Relative proliferation rate was assessed in SUM159 cells treated with olaparib (10 μM) for 5 d. (F) SUM159 cells expressing iNap3-mito were treated with olaparib (10 μM) for 4 d. The normalized ratio of fluorescence intensities excited at 420 and 485 nm (F_420nm/F_485nm indicates NADPH) was recorded by the microplate reader. (G) Olaparib-treated SUM159 cells were incubated with ¹³C₃-¹⁵N[NAM] for 1 h, and the abundance of M+4 isotopomers of NADP⁺ in anti-HA immunopurified mitochondria was analyzed by liquid chromatography–mass spectrometry (LC–MS).

Given that PARPi-mediated NAD⁺ accumulation can improve mitochondrial function in nononcological diseases (31), we explored the connection between NAD⁺ metabolism and PARPi response in BP cancer cells. BP^RP and BP^NRP cell lines stably expressing FiNad (32) were constructed to monitor cellular NAD⁺. Basal cellular NAD⁺ contents were comparable between these two groups (SI Appendix, Fig. S1O), and acute olaparib exposure increased NAD⁺ in both groups (SI Appendix, Fig. S1P). However, after 3 d of treatment, BP^NRP cells showed reduced NAD⁺, whereas BP^RP cells retained NAD⁺ abundance. NAD⁺ depletion (33) or NADP⁺ elevation (20) has been found to exacerbate PARPi cytotoxicity. However, our results do not support the reported correlation between cellular NAD⁺ or NADP⁺ contents and PARPi responsiveness (SI Appendix, Fig. S1O), suggesting alternative mechanisms may account for PARPi sensitivity in BP cells. We therefore tested if reduced NAD⁺ in olaparib-treated BP^NRP cells was attributed to its conversion to NADH or NADP(H) (SI Appendix, Fig. S1Q). Assessing NADH/NAD⁺ ratio with SoNar (34), cellular or mitochondrial NADPH levels with respective iNap1 or iNap3-mito reporter (35), and NADP⁺/NADPH ratio utilizing an adapted enzyme cycling assay (36), we found PARPi specifically increased mito-NADP(H) but not whole-cell NAD(H) or NADP(H) in nonresponsive cells (SI Appendix, Fig. S1 R–U), indicating a positive correlation between mito-NADP(H) pool and PARPi unresponsiveness in BP tumor cells.

To determine if PARPi-induced mito-NADP(H) pool expansion results from its compartment-specific de novo synthesis, i.e., the metabolic flux from NAD⁺ to NADP⁺ in the mitochondria. An isotope tracing experiment (36) by culturing BP^RP and BP^NRP cells in medium supplemented with [¹³C₃-¹⁵N]nicotinamide (NAM) was conducted (SI Appendix, Fig. S1V), followed by mitochondria isolation using the mitochondrial immunopurification (Mito-IP) method (37), and the content of M+4 mass isotopomers of NAD⁺ and NADP⁺ in mitochondria or whole cells was determined by LC–MS. As shown in SI Appendix, Fig. S1W, the purity of the mitochondrial fraction was rigorously validated using compartment-specific markers. BP^NRP cells showed an 80% increase in mito-NADP⁺ (M+4) (SI Appendix, Fig. S1X), with a marginal effect on whole-cell NAD⁺ (M+4) and NADP⁺ (M+4). In contrast, PARPi increased whole-cell NAD⁺ (M+4) by 30% in BP^RP cells (SI Appendix, Fig. S1Y). These results indicate that PARPi induces the de novo synthesis of mito-NADP⁺ from NAD⁺ exclusively in BP^NRP cells.

NAD phosphorylation, the only known reaction to generate NADP(H) from NAD(H), is catalyzed by NAD kinases (NADK) (38). Human cells express two NADKs, cytosolic NADK1 and mitochondrial NADK2, to catalyze the synthesis of compartment-specific NADP(H) (SI Appendix, Fig. S1Q) (39). To investigate whether PARPi-induced mito-NADP(H) pool affects PARPi response, we depleted NADK2 in SUM159 cells (BP^NRP) (SI Appendix, Fig. S1Z), and observed a significantly reduced cell proliferation following olaparib administration (Fig. 1E), accompanied by a drastic decrease in mito-NADP(H) pool (Fig. 1 F and G). These phenotypes were rescued by ectopic expression of a wild-type NADK2 (NADK2^WT), but not its catalytically dead mutant NADK2^D161A (40).

Interestingly, olaparib selectively induced NADK2, but not NADK1, in nonresponsive cells (SI Appendix, Fig. S1AA). We therefore mildly depleted NADK2 in olaparib-treated SUM159 cells to a level comparable to that in DMSO-treated cells (SI Appendix, Fig. S1AB). This moderate depletion was sufficient to inhibit PARPi-induced mito-NADP(H) pool (SI Appendix, Fig. S1AC), and sensitize cells to PARPi (SI Appendix, Fig. S1AD). By contrast, silencing NADK1 had no impact on mito-NADP(H) pool or cell proliferation following olaparib exposure. These findings indicate that PARPi-induced NADK2 is crucial for the subsequently increased mito-NADP(H) synthesis, which in turn renders cells unresponsive to PARPi.

PARP Inhibition Enhances mito-NADP(H) Pool-dependent De Novo mito-dTMP Biosynthesis to Prevent Uracil Misincorporation into mtDNA in BP^NRP Cells.

We were curious about how mito-NADP(H) modulates PARPi responsiveness. NADP(H) pool is believed to supply reducing power for antioxidant regeneration to neutralize ROS (41). However, olaparib-treated SUM159 cells lacking NADK2 display no increase in cellular ROS, mitochondrial superoxide (MitoSox), and mitochondrial H₂O₂ or GSH oxidation (SI Appendix, Fig. S2 A–E). NADP(H) also provides reducing power for reductive biosynthesis to support cell proliferation (26). We therefore attempted to rescue NADK2 depletion-induced proliferation defects in olaparib-treated SUM159 cells by supplementing individual metabolic intermediates whose scavenging or synthesis relies on NADP(H) availability (SI Appendix, Fig. S2F). Treatment with the ROS scavenger N-acetylcysteine (NAC), a ferroptosis inhibitor, fatty acids, or a mixture of nonessential amino acids (NEAAs) including single proline, which has recently been connected to mito-NADP(H) for its biosynthesis (42, 43), all exhibited negligible impacts on cell proliferation. Surprisingly, supplementation with dTMP, but not other dNMPs, yielded a significant rescue effect (SI Appendix, Fig. S2G). Orally administered dTMP, which can be catabolized to cell-permeable thymidine by nucleotidases (44), has been reported to treat mtDNA-depletion syndrome caused by an impaired dNTP pool (45–48). Indeed, [¹⁵N₂]dTMP isotope tracing showed extensive labeling of thymidine, dTMP, and dTTP in both whole cells and mitochondria (SI Appendix, Fig. S2H), supporting that exogenous dTMP can directly and effectively replenish mito-dTTP. Consistently, unbiased metabolomic profiling revealed that PARPi remarkably depleted mito-dTMP only in PARPi-responsive cells (Fig. 2A), where dTMP or thymidine replenishment reversed their proliferation defects (SI Appendix, Fig. S2I). These results indicate that maintaining mito-dTMP homeostasis is crucial for determining cell fate in response to PARPi.

Fig. 2.

PARP inhibition enhances mito-NADP(H) pool-dependent de novo mito-dTMP biosynthesis to prevent uracil misincorporation into mtDNA in BP^NRP cells. (A) Heatmap representing changes of the indicated metabolite levels measured by GC-MS in anti-HA immunopurified mitochondria from the indicated cell lines treated with olaparib (10 μM) for 4 d. The average of three biological replicates is shown. For each metabolite, values of olaparib-treated cells are shown as log₂(fold change) relative to the value of DMSO-treated cells. (B) Images (Left panel) and volumes (Right panel) of tumors collected from the indicated PDXs treated with olaparib (50 mg/kg), dTMP (3.2 mM), or a combination of both agents were shown. Tumor volume data were represented as mean ± SEM of n = 10 mice per group. (C) SUM159 cells were infected with the indicated lentiviruses, followed by olaparib treatment for 3 d. Cell lysates were subjected to IB analysis as indicated. (D) The indicated cells as described in (C) were treated with olaparib (10 μM) for 4 d, followed by labeling with [²H₄]dT or [¹³C-¹⁵N₂]dU for 8 h. Normalized peak areas of mito-dTTP (M+4), mito-dTTP (M+3), and total mito-dTTP were shown.

The steady-state level of mito-dTMP prevents uracil accumulation in mtDNA, which is essential for maintaining mtDNA integrity and mitochondrial function (49, 50). In agreement with the reduced mito-dTMP levels in olaparib-treated responsive cells, increased uracil content in mtDNA was observed in these cells as well (SI Appendix, Fig. S2J). As expected, in SUM159 cells lacking NADK2, dTMP supplementation effectively diminished uracil content in mtDNA (SI Appendix, Fig. S2K) and prevented mitochondrial disorders triggered by PARPi (SI Appendix, Fig. S2 L–O). These results were further confirmed in O1-BP-PDX^RP mice treated with olaparib, where dTMP replenishment rescued mitochondrial functions (SI Appendix, Fig. S2 P–S) and restored tumor growth (Fig. 2B). Taken together, these results indicate that PARPi-induced mito-NADP(H) pool in PARPi-unresponsive tumors maintains mtDNA integrity by preserving mito-dTMP levels, thereby counteracting the cytotoxicity of PARP inhibition.

To uncover the mechanism underlying the differences in mito-dTMP homeostasis between BP^RP and BP^NRP cells treated with PARPi, as well as its connection with PARPi-induced mito-NADP(H) pool, we first monitored the well-characterized salvage pathway synthesis of mito-dTMP from thymidine (SI Appendix, Fig. S2T), which is catalyzed by thymidine kinase 2 (TK2) (51), whose mutations are associated with mtDNA depletion syndromes and severe mitochondrial dysfunction in humans (52–54). To this end, we labeled tumor cells with [²H₄]thymidine (50) and determined that 8 h of labeling was sufficient for [²H₄]thymidine to reach its steady state (SI Appendix, Fig. S2U). Utilizing isolated mitochondria by mito-IP, we analyzed mito-dTMP synthesis. We noticed that both BP^RP and BP^NRP cells displayed largely decreased mito-dTTP (M+4) under PARPi (SI Appendix, Fig. S2V), suggesting a compromised salvage mito-dTMP synthesis, which needs further investigation.

We next turned to an alternate mito-dTMP generation pathway, known as de novo biosynthesis from deoxyuridine monophosphate (dUMP) (SI Appendix, Fig. S2T). This reaction is catalyzed by thymidylate synthase (TYMS) and produces dihydrofolate (DHF), which is further converted to tetrahydrofolate (THF) by dihydrofolate reductase 2 (DHFR2) in a NADPH-dependent manner (49). The regenerated THF is then used for subsequent rounds of mito-dTMP biosynthesis. Mito-IP (SI Appendix, Fig. S1W) and immunofluorescence staining assay (SI Appendix, Fig. S2W) showed clear localization of NADK2, TYMS, and DHFR2 in mitochondria. We labeled cells with [¹³C-¹⁵N₂]deoxyuridine (50) for 8 h (ensuring the isotopic labeling efficiency reaches steady-state levels, as validated in SI Appendix, Fig. S2X, reflecting the de novo biosynthesis of dTMP, which generates dTTP (M + 3). Strikingly, compared to BP^RP cells, BP^NRP cells exhibited largely increased mito-dTTP (M + 3) under PARPi (SI Appendix, Fig. S2V), implying an enhanced de novo mito-dTMP synthesis. As a result, the steady-state levels of total mito-dTTP were preserved in these cells, whereas BP^RP cells exhibited a substantial decrease in total mito-dTTP. Notably, in both responsive and unresponsive cells, olaparib treatment did not alter whole-cell dTMP generation (SI Appendix, Fig. S2Y). We next explored whether mito-NADP(H) pool maintains mito-dTTP homeostasis through its de novo biosynthesis. To this end, we silenced NADK2 or DHFR2 in SUM159 cells (Fig. 2C), the latter also functions exclusively in mitochondria, catalyzing THF metabolic process. Loss of either enzyme had no effect on whole-cell dTMP biosynthesis (SI Appendix, Fig. S2Z), but significantly impaired PARPi-induced de novo mito-dTMP generation (Fig. 2D), and the subsequent decrease of total mito-dTTP. These findings indicate that, in BP^NRP cells, NADP(H)-mediated de novo synthesis of mito-dTMP is essential for maintaining mito-dTTP homeostasis in response to PARPi.

FoxO1 Regulates PARPi Responsiveness Via Modulating Transcriptomic Signature Governing De Novo mito-dTMP Biosynthesis.

Consistent with the enhanced de novo mito-dTMP biosynthesis in BP^NRP cells upon PARPi, TYMS, NADK2, and DHFR2 were all transcriptionally up-regulated (SI Appendix, Fig. S3A). Analyzing publicly accessible databases (i.e., JASPAR, PROMO, and hTFtarget) identified FoxOs as potential candidate transcription factors (TFs) binding to promoters of all three genes (SI Appendix, Fig. S3B). Silencing each FoxO in SUM159 cells showed that FoxO1 depletion drastically enhanced PARPi sensitivity to an extent comparable to BRCA1 deficiency, and moderately weaker than BRCA2 silencing, whereas FoxO3 or FoxO4 knockdown had no effect (SI Appendix, Fig. S3 C–H). Importantly, in BRCA1- or BRCA2-depleted SUM159 cells, FoxO1 loss further enhanced PARPi-induced cytotoxicity (SI Appendix, Fig. S3H), suggesting a unique role of FoxO1 in modulating PARPi response, independent of BRCA status. To test whether NADK2, DHFR2, and TYMS were direct transcriptional targets of FoxO1, we conducted the RNA-seq analysis of olaparib-treated SUM159 cells depleted of FoxO1 (SI Appendix, Fig. S3I), and the ChIP-seq analysis of FoxO1 chromatin deposition using anti-HA antibody in 3× HA (hemagglutinin) tag knock-in SUM159 cells (SI Appendix, Fig. S3J). Olaparib treatment significantly enriched FoxO1 binding at promoter regions of all three genes (SI Appendix, Fig. S3K). Silencing FoxO1, but not FoxO3 or FoxO4, profoundly abrogated their induction in response to olaparib treatment (SI Appendix, Fig. S3 C, E, and L).

To evaluate the functional contribution of these mitochondrial enzymes in de novo mito-dTMP biosynthesis downstream of FoxO1, we individually introduced these targets in FoxO1-depleted SUM159 cells (SI Appendix, Fig. S3M). Upon olaparib treatment, adding back NADK2^WT but not NADK2^D161A completely restored mito-NADP(H) abundance (SI Appendix, Fig. S3N). However, ectopic NADK2 only partially improved FoxO1 depletion-induced growth arrest under PARPi (SI Appendix, Fig. S3O). We therefore expressed DHFR2 or/and TYMS in NADK2-expressing cells and found that de novo mito-dTMP synthesis was further enhanced, accompanied by completely restored mitochondrial function and PARPi unresponsiveness (SI Appendix, Fig. S3 P–W). By contrast, in cells lacking FoxO1, individually adding back either DHFR2 or TYMS had minimal effect, indicating that DHFR2 or TYMS functions to coordinate NADK2-dependent mito-NADP(H) generation, thereby facilitating de novo mito-dTMP biosynthesis.

PARPi-Induced Activation of STAT3 Transactivates FoxO1.

In most BP^NRP cell lines (16 out of 19), FoxO1 was induced upon olaparib exposure (Fig. 3A and SI Appendix, Figs. S1AA and S4A), concomitant with a selective up-regulation of NADK2, DHFR2, and TYMS. We next investigated the upstream signaling pathways responsible for PARPi-triggered FoxO1 induction. Unlike previous reports (55, 56), PARPi did not alter SIRT activity or FoxO1 acetylation in SUM159 cells (SI Appendix, Fig. S4 B and C), ruling out SIRT1-mediated FoxO1 activation. We adopted an in vivo nuclease-deficient Cas9 (dCas9)-targeted chromatin-based purification strategy, known as “CLASP” (Cas9 locus-associated proteome) (SI Appendix, Fig. S4D) (57), coupled to anti-HA ChIP (SI Appendix, Fig. S4E), followed by mass spectrometry. Eight DNA-binding candidates were enriched in the sgFoxO1-guided group (Fig. 3B and Dataset S2, Log₂FC > 0.6), including STAT3, which was also identified by analyzing JASPAR, PROMO, and hTFtarget databases (SI Appendix, Fig. S4F). STAT3 phosphorylation was remarkably induced by PARPi in 15 out of 19 BP^NRP cells, in parallel with the augmented expression of FoxO1 (Fig. 3A and SI Appendix, Fig. S1AA). Consistently, olaparib caused significant enrichment of phospho-STAT3 at the FoxO1 promoter region, measured by ChIP-seq (SI Appendix, Fig. S4G). Importantly, STAT3 depletion markedly blocked olaparib-mediated FoxO1 and its targets induction (SI Appendix, Fig. S4 H and I), sensitizing SUM159 cells to PARPi (SI Appendix, Fig. S4J).

Fig. 3.

PARPi-induced activation of STAT3 transactivates FoxO1. (A) Heatmap showing PARPi-mediated induction of phospho-STAT3 (Tyr705) (pSTAT3), FoxO1, NADK2, DHFR2, and TYMS in the indicated cell lines as shown in SI Appendix, Fig. S1AA (fold change of protein induction in olaparib-treated versus DMSO-treated cells, >1.5 fold). (B) Volcano plot of proteins interacting with dCas9–HA in olaparib-treated SUM159 cells infected with the indicated lentiviruses. DNA-binding proteins enriched in the sgFoxO1 guiding group were labeled as red dots. (C) SUM159 cells expressing FoxO1-flag were infected with the indicated lentiviruses, followed by olaparib treatment for 3 d. Cell lysates were subjected to IB analysis as indicated. (D) Images (Left panel) and volumes (Right panel) of tumors collected from the indicated SUM159 xenografts treated with olaparib (50 mg/kg) were shown. Tumor volume data were represented as mean ± SEM of n = 10 mice per group.

We next silenced STAT3 in SUM159 cells stably expressing FoxO1-flag (Fig. 3C) and found ectopic FoxO1 partially yet significantly rescued mito-NADP(H) pool, de novo mito-dTMP synthesis, PARPi-induced mitochondrial disorders, and cytotoxicity (SI Appendix, Fig. S4 K–R). The in vivo influence of the STAT3–FoxO1 axis on antagonizing PARPi efficacy was further confirmed using a xenograft animal model (Fig. 3D). Meanwhile, sgRNA-mediated knockout confirmed that STAT3–FoxO1 transcriptional cascade orchestrates a resistance hierarchy to PARPi across different cancer types (SI Appendix, Fig. S4 S–AJ). Notably, PARPi-induced NADK2 was codependent on both STAT3 and FoxO1 (Fig. 3C and SI Appendix, Fig. S4S). Further analysis revealed that NADK2 was also a transcriptional target of STAT3, as phospho-STAT3 bound to the promoter region of NADK2 with enhanced density upon PARPi (SI Appendix, Fig. S4AK). Ectopic expression of STAT3^WT, but not dominant negative STAT3 (STAT3^Y705F), fully rescued FoxO1 and its target levels in olaparib-treated SUM159 cells lacking STAT3 (SI Appendix, Fig. S4AL), indicative of a collaborative regulation of transcription between STAT3 and FoxO1 under PARPi.

Impaired PTPN1 ADPRylation Drives PARPi-Induced STAT3 Phosphorylation.

To examine how PARPi activates STAT3, we first assessed JAK signaling and found that neither olaparib nor PARP1 depletion activated JAK1/2 (SI Appendix, Fig. S5A). Meanwhile, STAT3 interactome analysis revealed an increased interaction between STAT3 and its known phosphatase, PTPN1 (58), following olaparib treatment (Dataset S3), which was validated by immunoprecipitation (SI Appendix, Fig. S5B). PTPN1 depletion significantly activated STAT3–FoxO1 axis–dependent transcriptional signature in nonresponsive cells (SI Appendix, Fig. S5C), even in the absence of PARPi. Importantly, olaparib failed to further activate STAT3–FoxO1 transcriptional program in PTPN1-deficient cells, indicating an epistatic relationship between PARP1 and PTPN1 in modulating this signaling.

To test whether PARPi may inhibit the phosphatase activity of PTPN1, we incubated separate pools of endogenous phospho-STAT3 and PTPN1 immunoprecipitated from SUM159 cells treated with DMSO or olaparib (Fig. 4A), and performed an in vitro phosphatase assay. PTPN1 purified from olaparib-treated cells displayed profoundly diminished phosphatase activity on phospho-STAT3 (Fig. 4B). By contrast, phospho-STAT3 purified from either DMSO-treated or olaparib-treated cells was readily dephosphorylated by PTPN1 purified from the DMSO sample, demonstrating that PARPi acts on PTPN1, rather than STAT3, to influence the dephosphorylation process. We next confirmed that PTPN1 interacted with PARP1 at endogenous levels (SI Appendix, Fig. S5B), which was sensitive to olaparib treatment, likely due to trapped PARP1 caused by PARPi being less accessible to PTPN1. ADPRylation of PTPN1 was readily detectable across various cancer types (SI Appendix, Fig. S5D), which was abrogated in the presence of PARPi. Consistently, PARP1 directly ADPRylated PTPN1 in vitro (SI Appendix, Fig. S5E), and ADPRylated PTPN1 dephosphorylated phospho-STAT3 in a dose-dependent manner, as detected by an in vitro phosphatase assay (SI Appendix, Fig. S5F). Therefore, PARP1-mediated ADPRylation of PTPN1 is crucial for its phosphatase activity toward phospho-STAT3.

Fig. 4.

Impaired PTPN1 ADPRylation drives PARPi-induced STAT3 phosphorylation. (A and B) Schematic of the strategy for in vitro phosphatase assay. Different phosphatase reactions were performed in vitro as in (A). The same amount of PTPN1 or pSTAT3 extracted from olaparib-treated SUM159 cells was incubated as indicated. The products of the reactions were subjected to IB analysis (B). (C) Schematic representation showing ADPRylated sites of PTPN1 determined by mass spectrometry. (D) PTPN1-deficient SUM159 cells were reconstituted with WT or PTPN1 mutants, followed by olaparib (10 μM) treatment for 3 d. Cell lysates were subjected to “denature IP” for detecting ADPRylated PTPN1. The immunoprecipitates were then subjected to IB analysis as indicated.

Utilizing site-specific LC–MS/MS proteomics, we identified twenty ADPRylated sites on PTPN1 (SI Appendix, Fig. S5G). Individual mutants were generated by substituting Asp or Glu residues with Asn or Gln, respectively, and reconstituted in PTPN1-deficient SUM159 cells (SI Appendix, Fig. S5H). Five mutants, including E97Q, E101Q, D181N, E186Q, and D229N, exhibited remarkably impaired phosphatase activity on phospho-STAT3. Further site mapping assays identified that E97, D181, E186, and D229 are responsible ADPRylation sites on PTPN1 (Fig. 4C and SI Appendix, SI Appendix, Fig. S5 I and J). Furthermore, functional validation established that PTPN1 ADPRylation at E97, E186, and D229 is particularly important for its phosphatase activity on phospho-STAT3 both in vivo and in vitro (Fig. 4D and SI Appendix, Fig. S5K).

Given that PTPN1 expression varied greatly in a cell type–dependent manner (SI Appendix, Fig. S1AA), we speculated that PTPN1 abundance may dictate PARPi-induced STAT3–FoxO1 activation, and hence PARPi responsiveness. Based on the inducibility of STAT3–FoxO1 axis upon PARPi exposure, 42 BP tumor cell lines were stratified to PARPi-uninducible (24) and PARPi-inducible (18) groups (Fig. 3A). Interestingly, PTPN1 was highly expressed in the majority of PARPi-inducible cells (15/18) (SI Appendix, Fig. S5L), which were mainly PARPi nonresponsive (13/15), whereas low PTPN1 expression characterized most PARPi-uninducible cells (18/24), which were largely PARPi responsive (16/18). Furthermore, we evaluated PTPN1 expression and its ADPRylation across a panel of BP PDX models (3 BP^RP and 7 BP^NRP) derived from multiple cancer types (SI Appendix, Figs. S1 M and N and S5M). We detected much less basal expression of PTPN1 in BP^RP tumor samples, compared to that in BP^NRP samples (SI Appendix, Fig. S5N). Analysis of PTPN1 ADPRylation revealed that basal levels of ADPRylated PTPN1 were positively correlated with total PTPN1 protein expression (SI Appendix, Fig. S5O). Although PARPi suppressed PTPN1 ADPRylation in both BP^RP and BP^NRP tumor samples (SI Appendix, Fig. S5P), PARPi only activated the STAT3–FoxO1 axis in BP^NRP tumors. These data indicate a tight correlation between PTPN1 basal abundance and PARPi response in BP tumors.

We next explored why STAT3–FoxO1 axis activation is compromised in BP^RP cells. PARPi efficiently suppressed PTPN1 ADPRylation in both BP^RP and BP^NRP cells (SI Appendix, Fig. S5Q). However, in sharp contrast to the robust STAT3 phosphorylation in BP^NRP cells depleted of PTPN1 (SI Appendix, Fig. S5C), BP^RP cells lacking PTPN1 failed to activate STAT3–FoxO1 axis (SI Appendix, Fig. S5R), phenocopying PARPi treatment, suggesting an intrinsic decoupling of STAT3 activation from PTPN1-mediated dephosphorylation in BP^RP cells. A plausible explanation might be attributed to the low basal abundance of PTPN1 in these cells, and the reliance on PTPN1 to fine-tune STAT3 activation upon PARPi treatment is mainly dependent on the basal expression level of this phosphatase.

Activation of STAT3–FoxO1 Transcriptional Cascade Drives PARPi Resistance Across Diverse Therapeutic Contexts.

Despite the therapeutic efficacy of PARPi in certain patients with BD tumors, a subset remains intrinsically nonresponsive. To determine whether PARPi-triggered STAT3–FoxO1 axis activation also underlies de novo PARPi resistance in BD tumors, we selected three BD^RP cell lines (including MB436/BRCA1^X1759_splice, CAPAN1/BRCA2^V1532Sfs*2, and UWB1.289/BRCA1^D825Efs*21), and three BD^NRP cell lines (including SUM1315/BRCA1^E23Vfs*17, HCC1954/BRCA1^V1809Cfs*20, and KURAMOCHI/BRCA2^R2318*) according to their IC₅₀ values of PARPi (SI Appendix, Fig. S1 A and B). All six cell lines were indeed HRD based on RAD51 foci formation and HR reporter assays (SI Appendix, Fig. S1 E–H). Similar to data obtained in BP cell lines, PTPN1 was preferentially highly expressed in BD^NRP cells (SI Appendix, Fig. S6A), and PARPi only activated the STAT3–FoxO1 axis and NADK2 expression in these cells, accompanied by selective up-regulation of TYMS and DHFR2, leading to increased mito-NADPH generation and enhanced de novo mito-dTMP biosynthesis (SI Appendix, Fig. S6 B and C). In contrast, upon PARPi exposure, BD^RP cells exhibited a pronounced reduction in total mito-dTTP, elevated uracil misincorporation into mtDNA, and mitochondrial dysfunction (SI Appendix, Fig. S6 D–I). Furthermore, we showed that the STAT3–FoxO1 transcriptional cascade drives PARPi resistance in BD^NRP context through the NADK2/TYMS/DHFR2–dTMP axis, as confirmed by genetic perturbation and rescue assays (SI Appendix, Fig. S6 J–AE).

To explore if STAT3–FoxO1 axis contributes to acquired PARPi resistance in BP and BD cells, we utilized A2780^RP (BP) and CAPAN1^RP (BD) cell lines to generate their PARPi-resistant derivatives (A2780^NRP-A and CAPAN1^NRP-A) by long-term treatment with olaparib (SI Appendix, Fig. S1 A and B). Strikingly, both resistant cell lines exhibited marked activation of the STAT3–FoxO1 axis and NADK2, TYMS, and DHFR2 expression compared to their sensitive counterparts, resulting in enhanced mito-NADPH production and augmented de novo mito-dTMP biosynthesis (SI Appendix, Fig. S6 AF–AH). Further genetic perturbation and rescue experiments demonstrated that the STAT3–FoxO1 transcriptional cascade confers acquired PARPi resistance via the NADK2/TYMS/DHFR2–dTMP axis (SI Appendix, Fig. S6 J–AE).

Notably, we observed a difference in terms of the contribution of PTPN1 in co-opting the activation of STAT3–FoxO1 signaling between the acquired and de novo PARPi resistance contexts. In both BP^NRP and BD^NRP cells (SI Appendix, Figs. S5C and S6AI), silencing PTPN1 robustly activated STAT3–FoxO1 axis. By contrast, PTPN1 depletion in BP^NRP-A or BD^NRP-A cells failed to do so. The precise alternative mechanisms driving the activation of STAT3–FoxO1, independent of PTPN1, in tumors with acquired PARPi-resistance warrant further investigation.

FoxO1 and PTPN1 Expression Predicts PARPi Responsiveness.

We were curious whether expression levels of PTPN1 or FoxO1 could predict clinical responses to PARPi. Analysis of data from TCGA and GTEx databases revealed underexpression of FoxO1 and PTPN1 transcripts in human ovarian tumor tissues (SI Appendix, Fig. S7A). Unfortunately, the correlation between gene expression and PARPi outcomes could not be analyzed utilizing clinical databases because of data unavailability. We therefore carried out a retrospective analysis of 208 ovarian cancer patients receiving clinical PARPi therapy (Dataset S4, including 120 HRD and 88 HRP patients), where pretreatment tissue samples were evaluable for IHC analysis of FoxO1 and PTPN1 expression (SI Appendix, Fig. S7B). As expected, HRD patients showed a better prognosis, compared to HRP patients (SI Appendix, Fig. S7C). Interestingly, we noticed that 53 out of 88 HRP patients displayed prolonged benefit from PARPi (SI Appendix, Fig. S7D). Further survival analysis indicated that low expression of FoxO1 or PTPN1 positively correlated with favorable survival in HRP patients (Fig. 5 A and B). Given our findings that PARPi can upregulate FoxO1 expression preferentially in PTPN1^high tumor cells (SI Appendix, Fig. S5 L–N), we further stratified FoxO1^low HRP patients as PTPN1^high and PTPN1^low groups by IHC. Patients with PTPN1^low tumors showed a better prognosis (Fig. 5C). Similar results were obtained in HRD patients receiving PARPi therapy (SI Appendix, Fig. S7 E–H), indicating that PTPN1/FoxO1 expression might be used as potential predictive biomarkers, independent of HR status.

Fig. 5.

STAT3 inhibitor elicits synthetic lethality with PARPi. (A–C) Kaplan–Meier survival analysis of the indicated groups. The number of patients at risk is listed below the survival curves for each time point. P values were calculated by the log-rank test. (D) Images (Left panel) and volumes (Right panel) of tumors collected from the indicated PDXs treated with stattic (10 mg/kg), olaparib (50 mg/kg), dTMP (3.2 mM), or a combination of two or three of the agents were shown. Tumor volume data were represented as mean ± SEM of n = 10 or 7 mice per group.

Previous studies have revealed that resistance to platinum-based chemotherapy is a strong predictor of PARPi resistance (59–61). We therefore analyzed TCGA data from ovarian serous adenocarcinoma patients who underwent platinum-based chemotherapy. Survival analysis revealed a positive correlation between low FoxO1/PTPN1 expression and PFS (SI Appendix, Fig. S7I), further highlighting that simultaneous measurement of PTPN1 and FoxO1 expression in pre-PARPi treatment samples may provide a more precise predictive value.

We next assessed the clinical relevance of STAT3–FoxO1 axis in recurrent PARPi-resistant ovarian cancer using 12 paired treatment-naive and post-PARPi-resistant HGSOC specimens from 6 patients (SI Appendix, Fig. S7J). WES analysis showed that all patients were BRCA1/2 WT except patient#6, who carried a BRCA1 nonsense mutation. Patients #1, #3, and #6 exhibited de novo PARPi resistance and displayed high basal PTPN1 expression in pretreatment samples (SI Appendix, Fig. S7 K and L), whereas Patients #2, #4, and #5 developed acquired resistance to PARPi and showed low basal PTPN1 expression prior to therapy. Intriguingly, elevated phospho-STAT3 and FoxO1 expression were detected by IHC in all post-PARPi tumor samples, and among them, 5 patients (Patient#1, #2, #4, #5, #6) displayed a selective up-regulation of NADK2, DHFR2, or TYMS analyzed by RNA-seq (SI Appendix, Fig. S7M). These results highlight a tight correlation between the activation of STAT3–FoxO1 axis–dependent transcriptional signature and clinical unresponsiveness in HGSOC patients treated with PARPi.

STAT3 Inhibitor Elicits Synthetic Lethality with PARPi.

We speculated that blunting STAT3 activation should elicit an effective response to PARPi in nonresponsive tumors. Stattic, a nonpeptidic small molecule that has been shown to selectively inhibit the function of the STAT3 SH2 domain (62), remarkably inhibited PARPi-triggered STAT3–FoxO1 axis and its transcriptional targets, and blocked the increased mito-NADP(H) pool, as well as de novo mito-dTMP synthesis, leading to mitochondrial disorders and synergistic cytotoxicity, in a FoxO1-dependent manner (SI Appendix, Fig. S8 A–I). Furthermore, synergistic effects were observed between stattic and PARPi in a series of BP^NRP, BD^NRP, BP^NRP-A, and BD^NRP-A cells that displayed PARPi-inducible STAT3–FoxO1 axis (SI Appendix, Fig. S8J). Consistently, Danvatirsen, a phase II STAT3 inhibitor that functions as an antisense oligonucleotide-mimetic agent promoting STAT3 degradation, recapitulated the synergistic effects of Stattic (SI Appendix, Fig. S8 K–S).

To test the synergistic cytotoxic effects in vivo, we assessed the efficacy of combination therapy utilizing eight PDX models across diverse human cancer types (SI Appendix, Fig. S1 M and N), including both HRD and HRP tumors with de novo or acquired resistance. These models exhibited hyperactivation of the STAT3–FoxO1 axis and NADK2 expression (SI Appendix, Figs. S5M and S8T), along with selective upregulation of TYMS and DHFR2 following PARPi exposure or resistance acquisition. Treating these PDX models with olaparib or stattic alone had little effect on tumor volume (Fig. 5D and SI Appendix, Fig. S8 U and V). However, combined treatment significantly ablated tumor growth, without obvious influence on the body weight of mice (SI Appendix, Fig. S8W). Meanwhile, IHC was conducted to assess the status of STAT3–FoxO1 axis in these PDX tumors (SI Appendix, Fig. S8X). When combined with olaparib, stattic not only profoundly blocked STAT3–FoxO1 axis but also resulted in largely decreased expression of NADK2, TYMS, and DHFR2 by qRT-PCR (SI Appendix, Fig. S8Y). Notably, in vivo replenishing dTMP restored tumor growth (Fig. 5D and SI Appendix, Fig. S8U), and significantly rescued mitochondrial disorders caused by combination therapy (SI Appendix, Fig. S8 Z–AC). Taken together, these data clearly demonstrate the combinational synergy of stattic and PARPi in treating multiple human cancer types with both acquired and de novo resistance to PARPi.

Discussion

Growing evidence has suggested that PARPis are beneficial for patients without BRCA1/2 mutations (5–18, 21, 22), advocating the potential utility of PARPis to a broad range of patients. Here, we report that increased de novo dTMP synthesis in the mitochondria triggered by PARPi-activated STAT3–FoxO1 axis not only drives the intrinsic unresponsiveness to PARP inhibition (SI Appendix, Fig. S9) but also promotes the acquired PARPi resistance, in both BP and BD tumors.

Previous findings have suggested that depleting cellular NAD⁺ can sensitize cells to PARPi (33). Given our observation that PARPi-induced NAD⁺ accumulation was retained in BP^RP cells but declined in BP^NRP cells following 3 d of PARPi treatment, accompanied by attenuated proliferation only in BP^RP cells, we propose that cellular NAD⁺ abundance may not necessarily dictate PARPi responsiveness. Instead, PARPi responsiveness in BP tumors is primarily determined by NADK2-mediated conversion of NAD⁺ to mito-NADP(H). These PARPi-induced NADP(H) fluxes, exclusively occurring in BP^NRP tumor cells, introduce a window of hypervulnerability to NADK2 inhibition. It is conceivable that NAD⁺ depletion (i.e., by FK866, a NAMPT inhibitor) may potentially inhibit the generation of NAD⁺ derivatives, including mito-NADP(H), which ultimately sensitizes tumor cells to PARPi. However, given the prominent and universal role of NAD⁺ as a cofactor for cellular metabolism and energy production, NAD⁺ depletion may interfere with normal physiological functions. Indeed, the decline in NAD⁺ levels is causally linked to numerous aging-associated diseases (26). Our study demonstrates that targeting FoxO1-NADK2 signaling, which is specifically up-regulated in PARPi-nonresponsive tumor cells, may help minimize intolerable side effects compared to NAD⁺ depletion in PARPi-based therapy.

Mitochondrial dNTP pool is essential for the faithful replication of mtDNA. Loss-of-function mutations in genes involved in mitochondrial nucleotide synthesis have been linked to mtDNA-depletion syndrome (MDS) (63, 64). Exogenous supplementation of nucleosides or nucleotides is a promising pharmacologic strategy for treating MDS caused by impaired mito-dNTP pools (45–48). Nucleotides are thought to be impermeable to cells. Noticeably, previous studies have postulated that orally administered nucleotides are rapidly catabolized to cell-permeable nucleosides in the small intestine by nucleotidases (45–47), which are located on the plasma membrane. Utilizing a [¹⁵N₂]dTMP tracing assay, we provide direct evidence that exogenously supplied dTMP can replenish cellular and mitochondrial dTMP levels. Based on this, we found that PARPi-triggered cytotoxicity in BP^RP cells can be alleviated by the administration of dTMP or thymidine, which restores mito-dTTP levels. These findings highlight that genetic ablation or pharmacological inhibition of mito-dTMP synthesis may serve as a potential therapeutic strategy to overcome intrinsic PARPi resistance.

We demonstrate that PARPi activates STAT3 via impaired dephosphorylation, rather than the enhanced phosphorylation process. The ablated ADPRylation of PTPN1 triggered by PARPi shifts the balance between phosphorylation and dephosphorylation toward the former. It is conceivable that PARPi may regulate PTPN1 through other mechanisms beyond ADPRylation, such as altering cofactor availability or protein–protein interactions. Further studies, including structural analysis, will be required to determine these possibilities.

Materials and Methods

Detailed Methods Description.

All the materials and methods conducted in this study are detailed in SI Appendix. This includes the following: cell culture, plasmid constructs and lentiviral infection, antibodies and reagents, immunoprecipitation and immunoblotting, quantitative RT-PCR, immunofluorescence, live-cell fluorescence measurement using the microplate reader, quantification of mtDNA copy numbers, clinical HGSOC samples collection, immunohistochemistry, electron microscopy, Mito-IP, Measurement of NAD(H) and NADP(H) using bioluminescent quantification assays, ATP level measurement by luminescence, SIRT activity assay, targeted metabolic flux analysis, measurement of uracil content in DNA, measurement of OCR, in vitro ADPRylation assay, identification of ADPRylation sites on PTPN1, in vivo ADPRylation assay, phosphatase assay, clonogenic assay, homologous recombination reporter assay, tagging of endogenous FoxO1 using CRISPR/Cas9-Mediated homologous directed repair (HDR), enrichment of dCas9–HA interaction proteins for proteomic analysis, cell labeling by SILAC and LC–MS/MS analysis, chromatin immunoprecipitation and sequencing, bulk RNA-seq analysis, whole exome libraries preparation and sequencing, whole-exome sequencing (WES) data analysis, survival analysis, comparing gene expression between normal and tumor tissues, animal studies, and statistical Analysis.

Study Design and Approval.

All mouse experiments were approved by the Institutional Animal Care and Use Committee of Xiamen University (XMULAC 20170331 and XMULAC20200227). A total of 208 ovarian cancer tissue samples were obtained from Peking University Cancer Hospital and Institute (2019KT100) and Sun Yat-sen University Cancer Center (B2022-422-01). All samples were deidentified prior to their use in this study. The requirement for informed consent was waived by the ethics committees due to the retrospective nature of the study and the use of anonymized clinical samples.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Dataset S05 (XLSX)

Data, Materials, and Software Availability

The RNA-seq, ChIP-seq, and Whole-exome seq generated in this study have been deposited in the Gene Expression Omnibus database under accession code GSE278907 (65), GSE278908 (66), and SUB14773295 (67). The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the iProX partner repository with the dataset identifiers PXD036386 (68) and PXD040255 (69). All other data are included in the manuscript and SI Appendix.