“Cysteine depletion induces pancreatic tumor ferroptosis in mice”

Abstract

Ferroptosis is a form of cell death resulting from the catastrophic accumulation of lipid reactive oxygen species (ROS). Oncogenic signaling elevates lipid ROS production in many tumor types and is counteracted by metabolites derived from the amino acid cysteine. Here we show that the import of oxidized cysteine (cystine) via system x_C^– is a critical dependency of pancreatic ductal adenocarcinoma (PDAC), a leading cause of cancer mortality. Cysteine was used to synthesize glutathione and coenzyme A which combined to control ferroptosis. Studying genetically engineered mice, we found that deletion of a system x_C^– subunit, Slc7a11, induced tumor-selective ferroptosis and inhibited PDAC growth. This was replicated through administration of cyst(e)inase, a drug that depletes cysteine/cystine, demonstrating a translatable means to induce ferroptosis in PDAC.

One sentence summary:

Genetic and pharmacological targeting of cystine import induces cancer-selective ferroptosis in pancreatic tumors of genetically engineered mice.

Pancreatic ductal adenocarcinoma (PDAC) is a deadly cancer that is resistant to traditional therapies. Over 90% of PDAC cases harbor mutations in KRAS that both promote proliferation and alter cellular metabolism. A byproduct of mutant KRAS signaling is the increased production of reactive oxygen species (ROS), which can damage cellular components. To compensate, PDAC cells upregulate metabolic programs that detoxify ROS using cysteine-derived metabolites such as glutathione (GSH) (1). Most cellular cysteine is acquired through the system x_C^– antiporter, which exchanges extracellular, oxidized cysteine (cystine) for intracellular glutamate. Yet, germline deletion of the system x_C^– gene, Slc7a11, is well tolerated in unstressed mice (2), suggesting normal cells have low basal cystine import requirements. We hypothesized that cystine import is a critical dependency of PDAC that may be selectively targeted as an anticancer therapy.

To investigate the role of cysteine metabolism in PDAC, we measured the viability of human PDAC cell lines cultured for 24 hours in media with varying concentrations of cystine or the system x_C^– inhibitor imidazole ketone erastin (IKE) (Fig. 1, A and B) (3). In four of five PDAC lines, cystine starvation reduced cell viability by >80%; this was largely prevented by addition of the lipophilic antioxidant Trolox. Cystine-starved cells underwent catastrophic destabilization of their plasma membranes, without visual evidence of nuclear fragmentation (movie S1). IKE treatment mimicked the effects of cystine withdrawal, quickly killing most cells from the four sensitive PDA lines in a manner visually identical to that of cystine starvation, but distinct from staurosporine-induced apoptosis (fig. S1A, movies S2 and S3). Neither cystine starvation nor system x_c^– inhibition (collectively referred to as “cysteine depletion”) induced caspase 3 cleavage (fig. S1B), indicating the cell death was not apoptotic. Rather, the oxidative cell death resembled ferroptosis, a form of iron-dependent, non-apoptotic cell death previously associated with system x_C^– inhibition (4). We found that co-treatment of human PDAC cells with either deferoxamine (DFO, an iron chelator), ferrostatin-1 (Fer1, a ferroptosis inhibitor), or N-acetylcysteine (NAC, a cell permeable analog of cysteine) significantly reduced cell death from cysteine depletion, whereas inhibitors of apoptosis or necroptosis had little impact on cell death, consistent with previous reports (Fig. 1C; fig. S1, C to E) (5). Autophagy inhibition had variable effects in different lines, likely reflecting the known impact of ferritinophagy on ferroptosis (6). Using the fluorescent probe C11-BODIPY, we observed a large increase in lipid oxidation (a hallmark of ferroptosis) prior to cell death in response to cysteine depletion; this was prevented by co-treatment with Trolox, Fer-1, NAC, and DFO (Fig. 1D; fig. S2, A and B). By contrast, elevated total ROS levels induced by cysteine depletion were not prevented by these agents, arguing against a more general oxidative process (Fig. S2C). We conclude from these experiments that most PDAC lines rely on cysteine to prevent ferroptotic cell death. An analysis of SLC7A11 expression across human datasets revealed a modest overexpression in PDAC versus normal tissues, enrichment in the malignant epithelial compartment of PDAC, and an association with signatures of redox stress (fig. S3). Across multiple human cancers, SLC7A11 was frequently overexpressed and associated with reduced survival (fig. S4).

Fig. 1. Pancreatic cancer cells require exogenous cystine to avert ferroptosis.

(A and B) Viability of human PDAC lines after 24 hours culture in varying concentrations of cystine (A) or IKE (B), alone or in combination with 100 μM Trolox. Student’s t-test comparing maximal cytotoxicity ± Trolox. (C) Viability of PANC-1 cells cultured for 24 hours in cystine-free media or treated with 10 μM IKE, alone or in combination with 100 μM Trolox (Tro), 500 nM ferrostatin-1 (Fer1), 100 μM deferoxamine (DFO), 1mM N-acetyl cysteine (NAC), 50 μM ZVAD-FMK, 1 nM Bafilomycin A1 (BA1) or 10 μM Necrostatin-1s (Nec1s). Tukey test. (D) Flow cytometry of C11-BODIPY fluorescence in PANC-1, AsPC-1, BxPC-3, and S-2013 cells after 6–8 hours of treatment with conditions from panel C. Tukey test. All data are means ±SEM of three independent experiments. * p < 0.05, x = no significant difference.

To learn whether pancreatic tumors in mice depend on system x_c^– for survival, we employed a dual recombinase genetic engineering strategy based on the “KPC” mouse model (7). Kras^FSF.G12D/+; Tp53^R172H/+; Pdx1FlpO^tg/+; Slc7a11^Fl/Fl; Rosa26^CreERT2/+ (KPFSR) mice (fig. S5) spontaneously develop PDAC driven by FlpO-dependent activation of mutant Kras and germline expression of mutant Tp53. These tumors are identical in genotype and phenotype to the KPC model, but administration of tamoxifen induces systemic deletion of Slc7a11 through the action of Cre recombinase expressed from the Rosa26 locus (figs. S5 and S6, A to C). We randomized KPFSR mice bearing 4–7mm tumors to receive six daily doses of vehicle or tamoxifen and monitored tumor growth by ultrasound (8). We found that deletion of Slc7a11 in established tumors of KPFSR mice nearly doubled median survival compared to vehicle treatment (Fig. 2A, p = 0.0295, n = 20). Most recombined tumors exhibited a period of stable disease or partial response—one underwent a complete regression; these responses were never observed in vehicle-treated mice (Fig. 2B; figs. S6, D to F and S7A). Critically, addition of NAC to the drinking water of tamoxifen-treated mice restored baseline survival and eliminated tumor responses, supporting a link to cysteine metabolism (Fig. 2, A and B). At necropsy, escaped tumors exhibited evidence of incomplete Slc7a11 recombination by PCR and restored protein expression by Western blotting, suggesting the outgrowth of unrecombined tumor cells (fig. S7, B to C).

Fig. 2. Deletion of Slc7a11 in KPC mice induces tumor ferroptosis and extends survival.

(A) Survival of KPFSR mice treated with vehicle (n = 11, median 15 days), tamoxifen (n = 9, median 29 days), or tamoxifen/NAC (n = 5, median 17 days). * p < 0.0295, log-rank. INSET: Survival of KPC mice treated with NAC alone (n = 8, median = 16 days) versus historical saline-treated controls (n = 10, median = 11 days). (B) Growth curves for each KPFSR tumor. (C) Hematoxylin and eosin (H&E) stained sections of tumor tissue from KPFSR mice treated with vehicle or tamoxifen. L = lumen of malignant epithelium, N = necrosis, yellow arrowheads = lipid droplets, black arrowheads = megamitochondria. Scale = 20 μm. (D and E) TEM images from tamoxifen-treated KPFSR tumors. LD = lipid droplets, N = nucleus, arrow indicates damaged mitochondrion. (D) scale = 1μm; (E) scale = 100 nm. (F) Gene set enrichment analysis. Top panel depicts enrichment of a published ferroptosis expression signature (Dixon) among genes differentially expressed in tamoxifen-treated KPFSR epithelia (Badgley) (p < 0.001). Bottom panel depicts the reciprocal comparison (p < 0.006).

The study of in vivo ferroptosis has been hindered by the lack of a validated, selective biomarker and the absence of a histopathological characterization of the phenomenon in tissues. Within tamoxifen-treated KPFSR tumors, we observed numerous lesions of ballooned epithelial cells with lipid droplet-like structures and intermittent megamitochondria, often juxtaposed to necrotic regions—a phenotype only occasionally observed in vehicle-treated KPFSR and untreated KPC tumors (Fig. 2, C to D; fig. S7C). These lesions exhibited no alterations in apoptosis or proliferation markers (fig. S8, A and B), but did display accumulation of 4-hydroxynonenal (4HN) (fig. S8, C to E), a byproduct of lipid peroxidation, making them candidates for in vivo ferroptosis. Critically, no pathologies were observed in non-pancreatic tissues of tamoxifen-treated KPFSR mice, indicating a tumor-selective phenotype. Transmission electron microscopy (TEM) and Oil Red O staining of tamoxifen-treated KPFSR tumors confirmed the presence of abnormally large lipid droplets (Fig. 2D and fig. S8F). TEM also revealed structural aberrations in mitochondria of malignant epithelial cells, including disrupted cristae and compromised membrane integrity (Fig. 2E), consistent with the results of prior in vitro studies (9). Finally, we performed laser capture microdissection and RNA-sequencing to isolate malignant epithelial cells from KPFSR tumors and found that genes upregulated in response to Slc7a11 deletion were enriched in a ferroptotic expression signature from erastin-treated HT-1080 cells (Fig. 2F) (10); apoptotic gene sets were not enriched (table S1). We conclude that the phenotype observed in tamoxifen-treated KPFSR tumors is a histologically identifiable, in vivo manifestation of ferroptosis.

Prior studies have indicated that cysteine regulates ferroptosis primarily through the synthesis of glutathione, a critical cofactor for the lipid peroxide-detoxifying enzyme GPX4 (3). Indeed, we found that cysteine depletion rapidly reduced GSH levels in two human PDAC cell lines (Fig. S9A). Furthermore, co-treatment with the membrane-permeable GSH analog glutathione ethyl ester (GSH-EE) prevented lipid oxidation and ferroptosis (fig. S9, B to E). However, inhibition of GSH biosynthesis using buthionine sulfoximine (BSO) (fig. S9F) did not induce lipid ROS or reduce cell viability (Fig. 3, A and B), demonstrating that GSH loss is not sufficient to induce ferroptosis in PDAC cells (11). To investigate whether additional cysteine-derived metabolites contribute to the regulation of ferroptosis, we traced the metabolism of exogenous cystine by using ¹³C-labelled cystine and measuring labelled metabolites by mass spectrometry. In addition to exhibiting rapid flux into glutathione pools, cystine was also converted to coenzyme A (CoA) over 24 hours (Fig. 3 C and D); no flux was observed into taurine, lactate, citrate, or glutamate (fig. S9G). CoA is synthesized from cysteine via the pantothenate pathway and plays a role in many metabolic pathways, particularly lipid metabolism. Both CoA and its derivative coenzyme Q₁₀ (CoQ₁₀) have been shown to impact sensitivity to ferroptosis (12, 13). We found that system x_C^– inhibition reduced CoA levels and increased levels of pantothenate (Fig. 3E), a metabolite upstream of cysteine incorporation in CoA synthesis. Moreover, treatment of PDAC cells with exogenous CoA (14) prevented IKE-induced ferroptosis (Fig. 3F) whereas pantothenate kinase inhibition with PANKi sensitized cells to IKE (fig. S10, A to C). Strikingly, PANKi combined synergistically with BSO to induce ferroptosis (Fig. 3G and fig. S10D). Co-treatment with idebenone (a membrane-permeable analog of CoQ₁₀) or a monounsaturated fatty acid blocked BSO/PANKi induced ferroptosis, whereas saturated or poly-unsaturated fatty acids did not (fig. S10E), similar to prior observations with IKE (15). Together, these experiments demonstrate that CoA and GSH cooperate to regulate ferroptosis in human PDAC cells (fig. S10H).

Fig. 3. Combination GSH and CoA inhibition induces ferroptosis in human PDAC cells.

(A) Flow cytometry for C11-BODIPY fluorescence in four human PDAC lines treated for 6 hours with 150 μM BSO. Paired t-test. (B) Viability of human PDAC cells treated for 24 hours with indicated concentrations of BSO. (C) Liquid chromatography time-of-flight mass spectrometry (LC-TOF-MS) analysis of GSH in PANC-1 cells labelled for 6 hours with ¹³C-cystine combined with vehicle or 5μM IKE. Student’s t-test. (D) LC-TOF-MS analysis of CoA in PANC-1 cells labelled for 6 hours with ¹³C-cystine, after 6 or 24 hours. Student’s t-test. (E) Liquid chromatography triple quadrupole tandem mass spectrometry (LC-QqQ-MS) measurements of CoA and pantothenate (Pant.) levels in Panc-1 cells treated with vehicle and IKE for six hours. Student’s t-test. (F) Viability of human PDAC cell lines treated with IKE, alone or in combination 200 μM CoA. Student’s t-test comparing maximal cytotoxicity ± CoA. (G) PANC-1 cells treated for 24 hours with combinations of 300 μM BSO and 5 μM PANKi along with Trolox, Fer-1, DFO, or CoA as described in Fig. 1C. Tukey test. In all panels, * p < 0.05, x = not significantly different. In panels A, B, E, and F, data are means ±SEM from three independent experiments. In panels C and D, data are means ± SD from three biological replicates. In (E), * p < 0.05 comparing maximal cytotoxicity in CoA treated vs. untreated conditions for each line, Student’s t-test. In (F), * p < 0.05, one-way ANOVA with posthoc Tukey test.

Finally, we sought a pharmacological means to target cysteine metabolism in pancreatic tumors. Drug delivery is compromised in PDAC due to the effects of fibrosis on tissue perfusion (16). Although current system x_C^– inhibitors are not optimized for the PDAC microenvironment, the engineered enzyme cyst(e)inase is well-tolerated in mammals, has a long half-life, and potently degrades both cystine and cysteine in circulation (17). In vitro, cyst(e)inase treatment induced lipid oxidation and reduced the viability of IKE-sensitive PDAC lines; this was largely prevented by co-treatment with ferroptosis inhibitors (Fig. 4, A to C; fig. S11, A and B). To determine the effects of cyst(e)inase on pancreatic tumors in vivo, we treated tumor-bearing KPC mice for 10 days with vehicle, low-dose cyst(e)inase, or high-dose cyst(e)inase (n=2 each). Histopathological examination of cyst(e)inase-treated tumors revealed a severe ferroptosis phenotype, with extensive lipid droplet formation, stromal disruption, decompressed blood vessels, and necrosis (Fig. 4D and E; fig. S12, A to B). TEM revealed enlarged lipid droplet formation, extracellular lipid droplets, and mitochondrial defects, preferentially in cyst(e)inase-treated KPC tumors (Fig. 4F to I). Ferroptotic lesions were generally 4HN positive and cleaved caspase 3 negative (fig. S13A). Finally, four additional KPC mice were treated with high-dose cyst(e)inase and their tumor growth was monitored by ultrasound. Notably, all four tumors exhibited stabilizations or regressions whereas historical vehicle-treated controls never stabilized (Fig. 4J). Thus, we conclude that the therapeutic depletion of cysteine/cystine can induce ferroptosis in Kras/p53 mutant pancreatic tumors in mice.

Fig. 4. Cyst(e)inase treatment induces tumor-selective ferroptosis in KPC mice.

(A) Viability of human PDAC lines cultured with varying concentrations of cyst(e)inase for 48 hours (AsPC1) or 72 hours (PANC-1, BxPC3, S2–013). (B) Viability of AsPC1 cells treated with 90 nM cyst(e)inase (C^–ase) for 72 hours, alone or in combination with indicated agents, under conditions described in Fig. 1C. (C) C11-BODIPY fluorescence was measured by flow cytometry in AsPC-1 cells after 24 hours of treatment with 90 nM cyst(e)inase, alone or in combination with indicated agents, under conditions described in Fig. 1C. (D and E) H&E stained sections of pancreatic tumors from KPC mice treated with vehicle or cyst(e)inase. Yellow arrowheads indicate lipid droplets. Black arrowhead indicates megamitochondrion. Bars = 20 μm. (F–I) TEM of pancreatic tumors from the KPC model treated with vehicle (F and H) or cyst(e)inase (G and I). Red arrowheads indicate mitochondria; LD = lipid droplets; N = nucleus. Bars = 1 μm. (J) Tumor growth curves from KPC mice treated with saline (historical controls) or 100 mg/kg cyst(e)inase, q.2.d., i.p.. For panels A–B, data depict mean of 3 biological replicates. * p < 0.05, x = not significant.

In summary, our data add to a growing body of evidence showing that certain cancers, including PDAC, rely on cysteine metabolism to avert ferroptosis. Previously, SLC7A11 deletion via CRISPR-Cas9 was shown to induce ferroptosis in cultured PDAC cells and slow xenograft engraftment and growth (18), and system x_C^– inhibition was shown to limit the growth of lymphoma xenografts, inducing a lipid oxidative signature and other indicators of ferroptosis (15). However, cysteine depletion in a PDAC xenograft model had little effect on tumor growth (19), perhaps indicating that the nutrient-deprived, hypoxic microenvironment of autochthonous pancreatic tumors may contribute to the tumor-selective cysteine dependency we observed in genetically engineered mouse models of PDAC. While it is not yet known whether human PDAC is also susceptible to ferroptosis from cysteine depletion, the clinical development of cyst(e)inase for treatment of the metabolic disorder cystinuria provides a pathway for future translation of this concept.