Since the ER membrane and mitochondrial membrane are in close contact during cell stresses (Bravo-Sagua et al., 2013), we first investigated whether the ER protein STING1 can be translocated to mitochondria in PANC1, a human pancreatic ductal adenocarcinoma (PDAC) cell line that expresses STING1 and is sensitive to ferroptosis (Zhu et al., 2017; Li C. et al., 2020). We used a classic ferroptosis activator, erastin, which not only binds to the amino acid antiporter system xc^– in the plasma membrane to inhibit cystine uptake with a 1:1 counter-transport of glutamate (Dixon et al., 2012), but also binds to voltage-dependent anion channels (VDACs) on the mitochondria (Yagoda et al., 2007), thereby increasing the permeability of the mitochondrial membrane and oxidative damage. Western blot analysis observed that after erastin treatment, the accumulation of STING1 in mitochondria increased, whereas STING1 expression in ER decreased (Figure 1A). Consistently, immunofluorescence analysis showed that erastin increased the localization of STING1 in the mitochondria of PANC1 cells (Figure 1B).

To define the mechanism of STING1 mitochondrial translocation, we treated PANC1 cells with sulfasalazine, which is a classic system xc^– inhibitor with activation that induces ferroptosis (Dixon et al., 2012). Unlike erastin, sulfasalazine failed to induce mitochondrial translocation of STING1 (Figure 1A), indicating that inhibiting system xc^– is not essential for mitochondrial translocation of STING1 during ferroptosis. Similarly, erastin (but not sulfasalazine) induced STING1 mitochondrial translocation in another human PDAC cell line (Capan2) (Figure 1A). As a control, we measured the inhibition of glutamate release and found that erastin and sulfasalazine both inhibited system xc^– activity in PANC1 and Capan2 cells (Figure 1C).

Since erastin can cause mitochondrial damage (Dixon et al., 2012), we further explored whether erastin-induced STING1 mitochondrial translocation requires a mitochondrial reactive oxygen species (mtROS) signal. To test this hypothesis, we used MitoTEMPO and MitoQ, which are mitochondria-targeted superoxide dismutase mimetic with superoxide and alkyl radical scavenging properties (Liang et al., 2010). MitoTEMPO and MitoQ inhibited erastin-induced mtROS production (Figure 1D), the accumulation of STING1 in the mitochondria (Figures 1E,F), lipid peroxidation (Figure 1G), and cell death (Figure 1H). It is worth noting that staurosporine, a classic apoptosis inducer, failed to induce STING1 mitochondrial translocation (Figure 1I). Collectively, these findings indicate that erastin induces mitochondrial translocation of STING1 by activating mitochondrial oxidative stress.

Consistent with previous findings that STING1 is a positive regulator of zalcitabine-induced ferroptosis (Li C. et al., 2020), the knockdown of STING1 by two different shRNA (termed as STING1^KD) (Figure 2A and Supplementary Figure 1) also limited erastin-induced cell death in PANC1 cells (Figure 2B and Supplementary Figure 1). Zalcitabine acts as a nucleoside deoxycytidine analog to induce ferroptosis by activating mitochondrial DNA damage (Li C. et al., 2020). However, erastin failed to cause significant mitochondrial DNA damage (Figure 2C) but did increase the production of mtROS (Figure 2D). Moreover, the knockdown of STING1 failed to affect zalcitabine-induced mitochondrial DNA damage (Figure 2C). These findings indicate that STING1-mediated erastin-induced ferroptosis may not be dependent on the mitochondrial DNA damage response. Moreover, STING1 failed to affect sulfasalazine-induced cell death (Supplementary Figure 1).

To define the mechanism of action of STING1 in ferroptotic cell death, we further analyzed the mtROS production and mitochondrial morphology (by staining with translocase of outer mitochondrial membrane 20 [TOMM20/TOM20]) of mitochondria in control and STING^KD cells. Erastin-induced mtROS production was inhibited in STING1^KD cells (Figure 2D), indicating that STING1 regulates mtROS production during ferroptosis. Unlike mitochondrial fission, which is one of morphological events during apoptotic cell death (Youle and Karbowski, 2005), mitochondrial fusion was significantly increased in erastin-treated control cells, but not STING^KD PANC1 cells (Figure 2E). These analyses indicate a potential role of STING1-dependent mitochondrial fusion in promoting ferroptosis.

Mitochondrial fusion requires the coordinated fusion of outer and inner membranes, which is mainly driven by OPA1 mitochondrial dynamin-like GTPase (OPA1) and mitofusins (including MFN1 and MFN2), respectively (Giacomello et al., 2020). Unexpectedly, an immunoprecipitation assay showed a direct interaction between STING1, MFN1, and MFN2 in the mitochondrial fraction of PANC1 cells following erastin treatment (Figure 3A). In contrast, erastin failed to cause an interaction between STING1 and OPA1 (Figure 3A). As expected, the knockdown of MFN1 and MFN2 (hereafter MFN1/2^KD) also limited erastin-induced mitochondrial fusion in PANC1 cells (Figure 2E). These findings suggest that STING1 is a regulator of mitofusin-mediated mitochondrial fusion machinery in ferroptosis.

We next investigated whether mitofusins have a function like STING1 in the regulation of ferroptotic response. Indeed, the suppression of MFN1/2 by shRNAs (Figure 3B) also inhibited erastin-induced cell death (Figure 3C) in PANC1 cells with reduced mtROS production (Figure 3D), decreased lipid peroxidation (Figure 3E), and diminished release of high mobility group box 1 (HMGB1, a representative damage-associated molecular pattern during ferroptotic cell death (Wen et al., 2019; Figure 3F). Compared with the control group, MFN1/2^KD PANC1 cells had similar sensitivity to zalcitabine-induced ferroptotic response (Figures 3C–F), indicating that the mitochondrial fusion machinery may not be involved in the regulation of mitochondrial DNA damage-induced ferroptosis. Moreover, erastin-induced cell death was also blocked in MFN1/2^–/– mouse embryonic fibroblasts (MEFs) (Figure 3G). Compared with the knockdown of MFN1/2 or STING1 alone, double knockdown of STING1 and MFN1/2 had similar inhibitory effects on erastin-induced cell death (Figure 3H). Together, these findings support the idea that a mitochondrial STING1-MFN1/2 complex promotes ferroptosis by mitochondria fusion-mediated mtROS production and lipid peroxidation.

Since STING1 is a promoter of autophagosome formation (Gui et al., 2019), and mitophagy contributes to the clearing of damaged mitochondria (Palikaras et al., 2018), we investigated whether STING1 is involved in ferroptosis by regulating mitophagy. Mitophagy was assayed by a commercial small molecule fluorescent probe, Mtphagy Dye, which was recently developed by Dojindo Laboratories (Iwashita et al., 2017). Under normal circumstances, Mtphagy Dye accumulated in the mitochondria and its fluorescence was weak (Iwashita et al., 2017; Figure 4A). When mitophagy was induced by erastin, the damaged mitochondria fused to lysosomes, causing the Mtphagy Dye to emit a high fluorescent signal (Iwashita et al., 2017; Figure 4A). This process was inhibited by bafilomycin A1 (BafA1) (Figure 4A), which is a vacuolar H + ATPase inhibitor used to block autophagic flux by decreasing autophagosome-lysosome fusion (Li J. et al., 2020). Although STING1^KD cells had lower microtubule-associated protein 1 light chain 3 beta-II (MAP1LC3B-II, a marker of autophagosome formation) expression (Figure 4B), there was no significant difference in the upregulation of Mtphagy Dye induced by erastin compared with control cells (Figure 4A).

PINK1 is a key regulator of mitophagy in mammalian cells by parkin RBR E3 ubiquitin protein ligase (PRKN, also known as PARK2) (Geisler et al., 2010). Erastin and STING1 depletion failed to affect the expression of PINK1 and PRKN in PANC1 cells (Figure 4B). In addition, the knockdown of PINK1 or PRKN by shRNA (Figures 4C,D) promoted uncoupler carbonyl cyanide m-chlorophenylhydrazone (CCCP, a classical mitophagy inducer)-induced cell death, but failed to affect erastin-induced cell death (Figure 4E). Collectively, these findings suggest that the classical PINK1-dependent mitophagy pathway is not involved in regulating erastin-induced ferroptosis.

To determine whether the STING1-mitofusion pathway regulates the sensitivity of ferroptosis in vivo, we implanted control, STING1^KD, or MFN1/2^KD PANC1 cell lines into the subcutaneous space on the right flank of nude mice. After 1 week, these mice were administrated imidazole ketone erastin (IKE), a metabolically stable analog of erastin in vivo (Zhang et al., 2019). Compared with the control group, IKE-mediated tumor suppression in the STING1^KD or MFN1/2^KD group was limited (Figure 5A). The activity of caspase-3 (an apoptosis marker) (Figure 5B) and the mRNA expression of STING1 (Figure 5C), MFN1 (Figure 5D), and MFN2 (Figure 5E) in tumor extracts was not changed by IKE. In contrast, IKE induced the upregulation of malondialdehyde (MDA, one of the products of polyunsaturated fatty acid peroxidation) (Figure 5F). The expression of ferroptosis markers, such as acyl-CoA synthetase long chain family member 4 (ACSL4) (Yuan et al., 2016b; Figure 5G) and prostaglandin-endoperoxide synthase 2 (PTGS2) (Yang et al., 2014; Figure 5H) in tumor extracts was reduced in the STING1^KD or MFN1/2^KD group. These animal studies further support the in vitro findings that the activation of the STING-MFN1/2 pathway promotes ferroptosis-mediated tumor suppression.

Zalcitabine (S1719), erastin (S7242), sulfasalazine (S1576), IKE (S8877), MitoQ (S8978), bafilomycin A1 (S1413), staurosporine (S1421), and CCCP (S6494) were purchased from Selleck Chemicals. MitoTEMPO (SML0737) was purchased from Sigma-Aldrich. The antibodies to STING1 (13647), TOMM20 (42406), MFN1 (14739), OPA1 (67589), PINK1 (6946), PRKN (4211), ACTB (3700), CALR (12238), and MAP1LC3B (3868) were purchased from Cell Signaling Technology. The antibody to MFN2 (ab205236) was purchased from Abcam. The antibody to mitHsp70 (MA3-028) was purchased from Thermo Fisher Scientific.

The PANC1 (CRL-1469), Capan2 (HTB-80), and MFN1/2^–/– MEFs (CRL-2994) were purchased from the American Type Culture Collection. These cell lines were grown at 37°C, 95% humidity, and 5% CO₂ according to standard mammalian tissue culture protocols and aseptic techniques. In brief, PANC1 and MEFs were cultured in Dulbecco’s Modified Eagle’s Medium (Sigma-Aldrich, D6429), whereas Capan2 was cultured in McCoy’s 5a Medium Modified (Sigma-Aldrich, M4892). All media were supplemented with 10% heat-inactivated (56°C for 30 min) fetal bovine serum (Sigma-Aldrich, F4135) and 5000 units/ml of penicillin and 5000 μg/ml of streptomycin (Thermo Fisher Scientific, 15070063). The identity of the cell line was verified by short tandem repeat analysis, and routine mycoplasma testing was negative for contamination.

A PANC1-derived xenograft tumor model was established as previously described. In short, wild-type, STING1^KD or MFN1/2^KD PANC1 cells (5 × 10⁶ cells) were injected subcutaneously into the dorsal side of NOD SCID mice (female, 8–10 weeks old). On the 7th day, these mice were given IKE (40 mg/kg, once every other day, intraperitoneal injection) for 2 weeks. The diameter of the tumor was measured twice a week with a caliper, and the tumor volume was calculated using the following formula: length × width² × π/6. At 14 days after the injection, the mice were euthanized, and the xenograft solid tumors were collected. All mice were maintained under specific pathogen-free conditions. All animal experiments were conducted in accordance with the institutional ethical guidelines related to animal care and were approved by the Institutional Animal Health and Use Committee.

A Countess II FL Automatic Cell Counter (Thermo Fisher Scientific, AMQAF1000) was used to determine the percentage of dead cells after staining with 0.4% trypan blue solution (Sigma-Aldrich, T8154). Trypan blue is a negatively charged dye that can only stain cells with damaged cell membranes, thereby indicating cell death.

The following human shRNAs were obtained from Sigma-Aldrich:STING1 shRNA (sequence: CCGGGCTGGCATGGTCA TATTACATCTCGAGATGTAATATGACCATGCCAGCTTTTT TG), STING1 shRNA-2 (sequence: CCGGGCTGTATA TTCTCCTCCCATTCTCGAGAATGGGAGGAGAATATACAG CTTTTTTG), MFN1 shRNA (sequence: CCGGGCTCCCATT ATGATTCCAATACTCGAGTATTGGAATCATAATGGGAGC TTTTTG), MFN2 shRNA (sequence: CCGGGTCAAAGG TTACCTATCCAAACTCGAGTTTGGATAGGTAACCTTTGAC TTTTTG), PINK1 shRNA (sequence: CCGGCGGACG CTGTTCCTCGTTATGCTCGAGCATAACGAGGAACAGCGT CCGTTTTTTG), and PRKN shRNA (sequence: CCGGGGCCT ACAGAGTCGATGAAAGCTCGAGCTTTCATCGACTCTGTA GGCCTTTTTG). Lipofectamine 3000 (Thermo Fisher Scientific, L3000001) was used for shRNA transfection according to manufacturer’s guidelines. In short, when the cells reached 70–80% confluence at the time of transfection, the DNA-lipid complex was added to the cells using serum-free Opti-MEM medium. After 48–72 h of transfection, the efficiency of RNAi was detected by qRT-PCR or western blotting.

According to the manufacturer’s instructions, we used an QIAGEN RNeasy Plus Micro Kit (74034) and iScript cDNA Synthesis Kit (Bio-Rad, 1708890) to extract total RNA and synthesize first-strand cDNA, respectively. We performed qRT-PCR using synthetic cDNA and primers (STING1: 5′-CCTGAGTCTCAGAA CAACTGCC-3′ and 5′-GGTCTTCAAGCTGCCCACAGTA-3′; PTGS2: 5′-CGGTGAAACTCTGGCTAGACAG-3′ and 5′-GCAAACCGTAGATGCTCAGGGA-3′; ACSL4: 5′-GCTATCT CCTCAGACACACCGA-3′ and 5′-AGGTGCTCCAACTCTGC CAGTA-3′; PINK1: 5′-GTGGACCATCTGGTTCAACAGG-3′ and 5′-GCAGCCAAAATCTGCGATCACC-3′; and PRKN: 5′-CCAGAGGAAAGTCACCTGCGAA-3′ and 5′-CTGAGGCTTC AAATACGGCACTG-3′). The data were normalized to RNA18S (5′-CTACCACATCCAAGGAAGCA-3′ and 5′-TTTTTCGTCACTACCTCCCCG-3′). Based on the untreated group, the relative concentration of mRNA was expressed in arbitrary units, and its assigned value was 1.

Mitochondrial DNA damage was assayed using a kit from Thermo Fisher Scientific (50-753-4394) according to the manufacturer’s instructions. This DNA damage detection kit was used to determine the damaged 8.1–8.8 kb mitochondrial DNA sequences by quantification of the replicated DNA with real-time PCR following qPCR analysis.

After treatment, whole cells or organelle fractions were harvested and lysed at 4°C in ice-cold Cell Lysis Buffer (Cell Signaling Technology, 9803) containing a protease inhibitor cocktail (Sigma-Aldrich, P0044) (Tang et al., 2010). A bicinchoninic acid (BCA) assay was used to detect protein concentration, and then 30 μg of protein in each lysate sample was subjected to SDS-PAGE (Bio-Rad, 3450124) at 100–120 V for 90 min, and then transferred to PVDF membrane (Bio-Rad, 1704273). The PVDF membrane was blocked with 5% non-fat dry milk, and then incubated with the indicated primary antibody (1:500-1:1000) at 4°C overnight. The membrane was washed three times in TBS-T buffer, and then incubated with an HRP-linked secondary antibody for 1 h at room temperature. After enhanced chemiluminescence exposure, a ChemiDoc Touch Imaging System (Bio-Rad, 1708370) was used for visualization and quantitative analysis of protein expression according to the manufacturer’s instructions. ACTB was used as a housekeeping control for whole cell lysates. TOMM20 and CALR (also known as calreticulin) were used as a mitochondrial or ER loading control, respectively. Mitochondria and ER isolation kits were obtained from Thermo Fisher Scientific (89874) or Sigma-Aldrich (ER0100), respectively.

After treatment, the cells were lysed at 4°C using ice-cold RIPA lysis buffer (Cell Signaling Technology, 9806), and the cell insoluble matter was removed by centrifugation (12,000 g, 10 min). BCA was used to detect the protein concentration, and then 200-300 μg protein in each lysate sample was pre-cleared for 3 h with Protein A Sepharose beads (Cell Signaling Technology, 9863) at 4°C. Then, in the presence of Protein A Sepharose beads, the sample with control IgG or a specific antibody (5 μg/mL) was gently shaken overnight at 4°C. After incubation, the Protein A Sepharose beads were washed thoroughly with PBS, and the protein was eluted by boiling in 2 × Laemmli sample buffer (Bio-Rad, 161-0737) before SDS-PAGE.

Commercially available assay kits were used to measure the concentrations or activity or release of MDA (Abcam, ab118970), HMGB1 (Shino-Test Corporation, 326070442), caspase-3 (Sigma-Aldrich, CASP3C), or glutamate (Thermo Fisher Scientific, A12221) in the indicated samples according to manufacturer’s guidelines.

Commercially available probes, including lipid peroxidation probe BODIPY-C11 (Thermo Fisher Scientific, D3861), mtROS probe MitoSOX (Thermo Fisher Scientific, M36008), and mitophagy probe Mtphagy Dye (Dojindo, MD01-10) were used in live cells according to manufacturer’s guidelines. Hoechst 33258 dye (Thermo Fisher Scientific, H3569) was used to stain cell nuclei. The signal was analyzed using a confocal microscope (ZEISS) or microplate reader (Tecan). The morphology of mitochondria was determined by staining TOMM20 and then a quantitative assay for mitochondrial fusion was performed as previously described (Bahat et al., 2018).

Data are expressed as mean ± SD unless otherwise stated. Unpaired Student’s t-test or an analysis of variance (ANOVA) were used to compare two or different groups, respectively. A P-value of < 0.05 was considered statistically significant.