Acsl4 contributed to lipidomic differences in gliomas
To investigate the underlying roles of ferroptosis in gliomas, microarray assays were firstly performed to identify lipid oxidation events by analyzing all major PE species in low-grade glioma (LGG) and glioblastoma (GBM) specimens. We found that AA- and AdA-containing PE (18:0/20:4 and 18:0/22:4, respectively) species were strikingly reduced in GBM versus LGG tissues (Fig. 1A) and cells (Fig. 1B). Then, we compared levels of oxidized lipids formed in erastin treated LGG and GBM cells. Similar to the results above, GBM cells suppressed the formation of doubly and triply oxidized AA- and AdA-containing PE species treated with erastin, in contrast with LGG cells (Fig. 1C). Because 12-HETE and 15-HETE are known as products of lipid peroxidation and are increasingly recognized as markers of ferroptosis, we found that the levels of both acids were decreased in GBM versus LGG specimens (Fig. 1D).
Previously published studies have revealed that some encoding proteins played crucial roles in lipid biosynthesis, especially in catalyzing PE-AA and PE-AdA, which represented preferred substrates for oxidation. Next, we detected endogenous levels of four proteins— Acsl4, GPX4, LPCAT3 and 15-LOX in LGG and GBM specimens using western blot (WB). Acsl4 was selected for further analysis because other proteins showed no difference in expression between LGGs and GBMs (Fig. 1E, Supplementary. Fig. 1A). Immunohistochemical (IHC) staining also confirmed that GBMs showed markedly decreased Acsl4 protein expression compared with LGGs (Fig. 1F). Then, analysis of Acsl4 expression in public databases such as The Cancer Gene Atlas (TCGA) and Rembrandt revealed its relatively low expression in GBM compared with LGG (Fig. 1G). Kaplan–Meier analysis also demonstrated that patients with low Acsl4 expression levels displayed reduced overall survival time, while gliomas high in Acsl4 expression favored prolonged survival (Fig. 1H). Patient-derived glioma cells PL1 and PG7 were respectively isolated from discarded LGG1 and GBM7 specimens using WB (Supplementary. Fig. 1B). Additionally, we found that various glioma cell lines such as U87, U251, T98, PL1, and PG7 showed downregulated Acsl4 protein expression, in contrast with normal human astrocytes (NHAs) in culture (Supplementary. Fig. 1C). Therefore, we selected PL1 and PG7 cells for subsequent experiments. Our general supposition was that GBM might escape ferroptosis via genetic deficiency of Acsl4, which plays an important role in glioma lipidomics and serves as a vital ferroptosis marker in glioma.
Acsl4 participated in mitochondrial-morphology regulation in ferroptosis
Kagan et al. has elaborated that genetic and pharmacological suppression of Acsl4 markedly increases the resistance of mitochondria to RSL3-induced outer-membrane rupture, possibly constituting an antiferroptotic rescue pathway. Given the mitochondrion is a main organelle for cellular oxidative phosphorylation and ROS production, we wished to unambiguously determine whether mitochondria were affected in Acsl4-dependent ferroptosis in glioma cells. Ferroptosis is known to differ from other forms of cell death that do not involve mitochondrial damage in cancer cells, but it contains significant morphological changes in mitochondria such as mitochondrial fragmentation and enlargement of cristae. We observed that compared with GBM cells, the mitochondrial morphology of LGG cells was liable to show more fragmentation accumulation around the nucleus in a dose-dependent manner in response to erastin toxicity (Fig. 2A). Quantification of mitochondrial length changed significantly at 1 µM in PL1 cells but showed no difference until 5 µM in PG7 cells (Fig. 2B). Furthermore, transmission electronic microscopy (TEM) revealed that PL1 cells treated with 1 µM erastin and PG7 cells treated with 5 µM erastin for 6 h had shrunken mitochondria and collapsed outer membranes with decreased microvilli, compared with cells that had been treated with the previous concentration (Fig. 2C). Therefore, we speculated that GBM cells tended to maintain a network of tubules, this morphology being characteristic of healthy and functional mitochondria, potentially due to low expression of Acsl4.
To investigate this possibility, we depleted PL1 cells Acsl4 of via short-hairpin ribonucleic acid (shRNA) knockdown, creating shAcsl4 cells, and stably transfected these into PG7 cells, creating Lv-Acsl4 cells. Protein and genetic levels of Acsl4 were detected via WB and quantitative reverse-transcription polymerase chain reaction (qRT-PCR; Fig. 2D and E).Since accumulation of lipid ROS is an end product of lipid peroxidation and a hallmark of ferroptosis in glioma, we estimated levels of lipid peroxidation using C11 BODIPY 581/591, a dye that is sensitive to lipid peroxidation. We found that after erastin treatment, ROS accumulation decreased to approximately one third in PL1 cells but roughly tripled in PG7 cells compared with control (Fig. 2F). We verified this finding using another independent probe, MDA (Fig. 2G). Moreover, both 12-HETE and 15-HETE levels were reduced when Acsl4 was knocked down in PL1 cells and increased when Acsl4 was overexpressed in PG7 cells (Fig. 2H and I). Additionally, as important antioxidants, reduced form GSH and GPX activity were upregulated in PL1–shAcsl4 cells, indicating that ferroptosis was suppressed in these cells, whereas they were downregulated in PG7–Lv-Acsl4 cells, indicating that ferroptosis was promoted therein (Fig. 2J and K). Furthermore, consistent with a prior study confirming subcellular localization of a lipid ROS probe via confocal-fluorescence microscopy (CFM), we found that in both PL1 and PG7 cells treated with erastin, the oxidized probe appeared in a distribution significantly colocalized with mitochondria and with the plasma membrane, with relatively high expression of Acsl4 (Fig. 2L). More importantly, the mitochondrial network of PL1–shAcsl4 cells was more elongated than that of PL1–shctrl cells, while PG7–Lv-Acsl4 cells became more fragmented and less elongated than PG7–vector cells. Therefore, we concluded that the mitochondrion was a primary site of Acsl4-dependent ferroptosis in glioma cells and that mitochondrial morphology could be affected by expression of Acsl4 in glioma ferroptosis.
Drp1 phosphorylation was essential for Acsl4-dependent ferroptosis
To establish potential regulators of Acsl4, we identified the proteins pulled down. Compared with control, we distinctly observed enrichment of proteins and a prominent band resolved at approximately 83 kDa, as shown in the sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) images in Figure 3A. Simultaneously, we analyzed whole-eluted samples without bias via mass spectrometry (MS). Next, we detected Drp1 as the most enriched protein. As expected, Drp1 was immunoprecipitated by sepharose-A coated with Acsl4 antibodies but not by immunoglobulin G (IgG; Fig. 3B), which was also confirmed by CFM (Fig. 3C).
Drp1 is well known for its role in regulating mitochondrial morphology, which was recently confirmed to be related to erastin-induced ferroptosis in melanoma cells. Additionally, Drp1 deficiency leads to mitochondrial elongation and mitochondrial tubules connection, similar to the morphology of mitochondria that are insensitive to ferroptosis . Therefore, we detected expression patterns of Drp1 in LGG and GBM samples using WB. Interestingly, we found no difference in expression of Drp1 protein between LGGs and GBMs (Supplementary. Fig. 2A and B). As is well known, Drp1 activity in glioma cells is regulated by post-translational modifications, mainly by phosphorylation: while serine 616 (Ser616) residue is an activation site, Ser637 residue is a repression site [23, 24]. We next determined levels of phosphorylated Drp1 (p-Drp1) at Ser616 and Ser637 in primary LGG and GBM clinical specimens using Western blot and IHC. In every model we tested, p-Drp1 (Ser637) levels were significantly increased in GBMs compared with LGGs, while p-Drp1 (Ser616) and Drp1 levels did not differ between the two specimen types, suggesting that GBM cells had attenuated activity of Drp1. Moreover, both LGG and GBM specimens showed decreased Drp1 Ser637 expression when treated with erastin (Fig. 3D–G), indicating that inactivation of Drp1 by phosphorylation at Ser637 was suppressed in erastin-induced ferroptosis. Then, to determine whether Drp1 phosphorylation was relevant to Acsl4-dependent ferroptosis, we generated a gain-of-function Drp1 containing both S637E (to mimic inhibitory phosphorylation) mutations in PL1 cells and S637A (to block inhibitory phosphorylation–dephosphorylation) mutations in PG7 cells (Supplementary. Fig. 2C and D). We found that after erastin treatment, both lipid ROS and MDA levels were significantly decreased in PL1 cells transduced by lentivirus expressing mutated Drp1S637E compared with control vector, synchronous with 12-HETE and 15-HETE levels (Fig. 3I and J, Supplementary. Fig. 3A and B). Moreover, reduced-form GSH and GPX activities were enhanced by the inactivation of Drp1 (Supplementary. Fig. 3C and D). CFM revealed that the morphology of mitochondria in PL1–Drp1S637E cells became more filamentous. However, when we overexpressed Acsl4 in PL1–Drp1S637E cells, the aforementioned changes in lipid mediators and mitochondrial morphology were reversed by Acsl4 restoration (Fig. 3H–J), suggesting that phosphorylation of Drp1 at Ser637 in LGG cells strongly inhibited Acsl4-dependent ferroptosis. Correspondingly, lipid ROS and MDA levels were markedly increased in PG7 cells transduced by lentivirus expressing mutated Drp1S637A compared with control vector cells (Fig. 3H–J), synchronous with 12-HETE and 15-HETE levels (Supplementary. Fig. 3A and B). In addition, reduced-form GSH and GPX activities were exhausted by activation of Drp1 (Supplementary. Fig. 3C and D). Moreover, PG7–Drp1S637A cells tended to show a fragmented mitochondrial phenotype. However, when we knocked down Acsl4 in PG7–Drp1S637A cells, those surrogate markers were once again reversed (Fig. 3H–J, Supplementary. Fig. 3A–D), suggesting that dephosphorylation of Drp1 at Ser637 in GBM cells strongly induced Acsl4-dependent ferroptosis. Taken together, these results demonstrated that dephosphorylation of Drp1 at Ser637 inhibited mitochondrial filamentation, which was essential for inducing Acsl4-dependent ferroptosis in glioma cells.
Hsp90 regulated Drp1 phosphorylation via calcineurin in gliomas
Through MS, we found that Hsp90 as well as Drp1 interacted with Acsl4 (Fig. 4A). Hsp90, as a global regulator of tumor cell metabolism in mitochondria including oxidative phosphorylation and redox networks, is defined as a common regulatory node in both necroptosis and ferroptosis[25, 26]. Co-immunoprecipitation (co-IP) experiments in PL1 and PG7 cells showed that Acsl4, Drp1, and Hsp90 interacted with each other (Fig. 4B), and confocal images showed that Hsp90 colocalized with Acsl4 and Drp1 in the mitochondrial outer membrane (Fig. 4C).
A previous study revealed calcineurin (CN) to dephosphorylate Drp1 at Ser637 in many types of cells; Hsp90 has been reported to bind to CN and stimulates its activity. Therefore, we next verified the effects of Hsp90 on CN and Drp1 in PL1 cells and in PG7 cells in the context of glioma, particularly the effects of Hsp90 level on the Drp1–Acsl4 axis in erastin-induced ferroptosis using WB. As expected, we found that knockdown of Hsp90 in PL1 cells promoted Drp1 (Ser637) phosphorylation while downregulating CN and Acsl4 expression. Of note, the expression level of Drp1 remained constant. Similarly, when we overexpressed Hsp90 in PG7 cells, we also found that the Drp1–Acsl4 axis was activated (Fig. 4D).
Promotion of the Hsp90–Acsl4 pathway enhanced erastin sensitivity in vitro
We sought to determine the relevance of the Hsp90–Acsl4 pathway in erastin-induced ferroptosis in vitro. First, we investigated the effect of Hsp90 on Drp1 and Acsl4 protein levels in erastin-induced ferroptosis. As shown in Figure 5A, shHsp90 significantly promoted Drp1Ser637 level and inhibited CN and Acsl4 expression; meanwhile, Drp1 expression remained unchanged in PL1 cells. Similarly, in PG7 cells, Lv-Hsp90 markedly inhibited Drp1 (Ser637) phosphorylation and promoted CN and Acsl4 expression, while Drp1 expression remained unchanged.
Next, we examined whether Hsp90 could sensitize glioma cells to Acsl4-dependent ferroptosis. As shown in Figure 5C and D, Hsp90 significantly ameliorated lipid ROS and MDA generation in PL1 and PG7 cells. The Hsp90–Acsl4 pathway also affected 12-HETE and 15-HETE levels (Fig. 5E and F). Moreover, GSH was exhausted by the promotion of Hsp90, with GPX activity downregulated in glioma cells (Fig. 5G and H). Furthermore, CFM revealed that mitochondria extended throughout the cell body to sites distal from the nucleus when the Hsp90–Acsl4 pathway was downregulated, and they showed more fragmentation accumulation around the nucleus when this pathway was upregulated (Fig. 5B).
We further investigated the mechanisms of the Hsp90–Acsl4 pathway on ferroptosis by studying the cytotoxic efficacy of erastin in glioma cells. Colony formation, EdU, and TUNEL experiments were performed to evaluate cell proliferation. These results showed that in all highly Acsl4-expressing cells (Lv-Hsp90, Drp1S637A, and Lv-Acsl4), the ability of erastin to inhibit cell proliferation was notably enhanced; meanwhile, in cells with reduced Acsl4 expression levels (shHsp90, Drp1S637E, and shAcsl4), this ability was comparably reduced (Fig. 6A–C). We also performed CCK-8 and cytotoxicity (LDH) assays to confirm that the Hsp90–Acsl4 pathway had a significant effect on cell proliferation in erastin-induced ferroptosis (Supplementary. Fig. 4A and B). Of note, apoptosis was not affected in PL1 and PG7 cells with changed Acsl4 levels, as indicated by the comparable activity and expression levels of cleaved Caspase-3 (Supplementary. Fig. 4C). Therefore, these results confirmed that Hsp90–Acsl4 pathway upregulation promoted ferroptosis and decreased proliferation of glioma cells. Conversely, downregulation of this pathway decreased ferroptosis and promoted proliferation of glioma cells.
Promotion of the Hsp90–Acsl4 pathway enhanced erastin sensitivity in vivo
To examine whether promotion of the Hsp90–Acsl4 pathway also increased tumor sensitivity to erastin in vivo, we first established mouse subcutaneous and orthotopic models via PG7 cells to confirm the sensitivity of parental GBM cells to different concentrations of erastin. Five days after PG7 implantation, mice were treated intraperitoneally (i.p.) with erastin at different concentrations (5, 10, 15, or 20 mg/kg−1/day−1 per mouse) or dimethyl sulfoxide (DMSO; 0.3%) every 2 days (Fig. 7A). Subcutaneous tumors transplanted with PG7 cells were visible at about 10 days and we found that erastin’s ability to slow tumor growth rate changed significantly between 10 and 15 mg (Fig. 7B). At day 35 after transplantation, tumors were collected and weighed (Fig. 7C and D), confirming previous results. At the same time, we traced tumor progression using in vivo bioluminescence imaging every 7 days; the images revealed that the antitumor effect of erastin did not make a difference until 15 mg (Fig. 7F). Similarly, mouse survival rate also exhibited an obvious difference between erastin 10 mg and erastin 15 mg (Fig. 7E).
Next, we assessed the therapeutic value of Acsl4 overexpression on GBM cells in vivo. Five days after PG7 implantation, mice were treated i.p. with erastin (10 mg/kg−1/day−1 per mouse) or DMSO (0.3%) every 2 days (Fig. 8A). We found that when erastin was combined with Hsp90 or Acsl4 overexpression, tumor inhibition was more significant. Drp1S637E inhibited the growth of erastin-treated Lv-Hsp90 PG7 cells to the same degree that it did to erastin-treated PG7 tumors (Fig. 8B–D). Moreover, bioluminescent imaging revealed that overexpression of Acsl4 effectively increased the sensitivity of GBM xenografts to erastin treatment. Mice receiving combined treatment showed considerably smaller tumor volume than other mice (Fig. 8G) and had dramatically prolonged lifespans (Fig. 8E). Orthotopic glioblastoma development markedly decreased mouse weight, which was mitigated by erastin administration (Fig. 8F).
We next assessed levels of primary Hsp90–Acsl4 pathway proteins in mouse tumors using IHC. Consistent with the in vitro results, Hsp90 overexpression mitigated p-Drp1Ser637 levels and enhanced Acsl4 expression, whereas Drp1 level did not change significantly (Fig. 9A). Additionally, mice with Hsp90–Acsl4 overexpression showed decreased levels of Ki-67 (Fig. 9B). TUNEL assays demonstrated that erastin mildly promoted cell death in vivo. Hsp90 and Acsl4 overexpression intensified erastin-induced PG7 cell death, but Drp1S637E limited this effect (Fig. 9C). Because mitochondria play a pivotal role in Acsl4-dependent ferroptosis, we assessed mitochondrial morphology in transplanted GBM tumors via TEM. Shrunken mitochondria and ruptured outer mitochondrial-membrane, possibly related to the typical mitochondrial changes caused by ferroptosis were found in the erastin, erastin + Lv-Hsp90, and erastin + Lv-Hsp90+ Drp1S637E +Lv-Acsl4 groups. These changes were alleviated in the erastin + Lv-Hsp90 +Drp1S637E group (Fig. 10A). Finally, we performed microarray assays to identify lipid oxidation events by analyzing all major PE species in mouse tumor specimens. We found that AA- and AdA-containing PE (18:0/20:4 and 18:0/22:4, respectively) species were strikingly increased in the erastin groups, especially the erastin + Lv-Hsp90 and the erastin + Lv-Hsp90 + Drp1S637E + Lv-Acsl4 group (Fig. 10B). Overall, these data demonstrated that Acsl4 could serve as a potential therapeutic target to enhance the benefits of erastin.