Abstract

   The gas therapy is drawing increasing attention in the treatment of
   many diseases including cancer. As one of gas signaling molecules,
   carbon monoxide (CO) has been proved to exert anti-cancer effects via
   triggering multiple cell death types, such as autophagy, apoptosis and
   necrosis. Here, we showed that low concentration CO delivered from
   CO-releasing molecule 3 (CORM-3) effectively induced ferroptosis, known
   as a novel proinflammatory programmed cell death, in vitro and in vivo.
   Mechanistically, we found that CO triggered ferroptosis by modulating
   the ROS/GSK3β/GPX4 signaling pathway, resulting in the accumulation of
   lipid hydroperoxides and the occurrence of ferroptosis. We think our
   findings provide novel insights into the anti-cancer mechanisms of CO,
   and suggest that CO could potentially be exploited as a novel
   ferroptosis inducer for cancer treatment in the future.

   Subject terms: Non-small-cell lung cancer, Cell death

Introduction

   Non-small cell lung cancer (NSCLC) accounts for ~80% of patients of
   lung cancer, which is the leading cause (~25%) of cancer-related deaths
   worldwide [[51]1]. Besides traditional treatments for NSCLC including
   surgery, chemo- and radiotherapy, new treatments, such as targeted
   therapy and immunotherapy, have been applied in recent years. However,
   the 5-year survival rate of NSCLC patients is still less than 15%.
   Consequently, there is an urgent need for the development of novel
   therapies for treatments of NSCLC.

   Recently, the treatment based on gaseous signaling molecules, which are
   involved in multiple critical functions via regulating signal
   transduction, has got intensive attention. Compared with traditional
   chemotherapy drugs, gas molecules, such as carbon monoxide (CO), nitric
   oxide (NO), hydrogen sulfide (H[2]S) and hydrogen (H[2]), display the
   in-comparable diffusion ability and stronger permeability in cancer
   therapy [[52]2–[53]4].

   CO was always considered as a kind of dangerous and lethal gas due to
   its strong binding with hemoglobin, then resulting in severe hypoxia
   [[54]5]. In the past decades, CO has been discovered to be generated by
   heme oxygenase (HO)-catalyzed heme degradation under physiological
   conditions. Moreover, it is worth noting that no distinct toxicity was
   detected after chronic exposure to even 500 ppm CO continuously for up
   to 2 years in rodents [[55]6, [56]7]. This also provides the
   possibility for clinical applications of exogenous low concentration
   CO. In recent years, low concentrations of endogenous CO have been
   proved to act as an important physiological signaling molecule to
   regulate the diverse physiological functions including the regulation
   of neurotransmitters and neuropeptide release, the relaxation of
   pulmonary vasculature, the anti-inflammatory effect, etc. In addition,
   increasing evidences support the anti-cancer effect of exogenous low
   concentration CO [[57]8–[58]11]. Moreover, CO has been reported to
   increase the tumor sensitivity to chemotherapy and to provide
   protection against the doxorubicin-induced cardiotoxicity. It raises
   more and more concerns that the biological functions and potential
   application in pharmacology of low concentration CO.

   Ferroptosis, an iron-dependent form of programmed cell death (PCD), is
   characterized by accumulation of lipid peroxides and imbalance of the
   redox system [[59]12, [60]13]. Morphologically, ferroptotic cells
   display mitochondrial atrophy, increased mitochondrial membrane density
   and reduced mitochondrial cristae [[61]14]. Some ferroptosis-inducing
   agents or stimulators, including sorafenib, artesunate and
   piperlongumine, are capable of effectively killing tumor cells,
   especially highly aggressive tumor cells [[62]15–[63]17]. Furthermore,
   given the non-apoptotic nature, ferroptosis-based cancer therapy may
   represent a novel strategy to overcome the resistance to
   apoptosis-inducing chemotherapy drugs [[64]18, [65]19]. In addition, as
   a form of immunogenic cell death, ferroptosis may elicit antitumor
   immune response via releasing various damage-associated molecular
   patterns [[66]20]. Therefore, inducing ferroptosis has been considered
   as a promising cancer treatment strategy. Accumulating evidences
   suggest that GPX4 can neutralize lipid peroxides, thereby protecting
   cells from ferroptosis. Inhibition or loss of GPX4 triggers ferroptosis
   by promoting accumulation of lipid peroxides [[67]21]. To date,
   applications of several GPX4 inhibitors are under different stages of
   clinical trials [[68]22].

   Previous studies have shown that CO could induce autophagy [[69]23],
   apoptosis [[70]24] and necrosis [[71]25]. In this study, we found that
   ferroptosis was induced by CO in a time- and dose-dependent manner in
   NSCLC cells. Mechanistically, we discovered that CO decreased the
   expression of GPX4 via the ROS/GSK3β axis, and then caused the
   accumulation of lipid peroxides to trigger ferroptosis. Our findings
   provide novel insights into the antitumor effects of low concentration
   CO, and demonstrate that the CO might work as a promising therapeutic
   medical gas against NSCLC cells via inducing ferroptosis.

Results

CO (CORM-3) induces cytotoxicity in various lung cancer cells

   To investigate the toxicity of CO to cancer cells, we evaluated the
   clone formation of various NSCLC (non-small cell lung cancer) cells
   after CO (CORM-3) treatment. As shown in Fig. [72]1A, CO treatment
   significantly decreased the clone survival of H1299, Calu-1 and H1975
   cells. Furthermore, clone survival of H1299 and Calu-1 cells in soft
   agar was also significantly reduced after CO treatment (Fig. [73]1B).

Fig. 1. CO (CORM-3) treatment decreased clone survival and cell viability of
NSCLC cells.

   [74]Fig. 1
   [75]Open in a new tab

   A Clone survival and corresponding quantified results of various NSCLC
   cells treated with CORM-3 (40 μM). B Representative images and
   corresponding quantified results of soft agar clone survival captured
   at 14 d after CORM-3 treatment (40 μM). C Cell viability of H1299 or
   Ca-lu-1 cells at 96 h after CORM-3 treatment (0–80 μM). D Cell
   viability of H1299 and Calu-1 cells at indicated time points after
   CORM-3 treatment (80 μM). E Cell viability of normal lung cells
   (BEAS-2B, MRC-5 and HBE) at 96 h after CORM-3 treatment (80 μM). ns not
   significant, *P < 0.05, **P < 0.01, ***P < 0.001.

   Next, we evaluated the dose and time dependence of CO-induced
   cytotoxicity in H1299 and Calu-1 cells with CCK-8 assay. The cell
   viability was gradually decreased with the increase of CORM-3
   concentration (10–80 μM). In addition, the cell viability was also
   reduced with the prolonged incubation time after CORM-3 exposure (Fig.
   [76]1C), and significant cytotoxicity appeared at 48 h after CORM-3
   treatment (Fig. [77]1D). These results suggest that CO significantly
   kills NSCLC cells in a dose- and time-dependent manner. Importantly,
   CORM-3 treatment did not affect the cell viability of normal human lung
   cells (BEAS-2B, MRC-5 and HBE), showing selective cytotoxic effects of
   CO in cancer cells (Fig. [78]1E).

CO (CORM-3) induces ferroptosis in NSCLC cells

   After CORM-3 treatment, we observed probable non-apoptotic cell death
   with microscopy (Fig. [79]2A). To further explore the type of cell
   death induced by CO exposure, we treated H1299 or Calu-1 cells with
   CORM-3 in the absence or presence of different cell death inhibitors.
   Interestingly, Ferrostatin-1 (Fer-1), a specific ferroptosis inhibitor,
   nearly completely rescued CO-induced cell death (Fig. [80]2B).
   Conversely, apoptosis inhibitor (Z-VAD-fmk) or necroptosis inhibitor
   (Necrostatin-1s) did not suppress cell death induced by CO. These
   findings suggest that CO might trigger ferroptosis in NSCLC cells, and
   this was also supported by the results that liproxstatin-1, another
   strong lipid oxidation inhibitor, nearly completely attenuated
   CO-induced cell death in H1299 and Calu-1 cells (Fig. [81]2C).
   Meanwhile, the cells treated with CORM-3 also displayed the
   characteristics of ferroptosis, accumulated lipid peroxidation and
   elevated MDA level (Fig. [82]2D–F). In addition, decreased GPX4
   activities were observed in H1299 cells at 72 h after CO treatment
   (Fig. [83]2G). Moreover, the images of transmission electron microscopy
   showed morphological features of ferroptosis, the shrunken mitochondria
   with disrupted cristae in CORM-3-treated cells (Fig. [84]2H).
   Collectively, our data suggest that low concentration CO induces
   ferroptosis in NSCLC cells.

Fig. 2. CO induces ferroptosis in NSCLC cells.

   [85]Fig. 2
   [86]Open in a new tab

   A Representative images of H1299 cells captured at 96 h after treatment
   with CORM-3 (80 μM). Scale bar: 200 μm. B Effect of indicated
   inhibitors, z-VAD-fmk (20 μM), Nec-1S (20 μM) or Fer-1 (20 μM), on cell
   viability after CORM-3 (80 μM) treatment in H1299 and Calu-1 cells. C
   Effect of Lipro-1 (10 μM) on cell viability after CORM-3 (80 μM)
   treatment in H1299 and Calu-1 cells. D, E Flow cytometry analysis and
   quantification of cellular lipid peroxidation in H1299 (D) and Calu-1
   (E) cells after CORM-3 (80 μM) treatment. F The level of MDA after
   CORM-3 (80 μM) treatment in H1299 cells. G Activity of GPX4 in H1299
   cells treated with RuCl[3] or CORM-3. H Mitochondrial morphology
   captured with transmission electron microscopy after CORM-3 (80 μM)
   treatment in H1299 cells. ns: not significant, *P < 0.05, **P < 0.01,
   ***P < 0.001.

CO drives ferroptosis through GPX4 downregulation and ROS production

   To determine the genes involved in CO-induced ferroptosis, mRNA-seq was
   performed after RuCl[3] or CORM-3 treatment in H1299 cells. After
   preprocessing the original data, with |log2FC | >0.32 and P < 0.05 as
   the standards, 2285 differentially expressed genes (DEGs) were obtained
   from the data set, including 1078 upregulated genes and 1207
   down-regulated genes (Fig. [87]3A). The results of Kyoto Encyclopedia
   of Genes and Genomes (KEGG) pathway analysis showed significant
   enrichment in the ferroptosis pathway after CO treatment (Fig. [88]3B).
   Furthermore, Gene Set Enrichment Analysis (GSEA) of these DEGs revealed
   that ferroptosis signaling pathway was significantly enriched after CO
   exposure (Fig. [89]3C). In the ferroptosis pathway, 12 genes including
   9 upregulated genes (MAP1LC3B, SLC3A2, HMOX1, FTH1, GCLM, FTL, SLC7A11,
   GCLC and SAT1) and 3 down-regulated genes (GPX4, STEAP3, SLC39A8)
   significantly changed in response to CO treatment (Fig. [90]3D). The
   PPI network of 12 ferroptosis-related genes was constructed by using
   the STRING database and the two hub genes, GPX4 and HO-1, were selected
   through the degree plugin of the Cytoscape software (Fig. [91]3E).
   Further validation using qPCR and western blot also confirmed the
   decreased GPX4 expression and increased HO-1 expression at mRNA and
   protein levels, in CORM-3-treated H1299 and Calu-1 cells (Fig. [92]3F,
   G).

Fig. 3. Data analysis of mRNA sequencing of CO-treated H1299 cells.

   [93]Fig. 3
   [94]Open in a new tab

   A Volcano maps of the differentially expressed genes in H1299 cells
   (RuCl3 vs CORM-3). B, C KEGG pathway enrichment analysis (B) and GSEA
   analysis (C) were based on the differentially expressed genes in H1299
   cells (RuCl[3] vs CORM-3). D Heatmap of ferroptosis-related
   differentially expressed genes identified by mRNA-seq in H1299 cells
   (RuCl[3] vs CORM-3). E PPI networks and the significant gene module in
   the ferroptosis-related gene set. F, G The mRNA (F) and protein (G)
   expression of HO-1 and GPX4 genes after CORM-3 treatment in H1299 and
   Calu-1 cells. **P < 0.01, ***P < 0.001.

   Numerous evidences have indicated that upregulation of HO-1 and
   downregulation of GPX4 is involved in ferroptosis induction [[95]21,
   [96]26]. To further confirm the critical role of HO-1 in CO-induced
   ferroptosis, HO-1 specific siRNA was employed to interfere with HO-1
   expression in H1299 cells (Fig. [97]4A). Unfortunately, knockdown of
   HO-1 did not rescue the cell viability after CO treatment (Fig.
   [98]4B), suggesting that CO-induced ferroptosis was independent of
   HO-1. Meanwhile, GPX4-overexpressing H1299 cell line (H1299-GPX4) was
   constructed to investigate the role of GPX4 in CO-induced ferroptosis.
   Compared with the H1299-vector cells, both the H1299-GPX4 cells with or
   without CORM-3 treatment displayed enhanced expression of GPX4 (Fig.
   [99]4C), which indicated that the decrease of GPX4 induced by CO was
   rescued by overexpressing GPX4 (Fig. [100]4F). As expected,
   overexpression of GPX4 mitigated cell death and decreased the
   accumulation of lipid peroxidation after CORM-3 treatment (Fig.
   [101]4D, E), suggesting that GPX4 inhibition was a major contributor to
   CO-induced ferroptosis.

Fig. 4. CO induces ferroptosis through GPX4 downregulation and ROS
production.

   [102]Fig. 4
   [103]Open in a new tab

   A Efficiency of knocking down HO-1 in H1299 cells detected with western
   blot. B Effect of HO-1 knockdown on the viability of H1299 cells
   treated with CORM-3. C Expressions of HO-1 and GPX4 in H1299/Vector and
   H1299/GPX4 cells at 72 h after CORM-3 treatment. D–F Levels of lipid
   peroxidation (D, E) and cell viability (F) in H1299-vector and
   H1299-GPX4 cells after CORM-3 treatment. *P < 0.05, **P < 0.01,
   ***P < 0.001.

CO down-regulates GPX4 expression via the activation of GSK3β

   Interestingly, we found that activation of GSK3β, one critical
   regulator of GPX4 expression, was enhanced after treatment with CO in
   H1299 and Calu-1 cells (Fig. [104]5A). To further determine the
   possible involvement of GSK3β in CO-induced ferroptosis, GSK3β specific
   siRNA was employed to interfere with the expression of GSK3β in H1299
   cells (Fig. [105]5B). As shown in Fig. [106]5C, GSK3β knockdown
   significantly increased the viability of H1299 cells treated with
   CORM-3, indicating the involvement of GSK3β in CO-induced ferroptosis.
   Furthermore, GSK3β knockdown dramatically alleviated the decrease of
   GPX4 protein expression after CORM-3 treatment (Fig. [107]5D),
   suggesting regulation of GPX4 by GSK3β. Similarly, accumulation of
   lipid peroxidation in response to CORM-3 treatment was also attenuated
   by knocking down GSK3β (Fig. [108]5E, F). The above results indicate
   that activation of GSK3β suppresses the expression of GPX4 and thus
   triggers ferroptosis. Moreover, NAC treatment nearly completely
   abrogated CO-induced cell death and accumulation of lipid peroxidation
   (Fig. [109]5G–I). The activation of GSK3β and decrease of GPX4
   expression induced by CO were also not observed after NAC treatment
   (Fig. [110]5J), indicating that ROS might be the primary cause of
   tumor-cell ferroptosis in response to CO treatment.

Fig. 5. Activation of GSK3β suppressed the GPX4 expression after CORM-3
treatment.

   [111]Fig. 5
   [112]Open in a new tab

   A Protein expressions of GSK3β and p-GSK3β in H1299 and Calu-1 cells at
   72 h after CORM-3 treatment. B Efficiency of GSK3β knockdown detected
   with western blot. C Effect of GSK3β knockdown on the viability of
   H1299 cells treated with CORM-3. D Effect of GSK3β knockdown on the
   expression of GPX4 in H1299 cells after CORM-3 treatment. E, F Levels
   of lipid peroxidation determined by C11-BODIPY 581/591 (10 µM) in
   GSK3β-knockdown H1299 cells treated with CORM-3. G–I Effect of NAC
   (5 μM) on the levels of lipid peroxidation (G, H) and cell viability
   (I) in H1299 cells treated with CORM-3. J Protein expression of GPX4
   and p-GSK3β in H1299 cells treated with CORM-3 in the presence of 5 μM
   NAC. *P < 0.05, **P < 0.01.

CO inhibits the growth of human lung cancer xenografts

   We further investigated the anti-tumor effect of CO in nude mice
   bearing H1299-Luc xenografts. Tumor-bearing mice were injected with
   either RuCl[3] or CORM-3 daily for 13 days as depicted in Fig. [113]6A.
   Compared with the results of the RuCl[3] treatment, significant
   suppression of tumor growth became apparent about 7 days after CORM-3
   treatment, and the inhibition became more distinctly with the time
   (Fig. [114]6B, C). At the end of experiment, a significant reduction of
   tumor volume and weight in CORM-3 treatment group was observed compared
   with the RuCl[3] treatment group (Fig. [115]6D–F), indicating that CO
   exerts anti-tumor activity in vivo, which was also supported by the
   decrease of Ki-67 proliferative index after CORM-3 treatment (Fig.
   [116]6G, H). Furthermore, we found CORM-3 treatment markedly elevated
   the levels of intratumoral MDA and 4-hydroxynonenal (4-HNE) (Fig.
   [117]6I, K), the secondary products of lipid peroxidation and the
   markers of ferroptosis. Consistent with our results in vitro, the
   expression of GPX4 protein in tumor was also down-regulated after
   treatment with CORM-3 but not RuCl[3] (Fig. [118]6G, J). Taken
   together, these results indicate that CO treatment effectively induce
   ferroptosis and suppress tumor growth in vivo via decreasing GPX4
   expression.

Fig. 6. CO inhibits the growth of H1299-Luc tumor xenografts.

   [119]Fig. 6
   [120]Open in a new tab

   A Schematic for experimental design. B Growth curves for H1299-Luc
   tumors in mice treated with CORM-3 or RuCl[3]. C Representative
   bioluminescence images of H1299-Luc tumors captured on days 1, 7 and 13
   after CORM-3 or RuCl[3] injections. D Macroscopic appearance of
   H1299-Luc tumors-bearing mice 13 days after CORM-3 or RuCl[3]
   treatment. E, F Gross appearances (E) and tumor weights (F) of excised
   tumors from mice sacrificed at 13 days after intratumoral injection
   with CORM-3 or RuCl[3]. G Representative IHC images of Ki-67, 4-HNE and
   GPX4 in tumors cut from mice. Scale bar: 100 μm. H–J Quantification of
   Ki-67 (H), 4-HNE (I) and GPX4 (J) protein expression. K The level of
   MDA in H1299-Luc tumors with CORM-3 or RuCl[3] treatment. *P < 0.05,
   **P < 0.01.

Discussion

   Recently, CO was found to have potential antitumor effects [[121]5].
   Accumulating evidences indicate that ROS and mitochondria have been
   implicated as the cellular targets for CO action [[122]27–[123]29].
   Wegiel et al. showed that CO inhibited tumor growth by targeting
   mitochondria and promoting ROS generation [[124]30]. This agrees with
   our results that CO treatment made mitochondrial morphology changes
   (shrunken mitochondria with disrupted cristae) and increased ROS
   production. It is noteworthy that CO also plays a negative role in ROS
   generation in previous studies. In osteoarthritic synoviocytes, CORM-2
   attenuates ROS generation and exerts an anti-inflammatory effect
   [[125]31]. In human gastric cancer cells, CORM-2 also abrogates
   IL-1β-induced ROS production [[126]32]. These suggest that CO
   functioning as a ROS inducer or ROS scavenger might be dependent on the
   cell types [[127]3]. Accordingly, further investigations are required
   to identify the relationship between CO and ROS generation.

   In our study, we found that CO treatment induced increased HO-1 and
   decreased GPX4 expression, which have been previously reported to
   trigger ferroptosis. In fact, HO-1 has dual roles in ferroptosis
   induction. As a negative regulator of ferroptosis induction, enhanced
   HO-1 expression prevents ferroptosis induced by glutamate in HT-22
   cells [[128]33]. Conversely, HO-1 enhances the ferroptosis induced by
   Bay117085 via promoting iron accumulation and ROS generation [[129]34].
   However, our results showed that knocking down HO-1 had no effects on
   CO-induced cell death, suggesting neither a pro-death nor pro-survival
   role of HO-1 in CO-induced ferroptosis. We speculate that there should
   be other factor(s) involved in iron metabolism to induce ferroptosis by
   developing iron overload. As a hydroperoxide scavenger which converts
   lipid hydroperoxides to non-toxic lipid alcohols, GPX4 prevents
   ferroptosis by suppressing lipid hydroperoxides. Some chemotherapeutic
   drugs including apatinib [[130]35], capsaicin [[131]36], paclitaxel
   [[132]37], etc., trigger ferroptosis via inhibiting GPX4 expression in
   several cancer cell types. Similar to previous studies, we also
   confirmed that downregulation of GPX4 contributed to CO-induced
   ferroptosis. However, overexpressing GPX4 only partially rescued the
   CO-induced ferroptosis, indicating the involvement of other signaling
   pathways in CO-induced ferroptosis which would require more in-depth
   studies.

   Although it is known that down-regulation of GPX4 promotes induction of
   ferroptosis, the upstream signal pathways modulating GPX4 expression
   are not yet fully understood. Several factors, such as ZEB1, Nrf2,
   ATF-4 and microRNA-15a, etc. [[133]38–[134]40], have been reported to
   regulate GPX4 expression. A previous work indicates that that CBS is
   responsible for the downregulation of GPX4 induced by CO in MCF-7 cells
   [[135]41]. In our study, inhibition of CBS was not observed after CO
   treatment in H1299 and Calu-1 cells, suggesting the independence of CBS
   activity in ferroptosis induction (Supplementary Fig. S[136]1). Our
   findings further confirmed that activation of GSK3β after CO exposure
   induces ferroptosis via down-regulating GPX4 expression, and this is
   different from a previous result that activation of GSK3β promotes
   ferroptosis via dominating iron homeostasis in Hela cells [[137]42].
   These findings suggest that GSK3β might regulate different the signal
   pathways to induce ferroptosis in various cell lines.

   In summary, low concentration of exogenous CO induced ferroptosis in a
   time- and dose-dependent manner in NSCLC cells rather than normal lung
   cells. Mechanistically, we found that CO treatment activated the
   ROS/GSK3β/GPX4 signaling pathway, then led to a dramatic decrease in
   GPX4 expression and the accumulation of lipid peroxides and induced
   ferroptosis finally (Fig. [138]7). We hope CO, identified as a novel
   ferroptosis inducer, may be applied as a potential chemosensitizer to
   overcome the resistance caused by apoptosis-inducing chemotherapy drugs
   in the future.

Fig. 7. Schematic diagram of CO-induced ferroptosis.

   Fig. 7
   [139]Open in a new tab

   CO increased the mitochondrial ROS production; the increased level of
   intracellular ROS promoted activation of GSK3β; GSK3β activation
   inhibited the transcription of GPX4, and then enhanced the accumulation
   of lipid peroxidation, resulting in ferroptosis.

Materials and methods

Cell culture

   H1299, Calu-1 and HEK-293T cells were obtained from the Cell Bank of
   Type Culture Collection of the Chinese Academy of Sciences (Shanghai,
   China). The H1975 cell line was purchased from the American Type
   Culture Collection (Manassas, USA). Calu-1, H1299 and H1975 cells were
   cultured with RPMI-1640 medium (Gibco, Carlsbad, USA). HEK-293T cells
   were cultured with DMEM medium (Gibco, CA, USA). All the culture media
   were supplemented with 10% FBS (Lonsera, Montevideo, Uruguay),
   100 µg/mL streptomycin and 100 U/mL penicillin. All cells were
   authenticated by STR profiling recently and verified to be free of
   mycoplasma and cultured at 37 °C in 5% CO[2].

Reagents and antibodies

   CORM-3 (S7448), one kind of water soluble CO-releasing molecules, was
   obtained from Selleck (Selleck Chemicals, Shanghai, China). RuCl[3],
   the negative control for CORM-3, was produced by completely releasing
   CO from CORM-3. Ferrostatin-1 (HY-100579), Liproxstatin-1 (HY-12726),
   Necrostatin-1 (HY-15760), Z-VAD-fmk (HY-16658B) and were purchased from
   MCE (MedChemExpress, New Jersey, USA).

   The primary antibodies used in the present study including anti-GPX4
   (52455), anti-HO-1 (43966S), and anti-p-GSK3β (9322 S) were purchased
   from CST (Cell Signaling Technology, Beverly, USA). Anti-GAPDH
   (60004-1-Ig), Anti-Ki-67 (27309-1-AP) and anti-β-actin (66009-1-Ig)
   antibodies were obtained from Proteintech (Wuhan, China). The antibody
   of anti-rabbit-GSK3β (WL01456) and anti-mouse-GSK3β (221162) antibodies
   were obtained from Wanleibio (Shenyang, China). The antibody of
   anti-4-HNE (mab3249) was obtained from Bio-Techne (Minneapolis, USA).
   The secondary antibodies, including anti-rabbit IRDye800® conjugated
   antibody (926-32211) and anti-mouse IRDye®680DX conjugated antibody
   (926-68070), were purchased from Li-COR (LI-COR Biosciences, Lincoln,
   USA).

Cell viability assay

   Cell viability was measured with Cell Counting Kit-8 (CCK-8) assay
   (K1018, ApexBio, Houston, USA) according to the manufacturer’s
   recommendation. Cell viability (%) = [(Absorbance of tested compound
   minus Absorbance of blank)/(Absorbance of control minus Absorbance of
   blank)] × 100%.

Colony formation assay

   The cells were plated into a 35 mm cell culture dish at a density of
   300 cells/dish followed by CORM-3 treatment. After 10 days incubation,
   the cells were fixed with 4% paraformaldehyde and stained with 1%
   crystal violet. The colonies containing ≥50 cells were counted and the
   survival fraction was calculated for each group.

Soft agar colony formation assay

   The prepared 0.6% agar in complete growth medium was plated in six-well
   plates as the base agar layer. After 30 min, re-suspended cells (1000
   cells/mL) in the growth medium containing 0.36% agar were seeded on the
   top of the solidified base layer, and then incubated at 37 °C and 5%
   CO[2]. After 2 weeks incubation, the number of colonies was counted
   with microscopy.

Western blotting

   The cells were lysed in RIPA buffer (P0013B, Beyotime Biotechnology,
   Shanghai, China) with PMSF (1 mM, ST506, Beyotime Biotechnology,
   Shanghai, China). The protein extracts were subjected to SDS-PAGE and
   transferred onto polyvinyl difluoride (PVDF) membranes. Subsequently,
   the membranes were incubated with primary antibodies at 4 °C overnight
   after blocking with 5% nonfat dry milk for 1 h at room temperature.
   After washing three times with TBST (0.1% Tween-20 in Tris-HCl buffer),
   the membranes were incubated with IRDye-labelled secondary antibodies
   for 1 h at room temperature, and the infrared fluorescence signals were
   detected with an Odyssey® CLx Infrared Imaging System (9140-00, Li-COR
   Biosciences, Lincoln, USA).

RNA extraction, mRNA sequencing (mRNA-seq) and quantitative real-time PCR

   Total RNA was isolated with RNeasy Mini kit (74004, Qiagen, Hilden,
   Germany) and cDNAs were synthesized from 2 μg total RNA with NovoScript
   1st Strand cDNA Synthesis kit (E047-01B, Novoprotein, Shanghai, China).

   Library preparation and data analysis of high throughput sequencing
   were conducted by Kangce Technology Co., Ltd (Wuhan, China). RNA-seq
   library was constructed by Illumina TruSeq RNA Sample Prep Kit version
   2 (Illumina, San Diego, CA) and RNA was sequenced by Illumina
   HiSEquation 2000 platform (Illumina). A gene is considered
   differentially expressed (DEG) if it has an FDR step-up p ≤ 0.05 and a
   fold-change ≥ ±2. KEGG pathway analyses
   ([140]https://www.kegg.jp/entry/map04216) were performed by using the
   DAVID website to study the function of DEGs. Values with p < 0.05 were
   considered statistically significant.

   Real-time PCR (qPCR) was performed with Hieff qPCR SYBR Green Master
   Mix Kits (11203ES03, Yeasen, Shanghai, China) on a Roche 480 Light
   Cycler (05015278001, Roche, Basel, Switzerland). The primers for PCR
   amplification were shown as follows: 5’-GAGGC AAGACCGAAGTAAACTAC-3’,
   5’-CCGAACTGGTTACACGGGAA-3’ (GPX4); 5’-AAGACTGCGTTCCTGCTCAAC’,
   5’-AAAGCCCTACAGCAACTGTCG-3’ (HO-1); and 5’-GGAGCGAGATCCCTCCAAAAT-3’,
   5’-GGCTGTTGTCATACTTCTCATGG-3’ (GAPDH). The GAPDH gene was used as a
   control, and the data were presented as fold changes of gene expression
   in the tested samples compared to the control with the 2^-ΔΔCt method.

Gene set enrichment analysis (GSEA)

   Gene set enrichment analysis (GSEA) was performed to clarify the
   enriched KEGG pathways by using the gene set “c2. cp.kegg.v7.4.
   symbols.gmt” as a reference. Normalized enrichment score (NES) absolute
   value greater than 1, false discovery rate (FDR) q < 0.25 and P < 0.05
   were used as the criteria in defining the significant difference.

Construction of protein-protein interaction (PPI) network

   STRING database ([141]http://string-db.org/) and Cytoscape software
   were used in the process of drawing the PPI networks of
   ferroptosis-related genes. Gene sets obtained by Venn analysis were
   inserted into STRING to generate PPI diagram, and the degree plugin was
   applied to screen pivotal genes.

Malondialdehyde (MDA) detection

   The MDA level was assessed with a lipid peroxidation MDA Assay Kit
   (S0131S, Beyotime Biotechnology, Shanghai, China) following
   manufacturer’s instructions. The cells were lysed with the lysis buffer
   and the isolated tumors were ground and lysed to obtain the tumor
   homogenate. After centrifugation, the supernatants were mixed with
   thiobarbituric acid (TBA) and incubated at 95 °C for 1 h. The
   absorbance was measured with a microplate reader at 532 nm to evaluate
   the MDA concentration.

Flow cytometry for lipid peroxidation detection

   The cells were collected and washed with PBS at 72 h after treatment.
   Subsequently, the cells were re-suspended in medium without FBS and
   stained with C11 BODIPY 581/591 (D3861, Invitrogen, Carlsbad, USA) for
   30 min, followed by flow cytometry analysis. Raw data were then
   processed with the FlowJo software.

Transmission electron microscopy

   The cells were fixed with 2.5% glutaraldehyde in phosphoric acid buffer
   for 2 h. After dehydration through an ethanol series, cells were
   embedded and then sectioned followed by staining with 3% uranium
   acetate and lead citrate. The images were then captured with a
   transmission electron microscope (JEM-1011, JEOL Ltd, Tokyo, Japan).

Establishing stable GPX4-overexpressing cell

   Human GPX4 plasmid was constructed by inserting GPX4 cDNAs
   ([142]NM_001039847.3) into a pCDH lentiviral vector (CD810A-1,
   Miaolingbio, Wuhan, China). To generate lentiviruses, the constructed
   plasmid was transfected into HEK-293T cells with pSPAX2 and pMD2G at a
   ratio of 5:3:2. Lentivirus-containing supernatant collected at 48 h
   after transfection was used to infect H1299 cells. At 48 h after
   infection, the cells were selected with 2 μg/mL puromycin for 14 days
   to obtain the GPX4-overexpressing H1299 cells (H1299-GPX4). To detect
   the tumor volumes in vivo, H1299 cells were stably expressed firefly
   luciferase by transfecting with pLentipuro3-TO-V5-GW-EGFP-Firefly
   Luciferase plasmid ([143]P21289, Miaolingbio, Wuhan, China).

Small interfering RNA transfection

   The siRNAs for HO-1 and the negative control siRNA were synthesized by
   GenePharma (Shanghai, China). The siRNA sequences of HO-1 were:
   siRNA#1, CCAGGCAGCTGATCATAGA; and siRNA#2, CGACCTGACTGCCAAGAAA. The
   siRNA sequences of GSK3β were: siRNA#1, CCACAAGAAGUCAGCUAUA; and
   siRNA#2, GGAU-CAGUUGGUAGAAAUA. The cells were transfected with
   double-stranded siRNAs by using the Lipofectamine®2000 transfection
   reagent (11668019, Thermo Fisher scientific, San Diego, USA) according
   to the manufacturer’s protocol.

Tumor growth and treatments

   Six-week-old male Balb/C-nude mice, obtained from GemPharmatech company
   (Nanjing, China), were grafted with 5 × 10^6 cells (H1299-Luc) by
   subcutaneous injection into the right flank. When tumors reached a size
   of ~50 mm^3, mice were randomized into three groups (n = 5) following
   random number table method and treated with PBS, RuCl[3] (30 mg/kg) or
   CORM-3 (30 mg/kg) via intratumoral injection once daily. Tumor growth
   was monitored by measuring the volume of tumors every two days. No
   sample animals were excluded from this study. The study was performed
   in non-blinded modality.

   The volume of a tumor was calculated as (L × W × H)/2, with the length
   L (the longest dimension), width W (the distance perpendicular to and
   in the same plane as the length), and height H (the distance between
   the exterior tumor edge and the mouse’s body) of each tumor was
   measured with vernier calipers. The mice were sacrificed at 13 days
   after CORM-3 treatment. Tumor xenografts were collected, photographed,
   and weighed and the ferroptosis was evaluated by immune-histochemistry
   (4-HNE) and MDA detection.

HE staining and immune-histochemistry (IHC)

   The tumor sections were deparaffined in xylene, rehydrated through
   gradient ethanol and then stained with H&E staining kit (Abcam,
   Cambridge, USA) as per manufacturers’ instructions. The expression
   levels of Ki-67, GPX4 and 4-HNE were detected with
   immunohistochemistry. The IHC results were scored as the percentage of
   positively stained cells by using Image J (IJ 1.46r).

Statistical analysis

   All experiments were performed at least three times. All analyses were
   performed with the GraphPad Prism 9 statistical software (GraphPad
   Software Inc., San Diego, USA). For all experiments, comparisons
   between two groups were based on t test, and one-way analysis of
   variance (ANOVA) was used to assess the difference between more than
   two groups. Data are shown as means ± SD. *P < 0.05, **P < 0.01 and
   ***P < 0.001; ns no significance.

Supplementary information

   [144]Original Data File^ (785.9KB, pdf)
   [145]Extended Data 1^ (2.7MB, tif)

Acknowledgements