Introduction

Acute myeloid leukemia (AML) is the most common acute leukemia in adults. Despite recent progress in molecularly targeted therapies, the 5-year overall survival rate is only about 30% [1]. This is mainly due to frequent relapse or refractory disease, which carries a dismal prognosis; these patients’ one-year survival rate is only 10% [2]. As a major cause of such therapy resistance, AML cells develop resistance to apoptosis induction through numerous mechanisms [3, 4], underlining the unmet need for the development of alternative therapeutic strategies that overcome apoptosis resistance.

Pathways of non-apoptotic regulated cell death are of growing interest in cancer research. One such form of regulated cell death is ferroptosis, which is characterized by iron dependency and triggered by the accumulation of lipid peroxidation on cellular membranes [5]. Increasing evidence from solid tumor studies supports the therapeutic potential of ferroptosis induction through its distinct mechanism of cell death independent of apoptosis. Major defense mechanisms against ferroptosis include the system xc–/glutathione/glutathione peroxidase 4 (GPX4) pathway [6, 7] and GPX4-independent pathways that utilize endogenous radical-trapping antioxidants such as coenzyme Q10 (CoQ) [8,9,10,11] or tetrahydrobiopterin [12, 13]. Among them, GPX4 is the only enzyme that can directly reduce lipid peroxides and its inhibition results in ferroptosis. GPX4 is a selenoprotein, whose translation is regulated by a unique protein synthesis mechanism that requires selenium. Because of its high rank in selenoprotein synthesis hierarchy, the expression of GPX4 is relatively retained even under selenium deficiency [14], suggesting its importance in cellular homeostasis.

AML cells exhibit increased reactive oxygen species production [15] and active iron uptake [16], both known determinants of ferroptosis susceptibility. We therefore hypothesize ferroptosis as a vulnerability in AML. While ferroptosis has been investigated recently in AML [17,18,19,20,21,22,23], the pathophysiological role of GPX4-mediated ferroptosis in AML remain largely unknown.

In the present study, we sought to elucidate the molecular mechanisms of GPX4 inhibition–induced ferroptosis in AML. We identified mitochondrial involvement in ferroptosis in AML cells, which provides a molecular rationale for targeting this disease-specific vulnerability to enhance the anti-leukemia effects of GPX4 inhibition.

Results

GPX4 is a potential therapeutic target in AML

To evaluate the therapeutic potential of targeting ferroptosis-regulating genes including GPX4 in a variety of cancer cell lines, we analyzed Cancer Dependency Map (DepMap) datasets using shinyDepMap [24], a tool that combines CRISPR and shRNA screening data of 423 cancer cell lines. This tool enables us to predict the essentiality of genes of interest across cancer cell lines as well as the selectivity of the gene essentiality among distinct cell types. In particular, we investigated ferroptosis suppressor genes annotated on FerrDb, a manually curated database of ferroptosis regulators and their disease associations [25, 26] (FerrDb V2, last access; 8/13/2022). GPX4 was the second-ranked gene with the high essentiality among the top 10 ferroptosis suppressors (efficacy score; X = −1.21), following MTOR (X = −1.454) (Fig. 1A). In addition, the selectivity of GPX4 (selectivity score; Y = 0.799) was much higher than that of MTOR (Y = 0.076), indicating that the essentiality of GPX4 is more cell line–specific. Importantly, when the cell lines were grouped by tumor types, AML was among those that are highly dependent on GPX4 (Fig. 1B). Meanwhile, consistently, MTOR exhibits high dependency not only in AML but more prominently in other tumor types (Fig. S1A). These analyses suggest that targeting GPX4 may have specific roles in AML.

Fig. 1: GPX4 is a potential therapeutic target in AML.
figure 1

A Efficacy/selectivity plot for the whole genes analyzed on shinyDepMap. The top 10 ferroptosis suppressor genes on FerrDb are annotated. B Dependency scores of GPX4 in all cell lines are clustered based on their lineages. A lower dependency score indicates a higher dependency on GPX4. Leukemia cell lines are subdivided into AML, B-cell ALL, and T-cell ALL, based on the original DepMap annotations. C GPX4 protein levels in 11 AML cell lines were determined by immunoblotting. The relative expression levels of GPX4 were assessed with densitometry as a ratio of GPX4 to α-tubulin (the loading control) and normalized to that of OCI-AML3. D, E GPX4 protein levels in peripheral blood mononuclear cells from 16 AML patients and bone marrow mononuclear cells from 4 healthy bone marrow donors (HD) were determined by immunoblotting (D). The relative expression levels of GPX4 were assessed with densitometry as a ratio of GPX4 to Histon H3 (the loading control), normalized to that of AML patient #17 that serves also as inter-membrane control, and plotted (E). High and low GPX4 bands suggest posttranslational modification, part of which appear as smears possibly due to partial degradation. For quantification, all the bands detected are included in the densitometry analysis. F GPX4 protein expression measured by mass spectrometry with tandem mass tag (TMT) labeling is plotted for AML samples (n = 44) and lineage-depleted normal bone marrow (NBM) samples (n = 3). Original data was obtained from a previous study [29].

Full size image

We next examined GPX4 protein expression in AML cells. In AML cell lines, the expression levels were found variable among 11 lines tested (Fig. 1C). Primary AML cells also exhibit distinct expression levels of GPX4 protein across 16 samples tested (Fig. 1D). GPX4 expression in normal bone marrow bulk mononuclear cells are significantly higher than that in AML cells (Fig. 1D, E). Of note, distinct levels of GPX4 bands were detected in some of the primary AML samples which may suggest post-translational modification of the protein [27], and some of the smear pattern which may reflect partial degradation of GPX4 protein. However, the exact nature of these bands remains unclear in the present study. We thus performed further validation and analyzed the differential expression of GPX4 in AML cells and normal cells using publicly available datasets with larger sample size. First, we utilized GEPIA platform, a web server for cancer and normal gene expression profiling [28]. We found that GPX4 gene expression of AML samples from The Cancer Genome Atlas (TCGA) dataset (n = 173) were significantly lower than that of normal bone marrow samples from The Genotype-Tissue Expression (GTEx) dataset (n = 70) (Fig. S1B). We also analyzed a recent proteome dataset comparing primary AML (n = 44) and normal bone marrow (n = 3) cells [29]. Consistently, GPX4 protein expression is significantly lower in AML samples compared to normal bone marrow samples (Fig. 1F).

Next, based on previous studies indicating a negative impact of elevated GPX4 gene expression on AML patients survival [30, 31], we analyzed TCGA dataset (n = 160) [32] and Beat AML dataset (n = 350) [33] to study the correlation. While TCGA shows worse survival in patients with high GPX4 expression (Fig. S1C), Beat AML did not show significant survival difference (Fig. S1D). In addition, analysis of a publicly available proteome dataset [34] revealed no significant survival difference between AML patients with high and low GPX4 protein levels (Fig. S1E). These analyses do not fully contradict the previous studies that reported the survival impact of GPX4 expression but also suggest that the correlation of GPX4 and AML patient survival should be more carefully concluded. Multiple aspects including confounding factors likely cause these distinct results, warranting further investigations to determine whether there are specific subpopulations of AML where GPX4 expression has a survival impact.

GPX4 suppression induces ferroptosis in AML cells in vitro and in vivo

To test the anti-leukemia effects of the pharmacological inhibition of GPX4, we utilized ML210, to date a well-established, proteome-wide specific inhibitor of GPX4 [35, 36]. We determined GPX4 inhibition–induced cell death as the percentage of cells positive for annexin V and/or DAPI to capture both apoptotic and non-apoptotic cell death including ferroptosis. In OCI-AML3 cells, ML210 increased both, annexin V–positive and DAPI-positive fractions, which were effectively suppressed by the lipophilic antioxidants liproxstatin-1 (Lip1) and α-tocopherol (aToc) as well as the iron chelator deferoxamine (DFO) (Fig. 2A). These data indicate that, although annexin V is usually used as a marker of apoptosis, annexin V–positive AML cells that result from GPX4 inhibition are ferroptotic. This is consistent with previous findings demonstrating that lipid peroxidation upon ferroptosis can cause phosphatidylserine externalization without caspase activation [37]. Consequently, the pan-caspase inhibitor z-VAD-FMK did not inhibit ML210-induced cell death (Figs. 2B and S2A, B). Lip1, aToc, and DFO consistently suppressed ML210-induced lipid peroxidation, which was detected with C11-BODIPY (Fig. 2C). The iron-dependent lipid peroxidation and cell death induced by ML210 were confirmed in MOLM-13 and OCI-AML2 cells as well (Fig. S2C–F).

Fig. 2: The pharmacological inhibition of GPX4 induces ferroptosis in AML.
figure 2

A, C OCI-AML3 cells were treated with 0.5 µM ML210 with or without 1 µM liproxstatin-1 (Lip1), 100 µM α-tocopherol (aToc), or 4 µM deferoxamine (DFO) for 48 hours. Cell death (A) and lipid peroxidation (C) were determined by annexin V (AnnV) and/or DAPI positivity or by C11-BODIPY581/591 staining, respectively, on flow cytometry. B OCI-AML3 cells were treated with 1 µM venetoclax (Ven) plus 0.2 µM AMG-176 (MCL-1 inhibitor) to induce apoptosis [65] or 0.2 µM ML210, with or without 25 µM z-VAD-FMK (zVAD), for 24 hours. D Parental AML cells were treated with ML210 for 24 hours, and specific cell death was calculated as described in Supplemental information. X axis represents ML210 concentration on a logarithmic scale. E The area under the cell killing curve (AUC) for ML210 (Fig. 2D) was correlated with the relative GPX4 expression level (Fig. 1C). Correlation coefficient (r) and statistical significance (p) on Pearson correlation analysis were indicated. F MOLM-13 cells with wildtype TP53 (WT), TP53 knockout (KO), or the indicated TP53 mutations were treated with ML210 for 72 hours. G Kasumi-1 cells transfected with shRNA targeting TP53 (shp53) or with scramble control shRNA (shC) were treated with ML210 for 72 hours. H Peripheral blood mononuclear cells from AML patients (n = 14) or bone marrow mononuclear cells from healthy bone marrow donors (HD) (n = 12) were treated with 10 µM ML210 for 48 hours. Cell death was determined for CD45+ bulk hematopoietic cell populations. I AUC for ML210 in primary AML cells (Fig. S2O) was correlated with the GPX4 expression level (Fig. 1D) for matching samples (n = 13).

Full size image

Next, we evaluated the efficacy of ML210 across 11 AML cell lines. Sub-micromolar concentrations of ML210 induced cell death in most cell lines (Fig. 2D). When analyzed with the GPX4 protein expression on Fig. 1C, their sensitivities were negatively correlated with the expression levels (Pearson correlation analysis; r = −0.739, p = 0.009) (Fig. 2E). Among the cell lines tested, MV-4-11 is relatively sensitive to ML210 despite its high GPX4 protein level, indicating that other cell line-specific mechanisms may also play roles in the determination of the sensitivity. On immunoblot, well-established ferroptosis regulators tested did not show any unique feature that could potentially increase the sensitivity in MV-4-11 cells (Fig. S2G, H). Further investigations for other molecules as well as their functional assays are required to elucidate the mechanisms involved. To assess the effects of endogenous GPX4 upregulation on ferroptosis, we utilized selenium to enhance GPX4 biosynthesis. Supplementing culture media with seleno-L-methionine (SLM) increased GPX4 expression levels in AML cells (Fig. S2I). We further generated Doxycycline (Dox)-inducible short hairpin RNA (shRNA) constructs to knockdown eukaryotic elongation factor, selenocysteine-tRNA specific (EEFSEC), which is required for selenoprotein translation [38]. The Dox-induced knockdown of EEFSEC abrogated SLM-induced GPX4 upregulation (Fig. S2J), confirming that GPX4 upregulation occurs through increased protein synthesis of GPX4. As expected, AML cells cultured in SLM-supplemented media are resistant to ML210 (Fig. S2K). GPX4-mediated suppression of ferroptosis through selenoprotein synthesis is summarized in Fig. S2L.

We also assessed the anti-leukemia effect of GPX4 inhibition in TP53-mutant AML cell lines, which served as an in vitro model of one of the most clinically challenging genetic subtypes of AML. ML210 had similar efficacy among isogenic MOLM-13 cells with TP53 deletion or hotspot mutations (Fig. 2F). Another GPX4 inhibitor, RSL-3, also showed the similar cell death induction in these cells (Fig. S2M). Immunoblot revealed that xCT expression was slightly higher in TP53 knockout (KO) or mutant cells (Fig. S2N) compared to TP53 wild-type (WT) cells. This is consistent with the previous reports showing that TP53 is a negative regulator of SLC7A11 [39, 40] while xCT protein levels did not correlate the cell sensitivities to ferroptosis in these conditions. The knockdown of mutant TP53 in Kasumi-1 cells harboring the TP53R248H mutation also did not alter sensitivity to ML210 (Fig. 2G). Taken together, GPX4 inhibition exerts anti-leukemia effects independently of TP53 mutation status.

We then treated primary cells derived from AML patients (n = 14) or healthy bone marrow donors (n = 12) with ML210. AML patients included those who relapsed after or developed resistance to multiple regimens and/or those with TP53 mutations (Table S1). Most of the primary AML samples tested were sensitive to ML210 (Figs. 2H and S2O). Three samples showed resistance to ML210, but we could not identify any common clinical features explaining the resistance, including prior regimens, cytogenetics, or mutations. A correlation analysis of AML cell sensitivity to ML210 and GPX4 protein levels in primary AML samples (n = 13) showed negative correlation (Fig. 2I), which suggests that GPX4 protein expression is a potential predictor of AML cell sensitivity to GPX4 inhibition. Independently, cell death induced in CD45+ mononuclear hematopoietic cells from healthy bone marrow donors (HD) was significantly lower than that induced in AML patient samples (Fig. 2H). Besides, we found that the effective dose of ML210 was higher in primary cells (at the levels of micromolar) compared to cell lines, for which the mechanisms are yet to be investigated. Of note, cell death induced by higher dose of ML210 in primary cells was inhibited by Lip1 (Fig. S2P), consistent with the on-target effects of ferroptosis induction.

We next introduced Dox-inducible shRNAs targeting GPX4 (shGPX4) into OCI-AML3 cells (OCI-AML3-shGPX4-1 or -2 cells) (Fig. 3A). Treatment with Dox for 96 hours (shGPX4-1) or 72 hours (shGPX4-2) resulted in lipid peroxidation (Figs. 3B and S3A), followed by cell death induction at 120 hours (Fig. 3C), demonstrating that GPX4 downregulation induces lipid peroxidation prior to the cell death. The findings were confirmed in MOLM-13 cells and OCI-AML2 cells transfected with the same shRNAs (Figs. 3D, E and S3B, C). Cell death was induced already at 96 hours post Dox treatment in these two cell lines, perhaps partly due to lower baseline GPX4 expressions compared to that in OCI-AML3 cells. In addition, GPX4 knockdown sensitized OCI-AML3-shGPX4 cells to ML210 at a dose that was not effective by itself (Fig. 3F), consistent with the aforementioned negative correlation between GPX4 protein expression and sensitivity to GPX4 inhibition. Cell death induced by GPX4 knockdown was abrogated by Lip1 and aToc but not by DFO (Fig. S3D, E). Of note, AML cells did not tolerate the dose range of DFO used in solid cancer ferroptosis studies (up to 100 μM) [6, 41] due to its inherent anti-leukemia effects (Fig. S3F). Given that lipid peroxidation induced by shGPX4 (Figs. 3B and S3A) was more profound than that by ML210 treatment (Fig. 2C), the lower dose of DFO in AML cells was possibly insufficient to block the robust ferroptosis induced by GPX4 knockdown.

Fig. 3: The genetic knockdown of GPX4 induces ferroptosis in AML in vitro and in vivo.
figure 3

A OCI-AML3 cells transfected with shRNAs targeting GPX4 (shGPX4) or with control shRNA (shC) were treated with 1 µg/ml doxycycline (Dox) for 72 hours, and GPX4 protein levels were determined by immunoblotting. B OCI-AML3-shGPX4-1 cells were treated with 1 µg/ml Dox for 96 hours, and cell death (left panel) and lipid peroxidation (right panel) were determined by flow cytometry. C–E OCI-AML3-shGPX4 (C), MOLM-13-shGPX4 (D), and OCI-AML2-shGPX4 (E) cells were treated with Dox for 120 hours (C) or 96 hours (D, E). F OCI-AML3-shGPX4 cells were treated with 0.1 µg/ml Dox for 72 hours in combination with 0.1 µM ML210 for the last 24 hours. G Survival curves of the NSG mice transplanted with luciferase-labeled OCI-AML3-shGPX4-2 cells and treated with vehicle or tetracycline (n = 7 for each group) were plotted. H OCI AML3-shGPX4-2 cells were harvested from moribund mice after treatment with vehicle or tetracycline for 6–8 weeks (n = 3 for each group) and purified for AML cells as described in Supplemental information. GPX4 protein levels were determined by immunoblotting.

Full size image

To determine the anti-leukemia effect of GPX4 inhibition in vivo, we labeled OCI-AML3-shGPX4-2 cells with luciferase and transplanted them into NSG mice. After engraftment was confirmed by bioluminescent imaging (BLI) (Fig. S3G), the mice were given regular water (vehicle) or tetracycline water to knockdown GPX4 in leukemia cells (n = 7/group). Tetracycline treatment significantly prolonged mouse survival (median survival 74 days vs 55 days) (Fig. 3G), confirming that GPX4 inhibition exerts anti-leukemia effects in vivo. Knockdown of GPX4 protein was verified for the leukemia cells harvested from mice on immunoblotting (Fig. 3H). The same tetracycline treatment did not extend survival of mice transplanted with OCI-AML3 cells with control shRNA (Fig. S3H).

Mitochondrial lipid peroxidation and electron transport chain complexes regulate AML cell ferroptosis

Because mitochondria are the centers of cellular redox and iron metabolism, we next investigated the involvement of mitochondria in AML cell ferroptosis. Morphologically, evaluation by transmission electron microscopy showed that, compared with control cells, ML210-treated OCI-AML3 cells had smaller mitochondria with an electron-dense mitochondrial matrix (Fig. 4A). This is consistent with previous ferroptosis studies in non-leukemia models [6, 39]. ML210-treated OCI-AML3 cells also had disorganized cristae with “vesicular” morphology, a phenomenon similar to the finding reported to occur during apoptosis induction in HeLa cells [42]. Functionally, ML210 increased mitochondrial superoxide production, which was largely blocked by Lip1, aToc, and DFO, in AML cells (Figs. 4B, C and S4A, B). This suggests that the mitochondrial superoxide induced by GPX4 inhibition is mostly iron-dependent lipid peroxidation. Indeed, flow cytometry with MitoPerOx staining, which specifically detects lipid peroxidation in mitochondria, revealed that ML210 consistently increased mitochondrial lipid peroxidation (Fig. 4D). Dox-induced GPX4 knockdown also increased mitochondrial superoxide production and lipid peroxidation in OCI-AML3 cells (Figs. 4E and S4C) and MOLM-13 cells (Fig. S4D).

Fig. 4: Mitochondrial lipid peroxidation and electron transport chain complexes regulate AML cell ferroptosis.
figure 4

A OCI-AML3 cells were treated with 0.5 µM ML210 or with DMSO as a control for 6 hours and then subjected to transmission electron microscopy to assess the mitochondrial ultrastructure. B, C OCI-AML3 cells were treated with 0.5 µM ML210 with or without 1 µM liproxstatin-1 (Lip1) (B), 100 µM α-tocopherol (aToc) (B), or 4 µM deferoxamine (DFO) (C) for 48 hours. Mitochondrial superoxide production was determined by MitoSOX staining and flow cytometry. The median fluorescence intensity (MFI) of the treated cells was normalized to that of the untreated cells. D OCI-AML3 cells were treated with 0.5 µM ML210 for 48 hours, and mitochondrial lipid peroxidation was determined by MitoPerOx staining and flow cytometry. E OCI-AML3-shGPX4-1 cells were treated with 1 µg/ml doxycycline (Dox) for 96 hours, and mitochondrial superoxide (left) and lipid peroxidation (right) were determined by flow cytometry. MEFs (F) or OCI-AML3 cells (G) were treated with 2.0 µM ML210 in combination with increasing concentrations of MitoQ or DecylQ as indicated on the x axis for 24 hours. Cell viability was determined as the percentage of annexin V– and DAPI-negative cells on flow cytometry. H OCI-AML3 cells were treated with 0.5 µM ML210 with or without 0.1 µM MitoQ or 10 µM MitoTEMPO for 48 hours. I OCI-AML3-shGPX4-1 cells were treated with 1 µg/ml Dox with or without 0.1 µM MitoQ or 10 µM MitoTEMPO for 120 hours. J Lysates from HL-60-RhoWT cells or -Rho0 cells were subjected to immunoblotting for the indicated proteins. HL-60-Rho0 cells or -RhoWT cells were treated with ML210 for 48 (K) or 24 (L) hours.

Full size image

A previous study using mouse embryonic fibroblasts (MEFs) demonstrated that the mitochondria-targeting antioxidant mitoquinone mesylate (MitoQ) is far less potent than the non-targeting antioxidant decylubiquinone (DecylQ) in protecting cells from GPX4 inhibition–induced ferroptosis [43]. This suggests that ferroptosis usually arises from a redox imbalance in cytoplasm but not in mitochondria. To further investigate the involvement of mitochondria in leukemia cells, we compared the anti-ferroptosis effect of MitoQ in MEFs and OCI-AML3 cells. The expression of GPX4 protein in those cells were determined by immunoblotting (Fig. S4E). We confirmed that the concentration of MitoQ required in MEFs to block ML210-induced cell death was more than 5-fold that of DecylQ (Fig. 4F), which is consistent with the previous report [43]. In contrast, and unexpectedly, MitoQ was as potent as DecylQ in blocking ferroptosis in OCI-AML3 cells treated with ML210 (Fig. 4G), suggesting that AML cell ferroptosis is regulated predominantly by mitochondrial, rather than cytoplasmic, lipid peroxidation. In addition to MitoQ, MitoTEMPO, another mitochondria-targeted antioxidant, also blocked cell death caused by ML210 treatment or by Dox-induced GPX4 knockdown (Figs. 4H, I and S4F, G), further supporting the mitochondrial regulation of ferroptosis in AML cells.

Next, to investigate the role of mitochondrial respiration, the center of mitochondrial redox metabolism, in AML cell ferroptosis, we studied AML cells with deficient electron transport chain (ETC) complex through chronic exposure to ethidium bromide to specifically deplete mitochondrial DNA (Rho0 cells). Since OCI-AML3 cells were not tolerated to Rho0 status (data not shown), we utilized well-established HL-60-Rho0 cells [44]. Immunoblotting revealed that Rho0 cells had non-detectable levels of ETC complex proteins (subunits in complexes I, II, III, and IV) and lower level of GPX4 expression (Fig. 4J). Notably, HL-60-Rho0 cells are nearly 10 times as sensitive to ML210 as their wild type (WT) controls (HL-60-RhoWT) (Fig. 4K). Furthermore, prominent mitochondrial lipid peroxidation was induced in Rho0 cells by almost 10-fold lower concentration of ML210 as RhoWT cells, confirming that Rho0 cells are more sensitive to GPX4 inhibition–induced mitochondrial lipid peroxidation in the AML cell line (Fig. 4L). Although GPX4 downregulation could also contribute to higher sensitivity to GPX4 inhibition in Rho0 cells, we reasoned that the intact expressions of ETC complex proteins may protect AML cells from GPX4 inhibition–mediated ferroptosis, perhaps uniquely in AML cells.

GPX4 inhibition–mediated ferroptosis is synergistically enhanced by hyperactivation of mitochondrial caseinolytic protease P

Given Rho0 AML cells’ higher ferroptosis sensitivity, we investigated the therapeutic potential of a combination of ferroptosis induction and ETC downregulation in AML. To this end, we hyperactivated the mitochondrial protease caseinolytic protease P (ClpP) by imipridones (ONC201 or ONC212) to induce the degradation of ETC subunits and exert cancer-selective lethality [45]. Indeed, the combination of ML210 and ONC201 induced synergistic cell death in OCI-AML2 and MOLM-13 cells (Fig. 5A). The combinatorial anti-leukemia effect was associated with the cumulative induction of lipid peroxidation and mitochondrial superoxide production (Fig. 5B, C). Next, we studied if the synergistic anti-leukemia effect is associated with changes in GPX4 protein. Immunoblot of GPX4 showed a size shift for all the ML210-treated samples (Fig. S5A), which reflects covalent binding of ML210 or its metabolite, as reported previously [36]. Meanwhile, changes in total GPX4 protein expression levels are distinct among three AML cell lines, suggesting that alteration in GPX4 protein expression by the treatment is not the primary mechanisms responsible for the synergistic anti-leukemia effects.

Fig. 5: GPX4 inhibition–mediated ferroptosis is synergistically enhanced by ClpP hyperactivation.
figure 5

A OCI-AML3, MOLM-13, and OCI-AML2 cells were treated with ONC201 and/or ML210 for 72 hours, and cell death was determined by flow cytometry. A combination index (C.I.) less than 1.0 indicates synergistic effects. OCI-AML3 cells were treated with ONC201 and/or ML210 for 48 hours, and lipid peroxidation (B) and mitochondrial superoxide production (C) were determined by flow cytometry. D Primary cells from AML patients (n = 8) were treated with ML210 and/or ONC201 for 48 hours. Combination indexes were calculated based on specific cell death in CD45+ population upon the combinatorial treatment for each sample and were plotted. One sample was excluded from C.I. assessment because ONC201 was completely ineffective and thus C.I. calculations were not applicable. The sample from patient #12, which was resistant to a single treatment with ML210 on Figs. 2H and S2O, is indicated by arrows. E Primary cells from AML patients (n = 8) or healthy bone marrow donors (HD) (n = 8) were treated with 10 μM ML210 and 5 μM ONC201 for 48 hours, and cell death was determined for CD45+ populations. The sample from AML patient #12 is indicated by an arrow. F OCI-AML3-CLPP-Y118A cells were treated with doxycycline (Dox) for 144 hours in combination with ML210 for the last 72 hours. OCI-AML3-shGPX4-1 cells were treated with Dox for 120 hours in combination with ONC201 (G) or ONC212 (H) for the last 72 hours. I A genome-wide CRISPR knockout screening performed on NALM6 cells treated with ClpP agonists [46] were re-analyzed and sgRNA frequency scores (RANKS scores) were plotted. J OCI-AML3, OCI-AML2, and MOLM-13 cells were treated with 2 µM ONC201 for up to 48 hours, and GPX4 expression was determined by immunoblotting. K HL-60-Rho0 cells or -RhoWT cells were treated with ML210 with or without 0.25 µM ONC212 for 48 hours.

Full size image

The combinatorial treatment was also tested in primary samples. Among the 14 primary AML samples tested for ML210 single treatment (in Fig. 2H), 8 samples (including one sample exhibiting resistance to ML210 alone) were tested for the combinatorial treatment as well. Combination indexes were below 1.0 for most of the treatment conditions, indicating synergistic anti-leukemia effects (Fig. 5D). The synergism was also observed for the sample that exhibited resistance to ML210 alone (indicated by an arrow in Fig. 5D). Importantly, primary cells from healthy bone marrow donors (n = 8) exhibit significantly less cell death upon the combinatorial treatment compared to those from AML patients (Figs. 5E and S5B).

Furthermore, synergistic cell death was induced when the genetic activation of ClpP using a Dox-inducible CLPP-Y118A hyperactivated mutant was combined with ML210 and when Dox-inducible GPX4 knockdown was combined with ONC201 and ONC212 (Figs. 5F–H and S5C, D). We also tested the combinatorial anti-leukemia effect of GPX4 knockdown and ONC212 in vivo. A tendency of extended survival in the combination arm was observed, however, the difference compared to the other arms was not statistically significant (Fig. S5E).

Supporting the combinatorial effects of GPX4 inhibition and ClpP hyperactivation, our re-analysis of a previously published CRISPR screening data of NALM6 leukemia cells treated with imipridones [46] revealed GPX4 to be one of the top hits; knockdown of GPX4 sensitized cells to ClpP hyperactivation (Fig. 5I). In addition to GPX4, genes essential for selenoprotein synthesis (SEPSECS, EEFSEC, PSTK, SEPHS2) [47] were also top hits, suggesting that the inhibition of GPX4’s function as a selenoprotein is important for the synthetic lethality of GPX4 inhibition and ClpP hyperactivation. We also found that ClpP hyperactivation by ONC201 upregulates GPX4 protein expression (Fig. 5J), suggesting that GPX4 exerts a cell-protective response against mitochondrial proteotoxic stress in AML cells. ClpP hyperactivation sensitized ETC-intact RhoWT cells to GPX4 inhibition, but the synergism was not exerted in Rho0 cells (Fig. 5K). This indicates that the induced synergism depends on ETC, supporting the notion that the regulated protein expression of ETC subunits helps protect cells from ferroptosis. Of note, pharmacological inhibition of each ETC complex (complex I by IACS-010759 [48], complex III by Antimycin-A, or complex V by Oligomycin-A) did not enhance but rather inhibited ML210-induced cell death (Fig. S5F). These data suggest that functional inhibition of single ETC complex does not necessarily sensitize AML cells to ferroptosis. Since Rho0 cells are characterized not merely by the impaired ETC function but also structural abrogation of ETC involving multiple complexes, loss of the intact 3D structures and loss of functions of multiple ETC complexes by imipridone-mediated degradation may similarly matter for the induction of synergism.

Discussion

An increasing number of studies have been reported regarding the roles of mitochondria and ETC complex in ferroptosis across various cancer models, resulting in diverse conclusions [6, 49,50,51]. This is perhaps due to the highly context-dependent nature of ferroptosis [52, 53]. Findings of the present study suggest that intact ETC complexes regulate GPX4 inhibition–mediated ferroptosis in AML cells in a manner distinct from that of other tumor types. Specifically, HL-60-Rho0 cells with ETC defects were more sensitive to GPX4 inhibition compared to their parental cells. This is in clear contrast with previous studies demonstrating similar or impaired efficacy of the classical GPX4 inhibitor RSL3 in Rho0 status of 143B osteosarcoma cells [6] or SK-Hep1 liver sinusoidal endothelial cells [49], respectively. Similarly, another study reported that, in HT-1080 fibrosarcoma cells, mitophagy-mediated mitochondrial depletion blocked erastin-induced ferroptosis while had no effect on RSL3-induced ferroptosis [50]. A CRISPR knockout screening for K562 chronic myeloid leukemia cells revealed synthetic lethal interactions between GPX4 deletion and ETC inhibition [51]. In the meanwhile, our data suggest that inhibitions of a single ETC complex rather reduced AML cell sensitivity to ferroptosis. The sensitivity was increased only when several complexes of ETC were simultaneously and structurally abrogated (by Rho0 condition or ClpP hyperactivation). Of note, in the CRISPR screening study of K562 cells, they demonstrate that the synthetic lethal interaction was only observed when cells are cultured with spend media but not regular media, which may have led to the difference with our data. Taken together, our findings suggest that AML cells may be distinctively dependent on ETC to protect themselves from ferroptosis.

This mechanism of ETC-dependent protection from ferroptosis could be a unique vulnerability in AML in which GPX4 has been inhibited. Indeed, we observed that the combination of GPX4 inhibition and ClpP activation–mediated degradation of ETC proteins had synergistic anti-leukemia effects in AML cells. We acknowledge that ClpP targets not only ETC proteins but also other proteins, including those involved in mitochondrial translation or metabolic pathways [45], and the inhibition of such proteins could contribute to the cells’ increased vulnerability to GPX4 inhibition–mediated ferroptosis. We had previously reported that ClpP hyperactivation induces atypical integrated stress response (ISR) and ATF4 upregulation [45, 54] and a recent study shows that ISR-mediated ATF4 activation sensitizes cells to GPX4 inhibition through alteration of GSH metabolism [55]. Involvement of these pathways should also be evaluated in future studies. Nevertheless, the combinatorial effects of ClpP hyperactivation and GPX4 inhibition observed in parental AML cells but not in ETC-deficient Rho0 cells strongly suggest that this synergism and the ferroptosis protection mechanism both depend on intact ETC complexes.

Furthermore, we demonstrated that MitoQ had unexpectedly potent anti-ferroptotic effects in AML cells but, as reported previously, not in MEFs. The importance of mitochondria-localized CoQ is also in line with ETC’s protective effects against mitochondrial lipid peroxidation and resultant ferroptosis because ETC complexes reduce mitochondria-localized CoQ to CoQH2, which exerts anti-oxidative effects [56]. MitoQ has been reported to have similar protective effects in erastin- or RSL3-treated SK-Hep1 liver sinusoidal endothelial cells, in which induced ferroptosis was also associated with mitochondrial lipid peroxidation [49]. In addition, recent studies by Boyi Gan’s group demonstrated that DHODH and GPD2 on the mitochondrial membrane, both regulate the recycling of mitochondria-localized CoQ in collaboration with ETC, protect against ferroptosis [10, 11]. Although the functional relevance of these non-ETC CoQ-regulating enzymes in AML cell ferroptosis remains to be investigated, we speculate that the maintenance of a reduced form of mitochondrial CoQ primarily contributes to ETC’s protective role against ferroptosis in AML cells.

Lastly, here we have reported significant in vivo anti-leukemia activity of GPX4 inhibition using a genetic approach. At present, none of the publicly available GPX4 inhibitors are fully developed for use in in vivo studies due to their suboptimal bioavailability in animal models [36]. The development of specific GPX4 inhibitors with superior pharmacokinetic and pharmacodynamic profiles in vivo is necessary for the clinical implementation of GPX4-targeted therapies. In parallel, potential toxicity of GPX4 inhibitors in non-hematopoietic systems must be evaluated carefully, as our genetic model does not reflect systemic GPX4 inhibition. Our in vitro data suggests preferential cell death induction by GPX4 inhibition in primary AML cells compared to normal BM cells, possibly due to their differential expression levels of GPX4 protein. Multiple genetic models also support the tolerability of conditional GPX4 deletion in normal hematopoiesis [57, 58] and hematopoietic stem cells [59], while a recent paper reported sensitivity of human normal hematopoietic stem cells to ferroptosis [60], warranting further investigation especially in vivo. GPX4 has been reported not only as essential for early embryonic development in mice [61, 62] but also as linked to a rare neonatal lethal disorder, Sedaghatian-type spondylometaphyseal dysplasia, when loss-of-function mutations occur in humans [63]. These studies alert that inhibition of GPX4 is potentially toxic to other organs when systemically targeted. To avoid the potential toxicity, development of cancer-specific targeting strategy for GPX4 inhibition (e.g., antibody-drug conjugates to target a cancer-specific surface protein) would be one of the most effective ways. Developing combinatorial strategies to minimize the dose of GPX4 inhibitors could be another solution, particularly by targeting other cancer-specific molecular mechanisms. Together with a recent study suggesting the combinatorial inhibition of GPX4 and FLT3 in FLT3-mutant AML [23], the present study suggests that targeting the mitochondrial regulation of ferroptosis could potentiate the therapeutic efficacy of GPX4 inhibition in AML.

Materials and methods

Detailed materials and methods can be found in the Supplementary information.

Cell lines and cell culture

Parental AML cells as well as HEK293T cells and MEFs were obtained and cultured as described in Supplemental information. OCI-AML3-CLPP-Y118A cells were established by us and described previously [45]. CRISPR-engineered isogenic MOLM-13 cells with TP53 WT, TP53 KO, or TP53 R175H or R248Q mutants were provided by Steffen Boettcher (University of Zurich, Zurich, Switzerland) [64]. HL-60-Rho0 and -RhoWT cells were provided by Peng Huang (Department of Molecular Pathology, MD Anderson Cancer Center) [44].

Primary samples

The mononuclear cells were purified from peripheral blood of AML patients (blast >60%) and bone marrow of healthy hematopoietic stem cell donors by Ficoll density-gradient centrifugation and were cultured in RPMI 1640 culture media (#MT10040CV, Corning, Corning, NY) supplemented with 100 ng/ml human Flt3-ligand (#300-19, Peprotech, Cranbury, NJ), 100 ng/ml human SCF (#300-07, Peprotech), 20 ng/ml human TPO (#300-18, Peprotech), and 20 ng/ml human IL-3 (#200-03, Peprotech), as well as 100 U/ml penicillin, 100 µg/ml streptomycin, 2 mM L-glutamine, and 10% heat-inactivated fetal bovine serum.

Immunoblot analyses

For the immunoblot analyses, we used following antibodies; anti-GPX4 (#ab125066, Abcam, Cambridge, UK), anti-α-tubulin (#2125; Cell Signaling), anti-Histon H3 (#3638, Cell Signaling, Danvers, MA), anti-β-actin (#A5316, MilliporeSigma, Darmstadt, Germany, or #4970, Cell Signaling), anti-AIFM2 (FSP1) (#HPA042309, MilliporeSigma), anti-SCD1 (#PA5-19682, Thermo Fisher Scientific, Waltham, MA), anti-DHODH (#14877-1-AP, Proteintech, Rosemont, IL), anti-XCT/SLC7A11 (#12691, Cell Signaling), anti-ACSL4 (#sc-271800, Santa Cruz Biotechnology), anti-EEFSEC (#HPA035795, MilliporeSigma), anti-p53 (#sc-126, Santa Cruz Biotechnology, Dallas, TX), and total oxidative phosphorylation rodent antibody cocktail (#ab110413, Abcam).