Search
Close this search box.

Phenolic compounds induce ferroptosis-like death by promoting hydroxyl radical generation in the Fenton reaction – Communications Biology

Identification of E. coli genes essential to PG tolerance

E. coli BL21(DE3) was grown in minimal salt medium (MSM) to late-exponential phase (OD600 2.5, Supplementary Fig. 1), and then treated with 0.5, 1.0, 2.0, and 4.0 g/L PG, respectively. About 5% cells survived after challenge of 0.5 g/L PG, and all cells were killed with presence of 4 g/L PG (Fig. 1a), showing that PG is highly toxic to E. coli. Adaptive laboratory evolution is a powerful method to elevate bacteria tolerance against environmental stresses and to identify genes essential to stress tolerance30,31. In this study, adaptive evolution was carried out using the GREACE method32, in which mutated dnaQ gene (encoding proofreading factor of the DNA polymerase) was introduced into BL21(DE3) strain for in vivo continuous mutagenesis. After 52 rounds of selection, PG concentration increased from 1 g/L to10 g/L (Fig. 1b), reaching its solubility, and the survival culture was spread onto LB agar plate to isolate single colonies. Then six colonies with the highest PG tolerance, which also showed improved phenol tolerance, were designated M01-06 and subjected to genome sequencing along with the parent strain Q3505.

Fig. 1: Identification of E. coli genes related with PG tolerance using adaptive laboratory evolution.
figure 1

a Tolerance of E. coli BL21(DE3) strain after treatment with PG at different concentrations for 4 h. A representative result was shown. b PG concentrations used in each round of adaptive evolution process. c Tolerance of E. coli BL21(DE3) wild-type strain and defined knockout and SNP mutants after treatment with PG at 1.3 g/L for 4 h (n = 3 biological independent samples). d Schematic diagram showing qmcA and fetAB genomic locations and sequence of their shared promoter region. The SNP mutation, previously reported fetA transcription starting site S1, newly identified initiation site S2 and its corresponding -35 and -10 regions were shown. e Relative mRNA level of qmcA, fetA and fetB determined by qRT-PCR in BL21(DE3) wild-type strain and qmcA-fetA SNP mutant (n = 4 biological independent samples with two technical repeats). f Tolerance of BL21(DE3) strains carrying empty vector, pqmcA and pfetAB after treatment with PG at 1.3 g/L for 4 h (n = 4 biological independent samples). Two-tailed Student’s tests were performed to determine the statistical significance for two group comparisons. Error bars, mean ± standard error of mean (SEM).

The sequences of both evolved mutants and the parent strain Q3505 were aligned to the E. coli BL21(DE3) reference sequence (NC_012978.2) to figure out the genetic mutations arose during evolution process. Totally, 28 mutations were identified in these six strains including 13 single-nucleotide polymorphisms (SNPs) and 15 insertions/deletions (Supplementary Table 1). Among these mutations, SNP in zipA gene resulted in a synonymous variant, gene B encodes a λ phage capsid protein, and lacZ, rbsD and insB27 are nonfunctional pseudogenes, so mutations related with these five genes were omitted from further characterization. To identify the contribution of each mutation to the improved PG tolerance, knockout mutants of genes fhuA, clpX, flgK, sodB, arcZ and yigB, site-directed mutants apaH P237Q and ftsZ R214H, and SNP mutants in the intergenic regions of qmcA-fetA, nfuA-gntT and yshB-glnG were constructed in the parent strain Q3505, respectively. Following PG tolerance assay, three mutations (ΔclpX, ΔsodB and qmcA-fetA SNP) proved of great assistance in the tolerance improvement (Fig. 1c).

As shown in Fig. 1d, qmcA gene and fetAB operon are divergent and the SNP locates in their shared promoter region. qRT-PCR results demonstrated that this SNP enhanced the mRNA level of fetAB about 20 times while it has no effect on qmcA transcription (Fig. 1e). Furthermore, qmcA and fetAB were cloned into vector pTrcHis2B and introduced into wild-type strain, respectively. The strain carrying pqmcA presented a survival rate similar with strain carrying empty vector, whereas overexpression of fetAB amazingly restored the viability of E. coli BL21(DE3) strain in treatment with PG (Fig. 1f).

The transcription start site of fetA in SNP mutant was mapped using RACE experiment. Different with the start site S1 reported previously33, fetA has a new transcription initiation site S2 located at 27 bp upstream of the start codon, and the SNP site is exactly situated in the putative -35 box of S2 promotor (summarized in Fig. 1d), which is the reason why this SNP mutation increased the transcription level of fetAB gene. Following, the function of clpX, sodB, and fetAB was confirmed by knockout and complementation (Supplementary Fig. 2). In summary, three genes/operon clpX, sodB and fetAB are essential to bacterial PG tolerance.

Iron and hydrogen peroxide are required for PG toxicity

The fetAB operon encodes an iron exporter, and overexpression of fetAB reduced intracellular iron levels in E. coli (Fig. 2a). Taking into account that BL21(DE3) strain carrying pfetAB are super tolerant to PG (Fig. 1f), iron is supposed to facilitate the PG toxicity. To illustrate iron’s role in PG killing, the cell-permeable iron chelator desferrioxamine was used to block intracellular iron. It turned out that desferrioxamine efficiently protected BL21(DE3) cells from treatment with PG (Fig. 2b). Subsequently, iron-free medium was used in PG killing assay, and iron and PG were supplemented into the medium alone or together. As shown in Fig. 2c, E. coli BL21(DE3) cells were killed only with the presence of both iron and PG. Furthermore, the cell viability declined in an iron concentration-dependent manner after exposure to 1.3 g/L PG and different concentration of iron (Fig. 2d). Hence, iron is essential for the PG toxicity, and fetAB overexpression enhanced exportation of intracellular iron, resulting in improved bacterial PG tolerance.

Fig. 2: Iron and H2O2 were required for PG toxicity.
figure 2

a Intracellular iron concentration in BL21(DE3) strain carrying empty vector and pfetAB (n = 3 biological independent samples). b Survival rates of BL21(DE3) wild-type strain after treatment with PG at 1.3 g/L for 4 h with the absence and presence of the cell-permeable iron chelator desferrioxamine (DFO) (n = 3 biological independent samples). c Survival rates of BL21(DE3) wild-type strain in iron-free medium after challenge of PG, Fe, and both of them (n = 5 biological independent samples). d Survival rates of BL21(DE3) wild-type strain after exposure to 1.3 g/L PG and different concentrates of iron (n = 3 biological independent samples). e Survival rates of BL21(DE3) wild-type strains carrying empty vector, pkatE and pkatG after treatment with PG at 1.3 g/L for 4 h (n = 3 biological independent samples). f Survival rates of BL21(DE3) wild-type strain and katE, katG, and katE katG mutants after treatment with PG at 1.3 g/L for 4 h (n = 3 biological independent samples). g Survival rates of the sodB mutant after challenge of PG, H2O2, and both of them (n = 3 biological independent samples). h Survival rates of BL21(DE3) strain and knockout mutants of genes encoding SODs after treatment with PG (n = 3 biological independent samples). Two-tailed Student’s tests were performed to determine the statistical significance for two group comparisons. Error bar, mean ± SEM.

The sodB gene encodes a FeSOD, and E. coli also has two other SODs: manganese- and copper, zinc-cofactored SODs (MnSOD and CuZnSOD encoded by sodA and sodC genes, respectively)34. It is confusing that ΔsodB mutant presented higher PG tolerance as SOD genes were upregulated upon phenol challenge and considered as antioxidant enzymes to resist phenol-induced oxidative stress in previous studies23,28. Superoxide radicals form as a harmful byproduct of aerobic metabolism, and is degraded by the SODs to hydrogen peroxide which is further converted into dioxygen by catalase KatG and KatE34. Based on the above information, the PG-tolerant sodB mutant should have a much lower intracellular concentration of hydrogen peroxide than wild-type strain, indicating that hydrogen peroxide is required for PG toxicity.

To test this hypothesis, katE and katG genes were cloned into plasmid vector pTrcHis2B and overexpressed in BL21(DE3) wild-type strain respectively, leading to a great enhancement of PG tolerance of this strain (Fig. 2e). In addition, katE katG double mutant was more susceptible to PG than wild-type strain (Fig. 2f). Furthermore, hydrogen peroxide was added into medium in sodB mutant’s PG killing assay. While neither hydrogen peroxide nor PG killed cells alone, their combination decreased the survival rate of sodB mutant dramatically (Fig. 2g). Next, knockout mutants of the other two genes encoding SODs were constructed, and the survival rate of sodC mutant was found to be 10 times higher than that of wild-type strain while the sodA mutant did not affect the PG susceptibility significantly (Fig. 2h). All these results demonstrated that hydrogen peroxide is required for PG toxicity, and SODs impaired E. coli tolerance to PG as it produces hydrogen peroxide.

PG-iron complex promotes the generation of hydroxyl radical

H2O2 can be reduced by iron via Fenton reaction to HO· that causes severe DNA, protein, and lipid damage35. In vivo experiments in E. coli suggested that DNA damage required H2O2 at a concentration of 1 to 2 mM36, which could only be achieved in catalase deficient E. coli strain35. Furthermore, the majority of iron inside the cells forms complex with iron-storage proteins FtnA, Bfr, and Dps that limit the potential for iron-dependent HO· formation35,37,38. So, the Fenton reaction is not significant in normal E. coli cells.

As reported previously, the generation of HO· in Fenton reaction can be promoted by a series of complexes composed of iron and salen/salophene, in which complexation of iron takes place by phenolic groups39,40. Therefore, it was wondered whether PG could complex with iron and stimulate the Fenton reaction. In FeCl2 solution with a pH of 7.0, Fe2+ could be oxidized to Fe3+ by dissolved oxygen, forming light brown flocs Fe(OH)3. With addition of PG gradually, the precipitate got dissolved to form homogeneous brown aqueous solution (Fig. 3a). This phenomenon was believed to be caused by the complexation between Fe3+ and PG, which was confirmed by the following UV-visible absorption assay (Fig. 3b) and NMR analysis (Fig. 3c and Supplementary Fig. 3).

Fig. 3: PG-iron complex promoted the generation of hydroxy radicals (HO·) in Fenton reaction.
figure 3

a Characteristics of ferrous chloride solutions at initial pH 7.0 with the absence and presence of PG. b UV-visible absorption spectra of solutions containing ferrous chloride alone, PG alone, and ferrous chloride and PG coupled at pH 7.0. c 1H NMR spectra of PG and PG-iron complex in acetonitrile-d3 solution. d Decolorization of Malachite Green with Fenton reaction containing 36 μM PG-iron complex, 36 and 360 μM FeCl3, respectively (n = 3). The decolorization efficiency after treatment of 5 min, 120 min, and 120 min with HO· scavenger t-butanol was shown. e Intracellular HO· concentration determined using hydroxyphenyl fluorescein of BL21(DE3) wild-type strain grown in iron-free medium after challenge of PG, Fe, and both of them (n = 5 biological independent samples). f Survival rates of BL21(DE3) wild-type strain after challenge of PG and Fe with dimethyl sulfoxide (DMSO) at different concentrations (n = 4 biological independent samples). Two-tailed Student’s tests were performed to determine the statistical significance of two group comparisons. Error bar, mean ± SEM.

To test whether PG-iron complex can enhance the production of HO· in Fenton reaction, Malachite Green (MG) decolorization experiment was carried out, in which MG will be oxidized by HO· and turned into colorless. As shown in Fig. 3d, the MG decolorization rate reached to 50.3% after 5 min with the Fenton reaction system containing 36 μM PG-iron complex, 2.2- and 1.7-time higher than with normal Fenton systems containing 36 and 360 μM FeCl3, respectively. After 120 min, 94.5% of MG were oxidized by the PG-iron complex-containing Fenton system, but more than two thirds of MG remained with the same concentration of FeCl3. To confirm that MG is oxidized by HO· rather than other substances, the HO· scavenger t-butanol was added into the MG decolorization reaction, and severely inhibited the oxidation of MG (Fig. 3d). All these results demonstrated that formation of the PG-iron complex could promote the production of HO· in Fenton reaction dramatically.

To identify the effect of PG-iron complex in vivo, E. coli BL21(DE3) strain was cultured in iron-free minimal medium supplemented with iron and/or PG, and the intracellular levels of HO· were determined. The cells grown with both iron and PG presented a HO· concentration at least 230-time higher than the others (Fig. 3e). Coupled with the fact that the HO· scavenger dimethyl sulfoxide rescued BL21(DE3) cells from the iron/PG killing assay (Fig. 3f), all above results demonstrated that the generation of HO· promoted by PG-iron complex was the main factor of PG toxicity to E. coli.

Knockout of clpX gene increases the protein stability of RpoS and Dps

Our results showed that knockout of clpX gene significantly enhanced E. coli tolerance to PG (Fig. 1c). ClpX is an ATP-dependent molecular chaperone in the ClpXP protease complex, in which the ClpP subunits form the proteolytic center and ClpX serves as a substrate-specifying adaptor to unfold and transfer the substrate protein to the catalytic site41. RpoS is the major regulator of general stress response in E. coli. In growing cells, RpoS has an extremely short half-life due to its proteolysis by ClpXP, and various stress conditions led to stabilization and accumulation of RpoS. In addition, a direct recognition factor RssB having specific affinity for RpoS and targeting RpoS to ClpXP is essential to the proteolysis of RpoS (Fig. 4a)42,43.

Fig. 4: Knockout of clpX gene improved the protein stability of general stress response regulator RpoS and miniferritin Dps.
figure 4

a Model illustrating the proteolysis of RpoS and Dps proteins mediated by ClpXP protease. b Survival rates of E. coli BL21(DE3) wild-type strain, clpX, rssB, rpoS, and clpX rpoS mutant strains after treatment with PG and iron for 4 h (n = 3 biological independent samples). c Western blot of RpoS protein in BL21(DE3) wild-type, clpX and rssB mutants with the absence and presence of PG. d Relative mRNA level of RpoS-dependent oxidative stress response genes in BL21(DE3) wild-type and clpX strains (n = 3 biological independent samples with two technical repeats). e Western blot of Dps protein in BL21(DE3) wild-type and clpX mutant with the absence and presence of PG. Two-tailed Student’s tests were performed to determine the statistical significance for two group comparisons. Error bar, mean ± SEM.

To test whether the increased PG tolerance of clpX mutant is related with the RpoS proteolysis, rpoS and rssB knockout mutants were constructed and subjected to PG tolerance assay. Although rpoS deletion did not affect cell growth (Supplementary Fig. 4), the rpoS mutant became much more susceptible to PG. In contrast, the PG tolerance of rssB mutant was significantly higher, similar with that of the clpX mutant (Fig. 4b). Furthermore, a double mutant clpX rpoS strain presented similar susceptibility to PG as the rpoS mutant, suggesting that the function of clpX mutation to increase E. coli PG tolerance is dependent on the RpoS protein.

RpoS protein level was determined by Western blot using a rabbit polyclonal anti-RpoS antibody. In E. coli cells at mid-log phase no matter the presence of PG, knockout of both clpX and rssB gene enhanced the RpoS protein level remarkably (Fig. 4c and Supplementary Fig. 5). Furthermore, clpX deletion notably enhanced the transcription of several RpoS-dependent genes involved in bacteria oxidative stress response including katE, uspB, yaiA, bsmA, yggE, yodB, and ychH (Fig. 4d). Especially, the katE mRNA level was 21 times higher in the clpX mutant, which certainly would protect E. coli from PG toxicity (Fig. 2a).

Besides RpoS, the ClpXP protease can also degrade the Dps protein, which is a miniferritin and sequesters free iron44. To monitor the Dps level, a his6-tag was fused to the C-terminal of Dps protein. As shown in Fig. 4e and Supplementary Fig. 6, much more Dps was detected in the wild-type strain than in the clpX mutant, that would help to reduce the intracellular concentration of free iron and the production of HO·. In summary, knockout of clpX stabilized the general stress response regulator RpoS and miniferritin Dps, and further rescued E. coli cells from oxidative stress.

Phenolic compounds induce ferroptosis-like death in diverse organisms

Above results demonstrated that excessive HO· produced in Fenton reaction killed E. coli cells, similarly to ferroptosis, an oxidative and iron-dependent regulated cell death that was described in eukaryotic organisms45,46. To confirm that PG induces ferroptosis-like death of E. coli, several assays were carried out. Lipid peroxidation is one of hallmarks of ferroptosis, and leads to accumulation of malondialdehyde46. As shown in Fig. 5a, the presence of PG in iron-containing medium increased the malondialdehyde level of E. coli cells by more than 70 times. Furthermore, PG exposure of E. coli cells resulted in a dramatic decline of the reduced glutathione (GSH) (Fig. 5b), which is a characteristic of ferroptosis in animals45. In consistent, addition of GSH restored the viability of E. coli cells in treatment with PG (Fig. 5c). Additionally, ferroptosis inhibitor ferrostatin-1 prevented E. coli cell death induced by PG (Fig. 5d). In animals, glutathione peroxidase 4 (GPX4) is a major ferroptosis defense system, detoxifying lipid hydroperoxides and inhibiting ferroptosis47,48, and E. coli also encodes a GPX4 homolog protein BtuE, sharing an identity of 35.5% and a similarity of 62.3% with human GPX4 (Supplementary Fig. 7). So, the btuE gene was cloned into plasmid vector pTrcHis2B and overexpressed in BL21(DE3) strain. The strain carrying empty vector presented a survival rate similar with that of BL21(DE3) wild-type strain, while overexpression of btuE dramatically increased E. coli tolerance to challenge of PG and iron (Fig. 5e). All experiments suggested that PG induced the ferroptosis-like cell death of E. coli.

Fig. 5: Phenolic compounds induce ferroptotic cell death in diverse species.
figure 5

a Accumulation of malondialdehyde (MDA) in E. coli BL21(DE3) cells grown in iron-containing medium after treatment with PG (n = 3 biological independent samples). Ec, E. coli. b Intracellular level of reduced glutathione (GSH) in E. coli BL21(DE3) cells grown in iron-containing medium after treatment with PG (n = 3 biological independent samples). c Survival rates of E. coli BL21(DE3) cells after challenge of PG and iron with the absence and presence of GSH (n = 3 biological independent samples). d Survival rates of E. coli BL21(DE3) cells after challenge of PG and iron with the absence and presence of ferroptosis inhibitor ferrostatin-1 (Fer-1) (n = 3 biological independent samples). e Survival rates of E. coli BL21(DE3) strain carrying empty plasmid vector and pbtuE after treatment with PG and iron (n = 3 biological independent samples). f Survival rates of E. coli BL21(DE3) strain upon exposure to different phenolic compounds with the absence and presence of iron. Concentration of each compound was indicated. (n = 3 biological independent samples). g Survival rates of Klebsiella pneumoniae (Kp), Salmonella typhimurium (St), and Saccharomyces cerevisiae (Sc) upon exposure to PG and phenol with the absence and presence of iron (n = 3 biological independent samples). h The cytotoxicity of iron and PG to HeLa cells (n = 6 biological independent samples). Two-tailed Student’s tests were performed to determine the statistical significance for two group comparisons. Error bar, mean ± SEM.

Next, some other phenolic compounds including phenol, catechol, resorcinol, pyrogallol, and naphthol, were tested on their abilities to induce E. coli ferroptosis-like death. Results in Fig. 5f showed that these phenolic chemicals could not kill E. coli BL21(DE3) strains alone, and supplementation of iron decreased the survival rates of BL21(DE3) strain one- to four-order of magnitude. Moreover, the toxicity of PG and phenol to bacteria Salmonella typhimurium and Klebsiella pneumoniae and eukaryote Saccharomyces cerevisiae was enhanced dramatically by the addition of iron as well (Fig. 5g). Although gram-positive bacteria are not tested, they are reported to be more susceptible to oxidative-ferroptotic death caused by iron complexes39. Most importantly, combination of PG and iron presented significant cytotoxicity to HeLa cells, while killing effect to HeLa cells was not detected with supplementation of PG or iron alone (Fig. 5h). All these results collectively demonstrated that phenolic compounds are capable to trigger the ferroptosis-like cell death pathway in diverse organisms, from bacteria to mammalian cells.

Repressing phenols-induced ferroptosis-like death benefits both biodegradation and biosynthesis of phenolic compounds

Phenolic compounds are industrially versatile commodity chemicals, also the most ubiquitous pollutants49,50. Recently, biosynthesis of desired phenolic compounds and biodegradation of phenolic pollutants have attracted more attentions due to their environmental friendliness and practical feasibility. However, most phenolic compounds are toxic to microorganism, and growth inhibition of phenols has become a major limiting issue for commercialization of phenols-related biochemical processes3,8,22,24.

To test whether this phenols-induced ferroptosis-like death affects phenol biodegradation process, Pseudomonas sp. DHS3Y strain was grown in M9 medium with 1 g/L phenol as sole carbon source under iron depleted- and iron rich-conditions, respectively. As shown in Fig. 6a, the strain under iron depleted-condition started to grow after 48 h, and reached an OD600 of 2.52 ± 0.04 after 72 h of inoculation, while its growth was not observed under iron rich conditions. In accordance with this, the phenol concentration decreased to 0.86 ± 0. 01 g/L after 60 h and phenol could barely be detected after 72 h of inoculation in iron depleted-medium, however there were still 0.93 ± 0. 01 g/L phenol left in the iron rich-medium after 72 h of inoculation (Fig. 6b). These results suggested that repression of phenol-induced ferroptosis-like death significantly improved the phenol biodegradation.

Fig. 6: Bacterial tolerance contributes to biodegradation and biosynthesis of phenols.
figure 6

a Growth of Pseudomonas sp. DHS3Y in M9 medium using phenol as sole carbon source under iron depleted and iron rich conditions (n = 3 biological independent samples). b Degradation of phenol by Pseudomonas sp. DHS3Y under iron depleted and iron rich conditions (n = 3 biological independent samples). c PG production of Q3595 (E. coli BL21(DE3) carrying PG biosynthetic pathway) and Q4333 (PG tolerant mutant carrying PG biosynthetic pathway) grown in MSM with the absence and presence of iron (n = 3 biological independent samples). d Growth of Q3595 and Q4333 in MSM with the absence and presence of iron (n = 3 biological independent samples). e Activity of aconitase and fumarase tested using whole cell lysate of E. coli BL21(DE3) cells grown with the absence and presence of iron (n = 3 biological independent samples). Two-tailed Student’s tests were performed to determine the statistical significance for two group comparisons. Error bar, mean ± SEM.

Furthermore, a PG-tolerant E. coli BL21(DE3) triple mutant strain (ΔclpX, ΔsodB, qmcAfetA SNP) was constructed (Supplementary Fig. 8). Then the plasmid pA-phlD/marA/acc carrying PG biosynthetic pathway12 was introduced into BL21(DE3) wild-type strain and triple mutant to generate strains Q3595 and Q4333, respectively. In shaking flask cultivation, the strain Q3595 produced 1.30 ± 0.03 g/L and 0.93 ± 0.02 g/L PG with the absence and presence of iron, whereas supplementation of iron in the strain Q4333 culture increased the PG production from 1.91 ± 0.03 g/L to 2.53 ± 0.05 g/L (Fig. 6c). Differently from this, iron enhanced the cell density of both strains in varying degrees (Fig. 6d). These results indicate that the toxicity of end-product PG is a major limiting factor for the wild-type strain Q3595, and the absence of iron impairs the ferroptosis-like death of E. coli, resulting in higher production of PG. However, for the resistant strain Q4333, PG is no longer a limiting factor and iron is essential for normal cell growth and metabolism. For example, aconitase and fumarase in TCA cycle need iron as cofactor, and iron depletion remarkably decreased their enzymatic activities (Fig. 6e). These also proved the necessity of resistant strain in bioproduction of phenolic compounds, and it is not practicable to increase bacteria tolerance to phenols simply by growing in an iron-free environment.

PG-induced ferroptosis suppresses tumor growth

Ferroptosis is considered as a new and promising option for cancer therapy as it remains functional in tumor cells that escape other forms of cell death such as apoptosis, necroptosis and pyroptosis45,46,51. To explore whether PG-induced ferroptosis can suppress tumor growth in vivo, tumor-bearing mice were generated by subcutaneous injection of human lung cancer H1299 cells and treated daily with water (control group), PG and PG-iron complex for 11 days as indicate in Fig. 7a. Administration of PG itself seemed to have some effect of suppressing tumor growth, but the statistically significant difference between control group and PG-treated group was not observed. Excitingly, complexation of PG and iron resulted in the lowest tumor growth rate and weight (Fig. 7b, c), suggesting that PG-iron complex dramatically reduced the H1299 tumor growth.

Fig. 7: PG-induced ferroptosis suppressed tumor growth.
figure 7

a Schematic experimental procedure for implantation of BALB/C nude mice and treatment with PG and PG-iron complex. b Growth of H1299 tumors in mice treated with water (control group), PG and PG-iron complex. Seven biological independent samples were tested in this experiment, and two-way ANOVA assays were performed to determine the statistical significance. c Weights of H1299 tumors in mice treated with water, PG and PG-iron complex, measured on day 11 (n = 7 biological independent samples). Two-tailed Student’s tests were performed to determine the statistical significance. Error bars, mean ± SEM. d Representative staining and immunochemical images from H1299 tumors in mice treated with water, PG and PG-iron complex. Prussian blue staining was used to detect ferric iron. Glutathione peroxidase 4 (GPX4) and 4-hydroxynonenal (4-HNE) were detected using corresponding antibodies. Scale bar represents 50 μm.

To confirm that ferroptosis was induced in tumor, tissue sections were stained with Prussian blue to detect the intracellular level of ferric iron, and results indicated the iron accumulation in tumors treated with PG-iron complex, but not in the other two groups treated with water and PG respectively (Fig. 7d). Lipid peroxidation is a hallmark of ferroptosis, producing some toxic products including 4-hydroxynonenal (4-HNE)52, while GPX4 is an effective defense system protecting biomembranes from peroxidation damage47,48. So, GPX4 and 4-HNE are both considered as biomarkers to monitor ferroptosis, and their level in tumors was determined by immunohistochemistry analysis. As shown in Fig. 7d and Supplementary Fig. 9, expression of GPX4 and accumulation of 4-HNE in tumors treated with PG-iron complex were much higher than those in control group and tumors treated with PG. These results proved that PG-iron complex-induced ferroptosis can suppress tumor growth effectively.