ENO1 suppresses cancer cell ferroptosis by degrading the mRNA of iron regulatory protein 1

α-Enolase 1 (ENO1) is a critical glycolytic enzyme whose aberrant expression drives the pathogenesis of various cancers. ENO1 has been indicated as having additional roles beyond its conventional metabolic activity, but the underlying mechanisms and biological consequences remain elusive. Here, we show that ENO1 suppresses iron regulatory protein 1 (IRP1) expression to regulate iron homeostasis and survival of hepatocellular carcinoma (HCC) cells. Mechanistically, we demonstrate that ENO1, as an RNA-binding protein, recruits CNOT6 to accelerate the messenger RNA decay of IRP1 in cancer cells, leading to inhibition of mitoferrin-1 (Mfrn1) expression and subsequent repression of mitochondrial iron-induced ferroptosis. Moreover, through in vitro and in vivo experiments and clinical sample analysis, we identified IRP1 and Mfrn1 as tumor suppressors by inducing ferroptosis in HCC cells. Taken together, this study establishes an important role for the ENO1–IRP1–Mfrn1 pathway in the pathogenesis of HCC and reveals a previously unknown connection between this pathway and ferroptosis, suggesting a potential innovative cancer therapy. Zhang and colleagues report a function of the glycolytic enzyme α-Enolase 1 in iron homeostasis. In this setting it promotes the mRNA decay of IRP1, thereby suppressing ferroptosis in hepatocellular carcinoma.

C ancer cells undergo extensive metabolic reprogramming including typical aerobic glycolysis, or the Warburg effect, which involves the switch of a multitude of metabolic enzymes to support the diversion of metabolites that facilitate glycolytic fermentation for energy metabolism even under ambient oxygen conditions [1][2][3] . Compelling studies have recently revealed that many glycolytic enzymes have moonlighting activities in regulation of gene expression 4,5 . For example, GAPDH, which converts glyceraldehyde-3-phosphate to d-glycerate-1,3-bisphosphate, has been unexpectedly found as the RNA-binding protein that regulates effector T cell function by binding to the IFN-γ messenger RNA to repress protein translation 6 . In addition to acting as a protein kinase to regulate gene expression, mitosis, cytokinesis and exosome secretion by phosphorylation of a variety of protein substrates 7 , PKM2 also acts as an RNA-binding protein to regulate mRNA translation in mouse embryonic stem cells 8 . Among the many glycolytic enzymes, ENO1 is the third to have been identified as a candidate of an RNA-binding protein from an 'interactome capture' study 9 . Nevertheless, its relevance to cancer development and the underlying mechanisms remain elusive.
Intriguingly, although ENO1 is a member of the RNA degradosome in Escherichia coli and facilitates RNA degradation 10 , whether this function is conserved in eukaryotic cells remains to be explored. Meanwhile, the functions of many genes are highly similar in eukaryotes and prokaryotes. For example, the basic machinery of DNA replication has high conservation in prokaryotes and eukaryotes 11 , despite the more complicated regulatory strategies in eukaryotic cells. Since ENO1 plays important roles in RNA degradation in prokaryotes, we endeavored to investigate whether ENO1 promotes mRNA degradation in eukaryotic cells.
Iron is an essential element for metabolic regulation, and its roles in the regulation of cell biological events are multifaceted. For example, iron also participates in potentially deleterious free radical reactions that damage lipids, proteins and DNA. Thus, iron homeostasis is maintained precisely in cells by iron transport, storage and regulatory proteins 12 . Both the beneficial and deleterious effects of iron have been reported during cancer development [13][14][15][16] . Ferroptosis is a form of regulated cell death characterized by the iron-dependent accumulation of lipid hydroperoxides 15,17 . Dysregulation of ferroptosis is associated with various human diseases, such as neurodegeneration, ischemia-reperfusion injury and cancer 15,[18][19][20][21][22] . Ferroptosis is morphologically, genetically and biochemically distinct from other forms of cell death such as apoptosis, necroptosis and autophagy 17 . ENO1 is known to be involved in the regulation of malignant phenotypes and cell proliferation, both in vitro and in vivo, through apoptosis 23,24 and autophagy 25 . However, it remains unclear whether ENO1 regulates tumor progression through ferroptosis and the regulatory mechanisms of ferroptosis in tumor biology remain largely unknown. Since iron is a key factor required for the accumulation of lipid peroxides, ultimately leading to ferroptosis 14 , it is essential to elucidate the detailed mechanisms by which cancer cells regulate iron homeostasis, metabolic reprogramming and ferroptosis to support cancer progression.
In this study we sought to identify moonlighting functions for the glycolytic enzyme ENO1. We found that ENO1, as an RNA-binding protein, conserves its RNA-degrading function from prokaryotes to α-Enolase 1 (ENO1) is a critical glycolytic enzyme whose aberrant expression drives the pathogenesis of various cancers. ENO1 has been indicated as having additional roles beyond its conventional metabolic activity, but the underlying mechanisms and biological consequences remain elusive. Here, we show that ENO1 suppresses iron regulatory protein 1 (IRP1) expression to regulate iron homeostasis and survival of hepatocellular carcinoma (HCC) cells. Mechanistically, we demonstrate that ENO1, as an RNA-binding protein, recruits CNOT6 to accelerate the messenger RNA decay of IRP1 in cancer cells, leading to inhibition of mitoferrin-1 (Mfrn1) expression and subsequent repression of mitochondrial iron-induced ferroptosis. Moreover, through in vitro and in vivo experiments and clinical sample analysis, we identified IRP1 and Mfrn1 as tumor suppressors by inducing ferroptosis in HCC cells. Taken together, this study establishes an important role for the ENO1-IRP1-Mfrn1 pathway in the pathogenesis of HCC and reveals a previously unknown connection between this pathway and ferroptosis, suggesting a potential innovative cancer therapy.
eukaryotes. We uncovered that ENO1 promotes mRNA degradation of IRP1 to maintain iron homeostasis in cancer cells, protecting these from ferroptotic cell death. We established a connection between the ENO1-IRP1-Mfrn1 pathway and ferroptosis, which is highly relevant to liver cancer progression. This knowledge will benefit the potential development of therapeutic strategies involving the induction of ferroptosis in liver cancer.

ENO1 degrades IRP1 mRNA as an RNA-binding protein.
To explore whether ENO1 has roles beyond its known glycolytic function, we asked whether ENO1 has conserved roles in RNA degradation in eukaryotes. We first performed RNA sequencing (RNA-seq) analysis in hepatoma PLC cells with ENO1 knockdown. We observed that many genes were regulated by ENO1 (Fig. 1a). Combining RNA-seq data with previous ENO1 crosslinking-immunoprecipitation-sequencing (CLIP-seq) data 9 , 17 genes were found to be negatively regulated and bound by ENO1 (Fig. 1b). Quantitative PCR with reverse transcription (RT-qPCR) confirmed that ANXA1, IRP1, KANK1, CAT, BIN1 and MYO6 among these 17 genes were strongly suppressed by ENO1 at the mRNA level (Fig. 1c). We further investigated the expression of these genes in paired clinical HCC lesions from patients with liver cancer, most of which were caused by the hepatitis B virus and their adjacent noncancerous tissues ( Fig. 1d and Extended Data Fig. 1a). The results showed that IRP1 mRNA was the most significantly decreased among HCC lesions (Fig. 1d). Therefore, we focused on IRP1 for further study.
RNA immunoprecipitation showed that ENO1 bound to IRP1 mRNA in PLC cells (Fig. 1e). ENO1 overexpression and knockdown systems consistently showed that ENO1 inhibited IRP1 expression in PLC cells (Fig. 1f, left and Extended Data Fig. 1b). Similar results were also observed in two additional HCC cell lines, HepG2 and Huh7 (Fig. 1f, right). Interestingly, IRP1 mRNA was significantly degraded in ENO1-overexpressing cells in the presence of actinomycin D, a transcription inhibitor, indicating that ENO1 negatively regulated IRP1 expression by disruption of its mRNA stability (Fig. 1g). RNA decay analysis using nascent RNA in living cells labeled with ethynyl uridine (EU) revealed that ENO1 accelerated IRP1 mRNA degradation (Fig. 1h). In addition, ENO1 promoted RNA decay of the other genes shown in Fig.1c (Extended Data Fig. 1c), demonstrating that ENO1 extensively regulates gene expression by acceleration of RNA degradation.
Recent studies have shown that the dual DNA-and RNA-binding capacity of a growing body of proteins may play an important role in modulation of gene expression 26 . ENO1 has a DNA-binding domain (DBD) 27 , and different deletion mutations of the DNA-binding region of ENO1 attenuated the inhibitory effect of ENO1 on both IRP1 mRNA and protein expression, as well as its ability to bind to IRP1 mRNA (Fig. 1i,j). Our results further showed that ENO1 S40A or ENO1 D245R , the catalytically dead mutants of ENO1 (ref. 28 ), exhibited IRP1 mRNA-binding capacity similar to wild-type (WT) ENO1 (Fig. 1j). Moreover, MBP1, an alternative translation product of ENO1 mRNA that contains the DNA-binding domain of ENO1 (ref. 29 ), also inhibited IRP1 via its DNA-binding domain (Fig. 1k). Taken together, we thus conclude that ENO1 binds to IRP1 mRNA via its DNA-binding domain. On the other hand, a dual-luciferase reporter assay demonstrated that the 5' untranslated region (5' UTR) of IRP1 mRNA is important for its association with ENO1 protein (Fig. 1l). More interestingly, consistent with a previous report that CpG dinucleotides located within the 5' UTR accelerated mRNA decay 30 , deletion of the CpG-rich region in the IRP1-5' UTR abolished the inhibitory effect of ENO1 on IRP1, indicating that this region is important for its association with ENO1 (Fig. 1m). In conclusion, ENO1, as an RNA-binding protein, binds to the CpG-rich region of IRP1-5' UTR via its DBD and ultimately promotes the degradation of IRP1 mRNA (Fig. 1n).
ENO1 recruits CNOT6 to degrade IRP1 mRNA. We then explored the mechanism of ENO1-mediated IRP1 degradation. Mass spectrometry data from a previous report showed that ENO1 interacts with the CCR4-NOT deadenylase complex, which plays a central role in mRNA regulation by catalyzing the removal of mRNA poly (A) tails 31 . Our analysis of poly (A) tail length showed that ENO1 promotes the deadenylation of IRP1 mRNA (Fig. 2a), suggesting the involvement of the CCR4-NOT deadenylase complex in ENO1-mediated IRP1 mRNA degradation. To further identify the factor(s) involved in ENO1-mediated RNA degradation, we screened the components of the CCR4-NOT complex that could interact with ENO1 using coimmunoprecipitation (CoIP) assays. As a result, CNOT3, CNOT6 and CNOT10 were found to interact with ENO1 (Fig. 2b). However, GST pulldown experiments showed that only CNOT6, rather than CNOT3 or CNOT10, interacted directly with ENO1 ( Fig. 2c and Extended Data Fig. 2a). This was further confirmed by CoIP experiments using Flag-ENO1 combined with HA-CNOT6 and vice versa ( Fig. 2d and Extended Data Fig. 2b). Importantly, CNOT6 significantly suppressed IRP1 at both the mRNA and protein level (Fig. 2e, left). Similar results were also observed in HepG2 and Huh7 cells (Fig. 2e, right). Furthermore, knockdown of CNOT6 promoted mRNA stability and protein expression of IRP1 (Extended Data Fig. 2c).
Further CoIP experiments revealed that the leucine repeat domain (LR) of CNOT6 mediated its interaction with ENO1 (Fig. 2f). On the other hand, the C-terminal region of the ENO1 . Blue and red dots indicate genes down-and upregulated, respectively, by shENO1 (fold change >1.5 and <0.67, respectively) (n = 1 sample). Immunoblot analysis confirmed knockdown of ENO1 (bottom). b, Venn diagram of rNA-seq and CLIP-seq showing that the mrNA of 17 genes was bound and downregulated by ENO1. c, mrNA levels of 17 screened genes were determined in PLC cells expressing NTC or shENO1 (n = 3 biological replicates). d, IRP1 mrNA levels were measured in 33 pairs of clinically matched tumor-adjacent noncancerous liver tissues (nontumor) and human HCC tissues (tumor) (n = 33 patients with HCC). right, n = 3 technical replicates. e, rNA immunoprecipitation analysis of the binding of endogenous IRP1 and ENO1 mrNAs by ENO1 in PLC cells (n = 3 biological replicates). f, IRP1 mrNA and protein expression in ENO1-overexpressing PLC, HepG2 and Huh7 cells (mrNA data: n = 3 biological replicates). g, IRP1 mrNA stability was determined in ENO1-overexpressing PLC cells treated with the transcription inhibitor actinomycin D (5 μM) for the indicated times (n = 3 biological replicates). h, Pulse-chase analysis of IRP1 mrNA in PLC cells with ENO1 overexpression (n = 3 biological replicates). i, IRP1 mrNA and protein expression in PLC cells expressing WT ENO1 or ENO1 deletion mutants (mrNA data: n = 3 biological replicates). j, Binding of endogenous IRP1 mrNA by ENO1 in PLC cells expressing WT ENO1 or ENO1 mutants (n = 3 biological replicates). k, IrP1 protein in PLC cells expressing ENO1, MBP1 or ENO1 △DBD mutant. l,m, Dual-luciferase analysis in HEK293 cells transfected with Flag-ENO1 and pGL3-IRP1-5' UTr or -3' UTr plasmids (l) or Flag-ENO1 and pGL3-IRP1-5' UTr plasmids with deletion of CpG-rich region (m) (n = 3 biological replicates). n, Working model showing that ENO1 binds to the CpG-rich region of IRP1-5' UTr via its DNA-binding domain to degrade IRP1 mrNA. Data presented as mean ± s.d. of three independent experiments (c,e-j,l,m) or mean ± s.e.m. (d). Statistical significance was determined by two-tailed unpaired Student's t-test (c-j,l,m). Immunoblot experiments were repeated three times independently, with similar results (a,f,i,k).rPKM, reads per kilobase million. shENO1 cells, cells expressing shENO1. EV, empty vector. P3h, pulse for 3 h; C24h, chase for 24 h. N, N-terminus; C, C-terminus. CTD, C-terminal domain.
protein was important for its interaction with CNOT6 (Fig. 2g,h and Extended Data Fig. 2d). This was further confirmed by CoIP experiments showing that the C-terminal region of ENO1 interacted with the LR of CNOT6 ( Fig. 2i and Extended Data Fig. 2e). Our data also demonstrated that the interaction between ENO1 and CNOT6 was independent of the presence of RNA (Extended Data Fig. 2f). More importantly, suppression of ENO1 abolished the inhibitory effect of CNOT6 on IRP1 expression, demonstrating that CNOT6 relies on ENO1 to inhibit IRP1 expression (Fig. 2j). In conclusion, ENO1 binds to the 5' UTR of IRP1 mRNA via its DNA-binding domain and recruits CNOT6 via its C-terminal domain, whereby the nuclease (NU) domain of CNOT6 promotes the deadenylation of the IRP1-3' UTR ( Fig. 2k).

ENO1 promotes liver cancer by inhibition of the IRP1-Mfrn1
axis. Consistent with a previous report that IRP1 barely affects the proliferation of cultured cancer cells 32 , we observed that IRP1 marginally inhibited the proliferation of PLC, HepG2 and Huh7 cells (Extended Data Fig. 3a). However, given that IRP1 is an important factor involved in iron homeostasis, we manipulated the iron concentration in cultured medium and found that, in the presence of iron, IRP1 dramatically inhibited cell proliferation (Extended Data Fig. 3a); meanwhile, knockdown of IRP1 promoted cell proliferation (Extended Data Fig. 3b). Notably, the same concentration of iron used alone in the above experiments (200 μM) showed no obvious effect on cell numbers (Extended Data Fig. 3c). Overexpression of IRP1 strongly impaired growth of tumor xenografts in mice ( Fig. 3a and Extended Data Fig. 3d,e). More importantly, we found that serum iron concentration in patients with clinical liver cancer was significantly higher than that of healthy subjects (Fig. 3b), which is consistent with a report showing that elevated iron levels in the body are associated with an increased risk of cancer 16 and that excess iron in the liver was more frequently observed in HCC patients than in healthy people 33,34 . Altogether, our results suggest that IRP1 sensitizes liver tumor cells to iron and serves as a tumor suppressor. IRP1 plays an important role in mitochondrial iron homeostasis and is essential for liver physiology and function 35,36 . However, because the relationship between IRP1 and mitochondria is poorly understood in tumors, we performed real-time PCR to analyze the expression of mitochondrial iron-sulfur cluster synthesis-related genes (Fig. 3c). Interestingly, Mfrn1, a mitochondrial channel protein that carries iron from the cytoplasm into the mitochondria 37 , was significantly downregulated by shIRP1 (Extended Data Fig. 3f). Similar results were observed in HepG2 and Huh7 cells (Extended Data Fig. 3g). Our data further showed that IRP1 enhanced Mfrn1 expression at both the mRNA and protein level (Fig. 3d). Similar results were observed in HepG2 and Huh7 cells (Fig. 3e). IRP1, together with a 4Fe-4S cluster, functions as a cytosolic aconitase by catalyzing the conversion of citrate to isocitrate. Our data showed that IRP1 C437S , the enzymatically inactive mutation of IRP1 (ref. 32 ), exhibited an effect similar to WT IRP1 on Mfrn1 expression and cell proliferation (Extended Data Fig. 3h). Consistently, overexpression of IRP1 C437S also strongly impaired growth of tumor xenografts in mice (Extended Data Fig. 3i,j), suggesting that regulation of Mfrn1 by IRP1 is independent of its enzymatic activity. Rather, our data suggested that IRP1 activated Mfrn1 transcription via CREB (Extended Data Fig. 4a,b). Intriguingly, Mfrn1 inhibited cell growth, especially in the presence of iron in PLC, HepG2 and Huh7 cells (Extended Data Fig. 4c). Importantly, ENO1 suppressed Mfrn1 expression at both the mRNA and protein level ( Fig. 3f and Extended Data Fig. 4d), which was abolished by IRP1 ( Fig. 3g and Extended Data Fig. 4e). Consistent with Fig. 1i, our results further showed that ENO1 S40A and ENO1 D245R , the catalytically dead mutants of ENO1, exhibited an effect similar to WT ENO1 on the inhibition of IRP1 or Mfrn1 (Extended Data Fig. 4f). Collectively, these data demonstrate that ENO1, as an RNA-binding protein, suppresses Mfrn1 expression in an IRP1-dependent manner.
Cell proliferation and mouse xenograft experiments further revealed that overexpression of IRP1 or Mfrn1 diminished ENO1-mediated promotion of cell proliferation and tumor growth ( Fig. 4a and Extended Data Fig. 5a-c). Moreover, overexpression of Mfrn1 abolished shIRP1-enhanced cell proliferation and tumor growth ( Fig. 4b and Extended Data Fig. 5d,e). Collectively, these data show that the IRP1-Mfrn1 axis is involved in ENO1-regulated cell proliferation and tumor growth both in vitro and in vivo.
Neither IRP1 nor Mfrn1 has previously been implicated in the pathogenesis of liver cancer. By employing a spontaneous mouse model of HCC 38 , we injected YAP-5SA plasmids alone or in combination with IRP1 plasmids into mouse liver and observed that IRP1 retarded tumor growth in YAP-5SA-induced liver cancer (Fig. 4c,d and Extended Data Fig. 6a). Using an NRAS/shp53 plasmid-induced liver tumor model, we also found that overexpression of IRP1 blunted tumor growth (Extended Data Fig. 6b-d). Since Mfrn1 -/mice were embryonic lethal 39 , we generated Mfrn1 +/− mice to explore the effect of Mfrn1 on HCC tumorigenesis (Extended Data Fig. 6e). In the YAP-5SA-induced HCC model we observed a significantly increased incidence of liver cancer and enhanced tumor growth in Mfrn1 +/− mice compared to WT mice (Extended Data Fig. 6f-i). To further evaluate the relationship between Mfrn1 and liver cancer, we generated Mfrn1 liver-specific knockout (LKO) mice. We observed a significantly increased incidence of liver cancer and enhanced tumor growth in Mfrn1 LKO compared to WT mice (Fig. 4e,f and Extended Data Fig. 6j). Immunoblot analysis confirmed the downregulation of Mfrn1 in tumor tissues from Mfrn1 LKO mice (Extended Data Fig. 6k). In the model using a diethylnitrosamine/carbon tetrachloride (DEN/CCl 4 )-induced HCC model 40 , we also observed that in Mfrn1 LKO mice it was much easier to facilitate HCC progression in vivo (Fig. 4g,h and Extended Data Fig. 6l). Collectively, our data strongly demonstrate that IRP1 and Mfrn1 serve as tumor suppressors in HCC.
The ENO1-IRP1-Mfrn1 axis regulates ferroptosis. Mfrn1 is known to promote the transfer of iron from the cytoplasm to tail length in PLC cells expressing Flag-tagged ENO1 using semi-qPCr. GAPDH served as negative control. b, CoIP assay in HEK293T cells transfected with HA-ENO1 and Flag-tagged CNOT3, CNOT6, CNOT7, CNOT9 or CNOT10 plasmids. c, Pulldown of His-CNOT6 by GST-ENO1 using proteins purified in E. coli. d, CoIP assay in HEK293T cells transfected with HA-CNOT6 and Flag-ENO1 plasmids using anti-Flag antibody. e, qPCr (n = 3 biological replicates) and immunoblot analysis of IrP1 expression in CNOT6-overexpressing PLC cells (left) and HepG2 and Huh7 cells (right). f-i, CoIP assay in HEK293T cells cotransfected with HA-ENO1 and Flag-tagged CNOT6, CNOT6-Lr or CNOT6-NU plasmids (f), HA-CNOT6 and Flag-tagged WT ENO1, ENO1-N terminus or ENO1-DBD + C terminus plasmids (g), HA-CNOT6 and Flag-tagged plasmids expressing the indicated domains of ENO1 (h) and HA-CNOT6-Lr domain plasmids together with Flag-ENO1-C terminus plasmids (i). j, IrP1 protein expression in CNOT6-overexpressing PLC cells with further knockdown of ENO1. k, Working model showing that ENO1 degrades IRP1 mrNA by recruitment of CNOT6. In brief, ENO1 binds to IRP1 mrNA via its DBD and interacts with the Lr domain of CNOT6 via its C terminus, whereby the NU domain of CNOT6 promotes IRP1 mrNA degradation by deadenylation of the IRP1-3' UTr. Data presented as mean ± s.d. of three independent experiments; statistical significance was determined by two-tailed unpaired Student's t-test (e). Experiments using immunoblot (b-j) or gels (a) were repeated three times independently, with similar results. mitochondria 37 . We observed that ENO1 decreased, while IRP1 and Mfrn1 increased, mitochondrial iron accumulation (Fig. 5a). Moreover, IRP1-mediated accumulation of mitochondrial iron was diminished by shMfrn1 and ENO1-mediated reduction of mitochondrial iron was reversed by overexpression of either IRP1 or Mfrn1 (Fig. 5b,c). Many iron-sulfur clusters are involved in the transmission of electrons in mitochondrial electron transport chains, and our data also showed that the ENO1-IRP1-Mfrn1 axis regulated mitochondrial complex I and II activities and mitochondrial respiration (Extended Data Fig. 7a-g). At the same time, because mitochondrial iron-sulfur cluster biogenesis promotes mitochondrial complex I and II activities and the stability of proteins FECH and POLD1 (refs. 35,41 ), we used xenograft tumor lysates generated from Huh7 cells overexpressing IRP1 (from Fig. 3a) to test mitochondrial iron-sulfur cluster biogenesis. The results showed that IRP1 promoted mitochondrial iron-sulfur cluster biogenesis in vivo by enhancement of mitochondrial complex activities and increased expression of FECH and POLD1 (Extended Data Fig. 7h,i). Consistent with a previous report that accelerated mitochondrial respiration led to excess accumulation of mitochondrial reactive oxygen species (ROS) in cancer cells 42 , we observed that both IRP1 and Mfrn1 induced mitochondrial ROS generation and led to cell death in the presence of iron ( Fig. 5d and Extended Data Fig. 8a), and that Mfrn1 knockdown abolished IRP1-mediated mitochondrial ROS accumulation (Extended Data Fig. 8b).

IRP1 and Mfrn1 deficiency predicts poor clinical prognosis.
Analysis of cell proliferation in vitro and liver cancer growth in vivo strongly indicated that IRP1 and Mfrn1 have tumor-suppressive effects. To further investigate the pathological significance of our findings, we examined ENO1, IRP1 and Mfrn1 mRNA expression in 33 paired clinical human HCC lesions and adjacent noncancerous tissue samples. The results showed that IRP1 and Mfrn1 mRNA levels were significantly decreased in HCC lesions compared to adjacent noncancerous tissue ( Fig. 1d and Extended Data Fig. 10a), but those of ENO1 showed the opposite trend (Extended Data Fig. 10b). Consistently, immunoblot analysis revealed decreased IRP1 and Mfrn1 protein levels but increased ENO1 protein levels in human HCC tissues compared to the corresponding adjacent noncancerous tissues (Fig. 7a). More interestingly, similar results were observed in YAP-5SA-induced mouse HCC samples (Fig. 7b) and NRAS/shp53-induced mouse HCC samples (Fig. 7c).
Next, immunohistochemistry (IHC) was employed for analysis of ENO1, IRP1 and Mfrn1 expression in a retrospective cohort of 135 clinicopathologically characterized HCC cases, including 14 of stage I (10.4%), 75 of stage II (55.6%), 30 of stage III (22.2%) and 16 of stage IV (11.8%), based on tumor, node, metastases (TNM) staging (Supplementary Table 1). IHC results demonstrated that IRP1 and Mfrn1 were abundantly expressed in normal liver tissues but weakly expressed in HCC tissues, while ENO1 showed the opposite trend (Fig. 7d). Quantitative analysis of IHC results revealed that IRP1 and Mfrn1 expression in clinical stage I-IV primary tumors was significantly decreased (Fig. 7e,f) but that ENO1 expression was significantly increased compared to that in normal liver tissue (Extended Data Fig. 10c). Furthermore, IRP1 and Mfrn1 were drastically downregulated in late-stage HCC (stages III and IV) compared to early-stage HCC (stages I and II) while ENO1 showed the opposite trend (Supplementary Tables 2-4). Moreover, Spearman analysis revealed correlations between the ENO1-IRP1-Mfrn1 axis and patient clinicopathological characteristics, including survival time, vital status, clinical stage and tumor size (Supplementary Tables  5-7), further suggesting strong association of ENO1, IRP1 and Mfrn1 expression with HCC clinical staging and patient survival. Finally, the Kaplan-Meier test indicated that patients expressing high IRP1 or Mfrn1 in their HCC lesions survived much longer than those with low expression (Fig. 7g,h), suggesting that IRP1 and Mfrn1 represent promising prognostic biomarkers in HCC, with ENO1 indicating the opposite outcome (Extended Data Fig. 10d). Further IHC analysis using serial sections of the same HCC tissues confirmed the correlation of high ENO1 and low IRP1 and Mfrn1 (Fig. 7i). Moreover, our IHC data showed markedly decreased expression of IRP1 and Mfrn1 in carcinoma tissues compared with paracarcinoma (Extended Data Fig. 10e,f). These results demonstrate therefore that the ENO1-IRP1-Mfrn1 axis is correlated with human HCC and predicts clinical prognosis.

Discussion
Enhanced glycolysis is a hallmark of cancer cells and is facilitated by oncogenes including MYC, HIF1, Ras and AKT, eventually leading to aberrantly induced expression of glycolytic enzymes in tumors 2,49 . Intriguingly, numerous metabolic enzymes have emerged as RNA-binding proteins with roles beyond the catalysis of glycolysis 4 . Although ENO1 is known to accelerate cancer progression as a glycolytic enzyme 50 , recent findings indicate that it may serve as an RNA-binding protein with moonlighting functions 9,51,52 . ENO1 is a member of the RNA degradosome in prokaryotes 10 , and this prompted us to speculate that it might mediate mRNA decay to promote cancer development. We demonstrate here that ENO1 accelerates mRNA degradation of many genes, including IRP1 in liver cancer cells (Fig. 1g,h and Extended Data Fig. 1c). These results indicate that the RNA-related activity of ENO1 could be a general phenomenon rather than liver cancer specific, although further independent study is warranted to consolidate its roles in other types of cancer. We further uncover that CNOT6 is a key factor driving ENO1-mediated IRP1 mRNA degradation (Fig. 2b,c,j). ENO1 serves as a sensor linking the energetic state to mRNA degradation in prokaryotes 10 . However, in eukaryotes we found here that ENO1 acts as an RNA-binding protein to degrade RNAs by recruitment of the RNA degradation factor CNOT6, demonstrating conservation and progression of the functional complexity of ENO1 during evolution. Thus, our findings identify a function of ENO1 that mediates mRNA degradation in mammalian cells, and these have enriched our knowledge on the role of ENO1 as a moonlighting enzyme. Iron metabolism is a fascinating research area because of the metal's dualities in multiple biological processes 53 . Iron biologists currently focus on understanding iron homeostasis and dyshomeostasis in malignant tumors, through which is enabled the identification of therapeutic approaches for cancer 13,14,16 . Although earlier studies suggested that iron metabolism promotes tumor initiation and progression 16 , recent studies demonstrate that excessive iron paradoxically has a deleterious effect on tumor tissue, leading to ferroptosis in particular 13,15,[54][55][56] . Ferroptosis is a unique type of cell death resulting from metabolic dysfunction involving the metabolisms of iron, lipids, oxidants and energy 14 . The mechanisms underlying ferroptosis during cancer development remain largely elusive. In our study, we found that IRP1 was suppressed by ENO1 via an RNA degradation mechanism, which led to decreased Mfrn1 expression. Significantly, we provide a mechanism by which IRP1 promotes ferroptosis via modulation of Mfrn1-induced mitochondrial iron enrichment (Figs. 5d-j and 6a-f). Although a previous study has shown, by RNA interference screening, that IRP1 may play a role in ferroptosis in mouse embryonic fibroblasts 57 , we demonstrated that suppression of IRP1 protected cancer cells from ferroptotic cell death. We also demonstrated the tumor-suppressive function of Mfrn1 through iron homeostasis, which is crucial in eliciting ferroptosis. Other mechanisms involved in Mfrn1-mediated tumor growth inhibition merit further exploration in future studies. We observed that the addition of iron dramatically enhanced the sensitivity of liver cancer cells to oxidative stress, which is consistent with previous studies employing iron as the ferroptosis inducer 58,59 . More interestingly, we found that serum iron concentrations of liver cancer patients were significantly higher than those of healthy individuals (Fig. 3b), indicating that iron metabolism is closely related to liver cancer progression.
Following analysis of clinical HCC samples and mouse tumor tissues, we found that IRP1 and Mfrn1 were significantly decreased but that ENO1 was significantly increased in HCC lesions compared to adjacent noncancerous tissue (Fig. 7a-c). Furthermore, IHC experiments revealed that expression of IRP1 and Mfrn1 in clinical primary tumors was significantly decreased, but that of ENO1 was significantly increased compared to that in adjacent noncancerous liver tissues (Fig. 7d-f and Extended Data Fig. 10c). These results suggest that the ENO1-IRP1-Mfrn1 axis is highly relevant to human HCC development and may represent a reliable HCC prognostic indicator. Hence, we have uncovered in this study an unexpected role for ENO1 in promotion of IRP1 mRNA degradation via CNOT6 in cancer cells, similar to a mechanism in prokaryotic cells. Importantly, we identified an ENO1-IRP1-Mfrn1 signaling axis critical in the regulation of ferroptosis and survival of cancer cells (Fig. 7j). Furthermore we demonstrated, by different Fig. 7 | IRP1 and Mfrn1 deficiency predicts poor clinical prognosis. a, ENO1, IrP1 and Mfrn1 protein levels were measured by immunoblot using paired adjacent noncancerous liver tissues (N) and human HCC tissues (T). Calnexin served as a loading control (n = 12 patients with HCC). b, ENO1, IrP1 and Mfrn1 protein levels were measured by immunoblot using paired adjacent noncancerous liver tissues (N) and cancerous liver tissues (T) in yAP-5SA-induced mouse HCC. Calnexin served as a loading control (n = 10 mouse liver tissues with HCC). c, ENO1, IrP1 and Mfrn1 protein levels were measured by immunoblot using paired adjacent noncancerous liver tissues (N) and cancerous liver tissues (T) in NrAS/shp53-induced mouse HCC. Calnexin served as a loading control (n = 8 mouse liver tissues with HCC). d, representative IHC images of ENO1, IrP1 and Mfrn1 staining in normal liver tissue (normal) and HCC specimens of different clinical stages (I-IV); scale bars, 50 μm. Insets: fourfold magnification; scale bars, 12.5 μm. e,f, Statistical quantification of MOD values for IrP1 (e) and Mfrn1 (f) staining in IHC assays between normal liver tissues and HCC specimens at clinical stages I-IV (healthy donors, n = 9; patients with HCC, stage I (n = 14), II (n = 75), III (n = 30) and IV (n = 16)). g,h, Kaplan-Meier curves with univariate analyses of patients with low versus high expression of IrP1 (g) (high IrP1, n = 62 patients; low IrP1, n = 73 patients) or Mfrn1 (h) (high Mfrn1, n = 66 patients; low Mfrn1, n = 69 patients). i, representative IHC images of ENO1, IrP1 and Mfrn1 staining in serial sections. j, Schematic model illustrating the role of the ENO1-IrP1-Mfrn1 axis in regulation of ferroptosis. The model shows that ENO1, as an rNA-binding protein, recruits CNOT6 to accelerate the mrNA degradation of IrP1 in liver cancer cells. Consequently, ENO1, via inhibition of the IrP1-Mfrn1 axis, suppresses mitochondrial iron-induced ferroptosis. Data presented as mean ± s.e.m. (e,f). Statistical significance was determined by two-tailed unpaired Student's t-test (e,f) or log-rank test (g,h).  I  II  III  IV   I  II III IV  I  II III IV in vivo models, that IRP1 and Mfrn1 function as tumor suppressors to predict the clinical prognosis of HCC, although more realistic spontaneous models of HCC will further advance this discovery. Our results shed light on the importance of the mechanisms counteracting ferroptosis in HCC progression, and may serve to encourage further mechanistic studies to identify innovative therapeutic approaches for patients with liver cancer. Plasmids and established stable cells. All shRNAs in the PLKO vector against ENO1, IRP1, Mfrn1, CNOT6 and CREB1 were obtained commercially (Sigma-Aldrich). shRNA targeting sequences are listed in Supplementary Table  8. ENO1 and its mutants IRP1, Mfrn1 and CNOT6 were subcloned into the pSin-3 × Flag or pSin-HA empty vector; they were then cotransfected with plasmids encoding VSVG and △8.9 into HEK293T packaging cells using PEI (Polysciences). PLC, HepG2 or Huh7 cells were infected with lentivirus containing polybrene and selected with 0.5 µg ml -1 puromycin to establish stable cells.

Methods
Immunoblotting. RIPA buffer (50 mM Tris-HCl pH 8.0, 150 mM NaCl, 5 mM EDTA, 0.1% SDS and 1% NP-40), supplemented with protease inhibitor cocktail, was used to lyse cells on ice for 30-45 min and protein concentration was determined using a Bradford protein assay kit. Extracted proteins were boiled at 100 °C for 5 min and then subjected to electrophoresis through 6-12% SDSpolyacrylamide gel electrophoresis. For detailed antibody information, please refer to the Reporting Summary.
qPCR. According to the manufacturer's instructions, total RNA was extracted from cells or tissues using TRIzol (Life Technologies) and complementary DNA was synthesized from 1-3 µg of RNA using an iScript cDNA Synthesis Kit (Bio-Rad). qPCR was performed using SYBR Green Master Mix (Vazyme). Primer sequences are shown in Supplementary Table 9. All samples were normalized to housekeeping genes (RPL13A, actin or 18s).
Chromatin IP assay. According to the manufacturer's instructions, chromatin IP assays were performed using an EZ-ChIP kit (Millipore). Briefly, cells were sonicated by an Ultrasonic Homogenizer JY92-IIN (Scientz) then DNA was immunoprecipitated with IgG or CREB antibody (Proteintech, no. 12208-1-AP) followed by qPCR analysis (Bio-Rad). The oligonucleotide sequences used are shown in Supplementary Table 10.
RNA IP. RNA IP assays were performed as described in Zhang et al. 60 . Briefly, anti-Flag and anti-ENO1 antibodies were used for precipitation of Flag-tagged ENO1 and endogenous ENO1, respectively. Detailed primer sequences of ENO1, IRP1, Actin and 18s are listed in Supplementary Table 9.
Dual-luciferase reporter assay. IRP1-5' UTR-, IRP1-3'UTR-or IRP1-5' UTR-deletion fragments were inserted into the pGL3-basic dual-luciferase reporter vector (Promega), designated as pGL3-IRP1-5' UTR, pGL3-IRP1-3' UTR or pGL3-IRP1-5' UTR deletion, respectively. HEK293 cells were seeded in 48-well plates. After overnight incubation, HEK293 cells were cotransfected with 4 ng of pSV-Renilla plasmid and 100 ng of firefly luciferase reporter plasmid containing the indicated IRP1 mRNA fragment. The activities of Firefly and Renilla luciferase were detected with the Dual-Luciferase Reporter Assay System (Promega). Renilla luciferase activity was used to normalize firefly luciferase activity. mRNA stability assay. PLC cells incubated with complete DMEM were treated with 5 μM actinomycin D for 0, 3 or 6 h. No decrease in cell viability was observed during the course of the experiment. Total RNA was collected with TRIzol, and mRNA levels were analyzed by qPCR. All samples were normalized with actin.
RNA decay and turnover experiment. The Click-iT Nascent RNA Capture Kit (no. C10365, Thermo) was used in this experiment. A slightly modified protocol was followed, as described previously 61 .
IP. HEK293T cells were lysed with IP buffer (0.5% NP-40, 20 mM HEPES pH 7.5, 150 mM NaCl, 2 mM EDTA, 1.5 mM MgCl 2 ) supplemented with protease inhibitor cocktail for 2 h on ice, and centrifuged at 16,000g for 10 min at 4 °C. The supernatant was incubated with the designated primary antibody at 4 °C overnight, then protein A/G-conjugated beads were added with incubation for a further 1 h. Immunoprecipitates were washed three times with 0.5% NP-40 IP buffer and then boiled with SDS buffer followed by immunoblot analysis.
Fusion protein pulldown experiment. The coding region of CNOT6 was cloned into the pSV282 vector, and cDNAs encoding CNOT3 and CNOT10 were cloned into the pET-22b vector. The coding region of ENO1 was cloned into the pGEX-4T-1 vector. The proteins CNOT6, CNOT3, CNOT10 and ENO1 were produced in E. coli (DE3). Pulldown experiments were performed using E. coli purified GST fusion proteins and His-tagged proteins using pulldown buffer (150 mM NaCl, 50 mM Tris pH 7.5, 0.1% NP-40, 5 mM dithiothreitol). After incubation for 20 min, the beads were pelleted and washed three times with pulldown buffer followed by protein elution and immunoblot analysis. Cell death assay. Cells were seeded in a 6-well plate 1 day before treatment. Cell death was assessed by labeling with PI (no. P4864, Sigma-Aldrich) and analyzed by flow cytometry (BD Biosciences).
Measurement of LDH activity. The culture medium of PLC cells was collected and LDH activity measured using the LDH-Glo Cytotoxicity Assay (no. J2380, Promega) following the manufacturer's instructions.
PAT assay. Analysis of poly (A) tail length by PCR (PAT assay) was performed as described previously 62 . The specific primer sequences are listed in Supplementary Mitochondrial iron assay. First, complete mitochondria were extracted using the methods mentioned above. Non-heme iron was measured as described 64 . In brief, equal amounts of mitochondrial proteins were mixed with protein precipitation solution (1:1 10% trichloroacetic acid plus 1 mol l -1 HCl) followed by heating for 1 h at 95 °C to release iron. After centrifugation at 4 °C and 16,000g for 10 min, the supernatant was mixed with an equal volume of chromogen solution (1.5 M sodium acetate, 0.5 mM ferrozine, 0.1% (v/v) thioglycolic acid). Absorbance at 562 nm was measured on a Thermo Mutiscan GO microplate reader.
Mitochondrial complex I and II activity. The activity of mitochondrial complexes I and II was measured as previously described 65 . In brief, complex I activity was measured by detecting the oxidation of NADH at 340 nm. NADH (0.13 mM), ubiquinone (65 µM) and antimycin A (2 µg ml -1 ) were added to the assay buffer (25 mM potassium phosphate pH 7.2 at 20 °C, 5 mM MgCl 2 , 2.5 mg ml -1 bovine serum albumin (fraction V)). Mitochondria (50 µg of protein) were then added followed by measurement of NADH/ubiquinone oxidoreductase activity for 3-5 min.
The specific activity of complex II was measured by detecting the reduction of 2, 6-dichlorophenolindophenol at 600 nm. Mitochondria (20 µg of protein) were preincubated in assay medium supplemented with succinate (20 mM) at 30 °C for 10 min, followed by the addition of 2 µg ml -1 rotenone, 2 µg ml -1 antimycin A and 50 µM 2, 6-dichlorophenolindophenol. After recording of baseline rate for 3 min, the reaction was started with 65 µM ubiquinone. The reduction of dichlorophenolindophenol catalyzed by enzyme was measured for 3-5 min.
Transmission electron microscopy. PLC cells were fixed with 2.5% glutaraldehyde for 12 h at 4 °C, then in 2% osmium tetroxide. After washing, samples were stained with 1% uranyl acetate aqueous solution, dehydrated in 50, 70, 90, 95 and 100% ethanol, respectively, and immersed in Eponate 12 resin. Samples were then cut into ultrathin sections and counterstained with uranyl acetate and lead citrate. Finally, images were obtained using a transmission electron microscope (120 kV; Tecnai G2 Spirit, FEI).

RNA-seq analysis.
Total RNA was extracted from cell lines using TRIzol Reagent (Life Technologies). RNA integrity was assessed by RNA integrity number and determined using an Agilent 2100 Bioanalyzer. A total amount of 3 µg of RNA per sample was used for analysis. Sequencing sampling was performed from one single replicate. Libraries were generated using a NEBNext Ultra RNA Library Prep Kit for Illumina (NEB). RNA-seq was performed on an Illumina NovaSeq 6000 platform by Novogene (Tianjin). Reads were aligned to the human genome hg19. TopHat2 v.2.1.0 and cufflinks v.2.2.1 were used to analyze RNA-seq data. Gene differential expression analysis was carried out with the DEGSeq R package (1.26.0). Gene set enrichment analysis was performed by DAVID Bioinformatics Resources.
Serum iron detection. Serum samples from 143 patients with liver cancer and 55 healthy individuals were measured using an automatic biochemical analyzer (Siemens, ADVIA 2400). Of these 143 patients, 93 were men, 50 were women and 103 were >50 years of age. Of the healthy donors, 24 were men, 31 were women and 21 were >50 years of age. Data on all 143 patients and 55 donors are provided in Supplementary Table 12. To use these clinical materials for research purposes, both the patient's previous written informed consent and the research approval of the Institutional Research Ethics Committee of the First Affiliated Hospital of the University of Science and Technology of China were obtained. An iron detection kit (Siemens, no. 501065) was used, and 200 µl of serum was assayed. All patients volunteered and were not given compensation.
Histological analysis. Liver tissue was separated, washed with ice-cold PBS and fixed with 10% neutral buffered formalin for 48 h, followed by dehydration and paraffin embedding. Sections (5 mm) were used for staining with hematoxylin and eosin.
KO mice. Mfrn1 +/− mice (C57BL/6J) were generated using CRISPR genome editing (Mfrn1 target sequence: GAACGTGATGATGATGGGTG) and were obtained from the animal facility of the University of Science and Technology of China. Mfrn1 LKO mice (C57BL/6J) were generated by crossing Mfrn1 floxed mice (Cyagen Biosciences, Inc.) with Albumin-Cre mice. All animals were housed at a suitable temperature (22-24 °C) and humidity (40-70%) under a 12/12-h light/dark cycle with unrestricted access to food and water. All animal studies were approved by the Animal Research Ethics Committee of the University of Science and Technology of China. The study is compliant with all ethical regulations of animal research.
Animal studies. All animals were housed at a suitable temperature (22-24 °C) and humidity (40-70%) under a 12/12-h light/dark cycle with unrestricted access to food and water for the duration of the experiment. All animal studies were approved by the Animal Research Ethics Committee of the University of Science and Technology of China. For xenograft experiments, 9 × 10 6 PLC cells stably overexpressing ENO1 with or without IRP1 or Mfrn1 overexpression, or expressing shIRP1 with or without Mfrn1 overexpression, were subcutaneously injected into 5-week-old male nude mice (BALB/c nude mice; Charles River Laboratory Animal Co.). Next, 5 × 10 6 HepG2 cells stably overexpressing IRP1 or Mfrn1 were subcutaneously injected into 5-week-old male nude mice, while Lip-1 was injected intraperitoneally at 10 mg kg -1 every other day, and 5 × 10 6 Huh7 cells stably overexpressing IRP1 were injected subcutaneously into 5-week-old male nude mice. We used a digital caliper to measure tumor volume every 3 or 4 days, and the following formula to calculate tumor volume: length (mm) × width (mm) × depth (mm) × 0.52. The xenograft tumor burden was less than the maximum tumor size (1 cm 3 ) approved by the Animal Research Ethics Committee of the University of Science and Technology of China.
For the DEN/CCl 4 -induced HCC model 40 , 2-week-old male mice (C57BL/6J; Cyagen Biosciences, Inc.) were intraperitoneally injected with DEN (25 mg kg -1 ) and 6 weeks later injected with CCl 4 (1:4 v/v in olive oil) at a dose of 2 ml kg -1 body weight twice per week for 14 weeks. The mice were sacrificed 72 h after the last injection of CCl 4 , and liver was used for biochemical, histological and molecular analysis.
Hydrodynamic injection. Four-week-old male ICR mice (Shanghai SLAC Laboratory Animal Co.) and 4-week-old male C57BL/6J mice (WT, Mfrn1 +/− or Mfrn1 LKO mice) were used. In the YAP-5SA-induced HCC model, 50 µg of plasmids expressing human YAP-5SA alone, or human YAP-5SA plus mouse-IRP1 together with 10 µg of plasmids expressing PB transposase, was diluted in sterile Ringer's solution with a volume equal to 10% of body weight. In the NRAS/shp53-induced HCC model, 25 µg of plasmids expressing mouse NRAS and 29 µg of plasmids expressing mouse shp53 or mouse NRAS/shp53 plus mouse-IRP1, together with 10 µg of plasmids expressing PB transposase and 37 µg of plasmids expressing SB transposase, were diluted in sterile Ringer's solution with a volume equal to 10% of body weight. The mixture was injected via the tail vein within 5-7 s.
Clinical human HCC specimens. Snap-frozen HCC tissues and corresponding noncancerous tissues that were at least 2 cm distant from the edge of tumors were collected from 33 patients with HCC in the First Affiliated Hospital of University of Science and Technology of China. Total RNA and protein were extracted from paired HCC and noncancerous tissues and then detected by qPCR and immunoblot, respectively. Formalin-fixed, paraffin-embedded primary HCC specimens from 135 patients were randomly selected from the archives of the First Affiliated Hospital of University of Science and Technology of China. Clinical data and pathological characteristics of patients, including age, gender, tumor size, tumor lymph nodes, serum alpha-fetoprotein, hepatitis B surface antigen, vascular invasion and liver cirrhosis, were recorded. Detailed patient data are shown in Supplementary Tables 1-7. The clinical staging of tumors was defined by the fifth edition of the American Joint Committee on Cancer/International Anti-Cancer Alliance TNM Classification System 66 . To use these clinical materials for research purposes, both the patients' previously obtained written informed consent and the research approval of the Institutional Research Ethics Committee of the First Affiliated Hospital of University of Science and Technology of China were obtained. All patients volunteered and recieved no compensation.

IHC.
Immunohistochemistry was performed as previously described 67 . In brief, samples were dewaxed with xylene and then rehydrated with graded ethanol. After antigen retrieval, sections were incubated with 0.3% hydrogen peroxide for 10 min to block endogenous peroxidase activity. Next, sections were preincubated in normal goat serum for 15 min to prevent nonspecific staining, then samples were incubated with anti-ENO1, -IRP1 or -Mfrn1 antibodies at room temperature. Four hours later, secondary antibody was used followed by incubation with DAB Chromogen dilution solution. The AxioVision Rel.4.6 computerized image analysis system was used for quantitative analysis of IHC staining, with the aid of an automatic measurement program (Carl Zeiss). HistoQuest was used for quantification (TissueGnostics). Six different staining fields of each section were analyzed to verify mean optical density (MOD), and t-test was used to compare average MOD differences between groups.
Statistics and reproducibility. The data are presented as either mean ± s.d. of three independent experiments or mean ± s.e.m. as stated. Statistical analyses were performed using either Prism 6 (GraphPad Software) or SPSS software Statistics 22 (IBM Corp.). Data distribution was assumed to be normal, but this was not formally tested. Two-tailed unpaired Student's t-test and one-, two-or three-way analysis of variance (ANOVA) were used to calculate P values, unless otherwise indicated in figure legends. The Tukey method was used to adjust multiple comparisons. Kaplan-Meier curves were used to depict survival function from lifetime data for human patients using the log-rank test. The relationship between expression of ENO1, IRP1 or Mfrn1 and clinicopathological characteristics was analyzed by chi-square test. P < 0.05 was considered significant. No statistical methods were used to predetermine sample size, but ours are similar to those reported in previous publications 68 . No data were excluded from the analyses. The experiments were not randomized, except that mice were randomly grouped before different treatments. Data collection and analysis were not performed blind to the conditions of the experiments, except for IHC score analysis.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
RNA-seq data have been deposited in the Gene Expression Omnibus under accession code GSE153989. Source data are provided with this paper. All other data supporting the results of this study can be obtained from the corresponding author upon reasonable request.