Epigenetic regulation of ferroptosis via ETS1/miR-23a-3p/ACSL4 axis mediates sorafenib resistance in human hepatocellular carcinoma

A tissue microarray chip containing 90 pairs of human HCC samples matched to their adjacent normal liver tissues and the associated clinicopathological information was purchased from Shanghai OUTDO Biotech Co., Ltd., Shanghai, China (LivH180Su08).

The protocol of double ISH was based on the previous publication [22] with some modifications. Briefly, the general rehydration steps were followed by 20-min digestion with 5 μL/mL proteinase K (Sigma P8044) at 37 °C. Samples were pre-hybridized in solution at 83 °C for 30 min. MiR-23a-3p probe (Qiagen 339,111) was denatured at 65 °C for 5 min and immediately chilled on ice for 5 min. Then the hybridization was performed in solution with 40 nM of miR-23a-3p probe at 53 °C overnight. The slide was washed stringently with 1x saline-sodium citrate buffer (SSC) at 53 °C for 10 min twice and 0.5x SSC at room temperature for 10 min. The staining steps followed the manual of Alexa Fluor™ 488 tyramide kit (Thermo Fisher B40932). It was processed with standard immunofluorescence protocol using the 594/555 secondary antibody and captured by LSM 780 confocal microscope (Carl Zeiss).

All animal protocols were approved by the Committee on the Use of Live Animals in Teaching and Research of the University of Hong Kong.

To identify dysregulated miRNAs associated with sorafenib resistance in HCC, we retrieved the miRNA expression profile from the GSE56059 in the GEO database. This dataset included HCC biopsies from patients administered with sorafenib and defined patients with progressive disease (pd) as non-responders, and others as responders to sorafenib. MiRNAs were analysed through WGCNA with soft-thresholding power (β) 7 (sample clustering and analysis of the scale-free fit index for β can be found in Fig. S1A, B). A total of four merged modules were identified by gathering co-expressed miRNAs, including blue [277], brown [76], yellow [74] and grey modules[176] (Fig. 1A). When associated ME with clinical traits in each module, blue ME showed a significant correlation with cirrhosis in HCC and a negative correlation with response to sorafenib. Therefore it was submitted to further analysis (Fig. 1B, C, Fig. S1C). The top 10 upregulated miRNAs are miR-30a, miR-30b, let-7 g, miR-200c, miR-886, miR-20a, miR-27b, let-7f, miR-29b, and miR-23a (Fig. 1D). The overall survival analysis on patients with liver cancer retrieved from the Kaplan-Meier Plotter indicated that higher miR-23a prominently contributed to worse survival of HCC (AUROC = 0.6101, P = 0.0052) (Fig. S1D). Moreover, patients with overexpression of miR-23a-3p showed inferior progression-free survival (PFS) with sorafenib treatment, suggesting the negative correlation of miR-23a-3p with patients response to sorafenib (Fig. 1E).

We also evaluated the clinical significance of miR-23a-3p in HCC by in situ hybridization of miR-23a-3p on a human HCC tissue microarray (TMA) containing 90 pairs of HCC specimens and their normal adjacent liver tissues (NATs). There was no significant difference in miR-23a-3p expression between HCC, NATs and different grades of HCC tissues (Fig. S1E, F). We observed higher expression of miR-23a-3p in patients with recurrent HCC (Fig. 1F, G). To uncover the association between clinicopathologic factors and HCC patient survival outcomes, we employed both univariate and multivariate analysis using the Cox regression survival model. At univariate analysis, poor overall survival (OS) and recurrence-free survival (RFS) were associated with overexpression of miR-23a-3p (Fig. 1H, I, Fig. S1G). On the other hand, miR-23a-3p expression was significantly associated with RFS [hazard ratio (HR) = 2.37, P = 0.006] at the multivariate analysis, indicating that miR-23a-3p was an independent risk factor of HCC relapse (Fig. 1J).

As upregulated miR-23a-3p was observed in HCC patients with poor sorafenib response, whether overexpression of miR-23a-3p was related to the development of sorafenib resistance deserves further clarification. A previous study showed that both intrinsic and extrinsic mechanisms could trigger drug resistance [28]. We therefore established an in vivo-generated sorafenib-resistant HCC cell line. In brief, HCC MHCC97L cells were subcutaneously injected into the right flank of NOD/SCID immunodeficient mice. Upon tumour growth, mice were orally administrated with sorafenib or vehicle daily. Mice in the vehicle group (WT) presented continuous tumour growth. However, mice upon sorafenib treatment (R1–5) presented a temporary reduction in size with rapid regrowth after long exposure to sorafenib, indicating the acquisition of sorafenib resistance (Fig. 2A). Their body weight gradually recovered from the temporary loss at the beginning under sorafenib treatment also suggested the adaptation towards sorafenib treatment (Fig. S2A). To detect their response to sorafenib, we isolated HCC cells from the tumours and measured the half-maximal inhibitory concentration (IC₅₀) of sorafenib via MTT assay. Those cells from the regrown tumours (R1–5) exhibited a higher IC₅₀ value of sorafenib, indicating the characteristic of sorafenib resistance (Fig. 2B). Relatively high level of miR-23a-3p was observed in those in vivo-generated sorafenib resistant cells (R1–5) (Fig. 2C). To determine whether the sorafenib resistant characteristics could be maintained after several passages, we subcutaneously re-injected three resistant lines (R1, R3 and R5) and parental cells respectively into NOD/SCID mice (n = 5) as shown in Fig. 2D. Tumours of resistant lines showed no responses to sorafenib treatment (Fig. 2E, body weight of mice can be found in Fig. S2B), suggesting the acquired sorafenib resistance in these cell lines could be maintained across passages. The expression of miR-23a-3p in sorafenib resistant cells was augmented over tenfold compared to the parental tumours (Fig. 2F). These results suggested that the increased expression of miR-23a-3p in both patients biopsies and in vivo-generated sorafenib resistant cells might be responsible for the acquisition of sorafenib resistance in HCC.

To identify whether the transcription of miR-23a-3p could be directly regulated by sorafenib, we treated two commonly used HCC cell lines, MHCC97L and PLC/PRF/5 with sorafenib. The IC₅₀ values of sorafenib were calculated by plotting cell viability versus drug concentration (Fig. S3A). The downregulation of phosphorylated extracellular signal-regulated kinase (ERK) indicated positive responses to sorafenib in HCC cell lines (Fig. S3B). At doses far lower than the IC₅₀ value of sorafenib, mature miR-23a-3p was induced in a dose-dependent manner (Fig. 3A). Its primary form (pri-miR-23a) was also increased by sorafenib (Fig. 3B), indicating that sorafenib treatment stimulated the transcription activity of miR-23a-3p.

So far, several transcription factors (TFs) associated with miR-23a-3p expression have been reported to facilitate miR-23a-associated regulations during cancer development [29‐31]. However, the key TF activated by sorafenib that regulates miR-23a-3p transcription remained unidentified. Proteomics analysis was performed on MHCC97L cells treated with and without sorafenib. A total of 3626 proteins in proteomics data were inputted to Enrichr using gene set library: ENCODE and ChEA consensus TFs from ChIP-X (https://​genome.​ucsc.​edu/​ENCODE/​) and 23 proteins were enriched as potential TFs. We integrated these 92 TFs with predicted TFs of miR-23a-3p in TransmiR [103] and CircuitDB [15] and obtained 3 overlapped TFs: ETS1, NFIC, SP1 (Fig. 3C). As the ETS Proto-Oncogene 1 (ETS1) was the highest-ranking TF in the proteomics data, thereby it was subjected to further validation (Fig. 3D).

It was well known that ETS1 is a member of the ETS family, which presents a conserved ETS DNA-binding domain recognizing GGAA/T sequence in target genes. High level of ETS-1 was reported to be associated with poor prognosis of patients with advanced HCC who were treated with sorafenib. High level of ETS-1 promoted sorafenib resistance by inducing the expression of multidrug resistance-related genes [32]. To investigate whether ETS1 was responsible for miR-23a-3p overexpression, we used ETS1 siRNA to inhibit ETS1 expression (Fig. S3C). It was found that inhibiting ETS1 reduced miR-23a-3p expression and effectively restrained the sorafenib-induced enhancement of miR-23a-3p in both MHCC97L and PLC/PRF/5 cells (Fig. 3E, F). To confirm the transcriptional regulation of ETS-1 on miR-23a-3p promoter, we detected the luciferase activity of the miR-23a-3p promoter. It was revealed that ETS1 inhibition potently neutralized sorafenib-induced transcriptional activation, which is consistent with the regulation on miR-23a-3p expression (Fig. 3G). Prediction for the binding motif of ETS1 in JASPER suggested that AGGAAG from − 360 to -365 nt before miR-23a-3p promoter was one of the key motifs. To experimentally evaluate their binding relationship, we conducted a chromatin immunoprecipitation experiment followed by quantitative PCR (ChIP-qPCR) with specific primers. The enrichment of predicted promoter fragments was significantly higher after the pulldown by ETS1 antibody (Ab) in the presence of sorafenib (Fig. 3H). These data indicated that ETS1 was the key TF that directly stimulated miR-23a-3p transcription under sorafenib treatment.

To further understand the role of miR-23a-3p in mediating sorafenib response in HCC, we established the miR-23a-3p knockout MHCC97L cell line (23a-KO) via CRISPR-Cas9 editing (Fig. S4A). The orthotopic hepatic tumour model formed by 23a-KO cells and its control pair (Scramble) was randomly grouped and received sorafenib (25 mg/kg) every other day for 4 weeks, as shown in Fig. S4B. The knockout of miR-23a-3p slightly delayed the growth of orthotopic tumours in the liver while dramatically decreased the growth of 23a-KO tumours in mice receiving sorafenib (Fig. 3A, the body weight of mice could be found in Fig. S4C). At the endpoint of the experiment, isolated livers confirmed the smaller size of tumours in 23a-KO mice than in Scramble mice administrated with sorafenib (Fig. 3B). To determine whether the inhibition of tumour growth by 23a-KO under sorafenib administration is caused by apoptosis induction in tumour cells, we stained the cleaved caspase-3 to indicate the apoptotic cell. It demonstrated that 23a-KO alone and combined with sorafenib showed significantly more apoptotic cells than their control group (Fig. 4C). The level of cleaved caspase-3 and PARP protein of 23a-KO tumours also supported the induction of apoptosis in 23a-KO cells treated with sorafenib (Fig. S4D).

In in vitro study, we transfected HCC cell lines with either miR-23a-3p mimics (miR-23a-3p) or its inhibitor (Anti-miR-23a) to up-regulate or suppress miR-23a-3p (Fig. S4E). MiR-23a-3p overexpression significantly led to the poor response of HCC cells to sorafenib, while miR-23a-3p suppression sensitized HCC to sorafenib (Fig. 4D). Such effects were independent of MEK/ERK signalling pathway (Fig. S4F). The apoptosis of HCC cells was detected by Annexin V/7-AAD and immunoblotting analysis. The results showed that miR-23a-3p overexpression inhibited cell death, which was manifested in reduced apoptotic cells and cleaved forms of caspase-3 and PARP; while miR-23a-3p suppression behaved the opposite way (Fig. 4E, F). These in vitro and in vivo observations suggested that overexpression of miR-23a-3p attenuated HCC cell response to sorafenib; and inhibition of miR-23a-3p potentiated sorafenib response.

Proteomic analysis was conducted in MHCC97L cells transfected with miR-23a-3p mimics and NC to examine potential pathways involved in miR-23a-3p-promoted sorafenib resistance. In addition, proteomics data on MHCC97L cells treated with sorafenib were also integrated for further analysis. A total of 178 differentially expressed proteins found in the three groups of cells were subjected to KEGG pathway enrichment analysis on Metascape (http://​metascape.​org/​). Notably, molecules associated with ferroptosis were enriched together with their associated pathways such as serine biosynthesis, glutamate metabolism, apoptosis, lysosome, and protein processing in the endoplasmic reticulum (Fig. 5A).

Recent studies suggested that ferroptosis is a form of iron-dependent programmed cell death (PCD) characterized by accumulation of peroxidised lipid products, iron overload and glutathione (GSH) synthesis which contributed to sorafenib resistance in HCC [33]. Therefore, ferroptosis might be involved in miR-23a-3p-mediated sorafenib resistance. A heatmap showing the protein expression patterns among these three groups of cells were grouped into eight clusters. The KEGG pathway of each gene in the clusters was annotated (Fig. 5B). According to the commonly inhibitory effect of miRNA and upregulation of miR-23a-3p by sorafenib, genes in cluster 4 might be potentially responsible regulators of ferroptosis. Hence, we determined the expression of the key regulator glutathione peroxidase 4 (GPX4), which inhibits ferroptosis by reducing lethal lipid peroxides (LPO) and Acyl-CoA synthetase long-chain family member 4 (ACSL4), a necessary enzyme for catalysing lipid peroxidation to trigger ferroptosis. We found that sorafenib treatment alone led to a decrease of GPX4, indicating induction of ferroptosis as supported by the previous study [34]. However, miR-23a-3p overexpression could remarkably enhance GPX4 but restraint the enhancement of ACSL4 by sorafenib, suggesting suppression of sorafenib-induced ferroptosis. The opposite effects caused by Anti-miR-23a further confirmed the above result (Fig. 5C). The cellular level of excessive chelatable iron was measured by fluorescent indicator Phen Green SK, which functions as an indicator of ferroptosis initiation leading to a dynamic fluorescence quenching. A significant reduction of intracellular iron in miR-23a-3p overexpressed HCC cells was observed in the presence of sorafenib. In contrast, an augmentation of intracellular iron induced by sorafenib was seen in Anti-miR-23a HCC cells (Fig. 5D). The detection of lipid peroxides deposition by BODIPY staining also suggested that miR-23a-3p-overexpression prominently attenuate sorafenib-induced ferroptosis (Fig. 5E). Moreover, the augmented sorafenib-induced cell death by Anti-miR-23a was diminished when we used ferrostatin-1 (Fer-1), an effective ferroptosis inhibitor (Fig. 5F). Taken together, these results suggested that miR-23a-3p could suppress sorafenib-induced ferroptotic cell death in HCC cells.

The protein expression pattern in the 4th cluster of the heatmap suggested that these proteins were associated with the effect of sorafenib treatment. Therefore, these proteins might be potential targets of miR-23a-3p as they were significantly decreased in miR-23a-3p overexpressed cells, and their expression upon sorafenib treatment was restrained by miR-23a-3p overexpression. Among these proteins, ACSL4 and CPOX (Coproporphyrinogen oxidase) were enriched in ferroptosis, which have been confirmed to be suppressed by miR-23a-3p under sorafenib treatment. Therefore, we narrowed the search scope of target genes to these two genes. The calculation by the algorithm of RNA hybrid showed a more favourable base pairing between the “seed” region of miR-23a-3p and the 3’UTR of ACSL4 mRNA compared to CPOX 3’UTR (Fig. 5A, Fig. S5A). Therefore, we selected ACSL4 for further validation. By detecting ACSL4 expression in HCC cells with miR-23a-3p mimics or Anti-miR-23a, and 23-KO HCC tumour tissues, it was found that miR-23a-3p negatively regulated the mRNA and protein expression of ACSL4 in HCC cells and tumour tissues (Fig. 6B, C). Luciferase assay demonstrated that miR-23a-3p significantly inhibited the luciferase activity of ACSL4 3’UTR (Fig. 6D), which confirmed that ACSL4 was a target gene of miR-23a-3p.

To further examine if ACSL4 was required for miR-23a-3p-suppressed ferroptosis in sorafenib-treated HCC, we co-transfected miR-23a-3p inhibitor with ACSL4 siRNA to HCC cells in the presence or absence of sorafenib (Fig. S5B, C). Suppression of ACSL4 markedly weakened the Anti-miR-23a-induced cellular iron deposition and the lipid peroxides accumulation in the presence of sorafenib (Fig. 6E-G). In addition, the induction of total reactive oxygen species (ROS) by co-treatment of sorafenib and Anti-miR-23a was also limited by ACSL4 inhibition (Fig. S5C, D). Sorafenib-induced ferroptotic cell death was determined by MTT assay. It showed that the augmentation of sorafenib-induced ferroptotic cell death by Anti-miR-23a could be attenuated by ACSL4 siRNA (Fig. 6H, I). The negative correlation between miR-23a-3p and ACSL4 was identified in human HCC TMA via dual-staining of miR-23a-3p and ACSL4 (Fig. 5E, F). These results demonstrated that ACSL4 was the target of miR-23a-3p that prominently regulated sorafenib-induced ferroptosis.