YTHDF2 reduction fuels inflammation and vascular abnormalization in hepatocellular carcinoma

Despite the m⁶A decrease in total RNA of human HCC, there is still no consensus about mRNA modification [8, 13]. To assess the global m⁶A-methylation in HCC, we examined 37 tumor paired with adjacent non-tumor samples. By performing dot blot or liquid chromatography-tandem mass spectrometry (LC-MS/MS), we observed a slight m⁶A decrease in total RNA, and yet a significant m⁶A increase in mRNA, while comparing tumor tissues with autologous paratumor tissues (Additional file 1: Figure S1A, B and Fig. 1a). Next, eight pairs of mRNA samples were immunoprecipitated using an m⁶A-specific antibody in combination with high-throughput sequencing (MeRIP-seq). An average of 70.80 and 29.20% m⁶A peaks showed a significant increase and decrease, respectively, in tumor relative to non-tumor samples (Additional file 1: Figure S1C). Through analysis of RNA sequencing (RNA-seq) data, we noticed most transcripts that gained in m⁶A manifested higher mRNA expression in tumor versus non-tumor counterparts. The superiority of ‘hyper-up’ population was found in 7 out of 8 patients, averagely taking up 69.27% amongst all the m⁶A-labelled transcripts (Fig. 1b and Additional file 1: Figure S1D, E). We then analyzed those high-frequency ‘hyper-up’ genes (at least shared by 3 patients), which are mainly responsible for cancer-promoting events, as indicated by Gene ontology (GO) analysis (Additional file 1: Figure S1F).

Using gene set enrichment analysis (GSEA), a previously established hypoxia signature [32] was strongly skewed toward tumor counterparts in the case of HCC (Fig. 1c). A recent study suggested hypoxia stimulated m⁶A demethylation⁶ whereas another report showed hypoxia favored m⁶A labelling [24]. Therefore, we exposed HCC cell lines to hypoxia and observed a gain of m⁶A in all four cell lines as compared to normoxic conditions (Fig. 1d). Using MeRIP-seq, we detected a larger number of m⁶A peaks in hypoxic SMMC7721 mRNA, preferentially enriched in 3′ untranslated regions (3’UTR) and coding regions (CDS) (Fig. 1e, f). As profiled by RNA-seq, transcripts gained in m⁶A upon hypoxia manifested profoundly increased expression (Fig. 1g), which recapitulated the ‘hyper-up’ pattern identified in HCC specimens. GO analysis of those ‘hyper-up’ genes suggested activation of hypoxia-related oncogenic pathways (Additional file 1: Figure S1G). Notably, gene signatures of cell-cell contact and ‘serine-type endopeptidase inhibitor activity’ are enriched in ‘hyper-up’ populations based on both tumor tissue and hypoxic cell profiling (Additional file 1: Figure S1F, G).

To analyze how hypoxia was linked to m⁶A-mRNA editing, we measured enzyme expression in HCC tissues as well as hypoxic cancer cells. Among methyltransferases, METTL14 mRNA displayed significant alteration in tumor (Additional file 1: Figure S2A), and yet its decrease could not explain the observed increase in global m⁶A. In concert with m⁶A hypermethylation, FTO and ALKBH5 mRNAs decreased in tumor tissues; however, they both increased under hypoxic conditions (Additional file 1: Figure S2A, B). Notably, YTHDF2 mRNA was reduced in both hypoxic cells and tumor tissues, which indicated an inverse correlation with the m6A/A ratio (Additional file 1: Figure S1H and 1I, J). Putatively, ‘hyper-up’ regulation of the m⁶A-epitranscriptome in HCC might result from the YTHDF2 reduction and consequential m⁶A-mRNA stabilization. We verified YTHDF2 protein levels were also decreased in hypoxic cells and tumor specimens (Additional file 1: Figure S1I, 1H and S2C). Importantly, in a setting of 200 HCC patients (recruited from 2006 to 2012), lower YTHDF2 protein levels were significantly associated with more multinodular tumors and microvascular invasion, higher TNM and BCLC stage classification, and shorter overall and recurrence-free survival period (Additional file 2: Table S1, Additional file 1: Figure S2D and 1 K). Accordingly, we hypothesized that a hypoxia-sensitive YTHDF2 reduction reprogrammed the m⁶A-edited transcriptome and promoted HCC development.

A recent report has preliminarily showed that overexpression of YTHDF2 could suppress HCC growth [19], but the importance of endogenous YTHDF2 in HCC cells was still not clear. Hence, we stably silenced YTHDF2 expression in human HCC cell line SMMC7721 and MHCC97H (Additional file 1: Figure S3A, B), and assessed their biological behaviors. Disruption of YTHDF2 greatly benefited cell viability but did not retard apoptosis (Fig. 2a and Additional file 1: Figure S3C, D). Impressively, this acquired growth competence was conserved under hypoxic conditions (Fig. 2a). In terms of hypoxia-related events, we further evaluated cell migration, stemness, metabolism, and tumor angiogenesis [33]. YTHDF2 knockdown resulted in a promotion of angiogenic sprouting of human umbilical vein endothelial cells (HUVECs) in a co-culture system (Fig. 2b and Additional file 1: Figure S3E), but unaffected other cellular behaviors (Additional file 1: Fig. S3F-H). We next exploited a subcutaneous xenograft mouse model to monitor in vivo tumor development. YTHDF2 silencing robustly facilitated tumor growth and metastasis in NPG (NOD-Prkdc^scidIl2rg^null) mice (Fig. 2c-e), and this was partially due to the effects on cell proliferation and tumor angiogenesis (Additional file 1: Figure S3I, K and Fig. 2f). It is well-documented that tumor vascular permeability and mimicry contributes to extravasation and metastasis [28, 34]. In line with this, YTHDF2 silencing in MHCC97H tumors led to increased dextran leakage and PAS⁺CD31⁻ fluid-filled channels lined by tumor cells, indicating intensified vascular permeability and mimicry (Additional file 1: Figure S3J, L and Fig. 2g). In contrast, overexpression of YTHDF2 not only arrested cancer cell growth but also reduced vessel density and permeability (Additional file 1: Fig. S4A-H).

To obtain a deeper insight into the impact of YTHDF2 deficiency on HCC, we generated the liver-specific Ythdf2 KO mouse strain (Ythdf2^LKO) and employed a chemical-induced HCC model. As compared with Ythdf2^F/F littermates, Ythdf2^LKO mice developed more advanced liver lesions, in terms of macroscopic number, size and histology, and more lung metastases (Fig. 2h-j and Additional file 1: Figure S4I). Tumor proliferation, apoptosis and angiogenesis were enhanced in the YTHDF2-deficient mouse liver (Additional file 1: Figure S4J, K and Fig. 2k). Nonetheless, no obvious vascular mimicry existed in either Ythdf2^F/F or Ythdf2^LKO liver tumors (data not shown). To our knowledge, the microvascular network in solid tumors is often compromised with functional abnormalities, which fuel metastasis formation [35]. In the absence of YTHDF2, NG2⁺ pericytes were significantly reduced, whereas CD31⁺ endothelial cells largely accumulated and formed microvessels in tumor regions (Fig. 2k), defining a fundamental role for YTHDF2 in vascular normalization [35].

We subsequently performed RNA-seq in control and YTHDF2-deficient SMMC-7721 cells. Taking into consideration that putative targets were retarded for degradation, we mainly focused on the upregulated transcripts (Fig. 3a). GO analysis revealed an enrichment of a cancer-promoting inflammation program, as characterized by the top GO term ‘secretion by cell’ and ‘positive regulation of tyrosine phosphorylation of Stat3 protein’ (Fig. 3b and Additional file 3: Table S2). We did not see epidermal growth factor receptor (EGFR) among the upregulated genes, although it has been recently reported elsewhere [19]. To find out the key determinants in a hypoxic cancer context, we used a combined screening approach by integrating YTHDF2-regulated transcriptional profiling and the hypoxia-responsive epitranscriptome. Therefore, we selected 182 genes that were increased by over 2-fold upon YTHDF2 knockdown, and 1122 hypermethylated transcripts that were upregulated by over 2-fold under hypoxic conditions (Fig. 3c). Precisely, these genes intersected for 21 candidate targets, among which IL11 and SERPINE2 were most relevant to cancer-promoting inflammation. IL-11 promotes STAT3 activation and inflammatory cancer progression in an autocrine manner [27, 36], while cancer cell-secreted serpin E2 confers invasive and metastatic processes by reprogramming the tumor vasculature [28, 37‐39]. We validated that both STAT3 phosphorylation and IL11 and SERPINE2 expression were upregulated in YTHDF2-deficient cells, and this effect could be reinforced by oxygen deficit (Fig. 3d-g and Additional file 1: Figure S5A-D). In parallel, YTHDF2 deficiency in mouse hepatocytes yielded IL11 and SERPINE2 expression (Additional file 1: Fig. S5E). Conversely, RNA profiling and qPCR detection of YTHDF2-overexpressed cells showed substantial downregulation of IL11 and SERPINE2, as well as phosphorylation of STAT3, in comparison to the control cells (Fig. 3i-j and Additional file 4: Table S3). Unexpectedly, short-term (12 h) exposure to hypoxia unleashed the repressive properties of YTHDF2 toward IL11 and SERPINE2 mRNAs (Fig. 3I).

Using m⁶A-specific qPCR, we uncovered that YTHDF2 induced m⁶A loss in IL11 and SERPINE2 transcripts (Fig. 4a). To investigate whether YTHDF2 directly recognized m⁶A-marked IL11 and SERPINE2 mRNAs, we performed mRNA-immunoprecipitation (RIP) for YTHDF2. Subsequent RT-qPCR identified IL11 and SERPINE2 as YTHDF2 substrates, particularly under hypoxic condition (Fig. 4b). We then asked whether YTHDF2 processes IL11 and SERPINE2 mRNAs to decay. After actinomycin D treatment, their lifetimes were prolonged in YTHDF2-silenced cells and were shortened in YTHDF2-overexpressed cells (Fig. 4c, d). Although hypoxia dramatically delayed IL11 and SERPINE2 degradation in control cells, YTHDF2 overexpression expedited the processing to a normal level. Since human HCC tissues exhibited high abundance of m⁶A modification to the consensus sites of IL11 and SERPINE2 3’UTRs (Fig. 4e), we inserted their transcript segments harboring m⁶A motifs into a pmirGLO reporter plasmid [17]. Under hypoxic conditions, introduction of exogenous YTHDF2 efficiently disrupted reporter activity in the presence of m⁶A motifs (Fig. 4f). Using fluorescence in situ hybridization (FISH), we visualized localization of YTHDF2 and target mRNAs to processing bodies (Fig. 4g, h). As expected, YTHDF2 affected both expression and distribution of IL11 and SERPINE2 mRNAs. Noticeably, 12 h of exposure to hypoxia induced YTHDF2 translocation to processing bodies, substantially supporting its provisional potency propelled by a lack of oxygen (Fig. 4g).

As a specific m⁶A ‘reader’ protein, YTHDF2 depends on the hydrophobic residues W432 and W486 in its carboxy-terminal YTH domain for selective recognition of m⁶A [17, 40]. To understand the molecular basis of our findings, we mutated the m⁶A recognition sites into W432A and W486A, which abrogate the specific binding affinity of YTHDF2 (Additional file 1: Figure S6A, B). A technical deviation is that the strain carrying a W432A mutation showed a little bit lower YTHDF2 protein level than wild-type control, but the W486A version was comparable to wild-type. In YTHDF2-deficient cells, re-expression of wild-type YTHDF2 but not the catalytically inactive mutant offset the acquired cancer-promoting inflammatory phenotypes (Fig. 5a, b), and reduced both phosphorylation of STAT3 and stabilization of IL11 and SERPINE2 mRNAs (Fig. 5c-e and Additional file 1: Figure S6C, D). In concert with this, redundant expression of catalytically dead YTHDF2 failed to inhibit in vivo tumor development (Fig. 5f and Additional file 1: Figure S6E). These results suggest that the m⁶A reader function is indispensable for the suppressive role of YTHDF2 in HCC.

To ascertain the importance of YTHDF2-IL11/SERPINE2 pathway in HCC development, we silenced IL11 or SERPINE2 in YTHDF2-deficient cells by using shRNAs. Removal of IL11 abrogated the growth competitiveness of YTHDF2-deficient SMMC7721 cells (Fig. 5g), while a SERPINE2 deficit attenuated the proangiogenic capacity of SMMC7721 cells (Fig. 5h), highlighting that IL11 and SERPINE2 are key targets of YTHDF2 in cancer-promoting inflammation. As anticipated, concomitant targeting of both IL11 and SERPINE2 counterbalanced the malignant capability endowed by YTHDF2 depletion (Fig. 5i and Additional file 1: Figure S6F). To examine this pathway in a clinical setting, we looked into human HCC tissues with relatively low or high YTHDF2 protein levels. As shown in Fig. 5j, IL-11 and Serpin E2 expression displayed negative correlations with YTHDF2.

Hypoxic tumor areas showed relatively low YTHDF2 expression in the xenograft mouse model (Fig. 6a), providing an in vivo evidence for hypoxia-mediated YTHDF2 reduction. A recent report showed that HIF-1α was responsible for this negative regulation [19]. However, when we sought to block the hypoxia signaling in vitro, only targeting HIF-2α completely rescued YTHDF2 expression in hypoxic cells (Additional file 1: Figure S7A-C and Fig. 6b, c). Containing the hypoxia-responsive elements (HREs, CGTG), the Ythdf2 promoter was assessed through chromatin immunoprecipitation (ChIP)-qPCR and demonstrated HIF-2α binding (Fig. 6d). Additionally, we inserted the Ythdf2 promoter into a pGL4 luciferase reporter plasmid. While Ythdf2 promoter activity was decreased upon hypoxia, this was reversed by a HIF-2α (but not HIF-1α) siRNA (Fig. 6e and Additional file 1: Figure S7D). In human tissues, HIF-2α was detected in the cytoplasm of non-tumor hepatocytes (Additional file 1: Fig. S7E), whereas it was aggregated in the nucleus of tumor cells, where it could regulate gene transcription.

Since HIF-2α often terminates further expansion of hypoxic tumors, a small molecule (PT2385) selectively targeting HIF-2α has been developed to treat renal cell carcinoma and hepatocellular carcinoma [29, 30]. We tested PT2385 in treating HCC cells in vitro and in vivo, and found it exhibited favorable therapeutic effects (Fig. 6f, g, k). PT2385 rescued YTHDF2 expression (Fig. 6h) but unaffected YTHDF2 localization in hypoxic cells (data not shown), accompanied by neutralization of IL11 and SERPINE2 expression, as well as STAT3 phosphorylation (Fig. 6i, j). Nonetheless, PT2385 administration to the mice bearing YTHDF2-deficient tumors failed therapeutically (Fig. 6l), which was indicative of an indispensable role of YTHDF2 in treating hypoxic HCC.

Tumor tissues and matched adjacent non-tumor tissues were obtained from liver cancer patients who underwent liver resection at the Department of Liver Surgery, and diagnosed as HCC by the Department of Pathology, Renji Hospital, School of Medicine, Shanghai Jiaotong University. All samples were collected with informed consent, and the experiments were approved by the ethical review committee of the World Health Organization Collaborating Center for Research in Human Production (authorized by the Shanghai Municipal Government). A total number of 200 patients enrolled from 2006 to 2012 were all treatment naïve before surgery, and were routinely followed up after curative liver resection. Portions of surgically resected tissues were immediately transferred to liquid nitrogen until RNA or protein extraction. Paraffin-embedded tumors were processed into tissue microarrays for histological staining and prognostic analysis. Detailed clinicopathologic features were summarized in Additional file 2: Table S1.

Poly(A)⁺ mRNA samples were denatured at 65 °C for 5 min in 3 sample volumes of RNA incubation buffer. An equal volume of chilled 20 × SSC buffer (Sigma-Aldrich) was then added before samples were spotted on the nitrocellulose membrane. After UV crosslinking for 10 min, the membrane was washed with 1 x PBST buffer, blocked with 5% non-fat milk and incubated with anti-m⁶A antibody (1:2000, Synaptic Systems) overnight at 4 °C. Then HRP-conjugated goat anti-rabbit IgG (Jackson Immuno Research Laboratories) was added to the blots for 1 h at room temperature and the membrane was developed with a ChemiDoc XRS system (Bio-Rad).

100–200 ng of mRNA was digested by nuclease P1 (2 U) in 25 μl of buffer containing 25 mM NaCl, and 2.5 mM ZnCl₂ at 42 °C for 2 h, followed by the addition of NH₄HCO₃ (1 M, 3 μl) and alkaline phosphatase (0.5 U) and incubation at 37 °C for 2 h. The sample was then filtered (0.22 μm pore size, 4 mm diameter, Millipore), and 5 μl of the solution was injected into the LC-MS/MS. The nucleosides were separated by reverse- phase ultra-performance liquid chromatography on a C18 column with online mass spectrometry detection using an Agilent 6410 QQQ triple-quadrupole LC mass spectrometer in positive electrospray ionization mode. The nucleosides were quantified by using the nucleoside-to-base ion mass transitions of 282 to 150 (m⁶A) and 268 to 136 (A). Quantification was carried out by comparison with a standard curve obtained from pure nucleoside standards run with the same batch of samples. The m⁶A/A ratio was calculated based on the calibrated concentrations.

Total RNA was extracted from human HCC tissues or SMMC7721 cells with TRIzol (Invitrogen), then polyadenylated RNA was enriched using an Oligotex® mRNA purification kit (Qiagen). In particular, additional DNase I digestion was applied to all samples to avoid DNA contamination. RNA fragmentation, m⁶A-IP and library preparation were proceeded according to previously published protocols [41]. Sequencing was conducted at the University of Chicago Genomics Facility on an Illumina HiSeq2500 machine in single-read mode with 50 base pairs per read. Enrichment of m⁶A-containing transcript segments was also analyzed through RT-qPCR. Primers targeting m⁶A-enriched regions of IL11 and SERPINE2 are listed: IL11_F, TCCAACAGTGAGGGTTAAGCAA; IL11_R, CCCTGAATGACAGTCCCTGC; SERPINE2_F, GGCCTCATGACAACATCGTG; SERPINE2_R, CGAGCTGCTTCTTGGTCCTG.

RNA-seq read counts for 14 genes that make up the hypoxia metagene signature (ALDOA, MIF, TUBB6, P4HA1, SLC2A1, PGAM1, ENO1, LDHA, CDKN3, TPI1, NDRG1, VEGFA, ACOT7 and ADM) [32] were analyzed with GSEA software v.2.0, which is available from the Broad Institute ().

A number of 5 × 10⁶ SMMC7721 or MHCC97H cells re-suspended in 100 μl of PBS were subcutaneously injected into the right flank of 6-week old male NCG mice (NOD-Prkdc^scid Il2rg^null) (Model Animal Research Center of Nanjing University). Tumor growth was monitored by caliper rule every 3–5 days. Primary tumor volume was measured using the formula: Volume = (Length × Width ²)/2. The mice were euthanized when the tumor length reached 15 mm or at the indicated date post tumor injection. Tumors and lungs were harvested for further histological study. For HIF-2 antagonism, PT2385 (c) was suspended in saline with 0.5% sodium carboxymethyl cellulose, 2.5% Tween 80 and 2.5% dimethyl sulfoxide. NPG mice were randomly distributed into 2 groups when the subcutaneous tumors reached an average size of 200 mm³, and administered vehicle or PT2385 (20 mg/kg o.p.d.) by gavage until endpoint (when the tumor size in control group reached 15 mm).

Ythdf2^F/F mice in C57BL/6 background were described elsewhere [48]. For liver-specific disruption, Ythdf2^F/F mice were crossed with mice harboring the Cre recombinase under control of the albumin promoter to obtain the Ythdf2^LKO strain. Male Ythdf2^F/F and Ythdf2^LKO littermates were injected intraperitoneally with a single dose of 25 mg/kg diethylnitrosamine (DEN) (Sigma-Aldrich) on postnatal day 14, and 6 weeks later administrated CCl₄ (1 ml/kg) once per week for 12 weeks. Mice were euthanized at 8.5-month old, and the livers and lungs were harvested for tumor assessment and histological experiments. The primers used for genotyping are listed: 5’Ythdf2_F, GCTTGCCTGCTACATAGTGAGA; 5’Ythdf2_R, AACTGAACTGCTTAACCTTCTGG; 3’Ythdf2_F, GAACGGTATTGTCGGTATTGTCA; 3’Ythdf2_R, AGACCACTCCAACACAGAACTT.

All mice were maintained under specific pathogen-free (SPF) conditions, on a 12 h light-dark cycle. All mouse experiments were approved by the Shanghai Administrative Committee for Laboratory Animals.

The procedure was adapted from the previous protocol [49] and was described in our recent work [16]. Briefly, a total of 5 × 10⁷ SMMC7721 cells expressing Flag-tagged YTHDF2 were collected and lysed in a buffer containing 150 mM KCl, 10 mM HEPES (pH 7.6), 2 mM EDTA, 0.5% CA-630, 0.5 mM DTT, 1:100 protease inhibitor cocktail, and 400 U/ml RNase inhibitor. Cell lysate was incubated with Flag M2 beads at 4 °C for 4 h. Then the beads were washed with NT2 buffer and incubated in elution solution containing Flag peptide (Sigma-Aldrich). After washing, bound RNAs were extracted using TRIzol reagent and analyzed by RT-qPCR.

SMMC7721 cells were exposed to hypoxia or normoxia, and 12 h later actinomycin D (MCE) was added at a concentration of 5 mg/ml. At the indicated time points, the cells were trypsinized and collected for RNA purification. RNA quantities were determined by RT-qPCR. The degradation rate of RNA (k) was calculated using the equation: where t is the transcription inhibition time, k is the degradation rate, and N_t and N₀ are the RNA quantities at time t and time 0. The RNA lifetime (t_1/2) can be calculated from the degradation rate as follows:

At least three biological replicates were used in each experiment unless otherwise stated. Data were analyzed with GrapPad Prism 7 and were presented as the mean ± standard error of the mean (SEM). Comparisons between two groups were assessed using unpaired or paired (for matched comparisons) two-tailed Student’s t-test, or non-parametric Mann-Whitney U-test. Multiple comparisons were assessed by one-way ANOVA. Survival rates were compared using the log-rank test. Pearson correlation coefficients (r) were calculated to assess correlation and statistical significance was assessed by a two-tailed t-test of r = 0. The statistical significance of clinicopathological differences among HCC patients were assessed by χ²-test.

More detailed materials and methods including Cell culture, Constructs and transfections, Immunohistochemistry, Immunoblotting, Real-time quantitative PCR, Immunofluorescence, Cell proliferation, migration and mammosphere assay, Tube formation assay, Enzyme-linked immunosorbent assay, Luciferase reporter assay, Chromatin immunoprecipitation, m⁶A-seq data analyses are provided in Additional file 5.