The m6A reader YTHDC2 inhibits lung adenocarcinoma tumorigenesis by suppressing SLC7A11-dependent antioxidant function

The biological functions of N6-methyladenosine (m6A) RNA methylation are mainly dependent on the reader; however, its role in lung tumorigenesis remains unclear. Here, we have demonstrated that the m6A reader YT521-B homology domain containing 2 (YTHDC2) is frequently suppressed in lung adenocarcinoma (LUAD). Downregulation of YTHDC2 was associated with poor clinical outcome of LUAD. YTHDC2 decreased tumorigenesis in a spontaneous LUAD mouse model. Moreover, YTHDC2 exhibited antitumor activity in human LUAD cells. Mechanistically, YTHDC2, via its m6A-recognizing YTH domain, suppressed cystine uptake and blocked the downstream antioxidant program. Administration of cystine downstream antioxidants to pulmonary YTHDC2-overexpressing mice rescued lung tumorigenesis. Furthermore, solute carrier 7A11 (SLC7A11), the catalytic subunit of system XC−, was identified to be the direct target of YTHDC2. YTHDC2 destabilized SLC7A11 mRNA in an m6A-dependent manner because YTHDC2 preferentially bound to m6A-modified SLC7A11 mRNA and thereafter promoted its decay. Clinically, a large proportion of acinar LUAD subtype cases exhibited simultaneous YTHDC2 downregulation and SLC7A11 elevation. Patient-derived xenograft (PDX) mouse models generated from acinar LUAD showed sensitivity to system XC− inhibitors. Collectively, the promotion of cystine uptake via the suppression of YTHDC2 is critical for LUAD tumorigenesis, and blocking this process may benefit future treatment.


Introduction
Lung adenocarcinoma (LUAD) is the most prevalent type of nonsmall cell lung cancer [1]. However, the mechanisms underlying LUAD tumorigenesis are largely unknown. Therefore, it is urgent to further explore the molecular pathogenesis of LUAD to develop novel treatment strategies and reduce mortality.
Numerous studies have demonstrated the crucial role of N6-methyladenosine (m 6 A) RNA methylation in various human cancers, including lung cancer [2][3][4][5]. m 6 A RNA methylation is the most common internal posttranscriptional methylation of mRNAs. It is a reversible process controlled by the m 6 A WER system, comprised of writers (W), erasers (E), and readers (R). The major m 6 A writer is a methyltransferase complex, while the eraser is an enzyme that catalyzes demethylation. In addition to the biological effect on target transcripts, the final outcome of m 6 A methylation is determined on the destiny after recognition and binding by the specific reader. Although the function of several m 6 A readers has been established in liver, colorectal, and ovarian cancer [4,6,7], to the best of our knowledge, the role of readers in LUAD is largely unknown. YTHDC2 functions as a reader to regulate spermatogenesis and suppress liver steatosis [8][9][10]. YTHDC2 is believed to be the final member of the YTH protein family. However, whether and how YTHDC2 exerts its function in LUAD has not yet been investigated. Although m 6 A methylation is known to be an important mechanism involved in cell proliferation, autophagy and apoptosis [11][12][13], its role in metabolic processes is still not known. Small molecules, such as erastin and sorafenib, can trigger lipid peroxidation by suppressing system X C − , a cystine/glutamate antiporter [14][15][16]. System X C − maintains redox homeostasis through the import of cystine, an essential precursor for glutathione (GSH) biosynthesis. Inhibiting system X C − impairs tumor growth and induces cell death [16,17]. System X C − is a heterodimer that consists of two core components: SLC7A11 and solute carrier 3A2 (SCL3A2). Previous studies have revealed that upregulation of SLC7A11 is negatively correlated with lipid peroxidation [14]. SLC7A11 can be regulated at the transcriptional and translational levels [16,18,19]. However, whether SLC7A11 is modulated at the m 6 A level in LUAD remains elusive thus far.
In this study, we found that YTHDC2 inhibits LUAD tumorigenesis in vivo and in vitro and is largely dependent on its YTH domain. Furthermore, SLC7A11 mRNA was modified by m 6 A and bound by YTHDC2. YTHDC2 promotes SLC7A11 mRNA degradation in an m 6 A-dependent manner. Using cell-based experiments, mouse models, and clinical LUAD specimens (total 382), our data demonstrate the critical tumor suppressor roles of the m 6 A reader YTHDC2 in LUAD tumorigenesis.

Clinical specimen
Tissues were obtained from 382 LUAD patients who had received curative surgery at Shanghai Chest Hospital and the Second Hospital of Shandong University. For cohort#1, 192 LUAD patients were recruited from January 2018 to March 2019; for cohort#2, 100 LUAD patients with acinar predominant subtype were recruited from December 2013 to December 2014; for cohort#3, 30 LUAD patients were recruited from north China; and for cohort#4, 60 LUAD patients were recruited from March 2017 to June 2017. Written informed consents were obtained from each enrolled patient, according to the guidelines of the declaration of Helsinki. This study was approved by Institutional Ethics Committee of Shanghai Chest Hospital.

Bioinformatics
We used GEPIA, UALCAN, TCGA and Oncomine database to analyze the transcriptional level of genes. The GEPIA and the Kaplan-Meier plotter databases were used to analyze survival information. The correlation between YTHDC2 and SLC7A11 was evaluated in GEPIA database.

Animal experiments
Kras LSL-G12D/+ ; p53 LSL-R172H/+ (KP) strains were purchased from the Jackson Laboratory. The KP mice were intranasal administered with 25 μl adeno-associated type 5 virus (AAV5) particles (2 × 10 12 viral particles/ml) encoding for Cre recombinase (Cre), thereby initiated lung tumorigenesis. The mice were also simultaneously intranasal infected with AAV5 encoding YTHDC2 WT (WT indicates wide-type of YTHDC2) or YTHDC2 ΔYTH (ΔYTH indicates YTH domain dele of YTHDC2) under anaesthesia. Meanwhile, the empty-AAV5 was parallel infected and treated as a negative control. Thereby, the KP-based pulmonary expressing YTHDC2 WT (KPY WT ), YTHDC2 ΔYTH (KPY ΔYTH ) and control KPE mice were generated. Afterwards, two mice/group were sacrificed weekly to monitor the incidence of lung tumor from 6 to 12 wk after initiation of tumorigenesis. For drug assessment, vehicle, N-acetylcysteine (NAC, 100 mg/kg/day, Sigma, #A7250, St Louis, MO, USA) or GSH (100 mg/kg/day, Solarbio, #G8180, Beijing, China) dissolved in distilled water were intraperitoneally injected into the KPE and KPY WT mice for 4 weeks beginning at 10 wk post AAV5 infection (n = 8 mice/ group). The survival rate was calculated by Kaplan-Meier method and statistical significance was assessed by log-rank tests.
For xenograft experiments, H1299 (1 × 10 7 ) or H1975 cells (1.5 × 10 7 ) with different treatment were subcutaneously injected into 4-6week-old athymic nude mice (n = 8/group). Tumor weight and cystine uptake were assessed after sacrificing the mouse at day 40 after implantation. The tumor volume was calculated as 0.5 × L × W 2 , (L indicates length, while W indicates width). All mouse experiments were performed according to the institutional guidelines of Shanghai Chest Hospital.
For cells in Cultrex® Basement Membrane Extract (BME)-based 3D culture system, firstly, 50 μl/well BME was added into 96-well plate. Subsequently, cell suspension (8000-10,000 cell/200 μl/well) was seeded on the plate coated with BME. Culture medium was changed every 3 days. Spheroid number and size were calculated after 7 days culture.

Study of m 6 A
For dot blot assay, total RNA was isolated with Trizol and mRNA was enriched using Dynabeads mRNA purification Kit (Invitrogen). mRNA was denatured by heating at 72 • C for 10min, followed by chilling on ice immediately. Then, mRNA was spotted on Biodyne Nylon Transfer Membranes (Pall, New York, NY, USA) and cross-linked by UVP for 10 min. The m 6 A level was measured using the m 6 A antibody (Synaptic Systems, #202003, Goettingen, Germany). For global m 6 A level, the quantity was measured by the m 6 A RNA methylation assay kit (Abcam, #ab185912). MeRIP-seq was performed by RiboBio (Guangzhou, China). For luciferase experiments, CDS and partial 3 ′ UTR contain consensus m 6 A motif of SLC7A11 mRNA were cloned into the pmir-GLO plasmids. For mutant reporter vectors, adenosine (A) in the m 6 A motif was replaced by a cytosine (C). The CDS, WT and Mut PCR products of SLC7A11 were synthesized by Generay Biotech Ltd (Shanghai, China). The pmir-GLO plasmids were transiently transfected into the indicated cells, and the luciferase activity was measured 24 h later using a kit from Promega (Madison, WI, USA). The primers were listed in Supplementary Table S1.

Promoter analysis
The human SLC7A11 -2 K promoter was cloned into the pGL4-Basic plasmids. Transfection efficiency was normalized by co-transfection with the Renilla plasmids. The primers were listed in Supplementary Table S1.

RNA and protein interactions
For m 6 A immunoprecipitation qPCR (MeRIP-qPCR), total RNA was extracted. After fragmentation, RNA was incubated with anti-m 6 A antibodies for immunoprecipitation using Magna methylated RNA immune-precipitation (MeRIP) m 6 A kit (Millipore, #17-10499, Billerica, MA, USA). Enrichment of m 6 A containing mRNA was then measured using RT-qPCR assay. For RNA immunoprecipitation (RIP) experiments, a Magna RIP Kit (Millipore, #17-700) was used. Briefly, 10 cm dish cell lysates were incubated with magnetic beads loaded with 5 μg antibodies of YTHDC2 (Abcam, #ab176846) or IgG overnight at 4 • C. The remaining RNA after proteinase K digestion was purified using TRIzol and measured by RT-qPCR. The primers for MeRIP and RIP-qPCR are supplied in Supplementary Table S1. For photoactivatable ribonucleotide crosslinking and immunoprecipitation (PAR-CLIP), control cells and cells stably expressing YTHDC2 WT -HA or YTHDC2 ΔYTH -HA were cultured with 4-SU (250 μM, Sigma) for 16 h, then irradiated with 365 nm UV light for crosslinking. The protein-RNA complex was labeled with biotin using the RNA 3 ′ end biotinylation kit (Thermo, #20160, Waltham, MA, USA). After washing three times with IP wash buffer, beads were resuspended and boiled at 95 • C for 10 min. To detect RNA-protein complexes, the samples were separated by SDS-PAGE and visualized by the chemiluminescent nucleic acid detection module (Thermo, #89880). RNA fragments were extracted by proteinase K digestion of the gel slices followed by ethanol precipitation. And the purified RNA pellets were dissolved in DEPC water and subjected to RT-qPCR detection. For in vitro RNA pull-down assay, synthesized partial SLC7A11 3 ′ UTR probes with or without m 6 A modification at GGAC motif or mutate GGAC motif to CCAG were labeled by biotin (Takara, Dalian, China). About 1 × 10 7 H1299 cells with control, YTHDC2 WT -HA or YTHDC2 ΔYTH -HA groups were washed with cold PBS and lysed in protein lysis buffer. Then incubated with 3 μg biotinylated SLC7A11 3 ′ UTR probes at 4 • C overnight. The biotin-coupled RNA-protein complex was pull-downed with streptavidin magnetic beads (Life Technologies, Carlsbad, CA, USA) for another 4 h. After incubation and washing five times, the streptavidin beads were boiled and used for the IB assay.

Measurement of metabolites
Metabonomics was performed by Luming Biotech Ltd (Shanghai, China). Cystine concentration were measured using an ELISA kit (Lichen Biotech). Glutamate release assay was determined using a glutamate detection kit (Solarbio, #BC1580). GSH/GSSG ratio was determined by GSH and GSSG levels measured by GSH (Solarbio, #BC1175) and GSSG (Solarbio, #BC1180) assay kits. The NAD + /NADH ratio was measured using the NAD + /NADH Assay Kit (Abcam, #ab65348) according to the manufacturer's instructions. For cystine uptake, medium was removed and replaced with cystine-free medium. L-14 C-Cystine (0.2 μCi/mL, PerkinElmer, #NEC854010UC, Shelton, CT, USA) was added and further incubated for 30 min before it was washed and lysed. To quantify 14 C radioactive cystine uptake, cell lysate was added into 1 ml scintillation fluid and radioactive 14 C counts per minute (CPM) were obtained in a liquid scintillation counter (Tri-Carb 2910 TR, PerkinElmer).

Statistical analysis
P values were calculated with Student's t-test between two groups. One-way ANOVA and two-way ANOVA were used to compare multiple groups. To investigate the correlation between two independent groups, the Chi-squared test was used. Survival curves were generated using the Kaplan-Meier method and compared using the log-rank test. Data are presented as means ± SEMs from three independent experiments. *p < 0.05 and **p < 0.01 were considered statistically significant.

Suppression of YTHDC2 is associated with tumor progression in LUAD
To evaluate the expression profile of m 6 A readers in lung cancer, we analyzed known readers, including YTHDF1/2/3, YTHDC1/2, HNRNPA2B1, HNRNPC/G, eIF3A, FMR1 and IGF2BP1/2, in LUAD and lung squamous cell carcinoma (LUSC) via the GEPIA database. Among these, only YTHDC2 was downregulated in both LUAD and LUSC compared to normal lung. However, IGF2BP2 was differentially expressed in LUAD and LUSC ( Fig. 1A and B, Figs. S1A and B). The Oncomine databases ( Fig. 1C and S1D-G) also revealed that YTHDC2 was commonly downregulated in LUAD. Analysis of the Kaplan-Meier plot further demonstrated that lower YTHDC2 correlated with poorer overall survival in LUAD ( Fig. 1D) but not in LUSC (Fig. S1C). These results led us to investigate the role of YTHDC2 in LUAD.
In cohort #1, which consisted of recruited LUAD patients in Shanghai, China, a significant downregulation of YTHDC2 mRNA and protein was observed in tumors compared to adjacent normal tissues Representative IHC images of YTHDC2 in paired adjacent-tumor tissues from cohort#1 TMA, scar bar 100 μm. IHC staining was scored and analyzed, and the percentage for patients with indicated alteration of YTHDC2 is also shown. (J, K) Correlations between YTHDC2 expression and tumor diameter and between YTHDC2 expression and tumor stage. Statistical analysis was performed using Student's t tests (E, F) and Chi-squared test (J, K). Data are means ± SEMs, *p < 0.05, **p < 0.01.

YTHDC2 exhibits antitumor activity via its m 6 A-reading domain
Conditional activation of oncogenic Kras and inactivation of p53 in KP mice by viruses that express Cre in lung epithelial cells results in a rapid development of LUAD [20][21][22]. To explore the m 6 A reader function of YTHDC2 in vivo, we achieved pulmonary overexpression of YTHDC2 with or without the m 6 A-recognizing YTH domain and generated KP-based and GFP-tagged KPY WT , KPY ΔYTH and control KPE mice ( Fig. 2A). In addition to strong GFP signals, YTHDC2 was confirmed to  Data are means ± SEMs, **p < 0.01, N⋅S.: no significant. be upregulated for a sustained period in tumors from KPY WT and KPY-ΔYTH mice compared to those of KPE mice 25 wk post infection (Fig. 2B), indicating long-lasting efficiency of the AAV5 expression system. Although YTHDC2 was unable to stop lung tumorigenesis, the tumor occurrence time was delayed, the overwhelming lung tumor burden was significantly suppressed, and the survival time was longer in KPY WT mice than in KPE mice and KPY ΔYTH mice (Fig. 2C-F).
We further investigated whether YTHDC2 is a tumor suppressor in human LUAD cells. H1975 and H1299 cells were chosen because they exhibited the highest and lowest YTHDC2 expression among the LUAD cell lines tested, respectively (Fig. S1H). IB assays confirmed the overexpression and knockout efficiency of YTHDC2 (Fig. 2G, S2A). While the number and size of the 3D spheroids and the tumor growth of xenografts decreased following YTHDC2 WT overexpression in H1299 cells, these effects were not observed upon YTHDC2 ΔYTH overexpression (Fig. 2H-K). In contrast, knocking out YTHDC2 in H1975 cells increased spheroid formation and xenograft growth in mice (Figs. S2B-E). Notably, when YTHDC2 expression was reconstituted using YTHDC2 sg2− resistant (res) , the positive roles of YTHDC2-sg2 in tumorigenesis were restored (Figs. S2B-E). Thus, YTHDC2 exerts a tumor suppressive function in LUAD via its m 6 A reading domain.
To exclude potential competitive interference with endogenous YTHDC2 function in our overexpression experiments, YTHDC2 activity was reconstituted by FLAG-tagged YTHDC2 sg2− res in YTHDC2-sg2mediated YTHDC2 knockout H1299 cells, and its expression was adjusted to a similar level in NC-sg-treated control cells before further overexpressing HA-tagged YTHDC2 sg2− res or YTHDC2 △YTH (sg2− res) (Fig. S2F). Because YTHDC2 has the ability to interact with the 5 ′ -3 ′ exoribonuclease XRN1 [23], the YTHDC2-XRN1 interaction was evaluated to assess YTHDC2 function in co-IP experiments. By assessing XRN1 levels in immunoprecipitates obtained using anti-FLAG antibodies, we observed that the YTHDC2-XRN1 interactions were not affected by HA-tagged YTHDC2 sg2− res or YTHDC2 △YTH (sg2− res) overexpression (Fig. S2G), indicating that competitive interference with endogenous YTHDC2 be ignored in our overexpression experiments. These results also explained why the significant expression of Ythdc2 in murine tumors (Fig. 2D) and YTHDC2 in H1299 cells (Fig. 2G) did not alter the transformative phenotypes when the mutant was overexpressed (Fig. 2D-K).

YTHDC2 inhibits cystine uptake and the downstream antioxidant program
Metabolic activity is typically observed in tumors with high protumorigenic properties. To investigate the impacts of YTHDC2 on metabolites, a metabolomic analysis was performed comparing LUAD specimens with low and high YTHDC2 expression (referred to as YTHDC2 low and YTHDC2 high , respectively) (n = 20/each group). GSH metabolism was among the top 20 KEGG pathways identified (Fig. 3A). Among the significantly altered metabolites, a 3.975-fold increase in cystine was observed in YTHDC2 low tumors compared to YTHDC2 high tumors (Fig. 3B). Moreover, a negative correlation between YTHDC2 and intracellular cystine was observed in cohort #1 (n = 100) (Fig. 3C). These data indicate that YTHDC2 reduces intracellular cystine.
To verify whether YTHDC2 reduces cystine uptake, we cultured primary patient-derived LUAD cells with either high (pHY#1-8) or low (pLY#1-8) YTHDC2 expression levels (Fig. S3A). Extracellular cystine levels in the culture medium were lower in pLY#1-8 cells than in pHY#1-8 cells, while intracellular cystine levels were higher (Figs. S3B and C). The L-14 C-cystine uptake assay results showed that YTHDC2, through its YTH domain, suppressed cystine uptake in xenografts generated by H1299 and H1975 cells, in LUAD formed in mice and in patient-derived LUAD cells (Fig. 3D-F, S3D). Upregulation of CHAC1 mRNA levels has been shown to indicate impaired cystine uptake [15]. Indeed, in the present, YTHDC2 was observed to increase CHAC1 mRNA levels (Figs. S3G and K). The cystine-glutamate antiporter system Xc − imports cystine with the counter transport of glutamate. The glutamate release assay results showed that YTHDC2 suppresses glutamate release ( Fig. S3H and L). These results suggest that YTHDC2 impairs cystine uptake via the inhibition of system Xc − . System Xc − is crucial for redox balance and metabolism [24,25], and alteration of the GSH/GSSG and NAD + /NADH ratios in H1299 and H1975 cells by the YTH domain of YTHDC2 (Figs. S3I and J, M and N) further suggested that YTHDC2 may impair system Xc − function.
Impaired system Xc − function was associated with decreased GSH levels (illustrated in Fig. 3G and Ref. 15,26). Data from patient-derived LUAD cells and tumors in mice demonstrated that GSH was indeed reduced by YTHDC2 (Figs. S3E and F). To investigate whether downstream cystine antioxidants compensate for the antitumor roles of YTHDC2, mice were administered GSH and NAC, a cysteine precursor. The tumor burden in the lung was significantly higher in KPY WT mice with NAC and GSH administration than in the vehicle-treated control mice, and the levels were similar to that of the KPE mice ( Fig. 3H and I). Interestingly, GSH and NAC administration only resulted in a significant increase in the GSH/GSSG ratio in the tumors from KPY WT mice compared to those from KPE mice (Fig. 3J). A potential reason for this result is that YTHDC2 impairs system Xc − , and we proposed that the impairment of system Xc − function might be critical for the facilitation of GSH and its raw materials other than cystine, such as cysteine or its precursor to enter into cell. This potential mechanism is consistent with the findings of a study describing that supplementation of GSH or NAC could effectively reverse the effects of cystine deprivation by cystine starvation or inhibiting system Xc − [27].
As reported in prior studies [28,29] and shown in Fig. 3G, the lack of GSH leads to an increase in lipid peroxidation. Therefore, a fluorescent probe (C11-BODIPY 581/591 ) was used to measure oxidized lipids in mouse LUAD. To specifically avoid GFP interference, we constructed KPE, KPY WT and KPY ΔYTH mice without a GFP tag. As shown in Fig. S3O, the fraction of oxidized C11-BODIPY 581/591 (green) was largely increased in the tumors of KPY WT mice compared to those of KPE mice; however, these levels were similar in the tumors of KPE and KPY ΔYTH mice. Moreover, 4-HNE and MDA levels, which are products of lipid peroxidation, were increased in the tumors of KPY WT but not in KPY ΔYTH mice compared to KPE mice (Fig. S3P). Notably, NAC and GSH supplementation reduced lipid peroxidation and shortened survival in KPY WT mice ( Fig. 3K and L). Overall, these results indicate the m 6 A reading function of YTHDC2 is essential to reduce the cystine downstream antioxidant program.
Next, we investigated whether YTHDC2 regulates SLC7A11. The protein and mRNA levels of SLC7A11 were both negatively regulated by YTHDC2 in H1299 and H1975 cells ( Fig. 4A and B, S4D and E). In mice, Slc7a11 was downregulated in tumors from KPY WT mice compared to those from KPE and KPY ΔYTH mice ( Fig. 4C and D). In addition, the YTH domain is a prerequisite for YTHDC2 to suppress SLC7A11 expression (Fig. 4A-D).
The clinical relevance of YTHDC2 and SLC7A11 in cohort #1 was also investigated. YTHDC2 was downregulated, whereas SLC7A11 was upregulated in tumors compared to adjacent tissues ( Fig. 4K and L, S4K). In addition, they were negatively correlated in tumors (Fig. 4M).

YTHDC2 accelerates SLC7A11 mRNA decay
The mechanisms underlying the regulation of SLC7A11 expression through YTHDC2 were further investigated. Because both the mRNA and protein levels of SLC7A11 are negatively regulated by YTHDC2 ( Fig. 4 and S4), we wanted to determine whether the mRNA and protein levels of SLC7A11 were both suppressed by YTHDC2. To address this, we first investigated whether YTHDC2 regulates SLC7A11 mRNA. mRNA homeostasis is maintained by the "production-degradation" balance [33,34]. The YTH domain played no role in suppressing SLC7A11 promoter activity when YTHDC2 was overexpressed in H1299 cells (Fig. S5A). Knocking out YTHDC2 followed by reconstituting its expression in H1975 cells also voted that YTHDC2 regulates SLC7A11 transcription (Fig. S5B). Then, we treated H1299 and H1975 cells with ActD, a pan transcription inhibitor, to examine whether YTHDC2 regulates SLC7A11 mRNA stability after transcription was blocked and found that the YTH domain is essential for YTHDC2 to shorten the half-life of SLC7A11 mRNA in H1299 cells (Fig. 5A). Loss of function of YTHDC2 by CRISPR/Cas9 technology followed by YTHDC2 reconstitution in H1975 cells further confirmed that YTHDC2 accelerates SLC7A11 mRNA decay (Fig. 5B). XRN2 and EXOSC10 are two major exonucleases responsible for 5 ′ -3 ′ and 3 ′ -5 ′ exonucleolytic activity, respectively [35,36]. Deletion of EXOSC10, but not XRN2, diminished the effects of YTHDC2 on shortening the SLC7A11 mRNA half-life (Fig. 5C, S5C and D), suggesting that YTHDC2 facilitates 3 ′ -5 ′ SLC7A11 mRNA degradation. At the pharmacological level, treating H1299 cells with DEPC and RNasin, two pan RNase inhibitors, also blocked YTHDC2 function (Fig. 5D). These findings establish a role for YTHDC2 in destabilizing SLC7A11 mRNA in LUAD cells.
Subsequently, we investigated whether the protein level of SLC7A11 is simultaneously suppressed by YTHDC2 independent of its mRNA. To investigate whether YTHDC2 affects SLC7A11 translation, the CDS region of SLC7A11 mRNA was cloned downstream of the firefly luciferaseencoding region of the pmirGLO plasmid. Unfortunately, the ratios between firefly luciferase and the internal Renilla luciferase were still comparable regardless of whether YTHDC2 was overexpressed or knocked out (Figs. S5E and F), suggesting that YTHDC2 is unable to regulate SLC7A11 translation. To explore whether YTHDC2 promotes SLC7A11 protein degradation, we blocked de novo generation of SLC7A11 protein by treating the indicated H1299 and H1975 cells with CHX, a protein translation inhibitor. The results showed that YTHDC2 was also unable to affect the half-life of the remaining SLC7A11 protein (Figs. S5G and H). Overall, the suppression of SLC7A11 protein by YTHDC2 might be the result of prior suppression of its mRNA.

SLC7A11 mRNA is suppressed by YTHDC2 in an m 6 A-dependent manner
YTHDC2 regulates SLC7A11 mRNA stability (Fig. 5A-D), and control of RNA decay is one of the important functions of m 6 A methylation [4,8,37]. However, m 6 A methylation of target transcripts is a prerequisite for YTHDC2 to control mRNA decay [8]. Whether the function of YTHDC2 in suppressing SLC7A11 relies on m 6 A methylation is still unknown. Among the three major writers, i.e., METTL3, METTL14, and WTAP, only a higher METTL3 expression level was found to correlate with a higher global m 6 A methylation level in H1975 cells compared to H1299 cells (Figs. S5I-K). The capacity of METTL3 to stimulate global m 6 A was also verified in H1299 and H1975 cells (Fig. 5E and F and S5L). Knocking down METTL3 in H1975 cells prolonged the half-life of SLC7A11 mRNA, whereas overexpressing METTL3 in H1299 cells accelerated SLC7A11 mRNA degradation (Fig. 5G). Furthermore, decreased METTL3 expression in H1299 cells blocked YTHDC2 to accelerate SLC7A11 mRNA decay, impair cystine uptake and stimulate lipid reactive oxygen species (ROS) generation (Fig. 5H-J), suggesting that the role of YTHDC2 in LUAD cells is m 6 A-dependent.

SLC7A11 mRNA is m 6 A methylated
To reveal the potential m 6 A methylation sites in SLC7A11 mRNA, MeRIP-seq was performed before and after METTL3 knockdown in H1975 cells. The GGAC motif was highly enriched within m 6 A sites in H1975 cells with or without knocking down METTL3 (Fig. 6A). m 6 A peaks were especially abundant in the 3 ′ untranslated region (3 ′ UTR) just near the stop codons of mRNAs (Fig. 6B), and SLC7A11 mRNA was no exception (Fig. S6A). MeRIP-seq revealed a putative m 6 A motif located within the 3 ′ UTR (+1793~+1796 relative to the transcription start site (TSS) of the SLC7A11 gene) (Fig. S6A). The m 6 A enrichment around this site was correlated with the m 6 A methylation level ( Fig. 5E  and F, Fig. S6A). To further confirm the m 6 A-dependent modification of SLC7A11 mRNA, MeRIP-qPCR was performed. m 6 A modification in SLC7A11 mRNA was significantly reduced specifically around the putative m 6 A site after METTL3 was knocked down in H1975 cells. In contrast, the opposite outcome was observed when METTL3 was overexpressed in H1299 cells (Fig. 6C). Thus, we provided evidence that SLC7A11 mRNA can be m 6 A methylated in LUAD cells.

YTHDC2 preferentially binds to m 6 A-methylated SLC7A11 mRNA
Although YTHDC2 functions as an m 6 A reader, whether YTHDC2 preferentially binds to and recognizes m 6 A-methylated SLC7A11 mRNA Fig. 6. YTHDC2 preferred to interact and destabilize m 6 A methylated SLC7A11 mRNA. (A) Predominant consensus GGAC motif was showed in H1975 cells with or without METTL3 knocked down from MeRIP-seq. (B) Density distribution of m 6 A peaks across mRNA transcripts, the percentages of m 6 A peaks were determined. (C) MeRIP-qPCR analysis of m 6 A levels at the indicated sites within SLC7A11 mRNA in control cells and H1975 cells with METTL3 knockdown, and H1299 cells with METTL3 overexpressed. (D) PAR-CLIP assay of RNA pulled down by HA-tagged YTHDC2 WT or YTHDC2 ΔYTH in H1299 cells with or without overexpressing METTL3. RNA labeled with biotin at 3 ′ end was visualized by the chemiluminescent nucleic acid detection module. SLC7A11 mRNA levels in the pulled down products were also verified by RT-qPCR. (E) In vitro partial SLC7A11 3 ′ UTR probe pulldown assay in H1299 control cells, HA tagged YTHDC2 WT and YTHDC2 ΔYTH overexpression cells. GGYC (Y = m 6 A or A), CCYG (Y = m 6 A or A). (F, G) RIP-qPCR analysis showing the binding between YTHDC2 and SLC7A11 mRNA in H1975 and H1299 cells under indicated treatment. (H-J) Schematic generation strategy for the pmir-Glo luciferase reporters containing WT and Mut (GGAC to GGCC) SLC7A11 3 ′ -UTR. Luciferase activities from the indicated pmir-Glo vector were measured in H1299 and H1975 cells with or without YTHDC2 overexpression or knockout. (K) Schematic presentation of the primer sets designed to detect WT or Mut F-Luc-SLC7A11 fusion mRNA. (L, M) The decay curve of exogenous F-Luc-SLC7A11 fusion mRNA in indicated groups. The F-Luc-SLC7A11 fusion mRNA level were normalized to that of internal Renilla luciferase mRNA. Statistical analysis was performed using Student's t tests (C, D, F, G), one-way ANOVA (I, J) and two-way ANOVA (L, M). Data are means ± SEMs, **p < 0.01, N⋅S.: no significant. is unknown. To this end, a PAR-CLIP experiment to assess whether the SLC7A11 mRNA-YTHDC2 interaction depends on m 6 A methylation was first performed. The results demonstrated that increasing the m 6 A methylation level by overexpressing METTL3 in H1299 cells facilitated YTHDC2 binding with SLC7A11 mRNA (Fig. 6D). In contrast, the SLC7A11 mRNA-YTHDC2 interaction was remarkably suppressed once METTL3 was knocked down in H1975 cells (Fig. S6B). Regardless of whether METTL3 was overexpressed or knocked down, deletion of the YTH domain blocked the SLC7A11 mRNA-YTHDC2 interaction ( Fig. 6D  and S6B). Next, we performed an in vitro RNA pulldown assay using synthesized partial SLC7A11 3 ′ UTR probes and incubated them with lysates from H1299 cells expressing HA-tagged YTHDC2 WT and YTHDC2 ΔYTH . YTHDC2 WT , but not YTHDC2 ΔYTH , bound preferentially to the 3 ′ UTR of SLC7A11 mRNA containing an m 6 A compared to the one with an unmethylated adenosine (Fig. 6E). Additionally, YTHDC2 is prone to bind the m 6 A consensus motif GGAC, which is the same as that located within the putative m 6 A site, because a robust reduction in the SLC7A11 3 ′ UTR-YTHDC2 interaction was observed when GGAC was replaced by a random CCAG motif (Fig. 6E). Finally, a RIP-qPCR assay was performed to evaluate whether m 6 A methylation modulates the intracellular SLC7A11 mRNA-YTHDC2 interaction. Indeed, higher global m 6 A methylation caused a stronger SLC7A11 mRNA-YTHDC2 interaction in H1975 cells than in H1299 cells ( Fig. S5J and K, S6C). From the gain-and loss-of-function experiments of METTL3 in H1299 and H1975 cells, we also concluded that the intracellular SLC7A11 mRNA-YTHDC2 interaction is determined by the m 6 A level ( Fig. 6F and G). These data suggest that YTHDC2 preferentially binds to m 6 A-methylated SLC7A11 mRNA.

YTHDC2 destabilizes SLC7A11 mRNA in an m 6 A-dependent manner
Then, we examined whether YTHDC2 destabilizes SLC7A11 mRNA via the putative m 6 A site. We cloned the wild-type (WT) SLC7A11 3 ′ UTR (+1724 ~ +1911 relative to the TSS) and mutant (Mut) 3 ′ UTR (GGAC to GGCC) downstream of the firefly luciferase (F-Luc) encoding region in the pmirGLO plasmid (Fig. 6H). In this experiment, firefly luciferase activities correlated with mRNA stability. The results suggested that only overexpression of YTHDC2 WT but not YTHDC2 ΔYTH reduced SLC7A11 mRNA stability in H1299 cells and that knocking out YTHDC2 could increase its stability in H1975 cells. However, compared to the those of the WT 3 ′ UTR, these effects were diminished for the Mut reporter ( Fig. 6I and J). Similar results were obtained after examining the RNA stability of the exogenous F-Luc-SLC7A11 fusion mRNA expressed by the pmirGLO plasmid ( Fig. 6K-M). Overall, the m 6 A site at the 3 ′ UTR is essential for YTHDC2 to destabilize SLC7A11 mRNA.

LUAD with YTHDC2 suppression is sensitive to system X C − -targeting therapy
To evaluate the importance of YTHDC2 suppression in LUAD, twenty YTHDC2 low LUADs were randomly selected from cohort #1, and 50% of them belonged to the acinar subtype (Fig. 7A). To understand the prevalence of YTHDC2 low in the acinar subtype, another cohort #2 was recruited specific for the acinar subtype. In cohort #2 (n = 100), YTHDC2 was downregulated in the tumor compared to the adjacent tissues, and 62% (62/100) of acinar tissue were classified as YTHDC2 low ( Fig. 7B and C). Overall survival was much shorter among acinar LUAD patients with YTHDC2 low compared to those without this expression pattern (Fig. 7D), further demonstrating that YTHDC2 suppression might be especially important for tumor progression of acinar LUAD.
YTHDC2 suppresses SLC7A11 (Fig. 4, S4); therefore, we speculated that SLC7A11 is simultaneously upregulated in acinar LUAD. Indeed, this is true in cohort #2 (Fig. 7E and F). Recently, utilizing ystem X C − inhibitors, such as erastin and its derivatives, has been proven to be an attractive strategy to treat cancer [16,38]. We wondered whether piperazine erastin (PKE), a stable in vivo derivative of erastin, has a potential therapeutic effect on acinar LUADs. In addition, sorafenib, an FDA-approved multikinase inhibitor and a potent system X C − inhibitor, was examined in parallel. The PDX mouse model is valuable to evaluate treatment sensitivity in patients [39]. Hence, PDX models that originated from 4 distinct acinar LUAD patients were administered PKE and sorafenib. No significant changes in histopathology and expression of YTHDC2 and SLC7A11 were observed due to mouse passage (Fig. S7A). Compared to that in the DMSO-treated control, significant tumor regression was observed when PKE and sorafenib were administered ( Fig. 7G-J, S7B). In addition, administrating mice with PKE and sorafenib also caused a significant increase in MDA and 4-HNE in tumors ( Fig. 7K and L), suggesting that induced lipid peroxidation might be the result of inhibiting tumorigenesis in acinar LUADs. Finally, we recruited another two cohorts to verify our conclusion. While cohort #3 LUAD patients were recruited from North China, cohort #4 was also recruited from Shanghai, China, but during a time frame distinct from that of cohort #1. Similar to the observations for cohort #1, MDA and YTHDC2 were suppressed, while both the mRNA and protein levels of SLC7A11 were elevated in the tumor compared to the adjacent tissues in cohort #3 (n = 30) (Fig. 7M), confirming that suppression of YTHDC2 in LUAD prevents lipid peroxidation by upregulating SLC7A11. In cohort #4 (n = 20/each stage), MDA and YTHDC2 were negatively associated, while SLC7A11 was positively associated with the stage (Fig. 7N), demonstrating that SLC7A11 elevation might be the key for YTHDC2 suppression to promote LUAD progression.

Discussion
Recently, as a hot topic in the field of epigenetic regulation, RNA m 6 A modification has been shown to participate in multiple cellular processes, such as the maturation and degradation of mRNA and protein translation [7,40]. The functions of m 6 A modification have been studied in multiple cancers, such as leukemia, liver, colon and breast cancer [6,[41][42][43]. High levels of the m 6 A writer METTL3 significantly promote liver and breast cancer progression [44,45]. By contrast, low levels of the m 6 A erasers FTO and ALKBH5 are linked with poor renal carcinoma prognosis [46]. In addition, m 6 A readers such as the YTH and IGF2BPs families play contradictory roles and act as either tumor promoters or suppressors [7,47,48]. Nevertheless, these studies highlight that the interplay in the WER system is essential to regulate cancer initiation and progression. However, how m 6 A works through its readers and what their targets are in LUAD remain unclear. Thus, from the perspective of the reader, the present study demonstrates that YTHDC2 has antitumor activity in LUAD and that SLC7A11 mRNA is its direct target.
Our current understanding of the YTH proteins, which consist of YTHDF1, YTHDF2, YTHDF3 and YTHDC1, suggests that this m 6 A reader family is critical for tumorigenesis [7,[48][49][50]. However, the role of YTHDC2 remains unclear. To address this, we investigated the function of YTHDC2 in LUAD tumorigenesis. Our results demonstrated that YTHDC2 suppression is frequent and is associated with tumor progression in LUAD. Furthermore, we showed that YTHDC2 impairs system X C − activity and cystine uptake through downregulation of SLC7A11. m 6 A methylation by METTL3 is a prerequisite for YTHDC2 to accelerate SLC7A11 mRNA decay. Hence, upregulation of SLC7A11 in LUAD is a result of YTHDC2 suppression. Because cystine is indispensable for the downstream antioxidant program [26,51], suppression of YTHDC2 thus promotes tumorigenesis by protecting LUAD cells from oxidative damage.
Metabolic active is one of the major characteristics that distinguishes tumor cells from normal cells [52,53]. Although ROS are physiologically important, excessive ROS are harmful for tumor cells [54]. Accumulated lipid peroxidation is one such oxidative damage [55]. The antioxidant system, including the system X C − /GSH axis, thus far has shown extreme importance to prevent excessive lipid peroxidation [26,56]. The activity of system X C − is often positively correlated with the expression of its catalytic subunit SLC7A11 [15,57]. SLC7A11 can be transcriptionally regulated by a series of transcription factors [18,58,59]. It can also be regulated at the protein level [16,17,60]. Our results further showed that YTHDC2 is associated with lipid peroxidation and that it regulates SLC7A11 mRNA and protein expression in a manner dependent on posttranscriptional regulation, in which SLC7A11 mRNA is destabilized in an m 6 A-and YTHDC2-dependent manner. Although prior studies have demonstrated that upregulation of SLC7A11 mRNA is a negative feedback to system X C − inhibition [14,19], overexpression of YTHDC2 only causes a downregulation of SLC7A11 mRNA. This does not necessarily mean that system X C − is not inhibited because SLC7A11 mRNA has a short half-life of less than 4 h; therefore, according to our model, YTHDC2-mediated reduction of system X C − activity might not be a slow process. Acinar subtype, the most common and heterogeneous type of LUAD, can be further divided into prognostically significant subsets [61,62]. Here, we suggest that the acinar subtype can be further classified according to YTHDC2 expression because acinar with low YTHDC2 expression is associated with poor survival. Fortunately, acinar LUAD is still sensitive to the system X C − inhibitors PKE and sorafenib. This might be because SLC7A11 is the potential target of both agents [16,63]. In other words, elevated SLC7A11 in acinar LUAD somehow increases the opportunity for PKE and sorafenib to exert their antitumor effects. Abnormal activation of oncoproteins often occurs in tumors. The application of small molecular inhibitors targeting these proteins in clinical oncology has shown great potential to improve patient outcomes [64][65][66]. Our preclinical model thus provides another example that molecularly selected patients might benefit from small molecule targeted therapy.
In conclusion, this study emphasizes the importance of the m 6 A reader YTHDC2 in LUAD. System X C − activity is induced to promote tumorigenesis possibly by suppressing YTHDC2 (Fig. 8). Furthermore, our results suggest that molecular subtyping based on YTHDC2 expression might be helpful to predict prognosis in LUAD. Inhibitors targeting system X C − might be helpful to treat LUAD patients with poor prognosis.
Thus, this study provides valuable insights regarding tumorigenesis, molecular subtyping, and drug sensitivity in LUAD.

Data availability
The high-throughput MeRIP sequencing data have been deposited in the Gene Expression Omnibus (GEO) database under the accession number GSE147171. All data supporting the findings of this study are included in the manuscript and are available from the corresponding author upon request.

Author contributions
JY Wang, YC Yu and LT Du conceived of the study and JY Wang carried out its design. LF Ma, TX Chen and X Zhang performed the experiments. YY Miao, XT Tian and KK Yu collected clinical samples. LF Ma, X Xu, YJ Niu, SS Guo, CC Zhang, SY Qiu, YX Qiao and WT Fang analyzed the data. LF Ma and JY Wang wrote the paper. YC Yu and LT Du revised the paper. All authors read and approved the final manuscript.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. The diagram reveals an m 6 A-dependent YTHDC2 regulation of SLC7A11 mRNA to modulate cystine uptake in LUAD cells. In our model, YTHDC2 destabilizes SLC7A11 mRNA via its m 6 A-reading YTH domain. METTL3-mediated m 6 A methylation of SLC7A11 mRNA at its 3 ′ UTR is prerequisite for this process. YTHDC2 is frequently suppressed in LUAD; thereby its anti-tumor activity to impair cystine uptake and downstream antioxidant program is blocked. This model may explain a possible mechanism underlying tumorigenesis in LUAD.