Understanding m6A changes in chromophobe renal cell carcinoma and predicting patient outcomes survival

Abstract

N6-methyladenosine (m⁶A) is a prevalent mRNA modification known for its implications in various cancer types, yet its role in chromophobe renal cell carcinoma (chRCC) remains largely unexplored. In this study, we performed m6A-SEAL-seq and RNA-seq analyses on tissues from three chRCC subjects, aiming to uncover m⁶A alterations in chRCC. Our findings revealed reduced expression levels of four m⁶A regulators in chRCC tissues and highlighted differences in m⁶A levels compared to normal tissues. Furthermore, we identified specific genes and cancer-related pathways affected by these differences, including notable candidates like NOTCH1 and FGFR1, implicated in chRCC development. Additionally, we developed a predictive model based on the expression level of m⁶A associated genes, demonstrating promising prognostic capabilities for patient survival prediction. Overall, our study provides valuable insights into the role of m⁶A in chRCC and its potential as a prognostic indicator.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-024-12956-6.

Introduction

According to the Global Cancer Observatory (GCO, https://gco.iarc.fr/en) database as of 2022, kidney cancer accounts for over 434 thousand new cases globally, with more than 155 thousand deaths attributed to the disease. Despite advancements in healthcare, kidney cancer remains a significant public health concern, ranking 16th in terms of incidence and 17th in terms of cancer-related mortality worldwide. The urgent need to decipher the molecular mechanisms underlying kidney cancer progression persists.

There are three primary types of kidney cancer: clear cell renal cell carcinoma (ccRCC), papillary renal cell carcinoma (PRCC), and chromophobe renal cell carcinoma (chRCC). Among these, ccRCC is the most prevalent form of kidney cancer. Significant advancements have been made in understanding the pathogenesis and therapeutic approaches for ccRCC [1–4]. Papillary renal cell carcinoma (PRCC) stands as the second most common type of renal carcinoma and is currently a well-studied subtype of renal cell carcinoma (RCC) [5, 6]. Whereas chRCC is an uncommon renal cell carcinoma (RCC) subtype, accounting for 5–10% of all cases of RCC [7]. Previous studies suggest that chRCC is a malignant neoplasm with a mortality rate of about 10%, and aggressive clinical course such as metastasis [8]. While the somatic genomic landscape of chRCC has revealed various genomic features [9], the understanding of the functional role of epitranscriptomics remains limited.

More than 170 types of modifications have been identified in the RNA molecule [10]. N⁶-methyladenine (m⁶A) emerges as the most abundant mRNA modification in eukaryotes, playing pivotal roles in various biological functions. These functions include the regulation of RNA stability [11–13], 3′-end processing [14, 15], alternative splicing [16, 17] and translation efficiency at the post-transcriptional level [13, 18, 19]. m⁶A is dynamically reversible in mammals and is governed by “writers” (methyltransferases), “erasers” (demethylases), and “readers” (binding proteins). The majority of m⁶A modifications are installed by the methyltransferase complex, primarily composed of the METTL3-METTL14 heterodimer [20, 21] and other subunits like WTAP [22–24], and removed by demethylases like FTO and ALKBH5 [25, 26]. m⁶A modification can also be recognized and bound by m⁶A binding proteins for the regulation of gene expression, like YTH domain family proteins [27–29]. The discovery of these diverse m⁶A regulators has significantly contributed to a deeper understanding of the physiological functions of m⁶A.

Numerous studies have underscored the close relationship between m6A modifications and tumor progression [30–33]. In pancreatic cancer, ALKBH5 inhibits cancer cell growth and progression by increasing PER1 mRNA levels through m⁶A removal, consequently inhibiting YTHDF2-dependent mRNA degradation [33]. In bladder cancer, METTL3 installs m⁶A on pri-miR221/222, expediting miRNA maturation and promoting cancer cell proliferation [34]. These findings highlight the pivotal roles of m⁶A modification in carcinogenesis through the regulation of RNA processing and metabolism, shedding light on new molecular mechanisms underlying cancer progression. m⁶A regulators exert significant impacts on clear cell renal cell carcinoma (ccRCC) [35]. METTL14, for instance, is downregulated in ccRCC tissue, and patients with lower METTL14 expression tend to exhibit worse prognoses [36]. Alterations in m⁶A regulators are associated with worse clinical characteristics [37]. In RPCC, a prognostic risk signature model incorporating three m⁶A regulatory genes, IGF2BP3, KIAA1429, and HNRNPC, accurately predicts survival outcomes [38]. The advent of high-throughput sequencing has enabled transcriptome-wide profiling of m⁶A distribution in various human carcinomas, offering insights into the molecular mechanisms underlying m⁶A modification and renal cell carcinoma (RCC). Transcriptome-wide m⁶A mapping in ccRCC has revealed the identification of unique m⁶A-related genes associated with cancer-related pathways and provided insights into potential mechanisms of m⁶A-mediated gene regulation [39]. However, the transcriptome-wide distribution of m⁶A in chRCC remains elusive. In this study, we utilized m6A-SEAL-seq to construct a landscape of m⁶A profiling in human chRCC [40], and identified novel m⁶A gene signature.

Methods

Patients and specimens

A total of three patients with chRCC were involved in our study. chRCC tissues and corresponding tumor-adjacent normal tissues were collected at the time of surgery from urology department, Peking University Third Hospital. All specimens were immediately separated into 1.5 ml RNase-free centrifuge tubes and stored at -80 ^◦C before RNA isolation. The study was approved by The Beijing Haidian Hospital Medical Ethics Committee, and the written informed consents were obtained from all the participants.

RNA preparation

Total RNA was extracted from tissue specimens using TRIzol reagent (Magen) and poly(A)⁺ RNA was isolated from total RNA using oligo(dT) 25 Dynabeads (Thermo Fisher Scientific). RNA concentration was determined using a Nanodrop ultraviolet-visible light spectrophotometer (Thermo).

qRT–PCR

Poly(A)⁺ RNA was isolated from tissues using TRIzol reagent (Invitrogen) and oligo(dT)₂₅ Dynabeads (Thermo Fisher Scientific). First-strand cDNA synthesis was carried out using SuperScript III (Thermo Fisher Scientific) and oligo(dT)₂₀ primer. Each sampling was performed with at least three biological replicates. The relative expression levels were normalized using GADPH as the reference gene. All the primers were designed by multiPrime at https://multiprime.cn [41]. All primer sequences are provided in Supplementary Dataset 1.

m⁶A-SEAL-seq and library construction

Poly(A)⁺ RNA for each sample was fragmented by a magnesium RNA fragmentation module (NEB). In a poly(A)⁺ RNA m⁶A oxidation assay, the reaction was performed in 300 µl aliquots of aqueous solution containing 300 µM of (NH₄)₂Fe(SO₄)₂·6H₂O, 2 mM of l-ascorbic acid, 300 µM of α-KG, 100 mM pH 7.0 HEPES, 0.2 µM FTO and 1 µg poly(A)⁺ RNA. After the FTO treatment at 37 °C for 5 min, RNA was purified by Oligo Clean & Concentrator column (Zymo Research). hm⁶A-modified RNA converted from m⁶A by FTO oxidation was treated by 200 mM freshly prepared DTT at 37 °C for 3 h in acidic aqueous solution (100 mM HEPES, pH 4.0). The product RNA was purified by ethanol precipitation. After ethanol precipitation, DTT treated RNA was washed by 75% ethanol and dissolved in 200 µl biotinylation buffer that contained 100 µM of MTSEA-XX-biotin (Biotum), 100 mM HEPES (pH 7.0), 1 mM EDTA and 20% DMF. The reaction was performed at 25 °C and 800 rpm in a ThermoMixer for 1 h. The product RNA was purified by phenol-chloroform extraction. 50 ng biotin-labeled RNA was saved as input, and the rest proceeded to affinity enrichment. After that, 20 µl Dynabeads MyOne Streptavidin C1 (Invitrogen) was washed twice by 200 µl 0.1 M NaOH to remove RNase contamination, and then washed with diethyl pyrocarbonate water to a neutral pH. The beads were resuspended in 100 µl binding solution containing 10 µl of high-salt wash buffer (100 mM Tris pH 7.5, 10 mM EDTA, 1 M NaCl, 0.05% Tween 20) and 90 µl diethyl pyrocarbonate water, and incubated with the biotinylated RNA for 1 h. The beads with biotinylated RNA were washed three times with 1 ml high salt wash buffer. 50 µl of 100 mM DTT was used to release the biotinylated RNA at 37 °C for 15 min on a ThermoMixer (800 rpm.). After collecting the supernatant, the second elution was performed with 50 µl of 100 mM DTT at 50 °C for 5 min to completely release the RNA. The twice-eluted RNA was combined and purified by ethanol precipitation. Library construction was performed using NEBNext Ultra II Directional RNA Library Prep Kit for Illumina according to the manufacturer’s protocol. Libraries were sequenced on the Illumina HiSeq XTen platform with a paired-end model (PE150).

Analysis of m⁶A-seq data

Sequencing reads were trimmed and mapped to the reference genome (GRCh38) by using Cutadapt (v1.18) [42] and HISAT2 (v2.1.0) [43], respectively. The m⁶A-enriched regions in chRCC and normal tissues were identified using the MACS2 [44] peak-calling algorithm based on enrichment criteria (IP/Input) ≥ 2 and FDR < 0.05. Confident m⁶A peaks were subjected to Hypergeometric Optimization of Motif EnRichment tools (HOMER) [45] for Motif Discovery. Genes with differentially methylated m⁶A sites were identified by MeTDiff [46] based on enrichment criteria fold change ≥ 2 and FDR < 0.05. Tissue analysis, Gene ontology (GO) and pathway enrichment analyses were performed by using DAVID.

Analysis of RNA-seq data

Adapter and low-quality reads were trimmed by using Cutadapt (v1.18) [42], and trimmed reads were aligned to the reference genome (GRCh38) using HISAT2 (v2.1.0) [43]. The differential expression genes between chRCC and adjacent normal tissues were screened by R package (DEseq2) [47] based on a cutoff criterion of fold change ≥ 2 and FDR < 0.05.

Risk stratification and survival analysis

A cohort of 65 chRCC cases from The Cancer Genome Atlas (TCGA) database was used to illustrate the relationship between the differential expressed DMMGs and chRCC patients. We randomly chose 40 samples from 65 cases as a training set to predict signature model and the rest samples form a testing set to verify the model (make sure the training set and testing set both contain tumor samples and normal samples). Firstly, we used least absolute shrinkage and selection operator (LASSO) to select candidate genes (glmnet package) in training set. Secondly, we performed the multi-variates cox regression and removed genes not supported by PH hypothesis using the selected candidate genes in training set. Thirdly, we performed the second regression (survival package) to calculate the coefficients between candidate genes and 5-years survival using the remaining candidate genes in training set to build a Cox model. The concordance index (C-index) was calculated to evaluate the prognostic power. Risk score of each sample was calculated through the sum of the product of each candidate gene fpkm-uq and its coefficient in training set. The patients were then classified into high-risk or low-risk group using the risk score where the difference value of true positive and false positive reaches to the maximum as the cutoff value. The Kaplan-Meier survival curve (survminer package) was performed to evaluate the 5-years survival, and the sensitivity and accuracy of the cox model to predict clinical outcome were evaluated by the area under curve (AUC) of the receiver operating characteristic (ROC) curve (survival ROC package). At last, we test the signature model in the testing set, ccRCC dataset (a cohort of 602 cases from TCGA database) and PRCC dataset (a cohort of 318 cases from TCGA database).

Results

The expression pattern of m⁶A regulators in normal and chRCC tissue

In order to determine whether m⁶A modification functions in chRCC, we first analyzed the expression levels of 8 m⁶A regulators, including 3 key writer subunits, 3 readers, and 2 erasers (m⁶A writer subunits: METTLE14, METTL3, and WTAP; m⁶A readers: YTHDF1, YTHDF2, and YTHDF3; m⁶A erasers: ALKBH5 and FTO) in three patients. The qPCR results showed that the expression levels of WTAP, YTHDF2, FTO, and ALKBH5 were downregulated markedly in chRCC tissues compared with corresponding tumor-adjacent normal tissues (termed normal tissues) (Fig. 1A-H). Furthermore, upon comparison with The Cancer Genome Atlas (TCGA database), we observed a notable decrease in the expression of the m⁶A writer WTAP as well (Fig. S1). The aberrant expression of the m⁶A regulators in chRCC, suggesting m⁶A might be dysregulated in chRCC.

Fig. 1.

Relative expression level of known m⁶A-related genes in Normal and chRCC tissues by qPCR. A, METLL3, B, METTL14, C, WTP, D, YTHDF1, E, YTHDF2, F, YTHDF3, G, FTO, H, ALKBH5. Data are presented as means ± SE, n = 3 biological replicates. *P < 0.05 by t test (two-sided)

Overview of m⁶A methylation feature in normal and chRCC tissues

To investigate whether m⁶A methylation landscape changes between the normal and chRCC tissues, we next constructed m⁶A-SEAL-seq using three samples and explored the distribution of m⁶A modification [40]. Approximately 87.1–13.2 million reads were generated from each library and 82.8–12.8 million reads were mapped to GRCh38 genome (Supplementary Dataset 2). m⁶A peaks were called in each sample using MACS2, based on fold enrichment (IP/input) ≥ 2 and FDR ≤ 0.05). The m⁶A peaks identified in all three replicates were classified as “confident m⁶A peaks”. Finally, we identified 15,024 confident m⁶A peaks corresponding to 11,396 transcripts/genes in normal tissues, whereas 12,841 confident m⁶A peaks corresponding to 10,102 transcripts/genes in chRCC tissue (Fig. 2A). To evaluate the reliability and performance of m⁶A-SEAL-seq, we compared our confident m⁶A peaks of normal tissues with MeRIP-seq of normal kidney tissues from published sequencing data (GSE122744) [48]. 5686 out of 7261 (78.3%) of the m⁶A peaks in MeRIP-seq were identified as confident m⁶A peaks in our m⁶A-SEAL-seq (Fig. S2A), and the metagene profile results of m⁶A-SEAL-seq and MeRIP-seq indicated that that our confident m⁶A peaks were highly enriched around MeRIP-seq peaks (Fig. S2B). We additionally computed normalized read coverages from m⁶A-SEAL-seq around MeRIP-seq peaks, revealing their co-enrichment (Fig. S2C, D). These findings suggest that m⁶A-SEAL-seq provides accurate and reliable results.

Fig. 2.

Characterization of m⁶ A modification in Normal and chRCC tissues reveals that decreased methylation number in chRCC Tissues. A, overlap of three biological replicates of m⁶A-SEAL-seq peaks identifying ~ 15,000 high-confident m⁶A peaks corresponding to 11,396 unique transcripts in normal tissues (left); Overlap of three biological replicates of m⁶A-seal peaks identifying ~ 12,800 high-confident m⁶A peaks corresponding to 10,102 unique transcripts in chRCC tissues (right). B, Metagene profile illustrating the region distribution of m⁶A peaks across the indicated mRNA segments. C, Pie chart depicting the fraction of the confident m⁶A peaks in each of the five non-overlapping transcript segments (5’UTR, start codon, coding sequence [CDS], stop codon and 3’UTR) in normal tissues (left); Pie chart depicting RNA types of m⁶A peaks in normal tissues (right). D, Pie chart presenting the fraction of the confident m⁶A peaks in eCRCC tissues (left); Pie chart depicting RNA types of m⁶A peaks in chRCC tissues (right). E, The number of m⁶A peaks in human chromosomes. F, The number of m⁶A peaks per gene. G, The number of m⁶A peaks in five non-overlapping transcript segments

The metagene profiles results demonstrated that confident m⁶A peaks in normal and chRCC tissue were both highly located within Coding Sequence (CDS) (Fig. 2B). However, the distribution pattern of m⁶A peaks in chRCC shifted from CDS region to stop codon and 3’UTR. To further locate confident m⁶A peaks, we divided the transcripts into five non-overlapping regions and assigned the confident m⁶A peaks into these regions. The fraction of confident m⁶A peaks of normal and chRCC tissues (Fig. 2C) in these five regions showed that they were dominantly enriched in 3′ UTR (40.46%, 40.55%), CDS (29.01%, 26.54%) and stop codon (17.32%, 18.59%). The motifs analysis results revealed that in both normal and chRCC tissues, the identified motifs are GGAYN (Y = C/U, N = U/A/C/G), AAACK (K = G/U) and NBNAH (N = U/A/C/G, B = G/C/U, H = C/U/A), RAACW (R = G/A, W = U/A). These motifs closely resemble the well-known m⁶A binding motif, RRACH (R = G/A, H = A/C/U).

Further we asked which RNA molecules prefer to contain m⁶A modification. We assigned confident m⁶A peaks to GRCh38 genome and found that 76.24% and 77.78% were mRNA, 18.42% and 17.16% were long non-coding RNA (lncRNA) in normal and chRCC tissues (Fig. 2D), separately. We also noticed that the confident m⁶A peaks number in chRCC tissues were less than that in normal tissues. Then we assigned m⁶A peaks to chromosome, genes and five non-overlapping regions. We found that the number of confident m⁶A peaks in chRCC tissues decreased globally among each chromosome (Fig. 2E). By analyzing the distribution of m⁶A peaks per gene, we found that most of m⁶A-motified mRNAs contained one or two m⁶A peak, while a small number of them contained three or more (Fig. 2F). Furthermore, m⁶A number per gene in chRCC tissues was all less than that in normal tissues. We also counted the number of m⁶A peaks among the five non-overlapping regions in normal and chRCC tissues (Fig. 2G) and found the number of m⁶A peaks decreased in chRCC tissues (except start codon) too. These results suggested that m⁶A modification level decreased in cancer tissues.

Differentially methylated m⁶A genes (DMMGs) participate in multi-cancer related pathways

To dissect the role of m⁶A modification, we subsequently identified DMMPs between normal and chRCC tissues. Compared to the normal tissues, the hyper- and hypo-methylated peaks were respectively regarded as hyper and hypo group. As to hyper group, we identified 644 shared differentially hyper-methylated m⁶A peaks corresponding to 593 genes in the three biological replicates. While as to hypo group, we identified 1304 shared differentially hypo-methylated m⁶A peaks representing 1137 genes (Fig. 3A). We further identified the m⁶A motif in hyper- and hypo- methylated m⁶A peaks within chRCC tissues by HOMER, and it revealed that the top consensus motifs in the hyper- and hypo-methylated m⁶A peaks were both similar to RRACH (Fig. S4).

Fig. 3.

Tissue analysis, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses of coding genes containing hypomethylation m⁶A peaks in chRCC tissues. A, Overlap of three biological replicates of hypermethylation m⁶A peaks identifying ~ 644 high-confident hypermethylation m⁶A peaks corresponding to 593 unique transcripts in chRCC tissues (left); Overlap of three biological replicates of hypomethylation m⁶A peaks identifying ~ 1,304 high-confident hypomethylation m⁶A peaks corresponding to 1,137 unique transcripts in chRCC tissues (right). B, Metagene profile illustrating the region distribution of hypermethylation and hypomethylation m⁶A peaks across the indicated mRNA segments. C, Pie chart presenting the fraction of the confident hypomethylation m⁶A peaks five non-overlapping transcript segments; D, Pie chart presenting RNA types (that is, transcript species) of the confident hypomethylation m⁶A peaks identified in chRCC. E-G, UP_TISSUE (E), Gene ontology (GO) (F), KEGG (G) analysis of the confident hypomethylation m⁶A genes

Then we examined the distribution profiles of DMMPs across the transcripts, and found that the density of DMMPs tended to be around the stop codon in both hyper and hypo group. In hypo group, the m⁶A peaks were especially enriched in the CDS and 3’ UTR near the stop codon; whereas in the hyper group, m⁶A peaks were enriched in the 3’ UTR near the stop codon (Fig. 3B). Further examination revealed that the m⁶A peaks in both groups were dominantly enriched within 3′ UTR (54.6% and 56.68%), CDS (25.92% and 23.14%) and around stop codon (13.19% and 11.02%) (Fig. 3C and Fig. S3A). Almost all of the transcripts containing DMMPs were mRNA (80.44% and 70.61%), only ~ 14–18% were lncRNA and a small proportion were other RNAs (Fig. 3D and Fig. S3B).

To explore the potential role of hypo-methylated m⁶A peaks in chRCC, we took advantage of DAVID to examine the most preferential expression tissues of genes with hypo-methylated peaks. The results showed that genes containing hypo-methylated m⁶A sites preferentially expressed in epithelium, followed by brain, placenta and renal cell carcinoma (RCC) (Fig. 3E), indicating the correlation between these genes and RCC. Furthermore, we performed GO enrichment analysis to uncover the functions of these genes. The results revealed that genes with hypo-methylated m⁶A peaks were enriched in many biological processes involved in kidney development and cancer pathogenesis, including transcription, androgen receptor signaling pathway, GTPase activity and cell-cell adhesion (Fig. 3F). Pathway analysis showed that genes with hypo-methylated m⁶A peaks were mainly enriched in pathways in cancer (Fig. 3G). These results suggested that genes with hypo-methylated peaks may participate in various pathophysiologic aspects of kidney cancer through different pathways.

Subsequently, we explored the potential function of hyper-methylated m⁶A peaks using the same method. The results showed that genes containing hyper-methylated m⁶A sites preferentially expressed in brain, followed by epithelium, duodenum, fetal kidney and ovary (Fig. S3C) which also indicating these genes is important in kidney development. GO biological process analysis revealed that genes with hyper-methylated m⁶A peaks were significantly associated with cancer-related biological processes, including protein phosphorylation, positive regulation of cholesterol efflux, ubiquinone biosynthetic process, regulation of mitophagy and so on (Fig. S3D). Pathway analysis showed that genes with hyper-methylated m⁶A peaks were mainly enriched in ubiquitin mediated proteolysis, metabolic pathways and adherents’ junction (Fig. S3E). These results further illustrated that gene with DMMPs played crucial roles in kidney cancer.

Differentially expressed genes involved in kidney development and cancer occurrence

We next identified the global mRNA expression patterns in normal and cancer groups. The results showed that a total of 3,911 mRNAs were significantly dysregulated in cancer group compared with normal group, including 2,344 down-regulated mRNAs and 1,567 up-regulated mRNAs (fold change ≥ 2, P < 0.05) (Fig. 4A). Hierarchical clustering depicted differential expression profiles in all the samples. (Fig. 4B). We then took advantage of the DAVID to examine the preferentially expressed tissues of these genes. The results showed that the dysregulated genes preferentially expressed in kidney, followed by liver and plasma (Fig. 4C). Then we performed GO analysis and KEGG pathway analysis. GO enrichment analysis demonstrated that dysregulated genes were significantly enriched in kidney development and cancer related biological processes involving metabolic process, cell adhesion and kidney development (Fig. 4D). While the KEGG pathway analysis revealed that dysregulated genes were significantly associated with metabolic pathways and protein metabolism (Fig. 4E).

Fig. 4.

Differential expression genes in chRCC tissues compared with adjacent normal tissues. A, Volcano plots showing the differentially expressed genes in chRCC tissues compared with those in adjacent normal tissues. B, Heatmap plots showing the differentially expressed genes in chRCC tissues vs. those in adjacent normal tissues. C-E, UP_TISSUE (C), Gene ontology (GO) (D), KEGG (E) analysis of the differential expression genes

Novel m⁶A gene signature identified by m⁶A-SEAL-seq and RNA-seq data

We comprehensively analyzed m⁶A-SEAL-seq and RNA-seq data in normal and chRCC tissues. The results demonstrated that among the 593 hyper-methylated genes detected by m⁶A-SEAL-seq, 44 targets tended to be down-regulated and 68 targets tended to be up-regulated (Fig. 5A, left); among the 1,137 hypo-methylated genes, 123 targets tended to be down-regulated and 51 targets tended to be up-regulated (Fig. 5A, right). Further, a significant increase of RPKM was observed in the hyper group compared to the hypo group (Fig. 5B).

Fig. 5.

Conjoint analysis of m⁶A-SEAL-seq and RNA-seq data. A, Overlap of hypermethylation genes with up-regulation genes and down-regulation genes (left). Overlap of hypomethylation genes with up-regulation genes and down-regulation genes (right). B, Cumulative distribution displaying the expression level changes in mRNAs with hypermethylation or hypomethylation m⁶A modification. C-D, Integrative genomics viewer (IGV) tracks showing the indicated m⁶A-seal reads distribution on target transcripts and the relative expression level in Normal and chRCC tissues. (C) NOTCH1, (D) FGFR1. F-G Survival analyses in the training set (F) and testing set (G). Log-rank p < 0.0001 showed a significant survival difference between the two sub-groups

To further analyze the role of DMMGs in cancers, we intersected DMMGs with Cancer Gene Census (CGC) database [49], a database consists of genes with strong indications of a role in cancer, and found 73 and 23 genes were annotated in hypo-methylated and hyper-methylated genes separately (Supplementary Dataset 3 and 4). Among these cancer related genes, 10 genes were differentially expressed genes, including NOCH1 and FGFR1. According to Integrative Genomics Viewer (IGV) software, the m⁶A modification level of NOTCH1 decreased significantly (Fig. 5C, left), then we measure the expression level of NOTCH1 and found in chRCC tissue, the expression level of NOTCH1 reduced significantly (Fig. 5C, right), suggesting NOTCH1 acts as a tumor suppressor gene in chRCC. Fibroblast growth factor receptor 1 (FGFR1) is a well-known oncogene. We found the m⁶A modification level of FGFR1 decreased significantly (Fig. 5D, left) and the expression level of FGFR1 increased in chRCC tissue (Fig. 5D, right). We observed differences in the expression patterns of these two genes, with both showing hyper-methylation in their modification levels but opposite expression patterns. This discrepancy indicates potential differences in m⁶A regulation pathways.

Thus, we conducted principal component analysis of the differential expressed DMMGs in 65 chRCC cases from The Cancer Genome Atlas (TCGA) database. Based on the expression of these genes, we could completely distinguish chRCC samples from normal samples (Fig. 5E). Cox regression screen and least absolute shrinkage and selection operator (LASSO) identified three m⁶A-dependent signatures (Supplementary Dataset 5) and defined a m⁶A-dependent cox model in the training set (Fig. S5A). Concordance index (C-index = 0.96) showed that the proposed model has a high prognostic power. In this model, we separated patients into high-risk or low-risk group according to their risk score, and patients with different 5-years survival could be distinguished completely between the two groups (Log-rank p < 0.0001) (Fig. 5F). AUC of ROC curve also confirmed the prognostic power of the m⁶A-dependent model (Fig. S5B). Then the proposed model was applied to the testing set for prediction. We calculated risk score of each patient in the testing set and assigned to high-risk or low-risk group according to the cut off value in the training set. The Kaplan-Meier survival curve and log-rank between the two groups showed significant difference (Fig. 5G), which demonstrated the high predictive ability of the m⁶A-dependent model. Furthermore, we also test the m⁶A-dependent model in ccRCC samples (a cohort of 602 cases from TCGA database) and PRCC dataset (a cohort of 318 cases from TCGA database). The results indicated that the m⁶A-dependent model is also suitable for ccRCC, but not PRCC (Fig. S5C, D).

Discussion

Despite having the lowest incidence and mortality rates among the three main types of kidney cancer, chRCC still contributes to thousands of deaths worldwide, according to statistics from the GCO in 2022. Over the past few years, there has been a growing effort to elucidate the mechanism of m⁶A modification in RCC, resulting in extensive accumulation of knowledge regarding the correlation between m⁶A modification and RCC. However, chRCC has been overlooked. Here, we demonstrated that m⁶A writer WTAP and m⁶A erasers FTO and ALKBH5 were downregulated in chRCC tissues, which may result in a different m⁶A modification pattern in chRCC tissues. By m⁶A-SEAL-seq, we confirmed that m⁶A modification pattern in chRCC tissues is distinct from that in normal tissues. The m⁶A peaks number and m⁶A peaks per gene in chRCC decreased and the distribution pattern of m⁶A peaks shifted to 3’UTR orientation in chRCC. Further functional studies showed that genes with hyper- or hypo-methylated peaks were mainly enriched in kidney development and cancer pathogenesis related pathway, which is a further proof of the fundamental role of m⁶A modification in chRCC. We also found m⁶A reader YTHDF2 were downregulated in chRCC tissues. The role of YTHDF2 in m⁶A-dependent gene regulation, particularly its impact on RNA stability, suggests the existence of m⁶A-dependent RNA degradation and gene dysregulation in chRCC. The down-regulation of NOTCH1 may be attributed to the down-regulation of YTHDF2. Additionally, the observed discrepancy in the expression pattern of FGFR1 indicates potential differences in m⁶A regulation pathways. Cumulative fraction of genes with hyper-methylated peaks and genes with hypo-methylated peaks indicated that these two group gene sets have different expression pattern in chRCC tissues, which may result from the dysregulation of m⁶A readers, such as YTHDF2.

In the present study, we depicted transcriptome-wide m⁶A profiling in normal and chRCC tissues by an antibody-free method. By comparing the m⁶A peaks in normal tissues and chRCC tissues, we revealed a total of 1730 DMMGs including 593 hyper- and 1137 hypo-methylated genes in chRCC tissues, which may be important factors in chRCC. For example, RARA, one of the hypo-methylated genes, has been demonstrated as a target of FTO in acute myeloid leukemia (AML). The m⁶A on RARA at UTRs influences the stability of RARA. The decreased m⁶A modification level of RARA destabilizes RARA mRNA, and then promotes tumorigenesis [50]. By CGC database analysis [49], we found 96 genes were causally implicated in cancers, such as ZEB1, which acts as an oncogene in ccRCC, PBRM1, ASXL2 and SETD2, which act as tumor suppressor genes in ccRCC. These results indicate m⁶A modification has a major influence on ccRCC, implying the roles of m⁶A modification in chRCC. Combined with our RNAseq data, we found 10 genes were differential expressed genes, including NOTCH1 and FGFR1. The level of m⁶A on NOTCH1 at 3’UTR and the expression level of NOTCH1 reduced significantly, which are similar to PER1 in pancreatic cancer and RARA in AML [33, 50]. PER1’s degradation is YTHDF2-dependent manner, while the mechanism underlying the degradation of NOTCH1 and RARA remains unknown. The level of m⁶A on FGFR1 at 3’UTR also decreased, but the expression level of FGFR1 increased significantly. It is reported that m⁶A reader IGF2BP3 enhances mRNA stability in an m⁶A-dependent manner [13, 51], which may account for the upregulation of FGFR1 in chRCC. Further functional studies are needed to clarify the molecular mechanisms of above-mentioned genes in the development of chRCC.

Despite the insights gained from our study, it is important to acknowledge a significant limitation: the small sample size. Our findings are based on data obtained from only three chRCC patients, which may not be sufficient to draw definitive conclusions about m⁶A modifications in chRCC. While the observed patterns in m⁶A regulation provide valuable initial insights, larger cohorts are needed to validate these results and ensure that they are broadly applicable to the wider chRCC patient population. Future studies with expanded sample sizes will be essential for confirming the role of m⁶A modifications in chRCC and for elucidating the underlying mechanisms with greater accuracy.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

The project was conceptualized and designed by Baoguo Li, Lulin Ma and Guifang Jia. Experiments were conducted by Zhigang Chen, Wei Zhang, Yang Qian, Nan Zhang，Min Lu, Liyuan Ge, Cheng Liu and Xiaojun Tian. The sequencing data were analyzed by Junbo Yang and Zixin Chen. Zhigang Chen, Junbo Yang and Wei Zhang wrote the manuscript. Baoguo Li, Lulin Ma and Guifang Jia revised the manuscript. All authors reviewed the manuscript and attest that they meet the current ICMJE criteria for authorship.

Funding

This work was supported by the National Basic Research Program of China (2019YFA0802201 and 2017YFA0505201), Beijing Natural Science Foundation (Z200010) and Beijing Haidian Hospital research fund (KYQ 2022018).

Data availability

The raw sequence data of RNA-seq (input), m⁶A-SEAL-seq have been deposited in the Genome Sequence Archive in the National Genomics Data Center (NGDC), Beijing Institute of Genomics, Chinese Academy of Sciences, under accession number (PRJCA004912).

Declarations

Ethical approval

Our research was conducted in accordance with the principles of the Declaration of Helsinki. The study was approved by The Beijing Haidian Hospital Medical Ethics Committee under the Project number M202231 and approval number 2022 − 112. Written informed consent was obtained from all participants prior to collecting clinical specimens for research purposes. All specimens were stored securely and handled using appropriate procedures to maintain their integrity. De-identified data were used in all analyses to protect patient privacy.

Competing interests

The authors declare no competing interests.