Interrogation of the interplay between DNA N6-methyladenosine (6mA) and hypoxia-induced chromatin accessibility by a randomized empirical model (EnrichShuf) Authors: Joseph Chieh-Yu Lai, Kai-Wen Hsu, Kou-Juey Wu Published: 2024 ABSTRACT Abstract N 6-Methyladenosine (6mA) is an epigenetic mark in eukaryotes regulating development, stress response and tumor progression. METTL4 has been reported as a 6mA methyltransferase induced by hypoxia. The detection and annotation of 6mA signals in mammalian cells have been hampered by the techniques and analytical methods developed so far. Here we developed a 6mA-ChIP-exo-5.1-seq to improve the sensitivity of detecting 6mAs in human cell lines. Furthermore, an EnrichShuf analysis tool for comprehensively comparing 6mA-ChIP-exo-5.1-seq, ATAC-seq, ChIP-seq and RNA-seq has been developed to annotate the functional relevance of 6mA in relation to chromatin accessibility and histone marks. Using a hypoxia-induced 6mA induction system as a model, we showed that hypoxic 6mA signals positively correlated with accessible chromatin regions. These 6mA signals correlate with their regulation by METTL4 under hypoxia, consistent with previous results. 6mAs also co-exist with H3K4me1, a histone mark for enhancers. Further analysis of enhancers using an ABC (active-by-contact) model shows that hypoxia-inducible factor-1α-induced H3K4me3 surrounds the 6mA/H3K4me1 site to augment active enhancers. These results suggest that correlation between 6mA and accessible chromatin regions plays a significant role in enhancer–promoter interactions during hypoxia-induced gene expression. Introduction The role of DNA N6-methyladenosine (6mA) in lower and higher eukaryotes regarding gene expression has been extensively characterized (1–10). In Caenorhabditis elegans, mitochondrial stress induced by antimycin leads to the induction of inheritable 6mA marking on mitochondrial response genes and induces gene expression, allowing their progeny to adapt to stress (6). In Mus musculus, 6mA played a pivotal role in development by repressing gene expression, achieved either by excluding chromatin remodelers or by incorporating the polycomb repressive complex 2 (PRC2) (5,8). In mammalian cells, 6mA could play a role in either gene activation or repression (7,9). METTL4 is proposed to be a putative 6mA methyltransferase and is induced by hypoxia to mediate nuclear 6mA deposition and promote tumor progression (9). The relationship between 6mA and chromatin accessibility has not been well delineated. Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq) is the obvious assay that can be used to answer this question (11). Histone marks have been shown to be linked to different chromatin accessibility, leading to different outcomes of gene expression status. For example, H3K4me3 is linked to gene activation, whereas H3K27me3 is linked to gene repression (12–14). Inherent 6mA can be incorporated with H3K4me3 and play a pivotal role in gene activation to adapt to stress (6). In glioblastoma (GBM) cancer stem cells (CSCs), 6mA is associated with gene repression and the presence of the H3K9me3 mark (15). Recent reports show that H3K4me1 and H3K27Ac represent the histone marks that are linked to enhancer and super-enhancer (16–18). A genomic region that contains concatenated (stitched) enhancers within 12.5 kb of each other is considered a super-enhancer (19). Although 6mA has been shown to be linked to gene activation or repression from different studies, its role in (super-)enhancer functions has never been elucidated. The challenge of 6mA detection in mammalian cells is caused by the low abundance of 6mA (20). Several methods have been developed for 6mA detection, including 6mA Methylated DNA immunoprecipitation (6mA-MeDIP), 6mA-Crosslinking-Exonuclease-sequencing (6mACE), single-molecule real-time sequencing (SMRT-seq) and 6mA-ChIP-exo-5.0 (7,21,22). In SMRT-seq, a high coverage of sequencing data is essential for 6mA detection, which can be costly (22). Other antibody-based methods may mislead the detection of 6mAs by non-specific binding of antibodies (20,23). Whole-genome amplification (WGA) is constantly used as a negative control in filtering out background 6mA signals in antibody-based detection methods (20). Although methods such as 6mACE or 6mA-ChIP-exo-5.0 can detect 6mA at higher resolution, UV photo-cross-linking introduces various types of DNA damage (7,21). These types of damage reduce the fidelity and yield of the sequencing library due to the reduction of the PCR amplification efficiency (24,25). For bioinformatic analysis of 6mA signals, various analytical methods have been developed to identify 6mA in different genomic regions (3,8,26,27). A χ2 test has revealed that 6mA positively correlates with both the expression of key stress-related genes in rice and the repression of LINE (long interspersed nuclear element) transposon expression in mice (3). Another randomized empirical model has been used to show that 6mA positively intersects with single-stranded DNA (ssDNA) in mice (8). A method remains to be developed to fully annotate 6mA signals together with other genomic datasets inside genomes. In this study, we characterized the functional relevance of 6mA by using a newly developed EnrichShuf, an R package designed for performing the comprehensive analysis of genomic features/regions and related expression traits. Our results showed that hypoxia-induced 6mAs positively correlated with accessible chromatin regions and H3K4me1. The interplay between 6mA, accessible chromatin regions and H3K4me1 may play a pivotal role in gene expression and promoter–enhancer interactions through hypoxia-induced METTL4 under hypoxia. Materials and methods Cell culture FaDu cells, a head and neck cancer cell line, were purchased from the Bioresource Collection and Research Center (BCRC, Hsinchu, Taiwan). The cell line was cultured in Dulbecco’s modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 1 mM sodium pyruvate and 1 × Antibiotic–Antimycotic (Gibco), under the conditions of 5% CO2 at 37°C. The absence of mycoplasma was tested by the Mycoplasma PCR Detection Kit (abm) before the experiments. Knockdown of METTL4 was achieved using a previously established method with lentivirus infection. For hypoxic conditions, cells were incubated in an environment of 1% O2, 5% CO2 and 94% N2 at 37°C for a period of 18 h. 6mA-ChIP-exo-5.1 sequencing and whole-genome amplification control DNA was extracted from cell pellets using a previously described method, followed by DNA fragmentation in the range of 100–200 bp according to the manufacturer’s protocol for Bioruptor Pico with 0.65 ml microtubes (9). 6mA-ChIP-exo-5.1 was developed based on a previously described 6mA-ChIP-exo-5.0 (7). From 30 to 100 μg of fragmented DNA was hybridized with 3 μg of the 6mA antibody (sysy-003, Synapse) overnight. The DNA:antibody mixture was cross-linked using UV-C irradiation in 1 × IP buffer (10 mM Tris–HCl pH 7.4, 150 mM NaCl, 0.1% Igepal-630) and pulled down by protein A Dynabeads (Invitrogen). Subsequent washes were performed with FA lysis buffer (50 mM HEPES-KOH pH 7.4, 150 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate), NaCl buffer (50 mM HEPES-KOH pH 7.4, 500 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate), LiCl buffer (100 mM Tris–HCl pH 7.9, 500 mM LiCl, 1% NP-40, 1% sodium deoxycholate), TE buffer (10 mM Tris–HCl pH 7.4, 1 mM EDTA) and Tris buffer (10 mM Tris–HCl pH 7.4) in the following enzymatic steps: DNA repair with PreCR Repair mix (M0309S, NEB), A-tailing with Klenow fragment (3′–5′, exo-) (M0212S, NEB), ligating the adapter with T4 ligase (M0202S, NEB) and T4 polynucleotide kinase (M0201S, NEB), fill-in with phi29 DNA polymerase (M0269S, NEB) and digesting with lambda exonuclease (M0262S, NEB). After a series of enzymatic reactions, DNA was eluted using an elution buffer (100 mM NaHCO3, 1% SDS) combined with RNase A. The sample was transferred to a new tube and digested with proteinase K at 56°C overnight. DNA was purified using clean-up with 3 × SPRI beads (Beckman), followed by a second adapter ligation using a splint adapter. The adapter-ligated sample was purified by clean-up with 3 × SPRI beads. Amplification was performed using 1 × Phusion High-Fidelity PCR Master Mix (NEB) and Illumina library primers (P-1.3 and P-2.1) for 12–14 cycles. The product was then purified using 1 × SPRI beads, resulting in a sequence library with an approximate concentration of 1–2 nM. Sequence data were generated by Illumina Nextseq 550 with 75 bp single-end sequence. WGA control was generated using a previously established method to filter out false-positive results caused by non-specific antibody binding (9). 6mA-ChIP-exo-5.1 analysis The sequenced reads were processed and trimmed using Trim Galore. To eliminate potential noise in the 6mA detection, reads aligned to mycoplasma and mitochondria were filtered out. The filtered reads were aligned to the hg38 using bowtie2 (28). After alignment, the aligned reads were filtered based on a mapping quality (MAPQ > 30), and PCR duplicates were removed by Picard. 6mA border sites under various conditions, including normoxia, hypoxia and METTL4-si under hypoxia, were identified using Model based Analysis of ChIP-Exo (MACE) (29). To perform 6mA signal analysis, hypoxic 6mA border sites were expanded into 300 bp peaks, and 6mA signals from each sample were counted in a stranded manner within these windows and normalized by edgeR (30). Positive hypoxic 6mA peaks were identified by comparing the hypoxic sample with the WGA sample, using a log2 fold change (FC) threshold higher than log2(1.1). Differential 6mA signals were analyzed using the criteria of log2FC higher than log2(1.1) for increased signals and lower than –log2(1.1) for decreased signals, comparing conditions of hypoxia versus normoxia and METTL4-si versus scramble-si under hypoxia. Assay for Transposase-Accessible Chromatin with high-throughput sequencing (ATAC-seq) Omni-ATAC was performed as previously described (31). Cells were detached from the plates using trypsin, and 5 × 104 detached cells were collected into a 1.5 ml tube. Cells were centrifuged at 500 g, 4°C for 5 min and washed twice with 1 × ice-cold phosphate-buffered saline (PBS). Cell pellets were resuspended in 50 μl of ATAC resuspend buffer (ATAC-RSB) (10 mM Trizma pH 7.8, 10 mM NaCl and 3 mM MgCl2), supplemented with 0.1% NP-40, 0.1% Tween-20 and 0.01% digitonin. The mixture was gently pipetted and incubated on ice for 3 min. To quench the lysis, 1 ml of ATAC-RSB supplemented with 0.1% Tween-20 was added. The samples were centrifuged at 500 g, 4°C for 5 min to pellet the nuclei. The nuclei were subjected to a transposase mix (2.5 μl of TDE1, 25 μl of 2 × TD buffer, 16.5 μl of 1 × PBS, 0.5 μl of 10% Tween-20, 0.5 μl of 1% digitonin and 5 μl of ultra pure water) and the reaction was performed at 37°C with shaking at 1000 rpm on a ThermoMixer for 30 min. After transposition, DNA was purified using the DNA Clean & Concentrator-5 kit (Zymo), following the manufacturer’s protocol. PCR was performed using the NEBNext High-Fidelity 1 × PCR Master Mix, supplemented with primers containing the sequence of Illumina adapters and indexes, to generate sequencing libraries. Libraries were sequenced on the Illumina Nextseq 550 platform, generating 75 bp paired-end reads. ATAC-seq analysis ATAC-seq was processed followed by standard ENCODE pipeline (https://github.com/ENCODE-DCC/atac-seq-pipeline). Tag-aligned bed files which did not contain mitochondria reads were selected for downstream analysis. Peaks were identified using MACS2 with the parameters ‘-f BED -g –shift -75 –extsize 150 –nomode –call-summits –keep-dup all –p 0.01’, targeting accessible chromatin regions (32). To identify conserved accessible regions across biological replicates for each condition, we called peaks from the merged tag-aligned files for each condition. Only peaks conserved across all replicates were selected for downstream analysis. Differential chromatin accessibility analysis was performed using DiffBind with tag-aligned and peak files from each sample with the cut-off by P-value < 0.05 and log2FC higher or lower than 0 (33). As the primary focus of this study was to investigate the correlation between hypoxic accessible chromatin regions and 6mA, we specifically extracted differential data within these hypoxic regions for subsequent analysis. RNA sequencing (RNA-seq) Cell pellets collapsed from the detached cells in the ATAC-seq procedure were washed with 1 × ice-cold PBS and lysed with 1 ml of Trizol (Ambion) at room temperature for 5 min. To the lysed sample, 200 μl of chloroform was added, and it was then mixed vigorously using a vortex. The mixture was placed on ice for 10 min. Following this, samples were centrifuged at 12 000 g at 4°C for 15 min. The supernatants were carefully collected and mixed with equal volumes of ethanol. The mixture was transferred to the column and RNA was extracted followed by RNA Clean & Concentrator-5 (Zymo) with DNase I treatment. The RNA-seq library was constructed using the RiboMinus with stranded RNA hyper prep kit (Zymo) with an input of 1 μg of RNA. Sequencing data were generated by Illumina Nextseq 550 with 150 bp paired-end sequence. RNA-seq analysis The RNA-seq data were processed using the standard ENCODE pipeline (https://github.com/ENCODE-DCC/rna-seq-pipeline). We selected the RNA-Seq by Expectation-Maximization (RSEM) gene or transcript quantification tables for downstream analysis (34). Data from each sample were imported using tximport, and differential gene analysis was conducted using both edgeR and limma with the cut off by P-value < 0.05 and log2FC higher or lower than 0 (30,35,36). Chromatin immunoprecipitation followed by sequencing (ChIP-seq) ChIP was conducted following previously established methods (9), using the antibodies H3K4me1 (ab8895, Abcam), H3K4me3 (ab8580, Abcam) and H3K27Ac (ab4729, Abcam). Each ChIP was carried out with 1 × 107 cells and 2 μg of antibody. ChIP DNA was purified using DNA concentrator-5 (Zymo). Library construction was performed using KAPA HyperPrep Kits. The libraries were sequenced on the Illumina Nextseq 550 platform to generate 75 bp single-end reads. ChIP-seq analysis Sequencing reads were processed according to the procedure used in the 6mA-ChIP-exo-5.1 sequencing analysis. Quality control of all alignment files and the estimation of fragment sizes were performed using the SPP. For H3K4me1, broad peaks were identified using MACS2 with the parameters ‘-f BED -g hs –nomode –broad –keep-dup all –p 0.01’ (32). For H3K4me3 and H3K27Ac, narrow peaks were identified using MACS2 with the parameters ‘-f BED -g hs –nomode –keep-dup all –p 0.01’ (32). Differential histone modification occupancy analysis was performed using the SICER with the comparison of hypoxia versus normoxia for each histone mark (37). Identification of predicted super-enhancers Super-enhancers were predicted using a modified version of the previously established ROSE algorithm, which applies H3K27Ac, a marker of enhancers, to facilitate the process (16,38). Using default parameters, H3K27Ac peak clusters were identified by extending the peaks to 12.5 kb and then concatenating them together. The H3K27Ac signal strength for each concatenated or single enhancer region was calculated to rank the signals. Predicted super-enhancers are defined as regions containing high H3K27Ac signals determined by ROSE. EnrichShuf EnrichShuf is a bioinformatics tool that utilizes the capabilities of valr to process genomic intervals (i.e. BED files), such as genomic features and regions, and generates values to calculate the significance of specific features, thereby providing insights into regulatory mechanisms. EnrichShuf constructs a randomized empirical model based on values generated from the processing of expected features (i.e. randomized observed features). To determine the significance and the enrichment of features, a z-test was applied using the mean and standard deviation (SD) from the randomized empirical model along with the observed value. In this study, EnrichShuf identifies features co-existing with regions if any part of them intersected with regions. For features not co-existing with regions, EnrichShuf calculates the distance between each feature and its nearest region using their edges. EnrichShuf can determine the significance of four types of correlations as described below. Assessing the correlation between features and regions across the genome Initially, a process is established to create a detailed table containing information on the distance between each feature and its nearest region. Several features can be obtained based on the specific range of distances from each feature to the nearest region. By inputting features into the processing, we were able to determine the number of features associated with regions within a specific range of distances (i.e. 0, 0–3 and 3–10 kb). The expected features were generated by randomly shuffling features in the genome with the same number of observed features. Subsequently, we derived values for both the observed and expected features by processing each distance range. Using the SD and the mean of the expected values, we assessed the significance and enrichment of the observed values. This enabled us to determine the correlation between the features and regions at each distance. Identifying regions enriched with features Based on the assessment of the correlation between features and regions across the genome, we obtained a range of distances with significant correlation between features and regions. In this section, we define the process of generating the number of features around each region within a specific range of distances. The same procedure used to assess the correlation between features and regions across the genome was applied to generate the randomized expected features. Through this procedure, we derived the observed and expected values for each region. Using the SD and the mean of the expected values, we assessed the significance and enrichment of the observed values for each region. This enabled us to determine the regions that were significantly positively and negatively enriched with features. Investigating the correlation of features with expression traits We first associated genomic loci with expression objects such as accessible chromatin regions for differential accessibility, promoters for differential gene expression and histone marks for differential occupancy. Next, we linked these ‘expression’ loci to features and defined them as observed expression traits. The expected expression traits were generated by randomly sampling the same number of expression traits as the observed traits. A cut-off value was set to identify significantly increased/decreased expression traits. The overlaps of the observed/expected and significantly increased/decreased expression traits were counted. Through this processing, we derived values for both the observed and expected values using our pre-defined cut-off. Using the SD and the mean of the expected values, we assessed the significance and enrichment of the observed values. This enabled us to determine the correlation between features and expression traits using cut-off values. Correlating the occurrence of features by regions Transcription machinery typically requires the participation of multiple factors or features. Therefore, we employed a randomized empirical model to assess the correlation between various types of features within or around specific regions. In this section, we typically require two types of features: one is the specific feature whose enrichment we aim to calculate, and the other is a feature with which we seek to establish an association. Initially, we associated regions with two types of features and generated a detailed table containing the distances of the regions to their nearest features. This allowed us to calculate the number of regions associated with the features within a specific range of distances. We identified the observed regions by processing specific features. Subsequently, by randomly sampling an equivalent number of regions to match the number of observed regions, we generated a set of expected regions. Using this procedure, we derived the values for both the observed and expected values. Using the SD and the mean of the expected values, we assessed the significance and enrichment of the observed values. This enabled us to determine the correlation between the two types of features within the surrounding regions. Application of EnrichShuf in chromosomal analysis Chromosomal enrichment analysis To analyze the enrichment of genomic features within each chromosome, we performed random shuffling for genomic features. The observed values were generated by counting the number of features present within each chromosome. By comparing the observed values with the expected values derived from the shuffling, we assessed the significance of chromosomal enrichment. This allowed us to determine whether specific genomic features were significantly enriched in certain chromosomes compared with their expected distribution across the genome. Correlation between genomic features and regions within each chromosome To analyze the correlation between genomic features and specific regions within individual chromosomes, we used the ‘assessing the correlation between features and regions across the genome’ method in EnrichShuf, which allowed us to evaluate how these features were distributed to regions within each chromosome. By combining the results (observed and expected values) across chromosomes, we can assess the correlation between features and regions across the genome, while accounting for the chromosomal distribution of genomic features. Chromosomal correlation analysis Various types of chromosomal lists according to genomic region or histone mark were obtained from analyses such as chromosomal enrichment analysis and the correlation between genomic features and regions within each chromosome. By applying a randomized empirical model, we can assess the correlation between these chromosomal lists to determine if the phenomena observed across various types of genomic features exhibited any chromosomal correlation. Active-by-contact (ABC) model for enhancer–promoter interaction prediction To predict enhancer–promoter interactions under hypoxia, we applied the ABC model to identify functional enhancers and their associated target genes (39). As the ABC model can predict enhancer–gene interactions without cell type-specific Hi-C data (39), we combined publicly available average Hi-C data (ftp://ftp.broadinstitute.org/outgoing/lincRNA/average_hic/average_hic.v2.191020.tar.gz) with our hypoxic ATAC-seq, H3K27Ac ChIP-seq and RNA-seq data to predict enhancer–promoter interactions. Since the publicly available Hi-C data were in the hg19 genome version, we re-aligned the ATAC-seq, H3K27Ac ChIP-seq and RNA-seq data to the hg19 genome for consistency. First, to compare with other hg38 results, we used the UCSC liftover tool to convert ATAC-seq peaks from hg38 to hg19. Using the hypoxic liftover ATAC-seq peaks and ATAC-seq alignment files, we identified potential enhancers and assessed their chromatin accessibility. Next, we quantified enhancer activity based on the H3K27Ac ChIP-seq alignment data. We then calculated the ABC score using gene expression and average Hi-C data. Finally, we considered an ABC score > 0.02 as indicating a significant enhancer–promoter interaction. CRISPRi for enhancer–promoter interaction identification We applied CRISPRi [clustered regularly interspaced palindromic repeats (CRISPR) interference] to target the enhancer predicted by the ABC model for enhancer validation based on a previous method (40). The p5w-dCas9 KRAB.pBsd and pU6-sgRNA.Ppuro expression constructs were obtained from the National RNAi Core Facility (Academia Sinica, Taiwan). Briefly, single guide RNAs (sgRNAs) targeting the MALAT1 enhancer were designed and cloned into pU6-sgRNA.Ppuro vector to construct pU6-sgRNA.Ppuro-MALAT1-6mA-1 (5′-AGGCGGATCACTTAAGGTC-3′) and pU6-sgRNA.Ppuro-MALAT1-6mA-2 (5′-TAGACTTCATTTGGGCCTA-3′) by the National RNAi Core Facility. The p5w-dCas9 KRAB.pBsd and pU6-sgRNA.Ppuro-MALAT1-6mA-1 or pU6-sgRNA.Ppuro-MALAT1-6mA-2 expression constructs were co-transfected into FaDu cells using Lipofectamine 2000. Following transfection, MALAT1 (F: 5′-GAATTGCGTCATTTAAAGCCTAG-3′; R: 5′-GTTTCATCCTACCACTCCCAATT-3′) expression was measured by quantitative PCR (qPCR). Glut (F: 5′-CGGGCCAAGAGTGTGCTAAA-3′; R: 5′-TGACGATACCGGAGCCAATG-3′) is the control for hypoxia, using 18S (F: 5′-GGCGGCGTTATTCCCATGA-3′; R: 5′-GAGGTTTCCCGTGTTGAG-3′) as the internal control. Statistical analysis All observed versus expected significant values were calculated using either a z-test or a binomial test with EnrichShuf. For the comparison of H3K27Ac signals, a Wilcoxon test was performed. Results Development of 6mA-ChIP-exo-5.1 to identify 6mA sites at higher resolution Since we showed that 6mA signals could be significantly detected under hypoxia (9), we used the hypoxia induction and METTL4 knockdown experiments as a model system in this study (see the Materials and methods). From previous studies, combination of UV photo-cross-linking and exonuclease digestion followed by 6mA MeDIP could identify 6mA at higher resolution (7,21,41). However, DNA damage caused by UV photo-cross-linking could lead to a decrease in PCR efficiency and the yield of the sequencing library during sequencing experiments (24,25). To improve this aspect, we added a step to repair the UV-damaged DNA and increase the yield of the sequencing library (designated as 6mA-ChIP-exo-5.1) (Figure 1A). To avoid the signal contamination from mycoplasma or mitochondria, we aligned the sequencing reads to the genome of mycoplasma and mitochondria and exclude these signals (see the Materials and methods) (Figure 1A; Supplementary Figure S1A, B). Using this improvement, the complex library generated enabled us to identify 6mA sites by MACE and called peaks by four steps, namely normalization and bias correction, signal consolidation and noise reduction, border detection and border matching (29). During peak calling by each algorithm (EM, entropic mean; AM, arithmetic mean; GM, geometric mean; SNR, signal to noise ratio), we found that GM could not pass the step of border matching (data not shown). Results of other algorithms showed that most of the peaks are single forward or reverse peaks (SFB and SRB) with a small percentage of Gale–Sharpley paired border peaks (GSBs) (data not shown). Therefore, combination of 6mA-ChIP-exo-5.1 and MACE could identify stranded 6mA sites, and 6mA usually appeared on one strand in the human genome, which was consistent with previous findings from the 6mACE (21). From this finding, we used the information in the border detection step in each algorithm for downstream comparison. The most highly conserved 6mA border sites were found between the algorithms of EM and AM (n = 101 252) (Supplementary Figure S1C). Finally, we select intersected peaks of EM and AM for downstream analysis (normoxia, n = 62 312; hypoxia, n = 101 591; METTL4-si under hypoxia, n = 56 397) (Figure 1B; Supplementary Figure S1D). The profile of 6mA-ChIP-exo-5.1 sequencing reads centered by 6mA border sites showed that peaks could be detected under stranded manners (Supplementary Figure S1E). Genome-wide analysis of 6mA-ChIP-exo-5.1 sequencing. (A) The flow chart of 6mA-ChIP-exo-5.1 sequencing illustrated three major components of the procedure for identifying 6mA sites: experimentation, quality control and data analysis. A step to repair damaged DNA caused by photo-cross-linking was added to increase the final yield of sequencing. (B) Bar graph showing the number of 6mA sites identified by MACE under each condition (gray, normoxia, n = 62 312; red, hypoxia, n= 101 591; cyan, METTL4-si under hypoxia, n = 56 397). The highest number of 6mA peaks was identified under hypoxia. (C) Pie plot showed the percentage of true-positive 6mA sites (56.84%) (red) and false-positive 6mA sites (43.16%) (gray). (D) The peak profile for hypoxic 6mA peaks surrounding gene bodies or transcription start sites (TSSs) indicated that true-positive regions (red) were generally more distal to the TSS compared with false-positive regions (gray). (E) Bar plot showing the percentage of true-positive (left) or false-positive (right) hypoxic 6mA peaks annotated across promoters, gene bodies and intergenic regions. The result showed that a lower percentage of true-positive regions could be found on promoters compared with false-positive regions. (F) The bar graph showed that compared with false-positive 6mA peaks (gray bars), a significantly higher percentage of true-positive 6mA peaks (red bars) showed the induction of 6mA signals by hypoxia and the decrease of 6mA signals by METTL4-si. We further performed 6mA-ChIP-exo-5.1 sequencing with WGA control to exclude false-positive signals (Figure 1A). To compare the signals in each condition, we extend the hypoxic 6mA border sites to 300 bp peaks and count the signals of each sample under stranded manners (Figure 1A). By subtracting the 6mA signals generated from WGA (43.16% of 6mA peaks; n = 43 992), 56.84% of hypoxia 6mA peaks were shown to be true positive (n = 57 599) (Figure 1C; Supplementary Figure S1F). The distribution of 6mA peaks on the genome from this subtraction showed that positive regions were located distal to transcription start sites (TSSs) (Figure 1D). Annotation of 6mA peaks showed lower percentages of positive regions at promoters compared with false-positive 6mA peaks (Figure 1E). Results in the scatter plot of log2FC comparison and bar plots showed that a higher percentage of 6mA peaks were regulated by METTL4 under hypoxia versus WGA control (Figure 1F; Supplementary Figure S1G). Development of an EnrichShuf algorithm to analyze the correlation between 6mA peaks and other markers To interrogate the correlation between 6mA and each genomic region with specific annotation (e.g. accessible chromatin regions, histone marks), we developed EnrichShuf, an R package (Figure 2). The principle of EnrichShuf is based on comparing the observed value versus the expected value through an empirical model (Figure 2A). Expected features (e.g. 6mA) were generated either by shuffling observed features across the genome or by randomly sampling specified genomic regions (e.g. accessible regions) or expression traits (expression associated with genomic loci, e.g. gene expression associated with promoter, chromatin accessibility associated with a genomic locus) (Figure 2B, C). Here, we showed each type of processing used to generate values that can be applied in downstream statistical analysis (Figure 2D–F). Firstly, one process quantified the number of genomic features located within a pre-defined distance range from genomic regions (Figure 2D). This involved correlating features with the nearest region and counting the features within a specified proximity threshold (Figure 2D). Secondly, the analysis included enumerating features associated with each genomic region (Figure 2E). This was accomplished by defining a cut-off distance, setting up the association where features were considered to be associated with a region and then tallying the features within this range (Figure 2E). Thirdly, the method identified the occurrences of genomic features (e.g. 6mA) when expression traits (e.g. chromatin accessibility or gene expression) exhibited significant differential information, or when genomic regions (e.g. accessible chromatin regions) co-existed with other markers (e.g. histone marks) (Figure 2F). With a different combination of randomization and processing, we built four major functions within EnrichShuf: assessing the correlation between features and regions across the genome, identifying regions enriched with features, investigating the correlation of features with expression traits and correlating the presence of features within regions (Figure 2G) (see the Materials and methods). Development of EnrichShuf to identify the correlation between genomic features (e.g. 6mA peaks) and regions. (A) The overview of EnrichShuf involved obtaining expected features by randomizing the observed ones. Values can be derived by processing these features, which enables the construction of a randomized empirical model with the expected features. This empirical model can be used to estimate the significance of the observed value. (B) The schematic plot illustrates the process of shuffling observed genomic features to obtain the expected ones. (C) The schematic plot illustrates the process of randomly sampling expression traits or genomic regions to obtain the expected outcomes. (D) The process involves two steps to generate values. First, it associates features with the nearest region on the genome and calculates the distance. Second, using the range of distances as a cut-off, we can determine the number of features. (E) The process is designed to determine the number of features associated with each region, based on a specific range of distances. (F) The process involves two steps to generate values. First, it associates features (e.g. 6mA) with expression traits (e.g. differential gene expression or chromatin accessibility) or regions (e.g. chromatin accessibility regions). Second, the list from the first step will be overlapped with differential expression traits or genomic regions associated with other markers (e.g. histone marks) and counts the occurrences. (G) EnrichShuf involves various functions, each resulting from different combinations of randomization and processing: (i) assessing the correlation between features and regions across the genome (combine B and D); (ii) identifying regions enriched with features (combine B and E); (iii) investigating the correlation of features with expression traits (combine C and F); and (iv) correlating the occurrence of features by regions (combine C and F). With the capability of ‘assessing the correlation between features and regions across the genome’ from EnrichShuf, we found that there was no consistently significant positive correlation between 6mA and TSSs at different distances (Supplementary Figure S2A). Furthermore, we correlated 6mA with transcript bodies and found that there was a positive enrichment of 6mA when 6mA was located in transcript bodies (Supplementary Figure S2B). In addition, across various distances of ranges from 1 kb to 1000 kb, there was no positive correlation between 6mA and genomic regions, such as TSSs and transcript bodies (Supplementary Figure S2C). This result suggests that 6mA was not directly correlated with TSSs or transcript bodies (Supplementary Figure S2) to regulate gene expression. Identification of the positive correlation between 6mA and accessible chromatin regions by EnrichShuf and ATAC-seq To explore the relationship between 6mA and gene expression under hypoxia, we applied ATAC-seq to precisely profile the distribution of accessible chromatin regions induced by hypoxia and correlated these regions with 6mA peaks (Figure 3A–C). The results showed that when correlating 6mA with accessible chromatin regions, there were significant positive enrichments of 6mA at distances of 0, 0–3 and 3–10 kb (Figure 3A). When correlating accessible chromatin regions with 6mA, there were significant positive enrichments of accessible chromatin regions at distances of 0, 0–3, 3–10 and 10–20 kb (Figure 3B). We next examined the correlation between 6mA and accessible chromatin regions across various ranges of distances from 1 to 1000 kb with the cut-offs of log2FC > 0.05 and false discovery rate (FDR) < 0.05 (Figure 3C). The results showed that when using 6mA as the genomic feature, the significantly positive correlations of accessible chromatin regions were observed within 27 kb (Figure 3C). Similarly, when using accessible chromatin regions as the genomic feature, the significantly positive correlations of 6mA were observed within 53 kb (Figure 3C). A positive correlation between 6mA and accessible chromatin regions was identified and 6mA peaks related to accessible chromatin regions were located upstream of the TSS. (A) The bar graph shows a significant correlation between 6mA and accessible chromatin regions at distances of 0 (P = 6.57 × 10−10), 0–3 kb (P = 1.3 × 10−18) and 3–10 kb (P = 7.07 × 10−21). (B) The bar graph shows a significant correlation between accessible chromatin regions and 6mA at distances of 0 (P = 2.39 × 10−4), 0–3 kb (P = 1.47 × 10−66), 3–10 kb (P = 2.18 × 10−91) and 10–20 kb (P = 6.85 × 10−14). (C) The profile showed the correlation between 6mA and accessible chromatin regions across various ranges of distances from 1 to 1000 kb. The result showed a significant observation within 27 kb (red line). In contrast, when correlating accessible chromatin regions with 6mA across various ranges of distances from 1 to 1000 kb, we observed that the significance was within 53 kb (cyan line). (D) The information of each bin showing the significant correlation between 6mA and accessible chromatin regions across various ranges of distances from 1 to 1000 kb. We identified 63 859 significantly positive and 48 369 significantly negative 6mA-enriched accessible chromatin regions (upper). Additionally, there were 22 892 significantly positive and 24 824 significantly negative accessibility-enriched 6mA peaks (bottom). (E) The flowchart shows the process of integrating information from 6mA to its nearest accessible chromatin regions, as well as from accessible chromatin regions to their nearest 6mA peaks, in order to construct the associated network. After constructing the association network, we filtered it to include only those associations with either positively or negatively 6mA-enriched accessible chromatin regions, or accessibility-enriched 6mA peaks. (F) The profiles showed the frequency of 6mA peaks in positive (red line) and negative (blue line) association networks surrounding gene bodies. (G) After categorizing 6mA peaks in the positive association network into different groups based on the annotation information of its associated accessible chromatin regions, the profiles showed the frequency of 6mA peaks across these groups, including promoter (red), gene body (cyan) and intergenic region (gray). (H) Bar graph showing a significantly higher percentage of positive accessibility-enriched 6mA peaks upstream of TSS compared with negative accessibility-enriched 6mA peaks. To examine the association between 6mA and accessible chromatin regions, we linked 6mA to the nearest accessible chromatin regions, or accessible chromatin regions to the nearest 6mA, and defined different groups, namely co-existence (0 kb), proximal (0–3 kb), distal (3–10 kb) or other (>10 kb) (Supplementary Figure S3A). After dividing 6mA and accessible chromatin regions into different groups, we profiled the distribution of the distances between these features/regions and TSSs (Supplementary Figure S3B). The results showed significantly higher percentages of 6mA nearby TSSs at the co-existence and proximal groups, compared with the groups of distal and others (Supplementary Figure S3B). However, this observation was not detected in accessible chromatin regions (Supplementary Figure S3B). This result suggested that the correlation between 6mA and TSSs could be influenced by the association with accessible chromatin regions, but the correlation between accessible chromatin regions and TSSs remained unchanged by 6mA (Supplementary Figure S3B). We next profiled the frequency of 6mA peaks around gene bodies for each group (Supplementary Figure S3C). The result showed that 6mA peaks were located around the TSS when they co-existed with accessible chromatin regions (Supplementary Figure S3C). With the correlation data between 6mA and accessible chromatin regions at various ranges of distances from 1 to 1 000 kb (Figure 3C), we conducted binomial tests (i.e. the function of ‘identifying regions enriched with features’ from EnrichShuf) to identify 6mA-enriched accessible chromatin regions and accessibility-enriched 6mA peaks (see the Materials and methods) (Figure 3D). For quality control of statistics, we first performed the binomial tests across 10–150 repetitions, with each repetition comprising 10 bins (Supplementary Figure S4A). The positive results of these tests indicate ∼98% conservation of 6mA-enriched accessible chromatin regions and 99% conservation of accessibility-enriched 6mA peaks across 100–150 repetitions, with each repetition comprising 10 bins (Supplementary Figure S4A). Similarly, the negative results show ∼90% conservation of 6mA-enriched accessible chromatin regions and 98% conservation of accessibility-enriched 6mA peaks in the same testing framework (Supplementary Figure S4A). In the results from 100 iterations of the binomial test, we identified a total of 63 859 significantly positive and 48 369 significantly negative 6mA-enriched accessible chromatin regions, as well as 22 892 significantly positive and 24 824 significantly negative accessibility-enriched 6mA peaks (Figure 3D). We then compared the previously identified groups (i.e. proximal, distal and others) of 6mA and accessible chromatin regions (Supplementary Figure S3A) with the accessibility-enriched 6mA peaks and the 6mA-enriched accessible chromatin regions for quality control (Supplementary Figure S4B). The results showed that the majority of positive 6mA-enriched accessible chromatin regions and accessibility-enriched 6mA peaks were found in the groups of proximal or distal groups (Supplementary Figure S4B). This result confirmed the authenticity of regions enriched with features, which were identified through binomial test (Supplementary Figure S4B). To identify the association between 6mA and accessible chromatin regions, we linked the 6mA peak with its nearest accessible chromatin regions and the accessible chromatin with its nearest 6mA peaks (Figure 3E). We combined these associations and divided them into a positive and negative association network based on either positive or negative 6mA-enriched accessible chromatin regions and accessibility-enriched 6mA peaks (Figure 3E). To minimize potential noise, we filtered out the associations that contained 6mA peaks co-existing with accessible chromatin regions in these networks. We then annotated the 6mA peaks and accessible chromatin regions within the filtered network, and selected the association by a criterion that required a 6mA peak and accessible chromatin regions to share the same nearest TSS. To examine the correlation between 6mA and accessible chromatin regions in gene regulation, we divided the positive association network into different groups based on the annotated location (e.g. promoter, intron, exon, downstream, intergenic) of 6mA and accessible chromatin regions, followed by ranking these groups according to the number of associations (Supplementary Figure S4C). The groups of association from the top third to the seventh showed that either 6mA peaks or accessible chromatin regions within these groups were located at the promoter regions (Supplementary Figure S4C), suggesting that the association between 6mA and accessible chromatin regions played an important role in gene regulation. Next, we profiled the 6mA peaks within the positive and the negative association networks around gene bodies (Figure 3F). The results showed significantly higher frequency of 6mA peaks around gene bodies in the positive association network compared with the negative association network (Figure 3F). To further investigate the details of 6mA in the positive association network, we categorized these associations into groups based on the location of accessible chromatin regions (e.g. promoter, gene body and intergenic) and separately profiled the 6mA peaks for each group (Figure 3G). The results showed that when accessible chromatin regions were located in either promoter or intergenic areas, their associated 6mA peaks were enriched around the gene body, especially upstream of the TSS (Figure 3G). Moreover, a comparison between positive and negative accessibility-enriched 6mA peaks showed that a higher proportion of positive peaks were enriched upstream of the TSS (Figure 3H). Positive correlation between 6mA and METTL4-regulated accessible chromatin regions highlights their crucial role in gene activation under hypoxia We first performed differential analysis for 6mA signals, gene expression and chromatin accessibility to identify patterns of gene activity across each comparison: hypoxia versus normoxia and METTL4-si versus scramble-si under hypoxia (Supplementary Figure S5A–C). In 6mA differential analysis, the signals of 43 657 6mA peaks were increased and those of 6955 6mA peaks were decreased in hypoxia versus normoxia, whereas the signals of 8062 6mA peaks were increased and 42 426 6mA peaks were decreased in METTL4-si versus scramble-si under hypoxia (see the Materials and methods) (Supplementary Figure S5A). In differential gene expression analysis, the expression of 3845 genes was increased and that of 3683 genes was decreased in hypoxia versus normoxia, whereas the expression of 3021 genes was increased and that of 2980 genes was decreased in METTL4-si versus scramble-si under hypoxia (see the Materials and methods) (Supplementary Figure S5B). In differential chromatin accessibility analysis, the accessibility of 40 389 accessible chromatin regions was increased and that of 2000 was decreased in hypoxia versus normoxia, whereas 6419 increased and 23 173 decreased accessible chromatin regions were identified in METTL4-si versus scramble-si under hypoxia from differential chromatin accessibility analysis (see the Materials and methods) (Supplementary Figure S5C). To assess the correlation between the dynamic change of 6mA and chromatin accessibility across various conditions, we applied the function of ‘investigating the correlation of features with expression traits’ from EnrichShuf to analyze the differential data of 6mA signals and chromatin accessibility (see the Materials and methods) (Figure 4A). The results showed that 6mA signals were increased under hypoxia through METTL4 when 6mA peaks co-existed with accessible chromatin regions (Figure 4A). Similarly, chromatin accessibility was increased under hypoxia through METTL4 when accessible chromatin regions co-existed with 6mA peaks (Figure 4A). We next correlated gene expression with either 6mA or accessible chromatin regions (Figure 4B). The results showed that gene expression was induced by METTL4 under hypoxia when the promoter was occupied by either a 6mA peak or an accessible chromatin region co-existing with a 6mA peak (Figure 4B). After observing that the co-existence between 6mA and accessible chromatin regions played an important role in gene activation (Figure 4B), we wanted to know whether the distal association between 6mA and accessible chromatin regions contributed to gene activation. To address this question, we correlated gene expression with either positive or negative 6mA-enriched accessible chromatin regions (Figure 4C). The result showed that compared with the negative 6mA-enriched accessible chromatin regions, gene expression was decreased in METTL4-si versus scramble-si under hypoxia when the promoter was occupied by positive 6mA-enriched accessible chromatin regions (Figure 4C). This result indicated that the distal association between 6mA and accessible chromatin regions (i.e. positive 6mA-enriched accessible chromatin regions) could affect genes activated by METTL4 (Figure 4C). We then correlated differential chromatin accessibility with its annotated locations, showing that knockdown of METTL4 significantly decreased chromatin accessibility at promoters under hypoxia (Figure 4D). Compared with differential chromatin accessibility of hypoxia versus normoxia, gene expression was decreased in METTL4-si versus scramble-si under hypoxia when the promoter was occupied by decreased accessible chromatin regions in METTL4-si versus scramble-si under hypoxia (Supplementary Figure S5D). These results indicate that METTL4 is the main factor that correlated with gene regulation through modulating chromatin accessibility via 6mA under hypoxia (Figure 4A–D; Supplementary Figure S5D). Transcripts activated by METTL4 under hypoxia are majorly affected by accessible chromatin regions. (A) The bar graph showed the correlation between differential 6mA signals and accessible chromatin regions (upper), and the correlation between differential chromatin accessibility and 6mA peaks (bottom). The results indicate that when 6mA co-existed with accessible chromatin regions (top), 6mA signals were increased (left, brown, P = 6.97 × 10−8) in hypoxia versus normoxia and decreased (sky blue, P = 2.24 × 10−4) in METTL4-si versus scramble-si under hypoxia. In accessible chromatin regions co-existing with 6mA (bottom), chromatin accessibility was increased (left, salmon, P = 8.93 × 10−25) in hypoxia versus normoxia and decreased (sky blue, royal blue, P = 1.81 × 10−57) in METTL4-si versus scramble-si under hypoxia. (B) The bar graph showed the correlation between gene expression and different types of features, including 6mA, accessible chromatin regions co-existing with 6mA and accessible chromatin regions. The results showed that when the promoter was occupied by 6mA (top), gene expression was increased (left, red, P = 2.33 × 10−8) in hypoxia versus normoxia and decreased (right, blue, P = 2.36 × 10−17) in METTL4-si versus scramble-si under hypoxia. In the promoter that was occupied by the accessible chromatin region co-existing with 6mA (middle), gene expression was increased (left, red, P = 3.4 × 10−8) in hypoxia versus normoxia and either increased (right, red, P = 2.14 × 10−3) or decreased (right, blue, P = 2.37 × 10−5) in METTL4-si versus scramble-si under hypoxia. In the promoter that was occupied by an accessible chromatin region (bottom), gene expression was either increased (left, red, P = 4.1 × 10−13) or decreased (left, blue, P = 9.04 × 10−7) in hypoxia versus normoxia and either increased (right, red, P = 7.01 × 10−8) or decreased (right, blue, P = 1.22 × 10−4) in METTL4-si versus scramble-si under hypoxia. (C) The bar graph shows the correlation between gene expression and each type of accessible chromatin region, such as positive and negative 6mA-enriched caccessible chromatin regions. The results showed that when the promoter was occupied with a positive 6mA-enriched accessible chromatin region (top), gene expression was increased (left, red, P = 2.57 × 10−18) in hypoxia versus normoxia and either increased (right, red, P = 8.93 × 10−3) or decreased (right, blue, P = 1.03 × 10−10) in METTL4-si versus scramble-si under hypoxia. In the promoter that was occupied with a negative 6mA-enriched accessible chromatin region (bottom), gene expression was increased (left, red, P = 6.09 × 10−15) in hypoxia versus normoxia and increased (left, red, 6.48 × 10−9) in METTL4-si versus scramble-si under hypoxia. (D) The bar graph shows the correlation between chromatin accessibility and each type of its annotated location, such as promoter, gene body and intergenic regions. The results showed that METTL4-si decreased chromatin accessibility in promoters (top, right, royal blue, P = 2.9 × 10−67) but increased the chromatin accessibility in intergenic regions (bottom, right, salmon, P = 1.33 × 10−38). (E) Scatter plot showing a negative correlation between hypoxia versus normoxia and METTL4-si versus scramble-si under hypoxia, and 2124 transcripts activated by METTL4 under hypoxia were identified. (F) After correlating differential chromatin accessibility with transcripts activated by METTL4 under hypoxia, we identified 279 decreased and 4 increased chromatin accessibility regions in METTL4-si versus scramble-si under hypoxia at promoter regions. (G) After identifying the transcripts activated by METTL4 under hypoxia and associated with regions of decreased chromatin accessibility in METTL4-si at the promoter, gene ontology analysis was performed. Results show that a large portion of genes participate in trans-Golgi, lysosome transport, tRNA process and the Wnt signaling pathway. After analyzing differential transcript expression, we found that there was a negative correlation between the log2FCs of hypoxia versus normoxia and METTL4-si versus scramble-si under hypoxia, identifying 2124 transcripts up-regulated by METTL4 under hypoxia (Figure 4E). We linked these transcripts with differential chromatin accessibility at their promoters (Figure 4F). The results showed that predominant accessible chromatin regions were decreased in METTL4-si versus scramble-si under hypoxia (Figure 4F). Gene Ontology analysis of transcripts whose promoters displayed decreased chromatin accessibility in METTL4-si under hypoxia shows that most of these genes were involved in trans-Golgi, lysosome transport, tRNA processing, and the Wnt signaling pathway, and they were induced by METTL4 under hypoxia (Figure 4G). A significant correlation between 6mA and H3K4me1/accessible chromatin regions is found within enhancers To test whether 6mA signals are linked to histone marks, histone 3 lysine 4 tri-methylation (H3K4me3), a marker of an activated promoter, was selected (42). It was based on prior studies showing that inherent 6mA can be incorporated with H3K4me3, playing a pivotal role in gene activation to adapt to stress (6). Previous results showed that accessibility-associated 6mAs were typically found upstream of the TSS (Figure 3F–H), suggesting a potential association with enhancers. Based on this observation, we selected two histone marks that are known to correlate with enhancers [e.g. histone 3 lysine 4 mono-methylation (H3K4me1) and histone 3 lysine 27 acetylation (H3K27Ac)] for possible linkage (17,43). After correlating 6mA with these histone marks, the results showed that 6mA was positively correlated with H3K4me1 (Figure 5A). Although there was no positive correlation of the co-existence between 6mA and other histone marks, including H3K4me3 and H3K27Ac (Figure 5A), we correlated 6mA with these histone marks across various ranges of distances from 1 to 1000 kb and showed the positive enrichment of 6mA at distal associations (Supplementary Figure S6A). After identifying the positive correlation of the co-existence between 6mA and each type of genomic region, including accessible chromatin regions (Figure 3A, B) and H3K4me1 (Figure 5A), we performed intersection analysis of accessible chromatin regions and H3K4me1 (Supplementary Figure S6B). The results showed that there were 103 573 accessible chromatin regions co-existing with H3K4me1, and 145 430 H3K4me1 peaks co-existing with accessible chromatin regions (Supplementary Figure S6B). We further correlated 6mA with the results of intersection analysis (Supplementary Figure S6C). The results showed that at various ranges of distances from 1 to 1000 kb, there was no significant difference in the correlation between 6mA and accessible chromatin regions that either co-existed or did not co-exist with H3K4me1 (Supplementary Figure S6C). Within super-enhancers, 6mA shows a significant correlation with H3K4me1 and accessible chromatin regions. (A) The bar graph shows the positive correlation of co-existence between 6mA and H3K4me1 (red, P = 3.64 × 10−41). (B) The bar graph shows that when H3K4me1 was occupied with 6mA, there was a positive enrichment of accessible chromatin regions (P = 6.84 × 10−131) and H3K27Ac (P = 1.3 × 10−101). In accessible chromatin regions occupied with 6mA, there was a positive enrichment of H3K4me1 (P = 9.54 × 10−32), H3K4me3 (P = 2.07 × 10−75) and H3K27Ac (P = 4.57 × 10−95). (C) The bar graph shows a positive correlation between 6mA and predicted super-enhancers at distances of 0 (P = 2.52 × 10−49), 0–3 kb (P = 2.54 × 10−15), 3–10 kb (P = 3.91 × 10−23) 10–20 kb (P = 1.67 × 10−24), 20–30 kb (P = 4.49 × 10−14) and 30–40 kb (1.65 × 10−5). (D) The box graphs show significantly higher H3K27Ac signals in super-enhancers co-existing with 6mA (P = 6.32 × 10−10). (E) The bar graph shows the correlation of co-existence between 6mA and various features/regions (i.e. 6mA and accessible chromatin regions) within or outside of super-enhancers (from left to right: accessible chromatin regions within super-enhancers, P = 4.27 × 10−72; accessible chromatin regions outside of super-enhancers P = 2.23 × 10−2; H3K4me1 within super-enhancers, P = 1.2 × 10−21; H3K4me1 outside super-enhancers, P = 4.91 × 10−55). (F) The profile shows the correlation between 6mA and accessible chromatin regions (top) or H3K4me1 (bottom) at the ranges of distances from 1 to 1000 kb within (red line) or outside (gray line) super-enhancers. (G) The bar graph shows the correlation between differential chromatin accessibility and each type of histone mark. In H3K4me1 (top), chromatin accessibility was increased (left, salmon, P < 1 × 10−360) in hypoxia versus normoxia and either increased (right, salmon, P = 1.5 × 10−5) or decreased (right, royal blue, P < 1 × 10−360) in METTL4-si versus scramble-si under hypoxia. In H3K4me3 (middle), chromatin accessibility was increased (left, salmon, P < 1 × 10−360) in hypoxia versus normoxia and decreased (right, royal blue, P < 1 × 10−360) in METTL4-si versus scramble-si under hypoxia. In H3K27Ac (bottom), chromatin accessibility was increased (left, salmon, P < 1 × 10−360) in hypoxia versus normoxia and decreased (right, royal blue, P < 1 × 10−360) in METTL4-si versus scramble-si under hypoxia. We next applied the function of ‘correlating features’ occurrence by regions’ from EnrichShuf to calculate the presence of 6mA within genomic regions (e.g. accessible chromatin regions and H3K4me1) when these regions were occupied by other histone marks (e.g. H3K4me3 and H3K27Ac) (see the Materials and methods). The results showed that when H3K4me1 co-existed with 6mA, there was a positive correlation between H3K4me1 and various types of regions/histone marks, such as accessible chromatin regions and H3K27Ac (Figure 5B). In accessible chromatin regions co-existing with 6mA, there was a positive correlation between accessible chromatin regions and various types of histone marks, including H3K4me1, H3K4me3 and H3K27Ac (Figure 5B). This result suggested that 6mA was significantly correlated with gene activation and the induction of chromatin accessibility induced by hypoxia (Figure 5B). To further examine the mechanistic possibility between 6mA and gene activation under hypoxia, we investigated the correlation of 6mA with super-enhancers that played an important role in the activation of oncogenes in cancer (44). Using H3K27Ac signals under each condition as input, we identified 1017 predicted super-enhancers under normoxia and 1269 under hypoxia (Supplementary Figure S6D) by using the ROSE algorithm (see the Materials and methods). After analyzing the correlation between 6mA and predicted super-enhancers under hypoxia, we observed the positive correlations at distances of 0, 0–3, 3–10, 10–20, 20–30 and 30–40 kb (Figure 5C). We also observed significantly higher signals of H3K27Ac in predicted super-enhancers occupied by 6mA induced by hypoxia (Figure 5D). To understand the interplay between 6mA, accessible chromatin regions, H3K4me1 and super-enhancers, we divided accessible chromatin regions and H3K4me1 into different groups, within or outside super-enhancers, followed by correlating them with 6mA (Figure 5E and F). The results showed a significantly higher correlation of co-existence between 6mA and each type of region/mark, including accessible chromatin regions and H3K4me1, within super-enhancers compared with regions outside of super-enhancers (Figure 5E). In addition, we correlated 6mA across various ranges of distances from 1 to 1000 kb with regions within or outside of super-enhancers and identified a significantly higher correlation within super-enhancers versus regions outside of super-enhancers (Figure 5F). This result suggested that the association between 6mA, accessible chromatin regions and H3K4me1 played an important role within super-enhancers (Figure 5E and F). We next compared the super-enhancers under normoxia versus hypoxia and identified 910 shared super-enhancers under hypoxia and normoxia and 359 hypoxia-specific super-enhancers (Supplementary Figure S6E). To examine whether the formation of super-enhancers was associated with changes in histone modification, we correlated different histone modification occupancy with various types of super-enhancers, including shared super-enhancers under hypoxia and normoxia, and hypoxia-specific super-enhancers (Supplementary Figure S6F). The results showed that the induction of H3K27Ac was correlated with hypoxia-specific super-enhancers (Supplementary Figure S6F). However, there was no difference in either induction or decrease of H3K4me1/H3K4me3 between shared super-enhancers under hypoxia and normoxia and hypoxia-specific super-enhancers (Supplementary Figure S6F). This result indicated that the induction of H3K27Ac by hypoxia was correlated with the formation of super-enhancers, but not H3K4me1 and H3K4me3 (Supplementary Figure S6F). We next correlated differential chromatin accessibility with H3K4me1, H3K4me3 and H3K27Ac (Figure 5G). The results showed that chromatin accessibility was induced by METTL4 under hypoxia within these histone marks (Figure 5G), suggesting that the induction of the chromatin accessibility by METTL4 under hypoxia was associated with these histone modifications contributing to gene activation. In addition to analyzing the correlation between 6mA and various genomic regions/marks, we examined whether these correlations are specific to certain chromosomes. Previous results identified a significant enrichment of 6mA peaks in Alkbh1 knockout embryonic stem cells and young full-length L1 elements on X chromosomes compared with other chromosomes (2). Based on the previous findings and our results showing that 6mA co-existed with accessible chromatin regions, H3K4me1 and super-enhancers (Figures 3A, 5A and C), we performed a similar analysis using randomized 6mA sites within their respective chromosomes to account for the distribution of 6mA across the genome (see the Materials and methods). The results showed that 6mA positively co-existed or was surrounded (0–3 kb) by these genomic features across each chromosome (Supplementary Figure S7A–C), which were consistent with our results (Figures 3A, 5A and C). To understand the chromosomal distribution of these genomic features, we performed chromosome enrichment analysis (Supplementary Figure S7D). Furthermore, to determine whether these genomic features co-exist with or are surrounded by 6mA within specific chromosomes, we correlated 6mA with these features across each chromosome (Supplementary Figure S7E). Through these analyses, we identified chromosomes with significant positive enrichment of various genomic features and their correlation with 6mA across each chromosome (Supplementary Figure S7F). Interestingly, not all chromosomes showed a significant correlation between each type of genomic region and 6mA, whether they co-existed with or were surrounded by 6mA (Supplementary Figure S7F). Next, we calculated the chromosomal correlation between different chromosomes and different positively enriched genomic features. The results showed that the chromosomal enrichment of 6mA was correlated with H3K4me1, whereas super-enhancers were correlated with accessible chromatin regions (Supplementary Figure S7G). Furthermore, we performed chromosomal correlation between different chromosomes and different positively enriched genomic features co-existing with 6mA or surrounded by 6mA. The results showed that there was a positive correlation between accessible chromatin regions co-existing with 6mA, super-enhancers co-existing with 6mA and super-enhancers surrounded by 6mA (Supplementary Figure S7H). By comparing the chromosomal correlations with different genomic features obtained from the above analyses (Supplementary Figure S7F), we observed that the enrichment of 6mA within each chromosome did not fully correspond to the correlation between 6mA and various genomic features, such as regions which co-existed with 6mA or were surrounded by 6mA (Supplementary Figure S7I). The results showed that, within each chromosome, there is a significant correlation between 6mA enrichment and accessible chromatin regions co-existing with 6mA (Supplementary Figure S7I). 6mA-related functional enhancers are associated with hypoxia-inducible factor-1α (HIF-1α) binding and induction of H3K4me3 under hypoxia To test the linkage between 6mAs and super-enhancers, our results showed that 6mA peaks located within super-enhancers were induced by METTL4 under hypoxia (Figure 6A). To further examine the interplay between 6mA, super-enhancers and gene expression, we overlaid the transcripts up-regulated by METTL4 under hypoxia and the transcripts whose promoters were occupied by 6mA-enriched super-enhancers (Figure 6B). The results showed that 185 transcripts were identified through overlapping of these two criteria (Figure 6B). By comparing 185 identified transcripts with the other 11 308 transcripts whose promoters were occupied by 6mA-enriched super-enhancers, we found that these 185 transcripts significantly correlated with the promoter's decreased accessibility in METTL4-si under hypoxia (Figure 6C). 6mA was associated with functional enhancers and accompanied by HIF-1α binding and the induction of H3K4me3 under hypoxia. (A) The bar graph shows the correlation between differential 6mA signals and super-enhancers. The results showed that within super-enhancers, 6mA signals were increased (left, brown, P = 3.2 × 10−9) in hypoxia versus normoxia and decreased (right, royal blue, P = 4.16 × 10−6) in METTL4-si versus scramble-si under hypoxia. (B) Venn diagram showing the distribution of transcripts up-regulated by METTL4 under hypoxia and those occupied by regions co-existing with 6mA and super-enhancers on promoters. The results showed that there were 185 transcripts present in both categories. (C) The bar graph shows that, in the category of transcripts whose promoters were located in 6mA-occupied super-enhancers and up-regulated by METTL4 under hypoxic conditions, chromatin accessibility was decreased (right, blue, P = 2.54 × 10−4) in METTL4-si versus scramble-si under hypoxia. (D) The pie plot shows that when associating accessible chromatin regions on the promoter with transcripts, there were 7096 accessible chromatin regions (12.51%) on the promoter driving expressed transcripts and 49 613 accessible chromatin regions (87.49%) on the promoter associated with non-expressed transcripts. (E) The pie plot shows that, when associating accessible chromatin regions outside of the promoter predicted from the ABC model, 4813 accessible chromatin regions (3.57%) were predicted to be functional enhancers, whereas 130 145 accessible chromatin regions (96.43%) were categorized as unknown. (F) The bar plots show that functional promoters were correlated with hypoxia-induced H3K4me3 (P < 1 × 10−360), whereas functional enhancers were correlated with hypoxia-induced H3K4me1 (P = 1.99 × 10−22), H3K4me3 (P = 6.35 × 10−43) and H3K27Ac (P = 3.33 × 10−86). (G) The bar plots show that accessible chromatin regions induced under hypoxia (left) were correlated with hypoxia-induced H3K4me1 (P = 5.7 × 10−4) and H3K27Ac (P = 2.31 × 10−8), whereas accessible chromatin regions reduced in METTL4-si versus scramble-si under hypoxia (right) were correlated with hypoxia-induced H3K4me3 (P = 2.63 × 10−2). (H) The bar plots show the correlation between the binding of HIF-1α and hypoxia-induced H3K4me3 (P = 4.21 × 10−21) and H3K27Ac (P = 2.87 × 10−7). (I) The bar plots show that the binding of HIF-1α was correlated with accessible chromatin regions increased under hypoxia (P = 3.77 × 10−10) and accessible chromatin regions decreased in METTL4-si versus scramble-si under hypoxia (P = 3.23 × 10−5). (J) The bar plots show that 6mA-related regions were correlated with the binding of HIF-1α (P = 3.68 × 10−3), but not with KDM5B-regulated regions. Although the interplay between 6mA, accessible chromatin regions and H3K4me1 contributes to gene activation (Figures 5 and 6A–C), super-enhancers are relatively rare in the genome. While the majority of 6mAs or histone marks were located outside of super-enhancers (Supplementary Figure S8A), it remains to be determined whether this interplay also has a role outside of super-enhancers. Our results showed that genomic regions that were occupied by H3K4me1 and 6mA were associated with accessible chromatin regions and H3K27Ac whether they were inside or outside super-enhancers (Supplementary Figure S8B). Genomic regions with both chromatin accessibility and 6mAs displayed increased levels of H3K4me1, H3K4me3 and H3K27Ac (Supplementary Figure S8B). 6mAs that co-existed with chromatin accessibility were induced by METTL4 under hypoxia outside of super-enhancers (Supplementary Figure S8C), consistent with previous observations (Figure 4C). The accessible chromatin regions that co-existed with 6mAs or H3K4me1 were also induced by METTL4 under hypoxia outside of super-enhancers (Supplementary Figure S8D). Previous analysis showed that a subset of accessible chromatin regions exhibited positive or negative enrichment of 6mAs in their vicinity (Figure 3D). We correlated positive or negative 6mA-enriched accessible chromatin regions with super-enhancers and various histone marks, and discovered that accessible chromatin regions intersecting with super-enhancers and H3K4me1 were positively enriched with 6mA peaks (Supplementary Figure S8E). Although we observed that 6mA was associated with super-enhancers and H3K4me1, a mark for enhancers (Figure 6A–C; Supplementary Figure S7), more detailed mechanisms underlying this relationship were explored. To address this issue, we applied the previously established ABC model to hypoxic ATAC-seq, H3K27Ac ChIP-seq, RNA-seq and publicly available average Hi-C data to predict potential enhancers (Supplementary Figure S9A) (39). The results showed that, among the total 20 544 expressed transcripts, 12 262 transcripts were associated with varying numbers of predicted enhancers (Supplementary Figure S9A). We categorized accessible chromatin regions into different groups (i.e. promoter, gene and intergenic region) based on their annotation locations (Supplementary Figure S9B). Next, we identified functional promoters by determining which expressed transcripts were associated with accessible chromatin regions at the promoter (Figure 6D). Additionally, we identified functional enhancers by overlapping the accessible chromatin region outside of the promoter with the predicted enhancer from the ABC model (Figure 6E). We correlated each type of histone modification (i.e. H3K4me1, H3K4me3 and H3K27Ac) with functional promoters or enhancers (Figure 6F). The results showed that H3K4me3 was increased under hypoxia within functional promoters, whereas H3K4me1, H3K4me3 and H3K27Ac were increased under hypoxia within functional enhancers (Figure 6F). Additionally, we correlated differential chromatin accessibility with functional promoters or enhancers (Supplementary Figure S9C). The results showed that chromatin accessibility within functional promoters was solely decreased by METTL4-si, whereas chromatin accessibility within functional enhancers was increased by hypoxia via METTL4 (Supplementary Figure S9C). Next, we correlated various types of accessible chromatin regions (e.g. increased under hypoxia and decreased in METTL4-si versus scramble-si under hypoxia) with each type of histone modification (i.e. H3K4me1, H3K4me3 and H3K27Ac) inside the functional promoters or enhancers (Figure 6G; Supplementary Figure S9D). Inside the functional enhancers, increased chromatin accessibility under hypoxia was accompanied by an increase in H3K4me1 and H3K27Ac (Figure 6G). On the other hand, hypoxia-induced H3K4me3 was observed in regions where chromatin accessibility decreased in METTL4-si versus scramble-si under hypoxia (Figure 6G), a pattern also seen within functional promoters (Supplementary Figure S9D). This result indicates the unexpected role of H3K4me3 in super-enhancers regulated by METTL4 under hypoxia. Previous studies have shown that the induction of H3K4me3 plays an important role at active enhancers in pre-TCR (T-cell receptor) signaling during T-lymphocyte differentiation (45). In addition, loss of RACK7 that recruits the H3K4me3 demethylase KDM5C could result in overactivation of enhancers through deposition of H3K4me3, supporting the role of H3K4me3 in enhancer activation (46). Under hypoxia, the induction of H3K4me3 in promoters is essential for gene activation by influencing chromatin structure through the stabilization of HIF-1α (47) and the loss of function of the oxygen-sensing histone demethylase KDM5A (48). However, the mechanism by which H3K4me3 is induced at enhancers remains unknown. We first correlated previous HIF-1α ChIP-seq data (9) with various types of histone modification, and found that the binding of HIF-1α was correlated with the hypoxia-induced H3K4me3 and H3K27Ac inside the functional enhancers (Figure 6H). We next correlated HIF-1α binding with differential chromatin accessibility, and the results showed that when regions were bound by HIF-1α, chromatin accessibility was increased under hypoxia via METTL4 inside the functional enhancers (Figure 6I). In addition to analyzing HIF-1α correlations, we also obtained KDM5A and KDM5B data from the ReMap database (49). We identified 3298 accessible chromatin regions that were regulated by both KDM5A and KDM5B, whereas 4006 regions were regulated by KDM5B alone within functional promoters and enhancers (Supplementary Figure S9E). After comparing their distribution, we found that 55.29% of the accessible chromatin regions regulated solely by KDM5B were located within functional enhancers, while the majority (94.15%) of regions regulated by both KDM5A and KDM5B were located within functional promoters (Supplementary Figure S9F). We performed a Venn diagram analysis between HIF-1α-binding regions and KDM5B-regulated regions inside the functional enhancers (Supplementary Figure S9G), as our focus is on the correlation between 6mA and enhancers. The results showed that there were 923 accessible chromatin regions only bound by HIF-1α and 1254 regions regulated by KDM5B alone (Supplementary Figure S9G). We then correlated 6mA-related regions (those co-existing with 6mA and positively enriched with 6mA) identified in previous analyses (Figure 3A and D) with various types of accessible chromatin regions (i.e. regions bound with HIF-1α and regions regulated by KDM5B) (Figure 6J). The results showed that 6mA was correlated with the regions bound by HIF-1α compared with the KDM5B-regulated regions (Figure 6J). METTL4-mediated gene activation under hypoxia is correlated with functional enhancers that are accompanied by 6mA, HIF-1α and the induction of H3K4me3 To further confirm the enhancer–promoter interaction, we downloaded the validated enhancer dataset from the Endb database, and identified 419 human experimentally validated enhancers in the hg38 genome version. We further compared the Endb-validated enhancers with ABC model-derived enhancers using various strategies (Supplementary Figure S10A–C). First, we found that 47.3% of Endb-validated enhancers overlapped with accessible chromatin regions in FaDu cells under hypoxia (Supplementary Figure S10A). Second, 12.6% of these overlapping regions could also be identified as enhancers by the ABC model (Supplementary Figure S10B). Finally, 68% of these overlapping enhancers shared the same regulated transcripts, indicating a 68% accuracy in our predicted enhancers (Supplementary Figure S10C), which aligns closely with a 70% precision reported from the ABC model (39). Finally, we closely examined the data to determine whether any validated enhancers regulate MALAT1 (Metastasis Associated Lung Adenocarcinoma Transcript 1) and discovered that it was regulated by an unreported super-enhancer through the hypoxia/METTL4/6mA axis (Figure 7E). The interplay between METTL4, 6mA and chromatin accessibility, along with HIF-1α binding and the induction of H3K4me3, occurs in functional enhancers under hypoxia. (A) By correlating differential chromatin accessibility with transcripts activated by METTL4 under hypoxia and predicted enhancer information, we identified 286 regions with decreased chromatin accessibility in METTL4-si versus scramble-si under hypoxia within predicted enhancers. (B) Gene Ontology analysis of transcripts which were activated by METTL4 under hypoxia and occupied with METTL4-si decreased chromatin accessibility regions on their predicted enhancers under hypoxia. The results showed that a large portion of genes belonged to the category of tRNA processing, wound healing and tissue migration. (C) Venn diagram showed the overlapping of METTL4-regulated enhancers with HIF-1α-binding sites, hypoxia-induced H3K4me3 regions and 6mA-related regions, identifying 31 accessible chromatin regions containing all three features. (D) The heatmap analysis shows that 58% of the enhancers (18 genes) were located within super-enhancers, whereas 48% of the enhancers (13 genes) were situated outside of super-enhancers. (E) The genome browser view demonstrated that the gene MALAT1 is situated within a super-enhancer, which was identified by high levels of H3K27Ac. This area displayed a high level of H3K4me1 and 6mA peaks upstream of MALAT1, along with a relative decrease in chromatin accessibility in METTL4-si within the super-enhancer. The ABC model predicted that the enhancer with 6mA peaks interacted with MALAT1. A high level of H3K4me3 around MALAT1 indicated that it was highly transcribed, and this was associated with HIF-1α binding to H3K4me3. A zoomed-in view of the upstream region of MALAT1 reveals SFB and SRB 6mA peaks, with conserved stranded signals (red, forward signals; blue, reverse signals) across three replicates. (F) The model depicted a comprehensive analysis of 6mA in relation to defined genomic regions and expression traits in enhancers. It indicated that under hypoxia, 6mA levels were increased by METTL4 and co-existed with H3K4me1 and accessible chromatin regions, where chromatin accessibility was increased. This event was associated with HIF-1α binding and the induction of H3K4me3 surrounding the 6mA site/H3K4me1. After observing that 6mA co-existed with H3K4me1 and was associated with HIF-1α binding and H3K4me3 induction under hypoxia, we further investigated the relationship between 6mA, H3K4me1 and H3K4me3. This was done by analyzing the peaks of H3K4me1 and H3K4me3 centered around 6mA sites linked to functional enhancers (Supplementary Figure S11A). The results showed that 6mA co-localized with H3K4me1, whereas H3K4me3 surrounded these 6mA sites (Supplementary Figure S11A). Furthermore, we linked previously identified transcripts up-regulated by METTL4 under hypoxia (Figure 4E) with differential chromatin accessibility inside the functional enhancers (Figure 7A). The results showed that predominant accessible chromatin regions were decreased in METTL4-si versus scramble-si under hypoxia inside the functional enhancers (Figure 7A). Gene Ontology analysis of transcripts whose enhancers displayed decreased chromatin accessibility in METTL4-si under hypoxia showed that most of these genes were involved in tRNA processing, wound healing and tissue migration, and they were induced by METTL4 under hypoxia (Figure 7B). We then associated these METTL4-regulated enhancers with HIF-1α, hypoxia-induced H3K4me3 and 6mA-related regions, and identified 31 accessible chromatin regions common to all three categories (Figure 7C). A heatmap analysis showed the gene expression pattern of these 31 genes associated with enhancers related to 6mA, bound by HIF-1α and linked to the induction of H3K4me3 under hypoxia (Figure 7D). The results also showed that more than half of the enhancers (n = 18, 58%) were located within super-enhancers (Figure 7D). As an example, the enhancer of MALAT1, a long non-coding RNA crucial for tumor progression under hypoxia, contained a super-enhancer and exhibited a high level of H3K4me1 and 6mA peaks (Figure 7E). The gene body of MALAT1 also exhibited a significant level of H3K4me3 (Figure 7E). Accessible chromatin regions within super-enhancers and around MALAT1 showed a significantly decreased chromatin accessibility in METTL4-si compared with scramble-si under hypoxia (Figure 7E). In addition, another example, F3 (Coagulation Factor III, also known as Tissue Factor), showed the same pattern inside the super-enhancer (Supplementary Figure S11B). An example of outside of the super-enhancer, KLHL36 (Kelch Like Family Member 36), was shown that could be regulated by a 6mA-related enhancer bound by HIF-1α (Supplementary Figure S11C). In addition to examining the enhancer and its associated genes under hypoxia, we also provided detailed analyses of H3K4me1 and H3K4me3 under both normoxia and hypoxia (Supplementary Figure S11D). The results showed that 6mA co-localized with H3K4me1 in these enhancers, accompanied by the surrounding localization of HIF-1α and hypoxia-induced H3K4me3 (Supplementary Figure S11D). This information suggested that although 6mA co-localized with H3K4me1, the level of H3K4me1 was not induced under hypoxia (Supplementary Figure S11D). Instead, this phenomenon was associated with HIF-1α binding, and its related H3K4me3 levels were induced under hypoxia (Supplementary Figure S11D). To further validate the correlation between 6mA on the enhancer and its target gene, we used MALAT1 as our model. We designed two sgRNAs targeting two 6mA sites on the enhancer of MALAT1 for CRISPRi experiments using dCas9-KRAB, following previously established methods (40) (Supplementary Figure S11E). The results showed that the co-transfection of dCas9-KRAB with each sgRNA abolished the induction of MALAT1 expression under hypoxia, thereby validating the enhancer–promoter interaction between 6mA and MALAT1 (Supplementary Figure S11F). A model is proposed in which METTL4 deposits 6mA upstream of the TSS within accessible chromatin regions co-localized with H3K4me1 inside the enhancers (Figure 7F). This process is associated with HIF-1α binding and the subsequent induction of H3K4me3, leading to the activation of gene expression under hypoxia (Figure 7F). Discussion The role of 6mA in gene expression, especially for mammalian cells, remains to be further elucidated in spite of numerous reports (1–10). In addition, its interplay with chromatin accessibility/histone marks remains elusive. In this report, we show that 6mA signals correlate with regions of chromatin accessibility and the H3K4me1 mark, suggesting its functional role in regulating hypoxia-induced gene expression. The low abundance of 6mA in mammalian cells means that it constantly evades detection by different methods, especially antibody-dependent approaches. This limitation is caused by the low complexity of the constructed library, a consequence of the extensive PCR cycles required during library preparation (24,25). In this report, we modified the ChIP-exo-seq method by incorporating a step to repair UV-damaged DNA resulting from cross-linking (Figure 1A). This modification improved the complexity of the ChIP-exo library generated from human hypoxic gDNA by reducing the number of PCR cycles, allowing the identification of 6mA sites by MACE (Figure 1). A WGA control was also employed to filter out potential false positives (Figure 1). We further evaluated the impact of the repair process added in 6mA-ChIP-exo-5.1 by comparing it with the previous 6mA-ChIP-exo-5.0 (7,9). The results showed that 6mA-ChIP-exo-5.1 exhibited higher library complexity (Supplementary Figure S12A), a lower duplication rate (Supplementary Figure S12B) and greater genome coverage (Supplementary Figure S12C) compared with the original 6mA-ChIP-exo-5.0 method, indicating that adding the repair process into the original 6mA-ChIP-exo-5.0 method can have a significant improvement in sequencing data quality (Supplementary Figure S12). Bioinformatics methods including χ2, binomial and Fisher’s exact tests were used to analyze 6mA signals (3,26,27). To broaden the scope of analysis not limited to 6mA signals, we developed EnrichShuf, an R package designed for comprehensive analysis of the correlation between 6mA and other genomic information (e.g. chromatin accessibility, histone marks) (Figure 2). EnrichShuf applies a randomized empirical model combined with a z-test to ascertain the functional relevance of 6mA signals with chromatin accessibility/histone marks. With EnrichShuf, we show that hypoxia-induced 6mA signals are positively enriched within or around accessible chromatin regions (Figure 2). 6mA-containing accessible chromatin regions are located at the promoter of cancer-related genes and upstream of the TSS (Figure 3). Furthermore, the interplay between hypoxia-induced 6mA signals and these accessible chromatin regions has a crucial role in gene activation mediated by hypoxia-induced METTL4 (Figure 4). We further discovered a positive correlation of the co-existence of 6mA and H3K4me1 (Figure 5A). H3K4me1 has been reported to be a histone mark associated with chromatin long-range interaction that involves highly transcribed genes with long-range interactions (known as super-enhancers) (50). Analysis of the H3K27Ac mark shows that a positive correlation could be found between 6mA and super-enhancers (Figure 5D). Furthermore, HIF-1α binding activated H3K4me3 that surrounds the 6mA signals located in the super-enhancers, consistent with reports that a broad H3K4me3 peak contributes to super-enhancer function (44). A model is proposed where METTL4 deposits 6mA upstream of the TSS-surrounding accessible chromatin regions and H3K4me1 within or outside super-enhancers followed by H3K4me3 induced by HIF-1α binding, thereby activating gene expression under hypoxia (Figure 6G). In this report, we applied the z-test as the main statistical method to determine the significance of our data (Figure 2A). To demonstrate the advantages of using the z-test, we compared different statistical methods used from published 6mA-related studies (Supplementary Figure S13A), including the χ2 test, Fisher’s exact test and the randomized empirical model (Useq). To evaluate the advantage and credibility of the z-test, we performed a detailed analysis of the correlation between the co-existence of 6mA and accessible chromatin regions (Supplementary Figure S13B–D). Firstly, we found that as the number of iterations to generate expected values increased, the actual density plot more closely resembled the normal distribution constructed from the expected values (Supplementary Figure S13B). This suggests that the z-test, which assumes normality, is appropriate for our data (Supplementary Figure S13B). Additionally, when the number of randomized iterations was set at 100, 1000 and 10 000, the normal distributions remained consistent across these different sample sizes (Supplementary Figure S13B). This consistency underscores the robustness of the z-test (Supplementary Figure S13B). We also performed the χ2 test and Fisher’s exact test on the same dataset (Supplementary Figure S13C). The χ2 test produced P-values ranging from 0.975 to 6.86 × 10−14, whereas Fisher’s exact test produced P-values ranging from 0.991 to 2.57 × 10−7 (Supplementary Figure S13C), indicating less stability compared with the z-test (P = 6.57 × 10−10) (Figure 3A; Supplementary Figure S13B). We further compared EnrichShuf with Useq and found that both methods produced similar observed values when accounting for chromosome distribution in EnrichShuf (Supplementary Figure S13D). However, EnrichShuf offers additional functionalities, such as performing statistical analysis for each chromosome individually (Supplementary Figure S7D–F), analyzing all chromosomes without considering chromosome distribution (Figure 3A), and calculating multiple associations (e.g. 0, 0–3, 3–10, 10–20, 20–30, 30–40 and 40–50 kb) simultaneously (Figure 3A; Supplementary Figure S7A). These capabilities provide a distinct advantage over Useq (Supplementary Figure S13D). This consistency confirms the reliability of EnrichShuf. In conclusion, we chose the empirical model with a z-test as our primary statistical method (EnrichShuf) since it provides a more stable and consistent measure of significance across different randomizations, making it a more reliable tool for our analysis. Super-enhancers are typically associated with highly transcribed genes, but it has been reported that they often coincide with destabilized DNA (51). It was caused by the disruption of RNA polymerase II elongation on enhancers, as indicated by the presence of ssDNA (52). 6mA has been reported that correlated with ssDNA and could influence DNA flexibility, thereby reducing nucleosome fuzziness (8,10,53). These studies provide insights into a significant correlation between hypoxic 6mA and super-enhancers. Rapid RNA polymerase II pause releasing is important in stimulating high expression levels of genes within super-enhancers (52). A recent study introduced a model termed ‘kiss and kick’, offering new insights into the importance of releasing the interaction between promoters and enhancers during transcription (54). It will be important to conduct further investigation to understand the potential role of 6mA within the ‘kiss and kick’ model under hypoxia from the functional relevance of 6mA in this report. Supplementary Material Acknowledgements We thank the National C6 RNAi Core Facility (Academia Sinica, Taipei, Taiwan) for assistance with sgRNA design and for providing the CRISPRi system. Author contributions: J.C.L. and K.J.W. conceived the project and designed the experiments, J.C.L. performed bioinformatics analysis, K.W.H. performed wet lab experiments, and J.C.L. and K.J.W. analyzed the data and wrote the manuscript. Contributor Information Joseph Chieh-Yu Lai, Graduate Institute of Biomedical Sciences, China Medical University, Taichung 406, Taiwan; Cancer Genome Research Center, Chang Gung Memorial Hospital at Linkou, Taoyuan 333, Taiwan. Kai-Wen Hsu, Institute of Translational Medicine & New Drug Development, China Medical University, Taichung 404, Taiwan. Kou-Juey Wu, Cancer Genome Research Center, Chang Gung Memorial Hospital at Linkou, Taoyuan 333, Taiwan. Data availability The sequencing data generated in this study are publicly available in the Gene Expression Omnibus (GEO) under the following accession numbers: GSE260869 (6mA-ChIP-exo-5.1), GSE260871 (ATAC-seq), GSE260872 (ChIP-seq) and GSE260873 (RNA-seq). HIF-1α ChIP-seq data from a previous study can be accessed under GSE171116 (9). Public Hi-C data for the ABC model are available from a previous study (39) (ftp://ftp.broadinstitute.org/outgoing/lincRNA/average_hic/average_hic.v2.191020.tar.gz). Code availability The source code for EnrichShuf is available on GitHub (https://github.com/JosephLaiC/EnrichShuf) and Zenodo (https://zenodo.org/records/13701277). The source code for the active-by-contact (ABC) model from a previous study (39) can be accessed on GitHub (https://github.com/broadinstitute/ABC-Enhancer-Gene-Prediction/tree/master). Supplementary data Supplementary Data are available at NAR Online. Funding The National Science and Technology Council [MOST 108–2321-B-182A-005, MOST 109–2326-B-182A-002, MOST 110–2326-B-182A-004, MOST 111–2326-B-182A-002, NSTC 112–2326-B-182A-004, and NSTC 113–2320-B-182A-007 to K.J.W.]; Chang Gung Memorial Hospital [OMRPG3I0011, OMRPG3I0012, OMRPG3I0013, NMRPG3J0671, NMRPG3J0672, NMRPG3J0673, CORPG3J0231, and CORPG3J0232 to K.J.W.]; the ‘Drug Development Center, China Medical University’ from The Featured Areas Research Center Program within the framework of the Higher Education Sprout Project (Ministry of Education, Taiwan) to K.W.H. Conflict of interest statement. None declared. References Associated Data Supplementary Materials Data Availability Statement The sequencing data generated in this study are publicly available in the Gene Expression Omnibus (GEO) under the following accession numbers: GSE260869 (6mA-ChIP-exo-5.1), GSE260871 (ATAC-seq), GSE260872 (ChIP-seq) and GSE260873 (RNA-seq). HIF-1α ChIP-seq data from a previous study can be accessed under GSE171116 (9). Public Hi-C data for the ABC model are available from a previous study (39) (ftp://ftp.broadinstitute.org/outgoing/lincRNA/average_hic/average_hic.v2.191020.tar.gz).