Multi-omics analysis reveals the crosstalk of epigenetic regulatory networks in cutaneous squamous cell carcinoma progression

Abstract

Background

Cutaneous squamous cell carcinoma (cSCC) is the second most common form of non-melanoma skin cancer (NMSC), with a steadily increasing global incidence, especially in populations with prolonged ultraviolet (UV) exposure. Advanced or metastatic cSCC carries a poor prognosis, while the application of targeted therapies and immunotherapies remains in the exploratory stage, due to limited understanding of the underlying mechanisms of cSCC occurrence and progression. Epigenetic dysregulation plays important roles in the progression of cSCC. However, how multi-dimensional regulatory networks reshape the epigenetic landscape and contribute to dysregulated gene expression in cSCC remains unclear.

Methods

In this study, we performed parallel RNA m⁶A sequencing, 850K DNA methylation arrays, whole transcriptome sequencing, and ATAC-seq chromatin accessibility profiling on samples from normal skin, actinic keratosis (AK), and cSCC. We analyzed the regulatory networks and pathways of epigenetic modifications on gene expression. We further explored the crosstalk of epigenetic regulatory networks by correlation analysis. By integrating single-cell RNA-seq data, we identified epigenetically upregulated candidate genes and confirmed their expression and functions with experimental methods.

Results

Our integrated multi-omics analysis provides a comprehensive and dynamic epigenetic map of cSCC progression. Further analysis revealed that DNA methylation and m⁶A modification jointly regulate gene expression through independent and synergistic ways. The identified epigenetically upregulated candidate genes IDO1, IFI6, and OAS2 were validated to be overexpressed in cSCC tissues and cell lines, and functional assays confirmed their potential key roles in regulating the processes of cell proliferation, migration and invasion in cSCC.

Conclusions

By integrating multi-omics data, this study systematically highlights the multi-layered epigenetic alterations and regulatory mechanisms involved in cSCC development. This multi-stage, multi-omics, and multi-resolution integrated analysis provides a theoretical basis and new insights for future personalized treatment strategies for cSCC.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-025-07262-z.

Background

Cutaneous squamous cell carcinoma (cSCC) is among the most common malignancies of keratinocyte-derived non-melanoma skin cancers (NMSCs) [1]. cSCC accounts for approximately 20% of NMSCs and is responsible for most NMSC-related deaths [2]. Metastatic cSCC shows significantly increased mortality rates, comparable to melanoma skin cancer (MSC). The three-year mortality rate for metastatic cSCC reaches 46% [3, 4]. Solar ultraviolet (UV) radiation is a major risk factor for cSCC development and induces genetic and epigenetic alterations in keratinocytes [5]. cSCC is believed to develop from UV-induced precursor lesions, known as actinic keratoses (AK), a progression supported by histological and molecular evidence [6]. As the epithelial-derived malignancy with the highest somatic mutational burden, cSCC harbors an average of approximately 50 mutations per megabase (Mb) of DNA [7, 8]. An early key event in cSCC pathogenesis is the UV-induced mutation and inactivation of the TP53 tumor suppressor gene, which is commonly observed in both AK and cSCC [9]. Additional genetic alterations in driver genes, such as NOTCH1-3, CDKN2A, PIK3CA, HRAS, and EGFR, influence signaling pathways that regulate the cell cycle, apoptosis, differentiation, and cell survival in cSCC pathogenesis [10]. However, mutations in these genes are also frequently found in microscopic clusters of keratinocytes within sun-exposed normal human skin [11]. This underscores the need to identify additional alterations that drive the progression from premalignant AK lesions to invasive cSCC.

Recent advances in genomics integrated with multi-omics data have propelled cancer research forward [12]. Elucidating the roles of genetic modifications and epigenetic regulation in the complex pathophysiology of cSCC progression could facilitate novel therapeutic interventions. Epigenetic modifications, including CpG island DNA methylation, RNA methylation, and histone methylation/acetylation, regulate gene transcription by altering chromatin accessibility, either individually or synergistically [13]. DNA methylation is among the most critical epigenetic regulatory mechanisms in various cancers. In cutaneous melanoma, specific promoter hypomethylation and intragenic hypermethylation have been shown to be associated with tumor aggressiveness [14]. In cSCC, Hervás-Marín et al. identified distinct epigenetic features of low-risk and high-risk cSCC using genome-wide DNA methylation profiling [15]. Li et al. reported that UVB exposure induced Inhibitor of DNA binding 4 (ID4) DNA methylation [16]. However, the mechanisms by which changes in DNA methylation drive cSCC development, especially the progression from normal skin to AK to cSCC, still require further investigation. N6-methyladenosine (m⁶A) is the most abundant and dynamically reversible post-transcriptional mRNA modification in mammals [17]. Dysregulation of global m⁶A levels and their associated regulators plays a critical role in cancer progression [18, 19]. In melanoma, m⁶A modifications have been implicated in immunoregulation of the tumor microenvironment (TME) and the enhancement of tumor immunogenicity [20, 21]. In cSCC, the m⁶A methyltransferase METTL3 has been identified as a key regulator of tumorigenesis, with elevated METTL3 expression promoting the proliferation of cutaneous squamous cells through increased ∆Np63 m⁶A methylation [22]. In a cohort study comparing normal skin and cSCC samples, immunofluorescence results demonstrated that UV exposure induces m⁶A modification. The level of m⁶A modification inversely correlates with the severity of skin photodamage, consistent with earlier in vitro and murine studies [23, 24]. However, m⁶A RNA methylation in cSCC has not been studied as extensively as in other related squamous cell carcinomas.

Recent researches have highlighted the importance of crosstalk among epigenetic modifications for the precise and synchronized regulation of gene expression [25]. A previous study revealed a regulatory mechanism by which RNA m⁶A modification together with DNA demethylation, modulates chromatin accessibility and gene transcription [26]. However, the roles of interactions among different epigenetic modifications in most biological processes, particularly cancer development, remain largely unclear. The pathogenesis of cSCC involves dynamic interactions among multi-stage molecular networks. Recent studies have shown that UV-induced malignant transformation of keratinocyte clones occurs through the accumulation of dysregulated key signaling pathways, jointly driven by genetic alterations and epigenetic remodeling. However, single-omics approaches often fail to systematically elucidate these complex mechanisms due to their limited perspective.

In this study, we conducted parallel analyses, including RNA m⁶A sequencing, 850K DNA methylation arrays, whole transcriptome sequencing, single-cell transcriptome profiling, and ATAC-seq chromatin accessibility profiling of samples from AK, cSCC, and normal skin. Through integrative multi-omics analyses, we characterized the epigenetic profiles across distinct stages of cSCC. Additionally, we identified potential key driver factors regulated by epigenetic modification during cSCC progression, which may serve as novel therapeutic targets for clinical treatments.

Materials and methods

cSCC and AK patient samples

Tissue samples of cSCCs (n = 5), AKs (n = 5) and normal skin (n = 10) were collected during surgical treatment at the Dermatology Department of the First Affiliated Hospital of Kunming Medical University (Yunnan, China). All samples were derived from the UV-exposed areas from immunocompetent patients, and excluded samples associated with viral infections, chemical carcinogens, ionizing radiation, and certain skin diseases such as scarring, chronic ulcers and familial genetic syndromes. And none of these patients had received any treatment before surgery. During surgery, most of the resected tissue was immediately put in cryogenic vials and snap-frozen in liquid nitrogen and stored at −80 °C for multi-omics sequencing. Some tissues were fixed in neutral-buffered formalin, embedded in paraffin, and used for histopathological diagnosis by two independent pathologists. This study was reviewed and approved by the Ethics Committee of the First Affiliated Hospital of Kunming Medical University [Approval Number (2020)-L-29], and written informed consent was obtained from all patients.

m⁶A-seq and data analysis

Total RNA was isolated and purified using the TRIzol reagent (Invitrogen, USA) following the manufacturer’s procedure. The RNA quantity and purity of each sample was assessed using the NanoDrop ND-1000 (NanoDrop, USA). The RNA integrity was assessed by Bioanalyzer 2100 (Agilent, CA, USA) with RNA Integrity Number (RIN) > 7.0. Poly(A) RNA is purified from 30 μg total RNA using Dynabeads Oligo (dT)25–61005 (Thermo Fisher, USA) using two rounds of purification. Then the poly(A) RNA was fragmented into small pieces using Magnesium RNA Fragmentation Module (NEB, USA) under 86 °C for 7 min. Then the cleaved RNA fragments were incubated for 2 h at 4 °C with m⁶A-specific antibody (Synaptic Systems, Germany) in IP buffer (50 mM Tris-HCl, 750 mM NaCl and 0.5% Igepal CA-630). Then the IP RNA was reverse-transcribed to create the cDNA by SuperScript™ II Reverse Transcriptase (Invitrogen, USA), which were next used to synthesise U-labeled second-stranded DNAs with E. coli DNA polymerase I (NEB, USA), RNase H (NEB, USA) and dUTP Solution (Thermo Fisher, USA). An A-base is then added to the blunt ends of each strand, preparing them for ligation to the indexed adapters. Each adapter contains a T-base overhang for ligating the adapter to the A-tailed fragmented DNA. Single- or dual-index adapters are ligated to the fragments, and size selection was performed with AMPureXP beads. After the heat-labile UDG enzyme (NEB, USA) treatment of the U-labeled second-stranded DNAs, the ligated products are amplified with PCR by the following conditions: initial denaturation at 95 °C for 3 min; 8 cycles of denaturation at 98 °C for 15 sec, annealing at 60 °C for 15 sec, and extension at 72 °C for 30 sec, and then final extension at 72 °C for 5 min. The average insert size for the final cDNA library was 300 ± 50 bp. The library preparations were sequenced on an illumina Novaseq™ 6000 (LC-Bio Technology Co., Ltd., Hangzhou, China), platform with a paired-end read length of 150 bp (PE150) according to the standard protocols.

Quality control of the raw sequencing data was performed using FastQC (version 0.11.9), and adapter trimming and filtering of low-quality reads were done using Trim Galore (version 0.0.1) with the parameters: -q 25 –phred33 –length 40 -e 0.1 –stringency 3. After quality control, the remaining reads were mapped to the reference genome (hg19). Peaks and differential peaks were called using R package ‘exomePeak2’ (version 1.4.2).

Peaks with RPM.IP > 1 & RPM.input > 1 & log2FC > 1.6 were considered significant peaks, while differential peaks with FDR < 0.05 & |log2FC| > 0.6 & width < 5000 were regarded as significant differential peaks.

Gene m⁶A level calculation

For each m⁶ A peak, the peak ration was normalized used exomePeak2 by statistically comparing IP signal and Input control signal. The m⁶A level of each gene was defined as the overall methylation level across all m⁶A peaks associated with that gene. In subsequent inter-group differential correlation analyses, genes lacking significant m⁶ A peaks were assigned a randomly generated value well below the significance threshold. The specific calculation formula is as follows [27]:

m⁶A topological distribution analysis

To profile the topological distribution of m⁶A, we clustered protein-coding genes into five groups based on the distribution of m⁶A peaks called by exomePeak2 along genic regions. The longest transcript of each gene was selected from R package “EnsDb.Hsapiens.v75”. Transcripts with 5′UTRs length < 50 bp or 3′UTRs length < 100 bp or CDS length < 100 bp were filtered out. Peaks overlapping with 5′UTRs were defined as ‘UTR5E’, and analogously, 3′UTRs as ‘UTR3E’, CDS as ‘CDSE’ and those peaks overlapping with more than one region were defined as ‘MULTIE’. When more than 50% of the peak regions overlap with a specific region, the peak is considered as the region-specific region. For visualization, the 5′UTRs, 3′UTRs, and CDS regions of each gene was divided into 10, 20, and 20 windows, respectively. We normalized the peak density to the filtered genes’ windows using the “normalizeToMatrix” function with “target_ratio = 1, k = 1” from R package “EnrichedHeatmap”.

Clustering analysis

The m⁶A distribution patterns of genes were used for patient clustering. The Gower method is employed to cluster categorical variables. The probes of the top 5000 variances were used for PCA methylation analysis. The 1000 most variable genes as determined by median absolute deviation of the TPM were used for PCA analysis. PCA analyses were implemented by the R packages FactoMineR and factoextra.

850K DNA methylation array and data analysis

Genomic DNA was extracted using a genomic DNA extraction kit (Tiangen, Beijing, China). DNA was quantified by Quant-iT PicoGreen dsDNA Reagent (Invitrogen, USA) and the integrity was analyzed in a 1.3% agarose gel. The EZ DNA Metrology Kit (Zymo Research,CA, USA) was used to transform genomic DNA into bisulfite to keep the methylated cytosine unchanged and the unmethylated cytosine into uracil. DNA after bisulfite transformation was detected by using the Infinium Metallation- EPIC BeadChip Kit (illumina), and whether the CpG site was methylated was determined by genotyping results. DNA extraction, sample pretreatment and Illumina Infinium MethylationEPIC BeadChip detection were performed by Novogene Bioinformatic Technology Co., Ltd (Beijing, China).

Genome-wide methylation profiling was performed by Illumina Infinium Human Methylation 850K BeadChip. The data (IDAT files) were mainly analyzed using the R package “ChAMP”. The β values (range, 0–1) was used to represent the DNA methylation level, which were calculated from the intensity ratio of the methylated signals to the total (methylated and unmethylated) signals for each site. Then the methylation levels were compared between the samples of normal skin, AK and SCC. Differentially methylated regions (DMRs) were identified using DMRcate. During the execution of the cpg.annotate function, the fdr was set to 1 to include all CpG probes in the subsequent DMR analysis for exploratory purposes. DMRs were defined as regions with p.adjust < 0.05 and |Δβ| > 0.05. Those with Δβ > 0.05 were classified as hypermethylated DMRs (hyperDMRs), while those with Δβ < −0.05 were classified as hypomethylated DMRs (hypoDMRs). Gene annotation information was obtained from the gencode.v19.annotation.gtf file. Promoters were defined as regions within TSS ±1kb regions. DMRs overlapping with promoters were defined as promoter-DMR and others overlapped with gene bodies but not promoter were classified as genebody-DMR.

Gene promoter methylation level calculation

Promoters were defined as regions within TSS ±1kb regions. The average methylation level of probes falling in the promoter range was calculated as gene promoter methylation level, and genes that drop fewer than 3 probes in the promoter region are excluded from analysis.

Gene expression analysis

RNA-seq data preprocessing and analysis were performed using standardized pipelines: Sequencing reads were aligned to the human GRCh38 reference genome using the STAR aligner. Following quality control, uniquely mapped reads were retained to generate the raw gene count matrix. This raw count matrix underwent filtration to remove genes with total counts below 10. The raw count matrix was normalized and transformed into a TPM matrix based on gene position information from the Ensembl database. Genes with TPM > 1 in at least one-fifth of the samples were retained as expressed genes. Differential expression analysis was subsequently conducted using the ‘DESeq2’ R package with default parameters. Pairwise comparisons used earlier-stage samples as the reference group by default, with log₂ fold changes calculated. Genes meeting the differential expression threshold (|log2FoldChange| > 0.6 and p.adjust < 0.05) were defined as differentially expressed genes. Gene identifier conversions between Ensembl ID, HUGO symbol, and Entrez ID were implemented using the ‘biomaRt’ package querying the Ensembl database.

Function enrichment analysis

Function enrichment analysis was performed by R package ‘clusterProfiler’ (version 4.2.2). Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome Pathway enrichment were calculated. with an adjusted p-value < 0.05 as the significance cutoff. All filtered precoding genes or genes captured by m⁶A input data were considered as universal genes.

Odds ratio calculation

The ‘oddsratio.wald’ function from the ‘epitools’ package was employed to calculate odds ratios (ORs) using the unconditional maximum likelihood estimation (Wald) method. This computed the ratio of the number of differentially upregulated/downregulated DNA methylation events in the m⁶A methylation gain/loss group versus the number in the group without m⁶A methylation gain or loss. Confidence intervals were calculated via the normal approximation (Wald) method. OR > 1 indicates that m⁶A methylation gain/loss tends to co-occur with differential upregulation/downregulation of DNA methylation. Similar OR calculations and tendency assessments were performed for DNA methylation changes. When OR > 1 but the 95% confidence interval includes 1, the tendency is considered non-significant; exclusion of 1 indicates significant tendency.

Linear regression model construction

Linear regression analysis was performed using the lm function from the ‘stats’ package in R. The model specified the expression change (log2FoldChange) as the dependent variable, with changes in m⁶A methylation levels, changes in DNA methylation levels, and their interaction term as independent variables. The ‘ggtexttable’ function from the ‘ggpubr’ package was used to generate three-line tables displaying model coefficients and evaluation metrics.

ATAC-seq and data analysis

1 ml of 1 XHB buffer was added into the tissue sample, after grinding and filtering, filtrate was centrifuged at 4 °C 500×g for 5 min, and then supernatant was discarded, density gradient centrifugation for 10 min, nucleus suspension was obtained. 50,000 nuclei were taken and resuspended in Tn5 transposase reaction mixture. The transposition reaction was incubated at 37 °C for 30 min. Equimolar Adapter1 and Adatper 2 were added after transposition, PCR was then performed to amplify the library. After the PCR reaction, libraries were purified with the AMPure beads and library quality was assessed with Qubit. The clustering of the index-coded samples was performed on a cBot Cluster Generation System using TruSeq PE Cluster Kit v3-cBot-HS (Illumina) according to the manufacturer’s instructions. The library preparations were sequenced on Illumina Novaseq platform at Novogene Bioinformatic Technology Co., Ltd (Beijing, China) and 150 bp paired-end reads were generated.

FastQC (version 0.11.9) was used to access the base quality of raw data and trim_galore (version 0.0.1) was used to trim the adaptor and low-quality reads with parameters -q 25 –phred33 –length 40 -e 0.1 –stringency 3. After quality control, the remaining reads were mapped to the reference genome (hg19) using bwa (version 0.7.17-r1188). High mapping quality reads with MAPQ≥30 were remained and then duplicated reads were filtered by sambamba (version 0.8.2). Peak calling was performed by using MACS2 (version 2.2.6) with parameters -g hs –nomodel –shift 100 –extsize 200 -q 0.05. Chromatin accessibility in promoter was used to represent the accessibility of genes which were obtained by calculating the TPM within the 1kb range of the upstream and downstream TSS.

Gene expression time series clustering

In order to explore the temporal expression of gene expression at three different stages, we used R package ‘ClusterGVis’ to conduct temporal clustering of expression levels at normal skin, AK and cSCC stages based on ‘fuzzy c-means’ method of ‘Mfuzz’. The optimal number of clusters is determined by the inflection point according to mean square and clustering heatmap.

scRNA-seq data analysis

The scRNA-seq data were from our previous published study. The analytical pipeline and cell annotation strategy can be referenced as described in this literature [28].

Expression and prognostic analysis of candidate genes in multiple squamous cell carcinomas

Expression data for cervical squamous cell carcinoma (CESC), esophageal squamous cell carcinoma (ESCA), and head and neck squamous cell carcinoma (HNSC) were obtained from TCGA. Differential expression visualization was performed using GEPIA. Survival data from TCGA were analyzed with R packages ‘survival’ and ‘survminer’ to conduct prognostic correlation analysis and generate Kaplan-Meier (KM) curves.

qRT-PCR

Quantitative Real-Time PCR (qRT-PCR) was used to verify the expression of key genes in tissues and cells lines. Total RNA was extracted with Trizol reagent (Invitrogen, Thermo Fisher Scientific, USA) and reverse transcribed into cDNA using FastKing One Step RT-qPCR Kit (Tiangen, Beijing, China) according to the manufacturer’s protocols. The qRT-PCR was performed using SYBR Green kit (Tiangen, Beijing, China). The primers are listed in Table 1. The RNA expression level of target genes was evaluated by 2^−∆∆Ct.

Table 1.

The primers of genes used for qRT-PCR

Gene	Forward primer 5’-3’	Reverse primer 5’-3’
IDO1	GCAGCGTCTTTCAGTGCTTT	ACAAACTCACGGACTGAGGG
IFI6	CTGATGAGCTGGTCTGCGAT	TACCTATGACGACGCTGCTG
OAS2	CTCAGAAGCTGGGTTGGTTAT	ACCATCTCGTCGATCAGTGTC
GAPDH	GGACCTGACCTGCCGTCTAG	GTAGCCCAGGATGCCCTTGA

Cell culture, transfections

Human immortalized epidermal keratinocytes cell line (HaCaT cells) and human cSCC cell lines SCL-I (purchased from GuanDao Biological Engineering Co. Ltd, Shanghai, China) were cultured in Dulbecco’s modified Eagle’s medium (DMEM, Gibico, USA), supplemented with 10% fetal bovine serum (FBS, Gibico) and 1% penicillium-streptomycin (Gibico), cultured in a 5% CO2 incubator at 37 °C. Using small interfering RNA (siRNA) transfection for RNA interference, to investigate the effects of silencing gene expression on cell growth, proliferation, invasion and metastasis. The sequences of various siRNA oligonucleotides used in this study were listed in Table 2. The overexpression plasmids pCMV-IDO1 (human)-EGFP-Neo were purchased from Yanming Co. Ltd., Shenzhen, China. The transfection efficiency was confirmed by RT-PCR.

Table 2.

The sequences of siRNA used in this study

	Sense (5’−3’)	Antisense (5’−3’)
si-IDO1-1	GGACAAUCAGUAAAGAGUACCAUAU	AUAUGGUACUCUUUACUGAUUGUCC
si-IDO1-2	CAGCUGCUUCUGCAAUCAAAGUAAU	AUUACUUUGAUUGCAGAAGCAGCUG
si-IDO1-3	CAAAGGAACUGGAGGCACUGAUUUA	UAAAUCAGUGCCUCCAGUUCCUUUG
si-IFI6-1	CCAUGGGUCUGCAGAGCAATT	UUGCUCUGCAGACCCAUGGTT
si-IFI6-2	GUAUUAAUUGGCUCUAUAATT	UUAUAGAGCCAAUUAAUACTT
si-IFI6-3	CACUAAUAGAACAAUCCUATT	UAGGAUUGUUCUAUUAGUGTT
si-OAS2-1	CCCACCAAACUAAAGGAUUUATT	UAAAUCCUUUAGUUUGGUGGGTT
si-OAS2-2	CGUGUUCCAUAACUCACUUAATT	UUAAGUGAGUUAUGGAACACGTT
si-OAS2-3	CCUGGAGCUGGUCACACAAUATT	UAUUGUGUGACCAGCUCCAGGTT

Cell-counting kit 8 (CCK8) assay

The transfected cells were seeded in 96-well plates at a seeding density of 2000 cells/well, cultured at 37 °C and 5% CO₂. Supernatants were removed after 24 h, 48 h, and 72 h, respectively, and each well was washed once with 200 μl of serum-free medium. CCK8 reagent (Beyotime, Shanghai, China) was added to each well, and then the cells were incubated for 2 h in CO₂ incubator. The absorbance (OD) of each well was measured at 450 nm by Microplate reader (BioTek, USA).

EdU cell proliferation assay

Cells in the logarithmic growth phase were inoculated in 24 plates (500 μl per well) at 2 × 10⁴ cells/mL density and cultured to normal growth phase. Cell suspensions were treated with BeyoClick™ EdU Cell Proliferation Kit with Alexa Fluor 555 (Cat# C0075S, Beyotime) according to the manufacturer’s instructions. Briefly, EdU labeling, fixation, EdU detection, nuclear staining, and then fluorescence detection were performed sequentially.

Cell migration assay

Cells were grown to confluence in 6-well plates, and the wounds were made in confluent monolayer cells using a sterilized 200 µL pipette tips. The cells were washed with PBS and cultured with the indicated treatment. Wound healing of different groups was detected at 0, 24, 48, and 72 h within the scraped lines, and representative fields were photographed at the different time points to assess the migratory ability of the cells.

Transwell invasion assays

The invasion ability of cSCC cells was evaluated by transwell assays using Transwell chambers (8 μm pore size, Corning Costar, USA) precoated with Matrigel. After transfection, 4 × 10⁴ cSCC cells in the 200 μL serum-free medium were added to the upper chambers, DMEM with 10% FBS was added to the bottom chambers. After incubation at 37 °C for 36 h, the cells invaded into the lower side of the inserts were fixed in 4% paraformaldehyde and stained with 0.1% crystal violet. Then counted and photographed under a microscope.

Plasmid transfection

Lentiviral systems were used to knockdown human IDO1 expression in SCL-II cells. Two single small hairpin RNAs (shRNAs, Table S5) targeting human IDO1 were synthesized (Ruibo Co. Ltd., Guangzhou, China) and cloned into GV248 vector, a gift kindly provided by Professor Xingding Zhang in School of Medicine, Sun Yat-sen University. For lentivirus packaging, HEK293T cells were co-transfected with the lentiviral transfer plasmid, packaging plasmid psPAX2 (Addgene #12260), and envelope plasmid pMD2G (Addgene #12259) using PEI (Yeasen, MW 40,000). Viruses were harvested 48 hours post-transfection and filtered through a 0.45 μm membrane.

Generation of IDO1 knockdown SCL-II cell line

Cells were infected with a 1:1 mixture of medium and viral supernatant in transduction medium (complete growth medium supplemented with 8 μg/mL polybrene [Solarbio, Cat. No. H8761]). After incubation at 37 °C with 5% CO₂ for 12 hours, the medium was replaced with fresh complete growth medium. Transduced cells underwent puromycin selection (1 μg/mL, InvivoGen™, ant-pr-1) for 72 hours.

Statistical analysis

All experiments were performed in triplicate technical replicates, and data are presented as mean ± standard deviation (SD). Differences among groups were analyzed using Student’s t test or one-way analysis of variance (ANOVA) for normally distributed data and the Kruskal-Wallis test for non-normally distributed data. And p < 0.05 was considered statistically significant while p < 0.01 was considered highly statistically significant. For bioinformatics analysis, non-parametric tests involving comparisons among multi-groups were performed using Kruskal-Wallis rank-sum test, comparisons between two-groups were performed using Wilcoxon rank-sum test. Differences in categorical variables were assessed using χ2/Fisher test. The statistical significance threshold was set at p.adjust < 0.05 for all tests, with significance levels denoted as follows: *p.adjust < 0.05, **p.adjust < 0.01, ***p.adjust < 0.001, ****p.adjust < 0.0001.

Results

The m⁶A profile in AK and cSCC

To explore the potential epigenetic mechanisms in AK and cSCC, we collected five AK samples, five cSCC samples and ten normal skin samples, covering both the epidermis and dermis. We then performed parallel analysis of RNA m⁶A-seq, 850K DNA methylation arrays, whole transcriptome, and ATAC-seq chromatin accessibility profiling. Based on the multi-omics data, we investigated the epigenetic regulatory network and identified key factors driving the progression from AK to cSCC (Fig. 1A).

Fig. 1.

The m⁶A profile in AK and cSCC. (A) Flowchart of overview of this study which explored the epigenome and transcriptome in human skin of actinic keratosis (AK) and cutaneous squamous cell carcinoma (cSCC) patients. (B)The m⁶A peak distribution in mRNAs. (C) PCA analysis of m⁶A levels of normal skin, AK and cSCC samples. (D) The proportion of genes with m⁶A modification and m⁶A genic distribution of different sample groups

To gain insight into the regulation of m⁶A methylation during cSCC development, we first performed m⁶A methylome analysis in AK, cSCC and normal skin samples. We identified m⁶A peaks and quantified their genic locations, finding that m⁶A peaks were markedly enriched in exons and 3’UTR regions near the stop codon. We identified 13,647, 13,946 and 14,564 significant m⁶A peaks in normal skin, AK and cSCC samples, respectively (Fig. 1B, Table S1). Principal component analysis (PCA) of all m⁶A modified genes clearly separated the cSCC samples from AK and normal skin samples. Moreover, while the AK samples clustered homogeneously, the cSCC samples exhibited greater heterogeneity (Fig. 1C). Across all three groups, nearly half of the genes harbored m⁶A peaks (Fig. 1D).

In our previous study, we found that distribution and topological transition of m⁶A across different genic regions of mRNAs play important roles in tissue development and cancer progression [29]. Therefore, to systematically investigate the regulatory effects of m⁶A topological transition during cSCC development, we first clustered the protein-coding genes according to their m⁶A distribution across genic regions. All genes were clearly grouped into five classes, each showing significant m⁶A peaks enrichment in specific genic regions (BLANK, UTR5E, UTR3E, CDSE and MULTIE; Table S2). The BLANK class contained no or very few m⁶A peaks, while UTR5E, UTR3E, and CDSE classes had m⁶A peaks assigned in 5’UTR, 3’UTR and CDS regions, respectively. The MULTIE class had m⁶A peaks located in at least two regions. In general, the BLANK class contained the largest number of genes across all samples. This was followed by CDSE class, while the UTR5E class consistently contained the fewest genes in all groups (Fig. 2A). The heatmap illustrates the five classes and the genic regions where m⁶A peaks were enriched within each cohort (Fig. 2B).

Fig. 2.

The m⁶A distribution along genic regions (A) The number of different classes of m⁶A modified genes based on distribution along genic regions. We defined those five classes as BLANK, CDSE, MULTIE, UTR3E and UTR5E according to the m⁶A deposited regions. (B) The distribution of m⁶A peak density across different gene types in normal skin, AK, and cSCC: the upper section displays the distribution profile, while the lower section presents the density heatmap, with each row representing a single gene

Next, we examined the stage specificity of the genes assigned to each class. Except for the UTR5E class, over 50% of the genes of each class overlapped across at least two groups, and we identified these as non-stage-specific genes. In the UTR5E class, 58.3% of genes were stage-specific (Fig. S1A). Functional enrichment analysis of genes with m⁶A in different genic regions showed that CDSE and MULTIE genes were similar across all three groups and were mainly associated with fundamental cellular processes, including DNA, RNA and metabolic pathways (Fig. S1B, C). The UTR3E class showed some stage specificity and was associated with vasculature development, cell migration, lipid metabolic, and autophagy (Fig. S1D). Genes in UTR5E class were strongly associated with skin-related functions, including keratinocyte differentiation, skin development, keratinization and epidermis morphogenesis, which were not observed in other non-5’UTR enriched genes (Fig. S1E). These results suggested that m⁶A modification is widespread across genes. In addition to the classic enrichment in CDS and 3‘UTR regions, a small proportion of peaks were also distributed in the 5‘UTR region. CDSE and MULTIE genes are relatively conserved among the groups, while the functions of UTR3E genes differ among the groups, reflecting the characteristics of the clinical stages. UTR5E genes displayed distinct tissue-related specificity. These genes may be more susceptible to environmental changes and may exhibit greater epigenetic heterogeneity during cSCC progression.

Correlation between m⁶A modification and gene expression during cSCC development

To investigate the impact of m⁶A modification on gene expression during cSCC development, we analyzed transcriptome data to quantify expression levels (transcripts per kilobase per million mapped reads, TPM) of genes with and without m⁶A peaks across all sample types. Correlation analysis revealed that genes with m⁶A modification exhibited a weak negative correlation between m⁶A methylation and gene expression (Fig. 3A). Among different gene classes, UTR5E, UTR3E and MULTIE genes all showed significant negative correlations, while the negative correlation for CDSE gene was weaker (Fig. 3B). Interestingly, box plots indicated that genes without m⁶A peaks had lower expression levels than those with m⁶A peaks (Fig. 3C). This finding aligns with our previous study, which suggests that genes with or without m⁶A modification are subject to distinct regulatory mechanisms [29].

Fig. 3.

Correlation between m⁶A modification and gene expression in cSCC development. (A) Scatter plots showed the correlation between m⁶A levels and gene expression levels in all m⁶A modified genes. (B) Scatter plots showed the correlation between m⁶A levels and gene expression levels in different classes of genes. (C) Box plots of expression levels (transcripts per kilobase per million mapped reads, TPM) of genes with or without m⁶A peaks in normal skin, AK and cSCC (Wilcoxon, ****p.adjust < 0.0001)

Next, differential expression analysis identified 1,067 upregulated and 852 downregulated differentially expressed genes (DEGs) between cSCC and normal skin. In comparison with normal skin, AK exhibited 554 upregulated and 278 downregulated DEGs. The comparison between cSCC and AK yielded the fewest DEGs, with only 181 upregulated and 156 downregulated (Fig. S2A, Table S3). Compared with normal skin, upregulated DEGs in both AK and cSCC were mainly enriched in immune-related pathways. Downregulated DEGs in cSCC were largely associated with skin development, epidermal differentiation, and keratinization. In contrast, downregulated DEGs in AK (vs. normal skin) were predominantly related to muscle tissue and structural development (Fig. S2B, Table S3). To explore dynamic m⁶A-mediated regulation of gene expression during cSCC development, we used the Mfuzz algorithm to identify potential temporal patterns and cluster gene sets with similar dynamic expression trends. Analysis of these temporal patterns showed that upregulated genes in AK and cSCC were mainly involved in immune functions, whereas genes related to skin development, keratinization, and epidermal differentiation were consistently downregulated during disease progression (Fig. S2C). This persistent upregulation of immune-related genes during disease progression indicates remodeling of the immune microenvironment during cSCC development.

m⁶A modification changes positively regulate gene expression in cSCC

To systematically investigate the impact of m⁶A modification changes on gene expression during cSCC progression, we assigned gain or loss status of m⁶A peaks to represent each gene’s overall m⁶A state for exploratory analysis. Genes transitioning from the BLANK category to other classes were defined as m⁶A gain (m⁶A.Gain) genes, whereas those transitioning from other classes to BLANK were defined as m⁶A loss (m⁶A.Loss) genes. Genes showing only positional changes of m⁶A peaks between groups were considered not to experience m⁶A peak gain or loss. Using this approach, compared to normal skin, the cSCC group showed m⁶A modification gain in 831 genes and loss in 781 genes. Compared to AK, the cSCC group showed gain in 837 genes and loss in 616 genes. Compared to normal skin, AK showed gain in 691 genes and loss in 862 genes (Fig. S3).

We categorized genes into distinct clusters based on the overlapping m⁶A peak gain or loss among groups at different stages (Fig. 4A). Genes in Cluster C2, which specifically gained m⁶A modification in cSCC, were primarily enriched in immune-related processes such as immunoglobulin isotype switching and leukocyte differentiation, as well as pathways related to elastic fiber assembly and cell cycle progression. Genes in Cluster C3, which specifically lost m⁶A modification in AK, were mainly enriched in protein and macromolecule modification, protein localization, the Toll-like receptor signaling cascade, and the p53/Wnt signaling pathways. Genes in Cluster C4, which gained m⁶A modification in both AK and cSCC, were mainly associated with processes such as metabolism, cellular stress response, and T cell activation, as well as the Toll-like receptor signaling cascade, immune system, and NOD-like receptor pathways (Fig. 4B). These results suggest that m⁶A modification may promote disease progression in AK and cSCC by regulating metabolism and immune responses. During the AK stage, m⁶A modification may affect protein homeostasis, chronic inflammation, and the p53/Wnt pathways, leading to aberrant accumulation characteristic of early carcinogenesis. At the cSCC stage, m⁶A modification may further modulate immune responses, cell proliferation, and tumor microenvironment remodeling, thereby facilitating tumor invasion and immune evasion.

Fig. 4.

Overlap of genes with m⁶A gain or loss across different stages and their functional enrichment. (A) Overlap of genes exhibiting m⁶A peak gain or loss across different stage comparisons. The overlapped groups are indicated at the ends of the dumbbell plot below. The bar graph above shows the corresponding number of overlapping genes. The bar on the left represents the total number of genes in each group. (B) Significantly enriched functional terms for overlapping gene sets. Enrichment results are shown for the GO (top), ReactomePA (bottom left), and KEGG database (bottom right)

Integrating the corresponding gene expression levels, we found that m⁶A modification gain tended to be coincide with increased expression, with a higher proportion of upregulated DEGs. Conversely, m⁶A modification loss was associated with decreased gene expression, showing a higher proportion of downregulated DEGs (Fig. 5A, B).

Fig. 5.

Expression Changes of Genes with m⁶A Gain or Loss. (A) Distribution of expression changes for genes exhibiting m⁶A peak gain or loss. The y-axis represents the log2FoldChange of differential expression between groups. (B) Number of significantly upregulated and downregulated genes among those with m⁶A peak gain or loss. From left to right: cSCC vs normal skin, AK vs normal skin, cSCC vs AK. (C) IGV browser view of m⁶A modification signals for the IDO1 and IFI6 genes

By merging genes with m⁶A peak gain or loss across normal skin, AK, and cSCC samples, we quantified m⁶A modification changes using read counts at peak positions and performed correlation analysis with differential gene expression. The results showed an overall positive correlation between changes in m⁶A modification and expression changes. This positive correlation was mainly driven by genes with m⁶A peak gain or loss. For genes without m⁶A gain or loss (genes that only showed changes in m⁶A peak positions or levels in specific pairwise comparison), the correlation between m⁶A modification and differential expression changes was weak (Fig. S4). Compared to normal skin, upregulated genes with m⁶A gain in cSCC were mainly associated with immune-related pathways such as interferon signaling, leukocyte-mediated immunity, and B cell activation. Comparing to AK, these genes were mainly enriched in pathways related to TGF-β signaling, IL-4 and IL-13 signaling, and extracellular matrix (ECM). Compared to normal skin, these genes in AK were mainly enriched in immune-related pathways including interferon signaling, lymphocyte activation, and T cell activation (Fig. S5). Genes with downregulation and m⁶A peak loss were fewer and showed no significantly functional enrichment. These results suggest that m⁶A topological transitions can reshape the immune and stromal microenvironment during disease progression by regulating gene expression, thereby promoting disease pathogenesis and progression. For instance, in cSCC samples, the significant m⁶A gain was observed in 3‘UTR and CDS regions of the immunosuppressive gene IDO1 and the interferon pathway-related gene IFI6 (Fig. 5C).

DNA methylation and gene expression during cSCC development

To analyze the DNA methylomes of human AK and cSCC, we conducted Infinium 850k methylation arrays on the same samples used for RNA methylation. PCA analysis of DNA methylation levels showed that cSCC samples were clearly separated from AK and normal skin samples. Moreover, while AK and normal skin samples clustered homogeneously, cSCC samples showed heterogeneous distribution (Fig. 6A). We then identified differentially methylated regions (DMRs) among cSCC, AK, and normal skin groups. Compared to the normal skin group, AK had only 8 hypomethylated DMRs (HypoDMRs) and 6 hypermethylated DMRs (HyperDMRs). However, there were 15,259 HypoDMRs and 8,302 HyperDMRs between cSCC and normal skin, 13,821 HypoDMRs and 8,302 HyperDMRs between cSCC and AK (Fig. 6B). Feature distribution analysis indicated that the proportion of HypoDMRs and HyperDMRs were similar in promoter regions, whereas the gene body contained more HypoDMRs (Fig. S6A). Both promoter and gene body regions showed high overlap of HypoDMRs and HyperDMRs among cSCC, AK and normal skin (Fig. S6B, C).

Fig. 6.

DNA Methylation Differences and Expression Changes in cSCC development A. PCA clustering of samples based on DNA methylation levels. B. Number of Hypo/HyperDMRs between groups. C. Scatter plot showing the correlation between DMR methylation change differences and the corresponding gene expression level changes (log2FoldChange) between the cSCC group and the normal skin group. D. Scatter plot showing the correlation between DMR methylation change differences and the corresponding gene expression level changes (log2FoldChange) between the cSCC group and the AK group. The results of the linear fit are indicated in the upper right corner of each plot

The average methylation levels of DMRs were very similar between normal skin and AK samples. Furthermore, the methylation change patterns of partitioned DMRs and the functional enrichment of associated genes were broadly similar between the two groups. Genes associated with promoter HypoDMRs were primarily enriched in biological processes such as cell killing and immune response. Conversely, genes associated with promoter HyperDMRs were mainly enriched in pathways related to embryonic organ development, muscle and nervous system development, and morphogenesis (Fig. S7A-D). Genes with gene body HypoDMRs were mainly enriched in processes such as synapse assembly and ion transport, whereas genes with gene body HyperDMRs were linked to the Wnt signaling pathway, GTPase cycle, and neural development (Fig. S7E-H). Overall, these results suggest that in AK, as a precancerous lesion, DNA methylation changes are mainly focal, with only a few genes showing differential methylation (e.g., AXIN2, a Wnt pathway regulator, and SIX1, a development-related gene, Table S4). Genome-wide methylation reprogramming has not yet occurred, indicating an early stage of molecular event accumulation. In contrast, the cSCC stage exhibits significant global methylation perturbation. This disruption of methylation homeostasis likely reflects the complete breakdown of keratinocyte proliferation/differentiation balance and microenvironment remodeling.

Next, we investigated the expression levels of DMR-associated DEGs. Direct quantitative analysis of DMR methylation changes and corresponding gene expression changes revealed that an inverse correlation in promoter regions, and a positive correlation in gene body regions (Fig. 6C, D).

The promoter is a crucial regulatory element for initiating gene transcription. Numerous studies have shown that DNA methylation in promoter regions negatively regulates gene expression. Further functional enrichment analysis of DEGs with promoter DMRs showed that HypoDMR-associated upregulated DEGs were mainly enriched in immune response pathways, whereas HyperDMR-associated downregulated DEGs were strongly linked to metabolic pathways (Fig. 7). For example, the core immune checkpoint gene PDCD1 and the B-cell development gene BST2 both exhibited significant promoter hypomethylation, whereas the lipid metabolism genes ALDH3A1 and EPHX3 showed marked promoter hypermethylation (Fig. S8).

Fig. 7.

Functional Enrichment of Negatively DEGs with Promoter Differential Methylation Between cSCC and Normal Skin or AK Groups. A. Significantly enriched terms for DNA methylation regulated DEGs between the cSCC group and the Normal Skin group. B. Significantly enriched terms for DNA methylation regulated DEGs between the cSCC group and the AK group. Terms are color-coded by database: purple for GO, orange for KEGG, and blue for ReactomePA. The red dashed line indicates the -log10p.adjust value corresponding to a p.adjust threshold of 0.05. Terms shown in the figure use p.adjust < 0.1 as the significance threshold

ATAC-seq can define the chromatin accessibility of genomic DNA. We conducted parallel ATAC-seq analysis on 4 cSCC samples and 5 normal skin samples. Distribution analysis showed that ATAC-seq peaks were also mainly located in promoter regions (Fig. S9A). Compared to normal skin, changes in promoter chromatin accessibility were positively correlated with gene expression changes of DEGs in cSCC (Fig. S9B). Meanwhile, chromatin accessibility was negatively correlated with changes in promoter DNA methylation levels between cSCC and normal skin (Fig. S9C). Compared to normal skin, chromatin accessibility of HypoDMR genes was increased in cSCC. Conversely, chromatin accessibility of HyperDMR genes was decreased (Fig. S9D). For example, in cSCC, promoter accessibility of tumor necrosis factor ligand superfamily member 14 (TNFSF14) and Aurora kinase A (AURKA) was increased compared to normal skin (Fig. S9E), along with changes in DNA methylation levels (Fig. S9F). TNFSF14 has been reported to stimulate T cell proliferation and trigger apoptosis in various tumor cells [30], whereas AURKA regulates cell cycle progression [31] and acts as a key component in the p53/TP53 pathway, particularly in checkpoint-response pathways critical for oncogenic transformation [32]. These results suggest that DNA methylation may regulate key gene expression by modulating promoter accessibility in cSCC.

The crosstalk of m⁶A and DNA methylation regulates gene expression in cSCC

Given previous reports that gene transcription can be regulated by RNA m⁶A modification coupled with DNA methylation [26], we investigated the potential synergistic effects of m⁶A and DNA methylation on gene expression in cSCC. Genes exhibiting neither m⁶A peak gain/loss nor DMRs were designated as the control group. The odds ratios (OR) were calculated respectively for the propensity of genes with m⁶A peak gain/loss to undergo DNA hypermethylation/hypomethylation, and for the propensity of genes with differential DNA hypermethylation/hypomethylation to exhibit m⁶A peak gain/loss. In comparisons of cSCC with AK and normal skin, we found that genes with m⁶A peak gain tended to show DNA hypomethylation, while genes with m⁶A peak loss tended to show DNA hypermethylation. The DMR-associated genes showed the same trend (Fig. 8A, B). Correlation analysis between changes in m⁶A modification and DNA methylation levels also revealed a significant negative correlation (Fig. 8C). This correlation was also validated by published DNA methylation data from Hervás-Marín et al. (Fig. S10) [33]. These results suggest a negative crosstalk between DNA methylation and m⁶A modification changes.

Fig. 8.

Correlation between m⁶A gain/loss and DNA methylation up/downregulation. A. Between cSCC and normal skin groups and B. Between cSCC and AK groups, odds ratios (OR) indicating whether m⁶A gain/loss tends to be associated with DNA methylation up/downregulation, or DNA methylation up/downregulation tends to be associated with m⁶A gain/loss. Red represents OR > 1, indicating a tendency. C. Scatter plot of the correlation between differences in m⁶A modification levels and differences in DNA methylation levels. The linear fit result is shown in the upper left corner. Point color represents the level of difference

The Venn diagram showed that in the cSCC versus normal skin comparison, the overlap between upregulated genes with m⁶A gain (m6AGain-UP) and upregulated genes with DNA hypomethylation (hypoDMR-UP) included 31 genes, accounting for 21.5% (31/142) of m6AGain-UP genes and 17.6% (31/176) of hypoDMR-UP genes. The overlap between downregulated genes with m⁶A loss (m6ALoss-DOWN) and those with DNA hypermethylation (hyperDMR-DOWN) included 22 genes, accounting 21.0% (22/105) of m6ALoss-DOWN genes and 16.4% (22/134) of hyperDMR-DOWN genes (Fig. S11A). Functional enrichment analysis showed that the overlapping genes of both m6AGain-UP and hypoDMR-UP were enriched in inflammatory diseases, including autoimmune disorders, graft-versus-host disease, as well as immune-related processes such as B-cell activation and differentiation, Toll-like receptor 4 signaling, monocyte proliferation and differentiation, and lymphocyte proliferation (Fig. S11B). For example, the expression of PRF1, which encodes a cytotoxic protein secreted by CD8+ T cells and NK cells, and OAS2, a gene involved in the innate immune response to viral infection, were significantly upregulated in cSCC. Both genes exhibited m⁶A peak gain in their 3‘UTRs and significant promoter DNA hypomethylation (Fig. 9). The significant promoter DNA methylation changes of DEGs was also validated by published data from Hervás-Marín et al. (Fig. S12).

Fig. 9.

The IGV diagram of DNA and m⁶A methylation levels for epigenetically co-regulated genes PRF1 and OAS2. (A) The DNA methylation levels for PRF1 and OAS2 respectively. (B) The m⁶A methylation levels for PRF1 and OAS2 respectively

To assess the potential combined effect of m⁶A and DNA methylation, we conducted a regression analysis. Differential expression changes between cSCC and normal skin were used as the dependent variable. Changes in m⁶A modification levels associated with peak gain or loss, changes in DNA methylation levels corresponding to DMRs, and their interaction term were incorporated into the regression model as independent variables. Using ANOVA to test the significance of main and interaction effects we found that increased m⁶A modification was associated with upregulated expression; decreased promoter DNA methylation was also associated with upregulated expression; and the interaction term contributed significantly to gene expression (Table 3). These results suggest that during cSCC development, a potential interaction exists between m⁶A modification and DNA methylation. Changes in m⁶A modification were statistical negatively correlated with changes in promoter DNA methylation, and these two epigenetic modifications may regulate gene expression through both independent mechanisms and synergistic effects. However, the details and causality of the cooperative regulation need more experimental validations.

Table 3.

Logistic regression model of cSCC vs normal skin

	Estimate	Std. Error	t value	Pr ( > ltl)
(Intercept)	−0.0794	0.0571	−1.3898	0.1657
m6A Diff	0.2071	0.0212	9.7808	0***
Beta Diff	−1.1967	0.5795	−2.0648	0.0399*
m6A Diff: Beta Diff	−0.4442	0.2186	−2.032	0.0431*

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘’ 1

Residual standard error: 0.908 on 271 degrees of freedom

(1235 observations deleted due to missingness)

Multiple R-squared: 0.3057,Adjusted R-squared: 0.298

F-statistic: 39.77 on 3 and 271 DF, p-value: < 2.2e-16

Identification of key driver genes regulated by epigenetic modification in cSCC development

The previous results have established that DNA methylation and m⁶A methylation participate in regulating multiple key biological processes during the cSCC development and progression, including epidermal differentiation, metabolic homeostasis, immune response, and stromal microenvironment remodeling. To further elucidate the molecular mechanisms underlying cSCC malignant transformation and identify novel epigenetically regulated target genes, we first screened for genes exhibiting significant epigenetic modifications and substantial upregulation. To effectively exclude confounding effects from infiltrating immune cells on gene expression, we leveraged our published single-cell transcriptomic data to identify genes specifically differentially expressed in key keratinocyte populations, including basal and proliferating basal cells. The intersection of epigenetically upregulated genes with upregulated genes in basal and proliferating basal subpopulations during the cSCC stage yields 44 candidate epigenetically upregulated genes. Notably, several overlapping genes have established roles in various cancers. For instance, matrix metalloproteinase genes MMP1, MMP3, and MMP13 have been extensively studied across multiple cancers including SCC [34]. The chemokine gene CCL5 has been reported to promote cell migration in cSCC [35]. Expression of IFI44 and IFI44L in keratinocytes is associated with HPV infection [36].

Through functional annotation analysis and integration of published literature, we ultimately selected three gens IDO1, IFI6, and OAS2, which were not previously linked to cSCC pathogenesis, as candidates with potential tumor-promoting roles for exploratory functional validation experiments. Single-cell transcriptomic data further confirmed that these three genes were significantly overexpressed in basal and proliferating basal cells (ProBasal) in cSCC (Fig. S13) [28]. Indoleamine 2,3-dioxygenase 1 (IDO1) is a heme enzyme involved in the catabolism of tryptophan to N-formyl-kynurenine [37]. A previous study observed significantly increased IDO1 expression in high-risk cSCCs with tumor thickness over 6 mm [38]. Interferon Alpha Inducible Protein 6 (IFI6) is an interferon-induced gene primarily involved in regulating apoptosis and antiviral responses. Although IFI6 has been extensively studied in various cancers, its specific function in cSCC remains poorly defined. 2’−5’-Oligoadenylate Synthetase 2 (OAS2) is an interferon-induced antiviral factor involved in innate immune defense. Studies have found that OAS2 expression is upregulated in highly invasive cSCC cell lines [39]. OAS2 has been more frequently studied in psoriasis, but no established association exists yet in cSCC. In this study, IDO1 and IFI6 were identified as m⁶A gain-upregulated genes. OAS2 was concurrently upregulated through both m⁶A peak gain in its 3‘UTR and DNA hypomethylation in its promoter region (Fig. 5 & 9).

Due to the lack of publicly available survival cohort data for cSCC, we leveraged other squamous cell carcinoma datasets from The Cancer Genome Atlas Program (TCGA) to analyze expression and prognostic associations for IDO1, IFI6, and OAS2. All three genes showed significantly higher expression in tumor tissues of cervical squamous cell carcinoma (CESC), head and neck squamous cell carcinoma (HNSC), and esophageal squamous cell carcinoma (ESCA) (Fig. S14). Notably, high IDO1 expression was associated with better prognosis in CESC and HNSC but poorer prognosis in ESCA. High IFI6 expression was associated with poorer prognosis in CESC and HNSC but better prognosis in ESCA. High OAS2 expression was associated with better prognosis in both CESC and HNSC, while showing no significant prognostic association in ESCA (Fig. S15). These findings suggest that IDO1, IFI6, and OAS2 may exert potential context-dependent tumor-promoting or tumor-suppressive functions through distinct molecular mechanisms in different squamous carcinomas. Their roles in cSCC pathogenesis urgently warrant further investigation.

The expression and functional characterization of IDO1, IFI6 and OAS2 in cSCC development

We first performed qRT-PCR on an independent set of clinical samples from normal skin and cSCC (n = 10 per group) to validate the expression of IDO1 and IFI6. The results showed that the expression levels of these genes were significantly increased in cSCC samples compared to normal skin (Fig. 10A).

Fig. 10.

The expression and functional characterization of IDO1 and IFI6 in cSCC development. (A) The expression of IDO1 and IFI6 was validated by qRT-PCR on an independent set of clinical samples of normal skin and cSCC groups. **p < 0.01, ***p < 0.001. (B) The expression levels of IDO1 and IFI6 in human normal skin HaCaT cells and cSCC cell lines (A431, SCL-I and SCL-II). *p < 0.05, **p < 0.01, ***p < 0.001, ns: not significant. (C) Effect of IDO1 and IFI6 on cSCC cell proliferation. The CCK-8 proliferation assay demonstrated a significant decrease in the proliferation of the si-IDO1and si-IFI6 groups compared with the si-NC groups. *p < 0.05. (D) The results of the EdU staining analysis, the cell proliferation rate of the si-IDO1 group and si-IFI6 group were reduced in SCL-I cells compared to the control group. **p < 0.01. (E) The result of cell scratch assays showed that IDO1 and IFI6 knockdown a shorter vertical migration distances of SCL-I cells after 72 hr compared with the control group. *p < 0.05, **p < 0.01. (F) Transwell invasion assay showed that IDO1 and IFI6 gene silencing reduced the invasion ability of SCL-I cells

We also evaluated their expression levels in HaCaT cells and human cSCC cell lines (A431, SCL-I and SCL-II). As shown in Fig. 10B, IDO1 and IFI6 expression was significantly upregulated in SCL-I and SCL-II cells compared to HaCaT cell, with the highest levels observed in SCL-I. Therefore, the SCL-I cell line was selected for subsequent experiments.

To investigate the effects of IDO1 and IFI6 on the proliferation, migration and invasion of human cSCC cells, SCL-I cells were transfected with small interfering RNA (siRNA) targeting these genes (Table 2).

As shown in Fig. 10C, the CCK-8 assay revealed significant reduced optical density (OD) values in the si-IDO1 and si-IFI6 groups compared with the control (p < 0.05). Additionally, EdU staining analysis (Fig. 10D) indicated a significantly reduced proliferation rate in both siRNA-IDO1 and siRNA-IFI6 groups compared to the control. These differences were statistically significant (p < 0.05), and consistent with the CCK-8 assay results, demonstrating that silencing IDO1 and IFI6 inhibits human cSCC cell proliferation.

The effect of gene interference on cSCC cell migration was assessed by cell scratch assay. The results showed that knockdown of IDO1 and IFI6 significantly reduced the migration distances of SCL-I cells after 72 h compared to the control group (p < 0.05), indicating decreased migration ability (Fig. 10E).

A transwell invasion assay was performed to evaluate the effect of gene interference on invasion ability. The results showed that silencing of IDO1 and IFI6 significantly reduced the invasion ability of SCL-I cells (Fig. 10F).

These results suggested that IDO1 and IFI6 may play crucial roles in cSCC by regulating cell proliferation, migration and invasion.

We also conducted similar expression and functional analyses of OAS2 in cSCC cell lines. Compared to HaCaT cells, OAS2 expression was also significantly upregulated in A431, SCL-I, and SCL-II cell lines (p < 0.05) (Fig. S16A). CCK-8 assay results demonstrated that cell viability was markedly reduced in the si-OAS2 group compared to the control (p < 0.05) (Fig. S16B). The cell scratch assay showed that OAS2 knockdown significantly reduced the cell migration distances after 72 h (p < 0.05) (Fig. S16C). Transwell invasion assays confirmed that OAS2 knockdown significantly impaired the invasive capacity of cSCC cells (p < 0.05) (Fig. S16D). Further treatment with doxorubicin (DOX) at 2 μM (slightly above the control IC₅₀ of 1.6 μM) resulted substantially lower cell survival in the si-OAS2 group compared to si-NC + DOX control group (Fig. S16E), suggesting that OAS2 knockdown may enhance cellular sensitivity to DOX. These findings indicate that OAS2 likely regulate proliferation and invasion in human cSCC cells and may contribute to tumor chemoresistance. Reduced OAS2 expression enhances DOX-induced cell death, suggesting that targeting OAS2 may represent a potential novel therapeutic strategy for cSCC, particularly in combination with chemotherapeutic agents.

Functional validation of IDO1 knockdown cells and changes in expression profiles

IDO1 is overexpressed in the vast majority of cancers and plays a critical role in helping tumor cells evade both innate and adaptive immune systems [40]. Combination therapy strategies targeting IDO1 and other immune checkpoints (such as PD-1/PD-L1) have demonstrated efficacy in various types of tumors, including melanoma [41]. Considering the role of IDO1 in tumor immunity, we established IDO1 knockdown (KD) SCL-II cell lines for further validation and investigation. The knockdown efficacy of IDO1 was confirmed by qRT-PCR (Fig. S17A). We then conducted the functional experiments on IDO1 WT and KD SCL-II cells. The results showed that IDO1 KD significantly influence the proliferation, migration and invasion of SCL-II cells. In addition, the phenotypes caused by IDO1 KD could be fully rescued by complementing with pCMV-IDO1 (human)-EGFP-Neo (Fig. S17 B-E).

Then we performed transcriptome sequencing on the IDO1 knockdown SCL-II cell lines to explored changes in the gene expression profile. Differential analysis revealed that after IDO1 knockdown, 331 genes were downregulated, while 128 genes were upregulated (Fig. S18A). The downregulated genes were enriched in interferon and cytokine-related pathways, the JAK-STAT signaling pathway, negative regulation of viral process, as well as biological processes associated with keratinization, epidermal differentiation and development, and skin injury repair. The upregulated genes were enriched in the Rho GTPase cycle and mitotic processes (Fig. S18B). Further examining the genes and enriched biological processes related to immunity or cell activation, we found several key immune-related genes, such as CD274 (PD-L1) and Interleukin 23 Subunit alpha (IL23A) were downregulated in response to IDO1 knockdown. Other downregulated genes included Suppressor of Cytokine signaling 1 (SOCS1) and Platelet-Derived Growth Factor Subunit B (PDGFB) (Fig. S18C). The downregulation of these genes following IDO1 knockdown suggests that IDO1 may promote an immunosuppressive and pro-tumorigenic environment by modulating key immune checkpoints, cytokine signaling, and stromal interactions. The presence of terms also supports the role of IDO1 in shaping immune responses withing the tumor microenvironment.

Discussion

The occurrence and development of tumors is an extremely complex, multistep process influenced by both genetic and environmental factors. Epigenetic regulation, serving as a critical interface bridging genotype and phenotype, can elucidate the spatiotemporal heterogeneity in the transcriptome that cannot be fully explained by genomic variations alone during tumor progression. In this study, we performed multi-omics profiling across different stages of cSCC. To minimize individual heterogeneity, we conducted parallel RNA m⁶A-seq, 850K DNA methylation array, whole transcriptome sequencing, and ATAC-seq chromatin accessibility assays on the same set of samples. To avoid batch effects, all samples had to meet the quantity requirements for these four assays simultaneously, which inevitably limited the number of qualified specimens and, to some extent, the comprehensiveness of the data. Future studies should aim to collect more samples to expand these analyses.

To our knowledge, our study is the first to perform m⁶A sequencing in cSCC and to delineate in detail how the topological distribution of m⁶A modifications regulates gene expression to influence cSCC progression. Based on the distribution of m⁶A deposition along genic regions, we identified five gene classes with distinct m⁶A topologies. Notably, m⁶A modifications within 5‘UTR regions are often overlooked due to their atypical distribution and low abundance. In our study, we found that genes with 5‘UTR m⁶A peaks were functionally enriched in processes related to skin development and epidermal differentiation, with clear pathological stage-specific patterns. This finding aligns with previous reports that demonstrated tissue-specific enrichment of m⁶A sites in 5‘UTRs [42, 43], suggesting that highly tissue-specific genes may be more susceptible to epigenetic alterations under different pathological conditions. Further analysis uncovered a dual regulatory effect of m⁶A on gene expression levels. Within m⁶A-modified genes, modification abundance was negatively correlated with expression levels, whereas the presence of m⁶A itself positively correlated with expression levels when comparing modified versus unmodified genes. This seemingly paradoxical pattern persisted across intergroup comparisons. The m⁶A gain genes tended to be upregulated, while m⁶A loss genes predominantly tended to be downregulated. This bidirectional regulatory characteristic reflects the complex and context-dependent mechanism of m⁶A modifications, which is strongly influenced by the spatial positioning of modification sites, the specific reader proteins recruited, and microenvironmental signaling networks crosstalk [44, 45]. This ‘dual functionality of a single modification’ pattern implies that m⁶A may possess bidirectional regulatory potential for both activation and suppression within the same system, with its ultimate biological effects determined by the spatiotemporal activation states of specific regulatory factors and crosstalk between signaling pathways, highlighting the complexity and context-dependency of epitranscriptomic regulation.

PCA analysis of DNA methylation levels clearly separated cSCC from AK and normal skin. cSCC samples exhibited significant global methylome disruption, whereas AK showed only localized epigenetic alterations. This phenomenon aligns intrinsically with epigenetic regulation patterns during keratinocyte differentiation. Previous studies indicate that terminal keratinocyte differentiation occurs independently of DNA methylation remodeling, which appears to preferentially regulate precursor cell fate decisions such as epidermal stem cell quiescence/activation transitions [46–48]. At the AK precancerous stage, lesional tissue primarily displays pathological features like epidermal thickening and hyperkeratosis while maintaining intact basal-spinous-granular layering, suggesting unimpaired proliferation/stemness homeostasis with preserved epidermal stem cell equilibrium. Progression to cSCC, keratinocytes undergo malignant transformation through acquired rapid proliferation capacity, dedifferentiation plasticity, and microenvironment remodeling, which may potentially dependent on DNA methylation reprogramming [16, 49–54]. Notably, the few differentially methylated genes in AK may hold critical clues: AXIN2, a core Wnt pathway regulator, showed mild hypermethylation in AK with further elevation in cSCC. AXIN2+ basal cells were found in mouse interfollicular epidermis (IFE) to maintain stemness through autocrine Wnt signaling and continuously generate keratinocytes [55]. This suggests AXIN2 methylation changes in AK may represent early epigenetic drivers of stemness perturbation. Furthermore, methylation signature-based studies classify AK into epidermal stem cell-derived and differentiated keratinocyte-originated subtypes [33, 56]. The cellular origin of AK/cSCC whether derived from stem-competent basal cells or differentiated keratinocytes still remains debated [57, 58]. In this study, we observed methylation similarity between AK and normal skin tissue. It requires deeper investigation with expanded datasets and further experimental evidence to reveal it is due to lesions originating from differentiated keratinocytes without sufficient epigenetic alterations to disrupt proliferation/stemness balance, or early events confined to epidermal stem cell subsets not yet triggering global methylation imbalance.

Recent studies have noticed crosstalk between DNA methylation and m⁶A RNA methylation. Research demonstrates that co-transcriptional m⁶A deposition recruits the demethylase TET1 via m⁶A reader protein FXR1, inducing local DNA demethylation to regulate gene expression [26]. In another study, reader protein YTHDC2 binds m⁶A-modified HERV-H RNA and recruits TET1 to maintain LTR7 hypomethylation, preserving its transcriptional activity [59]. Our study similarly reveals significant synergistic regulation between DNA methylation and m⁶A RNA methylation. The m⁶A gain genes concurrently exhibit reduced promoter DNA methylation, with both modifications synergistically driving transcriptional upregulation. However, the causal relationship and mechanisms of interaction between these epigenetic regulations have not been elucidated. The transcript levels of some key regulators, including DNMTs, TETs, YTHDFs, and YTHDCs, showed no significant intergroup differences. Future investigations may need integrate protein-DNA/RNA interaction data (e.g., ChIP-seq) to systematically dissect the molecular crosstalk mechanisms bridging these two modifications.

Cross-referenced with our published single-cell transcriptomic data, we identified significantly upregulated DEGs in keratinocytes with significant changes of epigenetic modification as potential driver genes in cSCC development. Expression and functional experiment confirmed the potential tumor-promoting roles of candidate genes IDO1, IFI6 and OAS2 in regulating the cSCC cell proliferation, migration and invasion. In phenotypic experiments on the epigenetically co-regulated gene OAS2, it was observed that cells became more sensitive to DOX treatment after OAS2 silencing. This suggests that OAS2 may play a role in regulating DOX resistance or cell survival. Although DOX is not a first-line standard treatment for cSCC, systemic chemotherapy regimens may be considered for some patients with advanced or metastatic cSCC, in which DOX is occasionally incorporated as part of combination therapies. Therefore, targeting OAS2 may confer certain therapeutic benefits in some cSCC cases. The role of IDO1 in tumor immunity has been extensively studied. Transcriptomic analysis of the IDO1 knockdown cSCC cell line identified differentially upregulated and downregulated genes following knockdown. Notably, downregulated genes enriched process “Negative regulation of interleukin-10 production” may reflect immune feedback mechanisms that help fine-tune anti-inflammatory responses, often dysregulated in cancer or chronic infection. Although the impact of these genes on cancer cell phenotypes has been confirmed in in vitro cellular experiments, their upstream epigenetic regulation and downstream phenotypic mechanisms require further investigation. Subsequently, we are going to sequence the knockdown cell lines to explore pathways/genes potentially influencing the phenotype and construct METTL3-knockout cell lines and 5-aza-treated cell lines to verify the regulation of epigenetic modifications on the expression of these genes.

In summary, we comprehensively described the epigenetic landscape and dynamic characteristics from normal skin to AK and cSCC and revealed the synergistic regulatory network between DNA methylation and m⁶A RNA methylation. At the pathway level, our findings elucidated how epigenetic regulation, through molecular networks involving immune microenvironment remodeling, metabolic reprogramming, and extracellular matrix remodeling, drives the malignant progression of cSCC. Through data analysis screening and in vitro cellular experiments, the epigenetically upregulated candidate genes IDO1, IFI6, and OAS2 were confirmed upregulated in cSCC and took important role in promoting cell proliferation, migration and invasion of tumor cells, which were potential novel targets for cSCC diagnosis and treatment (Fig. S19). This multi-stage, multi-omics and multi-layers integrative analysis provides a theoretical basis for future personalized treatment strategies for cSCC.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

L.H. designed and supervised the study. TS.G., J.X., and X.L. supervised the study and reviewed the paper. YZ.S., DD.Z., XJ.L., and JY.L. carried out gene sequencing, data analysis and figure/table presentation. YZ.S., DD.Z., XL.W., JH.F., K.H., YT.H., and XX.L. organized and carried out the experimental process. YZ.S., DD.Z., and XJ.L. drafted the manuscript. All authors contributed to the final version of the paper.

Funding

This study was supported by funding provided by Innovative Research Team in Ministry of Education of China (IRT17-R49), Yunnan Province Science and Technology Department & Kunming Medical University Joint Special Project for Applied Basic Research (202501AY070001-261), First-Class Discipline Team of Skin & Mucosal Regenerative Medicine of Kunming Medical University (2024XKTDTS10), Shenzhen Science and Technology Program (JCYJ20210324124808023), GuangDong Basic and Applied Basic Research Foundation (No.2024A1515013077 and 2025A1515011679), Guangdong Provincial Key Laboratory of Digestive Cancer Research (2021B1212040006), China Postdoctoral Science Foundation (2020M683073), Shenzhen Outbound Postdoctoral Research Project (SZBH202001), Yunnan Fundamental Research Projects (202401AU070030), Kunming University of Science and Technology & the First People’s Hospital of Yunnan Province Joint Special Project on Medical Research (KUST-KH2023023Y), High-level Talents Scientific Research Project of Yunnan Provincial Health Commission (2023-KHRCBZ-B13) and The First People’s Hospital of Yunnan Province (KHYJ-2023-6-08), Natural Science Foundation of Guangdong Province (2024A1515012364, 2025A1515010926), Natural Science Foundation of Shenzhen Municipality (JCYJ20240813150406009), National Natural Science foudation of China (32570772).

Data availability

The data of our study is available at Gene Expression Omnibus (GEO) database (850K DNA methylation array: GSE277273; ATAC-seq: GSE277274; m⁶A-seq: GSE277275). The scRNA-seq data was downloaded from GSE193304.

Declarations

Ethical approval and consent to participate

The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. All procedures performed in this study involving human participants were in accordance with the Declaration of Helsinki (as revised in 2013). This study protocol was approved by the Ethics Committee of the First Affiliated Hospital of Kunming Medical University (Approval Number (2020)-L-29), and written informed consent was obtained from all patients.

Conflicts of interest

The authors have no conflicts of interest to declare.