2.1 Public m⁶A RNA immunoprecipitation data

We used the Sequence Read Archive (SRA) database to obtain publicly available m⁶A RNA immunoprecipitation (MeRIP)-seq data from clinical samples of project PRJNA488293 (https://trace.ncbi.nlm.nih.gov/Traces/index.html?view=run_browser&acc=SRR7763558&display=metadata). Prior informed consent was obtained for the collection of all human tissue specimens in accordance with the Declaration of Helsinki and the University of Chicago's Institutional Review Board's ethical standards. The approved protocols were followed in the experiments. During primary debulking procedures performed at the University of Chicago, six omental tumour samples were collected from patients with recently identified advanced, metastatic HGSOC. Patients undergoing surgical procedures for benign gynaecological disorders yielded seven normal fallopian tube specimens.²⁶

2.2 Clinical tissue samples

Patients undergoing OC at the Third Affiliated Hospital of Chongqing Medical University provided normal fallopian tube epithelial tissues and OC tissues. Skilled pathologists confirmed the results after carefully analyzing the material. These patients were just starting on their treatment regimen and had not yet had any kind of surgery or chemotherapy. Every participant gave their informed consent after the study was approved by the Accreditation Committee of the Third Affiliated Hospital of Chongqing Medical University.

2.3 Sample processing of ovarian cell line RNA-seq

Trizol reagent (Invitrogen) was used to isolate total RNA from OC cells following recognized methods. The PolyATtract mRNA Isolation System (Promega) was subsequently used to separate the messenger RNA (mRNA) from the total RNA. New England BioLabs's NEB Next Ultra RNA Library Prep Kit for Illumina was used to build RNA libraries, and the Illumina NovaSeq 6000 Sequencing System generated an average of about 50 million pairs of paired-end reads, each of which was 150 nucleotides long.

2.4 Analysis of the RNA-seq data

The Illumina HiSeq X Ten platform was used for the sequencing, and the paired-end read length was 150 bp. Using the default settings of TopHat2 (v2.1.1), all reads were mapped to the Homo sapiens genome hg38. Reads that were perfectly matched were the only ones included; for downstream analysis, only reads that were properly paired and uniquely mapped were chosen.¹¹ Utilizing HTSeq (v2.2.1), the number of reads was measured. The differentially expressed genes’ log2 fold changes were determined with the help of DESeq2 (version 1.26.0).

2.5 Analysis of the MeRIP-seq data

Aligning clear reads from HGSOC samples to the Homo sapiens genome hg38 was done using Hisat2 v2.1.1.²⁷ Using the R package exomePeak2 with the default settings, we detected m⁶A-enriched peaks in each m⁶A -IP sample, using the corresponding input sample as a control.²⁸ Using exomePeak2 (version 3.7),²⁸ we compared the m⁶A peaks of metastatic and normal samples. Peaks with p-value < .05 were considered significant and annotated by the annotatePeaks function of HOMER.²⁹ To facilitate genomic visualization, the mapping reads were transformed into normalized BigWig files using Deeptools (version 3.4.1).³⁰ These files were then normalized using CPM, with read coverage for each 100 bp window calculated as the number of mapped reads per million. To see the m⁶A peaks, the Integrative Genomics Viewer (IGV) tool was used. For the integrative analysis of relative m⁶A modification level (tristile) and gene expression, the background transcriptome level was the count of reads in each m⁶A-input sample.

2.6 Identification of enriched motifs within m⁶A peaks

The motifs that were found to be more prevalent in the target m⁶A peak sequences, as opposed to the control sequences (shuffled m⁶A peak sequences), were found to be short (≤8 bp) and ungapped using the Discriminative Regular Expression Motif Elicitation (DREME) suite of MEME.³¹ Using the fastaFromBed tool in BEDTools software v2.28,³² the target m⁶A peak sequences were retrieved from the human reference genome hg38. While maintaining the nucleotide frequencies, the control sequences were created by randomly shuffling each m⁶A peak sequence.

2.7 Cell lines and cell culture

The American Type Culture Collection provided the HEK293T cells, while the Cell Bank of the Chinese Academy of Medical Sciences supplied the OVCAR3 and SKOV3 cell lines. OVCAR3 cells were grown in RPMI 1640 media (GIBCO), HEK293T and SKOV3 cells in DMEM (GIBCO), with 10% fetal bovine serum (GIBCO) added. The cells were cultivated at 37°C in a 5% CO₂ environment. Before conducting experiments, it was ensured that all cell lines were free of mycoplasma contamination.

2.8 Plasmids, cell transfection, and lentiviral infection

The pLKO.1 vector (Addgene) with puromycin selection was used to clone the short hairpin RNAs (shRNAs) targeting METTL3, FTO, YTHDF2, and RPS15A (sh1, sh2), as well as the negative control (shNC). In the supplemental table, you may find the sequences of the shRNAs. A plasmid called pCDH-PURO was used to clone full-length RPS15AP12 (Addgene). The pcDNA3.1 vector (Youbio) was used to overexpress RPS15AP12 and RPS15A cDNAs. Sangon Biotech of Shanghai sold a plasmid extraction kit. Using jetPRIME (Polyplus) as directed by the manufacturer, plasmid transfections were carried out. After 48 h of transfection, the quantities of messenger RNA and protein were measured. Utilizing jetPRIME transfection (Polyplus), lentiviral vectors were delivered into HEK293T cells alongside packaging vectors psPAX2 (#12260, Addgene) and pMD2.G (#12259, Addgene). After 48 h of transfection, the lentiviruses were collected.

2.9 Establishment of RPS15AP12 knockout (KO) cell lines

Through the use of CRISPR/Cas9 technology, RPS15AP12 KO cell lines were produced. To summarize, the LvSG06-RP15AP12-KO plasmid was constructed by cloning sgRNA#1 (GTCTTTCTGCACCACCACAA) and sgRNA#2 (TTTACAAATAAAATGCCCCG) targeting RPS15AP12 into the pCRISPR-LvSG06-puro (GeneCopoeia). To achieve RPS15AP12 KO in OVCAR3 and SKOV3 cells, cells in 6 cm pans were transfected with 2 µg of the LvSG06-RP15AP12-KO plasmid using jetPRIME (Polyplus) at a density of 60–70%. The cells were chosen using 2 µg/mL puromycin 48 h after transfection.

2.10 RNA extraction and quantitative real-time PCR

To extract total RNA, we used the manufacturer's instructions for Trizol (Sigma). Using Vazyme's HiScript II Q RT SuperMix for qPCR, cDNA was generated from whole RNA. We used a QuantStudioDx device from Life Technologies to perform quantitative real-time PCR (RT-qPCR) with SYBR Green Master Mix from Vazyme. The relative expression level of the gene was calculated using the 2^−△△Ct technique. In the supplementary table, the primers for RT-qPCR were purchased from Tsingke Biotech in Beijing.

2.11 Western blot

A BCA detection kit (Solarbio) was used to quantify protein contents after lysing cell pellets with RIPA buffer (Beyotime) that included 1% PMSF. Membranes made of PVDF (Millipore) were used for protein transfer after 10–12% SDS-PAGE separation. After 1 h of blocking with 5% skim milk, the membranes were incubated with primary antibodies at 4°C overnight. Primary antibodies were as follows: METTL3 (1:1000, 15073-1-AP, Proteintech), FTO (1:1000, 27226-1-AP, Proteintech), YTHDF2 (1:2000, 24744-1-AP, Proteintech), RPS15A (1:1000, TA369533S, ORIGENE), MDA5 (1:2000, 21775-1-AP, Proteintech), RIG-1 (1:1000, 20566-1-AP, Proteintech), TBK1 (1:1000, #3504, CST), p-TBK1 (1:1000, 82383-1-RR, Proteintech), IRF3 (1:5000, 11312-1-AP, Proteintech), p-IRF3 (1:1000, 29528-1-AP, Proteintech), GAPDH (1:10000, 60004-1-Ig, Proteintech). Membranes were incubated with the secondary antibody for 1 h at room temperature after being washed three times with TBST. Chemiluminescence on a western blot Imaging System was used to visualize the positive bands.

2.12 Enzyme-linked immunosorbent assays

Growth medium supernatants were collected from the cell samples and subjected to IFN-β measurement by enzyme-linked immunosorbent assays (Human IFN-beta RUNXIN, R&D Systems). The protein concentration of cell samples at harvesting was measured and used to normalize IFN-β read-outs.

2.13 Immunohistochemistry

Prior to heat-mediated antigen retrieval, paraffin-embedded tissue sections underwent deparaffinization. Incubation of the slides with primary antibodies against Ki-67 (AF0198, Affinity, 1:100), Caspase-3 (9662S, CST, 1:100), and RPS15A (TA369533S, ORIGENE, 1:50) followed by 30 min of blocking with goat serum (ZLI-9021, ZSBIO) at 37°C. The slides were then dried overnight at 4°C. The next step was to cure the samples for 45 min at 37°C with biotinylated secondary antibodies. A 30 min incubation with HRP-conjugated streptavidin followed by development with DAB (ZLI-9018, ZSBIO) was administered to the slides. The last step was to mount, dehydrate, and counterstain the sections with hematoxylin. In order to analyze the protein expression, a microscope (Eclipse 8i, Nikon) was used.

2.14 Cell proliferation assay

We used CCK-8 and colony formation assays to measure cell proliferation. The transfected cells were loaded into 96-well plates with a density of 2–3 × 10³ cells/well in order to conduct the CCK-8 test. Incubation was carried out for 3 h at 37°C after adding the CCK-8 solution (K1018, APExBIO) at 0, 24, 48, 72, and 96 h. Microplate spectrophotometer (Bio-Rad) absorbance measurements were taken at 450 and 630 nm. The six-well plates were seeded with 2–3 × 10³ cells for the colony formation experiment, and the media was changed every 4 days. Following an 8–12-day incubation period, the cells were fixed with 4% paraformaldehyde, stained with 0.1% crystal violet, rinsed with PBS, and subsequently counted using a light microscope.

2.15 Cell cycle assay

From the cell cycle kit (C1052, Beyotime), cells were collected and washed twice with PBS. The collected cells were fixed at 70% alcohol at 4°C for 24 h, propidium iodide staining solution was added, and incubated in a warm bath at 37°C away from light for 30 min. Finally, it is tested on the flow cytometer (BD Acurri C6). Modfit 5 software was used to analyze the results.

2.16 Transwell assay

To evaluate migration and invasion, 8 × 10⁴ cells were added to 200 µL of serum-free media in transwell chambers (8 µm, Corning Falcon). For invasion experiments, matrigel coating was utilized and 600 mL of medium containing ten percent FBS was added to the lower chamber. After 24 h, three randomly selected fields were used to count the cells on the bottom side of the membrane after staining them with crystal violet.

2.17 Wound-healing assay

The cells were grown in six-well plates until they reached confluence after being seeded with full media. Using a regular 200 µL pipette tip, a straight incision around 300–500 µm broad was made. The non-adherent cells were removed from the wounded monolayers by washing them twice with 1xPBS. At two prearranged intervals (0 and 12 h), the wound closure was photographed and recorded. The wound healing was measured as described: 12 h migration % = (0 h width – 12 h width of the wound)/(0 h width of the wound).

2.18 Xenograft mice experiment

Chongqing Medical University's Institutional Animal Care and Use Committee gave its approval to all animal experiments. The tumorigenesis assay involved injecting 5 × 10⁶ OVCAR3 cells subcutaneously into the right armpit region of female BALB/c nude mice that were 6 weeks old. We measured the tumour's width and length every three to four days. The mice were sacrificed 28 days after injection, and the weight of the tumours was noted. The formula V = .5×A×B² was used to get the tumour volume, with A and B standing for the longitudinal and latitudinal diameters, respectively. Furthermore, in order to facilitate further experiments, tumour tissues were embedded in 4% formalin. Nude mice were injected intraperitoneally with 5 × 10⁵ OVCAR3 cells transduced with shRNAs or empty vectors, which were suspended in 300 µL of PBS for the metastatic tests. The mice were euthanized after 3 weeks, and the number of metastatic nodules was determined.

2.19 RNA immunoprecipitation

A lysis solution containing RNA immunoprecipitation (RIP), DTT, PIC, and RNase inhibitor was used to lyse about 2 × 10⁷ cells. The buffer contained 20 mM HEPES, 150 mM NaCl, 10 mM KCl, 5 mM EDTA, 5 mM MgCl₂, 0.5% NP40, and 10% glycerol. Centrifuged after 30 min of ice incubation, 5% of the supernatant was set aside as the input sample. Overnight at 4°C, the lysate was incubated with 3 µg of specific or control IgG antibody. The mixture was incubated with rotation at 4°C for 4 h the following day after adding 50 µL of washed protein G beads (Invitrogen, 10004D). To provide quality control, 10% of the immunoprecipitated material was set aside for western blot after five 5 min washes. TRIzol reagent (Invitrogen) was used to extract the RNA that co-precipitated with the beads and input RNA. Then, ethanol was used to precipitate the mixture while glycogen was present. Following this, the standard operating protocol for RT-qPCR analysis was followed.

2.20 m⁶A RNA immunoprecipitation

Magna MeRIP m⁶A kit (Millipore) was used to carry out the MeRIP experiment. First, cells were lysed in RIP buffer. Then, either anti-m⁶A antibody (Abcam) or control IgG antibodies were used for immunoprecipitation overnight at 4°C. Finally, the RNA was purified. The RT-qPCR analysis was performed on the immunoprecipitated RNA.

2.21 RNA stability assay

Actinomycin D (AbMole) was added to transfected OVCAR3 or SKOV3 cells in a 12-well plate at a concentration of 5 µg/mL for 0, 2, 4, and 6 h intervals. After extraction, total RNA was reverse-transcribed and evaluated using RT-qPCR. As a means of measuring RNA abundance, the average Ct value at each time point was set to 0 h. GraphPad Prism 9.5 was used for nonlinear regression curve fitting, namely one phase decay, in order to establish the RNA half-life.

2.22 RNA fluorescence in situ hybridization

OVCAR3 or SKOV3 cells were subjected to RNA fluorescence in situ hybridization (FISH) with probes specific to human RPS15AP12 (RiboBio). RiboBio developed Cy3-labeled RPS15AP12 probes that were custom-designed. As internal controls for nuclear and cytoplasmic localization, respectively, FISH probes for human U6 snRNA and 18S rRNA were utilized.

2.23 Luciferase reporter assay

Co-transfection of pmirGLO, pmirGLO-WT for RPS15A, and pmirGLO-MUT for RPS15A constructs with miR-96-3p mimics or an RPS15AP12 overexpression plasmid was performed using jetPRIME reagent (Polyplus) into OVCAR3 or HEK293T cell lines. After 48 h of transfection, the relative luciferase activity was determined using a double-luciferase reporter assay kit (TransGen) and adjusted to Renilla luciferase activity. The results were shown as a fold change compared with the control groups that were matched, and each experiment was carried out three times.

2.24 Statistic methods

3.1 Integrative multi-omics analysis reveals dysregulation of pseudogenes in OC

Initially, a landscape analysis was conducted to evaluate the differential expression of pseudogenes in Pan-cancer compared with the corresponding normal controls. Approximately 1000–2000 pseudogenes were dysregulated in Pan-cancer, with a significant number upregulated in OC (Figure 1A,B). It was speculated that elevated expression of pseudogenes might trigger functional activation, and subsequently, univariate Cox analysis was performed to validate this prediction. According to the TCGA-OV cohort of RNA-seq, 911 and 878 pseudogenes were identified with prognostic values for OS and PFS, respectively, with higher expression correlating to poorer survival in most cases (Figure 1C).

Next, we investigated the genetic alterations that triggered the dysregulation of pseudogenes. Among 1246 differentially expressed pseudogenes (DEPGs) identified in HGSOC (Figure S1A), genetic changes were observed in only 73 pseudogenes, affecting less than 6% of patients. Of these, only two pseudogenes exhibited mutation frequencies of less than 1%, and four of 72 patients with copy number variation (CNV) had frequencies higher than 5% (Figure S1B). Among the pseudogenes exhibiting the highest frequencies of CNV, TSPY26P, and RPL13P5 showed a positive correlation between CNV and RNA expression (Figure S1C). Nevertheless, no significant correlation was found between their expression and patient survival outcomes (Figure S1D). Landscape statistics of 12 146 pseudogenes for genetic alteration were also conducted, showing that nearly all pseudogenes underwent negligible frequencies of genetic mutation (Figure S1E) and only 6 pseudogenes had frequencies of CNV greater than 10% (Figure S1F). Among those pseudogenes with high frequencies of genetic alteration, only the expression of eight pseudogenes were influenced (Figure S1G). A correlation was noted between the expression of a few pseudogenes and both gene copy number and OS (Figure S1H,I). These results demonstrate the dysregulation of pseudogenes in HGSOC, which may not be primarily due to genomic causes.

3.2 Epitranscriptome-wide profiling of m⁶A modification in pseudogene-derived RNAs

To discover the mechanism underlying the pseudogenes dysregulation in OC, we focused on epigenetic modulation. Since 72% of pseudogenes are processed through retro-transposition of mRNA transcripts, lacking introns, they are not subject to NMD RNA surveillance.¹¹ These NMD-resistant processed pseudogenes acquire de novo m⁶A sites under positive natural selection, establishing m⁶A-mediated degradation as a crucial RNA surveillance mechanism.¹¹ We examined m⁶A modification in pseudogenes across the genome using MeRIP data from HGSOC clinical specimens. m⁶A peaks are widely distributed across pseudogenes on all chromosomes, with a higher concentration on chromosome 1 (Figure 2A,B). Additionally, more m⁶A peaks were located in the coding sequence region (Figure 2C). HOMER analysis indicated that processed pseudogenes predominantly exhibited more m⁶A peaks in the exon region of omental tumour tissues, while more m⁶A peaks were found in the intron and intragenic regions in fallopian tube epithelium tissues used as normal controls (Figure 2D). We also analyzed the prevalence of m⁶A modification in pseudogenes in tumour and normal tissues, finding that fewer pseudogenes were m⁶A-modified in tumour tissues (Figure 2E). Similarly, fewer corresponding cognate parent genes of those m⁶A-modified pseudogenes showed modifications in tumour tissues and all 4078 paired parent genes from 12 146 pseudogenes (Figure 2E). Moreover, a higher abundance of m⁶A peaks was observed in fewer pseudogenes in tumours compared with controls (Figure 2F). Despite the generally low expression of ncRNAs, a higher abundance of m⁶A modification was observed in pseudogenes compared with their cognate parent genes (Figure 2G). These results support a redistribution and potential function of m⁶A modification in pseudogene-derived RNAs.

According to their origin, pseudogenes are classified into four types: processed, unprocessed, unitary, and polymorphic.² Unitary pseudogenes exhibited the highest abundance of m⁶A, while unprocessed pseudogenes had a higher abundance than processed pseudogenes (Figure 2H). We then used a Sankey Plot to visualize the landscape of m⁶A-positive and m⁶A-negative pseudogenes, and their correspondence to function and potential interplay with their cognate parent genes. Pseudogenes with m⁶A modifications had a higher proportion of m⁶A modifications in parent genes in OC, suggesting that more pseudogenes might retain those m⁶A sites from their cognate parent genes during evolution (57.9% vs. 40.3%, p < .001; Figure 2I). Through univariate Cox regression analysis, a higher expression of the majority of pseudogenes was correlated with poorer OS, regardless of pseudogene type and m⁶A modification (Figure 2J). Positive correlations were predominantly observed in most both m⁶A-positive and m⁶A-negative pseudogenes of expression with their paralogs, from which we speculated that ceRNA regulation could serve as a dominant posttranscriptional regulatory mechanism between pseudogenes and their paralogs (Figure 2K). We analyzed pseudogene expression across different m⁶A levels in two major subtypes, finding similar m⁶A-mediated down-regulation; both processed and unprocessed pseudogenes had m⁶A negatively correlated with expression as shown by the cumulative distribution function (CDF) plot (Figure 2L).

To evaluate the influence of m⁶A modification on the relevance between pseudogenes and their paralogs, we analyzed the correlation coefficients of gene expression between pseudogenes and cognate parent genes at varied levels of m⁶A modification in pseudogenes. Processed pseudogenes exhibiting elevated m⁶A levels demonstrated markedly reduced correlation coefficients, a trend that was not evident in unprocessed pseudogenes (Figure 2L). As controls, lncRNAs and mRNAs both showed m⁶A negatively correlated with expression, and higher m⁶A levels correlated with lower correlation coefficients (Figure S2). Overall, pseudogene subtypes could trigger varied m⁶A effects.

3.3 Identification of the critical Diff-m⁶A pseudogene RNAs in OC

We aimed to identify critical m⁶A-regulated pseudogenes with predictive and therapeutic potential. A total of 103 pseudogenes with differential m⁶A (Diff-m⁶A) levels in HGSOC compared with fallopian epithelium tissues were identified, with processed pseudogenes comprising 77.7% (Figure 3A). Most Diff-m⁶A sites were located in the intron, exon, and 3′UTR regions (Figure 3B). The typical RRACH motif (R = G or A; H = A, C, or U) was identified using MEME motif analysis (Figure 3C). Integrative m⁶A and gene expression analysis revealed that 15 pseudogenes exhibited significant differential m⁶A and stabilized RNA levels, with seven showing concurrent differential expression (Figure 3D). An IGV plot illustrated the differential m⁶A modification levels of representative pseudogenes (Figure 3E). RT-qPCR was conducted to confirm their expression in OC cells (Figure 3F). MeRIP-PCR confirmed the m⁶A modification of NPM1P8 and RPS15AP12 in OC cells (Figure 3G). Genomic profiling supported that RNA expression changes in these pseudogenes were not primarily due to genetic alterations, as very low frequencies of mutation and CNV ranging from 2% to 7% were present among only four pseudogenes (Figure S3A). Kaplan–Meier survival analyses showed that FAM86EP, BCASP2, RPL24P8, and RPS15AP12 had significant predictive value for survival (Figure 3H). Notably, RPS15AP12 was the highest ranked among up-regulated Diff-m⁶A pseudogenes in TCGA-OV versus GTEx (Figure S3B). Thus, a set of m⁶A-regulated pseudogenes potentially implicated in OC was identified.

3.4 Pseudogene RPS15AP12-lncRNA promotes OC progression

To address the role of m⁶A-regulated pseudogenes in OC further, we focused on the up-regulated Diff-m⁶A pseudogene RPS15AP12, a processed pseudogene derived from the mRNA reverse transcription of RPS15A (Figure S4A). To determine whether RPS15AP12 is translated into a protein, an Open Reading Frame (ORF) prediction was performed. Intriguingly, a potential ORF of 118 amino acids was identified in RPS15AP12 (Figure S4B). However, FLAG-tagging assays of the putative protein revealed that no corresponding proteins were detected in cells transfected with the RPS15AP12 ORF-FLAG fusion vector, suggesting RPS15AP12 might function as a long noncoding RNA (Figure S4C). Cellular fractionation and RNA-FISH assays revealed that RPS15AP12-lncRNA predominantly resides in the cytoplasm (Figure 4A,B).

In addition to its high expression in OC, RPS15AP12 was also overexpressed in the ascites of OC patients (Figure 4C), indicating its potential relevance to invasiveness. RT-PCR also confirmed that RPS15AP12 had higher expression in the OC samples than in the FTE samples (Figure S4D). To investigate the biological functions of RPS15AP12-lncRNA in OC, we knocked out the RPS15AP12 pseudogene in OC cells using the CRISPR-Cas9 genome-editing tool (Figure S4E,F). The growth, migration, invasion, and colony formation of OC cells were all dramatically reduced when RPS15AP12 was depleted (Figures 4D–G; Figure S4G) and the cell cycle of OC cells was changed (Figure S4I,J). Likewise, these carcinogenic tendencies were enhanced by the forced expression of RPS15AP12 (Figures 4H–K; Figure S4H). Using a subcutaneous xenograft model and a peritoneal metastasis model, we observed that depletion of RPS15AP12 substantially reduced subcutaneous xenograft formation both in volume and weight by OC cells, as well as the metastasis of xenograft tumours (Figure 4L,P). Forced expression of RPS15AP12 enhanced subcutaneous xenograft formation and metastasis (Figure 4 M,Q). Furthermore, immunohistochemistry (IHC) staining showed that KO of RPS15AP12 decreased the protein level of Ki-67 but increased the protein level of Caspase-3 in subcutaneous xenografts, whereas overexpression of RPS15AP12 had the opposite effects (Figure 4N,O). These results demonstrate that RPS15AP12-lncRNA facilitates OC progression.

3.5 FTO-mediated demethylation of m⁶A contributes to upregulation of RPS15AP12-lncRNA in OC cells

Integrative analysis indicated that m⁶A hypermethylation tended to reduce transcript levels in HGSOC compared with normal tissues, though not statistically significant (Figure S5A). Since RPS15AP12 was observed to be hypomethylated and upregulated in OC (Figure 3D), we investigated whether m⁶A modification regulated RPS15AP12. The Spearman correlation analysis of expression showed that RPS15AP12 was positively correlated with FTO and negatively correlated with METTL3 and YTHDF2 (Figure S5B), but no significant correlation was found between RPS15AP12 with ALKBH5 and METTL14 (Figure S5B). RIP assays using antibodies specific to METTL3 and FTO demonstrated that both could bind to RPS15AP12-lncRNA (Figure 5A,B). meRIP-PCR results showed that METTL3 knockdown decreased m⁶A modification on RPS15AP12-lncRNA, while FTO knockdown increased it (Figure 5C ,D). Upon METTL3 or FTO depletion in OC cells, RPS15AP12-lncRNA expression increased and decreased, respectively (Figure 5E,F). RNA decay assays indicated that METTL3 knockdown shortened the half-life of RPS15AP12-lncRNA, whereas FTO knockdown enhanced its stability (Figure 5G,H). YTHDF2, known as an m⁶A reader involved in RNA degradation, was found to increase both expression and stability of RPS15AP12 when FTO was knocked down, suggesting YTHDF2's involvement in m⁶A-mediated degradation of RPS15AP12-lncRNA (Figure 5I–K; Figure S5C–E).

3.6 RPS15AP12-lncRNA acts as a miRNA sponge to positively regulate RPS15A expression

RPS15AP12 was primarily localized in the cytoplasm (Figure 4A,B), implicating it may work as a miRNA sponge. The positive correlation between RPS15AP12 and RPS15A indicated potential ceRNA regulation (Figure 6A). RT-qPCR and western blot analyses revealed that RPS15A expression decreased following RPS15AP12 KO (Figure 6B,C). IHC staining showed that RPS15AP12 KO significantly reduced RPS15A protein levels, whereas RPS15AP12 overexpression significantly increased them (Figure 6D). A RIP assay using an anti-AGO2 antibody demonstrated that RPS15AP12, unlike control IgG, could bind with miRNAs (Figure 6E). Three bioinformatics databases (miRcode, TargetScan, and ENCORI) were utilized to predict miRNAs and their binding sites on RPS15A and RPS15AP12, revealing several potential co-binding miRNAs (Figure 6F). Among these, miR-96-3p showed a negative correlation with RPS15A expression (Figure S6A), and its inhibition significantly increased RPS15A expression (Figure S6B). The focus then shifted to miR-96-3p's role in RPS15AP12-lncRNA-mediated upregulation of RPS15A. RPS15AP12 contains the sequence UUGGCA, shared with RPS15A, matching the seed region of miR-96-3p (Figure 6G). Experimentally, a miR-96-3p mimic reduced RPS15AP12 expression at the RNA level in OC cells, while a miR-96-3p inhibitor increased RPS15A expression (Figure 6H,I). In addition, the down-regulation of RPS15A generated by RPS15AP12 KO could be undone by an inhibitor of miR-96-3p, while the upregulation of RPS15A caused by RPS15AP12 overexpression could be undone by miR-96-3p mimics (Figure 6J). To validate the binding of miR-96-3p to RPS15A, the 3′UTR region of RPS15A or a mutant version of it without the miR-96-3p binding site was inserted into a luciferase reporter vector. The results showed that luciferase activity was significantly reduced when co-transfected with miR-96-3p, but it remained unaffected when the binding site was mutated (Figure 6K–M). All things considered, these results point to the competitive binding of RPS15AP12-lncRNA to miR-96-3p as a mechanism by which RPS15A expression is enhanced.

To assess the impact of the RPS15AP12-lncRNA-miR-96-3p-RPS15A axis on OC cell proliferation, RPS15AP12-depleted OC cells were either overexpressed with RPS15A or treated with a miR-96-3p inhibitor. Depletion of RPS15AP12 significantly reduced colony formation in OC cells, but either re-expression of RPS15A or inhibition of miR-96-3p lessened the suppression of cell proliferation caused by RPS15AP12 KO (Figure S6C). Comparably, the application of either a miR-96-3p inhibitor or the re-expression of RPS15A resulted in a partial reversal of the inhibitory impact that RPS15AP12 KO had on the invasion and migration of OC cells (Figure S6D).

3.7 RPS15AP12-lncRNA suppresses innate immune response in OC

As expected, RPS15A was overexpressed in HGSOC compared with normal controls (Figure S7A), and RT-qPCR assays also confirmed its high expression as well as its strong positive correlation with RPS15AP12 expression in OC (Figure S7B,C). Besides, higher expression of RPS15A was correlated with poorer survival of OC patients (Figure S7D,E). Therefore, RPS15A could be regulated by RPS15AP12 to exhibit an oncogenic role either. Analogous to the KO of RPS15AP12, the knockdown of RPS15A markedly diminished the proliferation, migration, and invasion of OC cells and changed cell cycles (Figure S7F–L). Overexpression of RPS15A promoted the migration of OC cells (Figure S7M,N). Knockdown of RPS15A could also impair the ability of subcutaneous tumour formation and intraperitoneal metastasis of OC cells in xenografts (Figure S7O,Q). On the contrary, overexpression of RPS15A enhanced the ability of subcutaneous tumour formation and intraperitoneal metastasis of OC cells in xenografts (Figure S7P,R). To investigate the downstream targets of RPS15AP12 and RPS15A in OC cells, RNA sequencing was conducted after the knockdown of RPS15A or the KO of RPS15AP12. Knockdown of RPS15A caused the upregulation of 1049 genes and the downregulation of 1050 genes. In contrast, the KO of RPS15AP12 resulted in 248 upregulated genes and 344 downregulated genes (Figure 7A). The overlapping analysis identified 51 upregulated genes and 57 downregulated genes common to both RPS15AP12 KO and RPS15A knockdown (Figure 7B). Function enrichment analysis revealed that the differentially expressed genes (DEGs) exhibited significant enrichment in pathways associated with cancer and innate immunity (Figure S7S). Further enrichment network analysis revealed hub genes involved in innate immune-related pathways including IFN signalling, response to viruses, and IFN-α/β signalling, as well as links between them (Figure 7C). A heatmap displayed the expression patterns of 25 shared DEGs related to apoptosis, proliferation, and innate immunity (Figure 7D). RT-qPCR confirmed significant upregulation of RSAD2, IL6ST, UBE2E1, IFIT2, IFIH1, FGF2, DDX58, IFIT3, and CCN3 following either RPS15AP12 KO or RPS15A knockdown (Figure 7E,F). DDX58 (RIG-I) and IFIH1 (MDA5) are RNA sensors upstream of MAVS, whose activation triggers downstream IFN signalling pathways. The examination of RIG-I and MDA5 expression, along with the phosphorylation levels of TBK1 and IRF3, revealed an increase in response to either RPS15AP12 KO or RPS15A knockdown. This resulted in a markable enhancement of protein expression for RIG-I and MDA5 as well as elevated phosphorylation levels of TBK1 and IRF3 in OC cells (Figure 7G,H). We also found IFN-β levels were upregulated by RPS15AP12 KO or RPS15A knockdown (Figure S7T). These findings indicate that the RPS15AP12-RPS15A axis targets innate immune signalling in OC cells.