Full-length transcriptomic profiling reveals age-associated isoform remodeling and altered coding potential in the mouse ovary

Abstract

Ovarian aging, marked by a progressive diminution of oocyte quality and quantity, is a major contributor to declining female fertility and age-related reproductive disorders. However, transcript-level changes underlying this process remain incompletely understood. In this study, we applied Oxford Nanopore long-read RNA sequencing to profile full-length transcripts from granulosa cells and oocytes of young (6–8 weeks) and aged (10 months) mice, complemented by Illumina short-read sequencing for orthogonal support. We performed transcript annotation, differential expression analysis, alternative polyadenylation (APA) analysis, and weighted gene co-expression network analysis (WGCNA) to investigate age-associated transcriptomic changes. Comprehensive annotation classified 130,730 high-confidence transcripts, including over 100,000 putative novel isoforms, and revealed that aging was associated with a shift toward isoforms with lower predicted coding potential. Exploratory enrichment analysis suggested that transcripts with lower predicted coding potential were associated with biological processes such as protein synthesis and chromosome segregation. APA analysis identified age-associated 3'UTR shortening. Transcript-level differential expression and isoform-switching analysis uncovered 795 significant switching events across both cell types, frequently associated with predicted open reading frame changes and potential protein-domain loss. Exploratory WGCNA highlighted modules associated with aging and cell-type specificity, including an Esr1-derived hub transcript, TALONT000180938, from a gene previously linked to ovarian function and disease. Many disease-associated genes exhibited cell-type-specific isoform usage, with several novel isoforms undetectable at the gene level. Our results indicate that long-read sequencing improves isoform-level resolution of ovarian transcriptomic diversity and identifies candidate aging-associated transcript alterations that may be relevant to reproductive decline.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13048-026-02193-9.

Introduction

Reproductive aging, defined by the progressive decline in ovarian function, poses a major challenge to women's health. The ovary is among the first organs to exhibit age-related functional deterioration, with noticeable decline occurring as early as the age of 30 years [1]. Age-associated reductions in follicle and oocyte quality not only impair female fertility and increase the risk of infertility, but also elevate susceptibility to reproductive disorders such as premature ovarian failure (POF) [2, 3], ovarian endometriotic cysts [4], and ovarian cancer (OC) [5]. Follicles—the basic functional units of the ovary—are composed of oocytes and their surrounding granulosa and theca cells [6]. The growth, maturation, and maintenance of oocytes quality rely heavily on the metabolic support, signaling molecules, and nutrients provided by granulosa cells [6], while granulosa cell proliferation, secretory activity, and steroidogenesis are in turn regulated by oocyte-derived factors. This bidirectional communication is critical for proper follicular development and ovulation, yet becomes progressively impaired with advancing age [7]. Despite its importance, the transcriptomic mechanisms governing this regulatory interplay, and the ways in which aging disrupts it, remain poorly understood at the systems level.

Although short-read RNA sequencing has been used to investigate transcriptomic changes during ovarian aging, its limited read length hinders the accurate reconstruction of complete transcript structures. This limitation is particularly evident for resolving complex alternative splicing events, diverse transcription termination patterns such as alternative polyadenylation (APA), and precise annotation of novel transcripts. Moreover, most existing studies have focused primarily on overall gene expression differences [8–10], with insufficient exploration of deeper regulatory layers, including cell type–specific transcript structural variations, alterations in coding potential, and changes in noncoding RNA expression. In addition, oocytes and granulosa cells may display heterogeneous transcriptional responses to aging, highlighting the need for higher-resolution approaches capable of distinguishing these cell type–specific patterns.

In recent years, third-generation sequencing technologies—particularly long-read single-molecule sequencing based on Oxford Nanopore and PacBio platforms—have shown marked advantages in transcriptomic research. These technologies enable direct acquisition of full-length transcript sequences, precise resolution of complex alternative splicing events, and identification of previously unannotated transcript structures [11]. Unlike short-read sequencing, long-read approaches eliminate information loss caused by fragment assembly, thereby providing robust technical support for comprehensive characterization of transcript isoform diversity [12]. To date, this technology has been widely applied to explore transcript diversity in embryonic development [13], neurological disorders [14], and various cancers [15]. However, systematic application of long-read sequencing to deeply interrogate full-length transcript structures and expression dynamics in the ovary’s core functional cells—oocytes and granulosa cells-remains lacking. At cell-type-resolved resolution, the ability to capture transcript-level alterations, reconstruct splicing patterns, and identify potential disease-associated isoforms during aging represents a critical gap in the field.

Building on this rationale, we employed Oxford Nanopore long-read sequencing to generate full-length transcriptomic profiles of oocytes and granulosa cells from young and aged mice. The resulting transcripts were comprehensively annotated, including transcript classification, repeat element characterization, sequence conservation assessment, and coding potential evaluation. Our analyses revealed age-associated changes in predicted coding potential and candidate biological processes such as protein synthesis and chromosome segregation. Transcript-level expression analysis further uncovered age-related alterations not detectable by conventional gene-level approaches, along with numerous putative novel transcripts and genes, thereby refining the current mouse ovary transcriptome annotation. In addition, we integrated APA profiling to delineate the composition and expression characteristics of the ovarian transcriptome. Collectively, these findings underscore the value of long-read sequencing in capturing transcriptomic diversity and resolving candidate aging-associated transcriptomic alterations.

Methods

Animals

Female ICR mice, including young (6–8 weeks old) and aged (10 months old) groups, were obtained from Beijing Vital River Laboratory Animal Technology Co., Ltd. This age selection was based on previous evidence that ovarian reserve and reproductive function decline progressively during mouse aging [16, 17], while still allowing recovery of sufficient GV oocytes and granulosa cells for transcriptome profiling. All animals were maintained under specific pathogen-free conditions with a 12 h light/dark cycle and ad libitum access to food and water. The study was approved by the Ethics Committee of Reproductive Medicine of Reproductive Hospital of Shandong University (2022–138), and all experiments were conducted in accordance with the relevant regulations and guidelines of the Ethics Committee. In the present study, mice were sacrificed by cervical dislocation.

Granulosa cell and oocyte isolation

Granulosa cells (GCs) and germinal vesicle (GV) oocytes were isolated from young and aged mice by mechanically dissociating ovaries using a 30 G needle. Specifically, follicles on the ovarian surface were punctured in M2 solution to release cumulus-oocyte complexes (COCs). COCs were collected with a pulled glass Pasteur pipette and transferred to a clean dish containing M2 solution. The COCs were repeatedly aspirated and expelled to separate oocytes from GCs. GV oocytes were collected, the zona pellucida was removed using Acidic Tyrode's solution, and oocytes were transferred to PCR tubes. The GCs surrounding GV oocytes were collected into low-binding centrifuge tubes and centrifuged to obtain cell pellets.

RNA extraction, library construction, and sequencing

Following removal of the zona pellucida with EmbryoMax Acidic Tyrode's Solution (MR-004, Sigma), fifty oocytes were transferred into lysis buffer (0.2% Triton X-100 supplemented with 4 U RNaseOUT). Granulosa cells (GCs) were collected into low-binding centrifuge tubes, washed twice with DPBS, and total RNA was extracted using FreeZol Reagent (Vazyme, R711) according to the manufacturer’s protocol. According to previous methods [18], Poly(A) + mRNA was used to synthesize full-length cDNA with the SuperScript IV Reverse Transcriptase (Thermo Fisher, 18,090,050), and the pre-amplified cDNA was amplified using the KAPA HiFi Uracil + Kit (Roche, KK2802). For short-read RNA-seq, libraries were generated using the TruePrep DNA Library Prep Kit V2 for Illumina (Vazyme, TD502) and TruePrep Index Kit V3 for Illumina (Vazyme, TD203), and libraries that passed quality control were sequenced on the Illumina NovaSeq X Plus platform. For long-read Nanopore sequencing, cDNA was PCR-amplified for 13–14 cycles with barcoded adapters from the Oxford Nanopore PCR Barcoding kit (SQK-PBK004), 1D sequencing adapters were ligated to the cDNA, and libraries were loaded onto an R10.4.1 FLO-PRO114M flow cell for sequencing on the PromethION platform.

Quality control and data processing of short-read sequencing

Paired-end Illumina sequencing was conducted, and raw reads were quality-filtered and adapter-trimmed using Trimmomatic (v0.39) [19] in paired-end mode. The resulting cleaned paired-end FASTQ files were aligned to the mouse reference genome (GRCm39) using HISAT2 (v2.2.1) [20]. Post-alignment, unmapped reads and those with low mapping quality (MAPQ < 10) were removed using SAMtools [21], and the remaining mapped reads were coordinate-sorted to produce final BAM files. Gene expression quantification and transcript assembly were performed with Cufflinks (v2.2.1) [22] using the reference GTF annotation generated from long-read sequencing data, yielding corresponding read count and transcript assembly outputs.

Quality control and data processing of full-length transcriptome

Raw Nanopore sequencing data in fast5 format were basecalled and converted to FASTQ files using Dorado (v0.3.4 HAC model) [23], retaining reads with quality scores greater than 7. Full-length transcript sequences were identified and oriented using Pychopper (v2.7.10). The Full-length reads were then aligned to the UCSC mouse reference genome (mm39) with minimap2 [24] using the recommended long-read parameters (-ax splice -uf -k14 –secondary = no –MD), generating transcript assemblies for granulosa cells and oocytes.

Full-length transcriptome classification and quantification

Transcript and gene classification and quantification were performed using TALON (v6.0) [25]. Prior to TALON analysis, aligned reads were corrected for non-canonical splice sites using TranscriptClean (v2.1) [26] with default parameters. The talon_label_reads function was applied to annotate the poly(A) proportion within 20 bp windows for each read after alignment; reads with proportions greater than 0.5 were considered to contain internal primers and were excluded from downstream analyses. A TALON initialization database was then generated using Gencode vM36 as the reference annotation. Reads were mapped to the TALON database using default parameters. TALON classified transcripts into six major categories: Known, ISM, NIC, NNC, Antisense, and Intergenic. Newly discovered transcripts (all novel categories except known) were retained only if supported by at least five reads in the database; otherwise, they were excluded. Transcript quantification was performed with talon_filter_transcripts using the filtered database, and the corresponding GTF files were generated with talon_create_GTF. To improve the reliability of the transcript annotations, we further filtered for transcripts with TPM ≥ 0.1 in at least half of the samples in both granulosa cells and oocytes for subsequent analyses.

Coding potential prediction and repeat element detection

Coding potential of the obtained transcripts was assessed using CPAT [27] with default parameters. Following the authors’ guidelines, open reading frames (ORFs) with prediction scores greater than 0.44 were classified as protein-coding. Protein-coding ORFs were translated into amino acid sequences using the Biostrings package [28], and protein domain prediction was performed by searching these sequences against the Pfam database [29] using pfam_scan.py with default parameters. Repeat elements within transcript sequences were identified using RepeatMasker [28] with default parameters.

Conservation analysis

Conservation scores were obtained from the mm39.PhastCons.35way.bw file downloaded from UCSC, which was generated from multiple sequence alignments of the mouse genome (mm39) with 34 other vertebrate genomes, including chicken, cattle, dog, chimpanzee, and human. Background sequences were generated by randomly sampling 20,000 fragments of 1 kb each using the makewindows function in BEDtools [30]. Conservation scores for each transcript type, as well as for background sequences, were computed using ComputeMatrix, and the results were visualized with plotProfile.

Transcript switching events and alternative polyadenylation analysis

Transcript-level alternative splicing analysis for the identification of transcript switching events was performed using IsoformSwitchAnalyzeR (v2.8.0) [31, 32]. The official two-step analysis procedure was followed, incorporating CPAT-predicted ORF coding potential scores and Pfam-based protein domain predictions. SignalP [33] and IUPred2 [34] were applied to identify signal peptides and intrinsically disordered regions (IDRs) within ORFs, respectively. Transcript switching events with FDR < 0.05 and |dIF|> 0.1 were considered significant, and gene functional enrichment was performed using the extractConsequenceEnrichment function. Alternative polyadenylation (APA) analysis was conducted with DaPars2 [35] based on long-read transcriptomic data, calculating the percentage of distal polyadenylation site usage index (PDUI) for each gene. PDUI differences between aged and young granulosa cell and oocyte samples were compared, and genes with |ΔPDUI|> 0.1 were defined as showing 3'UTR lengthening or shortening. Because this PDUI threshold is effect-size based, APA results were interpreted as candidate age-associated shifts rather than definitive functional events.

Differential expression analysis and enrichment analysis

Differential expression analysis was performed using the R package DESeq2 [36]. For transcripts, differential expression between aged and young granulosa cell and oocyte samples was assessed using the criteria |log₂FC|> 1 and FDR < 0.05. Gene-level expression values were calculated from the summed expression of their constituent transcripts, and differential gene expression was evaluated using the same thresholds. Functional enrichment analysis was performed using clusterProfiler with Gene Ontology (GO) biological process terms. Enrichment P values were calculated using the hypergeometric test implemented in clusterProfiler [37], and Benjamini–Hochberg adjusted P values were also obtained. Because several comparisons yielded limited numbers of FDR-significant genes or transcripts, selected GO terms were visualized using nominal enrichment P values for exploratory biological interpretation. These enrichment results were therefore used to identify candidate biological processes associated with aging-related transcriptional changes and were not treated as definitive pathway-level evidence.

GO semantic similarity analysis

To compare enriched biological processes among differentially expressed gene (DEG), differentially expressed transcript (DET), and differential transcript usage (DTU) gene sets, Gene Ontology (GO) semantic similarity analysis was performed. For these comparisons, only GO biological process (BP) terms were utilized. Pairwise GO-term semantic similarity was calculated using the Rel method implemented in the rrvgo package [38], with org.Mm.eg.db serving as the annotation database; the resulting similarity between two GO-term sets was subsequently summarized as the mean best-match similarity in both comparison directions. This analysis was conducted separately for granulosa cells and oocytes to evaluate cell-type-specific DEG, DET, and DTU-associated BP profiles. Furthermore, for DET and DTU events, enriched BP profiles derived from known and novel isoforms were compared, and the direct GO-ID overlap between known- and novel-derived GO profiles was quantified using the Jaccard index.

WGCNA analysis of transcript co-expression networks

Weighted gene co-expression network analysis (WGCNA) was performed using the R package WGCNA (v1.72–1) [39] to identify transcript co-expression modules associated with aging and cell type. Transcript expression values from filtered transcript isoform quantification results were transformed to log₂TPM, and transcripts with low expression (average log₂TPM ≤ 1 across all samples) were removed. The expression matrix was constructed from 11 samples, comprising aged and young granulosa cell (GC) and oocyte groups. Outlier samples were detected by hierarchical clustering using average linkage and Euclidean distance metrics. Based on scale-free topology criteria (R2 > 0.9), an optimal soft-threshold power of 18 was selected using the pickSoftThreshold function. An unsigned network was then built using the blockwiseModules function, and the topological overlap matrix (TOM) was computed and stored for downstream visualization. Modules were assigned unique color labels, and module eigengenes (first principal component of each module) were calculated.

Transcript validation and support

Transcript 3′-end support was performed using the PolyASite 2.0 database [40], and 5′-end support was conducted using CAGE signals from the FANTOM5 database [41]. Genomic coordinates for poly(A) signals and CAGE peaks were obtained in the mm10 assembly from PolyASite and FANTOM5, respectively, and converted to mm39 coordinates using LiftOver [42] to ensure consistency with the reference genome used in this study. Selected key transcripts identified during the analysis were further validated by PCR. To specifically validate the ISM isoform, the HiScript-TS 5'/3' RACE Kit (Vazyme, RA101) was employed. RACE PCR primers were designed using the Vazyme Bioinformatics Cloud Platform (http://cloud.vazyme.com:83/) and SnapGene v8.0. Intergenic and NIC isoforms were validated by Phanta SE Super-Fidelity DNA Polymerase (Vazyme, P521). RNA and cDNA were extracted as described above. The specific primer sequences are shown in Supplementary Table 3 and Supplementary Table 4.

Results

Long-read RNA-seq reveals the full-length transcriptomic landscape of the mouse ovary

Oocytes and granulosa cells were isolated from 11 biological samples, comprising aged (10 months) and young (6–8 weeks) groups: two aged and three young oocyte samples, along with three aged and three young granulosa-cell samples. Both Oxford Nanopore Technologies (ONT) and Illumina sequencing were subsequently performed (Fig. 1A). After removal of low-quality and short reads, ONT produced 4.802–6.517 GB of data per sample, with a mean read length of 780 bp and an N50 of 1,184 bp; 84.78%−93.69% of full-length reads per sample aligned to the reference genome (Supplementary Table 1). Illumina sequencing yielded 4.324–7.241 GB per sample, with 86.70%−90.80% of reads mapping to the reference genome. In total, 66,119,964 short reads and 77,971,368 long reads were obtained.

Fig. 1.

Long-read sequencing unveils the complex landscape of transcriptomic remodeling during ovarian aging. A Workflow illustrating sample collection, sequencing analysis, and subsequent association with disease mechanisms. B TALON classification of transcripts identified by third-generation long-read sequencing into six categories. C Heatmap showing the Pearson correlation between oocytes and granulosa cells from aged and young groups in the third-generation long-read dataset. D Principal Component Analysis (PCA) of dimensionality reduction based on the third-generation long-read data. E Statistics of transcript numbers across different categories. F Statistics of the number of transcripts per gene in different categories. G Distribution of transcript lengths. H Distribution of exon numbers per transcript. I Ranked plot of unique transcript counts per gene. Each dot represents one gene. Green, yellow, and black dots indicate genes assigned to the Known, Antisense, and Intergenic categories, respectively; the predominance of green dots reflects the larger number of Known genes. Black labels indicate representative high-transcript-count genes, with “TALON”-prefixed labels denoting novel genes identified by TALON. Red labels indicate genes associated with ovarian aging-related diseases

Long-read RNA-seq data were processed using the TALON pipeline to generate high-quality transcript annotations (Fig. 1C, D). TALON categorized full-length transcripts into six classes: Known (matching a GENCODE vM36 transcript), ISM (incomplete splice match to a known transcript with novel 5' or 3' ends), NIC (novel in catalog, with new junction combinations at known splice sites), NNC (novel not in catalog, overlapping an annotated transcript but containing at least one novel splice site), Antisense (overlapping an annotated transcript on the opposite strand), and Intergenic (no overlap with annotated transcripts) (Fig. 1B; Supplementary Fig. 1A).

A total of 148,902 transcripts were identified from 43,539 genes (Supplementary Fig. 1B). After filtering out low-abundance isoforms (TPM < 0.1 in fewer than half of the samples within each group), 130,730 high-confidence transcripts from 31,349 genes were retained for downstream analysis. To evaluate the robustness of this filtering strategy, we further compared increasingly stringent expression thresholds and external evidence support rates. This analysis showed that stricter filters increased the proportion of transcripts supported by CAGE, PolyASite, or splice-junction evidence, but also reduced the number of retained isoforms, supporting the use of the current threshold as a balance between transcriptome coverage and annotation confidence (Supplementary Fig. 2). Of these, 21,558 matched known isoforms, whereas 109,172 were novel. In total, 112,473 transcripts originated from annotated genes, and 18,257 from previously unannotated genes (Fig. 1E, F). Compared with known isoforms, novel isoforms were generally shorter and contained fewer exons (Fig. 1G, H). Newly identified genes also tended to encode a larger number of isoforms compared with annotated genes (Fig. 1F).

To assess potential relationships with ovarian-related diseases, we curated genes associated with ovarian cancer, premature ovarian failure, and ovarian endometriosis by integrating data from the MalaCards and OMIM databases (Supplementary Table 2) [43, 44]. We then quantified their transcript numbers. These disease-associated genes generally exhibited fewer isoforms, suggesting lower splicing complexity and more conserved splicing patterns (Fig. 1I). Collectively, these findings demonstrate that the mouse ovarian transcriptome is diverse, and its annotation can be substantially refined through full-length transcriptomic approaches.

Comparative analysis of known and novel isoforms uncovers genomic features and coding capacities in the mouse ovary

To further characterize transcripts identified in mouse granulosa cells and oocytes, we assessed coding potential using CPAT, with transcripts scoring > 0.44 classified as protein-coding, following the tool's recommended threshold. Known isoforms exhibited the highest coding potential, significantly exceeding that of novel isoforms (Wilcoxon rank-sum test, p < 2.2 × 10⁻1⁶; Fig. 2A, C). Among transcript classes, coding potential was greatest in known isoforms, followed by NIC, NNC, and ISM, all of which were significantly higher than Antisense and Intergenic isoforms. Transcript length positively correlated with coding potential (Fig. 2A).

Fig. 2.

Genomic features, coding potential, and evolutionary conservation of known and novel ovarian isoforms. A Coding potential of transcripts (transcripts with a predicted coding potential > 0.44 were classified as protein-coding). B Expression levels of transcripts in different categories (Wilcoxon rank-sum test, Benjamini–Hochberg adjustment). C Coding efficiency of transcripts and the number of predicted protein domains translated from them. D Conservation scores of transcripts in different categories. E Proportional distribution of different families of repetitive elements (SINEs, LINEs, LTRs, etc.) within four types of exons: First Exon, Internal Exon, Last Exon, and Single Exon. The analysis is shown for each isoform category. The "Total" percentage on the right of each group indicates the total proportion of the exon's length that is occupied by repetitive elements

We next compared the expression levels between protein-coding and noncoding isoforms using Bonferroni-corrected Wilcoxon tests (Fig. 2B). Within the known class, coding isoforms displayed significantly higher expression levels than noncoding isoforms. In contrast, the opposite pattern was observed in ISM, NIC, and NNC classes, whereas Antisense isoforms showed no significant difference between coding and noncoding forms. In the Intergenic class, only a small subset of transcripts (n = 30; Fig. 2B, C) was predicted to be protein-coding, and their expression levels were comparable to that of intergenic noncoding transcripts (Wilcoxon with Bonferroni correction, p > 0.05).

Protein domain prediction using PfamScan revealed that known isoforms encoded proteins with a greater number of domains (Fig. 2C), consistent with their evolutionary conservation and functional complexity. In contrast, novel isoforms were generally domain-poor, with this trend particularly pronounced in Antisense and Intergenic classes, suggesting that many of these transcripts may exert regulatory roles primarily at the RNA level. Next, we assessed the evolutionary conservation of each transcript category using phastCons scores based on a 35-vertebrate multiple alignment [45]. The analysis showed that transcripts associated with known loci (Known, ISM, NIC, NNC) exhibit high conservation in their exonic regions (Fig. 2D). Conversely, transcripts from novel loci (Antisense, Intergenic) had significantly lower conservation, approaching background levels. Notably, transcripts from known loci displayed sharp conservation peaks near their transcription start sites (TSS) and transcription end sites (TES), which aligns with the conservation features of promoter and 3' UTR regions in the mouse genome. Overall, the conservation patterns of transcripts were highly consistent with the coding potential analysis, where higher conservation correlated with a greater likelihood of being protein-coding (Fig. 2C, D).

Finally, to investigate the contribution of transposable elements (TEs) to the transcriptome, we systematically analyzed the repetitive sequence content of exons using RepeatMasker. The analysis revealed a strong bias in the distribution of TEs within exons. Across all transcript categories, internal exons consistently showed the lowest TE coverage, while TEs were significantly enriched at both the 5' and 3' ends of transcripts (Fig. 2E). Specifically, known isoforms showed a significant 3' enrichment (First Exon: 13.4% vs. Last Exon: 36.3%). In stark contrast, novel isoforms exhibited a strong 5' enrichment (e.g., NNC, First Exon: 24.2% vs. Last Exon: 14.3%). This 5' enrichment in previously unannotated isoforms, which was primarily composed of LTRs, suggests that these elements are major contributors to the formation of transcription start sites and first exons. This finding is consistent with our conservation analysis, which shows that these TE-driven 5' regions in novel isoforms also have significantly lower sequence conservation (Fig. 2D, E).

Widespread 3′UTR shortening is associated with ovarian aging

To further refine the characterization of ovarian transcriptomic complexity, we evaluated support for 5' transcription start sites (TSS) and 3' transcript end sites (TES) using CAGE (FANTOM5) and polyadenylation signals (PolyASite 2.0), respectively. Known isoforms displayed the strongest external support, with approximately 60% supported at both 5' and 3' ends (Fig. 3A). Subsets of novel isoforms, particularly ISM, NIC, and NNC, showed partial TSS/TES support, indicating biological plausibility and potential functional relevance. In contrast, Antisense and Intergenic isoforms generally lacked such signals, consistent with atypical or low-abundance transcription events (Supplementary Fig. 1A).

Fig. 3.

Assessment of transcript integrity and age-associated 3'UTR usage patterns. A Support of transcript start sites (TSS) and end sites (TES) using CAGE and PolyASite databases, respectively. The bars show the percentage of isoforms in each category with support from CAGE (orange), PolyA signals (red), or both. B Short-read support of transcript expression using short-read Illumina sequencing. The stacked bar chart shows the proportion of isoforms in each category that were expressed (TPM > = 1), lowly expressed (0 < TPM < 1), or not expressed in the short-read data. C Genome browser view of the Igf1 locus on chromosome 10, showing both known (GENCODE) and novel (TALONT) isoforms identified in this study. Tracks for CAGE and PolyA signals are shown at the top. D Scatter plots comparing the Percentage of Distal Polyadenylation site Usage Index (PDUI) between aged and young samples in granulosa cells (left, purple) and oocytes (right, orange). Points above the diagonal line represent transcripts with increased distal polyadenylation-site usage in aged samples, whereas points below the line represent transcripts with reduced distal polyadenylation-site usage. Candidate age-associated PDUI shifts were defined using an effect-size threshold of |ΔPDUI|> 0.1

Short-read RNA-seq data, aligned using the long-read-derived GTF annotation, further supported expression levels across all transcript classes (Fig. 3B). Both start/end site evidence and short-read expression collectively provided orthogonal support for the majority of known and novel isoforms. Furthermore, we performed experimental validation on a representative set of transcripts selected from key categories. We employed RT-PCR to verify their internal splicing structures and 5'/3' RACE to confirm their transcriptional start and end sites. All tested transcript models were experimentally supported: RT-PCR assays produced amplicons of the predicted sizes, and RACE sequencing results were consistent with the annotated transcription start sites (TSS) and polyadenylation sites (Supplementary Fig. 9). As a representative example, Igf1, a key regulator in ovarian aging whose downregulation promotes granulosa-cell apoptosis and disrupts oocyte mitochondrial function, was found to express seven known and ten novel isoforms, most of which were supported by both CAGE and polyadenylation signals (Fig. 3C, Supplementary Fig. 12).

We next investigated age-associated differences in 3′ untranslated region (3′UTR) usage in granulosa cells and oocytes by analyzing distal polyadenylation-site usage (Fig. 3D). Polyadenylation site usage index (PDUI) values, calculated using DaPars2, reflect relative polyadenylation-site selection, with higher values indicating preferential use of distal polyadenylation sites and longer 3′UTRs. Using an effect-size threshold of |ΔPDUI|> 0.1, we defined 1,293 granulosa-cell transcripts and 1,032 oocyte transcripts as candidate age-associated PDUI-shift transcripts between aged and young mice. These candidates should therefore be interpreted as transcripts showing substantial changes in distal polyadenylation-site usage, rather than transcript-level statistically significant APA events.

At the global level, the distribution of PDUI changes was significantly shifted toward reduced distal polyadenylation-site usage in aged samples in both granulosa cells and oocytes, indicating a broad tendency toward proximal polyadenylation-site usage and shorter 3′UTRs during ovarian aging (granulosa cells: Wilcoxon rank-sum test, p < 2.2 × 10⁻1⁶; oocytes: Wilcoxon rank-sum test, p < 2.2 × 10⁻1⁶). Thus, while individual candidate transcripts were defined by an effect-size cutoff, the Wilcoxon tests support a cell-type-level distributional shift toward 3′UTR shortening. This pattern is consistent with transcriptome remodeling under DNA damage, oxidative stress, and other age-related cellular stimuli.

Notably, oocytes overall possessed shorter 3′UTRs compared with granulosa cells (paired t-test, p < 2.2 × 10⁻1⁶; Supplementary Fig. 3A). This difference may reflect more complete mRNA-processing activity in granulosa cells, favoring distal polyadenylation-site usage, versus the storage of incompletely processed maternal mRNAs in oocytes. A subset of candidate PDUI-shift transcripts was shared between the two cell types (Supplementary Fig. 3B), suggesting partially common 3′UTR usage changes associated with ovarian aging.

Isoform switching is associated with predicted ORF and domain changes despite stable gene expression

Previous transcriptomic studies were largely limited to gene-level analyses due to technical constraints. Leveraging long-read sequencing, we examined differential expression and isoform switching at the transcript level. Using DESeq2, we identified 338 differentially expressed transcripts (DETs) in granulosa cells and 382 in oocytes during aging (Fig. 4A; |log₂FC|> 1, FDR < 0.05). In granulosa cells, DETs were enriched in functions related to mRNA binding, protein folding, and IL-7 signaling, whereas in oocytes, DETs were associated with cytoplasmic translation, protein targeting, chromosome segregation, and IL-7 signaling (Fig. 4B). To clarify the relationship between transcript-level and gene-level functional signals, we compared GO enrichment across DETs, DTU-associated genes, and DEGs, and further evaluated known-versus-novel transcript contributions using semantic similarity and GO-term overlap. These comparisons showed that transcript-level analyses captured both shared and distinct biological themes relative to gene-level DEGs, supporting the added value of isoform-resolved analysis (Supplementary Fig. 5). Oocyte-derived IL-7 can act on neighboring granulosa cells to promote oocyte nuclear maturation via the PI3K/AKT pathway [46], and its age-associated increase has been proposed as a biomarker of follicular-fluid aging [47]. Furthermore, age-related increases in chromosome segregation errors in oocytes are well documented, leading to reduced oocyte quality and higher infertility risk [48].

Fig. 4.

Aging is associated with isoform switching and predicted functional consequences in genes without gene-level expression changes. A Differentially expressed transcripts. B GO enrichment analysis of differentially expressed transcripts. C Differential transcript usage between Aged and Young granulosa cells and oocytes. A significant isoform switch was defined based on a dual-criteria approach: (1) a gene-level statistical significance (FDR < 0.05) as determined by the DEXSeq method, and (2) a substantial effect size, defined as an absolute change in isoform fraction (|dIF|> 0.1) for at least one isoform. D Intersection of differentially used transcripts and differentially expressed transcripts between granulosa cells and oocytes. E Predicted consequence analysis of differentially used transcripts. F Intersection of differentially used transcripts. G GO enrichment analysis of differentially used transcripts

Several disease-associated risk genes linked to premature ovarian failure, ovarian cancer, and ovarian endometriosis exhibited pronounced transcript-level changes (Fig. 4A) that were either absent or only partially captured at the gene level (Supplementary Fig. 4A, B), underscoring the value of full-length transcriptomics for detecting subtle expression changes.

IsoformSwitchAnalyzeR identified 467 and 328 isoform-usage switching events (Differential Transcript Usage, DTU) in granulosa cells and oocytes, respectively, during aging (Fig. 4C; FDR < 0.05 and |dIF|> 0.1). These isoform switches were frequently accompanied by corresponding changes in transcript expression, with a subset shared between the two cell types (Fig. 4D, F). Functional predictions indicated that many switching events were associated with loss of open reading frames (ORF), protein domains, or a shift to noncoding isoforms (Fig. 4B, E, G). Notably, such isoform-level remodeling was frequently observed even in genes showing no significant differential expression at the gene level. For instance, in granulosa cells, Nasp (nuclear autoantigenic sperm protein) and Ganab (glucosidase II alpha subunit) maintained comparable total expression levels between young and aged groups (Supplementary Fig. 6E, F). However, both underwent marked isoform switching, shifting from dominant protein-coding isoforms in young mice to noncoding isoforms in aged mice. These findings highlight how traditional gene-level quantification can mask age-dependent transcript structural changes and suggest that isoform switching may contribute to reduced predicted protein-coding capacity in the aging ovary.

Co-expression networks identify aging-associated modules enriched for translational processes

To systematically investigate the relationships among ovarian transcripts, we constructed an unsupervised transcript-level co-expression network using weighted gene co-expression network analysis (WGCNA). WGCNA groups transcripts according to the similarity of their expression profiles across samples, rather than according to genomic location or gene annotation. In Fig. 5A, each terminal branch in the clustering dendrogram represents a transcript, and transcripts with similar expression patterns were grouped into the same color-coded module. The module labels therefore represent co-expression clusters that capture shared variation across age groups and cell types. For each module, a module eigengene was calculated as the first principal component of transcript expression within that module, providing a summary expression profile that was then used to compare module activity among young and aged oocytes and granulosa cells. Intramodular adjacency was further used to identify highly connected hub transcripts within each module (Fig. 5A, D).

Fig. 5.

Co-expression network analysis uncovers aging-specific modules associated with translational decline. A WGCNA clustering dendrogram showing transcript co-expression modules. Each terminal branch represents a transcript, and color blocks indicate modules defined by similar expression profiles across samples. Module eigengenes were used to summarize module-level expression patterns. B Expression patterns of phenotype-associated modules across experimental groups. C GO enrichment analysis of transcripts within selected modules. D Network plots showing the top 20 hub transcripts within each module based on intramodular connectivity

Clustering revealed five phenotype-related modules. Modules 1 and 2 captured cell-type specificity between oocytes and granulosa cells (Supplementary Fig. 7A, B), with oocytes showing high expression in Module 1 but low expression in Module 2, and granulosa cells exhibiting the inverse pattern (Fig. 5B). Modules 3, 5, and 10 were associated with ovarian aging, characterized by higher expression in young granulosa cells and oocytes compared with aged counterparts. Thus, Fig. 5A provides the network-level basis for defining transcript modules, while Fig. 5B shows how the resulting module expression patterns relate to age and cell type.

GO enrichment analysis supported these functional assignments (Fig. 5C). Module 1 was enriched for chromosome segregation and mitotic cell-cycle transitions, suggesting shared regulatory mechanisms between mitosis and meiosis. Module 2 was enriched for organic-compound oxidation and protein translation, consistent with the biosynthetic and secretory roles of granulosa cells, whereas oocytes rely more on stored maternal mRNAs and proteins. Modules 3, 5, and 10 were predominantly enriched for mRNA biogenesis and protein translation, paralleling the consequence analysis of differential transcript usage (DTU) events (Figs. 4E; 5C) and suggesting age-associated changes in translational programs at the isoform level in both cell types. These module-level associations should be interpreted as exploratory because of the limited number of biological samples. We therefore examined the sensitivity of WGCNA results to alternative parameter settings and quantified the known/novel isoform composition of key modules. The major phenotype-associated modules were broadly recoverable across reasonable WGCNA settings, and both known and novel isoforms contributed to the relevant module structures, although smaller modules showed greater parameter sensitivity (Supplementary Fig. 8).

RNA-splicing-related terms were broadly enriched across modules, supporting a role for alternative splicing in age-associated and cell-type-specific transcriptomic changes. Intramodular adjacency analysis identified candidate hub transcripts for each module (Fig. 5D). Notably, Esr1-derived TALONT000180938 was a central hub in Module 10; it exhibited both differential transcript expression and isoform switching during aging in oocytes and granulosa cells, and Esr1 itself is a risk gene linked to ovarian function and reproductive disorders—premature ovarian failure, ovarian cancer, and ovarian endometriosis (Fig. 6A), suggesting that this transcript may be relevant to aging-associated expression modules and warrants further functional investigation.

Fig. 6.

Age-dependent isoform switching nominates candidate transcript-level alterations in ovarian disease-risk genes. A Heatmap showing expression patterns of transcripts from disease-associated genes across samples, with blue labels indicating newly identified transcripts. POF: premature ovarian failure; OC: ovarian cancer; CC: chocolate cyst (endometriotic cyst). B, C Examples showing differential transcript usage of Esr1 in aging granulosa cells and oocytes, respectively. The differential transcript usage (DTU) events shown were identified using the IsoformSwitchAnalyzeR package. A significant isoform switch was defined based on a dual-criteria approach: (1) a gene-level statistical significance (FDR < 0.05) as determined by the DEXSeq method, and (2) a substantial effect size, defined as an absolute change in isoform fraction (|dIF|> 0.1) for at least one isoform. D Visualization and support of the key gene Esr1

Isoform-level changes nominate candidate links to age-related ovarian pathologies

To gain transcript-level insight into disease-gene associations, we examined three ovarian aging-related disorders, premature ovarian failure (POF), ovarian cancer (OC), and ovarian endometriosis (commonly referred to as chocolate cysts, CC), focusing on transcripts derived from associated risk genes curated from MalaCards and OMIM. Based on transcript-level expression profiles, disease risk genes were distinguishable (Fig. 6A). Many newly identified transcripts (prefixed "TALONT") were differentially expressed, and isoform-level patterns were broadly consistent across biological replicates. Several risk genes displayed pronounced cell-type specificity, such as Aff2 in POF, Mirlet7g in OC, and Hsd17b2 in ovarian endometriosis. Others exhibited more complex isoform usage; for instance, the POF risk gene Nr5a2 showed reciprocal isoform expression, with one isoform enriched in aged granulosa cells and another in young granulosa cells (Supplementary Fig. 11), highlighting transcript-level heterogeneity among disease-associated genes.

Esr1, encoding estrogen receptor-α, regulates downstream transcription in response to estrogen and plays an essential role in ovarian function; Esr1-deficient mice exhibit ovarian aging and infertility [49], and polymorphisms or methylation of Esr1 have been implicated in POF, OC, and CC [50, 51]. We found that the Esr1-derived transcript TALONT000180938 was expressed at high levels in young cells and low levels in aged cells in both granulosa cells and oocytes (Figs. 4A; 6A–C). Notably, this isoform was also supported by RACE evidence (Supplementary Fig. 9).

Isoform usage analysis revealed age-dependent switching in Esr1 expression. In granulosa cells, aged samples predominantly expressed a protein-coding isoform (TALONT000174509), whereas young samples expressed a noncoding isoform (TALONT000180938). Similarly, in oocytes, aged samples mainly expressed a protein-coding isoform (TALONT000177402), while young samples expressed a noncoding isoform (TALONT000170643) (Fig. 6B–D). This age-related shift toward protein-coding isoforms contrasts with the broader functional enrichment trends (Fig. 4E), suggesting a complex, cell-type- and age-specific regulatory landscape for Esr1 in the context of ovarian aging.

Discussion

Technical limitations have long hindered a comprehensive understanding of transcriptomic changes in ovarian aging. Although short-read sequencing provides high depth, its restricted read length limits the reconstruction of complex full-length transcripts and constrains analyses of alternative splicing, alternative polyadenylation (APA), and repetitive elements, keeping most studies at the gene-expression level. Advances in long-read technologies, such as Oxford Nanopore, now enable isoform-resolution profiling of full-length transcriptomes, revealing previously hidden transcript diversity. In this study, we used ONT to profile granulosa cells and oocytes from aged (10 months) and young (6–8 weeks) mice, annotating 130,730 high-confidence isoforms, including 21,558 known and 109,172 putative novel isoforms, thereby enriching current reference annotations. We identified age-associated transcriptome remodeling during ovarian aging, including alternative splicing, altered predicted coding potential, and APA shifts.

Age-related isoform changes may have functional relevance, although experimental validation is required. With advancing age, both granulosa cells and oocytes tended to express isoforms with lower predicted coding capacity, often due to open reading frame (ORF) disruption or premature termination codons. Igf1, a critical regulator of granulosa-cell function that acts synergistically with Fshr to promote proliferation and follicular development [52], displayed reduced coding isoform expression and a shift toward noncoding forms in aged granulosa cells, potentially weakening support for follicular growth. Oocyte aging is marked by frequent meiotic errors, while in granulosa cells, DNA damage accumulation is linked to Rad51 and Mcm9 dysfunction [53, 54]. Our transcript-level findings were consistent with these known aging-related themes (Fig. 4A; Supplementary Fig. 6C, D, Supplementary Fig. 10), with oocyte aging-related isoform changes enriched for chromosome-segregation pathways and Rad51/Mcm9 showing transcript-level differences undetectable at the gene level. In granulosa cells, Gas6 transcripts were differentially expressed with age. Prior work suggests that Gas6 ameliorates meiotic defects in aged oocytes, possibly through improved mitochondrial function [55], yet it has also been implicated in vascular and systemic aging [56], suggesting possible associations between granulosa-cell transcript changes and ovarian or systemic aging-related pathways.

Ovarian aging is not homogeneous but reflects cell-type-specific responses to intrinsic and extrinsic cues [57]. Although oocytes and granulosa cells reside within the same follicle, their metabolic activities, functional states, and stress responses differ markedly, resulting in limited overlap in transcript usage and expression. Our co-expression network placed the two cell types in distinct modules, reinforcing the need to resolve their aging trajectories separately. By dissecting isoform-level differences across ages in both cell types, this study provides a transcript-resolved resource for ovarian aging research.

Transcriptomic patterns observed during physiological ovarian aging may also inform candidate links to ovarian pathologies such as premature ovarian failure (POF), ovarian cancer (OC), and ovarian endometriosis (CC). Genetic studies link Esr1 polymorphisms to POF [58], while Esr1-knockout mice exhibit granulosa-cell dysfunction and infertility [59]. Aberrant Esr1 expression is associated with both OC and CC [51, 60]. Consistent with these observations, we identified age-dependent Esr1 isoform switching and differential expression in both cell types, changes detectable only at the isoform level, without significant differences in overall Esr1 gene expression. These findings suggest that transcript-level features of naturally aging wild-type mice may nominate candidate links to ovarian pathophysiology, rather than directly establishing disease mechanisms. Our co-expression analysis further highlighted a noncoding Esr1 transcript as a candidate hub in aging-associated modules, while other disease-risk genes exhibited minimal gene-level changes, emphasizing the utility of isoform-resolved analyses for generating testable hypotheses.

Several limitations should be noted. First, although we implemented stringent filtering and experimental validation for key transcripts (Fig. 3A, B; Supplementary Fig. 9), the large number of putative novel isoforms, particularly low-abundance transcripts, may include artifacts arising from incomplete reverse transcription, RNA degradation, or long-read annotation uncertainty [25]. Therefore, the biological authenticity and function of individual novel transcripts require further validation. Second, the number of biological replicates was limited (n = 2–3 per group), particularly for aged oocytes, which restricts statistical power and may affect the detection of subtle isoform-level changes. This small sample size, combined with the genetic heterogeneity of the outbred ICR strain, limits the generalizability of some low-abundance transcript findings. To mitigate this, we implemented a highly stringent filtering pipeline and performed experimental validation via RACE and RT-PCR to confirm the existence of key novel isoforms. Nevertheless, future studies with larger animal cohorts, single-cell or spatial long-read sequencing, proteomic validation, and isoform-specific gain- or loss-of-function experiments will be needed to distinguish correlative transcript changes from drivers of aging or disease.

In summary, this study provides an isoform-level view of molecular changes in granulosa cells and oocytes during ovarian aging, including reduced predicted coding capacity, isoform switching, and increased noncoding RNA expression. These findings advance transcript-level annotation of the aging ovary and nominate candidate isoforms and pathways for future functional studies of reproductive decline and gynecologic disease susceptibility.

Conclusions

In summary, this study provides a comprehensive, isoform-resolved landscape of the mouse ovarian transcriptome during aging. By employing long-read sequencing, we identified a large set of previously unannotated isoforms and found that ovarian aging is associated with isoform-level alterations, including reduced predicted coding potential, 3'UTR shortening, and isoform switching. Our findings suggest that age-related transcriptomic changes in oocytes and granulosa cells involve not only gene-level expression differences but also isoform-level structural remodeling of transcripts, particularly in key regulatory genes such as Esr1 and Igf1. These insights refine molecular annotation of the aging ovary and identify candidate isoform-level perturbations that may be relevant to reproductive disorders. Overall, this research provides a resource for future studies seeking biomarkers or therapeutic targets for monitoring and preserving female reproductive health.

Abbreviations

APA

Alternative polyadenylation

CAGE

Cap analysis gene expression

CC

Chocolate cyst (ovarian endometriotic cyst)

COCs

Cumulus-oocyte complexes

CPAT

Coding-Potential Assessment Tool

DETs

Differentially expressed transcripts

dIF

Difference in isoform fraction

DTU

Differential transcript usage

FDR

False discovery rate

GCs

Granulosa cells

GO

Gene Ontology

GVs

Germinal vesicle oocytes

ICR

Mice of Institute of Cancer Research

ISM

Incomplete splice match

LTR

Long terminal repeat

NIC

Novel in catalog

NNC

Novel not in catalog

OC

Ovarian cancer

ONT

Oxford Nanopore Technologies

ORF

Open reading frame

PCA

Principal Component Analysis

PDUI

Percentage of Distal Polyadenylation site Usage Index

POF

Premature ovarian failure

RACE

Rapid amplification of cDNA ends

RT-PCR

Reverse transcription polymerase chain reaction

TEs

Transposable elements

TES

Transcription end site

TPM

Transcripts per million

TSS

Transcription start site

UTR

Untranslated region

WGCNA

Weighted gene co-expression network analysis

Authors’ contributions

HW and XY designed the study, performed the experiments, analyzed the data, and drafted the original manuscript. HW and XY contributed equally to this work and should be considered co-first authors. MZ, KZ, YL, and YW contributed to the methodology, software implementation, and data curation. YY, XD, and YX performed the experimental validation and assisted in the investigation. HY contributed to the investigation and methodology. MZ, KW, SY, and CZ were responsible for the conceptualization, funding acquisition, project administration, and overall supervision of the study. All authors read and approved the final manuscript.

Funding

This work was supported by the National Key Research and Development Program of China (grant number 2024YFC2706904); the National Natural Science Foundation of China (grant numbers 32500714 and 323B2027); the Key R&D Program of Shandong Province, China (grant number 2024CXPT081); the Taishan Scholars Program of Shandong Province (grant number tsqn201909194); the Shandong Provincial Natural Science Foundation (grant number ZR2020QH046); the horizontal project (grant number 2220026002) from Suzhou Yuanyi Stem Cell Technology Co., Ltd. the Nanshan Scholar Start-up Fund of Guangzhou Medical University (06-445-1231); and the National Undergraduate Training Programs for Innovation and Entrepreneurship (grant number 202510570009).

Data availability

The datasets generated during and/or analysed during the current study are available in the Genome Sequence Archive (GSA) repository under project accession PRJCA044948. Processed datasets—including transcript annotations, expression matrices, and analysis scripts—are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The current study was carried out with the approval of the Ethics Committee of Reproductive Medicine of Reproductive Hospital of Shandong University (2022–138), and all experiments were performed according to relevant regulations and committee guidelines.

Competing interests

The authors declare no competing interests.