Long-read RNA-seq maps the isoform landscape of the trigeminal ganglion in an allergic rhinitis mouse model Authors: Zhiyao Han, Beirui Yu, Maoqia Shen, Ziyuan Tang, Xinran Li, Mingyang Wang, Fengxian Li, Xiaolong Cao Published: 2026 ABSTRACT BackgroundThe trigeminal ganglion (TG) plays a key role in allergic rhinitis (AR), mediating symptoms like nasal itching and sneezing via sensory neuron activation. Yet, the underlying transcriptomic mechanisms remain unclear. While short-read RNA-seq struggles to accurately capture complex transcript isoforms, long-read sequencing offers full-length coverage but limited quantitation. Here, we combine both technologies to comprehensively profile the TG transcriptome in an AR mouse model.ResultsWe identify 25,988 high-confidence full-length isoforms from 13,011 known and 237 novel genes, including 7,543 (29.02%) novel isoforms absent from reference databases. These are enriched in RNA splicing and processing genes. Alternative first exons are the most common splicing event and are especially frequent among novel isoforms. AR induces shifts in gene expression, splicing dynamics, and promoter activity, suggesting neuro-immune crosstalk. Single-cell RNA-seq integration further reveals that while some genes lack cell type-specific expression, their alternative splicing patterns are highly cell type-specific, suggesting that AS may play a crucial role in diversifying gene function across different cell types.ConclusionsThe findings provide a comprehensive full-length transcriptome dataset, facilitating in-depth exploration of the interplay between sensory alterations and AR in trigeminal ganglion.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12864-026-13024-y. Background Allergic rhinitis (AR), a chronic IgE-mediated nasal inflammation affecting over 400 million people globally, manifests as sneezing, congestion, rhinorrhea, and pruritus [1, 2]. Recent studies highlight the critical role of the trigeminal nerve in both sensing allergic stimuli and triggering allergen-induced sneezing [3]. The trigeminal nerve, the largest and most complex cranial nerve, mediates sensory input from the face, oral cavity, and nasal cavity [4]. Its ophthalmic and maxillary branches densely innervate the nasal mucosa, making it highly sensitive to mechanical, thermal, and chemical stimuli. Sensory signals converge at the trigeminal ganglion (TG), which houses the cell bodies of afferent neurons. In AR, TG nociceptive neurons transduce allergen-induced stimulation into enhanced action potentials, driving symptoms such as pruritus, and protective reflex like sneezing [3] . During the early phase of AR, allergens and inflammatory mediators (e.g., histamine, leukotrienes, prostaglandins, cytokines and chemokines) directly activate trigeminal nerve endings. This triggers the release of neuropeptides - including Substance P (SP), Calcitonin Gene-Related Peptide (CGRP), and Vasoactive Intestinal Peptide (VIP), Nerve Growth Factor (NGF), and Neuromedin U (NMU) which modulate immune responses [4]. For example, SP promotes vasodilation, glandular secretion, and vascular permeability [5, 6]. Conversely, CGRP suppresses inflammation by downregulating group 2 innate lymphoid cells [7]. These findings underscore the critical role of neuroimmune crosstalk in AR pathophysiology, highlighting TG as a key investigative target. While conventional short-read RNA sequencing approaches excel at gene-level quantification, they struggle to resolve transcript isoforms due to inherent read length limitations that cannot span full-length transcripts [8]. In contrast, full-length RNA-seq provides complete transcript sequences, enabling direct annotation of novel genes and splice variants without reliance on reference-based assembly [9]. Such annotations are vital for interpreting rare genetic variants [10] and elucidating mechanisms in complex neurological disorders (e.g., Alzheimer’s disease) [11]. However, the relatively lower throughput limits quantitative accuracy, necessitating integration with high-depth short-read data for robust transcriptome analysis [12]. Given the nervous system’s extensive alternative splicing [13, 14], TG likely exhibits significant isoform diversity to other neural tissues. Recent studies demonstrate that 36%-50% of isoforms in the human and mouse cerebral cortex are novel isoforms, with conserved patterns of isoform diversity [15]. Therefore, isoform-level characterization of TG is essential to understand its molecular complexity and role in AR pathogenesis. To elucidate these mechanisms, we performed multi-platform transcriptomic analysis (combining long-read and short-read sequencing) of TG tissues from naïve and AR model mice. Our data identified previously uncharacterized alternative splicing programs in TG that orchestrate neuroimmune communication during AR, identifying promising targets for precision therapeutic intervention. Methods Experimental animals All experimental procedures were approved by the Institutional Animal Care Committee of Zhujiang Hospital, Southern Medical University (Protocol LAEC-2023-076). C57BL/6J mice were obtained from GemPharmatech Co., Ltd. (Nanjing, China) and maintained under specific pathogen-free conditions (22–26℃) with a 12:12 h light-dark cycle and ad libitum access to food and water. For full-length transcriptome profiling aimed primarily at novel isoform identification, trigeminal ganglia were collected from around 3-week-old mice of both sexes. For allergic rhinitis modeling, short-read RNA-seq, and experimental validation, male mice aged 6–8 weeks were used. Induction of allergic rhinitis mouse model and tissue collection The ovalbumin (OVA)-induced AR model was established as previously described [3]. Briefly, mice received intraperitoneal injections (200 µL) of 0.02% OVA (Sigma, A5503) emulsified with Imject Alum adjuvant (OZ Biosciences, AH0050) on day 1 and 11. From day 14–18, sensitized mice underwent daily intranasal OVA administration (2 µL of 10% solution) under 2% isoflurane anesthesia. Control animals received PBS. TG tissues were harvested 24 h after the final challenge (day 19). Bulk RNA-seq library preparation and short-read RNA sequencing TG tissues were collected from four control groups and four OVA-treated AR groups, each group with six male mice aged 6–8 weeks. Total RNA was extracted using Trizol reagent (Thermofisher, 15596018) and assessed for quality (RNA Integrity Number > 7.0) using an Agilent 2100 Bioanalyzer. Poly(A) RNA was enriched from 5 µg total RNA using Dynabeads Oligo (dT) beads (ThermoFisher) and libraries were prepared following standard Illumina protocols. Final libraries (300 ± 50 bp insert size) were sequenced (PE150) on an Illumina NovaSeq 6000 platform (LC-Bio Technology CO., Ltd., Hangzhou, China). PacBio Iso-Seq library preparation and sequencing TG tissues were collected from one control group and one OVA-treated AR group, each group with six mice. Full-length cDNA libraries were prepared following the PacBio Iso-Seq protocol. Briefly, total RNA was assessed for concentration, integrity, and purity prior to library construction. First- and second-strand cDNA synthesis was performed, followed by size selection, SMRTbell adapter ligation, and exonuclease cleanup to enrich circular templates. Libraries passing quality control (QC) (Tapestation/Bioanalyzer size distribution and Qubit quantification) were sequenced on the PacBio Sequel II platform to generate HiFi reads according to the manufacturer’s recommendations. Long-read processing, isoform annotation, and quality assessment Circular consensus sequencing (CCS/HiFi) reads were aligned to the mouse reference genome (GRCm38/mm10) using uLTRA assisted by the GENCODE M25 annotation. Transcript reconstruction and quantification were performed with IsoQuant [16]. Isoforms were classified and quality-controlled using SQANTI3 [17], which assigns categories including FSM, ISM, NIC, NNC, antisense, fusion, genic, and intergenic. Only high-confidence isoforms (FSM, NIC, NNC; unless otherwise stated) were retained for downstream analyses. SQANTI3 metrics (e.g., splice-junction motif, CAGE support, PolyA support, RT-switching) were used to assess quality. Short-read processing, transcriptome reconstruction, and differential expression Short-read RNA-seq was performed with n = 4 biological replicates per condition (AR and control). Raw reads from TG were trimmed and filtered using fastp (v0.23.4) [18]. Cleaned reads were aligned to the GRCm38 mouse genome using STAR (v2.7.11b) [19]. Sample-level quality control was assessed by principal component analysis (PCA) on variance-stabilized gene expression values derived from DESeq2. The transcriptome was reconstructed using StringTie, and isoforms were annotated and classified via SQANTI3. Reads from TG and other mouse tissues from the ENCODE project were used as for gene and isoform quantification. Quantification was performed with RSEM (v1.3.3), with the gtf file with known genes from the GENCODE M25 annotation and all newly predicted novel isoforms passing SQANTI3 quality control as input. Differential expression analysis of transcripts and genes in TG samples were performed using DESeq2 with a design formula ~ condition based on RSEM-derived counts. Gene ontology (GO) enrichment analyses and gene set enrichment analyses (GSEA) were performed using the clusterProfiler R package (v4.15) [20] and visualized with aPEAR [21]. Long-read quantification and differential expression IsoQuant were used to quantify novel isoform expression with long-read data from the ENCODE project and TG of the current study. The “--no_model_construction” option was enabled to perform read assignment and quantification directly against a reference transcript model. Gene models curated after SQANTI3 QC were used as reference. For differential expression with TG long-read data, edgeR was applied at the gene level with pairwise comparison using the exact test with a fixed dispersion value (0.04). GSEA was performed using the clusterProfiler. Alternative splicing analysis Alternative splicing (AS) events were analyzed using SUPPA2 [22] to detect seven alternative splicing (AS) events, including A3/A5 (alternative splice sites), AF/AL (alternative first/last exons), SE (skipped exons), RI (retained introns), and MX (mutually exclusive exons). The generateEvent command processed SQANTI3-classified isoforms (FSM, NIC, NNC) and GRCm38 reference annotations to identify both known (reference-matched) and novel (long-read-specific) splicing events. Percent spliced-in (PSI) values were computed using SUPPA2 psiPerEvent with expression matrices derived from RSEM. Junction-level evidence for top AF/promoter switches was visualized using short-read sashimi plots in IGV, and robustness was assessed using an orthogonal isoform-switching framework (IsoformSwitchAnalyzeR) [23] applied to the same RSEM transcript-level quantification. Promoter activity analysis Promoter activity was inferred using proActiv (v1.1.18) [24] integrating STAR-aligned short-read BAMs with the full-length transcript annotation to estimate promoter-specific expression. Differential promoter activity between conditions was tested with DESeq2. Single-cell RNA-seq data analysis Processed single-cell RNA-seq (scRNA-seq) data, with the original study’s quality control (QC) and annotations, from were obtained from the website provided by the study [25]. For single-cell splicing junction (scSJ) analysis, raw data from the Martine Aubert dataset (NCBI GEO: GSE255436) were reprocessed [26]. Reads were aligned and quantified with STARsolo to generate cell-by-splice-junction (SJ) count matrices [27]. The SJ count matrix was imported into Seurat by treating each SJ as a feature (analogous to a gene) and using the SJ counts as the expression matrix. We then matched cells between the STARsolo-derived scSJ matrix and the annotated scRNA-seq object using the 10x cell barcode identifiers. Only cells present in the curated scRNA-seq object were retained for downstream scSJ analyses. Cell-type labels, including TG cell-type annotations, were transferred from the scRNA-seq object to the retained scSJ cells. For scSJ analysis, SJs present in GENCODE gene model were classified as known, whereas SJs derived from the TG transcriptome model were classified as novel. Cell-type–specific usage of SJs was assessed using the FindAllMarkers function in Seurat. Validation of novel isoforms by PCR and sanger sequencing To validate novel isoform structures, quantitative real-time PCR (qPCR) or PCR assays were designed to span novel exon–exon junctions or novel exons. Forward and reversed primer sequences were provided in Table S1. PCR products were gel-purified and subjected to Sanger sequencing. Results Full-length transcriptome profiling identified novel isoforms in TG tissues To comprehensively characterize the full-length transcriptome of TG, we performed PacBio Iso-Seq analysis on TG tissues dissected from both control (PBS-treated) and ovalbumin-induced AR model mice (Fig. 1A). Using Circular Consensus Sequencing (CCS) mode, we obtained high-fidelity (HiFi) reads and with > 99% bases with quality score better than Q20 (Figure S1). HiFi reads were aligned to the mouse reference genome (GRCm38/mm10) using uLTRA assisted by the GENCODE M25 mouse gene model [28], followed by transcriptome assembly and quantification using IsoQuant [16, 28]. Rigorous quality control and isoform classification were performed using SQANTI3 [17], which applies multiple orthogonal validation metrics and group isoforms to different categories by comparison with the annotated gene models: Full-Splice Matches (FSM), Incomplete-Splice Matches (ISM), Novel In Catalog (NIC), Novel Not in Catalog (NNC), fusion, genic, antisense and intergenic (Fig. 1B). Genes and isoforms matching the reference annotation were designated as “known” while those absent from the annotation were classified as “novel”. This analysis identified 25,988 high-confidence isoforms corresponding to 13,011 known and 237 novel genes (Fig. 1C). Among these isoforms, 7,543 were novel, comprising 7,278 isoforms from 4,589 known genes and 265 isoforms from the newly 237 identified genes (Fig. 1D). The assemble isoforms were categorized as: FSM (64.9%, n = 16,869), NIC (16.3%, n = 4,243), NNC (10.6%, n = 2,767), and ISM (6.1%, n = 1,576). Minor categories (antisense, genic, fusion, intergenic; < 3% combined) were excluded from most subsequent analyses. We predicted the coding functions of novel isoforms and found that most NICs and NNCs (> 90%) possessed coding functions, higher than isoforms from novel genes (approximately 50%) (Fig. 1E). Most of the coding novel isoforms encode protein sequences not found in the reference annotation, suggesting they contribute to the proteome diversity of TG (Fig. 1F). We quantified the number of isoforms expressed by genes in TG and found that 47.4% of the known genes have more than one isoform, while 36.0% express at least one novel isoform (Fig. 1G), which means three-quarters of genes with multiple detected isoforms produce novel variants beyond the reference annotation, highlighting the importance of full-length RNA sequencing for capturing previously uncharacterized transcript diversity. Analysis of isoform detection patterns across experimental conditions showed that the majority of known isoforms were present in both PBS control and OVA-treated AR groups. Novel isoforms from known genes exhibited similar distribution patterns to known isoforms, whereas novel isoforms from novel genes showed greater condition specificity, with more isoforms being uniquely detected in either the control or AR group (Fig. 1H). Fig. 1Full-length isoform landscape of the TG identified by long-read sequencing. A Overview of experimental design and analysis pipeline. B SQANTI3 isoform classes and structural categories. C–D Composition of all high-confidence isoforms passing SQANTI3 QC (C) and of the novel subset (D). E Predicted coding potential across novel isoform classes. F Fraction of novel isoforms predicted to encode protein sequences absent from the reference annotation. G Isoform complexity and novel isoform detection across known and novel genes. Left: Distribution of annotated (known) genes by the number of detected isoforms. Middle: Proportion of annotated genes expressing 0, 1, 2, or ≥ 3 novel isoforms, indicating that ~ 36% of known genes harbor at least one novel isoform. Right: Distribution of novel genes by the number of detected novel isoforms (1, 2, or ≥ 3). H Overlap of known isoforms and novel isoforms between control and AR groups. I Orthogonal QC metrics for predicted isoforms (splice-junction support, reverse transcriptase switching, 5’ CAGE support, 3’ PolyA evidence). NIC/NNC are comparable to FSM. J Sequence logos of splice donor and acceptor sites for known and novel isoforms. K–L Structural and regulatory features of novel (NIC/NNC) versus known (FSM) isoforms. K Novel isoforms tend to be longer, contain more exons, extend the coding sequences and have lower expression. L Novel isoforms more frequently use alternative transcription start (TSS) and termination (TTS) sites and are more often predicted targets of nonsense-mediated decay (NMD). All comparison between known and novel isoforms in (K–L) were significant (p-value < 2.2 × 10⁻¹⁶) Full-length isoform landscape of the TG identified by long-read sequencing. A Overview of experimental design and analysis pipeline. B SQANTI3 isoform classes and structural categories. C–D Composition of all high-confidence isoforms passing SQANTI3 QC (C) and of the novel subset (D). E Predicted coding potential across novel isoform classes. F Fraction of novel isoforms predicted to encode protein sequences absent from the reference annotation. G Isoform complexity and novel isoform detection across known and novel genes. Left: Distribution of annotated (known) genes by the number of detected isoforms. Middle: Proportion of annotated genes expressing 0, 1, 2, or ≥ 3 novel isoforms, indicating that ~ 36% of known genes harbor at least one novel isoform. Right: Distribution of novel genes by the number of detected novel isoforms (1, 2, or ≥ 3). H Overlap of known isoforms and novel isoforms between control and AR groups. I Orthogonal QC metrics for predicted isoforms (splice-junction support, reverse transcriptase switching, 5’ CAGE support, 3’ PolyA evidence). NIC/NNC are comparable to FSM. J Sequence logos of splice donor and acceptor sites for known and novel isoforms. K–L Structural and regulatory features of novel (NIC/NNC) versus known (FSM) isoforms. K Novel isoforms tend to be longer, contain more exons, extend the coding sequences and have lower expression. L Novel isoforms more frequently use alternative transcription start (TSS) and termination (TTS) sites and are more often predicted targets of nonsense-mediated decay (NMD). All comparison between known and novel isoforms in (K–L) were significant (p-value < 2.2 × 10⁻¹⁶) Quality assessment using multiple orthogonal metrics by SQANTI3 revealed novel isoforms (NIC/NNC) demonstrated comparable quality to FSMs in splice junction characteristics, reverse transcriptase switching (RTS) artifact frequency, 5’ transcript ends with CAGE (Cap Analysis of Gene Expression) support, and 3’ ends with PolyA evidence (Fig. 1I). ISM isoforms are typically considered incomplete splicing products, which is consistent with our findings that they showed significantly reduced CAGE support (41% vs 60% in FSMs) and reduced PolyA tail support, suggesting potential technical artifacts or degradation products [29], prompting their exclusion from subsequent analyses. Analysis of exon-intron boundaries showed that 98.20% of 5’-splice sites contained GT and 99.93% of 3’-splice sites contained AG among novel isoforms, with all sites conforming to canonical motifs (Fig. 1J). Comparative analysis revealed significant structural and functional differences between novel (NNC and NIC) and known (FSM) isoforms. Despite lower expression levels (median TPM (Transcript Per Million) 1.68 vs. 5.80, p = 2.0 × 10–26), novel isoforms exhibited increased structural complexity, including more exons (median 10 vs. 8, p = 9.6 × 10–77), longer transcripts (median 2,766 vs. 2,589 bp, p = 2.0 × 10–16), and extended coding sequences (median 1,338 vs. 1,251 bp, p = 2.7 × 10− 8; Fig. 1K). Novel isoforms exhibited distinct regulatory patterns compared to known isoforms, with 7.5% utilizing alternative transcription start sites (TSSs>100 bp from known sites vs. 0.2% in FSMs; p = 1.7 × 10–243) and 10.4% using alternative transcription termination sites (TTSs>100 bp from known sites vs. 0.7%, p < 4.7 × 10–280, Fig. 1L). This expanded use of alternative initiation and termination sites was associated with increased susceptibility to quality control mechanisms, as novel termination sites correlated with higher rates of premature stop codons targeted by nonsense-mediated decay (NMD) (9.9% vs. 2.5%; p = 7.5 × 10–125). These findings indicate that novel isoforms may represent transcriptional variants with altered regulatory properties, though many may be less stable compared to known isoforms, consistent with previous studies [29]. We compared the long-read transcriptome data with results from conventional short-read RNA sequencing performed on the same TG tissues. Short-read sequencing reads were assembled using StringTie, and the resulting isoforms were classified using SQANTI3 [17, 30]. Comparing the long-read and short-read isoforms with GffCompare [31], we found that 71.4% of novel high-confidence long-read isoforms (NIC, NNC) were missed by the short-read approach (Figure S2A). Although short-read sequencing identified a larger number of isoforms, these were predominantly fragments, exhibiting significantly shorter lengths (1,322 bp vs. 2,651 bp, p < 2.2 × 10–308), fewer exons (4 vs. 8, p < 2.2 × 10–308), and weaker support from CAGE (27.4% vs. 53.7%, p < 2.2 × 10–308) and PolyA evidence (53.9% vs. 92.2%, p < 2.2 × 10–308) (Figure S2B–E). At the expression level, gene-level quantification showed a moderate correlation between the two platforms (Pearson r ≈ 0.7, Figure S3A, B). However, the correlation was notably weaker for novel isoforms (r ≈ 0.5, Figure S3C–F), reflecting challenges in quantifying low-abundance complex isoforms with short-read. Together, these results demonstrate the superior accuracy and comprehensiveness of long-read transcriptome profiling. Characterization of the alternative splicing landscape and promoter activity in TG tissues To investigate the mechanisms underlying isoform diversity, we performed a comprehensive alternative splicing (AS) analysis on the full-length gene model, including FSM, NIC and NNC isoforms, using SUPPA2. We quantified seven major AS event types (Fig. 2A) and identified a total of 10,558 total AS events in total. Known AS events were defined as those that could be explained by the known gene annotations (i.e., splicing events predicted by the GENCODE gene model), while novel events were defined as other AS events that were introduced by the identified novel isoforms. Among the seven AS event types, alternative first exon (AF) events showed the greatest difference in frequency between known and novel AS events, with AF events comprising 12.89% of known and 32.46% of novel AS events (Fig. 2B). This significant enrichment of AF events in novel AS suggests preferential utilization of alternative promoter regions during the generation of novel isoforms. To quantify splicing variation, we calculated percent spliced-in (PSI) values using SUPPA2 [22]. While PSI distributions were broadly comparable between known and novel AS events across event types (Fig. 2C), we focused on alternative first exon (AF) events due to their notable differences in frequency (Fig. 2B) and because previous research has demonstrated that promoter selection is crucial for transcriptome diversity and functional regulation [24]. Fig. 2Alternative splicing landscape and promoter usage in TG. A Classification of alternative splicing (AS) events by SUPPA2. B Composition of known versus novel AS events across types. AF is the most common type of novel AS type detected. C PSI distributions for known versus novel events across AS types. Differences were tested within each AS type (Wilcoxon rank-sum) with BH-FDR correction across types. Significant shifts were observed for A5 and RI (adjusted p < 0.05). D Distribution of active promoters per gene inferred by proActiv. ~15% of genes contain multiple promoters. E Example of Arrb1, a novel promoter exhibits higher activity than the annotated promoter. Left, gene models and promoter positions; right, promoter activity estimates. F Major versus minor promoter usage among known and novel promoters. G Alternative promoters alter the structure of gene isoforms. The stacked bar charts show the percentage of shared and unique regions (5’UTR, CDS, 3’UTR) between isoforms driven by minor/major promoters or by novel/known promoters. H Examples where novel promoter usage modifies CDS and 3’ UTR sequences of Trpc6 and Tyro3 genes Alternative splicing landscape and promoter usage in TG. A Classification of alternative splicing (AS) events by SUPPA2. B Composition of known versus novel AS events across types. AF is the most common type of novel AS type detected. C PSI distributions for known versus novel events across AS types. Differences were tested within each AS type (Wilcoxon rank-sum) with BH-FDR correction across types. Significant shifts were observed for A5 and RI (adjusted p < 0.05). D Distribution of active promoters per gene inferred by proActiv. ~15% of genes contain multiple promoters. E Example of Arrb1, a novel promoter exhibits higher activity than the annotated promoter. Left, gene models and promoter positions; right, promoter activity estimates. F Major versus minor promoter usage among known and novel promoters. G Alternative promoters alter the structure of gene isoforms. The stacked bar charts show the percentage of shared and unique regions (5’UTR, CDS, 3’UTR) between isoforms driven by minor/major promoters or by novel/known promoters. H Examples where novel promoter usage modifies CDS and 3’ UTR sequences of Trpc6 and Tyro3 genes We used the R package proActiv to characterize promoter activity in the TG transcriptome based on its superior performance proved in previous study [24]. Promoters were classified as major or minor according to their predicted activity levels. Promoters from FSM isoforms were classified as known, others as novel. In total, we identified 15,014 active promoters from 12,707 genes, with 15% of genes (n = 1,906) possessing more than one active promoters (Fig. 2D). In our data, we observed cases where novel promoters show similar or higher estimated activity than annotated promoters. For example, the Arrb1 (Arrestin Beta 1) gene, which encodes a scaffolding protein that regulates G-protein coupled receptors (GPCR), modulates autophagy, and orchestrates receptor-mediated immune functions [32], showed higher estimated promoter activity in its novel promoter compared to its known counterpart (Fig. 2E). This finding suggests that novel promoter may also be important for its function. Additionally, novel promoters were more likely to be categorized as minor promoters than known promoters (45% vs. 10.5%, Fig. 2F). To evaluate the effects of alternative promoters on gene structure, we calculated the percentage of shared untranslated regions (UTRs) and coding sequences (CDS) between isoforms driven by different promoters. Isoforms from minor and major promoters shared 76% of their CDS, while the overlap in 5’ and 3’ UTR were much lower. Novel promoter modifies 7% of CDS regions and can alter protein sequences (Fig. 2G). For example, the receptor tyrosine kinase Tyro3 limits type 2 immune responses, and genetic variants that reduce its expression are important susceptibility factors for atopy and allergic rhinitis in human populations [33]. The cation channel Trpc6 plays a pro-inflammatory role in allergic rhinitis, and studies in murine models have demonstrated that inhibiting Trpc6 activity or expression alleviates AR symptoms and suppresses associated inflammation [34]. In both Tyro3 and Trpc6, novel promoters altered the CDS and 3’ UTR sequences (Fig. 2H). We also visualized the transcript structures using IsoformSwitchAnalyzeR (Figure S4A, B) and the RNA-seq read alignments in IGV (Figure S4C, D), to further support the presence of novel splice junctions specific to these isoforms. Importantly, PCR amplification and Sanger sequencing verified the expression of these newly identified transcripts at the experimental level (Figure S4E, S4F; S5). Together, these results provide robust computational and experimental evidence supporting the existence of novel transcripts and alternative promoter in Tyro3 and Trpc6. Tissue specificity and functional specialization of novel isoforms in TG To understand the function of novel isoforms in TG, we performed enrichment analysis with 4,589 genes expressing novel isoforms (Fig. 3A). Prominent enrichments were linked to neuronal identity and activity, including synapse organization, vesicle-mediated transport, and small GTPase-mediated signaling, suggesting these isoforms may fine-tune synaptic transmission, neuronal architecture, and signaling specificity. We observed strong enrichment for processes underlying essential cellular machinery, such as cytoskeletal organization, cellular metabolism, mRNA processing, protein synthesis, and post-translational modification and quality control. Together, these findings suggest novel isoforms support both general neuronal maintenance and specialized sensory functions, and contribute substantially to functional diversification in TG neurons. Fig. 3Tissue specificity and functional enrichment of TG genes expressing novel isoforms. A GO enrichment for genes expressing novel isoforms in TG. GO enrichment analysis was performed using clusterProfiler. Only terms with adjusted p-value < 1 × 10− 5 were included. B–C t-SNE of samples from TG and ENCODE tissues using long-read gene expression (B) and novel isoform expression (C). D Heatmap of novel isoform expression quantified from ENCODE long-read data across tissues. Isoforms are grouped into 10 clusters by hierarchical clustering based on tissue distribution. TPM values of samples from the same tissue were averaged. Representative GO enrichments for each cluster are annotated (tested against all genes with novel isoforms). Left sidebars denote average TPM of novel isoforms in the two TG samples Tissue specificity and functional enrichment of TG genes expressing novel isoforms. A GO enrichment for genes expressing novel isoforms in TG. GO enrichment analysis was performed using clusterProfiler. Only terms with adjusted p-value < 1 × 10− 5 were included. B–C t-SNE of samples from TG and ENCODE tissues using long-read gene expression (B) and novel isoform expression (C). D Heatmap of novel isoform expression quantified from ENCODE long-read data across tissues. Isoforms are grouped into 10 clusters by hierarchical clustering based on tissue distribution. TPM values of samples from the same tissue were averaged. Representative GO enrichments for each cluster are annotated (tested against all genes with novel isoforms). Left sidebars denote average TPM of novel isoforms in the two TG samples To characterize the unique transcriptional landscape of TG, we performed a comparative expression analysis between TG samples and 17 mouse organs using publicly available bulk RNA-seq data spanning the Mus musculus lifespan [35]. Dimensionality reduction plot shows that TG samples formed a distinct cluster positioned closest to brain tissues, reflecting shared global transcriptional features (Figure S6A). Hierarchical clustering of genome-wide expression patterns further confirmed that, while TG transcriptional profiles resemble those of the brain, they also exhibit clear tissue-specific signatures, with a substantial set of genes uniquely or highly expressed in TG (Figure S6B). To more accurately characterize the tissue specificity of novel isoforms identified in TG, we quantified the expression of genes and novel isoforms using long-read RNA-seq data from 9 mouse organs and cell lines from the ENCODE project [36]. Consistent with short-read data analysis, dimensionality reduction plots based on both gene expression (Fig. 3B) and novel isoform expression (Fig. 3C) demonstrated that TG samples clustered most closely with brain tissues (cerebral cortex and hippocampus). Hierarchical clustering analyses using genes (Figure S6C) and novel isoforms (Fig. 3D) confirmed that TG transcriptional profiles resemble those of brain tissue while maintaining distinct tissue-specific signatures. To dissect functional specialization, we grouped TG novel isoforms into 10 expression clusters based on tissue distribution (Fig. 3D) and performed GO enrichment analysis using all genes expressing novel isoforms as background. Clusters 1 and 2 exhibited high TG specificity, cluster 1 had high expression levels, while cluster 2 isoforms were largely TG-specific. Both were enriched for neuronal projection terms, but cluster 1 showed additional enrichment for actin binding and cytoskeletal protein binding, and cluster 2 for membrane proteins and ion channel activity, functions directly relevant to TG sensory neuron physiology. Clusters 3 and 4 were preferentially expressed in TG and brain, with cluster 3 more abundant in TG and cluster 4 in brain. Both were enriched for neuron projection terms, but Cluster 3 showed additional enrichment in axon and cell projection organization processes, whereas Cluster 4 was enriched for synapse-related functions. Clusters 6–10 predominantly contained isoforms associated with fundamental cellular processes. For instance, Cluster 6 isoforms were relatively highly expressed in the heart and enriched for metabolic processes. These results indicate that many novel isoforms are specialized for TG neuronal function, potentially reflecting adaptations for its unique sensory processing roles. To further delineate the functional roles of novel isoforms in TG, we performed GO enrichment analyses on the 500 most highly expressed genes containing novel isoforms and the 500 genes with the greatest isoform diversity in TG (Figure S7). The highly expressed set was enriched for core processes integral to TG physiology, including cellular supramolecular fiber organization, cellular metabolism, proteostasis, ion transport, and synaptic function. The most isoform-diverse genes were enriched for neuron-specific terms like postsynaptic specialization, axon development and function, and presynaptic organization. Both sets shared synapse-related pathways, underscoring the role of novel isoforms in synaptic plasticity and neuronal communication. Integrated long- and short-read transcriptomics reveals isoform-level remodeling of the TG in AR For short-read transcriptome data, we first performed basic quality control. An average of approximately 20.5 million reads per sample were successfully aligned to the reference genome. Alignment and quality control metrics demonstrated high data quality, with the unique alignment rate ranging from 93.05% to 94.68%. Reads discarded due to insufficient length represented 1.33–2.86% of total reads and the mismatch rate per base was approximately 0.2%. Sample-level quality control and batch assessment showed no evident batch-driven clustering between AR and control samples (Figure S8). To comprehensively assess transcriptomic changes associated with AR, we quantified isoform and gene expression using RSEM for short-read data and IsoQuant for long-read data, respectively [37]. We identified 63 differentially expressed genes and 537 differentially expressed isoforms in AR versus control (adjusted p-value < 0.05 and |log2FC| > 1) (Fig. 4A, Table S2). Of these, 17 were genes with novel isoforms and 282 were novel isoforms (Table S3). Notably, many neuronally relevant genes exhibited no significant change at the gene level despite significant isoform-level alterations (Fig. 4B, S9), indicating that AR induces regulatory remodeling at the isoform level. At the gene level, we observed increased expression of ion channel genes (Cacna1g, Kcng1, Kcne4) and solute carriers (Slc38a3, Slc22a8), alongside decreased expression of ion channel genes (Kcnq1, Kcng2) and components of monoaminergic neurotransmission (Th, Ddc, Slc6a2, Slc6a4, Slc10a4, Adra2b). To experimentally validate these transcriptomic changes, we performed qPCR analysis of Th and Ddc, which confirmed their significant downregulation in AR samples, consistent with the RNA-seq results (Figure S5; S10A, S10B). These patterns suggest a shift in ionic homeostasis and transporter activity that may enhance neuronal excitability (e.g., via reduced potassium current stability) and alter catecholaminergic/serotonergic modulation in TG. Importantly, some genes were with novel isoforms, further supporting disease-associated isoform remodeling. For example, qPCR validation of Pla2g3 and Mast1—two genes with AR-associated novel isoforms—revealed expression changes concordant with the computational analysis (Figure S5; S10C, S10D). Fig. 4AR model shows different gene and isoform expression and functional enrichment compared to control. A Volcano plots of differential expression between AR and control at the gene and isoform levels. B Example gene (Plxnb3) with no significant gene-level change but significant isoform-level differences, illustrating isoform-specific regulation. C GSEA of AR versus control. Only pathways with adjusted p-value < 1 × 10− 5 were included when using short-read data. For long-read data, only pathways with adjusted p-value < 0.05 were included. Pathways validated across platforms are circled in red. Sectors within nodes indicate the proportion of pathway genes that contain novel isoforms AR model shows different gene and isoform expression and functional enrichment compared to control. A Volcano plots of differential expression between AR and control at the gene and isoform levels. B Example gene (Plxnb3) with no significant gene-level change but significant isoform-level differences, illustrating isoform-specific regulation. C GSEA of AR versus control. Only pathways with adjusted p-value < 1 × 10− 5 were included when using short-read data. For long-read data, only pathways with adjusted p-value < 0.05 were included. Pathways validated across platforms are circled in red. Sectors within nodes indicate the proportion of pathway genes that contain novel isoforms Gene set enrichment analysis (GSEA) of the short-read dataset revealed significant upregulation of pathways related to protein synthesis and ATP biosynthetic processes, accompanied by downregulation of pathways governing cell development and differentiation, neurogenesis, proliferation, and cell–substrate adhesion (Fig. 4C). Approximately 30% of genes in altered pathways contained novel isoforms, indicating that isoform novelty is embedded within dysregulated programs. These findings support metabolic reprogramming in AR, potentially reallocating resources toward biosynthetic and energetic demands for ion pump activity, neurotransmission, and membrane excitability. To examine cross-platform concordance at the pathway level, we performed a rank-based GSEA using long-read–derived gene-level change metrics for gene ranking only, because the Iso-Seq dataset lacks biological replicates and therefore does not support reliable inferential testing of differential expression or splicing. In this exploratory concordance analysis, pathways related to ribosomal components and ATP biosynthetic processes showed concordant directional enrichment (Fig. 4C), consistent with the short-read results. Single-cell splice-junction profiling reveals cell type–restricted usage of novel isoforms in TG To assess the cell-type specificity of novel isoforms, we leveraged splice-junction evidence as a proxy for isoform usage at single-cell resolution. Because standard scRNA-seq protocols do not reliably recover full-length transcript structures, we focused on junctions introduced by NIC and NNC isoforms identified in our long-read data. Using the recently published harmonized TG cell atlas [25], we reanalyzed the Martine Aubert dataset [26] and generated per-cell splice-junction (SJ) count matrices with STARsolo (see Methods for details). We processed the resulting single-cell SJ (scSJ) matrices in Seurat, performed dimensionality reduction, and transferred cell-type labels from the matched scRNA-seq profiles. UMAP embeddings based on scSJ data recapitulated seven major TG cell types (Fig. 5A, B), indicating that SJ profiles alone can distinguish cell populations and reflect cellular identity beyond gene expression. We quantified the fraction of novel SJ reads per cell and observed a low overall proportion (mean ≈ 0.06%), with clear enrichment in neuronal subtypes (Th, Trpm8, Ntrk3, and Calca; Fig. 5C, D). The low global fraction is expected given the end-biased capture in scRNA-seq. Neuronal populations showed higher total reads and numbers of detected features (Figure S11A, B), which likely contributed to their higher detection of novel SJs. Despite these limitations, more than half of cells contained at least one novel SJ. Fig. 5Cell-type specific usage of novel isoforms from single-cell splice-junction data. A–B UMAP of TG single cells. A Embedding based on scRNA-seq gene expression. B Embedding based solely on scSJ profiles. C Per-cell fraction of reads mapping to novel junctions. D Violin plots showing the percentage of novel junction reads across seven major cell types. E Genes exhibiting no differential expression across neuronal subtypes (p-value ≥ 1 × 10− 5 and adjusted p-value ≥ 0.05) but displaying cell type–biased splice-junction usage (adjusted p-value < 0.05). F Experimental validation of Arhgap19 novel isoforms. Left, gel image of qPCR amplicons. Right, schematic of validated isoforms showing use of a previously unannotated first exon and the location of the PCR targets Cell-type specific usage of novel isoforms from single-cell splice-junction data. A–B UMAP of TG single cells. A Embedding based on scRNA-seq gene expression. B Embedding based solely on scSJ profiles. C Per-cell fraction of reads mapping to novel junctions. D Violin plots showing the percentage of novel junction reads across seven major cell types. E Genes exhibiting no differential expression across neuronal subtypes (p-value ≥ 1 × 10− 5 and adjusted p-value ≥ 0.05) but displaying cell type–biased splice-junction usage (adjusted p-value < 0.05). F Experimental validation of Arhgap19 novel isoforms. Left, gel image of qPCR amplicons. Right, schematic of validated isoforms showing use of a previously unannotated first exon and the location of the PCR targets To test whether isoform regulation is independent from gene-level abundance, we identified genes with no significant difference across four neuronal subtypes in the scRNA-seq data but with cell type-biased SJ usage in the scSJ analysis (Fig. 5E). This supports isoform-level regulation. We further examined high-expressed novel isoforms (TPM > 10 in both long- and short-read datasets) and detected cell type-specific novel SJs (Figure S11C). These isoforms arise from genes central to neuronal function, including ion channels (Kcnab2, Kcnmb1, Scn10a), cytoskeletal and extracellular matrix components (Dstn, Eml2, Epb41l3, Map7d2, Mxra7), and regulatory proteins (Arhgap19, Iqsec2, Mapkap1, Nap1l5), indicating that novel isoform usage is both cell type-restricted and functionally relevant. To validate novel transcript structures, we performed PCR and Sanger sequencing for Arhgap19, confirming two isoforms that utilize a previously unannotated first exon (Fig. 5F, Figure S11D). Together, these results integrate long-read transcript structures with single-cell junction evidence to demonstrate cell type–specific usage of novel isoforms in TG, providing a framework to link isoform diversity to TG cellular heterogeneity and AR-associated neuronal phenotypes. Discussion AR, a chronic inflammatory condition often comorbid with asthma, imposes significant health burdens.[2] TG is vital in AR pathophysiology by detecting nasal irritants via nerve endings and relaying signals to the brain [4]. Although direct studies on TG-AR interactions are limited, emerging evidence supports TG involvement in rhinitis. A full molecular characterization of TG requires isoform-level resolution beyond short-read RNA-seq. To address this, we integrated long- and short-read sequencing to profile TG transcriptomes in control and AR conditions, generating an isoform-resolved reference that revealed novel isoforms and splicing patterns (Figure S12). Our long-read analysis identified extensive isoform diversity in TG, including thousands of high-confidence novel isoforms with splice-junction quality comparable to annotated transcripts. TG transcriptomes showed the closest similarity to brain, yet also possessed distinct TG-enriched isoforms, suggesting that isoform diversification fine-tunes neuronal identity toward TG-specific functions. Consistent with its sensory role, TG exhibited strong enrichment for ion channels and GPCR activity, reflecting its specialized neuronal functions. Although AR is an immune-mediated disease, type 2 inflammation occurs primarily in the nasal mucosa rather than in the TG itself. Accordingly, the TG likely functions mainly as a sensory and neuroimmune relay, converting allergen- and inflammatory mediator-derived signals into neuronal and glial responses [4]. Because TG tissue is composed predominantly of neurons and ganglion-associated support cells, bulk transcriptomic changes are more likely to reflect neuronal excitability, glial communication, promoter usage, and isoform remodeling than broad canonical immune-pathway enrichment. Our findings are consistent with this model and support neuroimmune remodeling of the TG in AR. We found that alternative promoters alter the 5’ and 3’ UTR but with less change in the CDS (Fig. 2G), supporting a model that promoter selection typically reshapes posttranscriptional regulation rather than protein sequence. Promoter switching generally alters the 5’ leader sequence (through substitution of the transcription start site/first exon), potentially changing cis-regulatory elements and RNA secondary structure [38], thereby modulating translation initiation and mRNA stability. Recent large-scale analyses further support the use of alternative transcription start sites to fine-tune protein expression by altering the 5’ UTR sequence/structure, providing a possible mechanism for specific regulation in neuronal and neuroimmune settings [39]. Similar observations were reported in previous studies studying promoter activity in gastric cancer and brain isoform diversity in house mouse through long-read transcriptome sequencing [29]. Isoform-level changes in AR were more pronounced than gene-level alterations, with over half of differentially expressed isoforms being novel. We observed regulation of ion channels and monoaminergic pathways, aligning with increased neuronal excitability and altered neurotransmitter modulation. GSEA indicated metabolic reprogramming toward protein synthesis and ATP production, with nearly one-third of genes in dysregulated pathways containing novel isoforms, embedding isoform novelty within AR-related networks. To map isoform usage to cellular identity, we used SJ as a proxy for isoform usage at single-cell resolution with the published scRNA-seq database [25]. Although novel SJ reads were rare overall, they were enriched in neuronal subtypes, a pattern likely influenced by technical limits of scRNA-seq rather than biological absence. Future studies with full-length scRNA-seq will better resolve cell type-specific isoform expression. As an illustrative case, Macrod1 displayed six splice junctions with pronounced subtype-biased usage different from gene-level differences (Fig. 5E). Because most of these junctions are shared by two identified Macrod1 isoforms, this pattern likely reflects shifts in isoform stoichiometry and/or isoform-selective stability. Importantly, we identified genes with stable RNA expression but cell type-biased SJ usage, supporting isoform-specific regulation. Novel isoforms from ion channel, cytoskeletal, extracellular matrix, and regulatory genes exhibited cell type-restricted junction usage. These findings demonstrate neuron-enriched and cell type–specific usage of novel isoforms in TG and provide a framework for linking isoform diversity to neuronal subtype function and AR-associated phenotypes. Conclusion Our findings provide a comprehensive full-length transcriptome dataset, facilitating in-depth exploration of the interplay between sensory alterations and AR in trigeminal ganglion. Supplementary Information Supplementary Material 1. Supplementary Material 1. Supplementary Material 2. Supplementary Material 2. Supplementary Material 3. Supplementary Material 3.