Introduction

Menière’s disease (MD) is a chronic inflammatory disease of the inner ear characterised by episodes of vertigo usually associated with tinnitus and sensorineural hearing loss (SNHL)1. The disorder is prevalent across Eurasia, with higher incidence rates in the European rather than East Asian population2. Studies have found several endophenotypes, based on rare genetic variation (familial MD), or association to co-occurring conditions, such as migraine, autoimmune/autoinflammatory diseases and cytokine signature3.

Immune cells play an important role in the development of the disease. Three immune MD phenotypes have been described linked to a persistent inflammation and based on their cytokine profiles4,5,6,7. Previous research measured cytokines in MD patients after stimulation with a mix of Penicillium (brevicompactum, expansum, notatum and, roqueforti) and Aspergillus (oryzae, repens, niger and terreus) and found that interleukin (IL)-1β, IL-6 and tumour necrosis factor (TNF)-α were defined according to higher or lower levels of IL-1β, named MD High (MDH) and MD Low (MDL), respectively4.

Epigenetic research on MD peripheral blood mononuclear cells (PBMC) also showed two immune phenotypes according to their DNA methylation profiles. Interleukin-1β was measured, and patients were classified into MDH and MDL clusters. Whole genome bisulphite sequencing (WGBS) showed that MDH has over 10,000 differently methylated CpGs (DMC) and over 1500 differently expressed genes (DEG), whereas MDL has less than 200 DMCs and less than 30 DEGs. According to the imputed data, the active MDH cluster showed an inflammatory immune profile5.

Rare monogenic autoinflammatory disorders are a group of diseases characterised by genetic mutations that dysregulate the innate immune system, leading to episodic inflammation, febrile attacks, arthritis and pustular psoriasis8,9. The Inborn Errors of Immunity Committee and Infevers have reported 485 inborn errors of immunity, with 74 pathogenic genetic variants reported in defects in intrinsic and innate immunity10. The most common monogenic autoinflammatory disease is familial Mediterranean fever, which is caused by a mutation in the pyrin coding gene; MEFV. The mutation on MEFV induces a cytokine cascade, that leads to elevated levels of interferon (INF)-γ, TNF-α, IL-1, IL-6 and IL-89.

A mutation in the nucleotide binding domain-like receptor family pyrin domain 3 (NLRP3 gene) causes a dysregulation in the innate immune system that leads to the triggering of the NLRP3 inflammasome, which activates caspase-1 to cleave inactive pro-IL-1β to active IL-1β11,12. The activation of the inflammasome is characterised by elevated levels of IL-1β, IL-18 and other proinflammatory cytokines13. Symptoms associated with rare monogenic autoinflammatory diseases led by a dysregulation of NLRP3 inflammasome include arthralgia, arthritis, myalgias, urticaria and some skin conditions such as rashes9. In some diseases, such as Muckel-Wells syndrome and neonatal-onset multisystem inflammatory disease, SNHL is usually present14,15. However, these conditions do not show vestibular episodic symptoms.

A study using ELISA to measure cytokines in 103 Chinese MD individuals found that Th2 cytokines (IL-4, IL-5, IL-10 and IL-13) and IgE were upregulated compared to controls16. Using mass cytometry, an independent study in Spanish MD patients found two immunophenotypes based on their immune cell populations and cytokine profiles. The first group had a higher count of polymorphonuclear leucocytes, and lower levels of lymphocytes and monocytes. This endophenotype had the same cytokine profile as the one described in the Chinese cohort, and a type II immune response was observed. Conversely, the second group was characterised by monocyte differentiation into macrophages and, patients with elevated levels of IL-1β and autoinflammation6.

Furthermore, by using single-cell RNA sequencing (scRNAseq) and single-cell ATAC sequencing on PBMCs, it has been confirmed that there are two MD immunophenotypes, one inactive with no DEG when compared with controls, and one driven by DEG in the monocytes. The 'monocyte-driven' cluster had elevated levels of IL-1β’s receptors; IL1R1, IL1R2 and IL1RN. Furthermore, receptor CSF2RA of proinflammatory granulocyte-macrophage colony-stimulating factor and IL-8 were upregulated7.

The relationship between the immune response and rare variants in the genome of MD individuals has not been studied, thus, this study aims to investigate the link between rare genetic variants in immune response genes and the monocyte-driven autoinflammatory MD immunophenotype.

Results

A rare missense variant in the KIF1B gene shows association with autoinflammatory transcriptome in MD

We identified five differentially expressed genes (KIF1B, ARAP2, HIVEP3, DOCK8 and DOCK10) that overlapped between the MD scRNAseq dataset (Supplementary Data 1) and showed a rare variant in at least one of the genomic datasets (N = 965 patients). KIF1B (p value adjusted = 1.63 × 10−34, log2FC = 1.56), HIVEP3 (p value adjusted = 1.98E-17, log2FC = 2.29) and DOCK10 (p value adjusted = 3.26 × 10−23, log2FC = 1.54) were found to be upregulated on the monocytes, whereas DOCK8 (p value adjusted = 4.36 × 10−87, log2FC = -1.69), was downregulated in the monocytes. ARAP2 (p value adjusted = 6.93 × 10−10, log2FC = 2.07) was upregulated in the CD8+ T cells. No overlapping genes were found in the remaining PBMCs.

A total of 102 rare variants were identified across the five overlapping genes. Of the total number of rare variants, 97 were missense variants, and five were loss-of-function variants (one was a missense and splice region variant, one was a frameshift variant, two were stop gain variants and one was an in-frame deletion). The KIF1B variant (chr1:10374335 C > T; ENST00000676179.1, p.R1656W) was shared in the sporadic MD (SMD) and familiar MD (FMD) whole exome sequencing (WES) datasets, and the ARAP2 (chr4:36124946 C > A) and HIVEP3 (chr1:41583579 G > T) variants were found in the US and Spanish sporadic cohorts (Table 1). The variants found in the ARAP2 and HIVEP3 genes were excluded since their allelic frequency (AF) were >0.001, higher than the prevalence of MD. Conversely, the variant chr1:10374335 C > T found that the KIF1B gene was retained for further analyses. The KIF1B variant chr1:10374335 C > T found in three individuals, two SMD and one FMD, was associated with MD when it was compared with the AF in gnomAD-non-Finnish European (NFE) (Odds ratio (OR) = 83.39 (15.95–276.98, P value = 9.8 × 10−06).

Table 1 Missense variants found in the canonical transcript of the genes KIF1B, ARAP2 and HIVEP3 replicated in the Spanish and US MD cohorts
Full size table

Clinical Phenotype observed in KIF1B carriers

Patient 1 was a 54-year-old woman without a familial history of MD that started with symptoms at 21 years old. She has developed metachronic bilateral SNHL during the vertigo episodes and reported Tumarkin otolithic attacks and severe tinnitus. Furthermore, she did not report any history of episodic headache or autoimmune disease.

Patient 2 was a 63-year-old woman with a confirmed history of familial MD. Her sister also presented with MD, and her mother reported episodic vertigo. She developed the disease at 42 years old and presented bilateral synchronic SNHL, without migraine or any known autoimmune disease. Unfortunately, her sister’s DNA was not available to confirm an intrafamilial replication.

Patient 3 was also a 68-year-old female with a reported familial history of vertigo and SNHL. She developed MD in the left ear at 56 years old, with a moderate SNHL involving all frequencies, associated with persistent tinnitus without migraine or any known autoimmune disease. The hearing thresholds in the right ear were normal.

Transcripts of Interest for the KIF1B Gene

We used whole blood, EVB-modified lymphocytes and coronary artery gene expression data from the GTEx database to select alternative transcripts expressed in PBMCs. Three transcripts were identified to be highly expressed in the aforementioned tissues; ENST00000377093.8, ENST00000622724.3 and ENST00000377086.5. Transcript ENST00000377093.8 was too short to contain the variant, whereas both ENST00000622724.3 and ENST00000377086.5 included the variant (Fig. 1A). Of note, the canonical transcript (ENST00000676179.1) was not identified in these datasets.

Fig. 1: KIF1B ENST00000622724.3 transcript analysis.
figure 1

A Transcripts of interest for KIF1B gene. B Proportion (%) of RNA transcripts found in the MD sample and Controls. C Modelled protein for KIF1B ENST00000622724.3 transcript, highlighting the WT p.R1630 amino acid without any polar contacts and, the new amino acid p.R1630W which creates a new 2.9 Ä long polar contact with p.S1629.

Full size image

Differentially expressed transcripts of the KIF1B gene

The transcriptome of patient 3 was available to determine transcript isoforms and differential transcript usage (DTU). We found three transcripts in the KIF1B gene. ENST00000377086.5 and ENST00000377093.8 were both downregulated in the MD samples, whereas ENST00000622724.3 was overexpressed in the MD samples, two controls and absent in the remaining 13 controls (Fig. 1B).

All three differentially expressed transcripts were compared to the canonical transcript (ENST00000676179.1) as a reference. The transcript ENST00000377093.8, encoding the protein isoform α, is truncated and does not include the chr1:10374335 C > T variant region. Conversely, ENST00000377086.5, ENST00000622724.3 and ENST00000263934.10 include the variant.

KIF1B variant pathogenicity and splice site prediction

In silico prediction was conducted for the variant NM_001365951.3 (KIF1B): c.4966 C > T (p.Arg1656Trp) chr1:10374335 C > T in the KIF1B gene to evaluate its pathogenicity. ClinVar database classifies the variant as uncertain significance; conversely, PolyPhen2 score is 0.996 (probable damaging) and CADD score is 28.1 (likely pathogenic). Biological constraint measurements yielded significant results exhibiting a Z-score of 4.31, indicating substantial evolutionary conservation and potential functional relevance.

No splice site was predicted in transcript ENST00000377086.5 with SpliceAI or Pangolin for the KIF1B chr1:10374335 C > T variant.

Further analysis of the chr1:10374335 C > T variant in KIF1B selected transcripts was conducted using HSF Pro. This missense variant was demonstrated to disrupt and generate multiple splicing regulatory elements and transcription factor binding. A novel exonic splicing enhancer (ESE) was predicted to be created while simultaneously disrupting an existing ESE. Moreover, an exonic splicing silencer (ESS) was eliminated, and ten new ESS sites were generated (Supplementary Table 2).

Epigenomic changes in the KIF1B gene

Differential accessibility (DA) analysis between controls and MD “pseudo-bulk” ATACseq showed that KIF1B had one upregulated region (chr1:10378100-10379090, p value adjusted = 1.29 × 10−22, log2FC = 2.40). Furthermore, when MD was compared to external controls, KIF1B was found to have the same upregulated region (chr1:10378198-10379098, p value adjusted = 2.77 × 10−11, log2FC = 1.12) (Supplementary Data 2). We did not identify any differentially methylated regions in either MDH or MDL for any of the five selected genes.

Protein modelling

The Kinesin-like KIF1B protein has several isoforms (α, β, c), including the canonical c isoform. Transcripts ENST00000622724.3 and ENST00000263934.10 were selected for the modelling of both wild-type (WT) and mutated forms. ENST00000622724.3 encodes one of the c isoforms, that includes the C-terminal PH domain responsible for membrane localisation and differs from the canonical sequence. While ENST00000263934.10 encodes for isoform β.

The modelled proteins for transcripts ENST00000263934.10 and ENST00000622724.3 revealed that both the wild-type and mutated amino acids are in a large loop region of the protein. The pathogenicity heatmap generated based on data provided by the AlphaMissense database indicated a benign site for this variant (Supplementary Fig. 1). For both isoforms β and c, the chr1:10374335 C > T variant is 45 amino acids upstream from the PH domain N-terminal. The WT Arginine (p.R1610 and p.R1630, respectively) did not show any polar contacts. The variant in isoform β (p.R1610W) did not show any polar contacts either, suggesting the absence of structural alterations to this protein. However, the variant in transcript ENST00000622724.3 (p.R1630W) created a 2.9 Ä long polar contact with Serine-1629 (Fig. 1C). No other relevant structural changes were observed in the modelled proteins.

Identification of protein–protein interactions and biochemical pathways

We analysed the protein–protein interaction for Kinesin-like protein KIF1B. STRING’s network predicted 11 nodes (proteins), 24 edges (interactions), and it exhibited an average local clustering coefficient of 0.901, indicating a significant clustering amongst neighbouring proteins. Statistical analysis revealed a significant enrichment of interactions (PPI) (p value = 2.78 × 10−04) between the predicted proteins when compared to a random network equivalent in size (Supplementary Data 3).

Functional enrichment analysis results using Gene Ontology (GO) Biological Process (BP) for KIF1B yielded significant pathways 'microtubule-based movement' (GO:0007018, p value = 2.69 × 10−08, relative enrichment (RE) = 69.52), and 'cytoskeleton-dependent intracellular transport' (GO:0030705, p value = 7.02 × 10−07, RE = 150.63), whereas GO Cellular Component (CC) and Molecular Function (MF) pathways were 'kinesin complex' (GO:0005871, p value = 1.45 × 10−09, RE = 101.12) and 'kinesin binding' (GO:0019894, p value = 1.72 × 10−05, RE = 69.17), respectively (Supplementary Data 4–6).

No transcription factors (TF) were found to be enriched in KIF1B and their protein–protein interactions.

Discussion

Our study combines several MD-published genomic datasets with single-cell transcriptomic and epigenomic datasets to identify rare variants with functional impact in immune response-related genes. We have found five rare variants replicated across three independent MD cohorts (two sporadic cohorts of European ancestry and a third cohort of familial MD). The clinical characterisation of the Spanish MD cohort was published previously by Frejo et al. in 2016 and 20173,17.

Fifteen genes, including OTOG18, DTNA19, MYO7A20, and TECTA21 have been previously associated with MD, but none of them have been linked to the immune response in MD patients.

After prioritisation and filtering by AF, we have identified an ultrarare missense variant in three unrelated individuals with MD in the KIF1B gene based on protein pathogenicity, splice site prediction, and AF in NFEs. The variant has been reported in 49/1,614,030 individuals, none of them homozygous (global gnomAD AF 3.10−5).

When looked up in ClinVar and UniProt databases, we found that the chr1:10374335 C > T variant is reported twice for the ENST00000622724.3 transcript as p.R1630W, both being associated with Charcot–Marie–Tooth (CMT) and CMT-type 2. Also, the ENST00000377086.5 transcript has the same variant (as p.R1656W) reported three times, two have been associated with CMT and CMT-type 2, and one undetermined association. It was also deemed to be a variant of uncertain significance. Furthermore, when we performed DTU analysis, the transcript ENST00000622724.3 was found in the MD PBMC RNA sample and was absent in all but one of the controls, whereas ENST00000377086.5 was found to be more abundant in controls compared to MD.

The motor protein, Kinesin-like protein KIF1B, is expressed mainly in glial cells and neurons. It is well documented that it is involved in the synaptic vesicle transport and axonal outgrowth22,23. Rare variants in KIF1B have been associated with neuropathies, the most common being CMT, via impairing axonal growth24,25.

The role of Kinesin-like protein KIF1B in the immune response remains elusive. Two studies have shown that KIF1B is differentially expressed in patients who suffer from septic shock compared to controls26,27. The KIF superfamily of proteins has been found to transport mitochondria within cells28. Furthermore, KIF1B and YME1L1 metalloproteinase mediates mitochondrial fission to induce mitochondrial apoptosis29. Nitric oxide (NO) is a byproduct of the immune myeloid cells' inflammatory response30. Mitochondrial respiration is blocked by NO gas31, which can lead to cell stress, less production of ATP, and apoptosis32.

By using electroporation and small interfering RNA (siRNA) to silence genes KIF1A and KIF1B in mature dendritic cells (mDCs) derived from monocytes, no changes in maturation status or cell death was observed after 2 days. The study also showed that KIF depleted mDCs had lower levels of residual p62 proteins after HSV-1 infection, in contrast to controls, supporting higher levels of autophagy33.

The Kinesin-like protein KIF1B has two main isoforms, KIF1Bα and KIF1Bβ. These protein isoforms exhibit distinct functions, with the β isoform specifically involved in apoptotic signalling and potential tumour suppression34. Investigations have elucidated KIF1Bβ‘s involvement in neurodegenerative and hereditary conditions, due to its role in transporting vesicles associated with the IGF1R receptor, which is crucial for neuronal survival and regeneration25,35.

Structurally, Kinesin-like protein KIF1B comprises a conserved N-terminal motor domain that binds to microtubules and generates movement through ATP hydrolysis, as well as a C-terminal PH domain specialised in interactions with membrane lipids, particularly negatively charged phosphoinositide. This PH domain facilitates the protein’s targeting to specific membrane regions, serving as an anchor essential for its intracellular localisation and function36,37.

The R1610W variant, which substitutes arginine with tryptophan a few amino acids upstream of the PH domain, can induce significant functional alterations. Arginine, with its positive charge and high reactivity, is critical for electrostatic interactions with membrane lipids’ phosphate groups38. This interaction is essential for the precise localisation of proteins with PH domains, such as Kinesin-like protein KIF1B, which rely on this contact to direct vesicles and other intracellular cargo. Tryptophan, conversely, is hydrophobic with a bulky aromatic ring, which not only reduces the PH domain’s affinity for charged lipids but may also distort the domain’s local structure39. The establishment of a polar contact underscores this phenomenon, potentially enhancing the stability of the protein region and consequently further hindering its capacity to interact with the lipid membrane40.

The relationship between KIF1B and MD immune response in the autoinflammatory phenotype is yet to be fully elucidated. Menière’s Disease is considered a multifactorial disorder associated with a persistent inflammatory response with three immune phenotypes6,7,16,41,42. Although 15 genes had been associated with familial MD, none of them seem to be involved in the immune response. The contribution of rare genetic variation to immune response in MD has seldom been investigated43, mostly because genomic and transcriptomic data in the same patients are lacking.

This study shows some limitations. First, the number of individuals with the rare variant in the KIF1B gene is small, and the single-cell transcriptome dataset was obtained from a few MD patients. The autoinflammatory transcriptome was only available in one of the patients, and we cannot establish a causal relationship between the rare variant and autoinflammation. Simulating protein 3D structures with specific variants allows researchers to predict changes in stability, binding affinity, and function, improving understanding of variant pathogenicity44. Lastly, functional studies will be needed to confirm KIF1B function in the immune response. However, these methods depend on high-quality structural templates and struggle to accurately represent protein flexibility. Therefore, experimental validation is often necessary to confirm predicted interactions and binding affinities from these simulations.

In conclusion, a rare variant in the KIF1B gene is associated with autoinflammation in MD according to genomic, single-cell transcriptomic and ATACseq datasets. Although the function of KIF1B in the immune response is not known and further studies are needed, these findings anticipate a potential role for rare variation in KIF1B functional regulation which may contribute to autoinflammation in MD.

Methods

Human ethics

This study was conducted in compliance with the ethical standards of the Declaration of Helsinki, and the Granada Clinical Research Ethics Review Board (PI17/1644;PI20-1126) and the Human Ethics Research Committee (2023/HE000199) from The University of Sydney approved the protocol for this study. Written informed consent was obtained by the participants to participate in the original studies5,7,45,46,47.

Retrieval of preprocessed datasets

H5 data files of epigenomics and transcriptomics of MD PBMCs were retrieved from Gene Expression Omnibus (GEO) and previously published studies (GEO accession numbers GSE269114 and GSE269117). DEGs and differentially accessible regions in the chromatin were identified with a p value adjusted of <0.05 and log fold change (log2FC) of ±1.5. The original datasets were generated using scRNAseq and 'pseudo-bulk' ATAC sequencing7 (Supplementary Table 1 and Datasets 1, 5). Besides, we selected BAM files deposited in the European Nucleotide Archive5 to identify methylated cytosine regions in the MD epigenome. Peripheral blood cells were used for WGBS (Supplementary Table 1 and Dataset 2). In addition, we also retrieved rare variants from whole genome sequencing (WGS)45 (Supplementary Table 1 and Datasets 3, 4), and WES datasets from SMD and FMD previously published46. Finally, MD RNA sequencing FASTQ files were also used to compare them with single-cell RNAseq from PBMCs47 (Supplementary Table 1 and Dataset 6).

Identification of immune response genes with rare variants

The genome sequencing dataset (N = 511) and both SMD and FMD exome sequencing datasets (N = 454) were used to retrieve rare variants (AF <0.05) in overlapping genes. To locate immune response (IR) related genes, DEGs lists of single-cell populations of B cells, CD4+ T, CD8+ T, natural killer (NK) cells, and monocytes from the scRNAseq dataset were used7. Rare variants of IR genes were compared amongst the three datasets by manual inspection to find overlaps. 'Pseudo-bulk' ATACseq and WGBS datasets were used to identify epigenetic modifications in the selected genes5,7.

Statistical analysis

The AFs were obtained in both MD cohorts and compared with the AFs from NFE reference dataset from gnomAD (version v4.1.0) in order to calculate OR with 95% confidence interval (CI).

Transcripts of interest

The GTEx database (https://gtexportal.org/home/) was used to search for alternative transcripts expressed in immune cells for the genes of interest. ClinVar48 and UniProt49 were used to search for known effects on the variants of interest in the selected alternative transcripts.

Differential transcript usage

RNA-Seq FASTQ files were processed with Salmon v.0.10.050 and the GENCODE v28 transcript fasta51 file to quantify undanced and transcript lengths for controls and MD samples. Differential transcript usage was performed by uploading the quant.sf files to R and analysed using the default pipeline from the DRIMSeq v1.32.0 package52. The squile file was generated using R package GenomicFeatures v1.56.053 and GENCODE v28 gtf file.

Pathogenicity and splice site prediction

In silico pathogenicity predictions of selected variants was assessed utilising CADD score, PolyPhen254, and REVEL55 tools. Genomic constraint measurements were evaluated for 1 kb region surrounding selected variants using Z scores obtained from the gnomAD dataset v3.2.156.

Splice sites were predicted in the selected genes using SpliceAI57 and Pangolin58. The GRCh38 human genome was used as reference, and a 500-base pair (bp) upstream and downstream window from the variant position on the genome was applied. Reference and alternate scores for SpliceAI and Pangolin splice sites were calculated.

Human Splice Finder Pro59 (HSF Pro) was used to predict putative donor and acceptor splice sites, cis-acting elements such as ESE, ESS, intronic splicing enhancers (ISE) and intronic splicing silencers (ISS) and, branch points in the variants of interest.

Structural protein modelling

A comparative analysis of the transcripts of interest was conducted utilising Ensembl (http://www.ensembl.org/Homo_sapiens/Info/Index) to assess their suitability for modelling. The selected transcripts were evaluated against the canonical one based on their exon sequence. Protein modelling was employed to assess how the candidate variants affected the protein structure. Pathogenicity predictions for the missense sites were obtained from the AlphaMissense database60,61. Canonical sequences were retrieved from UniProt49 databank and from the AlphaFold2 Protein Structure Database60,61 and used to generate structural models of the wild-type and mutated proteins by homology, using MODELLER62 v10.5.

To ensure the quality and reliability of the generated structures, all the models underwent quality control assessments. Both Modeller and AlphaFold2 provide built-in validation metrics, including DOPE (discrete optimised protein energy), molpdf, GA341 (for Modeller) and sequence coverage and IDDT (for ColabFold). For Modeller, the lowest DOPE and GA341 scores were chosen to generate the best protein models. Additionally, all structures were subjected to further analysis using the SAVES v6.1 server (https://saves.mbi.ucla.edu/), which employs multiple validation algorithms to evaluate the quality of the protein structural models, including PROCHECK63, WHATCHECK (https://swift.cmbi.umcn.nl/gv/whatcheck/index.html) and ERRAT64. The first two tools evaluated the stereochemical quality of the modelled structure using Ramachandran graphs. ERRAT examines the atomic interactions and provides an overall quality factor for identifying incorrectly built regions in protein models.

Finally, the structural models were loaded at PyMOL (Schrodinger, LLC. 2010. The PyMOL Molecular Graphics System, Version 3.0.x) for their visualisation and to compare the WT and mutated tridimensional structures of the selected isoforms.

Functional analysis of IR genes with rare variants

The STRING v1265 and the Human Protein Atlas66 immune RNA databases were used to find protein–protein interactions and to identify RNA transcripts with similar levels of expression as the selected IR genes with rare variants. For STRING analysis, all nodes and edges were considered. GeneCodis v467 software’s functional annotations GO BP, GO CC, GO MF and KEGG Pathways and regulatory annotation CollecTRI TFs were used to identify active biochemical pathways and TF using the output from STRING and Human Protein Atlas.