In Silico Localization of Perilymph Proteins Enriched in Meńier̀e Disease Using Mammalian Cochlear Single-cell Transcriptomics

Abstract

Hypothesis:

Proteins enriched in the perilymph proteome of Meńier̀e disease (MD) patients may identify affected cell types. Utilizing single-cell transcriptome datasets from the mammalian cochlea, we hypothesize that these enriched perilymph proteins can be localized to specific cochlear cell types.

Background:

The limited understanding of human inner ear pathologies and their associated biomolecular variations hinder efforts to develop disease-specific diagnostics and therapeutics. Perilymph sampling and analysis is now enabling further characterization of the cochlear microenvironment. Recently, enriched inner ear protein expression has been demonstrated in patients with MD compared to patients with other inner ear diseases. Localizing expression of these proteins to cochlear cell types can further our knowledge of potential disease pathways and subsequent development of targeted therapeutics.

Methods:

We compiled previously published data regarding differential perilymph proteome profiles amongst patients with MD, otosclerosis, enlarged vestibular aqueduct, sudden hearing loss, and hearing loss of undefined etiology (controls). Enriched proteins in MD were cross-referenced against published single-cell/single-nucleus RNA-sequencing datasets to localize gene expression to specific cochlear cell types.

Results:

In silico analysis of single-cell transcriptomic datasets demonstrates enrichment of a unique group of perilymph proteins associated with MD in a variety of intracochlear cells, and some exogeneous hematologic and immune effector cells. This suggests that these cell types may play an important role in the pathology associated with late MD, suggesting potential future areas of investigation for MD pathophysiology and treatment.

Conclusions:

Perilymph proteins enriched in MD are expressed by specific cochlear cell types based on in silico localization, potentially facilitating development of disease-specific diagnostic markers and therapeutics.

Meńier̀e disease (MD) is a chronic, progressive disorder of the inner ear characterized by episodic vertigo, variably associated with fluctuating low- to mid-frequency sensorineural hearing loss, tinnitus, and aural fullness (1,2). The disease leads to substantial functional disability in affected individuals, along with decreased quality of life and significant direct and indirect healthcare costs (2). Endolymphatic hydrops (ELHs), expansion of the endolymph-containing scala media into the scala vestibuli, is considered a pathologic hallmark of MD. This finding, however, is insufficient for diagnosis and the specific cause of MD remains unclear (1). Without a clear understanding of the pathologic mechanisms of disease, current diagnostic modalities are unable to identify patients who may be more amenable to treatment (3,4); and treatment modalities often fail to adequately control symptoms, particularly those related to hearing loss (HL) (5–8).

Recent technological advancements have revealed novel disease-specific biomarkers and allowed for development of more targeted therapeutics. Numerous groups have demonstrated the feasibility of perilymph sampling for characterization of the cochlear microenvironment, including identification of biomolecular profiles unique to various otologic pathologies (9–18). Schmitt et al. recently analyzed perilymph from cochlear implant patients and found a unique proteomic expression profile for patients with MD compared to those with enlarged vestibular aqueduct (EVA) and otosclerosis (12). We still, however, lack a clear understanding of why the MD perilymph expression profile is distinct from that of other patients.

Single-cell RNA-sequencing (scRNA-Seq) and single-nucleus RNA-sequencing (snRNA-Seq) have emerged as techniques to better understand gene-regulatory networks by characterizing transcriptional profiles of individual cell types and different cellular states (19–21). Nelson et al. recently utilized publicly available scRNA-Seq and snRNA-Seq datasets from the mouse inner ear to localize genes associated with sudden sensorineural hearing loss to specific inner ear cell types (22). Similarly, we aimed to determine if RNA-Seq databases could provide additional insight into the unique proteomic perilymph profile of MD patients, specifically whether the genes associated with MD-enriched perilymph proteins might be expressed by specific cells within the inner ear. We hypothesized that comparing single-cell and single-nucleus transcriptome datasets from the mammalian cochlea to the perilymph proteomic profiles identified by Schmitt et al. would localize enriched perilymph proteins in MD to specific cochlear cell types.

MATERIALS AND METHODS

Perilymph Proteome Data Processing

We analyzed the complete raw perilymph proteomics data from the Hannover cochlear proteome dataset up to 2019, a portion of which was analyzed previously using relative label-free quantitation (LFQ) intensity values (12). This perilymph had been previously obtained from humans undergoing cochlear implantation with different etiologies of HL including vestibular schwannoma (n = 10), MD (n = 12), sudden hearing loss (n = 9), EVA (n = 14), otosclerosis (n = 10), and unknown etiology (n = 26). The proteomics data were visualized in a dimensionally reduced 2D uniform manifold approximation and projection (UMAP) plot with Scanpy (v1.8) function scanpy.tl.umap using default settings. The dataset was clustered both by using disease group designations and by using an unbiased modularity-based Leiden clustering algorithm (resolution = 1.0) implemented in Scanpy (v1.8), with clustering remaining unchanged despite the above approaches (see Supplemental Digital Content 1, http://links.lww.com/ONO/A10). Briefly, to explain dimensional reduction, each of the given proteins in the dataset represents a dimension. Therefore, each original patient sample is represented by 933 individual dimensions or proteins. This high-dimensional sample is transformed into a low-dimensional display UMAP plot that enables visualization in 2D while preserving the data’s global structure as much as possible. This dimensional reduction process is performed by using the Scanpy (v1.8) function scanpy.tl.umap with default settings. In this case, dimensional reduction simplifies this variability into a two-dimensional UMAP plot, where each dot represents in essence the variability of a given single sample; and the distance between each dot is representative of the similarity or difference between given patient samples, with similar samples “located” more closely to one another.

To compare protein expression and take advantage of the variability across samples, the nonparametric Wilcoxon rank test was performed on LFQ intensity values to generate a protein rank score utilizing the Scanpy function tl.rank_genes_groups with default settings. Protein rank score is the standard score (Z-score) of the test P value for each protein for each sample group. The value “0” indicates the mean of the P values and “1” indicates one standard deviation from the mean in the direction of greater significance. To designate enrichment for a given protein/gene in the perilymph sample, an arbitrary cutoff of Wilcoxon rank score > 1 was chosen for selection of MD-enriched proteins. This analysis established a hierarchy of characteristic proteins for each group of samples. It is important to note that the entire Hannover proteome dataset up to 2019 for the above diagnoses was analyzed, in distinction to Schmitt and colleagues, and that the advantage of utilizing a machine learning approach, versus the approach utilized previously (12), was that the variability across the samples could be leveraged in this analysis to determine characteristically enriched proteins for each diagnostic group without the need to average, hence all samples were utilized.

scRNA-Seq/snRNA-Seq Data and Software Availability

While single-cell transcriptional databases are not available for the human inner ear, datasets are available from an increasing number of single cell transcriptional datasets from the mouse inner ear including some adult mouse inner ear datasets. Our curated list of MD perilymph proteins was cross-referenced against the following previously published datasets. snRNA-Seq datasets of postnatal day 30 (P30) CBA/J mouse stria vascularis (SV) (23) were utilized (GEO Accession ID: GSE136196) (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE136196) and are available through the gene Expression Analysis Resource (gEAR) Portal, a website for visualization and comparative analysis of multi-omic data, with an emphasis on hearing research (https://umgear.org//index.html?layout_id=b50cae7a). A scRNA-Seq dataset from the P7 CD1 developing mouse cochlea, comprising inner hair cells (IHCs), outer hair cells (OHCs), and supporting cells that include inner phalangeal, pillar, and Deiters cells (24), was utilized (GEO Accession ID: GSE137299) and is available through gEAR (https://umgear.org//index.html?layout_id=f7baf4ea). Published scRNA-Seq datasets of P15 IHC, OHC, and Deiters cells obtained from C3HeB/FeJ mice (25) were utilized (GEO Accession ID: GSE114157) (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE114157) and are available through the Molecular Otolaryngology and Renal Research Laboratories (MORL) (https://morlscrnaseq.org/). A published scRNA-Seq dataset from P25-27 adult mouse spiral ganglion neurons (SGN) including Type 1A, 1B, and 1C SGN as well as Type 2 SGN (26) from mice consisting of C57Bl/6J and CD1 mixed background strain was utilized (GEO Accession ID: GSE114997) and is available at the following link (https://screen.hms.harvard.edu/harvard/), as well as through gEAR (https://umgear.org//index.html?layout_id=fee360e8). Finally, a published scRNA-Seq dataset from 8- to 12-week-old C57Bl/6 adult mouse dorsal root ganglia (DRG) satellite glial cells and Schwann cells (27) was utilized (GEO Accession ID: GSE139103).

Data Visualization

P30 SV snRNA-Seq

Previously published P30 SV snRNA-Seq data were preprocessed by Scanpy (v1.5.1) with criteria as previously described (23).

P7 IHC, OHC, and Supporting Cells

A previously published normalized dataset from Kolla and colleagues (24) was processed by Scanpy (v1.5.1). “Rik” and “Gm-” genes were filtered in all downstream analyses. Cell clustering and annotation was performed using modularity-based clustering with Leiden algorithm (resolution = 0.8) implemented in Scanpy.

P15 IHC, OHC, and Deiters Cells

A data expression matrix from previously published P15 IHC, OHC, and Deiters cell scRNA-Seq datasets (25) was normalized by Scanpy (v1.5.1) function pp.normalize_total with parameter “exclude_highly_expressed” set as True. Normalized counts were scaled utilizing Z-score prior to plotting on heatmaps.

P25-27 SGN scRNA-Seq

A previously published normalized dataset from the P25-27 mouse SGN scRNA-Seq dataset from Shrestha and colleagues (26) was processed using Scanpy (v1.5.1). Cell clustering and annotation was performed using modularity-based clustering with Leiden algorithm (resolution = 2.0) implemented in Scanpy. Normalized counts were scaled utilizing Z-score prior to plotting on heatmaps.

Eight- to 12-week-old DRG Satellite Glial Cell and Schwann Cell Single-cell Transcriptional Profiles

A previously published normalized dataset from Avraham and colleagues (27) was processed and visualized as described above.

All heatmaps were plotted with Seaborn (v0.11).

Gene Ontology Analysis of MD-enriched Perilymph Proteome

MD proteins from the perilymph proteome dataset with a Wilcoxon rank score > 1 were selected for corresponding gene ontology (GO) analysis utilizing Enrichr (https://maayanlab.cloud/Enrichr/) (28–30). Enrichr is an integrated web-based application that includes updated gene-set libraries, alternative approaches to ranking enriched terms, and a variety of interactive visualization approaches to display the enrichment results as previously described (19,29,31,32). Enrichr employs 3 approaches to compute enrichment as previously described (32). The combined score approach, where enrichment was calculated from the combination of the P value computed using the Fisher exact test and the Z-score, was utilized, as this measure has greater ability to detect true transcriptional targets of perturbations based on published benchmarking analyses (29). Top biological process GO terms were chosen by utilizing the combined score as described. Genes corresponding to proteins with enrichment in SGN and SV cell types, as determined by Wilcoxon score described above, were also identified, and biological process GO terms were identified as previously described (29). The complete list of enriched GO terms is provided (Supplemental Digital Content 2 for SGN, http://links.lww.com/ONO/A11 and Supplemental Digital Content 3, http://links.lww.com/ONO/A12 for SV cell types), with GO terms plotted utilizing the Python Seaborn package (v0.11).

RESULTS

Hierarchical clustering analysis of the published perilymph proteomics data (12) demonstrates distinct clustering of perilymph samples from MD patients, with the exception of one sample, separately from patients with other etiologies of HL (Fig. 1A). The top 20 MD-enriched genes, as determined by Wilcoxon rank scores, are shown in Figure 1B and Table 1. The MD condition had the highest rank scores, with scores >3 for the top 20 proteins and > 4 for 7 proteins. To localize candidate gene expression to the single-cell level, publicly available scRNA-Seq and snRNA-Seq datasets from the mammalian cochlea were cross-referenced against more highly abundant protein expression for: (1) MD compared to EVA and otosclerosis, as published by Schmitt and colleagues (12) (Fig. 2), (2) MD and EVA compared to otosclerosis, as published by Schmitt and colleagues (12) (Fig. 3), and (3) MD compared to other HL etiologies, as identified by hierarchical clustering described above (Fig. 4).

FIG. 1.

Unbiased machine learning-based clustering of perilymph proteome samples with samples color coded by disease entity. Disease entities included vestibular schwannoma (V-Schwann), Ménière disease, sudden hearing loss (SHL), enlarged vestibular aqueduct (EVA), otosclerosis, and unknown etiology. A, Uniform manifold approximation and projection (UMAP) plot demonstrates clustering of Ménière disease perilymph proteome samples from other disease etiologies. B, Ranked expression as determined by Wilcoxon rank score of the top expressed perilymph proteins by disease etiology are shown in bar plots.

TABLE 1.

Top 20 ranked proteins as determined by Wilcoxon rank score in Ménière disease subgroup based on hierarchical clustering

Protein name	Gene name
Chloride intracellular channel protein 1	CLIC1
Short-chain dehydrogenase/reductase family 9C member 7	SDR9C7
Triosephosphate isomerise	TPI1
Complement factor H-related 1	CFHR1*
Glutathione S-transferase P	GSTP1
Actin, cytoplasmic 1	ACTB
S-formylglutathione hydrolase	ESD
Protein FAM83H	FAM83H
Fibrinogen β chain	FGB
Peroxiredoxin 6	PRDX6*
α-mannosidase 2	MAN2A1
Complement C1r subcomponent	C1R
γ-enolase	ENO2
Protein disulfide isomerase family A member 3	PDIA3*
Vinculin	VCL*
Rho GDP dissociation inhibitor α	ARHGDIA*
Actinin α-1	ACTN1*
Sarcoglycan ε	SGCE*
Filamin-B	FLNB
Carboxypeptidase N subunit 1	CPN1*

^*Proteins that were not identified as highly abundant proteins for the MD condition by Schmitt et al.

FIG. 2.

Perilymph proteins highly abundant in Ménière disease but of low abundance in enlarged vestibular aqueduct and otosclerosis, cross-referenced with (A) postnatal day 30 (P30) stria vascularis (SV) single-nucleus RNA-sequencing (snRNA-Seq); (B) P7 Organ of Corti single-cell RNA-sequencing (scRNA-Seq); (C) P25-27 spiral ganglion neuron (SGN) scRNA-Seq. Each heatmap shows cell types along the horizontal axis and MD-implicated genes along the vertical axis. Expression is represented using scaled Z-scores; expression of a given gene is averaged across all cell types to establish a mean, and the Z-score represents the differential expression of that gene as a standard deviation from the mean for a given cell (where red is above the mean and blue is below the mean).

FIG. 3.

Perilymph proteins highly abundant in Ménière disease and enlarged vestibular aqueduct but of low abundance in otosclerosis, cross-referenced with (A) postnatal day 30 (P30) stria vascularis (SV) single-nucleus RNA-sequencing (snRNA-Seq); (B) P7 Organ of Corti single-cell RNA-sequencing (scRNA-Seq); (C) P25-27 spiral ganglion neuron (SGN) scRNA-Seq. Each heatmap shows cell types along the horizontal axis and MD-implicated genes along the vertical axis. Expression is represented using scaled Z-scores; expression of a given gene is averaged across all cell types to establish a mean, and the Z-score represents the differential expression of that gene as a standard deviation from the mean for a given cell (where red is above the mean and blue is below the mean).

FIG. 4.

Perilymph proteins highly abundant in Ménière disease based on hierarchical clustering, cross-referenced with (A) postnatal day 30 (P30) stria vascularis (SV) single-nucleus RNA-sequencing (snRNA-Seq); (B) P7 Organ of Corti single-cell RNA-sequencing (scRNA-Seq); (C) P25-27 spiral ganglion neuron (SGN) scRNA-Seq. Each heatmap shows cell types along the horizontal axis and MD-implicated genes along the vertical axis. Expression is represented using scaled Z-scores; expression of a given gene is averaged across all cell types to establish a mean, and the Z-score represents the differential expression of that gene as a standard deviation from the mean for a given cell (where red is above the mean and blue is below the mean).

For condition (1) [MD compared to EVA and otosclerosis] (12), the heatmap displaying MD-implicated gene expression amongst P30 SV cells (Fig. 2A) shows enrichment of Flnb, Mdh1, and Atrn in marginal cells, and Actb in basal cells. The P7 Organ of Corti heatmap (Fig. 2B) demonstrates relative enrichment of Atrn and Chga in IHC and OHC, and Ppia across IHC, OHC, and Schwann cells. The P25-27 SGN analysis (Fig. 2C) shows generalized enrichment of nearly all MD-implicated genes among SGN, with the highest relative expression for Clic1 and Gstp1 in Type 2 neurons.

For condition (2) [MD and EVA compared to otosclerosis] (12), MD-implicated gene expression amongst P30 SV cells (Fig. 3A) shows enrichment of Qsox1, Lamb1, and Atrn in marginal cells; Nrcam in intermediate cells; and Hspg2, Ptgds, Fn1, Itih5, and Cfh in basal cells. The P7 Organ of Corti heatmap (Fig. 3B) demonstrates enrichment of Hspg2, Lamb1, and Itih5 in Schwann cells; Atrn in IHC and OHC; and Cntn1 in pillar cells. In the P25-27 SGN analysis (Fig. 3C), there is generalized enrichment of approximately half of implicated genes amongst SGN, though this is not as striking as for condition (1).

For condition (3) [MD compared to other HL etiologies based on hierarchical clustering], MD-implicated gene expression amongst P30 SV cells (Fig. 4A) shows enrichment of Flnb in marginal cells, similar to condition 1; as well as enrichment of Actn1, Actb, and Eef2 in basal cells, and Actn1 in Reissner’s membrane cells. The P7 Organ of Corti heatmap (Fig. 4B) demonstrates enrichment of Eef2 across all cell types. Similar to conditions 1 and 2, the P25-27 SGN analysis (Fig. 4C) shows enrichment of the majority of MD-implicated genes amongst SGN, again with the highest relative expression for Clic1 and Gstp1 in Type 2 neurons.

The analysis for P15 IHC, OHC, and Deiters cells and 8-12-week-old DRG cells did not demonstrate any clear trends in differential expression amongst cell types (see Supplemental Figures 4 and 5, http://links.lww.com/ONO/A13). Together, the above data demonstrate that a range of different cells contribute to the proteome that distinguishes MD perilymph from other disorders.

GO analysis of MD-enriched proteins as determined by hierarchical clustering demonstrated enrichment for proteins involved in regulation of complement activation (GO:0030449), regulation of immune effector process (GO:0002697), canonical glycolysis (GO:0061621), glucose catabolic process to pyruvate (GO:0061718), and classical pathway complement activation (GO:0006958) (Fig. 5A). GO analysis of MD-enriched proteins in SV and SGN cell types (Fig. 5B, C, respectively) demonstrated similar enrichment for proteins involved in canonical glycolysis (GO:0061621), glucose catabolic process to pyruvate (GO:0061718), glycolytic process (GO:0006096), glycolytic process through glucose-6-phosphate (GO:0061620), and platelet aggregation (GO:0070527).

FIG. 5.

Gene ontology (GO) analysis of perilymph proteome enriched in Ménière disease (MD). A, Biological process (BP) GO analysis of all MD-enriched perilymph proteins. B, BP GO analysis of MD-enriched perilymph proteins expressed in the stria vascularis (SV). C, BP GO analysis of MD-enriched perilymph proteins expressed in spiral ganglion neurons (SGN). Scores on horizontal axes represent the combined score metric used in Enrichr (29) that takes into account both the P value and Z-score for deviation from the expected rank.

DISCUSSION

MD is a clinical entity whose diagnosis requires presence of a constellation of otovestibular symptoms (1,2). Application of diagnostic criteria remains challenging, however, due to the disease’s fluctuating and variable clinical presentation, both between different patients and across time for individual patients. Furthermore, other diseases, such as autoimmune inner ear disease, otosyphilis, and vestibular migraine may mimic MD clinically and histopathologically (2). Enhancing our understanding of the pathophysiology underlying MD and its variable phenotypic expression is ongoing, now with new technologies facilitating characterization of biomolecular “signatures” of disease (9,10,12,15,17–19,22,23,33,34). Single-cell and single-nucleus transcriptomics is one such domain allowing for both characterization of gene expression at an individual cellular level and in silico localization of pathology-implicated genes to specific cell types (21,35). In this study, we demonstrate in silico localization of MD-implicated proteins from human perilymph (12) to specific inner ear cells using previously published scRNA-Seq/snRNA-Seq datasets.

Interspecies comparison of murine scRNA-Seq/snRNA-Seq datasets and human perilymph proteomic profiles, along with a lack of transcriptomic profiles for all cochlear cell types and developmental stages, may limit the applicability of our findings. However, single-cell transcriptomic databases for human cochlear cells are not available due to difficulties accessing the bony labyrinth and obtaining sufficient tissue for analysis, in addition to the unacceptable morbidity associated with obtaining cochlear tissue from living humans (36). The mouse has long been utilized as a model to improve our understanding of mammalian biology and human diseases. Due to similarities in the mouse and human cochleae, genomic analyses have allowed for discovery of candidate genes in human disease, such as the Pou4f3 mouse gene and its associated knockout leading to discovery of the orthologous human gene for DFNA15 (37).

Multiple murine transcriptomic databases are available and have been well-validated across multiple organ systems, including the inner ear (19,23–25). Furthermore, a number of other medical subspecialties have compared murine and human transcriptomic data to elicit new scientific hypotheses and better understand the physiologic underpinnings of organ function (38–47). In light of interspecies comparisons across multiple fields of medicine, we believe that our experimental approach is a valid method to localize implicated MD biomarkers to expression at a single-cell level to generate hypotheses for future research. We did attempt to include a wide variety of available cochlear cell databases (hair cells, supporting cells, SGN, SV cell types, and endothelial cells and pericytes from the DRG) in our analysis. The subsequent discussion highlights patterns in which perilymph protein expression was particularly enriched.

Our hierarchical clustering analysis of the perilymph proteomic data from Schmitt et al. (12) identified 20 top-ranked proteins in the MD subgroup (Table 1). Seven proteins had a Wilcoxon score >4, including CLIC1, SDR9C7, TPI1, CFHR1, GSTP1, ACTB, and ESD. Eight of the 20 proteins identified by our analysis were not cited in the list of 33 highly abundant proteins for MD published by Schmitt and colleagues; however, all but 1 (CFHR1) of the 7 proteins with a Wilcoxon score >4 were captured in their list of 33 proteins (12). Because Schmitt et al. selected for proteins that were expressed in at least eight perilymph samples per disease group, it is possible that we identified additional MD-implicated proteins because all perilymph datasets were included in our analysis, and because our analysis leveraged, rather than ignored, the variability across all samples to identify additional markers.

In silico cross-referencing of the P30 SV snRNA-Seq dataset to the MD-implicated proteins identified both by Schmitt et al. and by our hierarchical clustering analysis showed enhanced expression of Flnb by marginal cells and Actb by basal cells. Gu et al. recently performed a similar in silico localization utilizing P30 SV scRNA-/snRNA-Seq datasets and MD-implicated genes identified in a systematic literature review (35). The implicated genes identified by Gu et al. do not coincide with the genes identified in our study. This discrepancy may reflect differences in study sample origins, as Gu et al. obtained a large proportion of their data from peripheral blood cell samples, while our data reflect the microenvironment of human perilymph exclusively. In regard to our findings, Flnb encodes filamin B, a protein involved in cell signaling and cytoskeletal changes. Mutations in this gene have been linked to Larsen syndrome, a condition characterized by skeletal anomalies and HL (48). Actb encodes β-actin, a ubiquitously expressed actin within cytoplasm, important for maintaining cell structure. β-actin has been shown to localize in basal cells of the SV and to have a role in hair cell stereocilia maintenance, with protein loss associated with progressive HL (49,50). While the significance of these genes and their associated cytoskeletal proteins is still unclear, it is possible that alterations in their expression can contribute to ELH by affecting intercellular junction integrity, maintenance of the endocochlear potential, and/or other aspects of SV function.

Interestingly, our analysis demonstrated an enrichment of protein expression by all SGN cell types, particularly in conditions 1 (MD compared to EVA and otosclerosis) and 3 (hierarchical clustering). This suggests a possible neural component to MD pathology, and a potential contribution to the sensorineural hearing loss and marked reduction in speech discrimination of patients with MD, clinical features that have not been clearly linked to the ELH found in MD-affected temporal bones (51–55).

A neural pathophysiologic mechanism in MD was first suggested by Nadol in 1987. His microscopic analysis of temporal bones from a patient with unilateral MD demonstrated no significant difference in SGN density between the two ears, but the diseased ear did show fewer afferent nerve endings associated with IHC and OHC and fewer afferent synapses per IHC. This finding implicated a phenomenon in which degeneration of neuronal peripheral processes precedes and eventually leads to cell body death (56). Subsequent studies again noted patterns of neuronal degeneration, including significantly decreased SGN nuclei and axon diameters in MD-affected temporal bones compared to normal controls (57,58). Sperling et al. and Kariya et al. later noted greater SGN loss in temporal bones of MD patients compared to patients with ELH without classic MD symptoms and normal controls, respectively (53,59). In animal models of ELH, significant apical SGN and OHC loss has been observed, with SGN degeneration tending to exceed the magnitude of IHC loss (60–63). Though the significance of globally enhanced SGN protein expression in our study remains to be determined, the histopathologic literature lends support to our findings of a potential SGN contribution to MD pathophysiology.

It is important to note that many of these perilymph-enriched proteins are intracellular or membrane-spanning. Detection of these proteins in the perilymph may suggest that contents of dying cells are being shed into the perilymph and/or that implicated cell types have amplified secretory responses to the pathophysiology occurring within the MD-affected cochlea. While the exact mechanism of cochlear cell responses remains to be elucidated, it is possible that extracellular vesicles (EVs) are contributing to the identified perilymph proteomes. EVs are membrane-bound nanovesicles of varying size (30–4000 nm) that have been identified in recent years as important biomolecules for intercellular communication, transferring their contents (eg, intracellular proteins, RNA, lipids) to near and sometimes distant organs in states of health and disease (64,65). In Alzheimer disease (AD), EVs have been implicated in the pathogenesis of amyloid plaque formation and have also been found in human serum, suggesting a role for their vesicular contents as disease biomarkers (64,66,67).

The literature on EVs in the inner ear is overall lacking, but Wong et al. recently used an ex vivo rat model of the inner ear and successfully isolated EVs from culture supernatants, with the ototoxicity treatment group demonstrating particular enrichment for Tmem33, a transmembrane protein localized in the endoplasmic reticulum and nucleus and previously described in noise-traumatized rat cochleae (65). A later study by Warnecke and colleagues showed that treatment with human mesenchymal stromal cell-derived EVs led to increased survival of rat SGN in vitro and to hair cell protection from noise trauma in murine in vivo models. While the exact mechanisms of protection remain to be elucidated, the authors draw on other EV-related literature and postulate a potential EV-mediated modulation of inflammatory cytokines and/or cellular signal transduction, the latter via EV-abundant tetraspanins (68). More recently, Zhuang et al. successfully isolated EVs from human perilymph using a novel microfluidics technology (69). Though characterization of human perilymph EV content is still needed and our study does not experimentally validate the findings of our in silico analysis, the above cited literature related to other neurodegenerative diseases and EVs specifically, provides a possible framework from which to generate hypotheses for further experimentation related to our study findings.

Our GO analysis of MD-enriched proteins suggests potential contributions of inflammation to MD pathology, in particular mechanisms related to complement activation, the immune effector process, and platelet aggregation. The complement system has long been recognized as a key component of the innate immune system, enhancing the ability of immune effector cells to clear pathogens and damaged cells from the body (70). Complement also plays an important role in the adaptive immune system, modulating B-cell maturation and T-cell effector functions (71–73). In addition, complement activation induces platelet activation and aggregation, and platelet aggregation may activate the complement system as well (74,75). Though the role of complement has not been studied in MD, several investigators have cited potential autoimmune/autoinflammatory pathologic contributions to some MD phenotypes (76–78). Existing literature on complement system activity in other pathologic conditions has demonstrated that its dysregulation contributes to autoimmune diseases such as systemic lupus erythematosus (SLE) and myasthenia gravis; ocular pathologies such as age-related macular degeneration, diabetic retinopathy, and uveoretinitis; and neurodegeneration characteristic of AD, Parkinson disease, and amyotrophic lateral sclerosis (70,79–85).

In addition to its immune function, the complement system is involved in neurogenesis and synaptic remodeling, with a basal level of regulatory complement protein activity essential to organ homeostasis (70,79,80). In neurodegenerative conditions, complement component 3 (C3) and complement component 1q (C1q) appear to be overexpressed by astrocytes (supportive cells), leading to microglial-dependent phagocytosis and synapse loss (70,81). Complement inhibition in animal models has led to varied results, with many studies showing decreased inflammation and neuronal loss and others demonstrating greater neuronal toxicity (70,79,81). With this literature in mind, complement proteins could serve as future experimental targets to determine potential impacts on cochlear cell survival and whether the complement system is involved in MD pathology.

Our GO analysis for condition 3 and for MD-enriched proteins in SV and SGN cell types also demonstrated enrichment of proteins involved in glycolysis and glucose catabolism. Studies are lacking and overall inconclusive regarding the role of glucose in MD (86,87), but as potential autoimmune/autoinflammatory and neurodegenerative contributions to MD pathology are explored in the future, the literature regarding glucose metabolism in other autoimmune and neurodegenerative conditions can be informative for future experimentation (71,83,88–94). In SLE, activation of the mammalian target of rapamycin (mTOR) pathway, a central regulator of metabolism and immune cell function, has been linked to enhanced glucose metabolism and pathologic T-cell differentiation; and some data suggest that reductions in pathway activation occur via inhibition of glucose metabolism (83). In AD-affected brains, glucose uptake decreases at a significantly greater rate compared to nonaffected brains, resulting in altered neurotransmission and neurotoxicity (92,94). This is thought to be related to alterations in the neuroprotective functions from astrocyte glycolytic activity. Preferred astrocyte glycolytic pathways lead to generation of antioxidants for ROS neutralization, but disruptions in astrocyte glucose metabolism may lead to reduced levels of antioxidants for neuronal homeostasis (90–92,94). Taken together, existing literature supports an evolving understanding of potential connections between dysregulation of metabolism and the immune system in autoimmune and neurodegenerative conditions. Our GO analysis implicates a potential contribution of altered glucose metabolism, complement activation, and immune dysregulation to the pathology of MD, which all may be areas of further research in attempts to better understand MD pathology.

This study has several limitations. The analyzed perilymph samples were obtained from MD patients undergoing cochlear implantation, suggesting significant HL associated with late-stage disease. Additionally, these samples may not have fully captured the varied phenotypic expression of MD, and it is possible that different pathogenic mechanisms are involved in earlier stage disease and/or alternate disease phenotypes. We also recognize that potential mechanistic pathways suggested by our GO analysis do not have robust support within the existing MD literature. Prior to this study, however, much of the literature examined gene expression changes from peripheral blood samples, while this study looks specifically at perilymph samples. Additionally, the related literature for other neurodegenerative/autoimmune conditions provides a starting context for understanding potential unexplored mechanisms of MD. We acknowledge that differential gene expression via RNA-sequencing does not directly establish functional significance of implicated genes, and that interspecies comparison of human data with transcriptomic neonatal murine datasets certainly may limit the applicability of our findings. Furthermore, proteomic analysis of human perilymph is inherently subject to sampling error, in large part due to exceedingly small sample volumes. Despite this limitation, we attempted to extend the utility of datasets obtained from perilymph, a limited source of data but one nevertheless worthy of further exploration to better understand inner ear physiology. As noted earlier, our methodology has been applied in prior studies and, as the first comparison of its kind, this study attempts to synthesize relevant literature and localize expression of genes to specific cochlear cell types in order to allow for further experimentation and characterization of potential contributions to MD pathogenesis.

In conclusion, we demonstrate that in silico analysis of single-cell transcriptomic datasets localizes expression of a subset of MD-enriched perilymph proteins to a variety of endogenous cochlear cells types including SGN and SV cell types and to some exogeneous cells involved in the immune effector pathway. This suggests a possible pathologic role for these cell types in the late-stage MD, which will need to be validated in future scientific studies.

FUNDING SOURCES

This research was supported (in part) by the Intramural Research Program of the NIH, NIDCD to M.H. (ZIA DC000088), the DFG Cluster of Excellence EXC 2177/1 “Hearing4all” to A.W., and the University of Kansas Medical Center Hearing Research Fund to H.A.S.

CONFLICT OF INTEREST

M.H. holds the position of Associate Editor for Otology and Neurotology Open and has been recused from reviewing or making decisions for the article. The remaining authors disclose no conflicts of interest.

DATA AVAILABILITY STATEMENT

Some of the datasets generated during and/or analyzed during the current study are publicly available. The remaining datasets generated during and/or analyzed during the current study that are not publicly available, are available from the corresponding author on reasonable request.