B6.B10ScN-Tlr4lps-del/JthJ and C.129(B6)-Tlr2^tm1Kir/J mice (The Jackson Laboratory, Bar Harbor, ME, United States) were used for Tlr2/4 double knockout. Only 8-week-old male mice without a history of ototoxic damage or noise exposure were used for model establishment. The animals were housed under a standard 12-h light/dark cycle and were allowed free access to water and a regular mouse diet. The cochleae of each animal were collected and assigned to different assays to reduce animal usage and maintain sufficient sample sizes. The following genotyping primers were used to examine the Tlr2 knockout allele: 5′-CTTCCTGAATTTGTCCAGTACA-3′; 5′-GGGCCAGCTCATTCCTCCCAC-3′; 5′-ACGAGCAAGATC AACAGGAGA-3′ (mutation allele 334 bp, WT allele 499 bp). The following genotyping primers were used to examine the Tlr4 knockout allele: 5′-GCAAGTTTCTATATGCATTCTC-3′; 5′-CCTCCATTTCCAATAGGTAG-3′; 5′-ATATGCATGATCAA CACCACAG-3′; 5′-TTTCCATTGCTGCCCTATAG-3′ (mutated allele 140 bp, WT allele 390 bp). The procedures were approved by the Ethics Committee of Xinhua Hospital, affiliated with Shanghai Jiao Tong University School of Medicine (Shanghai, China).

Mice of the SGN degeneration group were subcutaneously injected with 1 g/kg kanamycin sulfate (Sigma-Aldrich, E004000), and intraperitoneally injected with 0.4 g/kg furosemide (Tianjin Pharmaceutical Group, H12020527; 10 mg/ml) after 30 min (Hu et al., 2017; Ye et al., 2019). Mice in the control group were simultaneously and intraperitoneally treated with equal doses of saline. The mice were sacrificed 7, 15, and 30 days after drug administration, and the temporal bones were removed and dissected in PBS at 4°C.

The mice were anesthetized with a ketamine (0.1 mg/g)/xylazine (0.01 mg/g) mixture. Body temperature was maintained at 37°C with a warming blanket. An active needle electrode was placed in the midline of the vertex of the skull, a reference electrode at the mastoid areas, and a ground electrode in the low back area. The ABRs were provoked with tone bursts of 4, 8, 11, 16, 22, and 32 kHz, generated by a D/A converter (RP2.1; TDT) and relayed to an attenuator (PA5; TDT), an amplifier (SA1; TDT), and a magnetic speaker (MF1; TDT). Mouse auditory brainstem responses were filtered (100–3,000 Hz), amplified, and averaged using TDT hardware and software. Responses were recorded from 90 dB SPL to 10 dB below the threshold level in 5 dB descending steps. The ABR threshold was defined as the lowest intensity that reliably elicited a detectable response. The average ABR threshold shifts between wild-type and DKO mice were compared using two-way ANOVA with the two factors group × frequency. If significant group effects were identified, the Welch two-sample t-test was used to evaluate each frequency between the two groups.

The excised cochlea was immersed in 4% paraformaldehyde in phosphate-buffered saline solution for 12 h and decalcified in 10% EDTA for 5 days. Specimens were sliced into 4 μm sections, mounted onto silane-coated slides, stained, and observed under a light microscope. The evaluation of cochlear histology included the apical, middle, and basal regions in the Rosenthal canal and the organ of Corti. Every fifth modiolar section for a certain cochlea (four in total) was subjected to histopathological assessment. The same animals were used for the IHC staining. Primary antibodies against the following antigens were used: Caspase 3 (Abcam, ab44976), p65 (Abcam, ab32536), and p38 (Abcam, ab170099). For immunofluorescence staining, frozen sections were blocked with 10% donkey serum (Jackson, 017–000-121) and 0.5% Triton X-100 (Sigma-Aldrich, 9002–93-1) and then incubated with rabbit anti- Caspase 3 (Abcam, ab44976; 1:400) or p65 (Abcam, ab32536; 1:500) overnight at 4°C. After the sections were washed 3 times with PBS, donkey anti-rabbit Alexa Fluor 594 (Jackson, 711–585-152; 1: 500) were used to incubate samples at 37°C in the dark for 30 min. Finally, the sections were stained with DAPI (Beyotime, C1002; 1:2,000) for 5 min. For immunohistochemical staining, paraffin sections were heated at 67°C for 2 h and then deparaffinized in deparaffinized in xylene and rehydrated in alcohol. After the sections were washed 3 times with 0.1 M PBS, they were incubated with 3.0% H2O2 for 30 min at room temperature. The sections were then incubated with the primary p38 (Abcam, ab170099; 1:200) antibody overnight at 4°C. The secondary horseradish peroxidase-labeled goat anti-rabbit IgG antibody (Beyotime, A0208, 1:500) was added for 1 h, and finally, the sections were stained with 3,3N-diaminobenzidine (Beyotime, P0203). Unless otherwise stated, middle cochlear tissues (11–16 kHz region) were used for analyses.

Total RNA from Cochlear modiolus tissues was extracted using the QIAGEN RNA Extraction Kit (Cat. No.74104). Agarose gels (1.5%) were used to detect RNA degradation and contamination. Illumina Hiseq 2,500 and Hiseq 4,000 were used to characterize the transcriptome landscape of SGN following the manufacturer’s recommendations. The following procedures were performed using Hisat2 (version 2.1.0) to align the clean reads with the reference genome of mice (Kim et al., 2015) and StringTie (version 1.3.3b) to evaluate gene expression (Pertea et al., 2015).

Gene set enrichment analysis (GSEA) (Subramanian et al., 2005; Yu et al., 2012) was used to extract biological insight from RNA-seq data. GSEA aims to determine whether members of gene set S tend to occur in the extremes (top or bottom) of a given list from RNA-seq results. The enrichment score (ES) indicates the degree of a set S overrepresentation at the extremes of the gene ranked list L. ES was calculated as follows:

P-value was calculated by estimating the statistical significance of the ES by empirical phenotype-based permutation test procedures.

Terms used for functional annotation include Gene Ontology (Ashburner et al., 2000) and Kyoto Encyclopedia of Genes and Genomes (KEGG) (Kanehisa and Goto, 2000). Functional annotation of GO, including biological processes, cellular components, and molecular functions, and KEGG is a common omnibus to systematically interpret gene functions, facilitating an intensive understanding of genomic information and higher-order functional information.

The construction of PPI networks imported the Search Tool for the Retrieval of Interacting Genes database (STRING) (Szklarczyk et al., 2021), and was analyzed using Cytoscape (version 3.6.1) software (Shannon et al., 2003). STRING (version 11.0) covers 2,000 million interactions of 24.6 million proteins from 5,090 organisms. When fed with differentially expressed genes, STRING was able to identify the most significantly interactive associations, in which a criterion of a combined score > 0.4, was set as the significance level. Cytoscape plugin molecular complex detection (MCODE) can illustrate the core modules in the PPI network. The false degree cut-off, node score cut-off, haircut, false K-core, and maximum depth from seed were set at 2, 0.2, true, 2, and 100, respectively (Wang et al., 2018, 2020; Yan et al., 2019).

Computations and data visualization were performed using R 4.0.2. Distribution of the data was assumed to be normal, but this was not formally tested. No statistical methods were used predetermine sample sizes, but our sample sizes are like those reported in our previous publications (Hu et al., 2017; Ye et al., 2019). Statistical differences in auditory physiology, SGN counts were analyzed using two-way ANOVA, followed by Welch t-test. All statistical analysis results were expressed as the mean ± standard error of the mean (SEM). Statistical significance was defined as P < 0.05.

SGN degeneration was successfully induced by the destruction of cochlear hair cells, which is consistent with the results of previous studies (Hu et al., 2017; Ye et al., 2019). Apoptotic activities in hair cells were significantly upregulated within the following 7 days, whereas no change was seen in SGNs (Figure 1A). The structure of the organ of Corti was almost completely collapsed and destroyed, and nearly no sensory epithelial cells remained in the middle turn on the 30th day after kanamycin and furosemide injection, while almost only a continuous layer of flattened cubic epithelial cells remained on the basilar membrane (Figure 2B). A time-series evaluation of SGN density was conducted on the 7th, 15th, and 30th day after kanamycin and furosemide injection (Figure 1B, auditory physiology data in Supplementary Figure 1 and Supplementary Table 1). On the 30th day after kanamycin and furosemide injection, SGN density has significantly decreased in the apical, middle, and basal region of the cochlea (Welch two sample t-test, apical: t = 8.4, P = 0.00009; middle: t = 7.53, P = 0.0005; basal: t = 10.33, P = 0.00003). Thus, this time point was used to decipher the transcriptome changes during SGN degeneration.

We identified 674 upregulated and 390 downregulated genes in a background of 34147 non-significant genes during SGN degeneration (Supplementary Table 1). Among them, 972 differentially expressed genes (DEGs) had clear autosome positions, and chromosomes 1–19 had 60, 73, 52, 60, 76, 63, 60, 52, 58, 29, 65, 35, 30, 44, 52, 53, 42, 29, and 39 DEGs, of which the most significant ones were Atf3, Prrg4, Gatad2b, Sdr16c5, Ocm, Gm5737, Cep41, Car7, Mmp13, Tspan8, Ccl7, Serpina3n, Bmp6, Clu, Mal2, Gap43, Tnfrsf12a, Lox, and Ptar1 (Figure 3A). Among the differentially expressed genes, there were 76 genes encoding transcription factors (Figure 3B), of which the 10 most significant were Atf3, Stat3, Chgb, Irf6, Scrt1, Hipk2, Zfp871, Rxrg, Lmx1a, and Setbp1.

Gene set enrichment analysis (GSEA) showed that the top30 significantly enriched KEGG pathways included 20 upregulated pathways (Figure 4A), including complement and coagulation cascades, ubiquinone and other terpenoidquinone biosynthesis, linoleic acid metabolism, non-homologous end-joining, basal transcription factors, butanoate metabolism, inflammatory mediator regulation of TRP channels, phenylalanine metabolism, Hippo signaling pathway, endometrial cancer, EGFR tyrosine kinase inhibitor resistance, ErbB signaling pathway, cytokine-cytokine receptor interaction, hepatitis C, arachidonic acid metabolism, neuroactive ligand-receptor interaction, retinol metabolism, JAK-STAT signaling pathway, coronavirus disease (COVID-19), and TNF signaling pathway. The other 10 downregulated KEGG pathways (Figure 4A) were ribosome biogenesis, ribosome, oxidative phosphorylation, collecting duct acid secretion, glycosaminoglycan degradation, glycolysis/gluconeogenesis, regulation of lipolysis in adipocytes, DNA replication, cGMP-PKG signaling pathway, and mTOR signaling pathway. Autophagy and mTOR signaling pathways have been validated to regulate SGN degeneration in our previous work (Ye et al., 2019), which is consistent with the transcriptome results.

GO analysis showed that the dysregulated functions included positive regulation of peptidyl-serine phosphorylation of STAT protein, type I interferon receptor binding, T cell activation involved in immune response, B cell proliferation, natural killer cell activation involved in the immune response, membrane disruption, negative regulation of B cell apoptosis, and embryonic brain development (Figure 4B). The analysis of PPI interactions indicated six key modules (Table 1), and 10 molecules were the most significant, including COMP, COL1A1, STAT3, HAPLN1, HBEGF, CCL2, MMP13, SMAD4, CXCL1, and CTSK. We noticed that these enriched activities of immune responses, TNF signaling pathway, and cytokine-cytokine receptor interaction could be regulated by Tlr molecules (Peek et al., 2020). At the same time, Tlr2 and Tlr4 disrupted the SGN degeneration signaling network (Figure 3C). Based on these results, we hypothesized that Tlr might contribute to the regulation of SGN degeneration.

Given that Tlr may play an important role in SGN degeneration, we established Tlr2 and Tlr4 knockout mice. We detected the hearing threshold of Tlr2/4 DKO mice and wild-type littermates before and 30 days after SGN degeneration. The Tlr DKO group of mice treated with normal saline (NS) and the wild type (WT) mice maintained normal hearing thresholds (Figure 2A), and there was no significant difference in the hearing thresholds of the two groups of mice (P = 0.748, F = 0.105, two-way ANOVA). The Tlr DKO mice of the SGN degeneration group exhibited hearing threshold advantages (P = 0.00446, F = 8.700, two-way ANOVA) on the 30th day after kanamycin and furosemide injection, and the hearing preservation advantage was mainly manifested in 4–16 kHz (4 kHz, t = –3.16, P = 0.012; 8 kHz, t = –3.12, P = 0.0012; 11 kHz, t = –2.51, P = 0.041, 16 kHz, t = –3.77, P = 0.0059; 22 kHz, t = –1.44, P = 0.19, 32 kHz, t = –0.22, P = 0.83; N = 17, Welch Two Sample t-test) (Figure 2A and Supplementary Table 1).

We also observed morphological changes in the Organ of Corti and spiral ganglion neurons. The Tlr2/4 DKO mice and their wild-type littermates treated with normal saline (NS) maintained relatively intact inner and outer hair cells, while the Tlr DKO mice and their wild-type littermates under the SGN degeneration model lost most inner and outer hair cells, and the tunnel of Corti collapsed (Figure 2B).

The Tlr DKO mice and their wild-type littermates treated with NS maintained a relatively normal density, with no significant difference in neuron density (P = 0.675, t = 0.43, N = 24, Welch Two Sample t-test). SGN density significantly decreased in Tlr DKO mice and wild-type mice under the SGN degeneration model (Figure 2C), but the SGN density was significantly higher in the DKO group mice (P < 0.001, t = 11.71, N = 32, Welch Two Sample t-test) (Figure 2D). These results suggest that the hearing preservation advantage of Tlr2/4 DKO mice mainly relies on the protection of spiral ganglion neurons.

The p38-mediated MAPK and p65-mediated NF-κB signaling pathways are canonical signaling pathways of Tlr molecules (Li et al., 2010). We used immunohistochemistry to detect the expression of these genes in SGNs. SGN p38 expression was found in both DKO and WT mice, mainly in the cytoplasm and nuclei. There was no significant difference in p38 expression between the two groups on the 30th day of degeneration induction (Figure 2E, neuron density evaluation in Supplementary Figure 2). The SGNs of DKO and WT mice express p65 (RelA), which is mainly located in the cytoplasm and nuclei. P65 expression decreased 30 days after SGN degeneration, but no significant differences in expression were identified between the two groups (Figure 2F, neuron density evaluation in Supplementary Figure 2).

Transcriptomic analysis was performed on the modiolus of DKO and WT mice on the 30th day after kanamycin and furosemide administration. We identified 1029 upregulated genes and 833 downregulated genes with the criteria of P < 0.01, and | log₂ fold change | > 0.2) (Figure 5A; Supplementary Table 1). Among them, 1,747 had a clear autosomal location and there were 136, 133, 86, 103, 125, 107, 126, 82, 96, 80, 149, 70, 65, 78, 76, 56, 75, 39, 65 on chromosome 1–19, respectively. The most strongly differentially expressed genes on autosomal chromosomes were identified, including Sned1, Ttn, Gbp2, Tlr4, Jchain, Clec7a, Itgax, Nlrc5, Mmp3, Dcn, Gh, Serpina3n, Sfrp4, Dnah12, Glycam1, Parp14, C4b, Iigp1, and Ifit3 (Figure 5B). Figure 5C shows the differentially expressed transcription factors.

In general, Tlr2/4 DKO mice showed downregulated biological activities during SGN degeneration compared to WT mice. The top 30 enriched KEGG pathways included 1 upregulated and 29 downregulated pathways (Figure 6A). Oxidative phosphorylation was downregulated during SGN degeneration, but upregulated in DKO mice compared with WT mice, which is called inverted gene set enrichment. The top 10 KEGG terms with inverted gene set enrichments included arachidonic acid metabolism, complement and coagulation cascades, coronavirus disease—COVID-19, cytokine-cytokine receptor interaction, hepatitis C, inflammatory mediator regulation of TRP channels, JAK-STAT signaling pathway, neuroactive ligand-receptor interaction, TNF signaling pathway, and oxidative phosphorylation (Figure 6B).

Similarly, the top 30 enriched GO terms included one upregulated and 29 downregulated activities (Figure 6C). The top 10 KEGG terms with inverted gene set enrichments included iron ion binding, monooxygenase activity, negative regulation of endopeptidase activity, platelet aggregation, positive regulation of myoblast differentiation, protein polymerization, response to cytokines, response to peptide hormones, serine-type endopeptidase activity, and serine-type endopeptidase inhibitor activity (Figure 6D).

In order to detect whether DKO mice had consistent morphology across the entire signaling network, i.e., whether there is an inverted gene set enrichment relationship with the primary auditory neurons degeneration (PAND)-specific gene set, we analyzed the different genes in the PAND-specific gene set. We found that there was an inverted GSEA relationship between the PAND process and the PAND-specific gene set in DKO mice, and the PAND-associated DEGs were relatively reverse regulated in the progression of PAND in DKO mice (NES = –1.84, P < 0.001).

We further analyzed the attenuation of SGN degeneration in DKO mice from the perspective of the entire signaling network (Supplementary Figure 3). We found that 130 of the 661 genes specifically upregulated during the degeneration of SGN in WT mice were downregulated mice during the degeneration of spiral neurons in DKO mice, and 380 genes were specifically downregulated during the degeneration of SGN in WT mice; 116 genes were downregulated during the degeneration of spiral neurons in DKO mice.

Protein interaction analysis of the proteins encoded by these genes showed that DKO mice had a specific anti-degenerative signal network compared to WT mice. The disease module analysis showed that there were two key modules (Table 2). The score of the first module was 3.667, including seven genes, namely EGF, STAT3, CALB2, LOX, SNAP25, CAV2, and SDC4. The score of the second module was 3.5, which included five genes: MYL1, NCS1, PVALB, TPM4, and TMOD4. In addition, EGF, STAT3, CALB2, SNAP25, SDC4, TPM4, and TMOD4 are key molecules involved in SGN degeneration. CALB2 is a specific marker of the SGN Ia and Ib subtypes in mice (Shrestha et al., 2018), PVALB is a neuron-specific marker (Petitpré et al., 2018), and the interaction of these molecules may contribute to the attenuation of SGN degeneration in DKO mice.