Noise-Induced Dysregulation of Quaking RNA Binding Proteins Contributes to Auditory Nerve Demyelination and Hearing Loss

Functional declines and pathological alterations in the cochlea in response to noise exposure

We first characterized pathophysiological alterations in cochleae of adult CBA/CaJ mice exposed to octave-band noise (8–16 kHz) at 100, 106, or 112 dB SPL for 2 h. Auditory brainstem responses were measured at several time points after noise exposure to evaluate AN function. As shown in Figure 1a, a 100 dB (SPL) exposure induced wave I threshold shifts up to ∼55 dB immediately after exposure. However, thresholds improved rapidly with nearly complete recovery by 7 d after exposure. For noise exposure at SPLs of 106 and 112 dB, threshold shifts up to ∼65 dB were seen Im (Fig. 1b,c). Although wave I thresholds improved by day 3, permanent threshold shifts (PTSs) of >20 dB were present at 16 kHz and higher-frequency regions for the 106 dB exposure (Fig. 1b), and of >35 dB at 5.6 kHz and higher-frequency regions for the 112 dB exposure (Fig. 1c). A comparison of threshold shifts for the 11.3 kHz stimulus at all three SPL levels of noise exposure is provided in Figure 1d. Because the 106 and 112 dB noise levels induced PTS in most frequency regions, these two noise levels were chosen for use in subsequent experimentation.

The effect of noise exposure at 106 dB SPL on afferent synaptic ribbons was evaluated quantitatively by immunostaining flat preparations of the basilar membrane with anti-C-terminal binding protein (CtBP2; Khimich et al., 2005). CtBP2⁺ presynapse losses of ∼50% were seen for frequency regions >16 kHz (Fig. 1e) at both days 1 and 14 (Fig. 1f). Electron micrographs revealed that the synapse ribbons, which remained at day 1, appeared to have normal ultrastructural features (Fig. 1g,h). Evaluation of hair cells by immunodetection of PCP-4 (Purkinje cell protein 4), a recently identified putative hair cell marker (Ku et al., 2014), showed that 106 dB noise elicited a loss of outer hair cells at frequency regions ≥22.6 kHz at day 14 after exposure (Fig. 1i,j). Losses of inner hair cells were detected only for frequency regions ≥45.2 kHz (Fig. 1i,j). A significant reduction of ABR wave I amplitudes was seen at 5.6 kHz and higher-frequency regions after exposure (Fig. 1k and Fig. 1-1, exact statistical output).

Structural changes in myelin and axo-glial junctions following noise exposure

Our previous studies of mouse and human cochleae demonstrated that myelin abnormalities occur before a significant AN loss, suggesting that glial dysfunction plays a critical role in the early phase of AN degeneration in sensorineural hearing loss (Xing et al., 2012). To investigate this supposition, we examined ultrastructure of myelinating glia in mice exposed to 106 or 112 dB noise. Since this exposure paradigm induces PTS in the frequency range of ≥16 kHz (Fig. 1b,c), sections from 16 to 45.6 kHz regions of the cochlea were selected for TEM examination. Electron micrographs from control mice showed that type I SGNs are ensheathed with multiple layers of compact myelin lamellae proximal to the glia and multiple layers of noncompact myelin lamellae proximal to the neuron (Fig. 2a,b). Myelinating cells enclosing axons often have only compact myelin (Fig. 2b). Myelin paranodal loops were seen to be in tight contact with axons through axo-glial junctions (Fig. 2b). Loss/disruption of myelin was frequently observed in the AN nerve within RC Im and on day 1 (Fig. 2c–e), whereas no gross pathological alterations were seen in the neuronal bodies (Fig. 2). Disruptions were also present in paranodal axo-glial junctions (Fig. 2f). Myelin abnormalities were more pronounced at day 14 and included loss/disruption of compact myelin lamellae, the appearance of degenerative myelin whorls (Fig. 2g,k), and disruption of and, in some cases, a total loss of myelin sheaths (Fig. 2h,i). At day 30, pathological alterations in myelin were still present in the mice exposed to 106 dB noise (Fig. 2l). In randomly selected sections from the middle turn and middle portion of the basal turn of noise-exposed cochleae among all the mice examined, 225 of 288 identifiable glial cells exhibited at least one of the myelin abnormalities described above. These myelin abnormalities were found in 165 of 215 glial cells at day 1 and in 60 of 73 glial cells at day 14. By comparison, only 41 of 143 glial cells seen in control samples exhibited similar myelin abnormalities. Approximately 77% of glial cells observed in ears on day 1 and 29% of glial cells in controls had abnormal myelin levels (Fig. 2n; n = 3 mice/group).

Neuronal cell adhesion molecule (NrCAM) is expressed in the nodal axolemma and plays an important role in maintenance of the axon initial segment and the nodes of Ranvier (Amor et al., 2014). Immunostaining of the AN with anti-NrCAM antibody at day 1 revealed a loss of axon initial segments (heminodal) at the habenula (Hab) opening in regions encoding frequencies of ≥8 kHz (Fig. 2m). A significant delay of ABR wave I latencies was seen in frequency regions ≥11.3 kHz after noise exposure (Fig. 2o and Fig. 2-1, exact statistical output). These results were consistent with the appearance of myelin abnormalities and the disruption of paranodal structures at similar frequencies, perhaps leading to slower conduction velocities.

Involvement of macrophages in demyelination after noise exposure

Macrophages contribute to the process of nerve degeneration by phagocytosing myelin debris and axonal remnants. Macrophage recruitment/migration into the cochlea has been reported to occur following noise-induced cochlear insult (Hirose et al., 2005; Sautter et al., 2006), ototoxic drug-induced SGN degeneration (Lang et al., 2016), and genetic modification-induced hair cell loss (Kaur et al., 2015). To validate the presence of macrophages in the AN, sections from control and noise-exposed ears were immunostained with Iba-1, a key calcium binding protein that is expressed by microglia/macrophages, but not neural cells (Nakamura et al., 2013). Iba-1⁺ macrophages were present in several regions of the AN in noise-exposed cochleae (Fig. 3a–e). Like monocytes/macrophages, neutrophils are also often seen in injured tissues (de Oliveira et al., 2016). Immunostaining for lymphocyte antigen 6 complex locus G6D (Ly6G), a commonly used marker for granulocytes/neutrophils (Daley et al., 2008), identified Ly6G⁺ cells in the control cochlear bone marrow (Fig. 3f) but did not identify any Ly6G⁺ neutrophils in the ANs of either noise-exposed or control mice (Fig. 3g–i). These findings suggest that monocytes/macrophages, but not granulocytes/neutrophils, are involved in AN degeneration after exposure. Further support for this is provided by examination of ultrathin sections from control and noise-exposed mice. Macrophages were often present in regions of myelin degeneration at both days 1 and 14 after exposure (Fig. 3k–p). Ensheathment of abnormal-appearing neurites by macrophage processes (Fig. 3m) supports the involvement of activated macrophages in demyelination following noise exposure. Very few macrophages were seen in the ANs of controls or at day 30; when present, these macrophages were surrounded by normal-looking axons (Fig. 3j,q).

Gene expression analysis identifies noise-induced effects on genes associated with inflammation and myelin

To understand molecular changes following noise exposure, we performed microarray transcriptional profiling on ANs from control adult mice and mice Im, or on days 1, 7, or 14 (n = 3 per group). Differential expression analysis identified 1247 probe sets corresponding to 901 unique genes that were significantly different among the pairwise relationships (Fig. 4 and Fig. 4-1). Notably, the comparison excluded 48 genes detected as differentially expressed based on animal sex in a preliminary analysis of expression data (Fig. 4-2). Examination of expression patterns for the noise-responsive genes showed that most exhibited transient upregulation or downregulation after exposure that peaked in Im or day 1 samples but returned to near-normal levels in day 14 samples (Fig. 4a). Analysis of biological processes enriched among all of the differentially expressed genes detected a core set of related processes associated with wounding and inflammation (Fig. 4b, top network), including the process that had the overall highest enrichment score, immune response (p_adj = 5.9E-08). Also enriched among the differentially expressed genes was a set associated with apoptosis and DNA metabolism (Fig. 4b, middle network) and independent processes that included cell cycle and positive regulation of development (Fig. 4b, bottom).

To address the myelin pathologies observed following noise exposure, we specifically examined expression data for myelin-related genes. The list of myelin genes interrogated was compiled from several sources, including the gene ontology database and datasets produced from recent RNA-sequencing profiling of myelinating cells (Zhang et al., 2014; Thakurela et al., 2016). The compiled myelin list comprised 1849 unique genes that in turn corresponded to 3510 probe sets for the microarray platform used in our experiments. Analysis of the expression of these genes in response to noise found that approximately half were statistically different (p < 0.05) for at least one pairwise relationship in our dataset (1063 of 1849 unique genes = 57%; 1560 of 3510 probe sets = 44%; Fig. 4-3). Clustering analysis of expression patterns for these genes again showed that many underwent transient upregulation or downregulation that peaked in Im or day 1 samples (Fig. 4c).

Quaking and its potential target genes in the AN are affected by noise exposure

Among genes differentially expressed in response to noise, the QKI gene is well characterized as a key regulator for the formation and maintenance of myelin and axo-glial junctions in the CNS (Darbelli et al., 2016). The QKI gene generates three major spliced transcripts encoding QKI-5, QKI-6, and QKI-7 isoforms (Ebersole et al., 1996). Here we examined the expression of Pan-QKI and specific variants in response to noise. Quantitative PCR analysis of isoform expression in the mouse AN 1 d after noise exposure revealed significant change in QKI-7 but not in QKI-5 and QKI-6 (Fig. 5a). These data, coupled with our findings of differential expression of QKI by microarray, demonstrated that QKI isoform expression is altered in response to noise and suggests that targets of QKI may also be affected. We therefore reanalyzed the microarray expression data to examine putative targets of QKI, compiling a list of 1416 mRNAs containing the QRE, a bipartite consensus binding sequence (Galarneau and Richard, 2005), and 28 RNAs detected as differentially spliced in mouse oligodendrocytes (Darbelli et al., 2016). Collectively, the list of putative QKI targets corresponded to 3102 probe sets on the microarray platform used for our study. Query of these probe sets revealed that 1297 (42%) were statistically different (p < 0.05, Student's t test, unpaired) for at least one pairwise relationship in our dataset (Fig. 5-1). Clustering analysis of expression patterns for these putative QKI targets regulated by noise again showed that most underwent transient upregulation or downregulation peaking Im or at day 1 after noise exposure (Fig. 5b).

Given the established relationship between QKI and myelin (Darbelli and Richard, 2016), we determined the overlap between the myelin-related genes affected by noise and the putative QKI targets affected by noise. Of the 1560 differentially expressed myelin-related probe sets and the 1297 differentially expressed QKI target probe sets, 143 were both myelin-related and putative QKI targets (Fig. 5c,d and Fig. 5-2). Unsurprisingly, QKI was one of the genes identified by this analysis; other genes of interest included Cdc42, Irgn1, Ugt8a, Egr2, Mal, Kcna1, NF1, and Nfasc.

QKI and its potential RNA targets in the adult cochlea fail to fully recapitulate their developmental regulatory patterns after noise exposure

Our previous study revealed that acute nerve injury caused transcriptional effects in the AN resembling a developmental transition from the mature to the immature state (Lang et al., 2015). For example, AN injury caused upregulation of Sox2 and other neural progenitor-related genes and increased the proliferation of a subset of glial cells. Specific temporal patterns of QKI expression in developing nerve tissue are associated with the states of glial cell differentiation and myelination. For example, the expression of QKI-6 and QKI-7 peak at approximately P14, coinciding with myelination of brain tissues (Ebersole et al., 1996; Hardy et al., 1996). We next examined whether this select group of genes, which are myelin related, and putative QKI targets, which are responsive to noise injury, also exhibited evidence of developmental regulation. To address this, we queried our previously described microarray dataset that profiled mouse auditory development from P0 to P21 (Lang et al., 2015), which covers a critical period for glial maturation and myelination. Clustering analysis of these genes (Fig. 5d,e) revealed that nearly all showed vivid patterns of upregulation or downregulation during the developmental period. Inspection of the heatmap involving both injury and development profiles identified three prominent clusters, with two (clusters 1 and 2) showing upregulation by noise and downregulation during development and the third (cluster 3) showing the opposing pattern of downregulation by noise and upregulation during development (Fig. 5e). Among these clusters, only cluster 1 showed temporal patterns for development and noise exposure that were fundamentally similar. For clusters 2 and 3, the developmental and injury temporal response patterns were largely opposing. Of the 143 myelin-related, putative QKI target, noise-responsive genes, 98 were present in these three clusters (69%). Opposing temporal relationships were also apparent in the unclustered QKI targets, which contain 31% of myelin-related QKI targets. These results highlight that there is similarity but also discordance in how putative QKI targets are regulated in the developmental, promyelination state, versus the injury response state.

Validation of differential expression was performed for several of the genes identified by this focused analysis of myelin and QKI genes as well as genes identified in the original, unbiased differential expression analysis that had highlighted an effect on immune response. As demonstrated in Figure 5f, fold changes measured by qPCR for day 1 versus control samples showed a strong correlation with fold changes measured by microarray analysis (Fig. 5-3, exact statistical output).

Expression patterns of QKI isoforms in postnatal and adult ANs

To extend our observation of QKI regulation in AN during postnatal development, we analyzed the expression of total QKI (pan) and the three different isoforms, focusing on the following three critical stages of myelination: P3, P7, and P14 (Fig. 6a and Fig. 6-1, exact statistical output). Pan-QKI expression increased gradually through postnatal development, with a significant increase in expression between P3 and P14. All three of these transcripts were detectable throughout postnatal development, with QKI-7 being more abundant than QKI-5 and QKI-6. QKI-6 and QKI-7 were both significantly upregulated during postnatal development, while QKI-5 was significantly downregulated. These patterns match findings for QKI isoform expression in the mouse brain (Ebersole et al., 1996), with QKI-6 and QKI-7 increasing and QKI-5 decreasing in the mouse AN at times that coincide with the onset of myelination.

We further examined auditory QKI expression, detecting total and variant QKI proteins in mouse and human cochlear sections. Analyses in mice were focused on RC (Fig. 6b), which contains satellite glial cells (SGCs) that ensheath soma of SGNs and Schwann cells (SCs) that myelinate fibers. Immunohistochemical analyses were performed on cochlear sections from 2-month-old young adult mice (Fig. 6c,d,f) and P14 mice (Fig. 6e), and both ears from a 38-year-old human donor (Fig. 6g). QKI-5 was expressed in nuclei of both SCs and SGCs (Fig. 6c). QKI-6 and QKI-7 were expressed predominantly in the cytoplasm of glia surrounding SGNs and fibers (Fig. 6d,e). Pan-QKI detection in the mouse confirmed that total QKI protein was distributed in both nucleus and cytoplasm but that it was limited to glial cells (Fig. 6f). Pan-QKI detection in human temporal bones revealed a similar distribution, with QKI present in nuclei and cytoplasm of SCs and SGCs, but not of neurons (Fig. 6g).

Reduction of QKI protein expression in AN glial cells of adult QKI^{FL/FL;PLPCreERT} mice

To investigate the importance of QKI RBPs to glial cell function in the auditory system, we examined the effects of QKI deficiency by using the QKI^{FL/FL;PLPCreERT} mouse model that targets all QKI variants in glial cells. The QKI^{FL/FL;PLPCreERT} mouse model was generated by crossing QKI^FL/FL with an inducible Cre knock-out line, PLP-CreERT, that induces QKI knockout in Plp1-expressing glial cells after being activated by injections of OHT (Doerflinger et al., 2003).

Immunostaining analyses of RC and OSL regions were performed on cochleae of 15-week-old OHT-treated QKI^{FL/FL;PLPCreERT} mice and compared with QKI^FL/FL;- littermate controls (Fig. 7a). Immunostaining for Pan-QKI showed that nuclear QKI was notably reduced in SGC nuclei in QKI-deficient cochleae compared with glia in control cochlea (Fig. 7b). QKI was also substantially reduced in glial cytoplasm of QKI^{FL/FL;PLPCreERT} mice compared with controls, which had a QKI distribution comparable to that seen in CBA/CaJ mice (Fig. 6f). Figure 7c illustrates a substantial loss of QKI in nuclei and cytoplasm of SCs ensheathing peripheral fibers in the OSL. To determine QKI reduction specifically in the glial cells, dual immunostaining was performed for Pan-QKI with Sox10, a marker of glial cell nuclei (Britsch et al., 2001; Breuskin et al., 2010). We chose the prodifferentiation/myelination factor Sox10 because it is expressed in fully differentiated, mature SCs in the adult peripheral nervous system (Kuhlbrodt et al., 1998; Britsch et al., 2001; Finzsch et al., 2010). In control mice, QKI colocalized with Sox10 in nuclei of SGCs and SCs in RC (Fig. 7d). However, only a few QK⁺ cells stained positively with Sox10 antibody in QKI-deficient mice (Fig. 7d). Quantification of Sox10⁺ glia cells expressing QKI showed that there was a profound reduction in Pan-QKI⁺/Sox10⁺/DAPI⁺ cell density in the OSL of QKI^{FL/FL;PLPCreERT} mice compared with controls (Fig. 7e and Fig. 7-1, exact statistical output). Quantification of Pan-QKI⁺/Sox10⁺/DAPI⁺ cells in RC also showed a significantly reduced density in QKI-deficient mice compared with controls (Fig. 7f and Fig. 7-1, exact statistical output). Total glia cell counts were not significantly different between QKI^FL/FL;- and QKI^{FL/FL;PLPCreERT} ANs.

To investigate further the consequences of QKI deficiency in cochlear glial cells, we evaluated the distribution of Sox2, a glial progenitor marker (Graham et al., 2003; Ferri et al., 2004), in control and QKI-deficient mice. Sox2 is a marker of immature, undifferentiated glia, and negatively regulates myelination by inhibiting SC differentiation (Graham et al., 2003; Le et al., 2005) and promotes a progenitor-like glial state associated with neuronal repair after injury (Parrinello et al., 2010). Sox2 expression declines as glia mature (Le et al., 2005), but is present in satellite glial cells in adult nerves (Koike et al., 2015). In our studies, OSL of control mice contained only a few Sox2⁺ cells, and there was no colocalization of Sox2 with QKI in SCs (Fig. 7g). Sox2⁺ cells were more abundant in RC of control mice, and these Sox2⁺ SGCs stained positively for Pan-QKI antibody (Fig. 7i). Interestingly, QKI-deficient mice showed a substantially different profile of Sox2 expression. In the OSL, there was a notable increase in the number of Sox2⁺ SCs (Fig. 7h), although there was still no evident colocalization of Sox2 and residual QKI in the mutant animals. In RC of QKI-deficient mice, Sox2⁺ SGCs appeared to be increased compared with controls (Fig. 7j). These findings suggest that QKI deficiency may lead to dedifferentiation in a subset of cochlear glia.

QKI deficiency results in demyelination and disruption of paranodal structures in the adult AN

To examine further the role of QKI RBPs in AN architecture and function, we assessed the ultrastructure of myelin and flanking paranodes in QKI^{FL/FL;PLPCreERT} mice in OSL, Hab), and RC regions (Fig. 8e). We examined 41 nodal structures associated with 54 glial cells of 3 QKI^{FL/FL;PLPCreERT} mice and 17 nodal structures associated with 34 glial cells of 3 QKI^FL/FL;- control mice. QKI-deficient mice showed evidence of hypomyelination, with a reduced number of lamellae wrapping some type I AN fibers (Fig. 8a,f). Mutants also exhibited aberrant myelin structure and the presence of unmyelinated fibers (Fig. 8f). In addition, QKI-deficient mice exhibited pathologies in the paranodal region flanking the node of Ranvier. In the control mice, terminal myelin loops of each lamella were well organized, closely apposed to each other, and appeared to have proper septate-like junctions connecting the loop heads to the axolemma (Fig. 8b). In the QKI-deficient animals, the paranode had thinner terminal loop heads that were more broadly spaced, with some large gaps present within the structure (Fig. 8g). In the habenular region, where AN fibers form heminodes, demyelinate naturally, and connect to the organ of Corti, the control AN showed thickly myelinated fibers close to the bony perforation, along with unmyelinated heminodes (Fig. 8c). In contrast, in the QKI-deficient sample there were few myelinated fibers, which were in different states of dysmyelination, and there was a large open space clear of heminodes and axons (Fig. 8h). Soma of type I SGNs were densely myelinated and appeared healthy in control mice (Fig. 8d), whereas type I neurons in the QKI-deficient mice showed dysmyelination, with debris-laden spaces separating the cell membrane from the lamellae (Fig. 8i). Additionally, regions of hypomyelination with limited, noncompact lamellae could be seen in close proximity to densely myelinated soma (Fig. 8j). The observation of hypomyelination in the QKI-deficient animals was supported by a measured increase in the G ratio compared with controls (Fig. 8k).

To understand what may be causing the structural aberration in the axo-glial junctions at the paranodes in QKI-deficient animals, we performed coimmunostaining for contactin1 (Cntn1), an axo-glial connector protein (Boyle et al., 2001), and neurofilament 200 (Dillman et al., 2009). Sections from the OSL and RC in control samples showed characteristic paranodal patterns throughout the nerve fibers (Fig. 8l,m, top panels). Conversely, Cntn1 staining and the distinctive patterns were largely absent in the QKI-deficient samples (Fig. 8l,m, bottom panels). As another measure of the effect of QKI deficiency on nodal structures, we performed dual immunostaining for NrCAM and Cntn1. In QKI-deficient animals, paranodes flanking the heminodes were devoid of Cntn1 at the habenula (Fig. 8n, bottom) and were mostly absent on the flanks of axonal nodes of Ranvier in the OSL compared with controls (Fig. 8o). Some Cntn1⁺ paranodal structures were seen in the RC in QKI-deficient cochleae (Fig. 8p).

AN function was tested in QKI-deficient animals by ABR measurements. Comparison of controls with QKI-deficient animals showed that latencies of ABR wave I responses were significantly delayed for 8 and 16 kHz stimuli in QKI-deficient animals (Fig. 8s and Fig. 8-1, exact statistical output). However, ABR thresholds of wave I were not significantly different between control and QKI-deficient groups for 8 and 16 kHz stimuli (Fig. 8q), and neither were the measured amplitudes (Fig. 8r). Together, these findings show that QKI-deficient mice have ABR defects consistent with the dysmyelination observed in these animals.