Effects of cochlear hair cell ablation on spatial learning/memory

Current clinical interest lies in the relationship between hearing loss and cognitive impairment. Previous work demonstrated that noise exposure, a common cause of sensorineural hearing loss (SNHL), leads to cognitive impairments in mice. However, in noise-induced models, it is difficult to distinguish the effects of noise trauma from subsequent SNHL on central processes. Here, we use cochlear hair cell ablation to isolate the effects of SNHL. Cochlear hair cells were conditionally and selectively ablated in mature, transgenic mice where the human diphtheria toxin (DT) receptor was expressed behind the hair-cell specific Pou4f3 promoter. Due to higher Pou4f3 expression in cochlear hair cells than vestibular hair cells, administration of a low dose of DT caused profound SNHL without vestibular dysfunction and had no effect on wild-type (WT) littermates. Spatial learning/memory was assayed using an automated radial 8-arm maze (RAM), where mice were trained to find food rewards over a 14-day period. The number of working memory errors (WME) and reference memory errors (RME) per training day were recorded. All animals were injected with DT during P30–60 and underwent the RAM assay during P90–120. SNHL animals committed more WME and RME than WT animals, demonstrating that isolated SNHL affected cognitive function. Duration of SNHL (60 versus 90 days post DT injection) had no effect on RAM performance. However, younger age of acquired SNHL (DT on P30 versus P60) was associated with fewer WME. This describes the previously undocumented effect of isolated SNHL on cognitive processes that do not directly rely on auditory sensory input.


Vestibular, retinal, and central exclusion. While ablation of cochlear hair cells induced hearing loss,
it was important to spare vestibular hair cells so that the effect of hearing loss on spatial learning and memory would not be confounded by vestibular deficits. As previously demonstrated, the ablation of vestibular hair cells required a much higher dose of DT than cochlear hair cells in DTR animals 17,19 . A small dose of DT (6.25 ng/g) caused rapid cochlear hair cell loss but did not cause vestibulomotor dysfunction phenotypes such as headbobbing, staggering gait, circling or rolling behavior at any point after DT injection.
Long term vestibular function was objectively assessed using VsEPs (Fig. 2a), histology, and running speed. Results were obtained from 10 animals per condition. WT animals that received DT injection had identical VsEP thresholds compared to DTR animals at 90 dpi (Fig. 2b). Utricle bundle counts at 90 dpi revealed that there was a significant difference between WT and DTR (one-tailed t-test: 115.75 [95% CI 112. 18 Fig. 2c). However, the mean utricle hair cell loss of 8.96% was less than the > 60% loss necessary to be physiologically detected 20,21 . Furthermore, when comparing movement metrics between WT and DTR, there was no significant difference in mean movement velocity (one-tailed t-test: 5.766 cm/s [95% CI 5.512, 6.020] vs 5.848 cm/s [95% CI 5.648, 6.048], p = 0.610) and subjective assessment of gait on visual and center-point video tracking analysis revealed no apparent differences between groups.
In the eye, Pou4f3 is expressed in small subset of retinal ganglion cells (RGCs) and transgenic deletion of Pou4f3 caused hearing and vestibular defects but no retinal defects [22][23][24][25] . It has been previously suggested that DT would not be lethal to RGCs in mature DTR animals 26 . Further considering that the dose response allowed for vestibular sparing despite Pou4f3 expression in 100% of vestibular hair cells, we hypothesized that the current dose would spare RGCs. To experimentally confirm retinal exclusion, mature mice (P30) were injected with 6.25 ng/g DT and retinas were harvested at 2 and 10 dpi. RGC counts were obtained in an automated fashion along a 210 μm section of the RGC layer (40 μm original slice thickness; Fig. 2d,e) and the average of 6 evenly spaced sections was computed per eye. Results reported were from 8 eyes from 4 animals per condition. There was no difference in RGC counts between WT and DTR animals at 2 and 10 dpi ( Fig. 2f; one-way ANOVA: F [2,21] = 0.09, p = 0.9133). All animals reacted to visual confrontation and demonstrated seeing behavior throughout the experiment.
While Pou4f3 mRNA expression has been reported centrally in the striatum and olivary nuclei 17 , there are no prior reports of the gene product (Brn3c) expressed centrally in mature animals to our knowledge. To experimentally confirm central exclusion, mature mice (P30) were injected with 6.25 ng/g DT and brains were harvested at 2 and 10 dpi. Striatal cells were counted in an automated fashion in 420 μm 2 sections (40 μm original slice thickness; Fig. 2h,i). DAPI was used to label nuclei of all cell types including neurons and glia, while anti-NeuN labeled perinuclear cytoplasm of neurons specifically (Fig. 2g) and the average of 8 equally spaced, atlas-matched Figure 1. Characterization of cochlear damage. Error bars represent standard deviation, *P < 0.05, **P < 0.001. All scale bars represent 5 μm. (A) ABR thresholds from 0-4 days post injection (dpi) show that profound hearing loss occurs by 4 dpi in mature animals (0 dpi = P30, n = 8 ears from 4 mice per group). (B) ABR thresholds at 90 dpi show that injection of WT animals have no long term effects on hearing while profound hearing loss is sustained in DTR animals (0 dpi = P30, n = 8 ears from 4 mice per group). (C) Auditory thresholds were determined to be lowest intensity at which a characteristic Peak I could be identified. Tracings represent the average of 256 responses. Example is from 8 kHz stimuli. (D) Schematic for where cochlear whole mount sections for apical, middle, and basal turns were obtained. Counts for inner (E) and outer (F) hair cells for each turn show near-total inner and sub-total outer hair cell ablation, respectively, by 4 dpi. (G) Counted hair cells have a DAPI-stained nucleus associated with an anti-Myosin 7a-stained cell body. While some damaged inner hair cells are still counted based on these parameters, they may have a swollen and unhealthy morphology. Cochlear hair cell ablation caused a spatial learning/memory deficit. Spatial learning/memory was assayed using a fully automated 8-arm radial maze and objectively scored using video tracking software (Fig. 3a,b). Food-deprived mice were given 4 daily habituation trials in the radial arm maze (RAM) where mice learned that a single bait exists at the terminal end of 8/8 RAM arms followed immediately by 10 testing days where a single bait was placed at the end of same 4/8 RAM arms each trial (Fig. 3c). The maze was scored in two ways. First, working memory errors (WMEs) were defined as any re-entry into a previously entered arm www.nature.com/scientificreports/ (Fig. 3d). Working memory informed mice which arms had been previously visited in a given trial. WMEs were independent of baiting pattern and could be scored during habituation and testing. Second, reference memory errors (RMEs) were first entries into never-baited arms (Fig. 3e). Reference memory informed mice which arms would be baited. RMEs relied on the 4/8 RAM configuration and had a maximum score of 4 per trial. Due to the nonparametric nature of the data, group performance on single days were compared using Kruskal-Wallis tests and longitudinal performance was compared using the nonparametric analysis of longitudinal data (nparLD) package for R 27 . Presence of a nose-point in a GZ was scored as an "entry". Configuration of visual cues are shown by an overlaid schematic. (C) RAM training timeline consists for 4 days of "habituation" in a fully-baited RAM immediately followed by 10 days of "testing" in a partially-baited maze, where baited arms remained constant throughout the testing period. No training day was assigned "day 0". (D) For working memory, correct entries were first entries into any previously un-entered arm during a trial, while errors were counted each re-entry into a previously visited arm. (E) For reference memory, correct entries were first entries into baited arms, while errors were counted for each first entry into an un-baited arm (maximum of 4 reference memory errors per trial).
For reference memory, DTR animals performed worse than WT animals longitudinally during the testing period (nparLD: F[2.47,25.41] = 11.27, p = 0.0001). By the end of the testing period on training day 10, DTR animals committed more errors than WT animals (Kruskal-Wallis test: χ 2 = 7.985, p = 0.0463). These data demonstrate that lesions isolated to the auditory periphery cause impairment of central processes that do not directly rely on auditory sensory input.

Effects of hearing loss duration.
In order to determine whether the duration of hearing loss had an effect on spatial learning/memory, DTR animals (P60-DT) were divided into two groups based on age at RAM training start (P90 and P120), which represented different durations of hearing loss (30 dpi and 60 dpi, respectively). WT animals were also divided into groups that were age-matched with the DTR groups with respect to age at RAM training start (P90 and P120).
For working memory (Fig. 4a) 44.01] = 0.446, p = 0.705), suggesting that duration of hearing loss had no effect on performance.
Similar associations were observed for reference memory (Fig. 5a) These data suggest that while hearing loss causes a spatial learning/memory deficit, the duration of hearing loss does not affect the severity of the cognitive deficit in age groups considered.  For reference memory (Fig. 5b) for DTR-P60 DT). However, there was no difference in performance observed between DTR groups (nparLD: F[1,28.73] = 0.57, p = 0.455). These data suggest that age of acquired hearing loss affects the degree of central deficit and that younger age is associated with attenuated deficit. Since neural plasticity is greater at younger ages, it is likely that increased plasticity compensates for the loss of auditory sensory input.

Discussion
The results of this study show that isolated sensorineural hearing loss produced via selective cochlear hair cell ablation causes spatial memory impairments. The duration of hearing loss had no impact on the degree of memory impairment in the timelines studied; mice that experienced 1 month of hearing loss demonstrated similar memory deficits to mice that experienced 2 months of hearing loss. However, the age of hearing loss onset affected the degree of memory impairment; mice that lost hearing at P60 had worse memory than mice that lost hearing at P30. These data suggest that the sudden hearing loss compromises cognitive function and this effect may be modulated by the age of hearing loss onset. This finding offers novel insights into the putative pathway by which hearing loss causes cognitive decline in humans.
Rodents rely heavily on auditory information in the form of vocalization for and social integration and establishment of social structures (Brudzynski, 2015). These vocalizations are predominantly ultrasonic (> 20 kHz); while human hearing typically extends from 0.02 to 20 kHz with fine tuning to speech frequencies (0.08 to 2 kHz), www.nature.com/scientificreports/ rodents hear sounds ranging from 0.2 to 90 kHz 28,29 . Calls at 22 kHz are associated with threats while calls at 50 kHz are associated with rewards 30 . Mouse pups produce isolation vocalizations in the 50-200 kHz range 31 and reproductive vocalizations in adult mice are recorded at intensities over 100 dB SPL at 40 kHz 32 . Taken together, hearing is an important sensory modality in mice for interpretation of social behaviors, thus making it plausible that acquired hearing loss could have central consequences in mice.
Prior data in mice has demonstrated that the degree of age-related hearing loss characteristic of the C57BL/6 strain closely correlates to the decrease in spatial memory performance and hippocampal plasticity 1,10,11 . This is consistent with human neuroimaging data that demonstrates that age-related hearing loss is associated with decreased brain volume [33][34][35] . The hippocampus receives direct input from the auditory cortex 36,37 , which facilitates memory formation and consequently guides behavior [38][39][40] , so it is conceivable that global cognitive function may be affected by auditory deprivation. However, hearing loss experienced in aging is cumulative in nature and its global consequences may be confounded by other effects of age. Studying rapid-onset hearing loss would therefore be informative in this regard. Prior studies where rapid-onset hearing loss was induced relied on noise as a damage model and found that noise-induced hearing loss (NIHL) caused impairments to spatial working and reference memory [12][13][14] . However, permanent NIHL requires noise exposure of over 120 dB SPL for several hours. This degree of noise exposure may cause profound stress on central pathways, leading to cognitive and emotional consequences that in turn may affect performance on behavioral assays. Distinguishing the effects of the noise exposure itself from subsequent NIHL is challenging for several reasons. First, the adverse impact of noise on learning and memory has consistently been reported previously [40][41][42][43][44] . Second, impulse noise exposure itself causes long term central effects even in the absence of permanent NIHL 16,45 . Therefore, assessing the effects of peripheral sensory loss requires a model where conditional hearing loss could be induced in adulthood with no or minimal central stress.
Here we used a transgenic mouse model that allowed for conditional and selective cochlear hair cell ablation by insertion of the human DTR gene after the Pou4f3 promoter. This promoter is robustly expressed in inner hair cells with lower expression in outer and vestibular hair cells and minimal expression in the central nervous system 17 . Use of a very low dose of DT caused damage that correlated with the degree of Pou4f3 expression; there was near complete inner hair cell damage that caused profound hearing loss, minor utricle hair cell loss that was not physiologically detectable, and negligible central loss. Therefore, the dominant insult in this model was sensorineural hearing loss due to inner hair cell ablation. Here, we find that sudden sensorineural hearing loss causes a spatial memory and learning deficit.
In the current study, age of hearing loss onset affected the severity of the cognitive impairment. The cohort that experienced hearing loss at P60 had more severe working memory impairment than the P30 cohort. Since working memory and neural plasticity both decrease with age, it is possible this modulates the effect that age of hearing loss has on memory. Mice at P30 may have less developed dependency on hearing and consequently a lower barrier to adapt to hearing loss. Alternatively, the P30 brain may be inherently more plastic than the P60 brain and is therefore better able to adapt to sensory loss. It is likely that a combination of these possibilities is at play. Since selective cochlear hair cell ablation in perinatal mice could not be consistently achieved with the current model, future investigation into the effect of congenital versus acquired hearing loss on cognition would be informative. This effect was not observed for reference memory; however, it is likely that this was influenced by the limited resolution of the reference memory assay.
Finally, the duration of hearing loss did not affect the degree of cognitive impairment up to 2 months after onset. It is possible that in sudden hearing loss, the maximal cognitive effect is rapidly reached and persists for the duration of hearing loss. This is consistent with prior data suggesting that partial hearing recovery occurring 12 months after noise-induced hearing loss is accompanied by partial recovery of spatial working memory but not reference memory 14 . Since hearing loss is permanent in cochlear hair cell ablation, it is possible that the cognitive deficit may be alleviated by hearing restoration. Alternatively, it is possible that the current study period falls within the transitional phase following hearing loss, as the mammalian brain is adaptable to other forms of sensory loss. Specifically, visual loss similarly causes cortical and subcortical reorganization that hinders cognitive performance 46 . However, long-term cognitive recovery following perpetual visual loss occurs [47][48][49] , suggesting that long-term recovery following hearing loss may also occur beyond the observation period.
In conclusion, this work demonstrates that spatial learning and memory impairment follows mature-onset cochlear hair cell ablation in mice, suggesting that sensorineural hearing loss causes a cognitive impairment. It is likely that hearing loss has a similar effect on human cognition and behavior given the strong association observed between hearing loss and cognitive decline. It would be of interest to know whether hearing loss onset before the critical period would have a similar effect on behavior, whether long term recovery occurs, or if this effect could be mitigated by auditory rehabilitation. Additionally, this experimental approach provides the opportunity to investigate the neuroanatomical networks and neurotransmissions involved and influenced by cognitive decline due to hearing loss.

Methods
Animals. All experimental procedures were performed on the B6.Cg-Pou4f3tm1.1(HBEGF)Jsto/RubelJ transgenic mouse line (DTR; The Jackson Laboratory, stock 028673). The generation and characterization of these mice were previously described 17,19 . In brief, this transgenic line was generated on a C56BL/6 J background and expressed the gene for human diphtheria toxin receptor (DTR, heparin-binding epidermal growth factor receptor) under the regulation of the Pou4f3 promoter. The gene product of the Pou4f3 promoter (Brn3c) is ubiquitously expressed during development but expression in mature animals is limited to cochlear hair cells, vestibular hair cells, and a small subset of retinal ganglion cells (RGCs) 22,50 . While there have been reports of Pou4f3 mRNA expressed in the striatum and olivary nuclei 17  www.nature.com/scientificreports/ mature brain to our knowledge. Prior work has shown that ablation is dose sensitive 17 , and that low expression of Pou4f3 in off-target tissues such as RCGs allow for off-target exclusion 26 . Additionally, the lack of central Brn3c expression in mature animals would suggest that the central nervous system is also spared in this model. Systemic treatment of adult, heterozygous DTR mice with a low dose of diphtheria toxin (DT) resulted in specific ablation of cells with high Pou4f3 expression (i.e. cochlear hair cells) while having no effects on wildtype (WT) littermates. Ablation of vestibular hair cells in DTR mice required a much higher dose of DT than cochlear hair cells 17,19 , thus a single intraperitoneal injection of low dose (6.25 ng/g) DT (Sigma-Aldrich, catalog 322326) at P30 or P60 was chosen to cause hearing loss without apparent vestibulomotor dysfunction. The low dose was then experimentally confirmed to spare RGC and striatal cells. Genotypes were established using PCR as described previously 51 . Animals from both sexes were used. All procedures were approved by the Animal Care and Use Committee at Stanford University. All methods were carried out in accordance with relevant guidelines and regulations.
Auditory brainstem responses. Auditory brainstem responses (ABRs) were measured as previously described 52 . In brief, mice were anesthetized with an intraperitoneal injection of 10 mg/kg ketamine and 10 mg/ kg xylazine and placed on a thermostatic heating pad to maintain body temperature between 37.5-38 °C. ABRs were recorded from a needle electrode placed inferior to the tympanic bulla and referenced to an electrode on the cranial vertex, with a ground electrode placed into the ipsilateral hindlimb. Tone pip stimuli (5 ms duration, 0.2 ms gate time) were presented from an open-field speaker in 10 decibel sound pressure level (dB SPL) steps up from 0 to 90 dB SPL in frequencies ranging from 5.7 to 32 kHz. Waveforms for each condition were generated by averaging 256 responses. ABR thresholds for each frequency were defined as the intensity at which the typical peak I could be identified. A lack of response was designated at the highest sound level, 90 dB SPL. Vestibular evoked potentials. Linear vestibular evoked potentials (VsEP) responses were recorded as previously described 53 . Mice were anesthetized (10 mg/kg ketamine and 10 mg/kg xylazine) and had electrodes placed in an identical manner as described for ABRs. The head was clipped to a mechanical shaker, which delivered linear jerk stimuli in the naso-occipital axis ranging from 0.125 to 2.0 g/ms. Broadband noise from an open-field speaker (90 dB SPL) was used to mask responses from the auditory components of cranial nerve VIII. Responses from normal and inverted vestibular stimulus polarities were summed for 256 sweeps per waveform. VsEP thresholds were defined as the intensity at which a typical positive response peak I could be identified.
Tissue preparation. Mice were anesthetized as previously described and were then perfused via the intracardiac route with 25 mL of 4% paraformaldehyde (PFA). Cochleae, eyes, and brains were then immediately dissected. Cochleae were perfused with 0.5 mL of PFA via the intralabyrinthine route through openings made in the round and oval windows, immersed in PFA for 45 min, decalcified in 1 M EDTA for 3 days at 4 °C, and finally dissected into 3 turns 54 . Turns were triple-washed in PBS with 0.5% triton (PBST) then incubated for 1 h in blocking solution (4% bovine serum albumin in PBST), overnight at 4 °C in rabbit anti-myosin 7a (1:500 in PBST; Proteus Bioscience; catalog 25-6790), then for 1.5 h in Alexa Fluor donkey anti-rabbit 546 (1:600; Invitrogen, catalog A10040) and DAPI (1:10,000; Invitrogen, catalog D1306).
Eyes were immersed in PFA for 15 min, then 30% sucrose in PBS overnight at 4 °C, embedded in optimal cutting temperature (OCT) compound (VWR International, catalog 4583), then cryosectioned along the transverse axis in 40 μm-thick sections onto slides. Slides were then stained with hematoxylin and eosin (H&E; Fisher Scientific, catalog 9990001) or DAPI.
Brains were immersed in PFA overnight, then 30% sucrose in PBS for 48 h at 4 °C, embedded in OCT, then cryosectioned along the antero-posterior axis in 40 μm-thick sections into wells of PBS. Free floating sections were then triple-washed in PBST, incubated for 1 h in blocking solution, 2 h in rabbit anti-NeuN (Invitrogen, catalog PA578639), then 1.5 h in Alexa Fluor donkey anti-rabbit 546 and DAPI.
All tissues were imaged as Z-stacks on a Zeiss LSM700 confocal microscope. Images were captured using Zen Black software (Zeiss) and analyzed with Zen Blue software (Zeiss) and ImageJ (NIH) or Imaris (Oxford Instruments).
Cell counting. Cochlear hair cells were counted along 200 μm sections along the contour of the tunnel of Corti. Cells with DAPI-stained nuclei associated with myosin 7a-stained cell bodies were counted. Counts were statistically compared between groups using one-way ANOVAs. Vestibular hair cells were counted in 130 μm × 130 μm sections of the utricle. Cells with Phalloidin-stained bundles were counted in a striolar and an extrastriolar section and averaged for each utricle. Counts were statistically compared using t-tests.
For RGCs, nuclei in the RGC layer were counted along 210 μm sections along the contour of the inner limiting membrane. Counts were performed in an automated fashion using spot detection in Imaris. The automatic threshold at which spots were inserted in the RGC layer was manually adjusted on a representative image, the parameters saved, then batch applied across all retinal sections for objective counts relative to the reference image. For each eye, the RGC count was the average of six evenly spaced sections along the transverse axis. Counts were statistically compared between groups using one-way ANOVAs.
Striatal cells were counted in 420 μm × 420 μm sections. Counts were performed in an automated fashion using spot detection in Imaris. Automatic thresholds at which spots were inserted in the field of view were manually adjusted on a representative image separately for NeuN and DAPI channels, the parameters saved, then batch applied across all striatal sections for objective counts relative to the reference image for each channel. For each hemisphere, the striatal cell count was the average of eight equally spaced, atlas matched (Allen Mouse