Abstract
Several lines of evidence have suggested that steeply sloping audiometric losses are caused by hair cell degeneration, while flat audiometric losses are caused by strial atrophy, but this concept has never been rigorously tested in human specimens. Here, we systematically compare audiograms and cochlear histopathology in 160 human cases from the archival collection of celloidin-embedded temporal bones at the Massachusetts Eye and Ear. The dataset included 106 cases from a prior study of normal-aging ears, and an additional 54 cases selected by combing the database for flat audiograms. Audiogram shapes were classified algorithmically into five groups according to the relation between flatness (i.e., SD of hearing levels across all frequencies) and low-frequency pure-tone average (i.e., mean at 0.25, 0.5, and 1.0 kHz). Outer and inner hair cell losses, neural degeneration, and strial atrophy were all quantified as a function of cochlear location in each case. Results showed that strial atrophy was worse in the apical than the basal half of the cochlea and was worse in females than in males. The degree of strial atrophy was uncorrelated with audiogram flatness. Apical atrophy was correlated with low-frequency thresholds and basal atrophy with high-frequency thresholds, and the former correlation was higher. However, a multivariable regression with all histopathological measures as predictors and audiometric thresholds as the outcome showed that strial atrophy was a significant predictor of threshold shift only in the low-frequency region, and, even there, the contribution of outer hair cell damage was larger.
SIGNIFICANCE STATEMENT Cochlear pathology can only be assessed postmortem; thus, human cochlear histopathology is critical to our understanding of the mechanisms of hearing loss. Dogma holds that relative damage to sensory cells, which transduce mechanical vibration into electrical signals, versus the stria vascularis, the cellular battery that powers transduction, can be inferred by the shape of the audiogram, that is, down-sloping (hair cell damage) versus flat (strial atrophy). Here we quantified hair cell and strial atrophy in 160 human specimens to show that it is the degree of low-frequency hearing loss, rather than the audiogram slope, that predicts strial atrophy. Results are critical to the design of clinical trials for hearing-loss therapeutics, as current drugs target only hair cell, not strial, regeneration.
Introduction
Cellular pathology in the inner ear underlying sensorineural hearing loss cannot be imaged noninvasively, and the relevant tissues cannot be biopsied. Thus, our understanding of the functionally important structural changes causing human hearing loss is built on postmortem histopathological studies and their correlation with premortem audiological data and medical history (e.g., Merchant and Nadol, 2010).
From such human work, and myriad animal studies, we know that the two main causes of threshold elevation in sensorineural hearing loss are (1) damage to the hair cells, which transduce sound-evoked mechanical vibrations into electrical signals that excite auditory-nerve fibers (Dallos and Harris, 1978; Liberman and Dodds, 1984), and (2) atrophy of the stria vascularis, a region of the cochlear duct specialized for ion transport, which creates the large positive resting potential in the endolymphatic spaces that drives the hair cells' sound-evoked transducer currents (Sewell, 1984; Schulte and Schmiedt, 1992).
Longstanding dogma in otopathology holds that strial atrophy can be diagnosed by a “flat audiogram” (i.e., a roughly equal loss of threshold sensitivity across all test frequencies from 250 to 8000 Hz) (Schuknecht et al., 1974; Pauler et al., 1988). However, the studies underlying this dogma were based on relatively few exemplar cases and did not include explicit comparison of strial degeneration patterns in ears with different audiometric shapes. Furthermore, recent studies from our laboratory showed that the degree of hair cell degeneration was vastly underestimated in classic otopathology studies (Wu et al., 2020b), and that, contrary to dogma, a statistically rigorous assessment of a large number (>100) of cases of human age-related hearing loss (ARHL) showed no significant difference in audiogram shape when comparing those individuals with the most versus the least strial degeneration (Wu et al., 2020a).
The ability to infer the nature of inner ear pathology from audiometric data and medical history is growing in importance as preclinical studies demonstrating therapeutic approaches to hair cell regeneration (Mizutari et al., 2013) are moving to clinical trials (McLean et al., 2021), and treatments for monogenic disorders of hearing are moving from bench to bedside (Jiang et al., 2023). It is particularly critical to be able to infer the degree of strial dysfunction, since most of the regenerative and reparative approaches are focused on hair cells, and restoring hair cell function will not restore hearing if there is significant strial degeneration.
These considerations inspired us to revisit the ideas relating strial degeneration with a flat audiogram, expanding our inquiry to include cases other than those with ARHL (i.e., including those with explicit otologic disorders arising from ototoxic drugs, Ménière's disease, genetic mutations, and other more complex causes). Here, we added 54 additional cases to the 106 ARHL cases previously studied, by combing the temporal bone archive at the Massachusetts Eye and Ear explicitly for cases with “flat” audiograms. The resultant quantitative analyses show that, when a wide variety of hearing loss etiologies is considered, it is not the flatness of the audiogram that is predictive of strial atrophy; it is the degree of low-frequency hearing loss.
Materials and Methods
Case materials
This study was based on 160 temporal-bone cases from the archives at the Massachusetts Eye and Ear (Merchant et al., 2008), comprising 92 male and 68 female ears. Each temporal bone, including all structures of the middle and inner ear, was removed at autopsy and fixed by immersion for 1-2 weeks in 10% neutral buffered formalin, decalcified, dehydrated in ascending grades of alcohol, and embedded in 12% celloidin. Complete histologic protocols are described elsewhere (Schuknecht, 1993). Each case consists of a set of every 10th section, of 20 µm thickness, cut in the horizontal plane and stained with H&E. Since the intervening sections are stored in 80% alcohol, they can be retrieved and immunostained for a variety of purposes (O'Malley et al., 2009).
Of the 160 specimens, 106 cases were from a prior study of normal-aging ears (Wu et al., 2020a) (i.e., those without any history of otologic disorder or disease) to which we added 54 additional cases, selected by filtering the audiological records for those with “flat” and “gently sloping” audiograms regardless of otologic history (for further details, see Results). Cases were included in the final dataset only if the interval between the last audiometric test and death was <6 years (mean = 1.6 ± 0.16 years), and only if there was no evidence of significant conductive hearing loss, that is, mean air-bone gap across all test frequencies <15 dB (mean = 3.8 ± 0.62 dB). The mean time between death and tissue harvest was 14.1 h (± 0.61 h). Only those cases with excellent tissue preservation (i.e., where hair cell populations could be unambiguously assessed) were used for this study. All procedures and protocols for the study of archived human tissue were approved by the Institutional Review Board of the Massachusetts Eye and Ear.
Histochemistry and immunohistochemistry
Selected unstained sections were retrieved for histochemical or immunohistochemical analysis. After decelloidinization (O'Malley et al., 2009), these sections were slide-mounted and incubated in different combinations of stains and antibodies.
To infer strial health and to validate the boundaries between strial and spiral ligament cells, we first placed the sections in a blocking buffer (PBS with 5% normal horse serum and 0.3%-1% Triton X-100 for 1 h at room temperature) followed by an overnight incubation at room temperature in the first primary antibody (i.e., anti-smooth muscle actin [SMA]; Abcam #32575, at 1:1000). On day 2, a 1 h incubation in a biotinylated donkey anti-rabbit Fab fragment (Jackson ImmunoResearch Laboratories #711-067-003) at a 1:400 dilution was followed by a 1 h incubation in streptavidin 568 at 1:500 at room temperature, then a 1 h incubation in donkey anti-rabbit Fab Fragment (Jackson ImmunoResearch Laboratories #711-007-003) at 1:100 at room temperature. These steps were followed by overnight incubation in two primary antibodies at room temperature: (1) anti-Na+K+ATPase (Invitrogen, #PA5-52729, at 1:100) and (2) anti-cochlin (EMD Millipore #MABF267, at 1:100). On day 3, we linked the primaries to appropriate secondary antibodies by incubation for 1 h at room temperature (i.e., with anti-rabbit AlexaFluor-405 at 1:200 and anti-rat AlexaFluor-488 at 1:1000). This step was repeated, with addition of Topro at 1:5000 to stain nuclei. Last, the tissue was coverslipped with Vectashield. All antibody solutions also contained 0.3% Triton X-100 to improve penetration, and three PBS rinses were interleaved between each antibody incubation step.
To count auditory nerve peripheral axons, we incubated sections in a fluorescent membrane dye, Cellmask (Invitrogen #C10045), at 1:1000 dilution for 5 min before coverslipping in Vectashield.
Histologic analysis
Strial analysis consisted of measuring its area with a 20× objective (N.A. = 0.95) on a Nikon E600 microscope, using computer-aided anatomy software (Neurolucida). Measurements in each case were made at 14 locations, positioned to be as close to cross-sectional as possible and as equally spaced along the cochlear spiral as possible. A geometric correction was applied to those area measurements which were nonorthogonal. At each locus, the extent to which the cutting angle deviated from orthogonal (α) was estimated from the standard two-dimensional reconstructions of the cochlear spiral, based on identifying the section numbers of the tangential sections through each half turn and representing the connecting segments as semicircles (Merchant and Nadol, 2010). The actual area was then defined as the observed area multiplied by cos(α), then expressed as percent survival by normalizing to the place-appropriate mean data from the normal-hearing ears in this study.
For peripheral axon counts, confocal z stacks of the osseous spiral lamina region were acquired at each of five cochlear locations, using the tangential section through each half turn of the cochlear spiral that contained the junction between the limbus and tectorial membrane. These five locations correspond to cochlear frequency values of ∼0.2, 0.4, 0.8, 2.0, and 8.0 kHz. At each ROI, the top 5 µm of the section was imaged at 0.33 µm z spacing, using a 63× glycerol objective (N.A. = 1.3) on a Leica SP8. The brightest image slice was chosen, and myelinated fibers were counted within a 250 µm mask placed exactly in the middle of the lamina. Raw data were normalized to average counts from the appropriate cochlear region in young normal cases, aged 0-6 years, as described in more detail previously (Wu et al., 2020a).
Hair cells in each case were counted in every 10th section through the cochlear duct using a 100× oil-immersion objective (N.A. = 1.3) on a Nikon E800 microscope. In each section, the fractional survival was estimated, as has been described in a prior publication (Wu et al., 2020b). Data from all three outer hair cell (OHC) rows were combined into an ensemble average. Data from both inner hair cells (IHCs) and OHCs in each case were binned in 5% increments of cochlear length.
For all histologic metrics, cochlear location was converted into frequency according to a map derived by Greenwood (1990), modified to produce an apical-most best frequency of 100 rather than 20 Hz. This modification was introduced because (1) the lowest best frequencies ever recorded from the mammalian auditory nerve are all >100 Hz, including macaque (Shera et al., 2011), cat (Liberman, 1978), gerbil (Schmiedt, 1989), guinea pig (Tsuji and Liberman, 1997), and chinchilla (Henry et al., 2016); and (2) if human auditory-nerve fibers are comparable, those tuned to 100 Hz would respond to 20 Hz tones at sound pressure levels appropriate to the human audiogram.
Statistical analysis
Regression analyses and t tests were conducted in Excel and for correlation matrix GraphPad Prism version 10 was used. The LASSO multivariable regression was conducted in R studio, using the glmnet package in R 3.6.1. Before fitting in the LASSO linear regression, the data were preprocessed as follows: (1) in each case, each of the histopathological measurements (OHCs, IHCs, stria, and auditory nerve fibers) was averaged across low (0.25, 0.5, and 1.0 kHz) and high (2.0, 4.0, and 8.0 kHz) frequencies to produce a mean survival and treated as continuous variables; and (2) continuous variables (i.e., histopathological measurements, age, and hearing level) were normalized (z score). The best model was selected via nested 5 × 10-fold cross-validation.
Results
Defining strial area as a measure of strial atrophy
In H&E-stained sections, the stria vascularis appears as an epithelial strip on the lateral wall of the cochlear duct, nestled between the spiral prominence and Reissner's membrane (Fig. 1A). As its name implies, the stria is a highly vascularized structure (Fig. 1B, arrowheads). It consists of three cell layers: marginal cells with round nuclei on the endolymphatic surface, basal cells with flattened nuclei, at the border with the spiral ligament, and rarer intermediate cells with irregular nuclei located between the other two layers (Fig. 1B). This trilayer epithelium, separated from both the spiral ligament and the scala media by tight junctions (Gow et al., 2004), is specialized to pump K+ ions across it, thereby creating the positive resting potential in the endolymphatic spaces (Salt et al., 1987). The intraepithelial surface areas of marginal and intermediate cells are greatly expanded via an elaborate interdigitating network of membranous folds (Hirose and Liberman, 2003), where the Na+K+ pumps and the Na+K+Cl– cotransporters driving the ion fluxes are localized (Crouch et al., 1997; Lang et al., 2007). In H&E-stained material, these elaborated membranes have a characteristic striated appearance, which distinguishes the marginal and intermediate cell layers from the basal cell layer, which has a more homogeneous appearance: compare images in Figure 1B to Figure 1C, where a blue line has been drawn to enclose the marginal and intermediate layers.
Measuring strial area in celloidin sections stained with H&E. A, Photomicrograph of the cochlear duct from the mid-cochlear (3.7 kHz) region of a 24-year-old male showing the appearance of the stria vascularis and other structures of the lateral wall. B, Higher-magnification view of the stria vascularis from A, showing nuclei from each of the three cell layers, and highlighting several cross sections through blood vessels (open arrowheads). C, The same image as in B, but with the strial area outlined (blue), as it was done in the present study.
In these H&E-stained sections, it was easier to reproducibly define the border between the marginal/intermediate cell layers and the basal cell layer than to define the border between the basal cell layer and the spiral ligament; we chose to measure the area of the marginal/intermediate cell layer in each sampled cochlear location as our metric of strial survival (Fig. 1C).
To further validate this approach as the best estimate of functional strial tissue extractable from these archival temporal-bone sections, we compared the H&E images to adjacent sections immunostained for several key markers, from selected individuals with normal audiometric thresholds and those with sensorineural hearing loss (Fig. 2). Immunomarkers included the Na+K+ATPase (Atp1b2) that powers the ion transport in the strial epithelium and is a necessary (but not sufficient) component of the strial “battery” (Schulte and Schmiedt, 1992; Gratton et al., 1997). We also included SMA, Acta2 because it has been shown to be a marker for strial basal cells (Nakazawa et al., 1996), and cochlin (Coch) because it is a secreted protein that is widely distributed in regions in ready communication with perilymph, including the spiral ligament, and should be excluded from the strial epithelium by the tight junctions that surround it (Robertson et al., 2001). These immunostains confirmed (1) that the region we outline in the H&E sections corresponds to the region where the Na+K+ATPase is expressed (e.g., Fig. 2A–C,E), (2) that when we conclude from the H&E sections that there is no strial tissue remaining, there is also no Na+K+ATPase expressed (Fig. 2D), and (3) that in those sections where both SMA and cochlin antibodies were used, the actin-positive (basal cell) region forms a distinct buffer zone between the cochlin-positive (spiral ligament) and the Na+K+ATPase -positive marginal/intermediate cell layers.
Immunohistochemical validation of the correspondence between the surviving strial regions identified in the H&E-stained archival sections (A1–E1) and the regions still expressing Na+K+ATPase (A2–E2). Each column represents a pair of images. A1–E1, H&E tissue. A2–E2, Immunostained nearby section. A2–D2, Stained with antibodies to cochlin and SMA, and are from the 3.7, 1.1, 0.6, and 0.2 kHz regions, respectively, from a 24-year-old male with a normal audiogram. E2, Stained only with anti-Na+K+ATPase and SMA, from the 3.7 kHz region of a 52-year-old female with a Flat Moderate audiogram. Scale bar in A2 applies to all panels.
The micrographs in Figure 3 also illustrate that, in addition to complete absence, partial strial atrophy can manifest as a localized thinning (Fig. 3D, at arrow) and/or a retraction of the healthy strial tissue from both Reissner's membrane and the spiral prominence (Fig. 3E). In all ears older than a few years, there is marked disorganization of the cell layers, which appear remarkably orderly in the neonate (Fig. 3A). In the oldest ears (i.e., 75-100 years), there is also an increased accumulation of pigment granules in the strial tissues (Fig. 3E).
Examples of strial dysmorphology with increasing age. All images are from a mid-modiolar section through the upper basal turn, corresponding to the 3.7 kHz region. The audiogram shapes associated with the images were unmeasured (A,B), Sensory (C), Normal (D), and Descending (E). D, Arrow points to a region where the strial epithelium is thinned. D, E, Arrowheads point to clusters of pigment granules. Scale bar in E applies to all panels.
Strial survival as a function of audiometric pattern
The idea that strial atrophy is diagnosable via the presence of a “flat,” or “gently sloping,” audiogram originated with Schuknecht and colleagues (Schuknecht, 1964; Schuknecht et al., 1974; Pauler et al., 1988) and was further refined by Dubno and colleagues (Schmiedt et al., 2002; Dubno et al., 2013; Vaden et al., 2022). The latter developed a set of explicit algorithms for the study of ARHL to distinguish “older normals” from those with (1) “sensory” lesions, dominated by hair cell dysfunction, from (2) “metabolic” lesions, dominated by strial dysfunction, and from (3) mixed lesions combining both types of underlying pathology in differing ratios (Dubno et al., 2013; Vaden et al., 2017, 2022).
Many audiograms from the present study failed to fit into prior schemes because the overall level of hearing loss is greater in our collection of temporal-bone donors than in the larger studies of living patients (Dubno et al., 2013; Vaden et al., 2017, 2022). Therefore, we devised an alternate strategy, inspired by these prior studies, but designed to accommodate a wider variety of audiometric shapes and degrees of hearing impairment. As illustrated in Figure 4A, that simple strategy separated audiograms along two axes: the SD of all the thresholds, in dB HL as measured in clinic, and the low-frequency pure tone average (PTA) (i.e., mean threshold at 0.25, 0.5, and 1.0 kHz). Arbitrarily dividing the SD axis at 16, and the PTALOW axis at 25, 50 and 75 dB HL yielded five groups, for which the audiogram shapes were quite distinct (Fig. 4B). Many of the “Metabolic” lesions from the Dubno group would be included in our “Normal” group, and their “Sensory” and “Combination” audiograms would generally fall into our “Sensory” group. The low-frequency thresholds in our other three groups are generally worse than those considered in the Dubno group's studies (Vaden et al., 2022).
The relations between audiometric pattern and the pattern of strial survival as a function of cochlear frequency. A, To separate audiograms by shape, we used an algorithm based on “flatness” (i.e., the SD of hearing levels at all test frequencies) and “severity” (i.e., the PTA hearing level at low frequencies, 0.25, 0.5, and 1.0 kHz). SDs were divided into two groups at 16, and PTAs were divided into three groups at 25, 50 and 75 dB, as indicated by the dashed lines. Five cases were unclassified by this metric (gray points). B, Spaghetti plots of all the audiograms falling into each group, with group names indicated and color coding the same as in A. C, Mean strial survival (±SEM) for each of the five groups. Inset, Significance of the pairwise differences in mean strial survival: *p < 0.05; **p < 0.005; ****p < 0.0001; two-way ANOVA. D, A dot plot of subject ages in each group along with the mean. E, The percentage of each group that was male.
To compare strial survival as a function of audiometric shape, we normalized all the strial areas to mean values in the Normal group. As seen in Figure 4C, strial atrophy, when expressed as percent loss of cross-sectional area, is generally worse in the apical half of the cochlea than in the basal half, and many of the differences between audiometric groups are statistically significant (Fig. 4C, matrix inset). Although the Flat Severe group has the worst strial atrophy, and the Sensory group has the least, the differences between Descending and Flat Moderate groups are not significant despite the marked differences in the flatness of their audiograms (Fig. 4B).
Not surprisingly, the Normal group is significantly younger than all the other groups (Fig. 4D; p ≪ 0.001). The Sensory and Flat Moderate groups are significantly younger than the Flat Severe (p < 0.01 and p < 0.05, respectively). No other intergroup age differences are significant. Interestingly, the Sensory group is overwhelmingly male, and the Flat Moderate group is predominately female (Fig. 4E).
The audiometric patterns must also be influenced by the degree of hair cell loss, and those patterns are summarized in Figure 5. The Flat Severe group, with the worst PTALOW and the worst apical strial atrophy, also shows the most IHC and OHC loss in the apical half of the cochlea. The Descending and Sensory groups, with the worst PTAHIGH, correspondingly show the most ICH and OHC loss in the basal half of the cochlea. The aspect of hair cell loss least well aligned with the threshold patterns is the greater degree of OHC loss in apical regions of the Sensory group despite a PTALOW similar to that in the Normal group. The loss of peripheral axons varies little among the different audiometric groups (Fig. 5C), except that all those with sensorineural hearing loss have more neural degeneration across the cochlea than the Normal group.
The relations between audiogram shape and hair cell or neural survival as a function of cochlear frequency. A, Mean audiometric thresholds (±SEM) for each of the five groups defined in Figure 4. B–D, Mean survival (± SEM) of IHCs, auditory nerve fibers, and OHCs, respectively, for each of the five audiometric groups shown in A. Symbol key in A applies to all panels.
Simple and multivariable regression analyses
A more granular look at the correlations between strial atrophy and audiometric thresholds is shown in Figure 6, where each point in each scatterplot represents a different case. The highest correlation is seen when comparing the mean strial atrophy in the apical half of the cochlea to the mean threshold level in the corresponding frequency regions (Fig. 6A). The correlation between basal strial atrophy and threshold is also significant (Fig. 6B), but there is no significant correlation between audiogram “flatness” and strial atrophy in either the apical or the basal half of the cochlea. The strial atrophy in both apical and basal cochlear regions is also weakly, but significantly, correlated with OHC loss in the same regions (data not shown).
Low-frequency PTA is predictive of strial atrophy (A,B), not audiogram flatness (C,D), and strial atrophy is generally worse in women (E,F). Each scatterplot represents the correlation between mean apical (A,C) or basal (B,D) strial area and the low- or high-frequency PTA (A,B) or audiogram flatness (C,D). Audiometric groups are as defined in Figure 4. Best-fit straight lines (brown) are shown where the correlation is significant (A,B). Each point represents a different case; apical strial area is averaged over the apical nine measurements in each case (see Fig. 4C) to match the low-frequency PTA (0.25, 0.5, and 1.0 kHz), while basal stria is averaged over the remaining five measurement points to match the high-frequency PTA (2.0, 4.0, and 8.0 kHz). E, Comparison of the mean apical and basal strial areas (± SEMs) in males versus females. F, The mean ages of the male and female cohorts. **p < 0.005. *p < 0.05.
As hinted at by the sex ratios of the different audiometric groups (Fig. 4E), there is significantly more strial atrophy in females than males (Fig. 6E), with a larger and more significant difference in the apical half of the cochlea (females ∼20% smaller, p < 0.005) than the base (females ∼10% smaller, p < 0.05). This difference is not confounded by age (Fig. 6F): mean ages for the males and females studied here were almost identical (70.7 ± 1.9 vs 70.7 ± 2.7 years, respectively).
Although there is a significant correlation between age and the degree of strial atrophy among the cases in the present study (Fig. 7A), the correlation is weaker than that seen between age and degeneration of any of the other three cochlear structures analyzed. The correlations between age and auditory nerve fiber degeneration, as well as OHC loss, are particularly strong (Fig. 7B and Fig. 7D, respectively), and the inferred rates of age-related degeneration in both these structures is roughly twice that seen for either the stria or the IHCs.
Strial atrophy is more weakly correlated with age than any of the other histopathological measures in the present study (A–D) but is more strongly correlated with IHC (E) or OHC (F) degeneration. For each of the four measures, the histopathological data were averaged over all cochlear locations corresponding to the audiometric frequency range (i.e., 0.25-8.0 kHz). Audiometric groups are as defined in Figure 4. Best-fit straight lines are shown for each regression.
Since both hair cell loss and strial degeneration can directly affect the sensitivity of the auditory periphery, it is instructive to consider all the histopathological variables as predictors in a multivariable regression to determine which combination and relative weights best predict the audiometric thresholds. For this analysis, histologic data for each metric in each case were binned to correspond to each audiometric frequency; thus, each point in Figure 8A represents one audiometric frequency region from one case. The range of measured thresholds for the Normal, Flat Moderate, and Flat Severe groups is smaller than for the Sensory and Descending groups because the audiograms are flatter (Fig. 8B). To directly compare to our prior study of ARHL ears, we also add age as a predictor. The resultant weighting factors that produced the best fit between predictors and outcome; that is, the LASSO coefficients (Fig. 8C) show that OHC loss was the most important threshold determinant in both apical and basal cochlear regions, whereas strial atrophy significantly contributed to the audiometric threshold only in the apical half of the cochlea. The matrix in Figure 9 shows the collinearity among the variables in this regression analysis. The highest pairwise correlation in each half of the cochlea is between IHC and OHC survival. In contrast, the correlations between strial survival and each of the other histopathological metrics is relatively low.
Assessment of strial atrophy adds significantly to the predictive power of a multivariable LASSO regression model of hearing threshold only at low frequencies. The regression analysis included all cases in the present study and was conducted separately for low and high frequencies, with the histologic data separated at 1.4 kHz, the geometric mean between 1.0 and 2.0 kHz. A, For the analysis, histologic data for each metric in each case were binned to correspond to each audiometric frequency. Each point represents one audiometric frequency region from one case. The regression line (plus 95% CIs) is shown for both frequency regions combined. B, The data from A are replotted, group by group, to better reveal the differences in predictive power of the model. C, The LASSO coefficients for each of the predictor variables are the weighting values that the multivariable analysis found to provide the best prediction of audiometric thresholds.
Correlation matrix showing the degree of collinearity for the variables in the regression model of Figure 8. The division between high- and low-frequency cochlear regions is made at 1.4 kHz. All r values have been converted to absolute values to facilitate comparison between variable pairs with positive versus negative correlations.
Discussion
Prior studies of audiogram shape and strial atrophy
Decades ago, Schuknecht et al. (1974) evaluated the histopathology in ears with ARHL and divided them into four classes: (1) Sensory (12% of cases), with a down-sloping audiogram and basal-turn hair cell loss; (2) Strial, also called Metabolic, (35%), with a flat audiogram, strial atrophy, and minimal hair cell loss; (3) Neural (31%), with exceptional loss of spiral ganglion neurons; and (4) Inner ear conductive (23%) without any visible histopathology. In these studies, hair cells and neurons were crudely counted (see below), and strial condition was evaluated in semiquantitative and unblinded fashion. Although Schuknecht's criteria for sorting cases were never explicitly described, it appears that a case was classified as Strial only if it showed exceptional strial atrophy and a flat audiogram. Indeed, a different otopathology laboratory selecting cases simply based on audiogram shape found no significant association between a flat audiometric pattern and strial atrophy (Nelson and Hinojosa, 2003).
Recently, we revisited these issues by re-evaluating the same cases from the Massachusetts Eye and Ear archive studied by Schuknecht (Wu et al., 2020a). Although we quantified strial survival in essentially the same way, our hair cell analysis used high-power microscopy to through-focus each section and assess fractional survival rather than rating hair cells as either present or absent (Wu et al., 2020b), as in all prior human studies. The binary rating scale used in older studies vastly underestimated hair cell loss: even fractional losses as high as 80% were previously scored as 100% “present.” In our re-evaluation, we used unbiased cluster analysis to separate two groups with relatively good versus relatively poor strial survival, and found no significant difference in mean audiograms (Wu et al., 2020a). Furthermore, a multivariable regression model predicting audiometric thresholds from histopathological metrics found major contributions from hair cell survival but no significant contribution of strial atrophy (Wu et al., 2020a).
The apparent lack of effect of strial pathology on threshold was puzzling, given evidence that strial atrophy decreases the resting endolymphatic potential (EP) in scala media, and therefore elevates thresholds by decreasing current flow through hair cell mechano-electric transduction channels (Sewell, 1984; Gratton et al., 1996, 1997; Schmiedt et al., 2002). One of the best-studied animal models of ARHL, the quiet-aged gerbil, shows minimal hair cell loss outside of the extreme apex, but reduced EP and marked strial atrophy throughout the cochlea (Tarnowski et al., 1991), as assessed either by measures of cross-sectional area or by assays of a key strial ion pump, the Na+K+ATPase (Schulte and Schmiedt, 1992; Gratton et al., 1997). Based on this animal work, Schmiedt et al. (2002) proposed that pure strial dysfunction should result in a gently down-sloping audiogram, with low-frequency thresholds elevated by ∼20 dB and high-frequency thresholds elevated by ∼60 dB. The frequency dependency arises because the cochlear amplifier, powered by the electromotility of OHCs, requires a strong EP (Ruggero and Rich, 1991), but contributes less mechanical amplification at low versus high frequencies (Liberman et al., 2002).
These insights from animal work inspired Dubno and colleagues to devise algorithms to estimate the degree of strial atrophy versus sensory cell loss from audiogram shape (Dubno et al., 2013; Vaden et al., 2017, 2022). In a recent study (Vaden et al., 2022), they combed the same temporal-bone database we used here, extracted 35 audiograms that their algorithm identified as pure “metabolic” (strial), pure “sensory” (hair cell), or “combination” lesions, and then used the qualitative histopathological summaries included in the public-facing database to test their ideas. According to our classification scheme, their “metabolic” cases fall into our Normal (n = 6) or Flat Moderate (n = 3) groups, and all 9 of 9 “sensory” and 16 of 17 “combination” cases fall into our Sensory group. None of their audiograms fall in our Descending or Flat Severe groups, all of which have worse low-frequency thresholds, as do most of the cases in our Flat Moderate group.
The aim of the present study was to further test ideas about these relations by pulling a large number of cases from the Massachusetts Eye and Ear temporal-bone archive, covering a wide range of hearing impairments, and rigorously quantifying the histopathology and its correlations with thresholds and audiogram shape. Our prior ARHL study did not include many cases with flat or gently sloping audiograms, since the classic audiogram of aging is a down-sloping pattern, and most flatter audiogram cases have frank otologic disease. Thus, here, we supplemented our prior study with a large number of flat-audiogram cases that were excluded from the prior study, because their otologic history included diagnoses other than ARHL.
Inferring strial condition from audiometric data
The present study of 160 cases with a wide range of audiometric patterns and hearing-loss etiologies confirms the longstanding idea that flatter “metabolic” audiometric shape is associated with greater strial atrophy than the down-sloping “sensory” pattern (Fig. 4). However, it is not the flatness of the audiogram that is predictive of strial degeneration; it is the degree of low-frequency threshold shift (Fig. 6).
Humans generally lose high-frequency hearing sensitivity as they age (Grant et al., 2022), because of hair cell and neuronal atrophy (Fig. 8). Strial atrophy tends to flatten the audiogram (i.e., add low-frequency hearing loss to the impairment) because it is more common in the apex than the base, and because strial atrophy in the base can also reduce the EP in the apical half of the cochlea (Wu and Hoshino, 1999). There is often OHC loss in the apical half of the cochlea, but even complete loss of OHC amplification will not raise low-frequency thresholds by >20 dB (Schmiedt et al., 2002), and a 20 dB threshold elevation is still considered within the “normal” range. There tends to be minimal loss of IHCs in apical half of the cochlea (Fig. 5), and stereocilia damage on surviving IHCs in the apical half of the cochlea tends to be minimal as well (Wu and Liberman, 2022). Auditory-nerve fiber loss has little effect on threshold until it is nearly complete (Schuknecht and Woellner, 1953), which rarely happens in the apical half of the cochlea (Fig. 5); thus, the main way to produce exceptionally poor low-frequency thresholds is by adding strial atrophy, and the resultant EP reduction, to the mix. In contrast, at high frequencies, OHC loss can raise thresholds by up to 60 dB, there is significant IHC loss, and surviving IHCs in the aging ear have severe stereocilia pathology that can raise thresholds further still (Wu and Liberman, 2022). Furthermore, there is less strial pathology when expressed on a % survival basis (Fig. 4), and the absolute cross-sectional area of the basal stria is more than twice that of the apical stria (Nelson and Hinojosa, 2003); thus, there may be more reserve capacity.
Prior epidemiological studies noted a correlation between low-frequency threshold shift and cardiovascular disease (Friedland et al., 2009). The stria is a highly vascularized structure, and the strial vasculature becomes progressively sparser toward the apex (Glueckert et al., 2018). Thus, it may be vascular disease that leads to the predilection of the apical half of the cochlea toward strial pathology, as seen in the aging gerbil (Thomopoulos et al., 1997) and, therefore, toward low-frequency hearing loss. In the present study, we saw that the flatter audiometric patterns were more commonly seen in women than men, and that the reverse was true of the descending patterns (Fig. 4). These trends are mirrored in a much larger study of audiometric thresholds in >80,000 patients at the Massachusetts Eye and Ear (Grant et al., 2022), which showed that, beyond age 60, mean thresholds for women begin to be significantly worse than those for men at the three lowest audiometric frequencies (0.25, 0.5, and 1.0 kHz).
Ramifications for therapeutics
Therapies designed to reverse sensorineural hearing loss have focused mainly on regenerating hair cells (e.g., Mizutari et al., 2013), with much less work on strial regeneration (Wang et al., 2019). Data from the present study show that there is a correlation between hair cell degeneration and strial atrophy; thus, there is a good chance that any ear with hair cell degeneration may also have strial degeneration, and regeneration of hair cells would not have the desired therapeutic effect if there is not a strong enough EP to drive mechano-electric transduction. Furthermore, the restoration of hair cells in regions of strial atrophy can further reduce the EP throughout the cochlea, according to the “battery” model of strial function (Eckert et al., 2021). However, the strial degeneration is generally less severe in the basal half of the cochlea where the hair cell degeneration is most pronounced (Figs. 4, 5) and where, arguably, local delivery of therapeutics (e.g., at the round window) might be most effective.
One aim of this study was to determine whether strial atrophy can be predicted from premortem measurements. Our analysis shows a modest, but statistically significant, correlation between low-frequency thresholds and strial degeneration. The unavoidable variations in postmortem time and audiogram-death interval likely added noise to the data. Although the noisiness of the correlation precludes diagnosis of strial pathology on a case by case basis, in general, Sensory audiograms (i.e., down-sloping with good low-frequency thresholds) are associated with better strial survival than audiograms with lower slope and worse low-frequency thresholds (Fig. 4), commonly referred to as “metabolic” (Dubno et al., 2013). The Sensory pattern is more common in males (Fig. 4), for whom noise damage tends to be a more important contributor to the threshold shift; correspondingly, noise damage does not tend to cause strial degeneration (McGill and Schuknecht, 1976; Hirose and Liberman, 2003; Wu et al., 2021).
However, in a therapeutic context, it is also important to consider the supporting cells in the organ of Corti, since preclinical success at hair cell regeneration involves forced transdifferentiation of surviving supporting cells into hair cell/supporting cell hybrids (Mizutari et al., 2013). A study of supporting cell survival in human ears (Kaur et al., 2023), including many of the same cases studied here, showed excellent survival in the apical half of the cochlea, but much more variable survival in the base. The best predictor of high-frequency supporting cell survival was the slope of the audiogram: down-sloping “sensory” audiograms had significantly worse supporting cell survival than the flatter “metabolic” type, even for the same level of high-frequency threshold shift. The present study suggests that the trend observed for supporting cell survival may reflect the fact that more of the high-frequency threshold shift in metabolic cases is because of strial atrophy (and less to hair cell loss), all of which points to the difficulties there will be in choosing the right participants for future clinical trials of regenerative therapies.
Footnotes
This work was supported by National Institute on Deafness and other Communicative Disorders Grant P50 DC 015857. The data that support the findings of this study are available from the corresponding author upon reasonable request.
The authors declare no competing financial interests.
- Correspondence should be addressed to Charanjeet Kaur at Charanjeet_Kaur{at}meei.harvard.edu
