Introduction

Mouse deafness mutants have provided significant insights into the genetic basis of hereditary deafness, ready access to cochlear pathology resulting from the mutant protein, and insights into the regulatory circuits required for normal cochlear development, homeostasis, and function. The extensively studied shaker 2 (sh2) mouse mutant (Deol 1954, 1955) has defects in stereocilia maturation, leading to profound deafness as a result of a missense mutation in the motor domain of myosin XV (Probst et al. 1998). This point mutation causes an amino acid change of cysteine to tyrosine. A second allele, sh2 J, harbors a deletion of a predicted C-terminal FERM domain of the myosin XV tail, which has been suggested to interact with cytoplasmic domains of integral membrane proteins (Anderson et al. 2000). Homozygous sh2 and sh2 J mice have identical cochlear pathology, i.e., abnormally short stereocilia at the apical surface of cochlear and vestibular hair cells, and abnormal actin filament-rich structures (“cytocauds”) that originate at the cuticular plate and project out from the basal end of the hair cell (Probst et al. 1998; Anderson et al. 2000; Beyer et al. 2000). Stereocilia lack the graded, stair-step organization, in spite of the normal spatial organization of stereocilia bundles (Anniko et al. 1980; Belyantseva et al. 2003). Mutations in human MYO15A also cause congenital, profound deafness in several DFNB3 families (Wang et al. 1998; Friedman et al. 2002). Despite the identification of MYO15A as the defective gene, we do not understand at the molecular level how defective myosin XV proteins lead to stereocilia malformation or to subsequent long-term effects on cochlear pathology in affected members of DFNB3 families or in sh2 mice.

Unconventional myosins bind various cellular constituents via their divergent C-terminal “tail” domains and transport these cargos along polymerized actin filaments (Mermall et al. 1998). Myosin XV is normally located at the tips of stereocilia and within the actin-rich cuticular plate that anchors the bases of the stereocilia (Liang et al. 1999; Belyantseva et al. 2003). Myosin XV was recently shown to be essential in transporting whirlin to the tips of stereocilia (Belyantseva et al. 2005; Delprat et al. 2005; Kikkawa et al. 2005). Defective myosin XV proteins in sh2 mice disrupt relevant structural interactions and/or biochemical pathways associated with this transport system and lead to stereocilia malformation.

Myosin XV may provide additional functions in sensory hair cells, such as in the active process of actin polymerization, leading to consequences beyond those immediately apparent changes in cytoskeletal architecture and to functional changes through alterations in gene expression. Small guanosine triphosphate (GTP) binding proteins such as Rho, known to regulate actin polymerization events, are also known to regulate gene transcription (Hall 1998). Activation of serum response factor, a key transcription factor, depends on appropriate control of actin polymerization (Sotiropoulos et al. 1999). We hypothesized that the Myo15 mutation in sh2 mice directly or indirectly affects homeostatic mechanisms, leading to changes in the steady-state transcript levels of genes that operate in compensatory pathways. To identify these pathways, we used Affymetrix GeneChips to profile gene expression in cochlear RNA from deaf sh2/sh2 mice at 3 weeks and 3 months of age, and in normal-hearing heterozygotes (+/sh2) to control for age-related changes.

Materials and methods

Animals

Heterozygous sh2 (+/sh2) females with normal auditory function were mated to sh2/sh2 deaf males to generate +/sh2 and sh2/sh2 offspring. C57BL/6J mice were obtained from Jackson Laboratory (Bar Harbor, ME, USA). Mice of both genotypes were euthanized for cochlear RNA isolation at 3 weeks or 3 months of age. All animals were housed, cared for, and used in accordance with regulations of the University Committee on Use and Care of Animals (UCUCA).

Scanning electron microscopy

Morphological analysis of the inner ear was performed by scanning electron microscopy (SEM) and as described previously (Beyer et al. 2000). Three +/sh2 and three sh2/sh2 littermates were examined at 3 months of age. Briefly, mice were subjected to transcardiac perfusion using 2.5% glutaraldehyde in 0.15 M cacodylate buffer. The apex of the otic capsule was pierced to allow local perfusion through the round and oval windows. After an overnight fixation, cochleae were dissected further to reveal the surface of the organ of Corti and were stained with alternating osmium tetraoxide and thiocarbohydrazide, followed by dehydration in a graded series of ethanol and critical point drying with CO2 in a SamDri 790 (Tousimis, Rockville, MD, USA). Surface morphology of the organ of Corti was viewed and photographed by SEM (Amray 1910 FE).

Cochlear dissection and RNA isolation

For microarray experiments, we pooled cochleae from 7 to 10 mice. Three independent pools were isolated from each of the four experimental groups (+/sh2 at 3 weeks, +/sh2 at 3 months, sh2/sh2 at 3 weeks, and sh2/sh2 at 3 months). Cochlear tissues, including the otic capsule but excluding vestibular tissues, were dissected either in phosphate-buffered saline or RNAlater (Ambion, Austin, TX, USA), and RNA was isolated by using TRIzol (Invitrogen, Carlsbad, CA, USA). RNA subjected to reverse transcriptase-quantitative polymerase chain reaction (RT-qPCR) assays was first treated with RNase-free DNase I (10 units; Roche Applied Science, Indianapolis, IN, USA) at 37°C for 15 min, followed by purification on RNeasy spin columns (Qiagen, Valencia, CA, USA). RNA quality was assessed with Agilent Bioanalyzer 2100 (Agilent Technologies; New Castle, DE, USA); RNA concentration was determined by UV spectrophotometry. RNA yields averaged 0.5 μg per cochlea.

Microarray analysis

Cochlear gene expression was assessed with MG_U74Av2 GeneChip oligonucleotide arrays (Affymetrix, Santa Clara, CA, USA) on total cochlear RNA (3.5–5 μg) from three pools each of four experimental groups. Synthesis of cRNA, hybridization to chips, and washes were performed in the University of Michigan Diabetes Research and Training Center (MDRTC) Microarray Core according to the Affymetrix protocol. GeneChips were processed in two separate batches. The first batch included one RNA sample for each of the four experimental groups; the second batch included two RNA samples for each group. GeneChips were scanned at 3-μm density with GeneArray Scanner (Affymetrix); images were inspected to ensure that all 12 chips had low background and bright hybridization signals.

Data analysis

Data analysis was performed at the University of Michigan Comprehensive Cancer Center Affymetrix and cDNA Microarray Core. We compared two different methods for data analysis: Robust Multiarray Average (RMA) analysis and quantile normalization of the trimmed means of the perfect match–mismatch (PM–MM) signals (differences between PM and MM values) (Giordano et al. 2001). In the former, expression values were calculated by using the RMA method, as implemented in the Bioconductor library of the R statistical language (http://www.bioconductor.org). RMA has been shown to give more reliable estimates of relative expression values when using “spike-in” samples that only differ for a known subset of probe sets. The same is true for a set of samples that were serially diluted before hybridization (Irizarry et al. 2003). We also demonstrate in this study that the RMA method is more reproducible, particularly for genes expressed at lower levels (details in Results and Fig. 2). Therefore, expression values derived from the RMA analysis were used in the following analyses. Expression values, y ij(k), were fitted to an ANOVA model on a gene-by-gene basis.

$$ \gamma _{{ij{\left( k \right)}}} = \alpha _{{i{\left( k \right)}}} + \beta _{{j{\left( k \right)}}} + {\left( {\alpha \beta } \right)}_{{ij{\left( k \right)}}} + \varepsilon _{{ij{\left( k \right)}}} $$

where α represents differences due to genotype (sh2/sh2 vs. +/sh2); β indicates differences due to age (3 months vs. 3 weeks); αβ accounts for differences due to a genotype × age interaction; \(\varepsilon \) is residual effects; and k indicates that each of these factors was nested in a batch effect, since the 12 samples were processed at two separate times. A nested model was used to account for obvious nonbiological differences between the batches. Data were analyzed with SAS Proc GLM, which uses sums of squares. Data were also analyzed using Proc Mixed, which fits a restricted maximum likelihood model. Both analyses yielded similar results.

Expression values from one GeneChip (+/sh2 at 3 months) exhibited different density dynamics compared to the other 11 chips; therefore, this chip was excluded from subsequent analyses. Additional data analyses using dChip and a hierarchical clustering algorithm “Heat Map” (Eisen et al. 1998) indicated significant differences between the two batches of GeneChips: batch 1 (four chips) vs. batch 2 (eight chips), which were processed and hybridized at separate times to GeneChips of different lots. In other words, reproducibility was greatly improved when tissue samples were processed in parallel for RNA preparation and microarray hybridization using GeneChips of the same lot. Because the outlier chip came from the first batch, we also performed analyses with only the second batch of eight chips, i.e., two chips per group. Analyses derived from 11 chips and 8 chips were evaluated for consistent results. For each comparison (age × genotype interaction, genotype, and age main effects), p values were adjusted by using false discovery rate (FDR) (Storey 2002); significance was defined as p < 0.01. Fold changes between +/sh2 and sh2/sh2 or between 3-week- and 3-month-old mice were calculated to determine the direction and magnitude of changes.

Quantitative RT-PCR

Genotype-related changes in gene expression were confirmed by RT-qPCR assays on additional pools of RNA from 3-week-old +/sh2 and sh2/sh2 mice (4 pools for each genotype; 4–6 mice per pool), independent of those used in the GeneChip analysis. Age-related changes that were independent of the Myo15 mutation were confirmed on cochlear RNA from C57BL/6J mice (4 mice per pool, 4 pools per age) at both ages. First strand cDNA was synthesized from total RNA (1 μg) with 200 units of RNase H−-MMLV reverse transcriptase (SuperScript III; Invitrogen) and oligo (dT)12–18 (500 ng) in the presence of RNasin (20 units; Promega, Madison, WI, USA).

Real-time PCR was performed in a Prism 7000 Sequence Detection System (Applied Biosystems, Foster City, CA, USA) with TaqMan probes (Assays-on-Demand or Assays-by-Design; Applied Biosystems). For each gene assayed, triplicate cDNA samples derived from each RNA pool were assayed for quality control. To minimize experimental variability, multiple pools of RNA samples were simultaneously assayed, i.e., in the same PCR plate, for any given gene. As an internal control, we also determined the level of S16 mRNA in each RNA sample. C T, the threshold cycle, is defined as the PCR cycle number at which the fluorescence intensity is appreciably above the background level but is still in the early exponential phase of amplification. The difference in C T between the gene of interest (gene “X”) and S16 for any given RNA sample was defined as ΔC T(X). Individual ΔC T(X) was adjusted to the averaged ΔC T(X) of four replicates (derived from four independent pools of tissues) in the control group, i.e., 3 weeks in comparisons between two ages and +/sh2 in comparisons between two genotypes. These adjusted values (ΔC T(X)adj) were subjected to statistical analysis by unpaired t-test (StatView; SAS Institute, Cary, NC, USA). The difference in ΔC T(X)adj between two groups (e.g., 3 months vs. 3 weeks, sh2/sh2 vs. +/sh2) was defined as ΔΔC T(X)adj, which represented a relative difference in expression of gene X (Livak and Schmittgen 2001). Fold difference was calculated as \( 2^{{\Delta \Delta CT{\left( X \right)}adj}} \), assuming doubling of products every amplification cycle or 100% amplification efficiency.

Results

Inner ear morphology

At 3 months, the organ of Corti of +/sh2 mice had a normal appearance throughout most of the cochlea (Fig. 1A). Toward the basal hook region, we observed in two of the three mice examined occasional outer hair cell loss, while the inner hair cells remained intact (Fig. 1B). In sh2/sh2 mutants, the surface morphology in the apex at 3 months closely resembled that observed at 3 weeks (Karolyi et al. 2003). Most outer and inner hair cells were present and could be identified by their stubby and poorly organized stereocilia on the apical surface (Fig. 1C). On the other hand, in the basal hook region most outer hair cells were no longer visible on the surface view (Fig. 1D). There is a gradual change in organ of Corti surface morphology along the cochlea. For example, in the lower apex and upper base regions, there was some loss of both inner and outer hair cells although most survived (Fig. 1E). Supporting cells appeared normal in general and expanded to fill spaces in areas of missing outer hair cells. Hensen cells had an irregular apical contour (Fig. 1E). Upon closer examination of the apical region, the outer hair cell surface was smaller in diameter and had extensive loss of stereocilia (Fig. 1F). On the inner hair cell surface, extra rows and ectopic location of stereocilia were commonly observed (Fig. 1G). Inner phalangeal cells had an excessive number of microvilli that were densely matted. Pillar cells survived throughout the length of the cochlea.

Fig. 1
figure 1

Surface morphology of +/sh2 and sh2/sh2 cochleae at 3 months of age by SEM. (A, B) +/sh2 mice. In +/sh2 mice, the organ of Corti appeared normal in the apex (A). A few outer hair cells were missing in the basal hook region (B). (C–G) sh2/sh2 mice. Most hair cells were present in the apex (C), but the apical diameter of outer hair cells appeared smaller than normal, and the supporting cells were expanded. In the basal hook region (D), there was considerable hair cell loss with most outer hair cells degenerated and few inner hair cells surviving. In the lower apex of the cochlear duct, Hensen cells had an irregular contour (E). Surviving outer and inner hair cells could be identified by short, disorganized stereocilia (F and G, respectively). The density of microvilli in inner phalangeal cells appeared to have increased (G), compared to normal cochleae (B).

Full size image

Dynamic range of gene expression in the mouse cochlea

We used Affymetrix mouse U74Av2 GeneChips to profile gene expression in mouse cochlear RNA and compared two different methods for data analysis. For genes with low expression values, expression values derived by RMA analysis were more reproducible (Fig. 2A) than data obtained by quantile normalization of the trimmed means of the PM–MM signals (Fig. 2B). Based on the average expression value for the 12 chips analyzed, all probe sets were ordered from the lowest (2.45) to the highest (14.42) value and grouped into four quartiles (Fig. 2C). The total number of probe sets in each quartile was counted to determine transcript frequencies. Table 1 presents the percentage of the 12,488 probe sets with expression values in each quartile. Genes expressed in limited cell types within the cochlea (Table 1) were distributed throughout this entire range. Prph, a marker of type II spiral ganglion neurons, was in the lowest quartile, consistent with type II neurons comprising only 5% of the spiral ganglion. Myo6 and Myo7a, genes primarily expressed in hair cells in the cochlea, and supporting cell-specific genes Hes5, Tecta, and Tectb (α and β tectorins), were in the second quartile. Many genes with broader expression patterns, including Pou3f4 (Brn4) in the otic capsule and Gjb6 (Cx30) in the supporting cells and stria vascularis, were within the third quartile. Only 1.1% of the probe set had high expression values (the top quartile). Gjb2 (Cx26) is widely distributed in the sensory epithelium, as are genes for the housekeeping proteins GAPDH, β-actin, and ribosomal proteins S16 and S18. Surprisingly, Pou4f3 (Brn3.1), encoding a POU-domain transcription factor found exclusively in hair cells, was also highly expressed. Thus, although analysis of microarray data by previous methods based on PM–MM values failed to detect low abundance transcripts, RMA analysis of microarray data reproducibly detects genes with a wide range of expression values, as well as those expressed in restricted cell types in the cochlea.

Fig. 2
figure 2

Expression profiles of cochlear RNA from +/sh2 and sh2/sh2 mice. (A) Scatter plot of expression values derived from perfect match (PM) values only using RMA analysis. Expression profiles of two independent pools of cochlear RNA from the same experimental group, i.e., sh2/sh2 at 3 weeks of age, were compared to each other. Note the high correlation between values, even at low signal intensities. (B) Scatter plot of signal intensity derived from geometric means of differences between PM and mismatch (MM) values. The same data extracted from Affymetrix GeneChips as in Figure 1A were processed by the geometric means of PM–MM values and transformed to log base 2. Note the poor signal-to-noise ratios at low signal intensities. (C) Dynamic range of gene expression values in the mouse cochlea. Averaged expression values, derived from PM signals of all 12,488 probe sets on all 11 mouse U74Av2 GeneChips and processed by RMA, were expressed to the log base 2 (Y-axis) and plotted (X-axis) from the lowest (2.45) at the left to the highest (14.42) at the right.

Full size image
Table 1 Distribution of expression values of all 12,488 probe sets on MG_U74Av2 GeneChip oligonucleotide arrays
Full size table

Differential gene expression in the sh2/sh2 cochlea: genotype effects

To identify differentially expressed genes, we subjected each gene's expression values to two-way ANOVA. Two data sets were analyzed. Set 1 included data from all 11 GeneChips, excluding data from one GeneChip (3-month-old +/sh2) that exhibited different signal dynamics. Set 2 included only data from the eight RNA samples hybridized simultaneously to GeneChips (batch 2). Overall, 327 probe sets in data set 1, and 201 probe sets in data set 2, showed significant differences (p < 0.01; Fig. 3A) due to genotype (sh2/sh2 vs. +/sh2), age (3 weeks vs. 3 months), or an age × genotype interaction. We conservatively focused on 123 probe sets identified in both data sets.

Fig. 3
figure 3

Analysis of differentially expressed genes. (A) Comparison of different data sets. Two collections of differentially expressed probe sets were identified when we used expression values from all 11 chips (left diamond) vs. those from 8 chips (right diamond). There were 123 probe sets (overlap) common to both data sets. These 123 probe sets were analyzed further. (B) Venn diagrams of the number of probe sets with statistically significant differences (p < 0.01) due to age (96), genotype (11), or age × genotype effects (23). Some of the 123 probe sets showed significant differences in expression values between the two age groups or between the two genotypes, whereas others had different trends of age-related changes between the two genotypes (genotype × age interaction).

Full size image

Only 11 genes (9%) showed expression significantly different in sh2/sh2 vs. +/sh2 mice (genotype effects) (Fig. 3B). Nine genes had modest increases (1.1- to 1.7-fold), whereas two genes had very small decreases (1.1- to 1.2-fold) in the sh2/sh2 mice (Table 2). Based on their potential function and relevance in hair cells, we chose four genes for further analysis. These were Tm4sf3, Aim1, Dbi, and Gas6. TM4SF3, a small tetraspan membrane protein, interacts with other tetraspanins as well as a subset of β1 integrins (α3β1, α4β1, α6β1) (Serru et al. 1999). Aim1 (absent in melanoma 1) encodes a nonlens member of the βγ-crystallin superfamily (Ray et al. 1997; Rajini et al. 2003) that may interact with the cytoskeleton, based on AIM1's homology to a member of the gelsolin superfamily of actin-binding proteins (Teichmann et al. 1998). Dbi (Diazepam binding inhibitor) encodes a small, 10-kDa protein that competes for the benzodiazepine binding site on type A GABA receptors in the brain and is thought to down-regulate GABA-mediated inhibitory responses (Costa and Guidotti 1991). Gas6 (Growth arrest-specific gene 6) encodes a ligand for receptor tyrosine kinases Mer, Axl, or Sky to promote proliferation in several cell types (for review, see Melaragno et al. 1999; Yanagita 2004). Furthermore, Gas6 and its receptor Mer play a critical role in maintenance of vision function (Hall et al. 2002; Hall et al. 2003). Although we could not confirm the 20% decrease in Gas6 mRNA in this whole cochlear preparation, we confirmed that Tm4sf3, Aim1, and Dbi had increased expression in sh2/sh2 cochlear RNA. Tm4sf3 mRNA was 1.7-fold higher in sh2/sh2 cochlear RNA by microarray analysis and 2.2-fold higher (p < 0.01) by RT-qPCR. We observed a 1.2-fold increase for both Aim1 and Dbi with microarrays, but greater increases (1.9- and 1.5-fold, respectively) by RT-qPCR (Table 2).

Table 2 Genes differentially expressed between +/sh2 and sh2/sh2 mice
Full size table

In comparison to the two-way ANOVA analysis with combined grouping, we also performed direct comparisons between the two genotypes at 3 weeks and at 3 months. Direct comparisons failed to identify the aforementioned genes, except for Tacstd2 (p < 0.05). This result was not surprising, because combining the groups in two-way ANOVA gives a larger sample size (n) and is thus more powerful than comparing the individual groups with smaller sample numbers.

Age-related changes in genes for collagen and extracellular matrix proteins

A surprisingly large number of genes showed age-related changes in expression. Of the 123 probe sets, 96 (79%) had statistically significant (p < 0.01) differences in expression at 3 weeks vs. 3 months in both genotypes. Fifty-seven probe sets (corresponding to 44 genes) showed at least a 1.5-fold change (Table 3). Many genes with decreased expression at 3 months encoded collagen subunits or enzymes involved in collagen maturation. The mouse U74Av2 Chip contains probes for 28 collagen genes, whose expression values range from the lowest quartile (Col3a1, Col4a3, Col7a1, Col14a) to the highest (Col1a1) (Table 1). Expression of Col13a1 showed only a moderate 1.5- to 1.7-fold decrease at 3 months, whereas Col1a1, Col1a2, Col3a1, Col5a2, Col11a1, and Col11a2 consistently decreased at least 2-fold (Table 3). Also down-regulated were genes for proteins involved in collagen cross-linking and maturation (Lox), in bone mineralization (Akp2, Bglap1) (Gopalakrishnan et al. 2001; Hessle et al. 2002), and in the extracellular matrix (ECM) (Sparc, Npnt). Most genes with higher expression at 3 months encoded cell antigens in either the histocompatibility or the immune response (Table 3). These increases in expression likely reflected changes in bone marrow maturation and blood cell infiltration in the apex of the otic capsule included in the dissected tissue.

Table 3 Genes with differential expression in the cochlea at 3 months vs. 3 weeks
Full size table

We used real-time RT-PCR assays to confirm the age-related changes in a subset of genes, plus Col15a1, Pou2af1, and Ifit1 on mouse cochlear RNA from C57BL/6J mice (Table 3). Of the 13 genes tested, 12 showed differential expression at these two ages at p<0.01; the changes in Pou2af1 were significant at p< 0.05 (Table 3). The magnitude of the changes determined by RT-qPCR was often greater than that determined by the microarray analysis.

Genes with age and genotype effects

In addition to differences in gene expression between sh2/sh2 and +/sh2 (the genotype effect), we were interested in differential responses over time between two genotypes associated with progressive deterioration in sh2/sh2 mice, but not in +/sh2. These differential responses would be reflected in the genotype × age interaction term. Twenty-three probe sets had a genotype × age interaction term in two-way ANOVA (p < 0.01; Table 4), indicating that either the direction or extent of the change in expression differs with age in +/sh2 vs. sh2/sh2 mice. For 20 of those genes, there was no significant difference in expression due to either genotype or age alone. However, three of these genes, i.e., P2rx4, Tcf3, and Sparc, also showed genotype-independent age effects. P2rx4, a P2X purinergic receptor, was previously shown to be expressed in the rat organ of Corti and in both vestibular and cochlear spiral ganglion neurons (Xiang et al. 1999). In the microarray analysis, we observed a 55% increase in P2rx4 expression with age in +/sh2 heterozygotes, but only a 24% increase in sh2/sh2 mice. Tcf3 encodes a transcription factor implicated in Wnt/β-catenin signaling in early embryogenesis (Korinek et al. 1998). In this study, Tcf3 showed a small but consistent decrease (14%) with age in sh2/sh2 mice, but no age-related change in +/sh2 mice. Sparc, one of the most abundant transcripts in both human fetal cochlear (Skvorak et al. 1999; Resendes et al. 2002) and mouse organ of Corti cDNA libraries (Beisel et al. 2004; Pompeia et al. 2004), encodes a Ca2+-binding glycoprotein whose high-affinity binding to hydroxyapatite contributes to bone ossification (Bolander et al. 1988). Expression of Sparc decreased with age in +/sh2, sh2/sh2, and C57BL/6J mice (Table 3). In addition, its expression, determined byRT-qPCR, was 53% higher in sh2/sh2 than +/sh2 mice at 3 weeks (p < 0.05), consistent with the age × genotype effect detected in the microarray data. These genes with age × genotype effects might encode proteins involved in cochlear pathology in sh2/sh2 mice.

Table 4 Genes with a genotype x age interaction, indicating different trends of age-related changes in +/sh2 vs. sh2/sh2 mice
Full size table

Discussion

Gene expression profiling

Analysis of mammalian cDNA libraries derived from human fetal cochlea (Robertson et al. 1994) and from rodent cochlea or organ of Corti (Ryan et al. 1993; Harter et al. 1999; Beisel et al. 2004; Pompeia et al. 2004) has identified likely cochlea-specific genes and proteins important for auditory function. Gene expression profiling with cDNA microarrays further defined the overall pattern of gene expression in the mature cochlea (Cho et al. 2002) and changes in gene expression during cochlear development (Chen and Corey 2002a, b; Hertzano et al. 2004). These approaches contributed substantially to our understanding of pathways critical for development and maintenance of the cochlea and to the identification of genes associated with hearing loss. We now extend these studies to expand our knowledge on molecular events and cellular pathways leading to cochlear pathology.

Gene expression profiling of complex tissues such as the mammalian cochlea presents several challenges, primarily detecting small changes when genes are expressed in only a subset of cells, such as hair cells. For instance, one mouse cochlea is estimated to contain about 3000 hair cells, less than 5% of the entire cell population (Chen and Corey 2002a, b). We demonstrated here, using Affymetrix GeneChips in combination with RMA analysis, that it is possible to reliably detect genes expressed at a wide range of levels, including those expressed in only a limited subset of cell types. We routinely confirmed even small changes in gene expression, such as those observed in Col15a1 and Pou2af1, by RT-qPCR.

Genes with genotype effects

Genes with differential expression in the cochleae of sh2/sh2 mice might encode proteins that interact directly or indirectly with myosin XV and/or its known interacting proteins. Myosin XV is required to transport whirlin to the tips of stereocilia, probably along actin filaments. Are there other cargos transported by myosin XV? Delprat et al. (2005) proposed that whirlin interacts with a transmembrane protein, NGL-1, isolated from a P2–P6 mouse inner ear cDNA library. In light of similarities between the unconventional myosins XV and VIIa, the PDZ domain proteins whirlin and harmonin, and the transmembrane proteins NGL-1 and cadherin 23, these proteins may form two similar complexes required for stereocilia morphogenesis. These and other interacting proteins are potential candidates that might be affected in the sh2/sh2 mutants. We found that Whrn expression was not affected at the transcriptional level by the Myo15 mutation in sh2/sh2 mice at 3 weeks of age (data not shown).

Genes that exhibited differential expression in sh2/sh2 mice might correspond to proteins involved in cochlear pathology, in altered homeostatic mechanisms, or in compensatory pathways. For example, induction of Dbi, which is believed to down-regulate GABA-mediated inhibitory responses, might represent a mechanism to increase neuronal activity in either hair cells or spiral ganglion neurons to compensate for decreased signals from nonfunctional hair cells.

One of the most interesting differentially expressed genes encodes TM4SF3, a member of the superfamily of membrane proteins containing four transmembrane domains (Maecker et al. 1997). Tetraspanins are known to interact with similar membrane proteins and with integrins at cellular junctions and to assemble small proteins into complex, extensive networks involved in cell–cell signaling. In cancer cell lines, TM4SF3 forms primary complexes with CD9P1 and two other tetraspanins, CD9 and CD81. The primary complexes then interact with yet another tetraspanin CD151, which probably mediates the interaction of the core complex with α3β1 integrin and signaling molecules, such as PI4 kinase and EpCAM (Claas et al. 2005). Mutations in CD151 cause syndromic sensorineural deafness (Karamatic Crew et al. 2004). These and other tetraspanins may be important in maintaining proper cell–cell interactions and thus crucial for auditory function.

Genes with age-related changes

Genes with age-related changes in mice of both genotypes likely reflect aspects of normal cochlear maturation. The age-related decreases in genes encoding collagens and other ECM proteins are consistent with active remodeling of the ECM in the sensory epithelium or other cochlear structures. We observed decreased expression of genes for collagens types I, III, V, XI, XIII, and XV at 3 months in sh2/sh2 mutants, as well as in normal-hearing C57BL/6J mice, and BALB/C mice (T.W. Gong, unpublished data), indicating that these decreases are independent of mouse substrain or genotype. An earlier study detected decreased expression of collagen genes in mouse cochlea between postnatal day 2 (P2) and P32 (Chen and Corey 2002a, b). Our findings indicate that down regulation of collagen genes continues beyond 3 weeks of age. Decreased expression of collagen type V may reflect changes occurring either in the tectorial membrane, the otic capsule, or both. In addition, decreases in expression of type I and type XI collagens may represent, at least in part, changes specifically in the sensory epithelium, since these genes are also highly represented in a cDNA library derived from the organ of Corti and adjacent supporting cells (Pompeia et al. 2004). Sparc expression is high during embryogenesis and later is primarily restricted to tissues that undergo renewal, remodeling, or repair (for review, see Sage et al. 1989; Schellings et al. 2004). In late embryonic to early postnatal stages of cochlear development, Sparc is expressed in the ossifying otic capsule, in stria vascularis, spiral limbus, and spiral ganglion neurons (Mothe and Brown 2001). SPARC is also implicated in cell growth and differentiation via cell–ECM interactions (Yan and Sage 1999).

Two genes in the Wnt/β-catenin signaling pathway, i.e., Wisp1 and Tcf3, were also down-regulated with age. β-Catenin and TCF3/LEF DNA-binding proteins form bipartitate complexes that are translocated to the nucleus to transactivate downstream genes, such as WISP1 (Xu et al. 2000). In the absence of stabilized β-catenin, TCF3 can act as a repressor and plays a critical role in axis formation in the early embryo (Korinek et al. 1998; Merrill et al. 2004). WISP1 plays an important role as an osteoblastic regulator, promoting ostoblast differentiation and chondrocyte proliferation during embryonic skeletal development and repair of fractured bones (French et al. 2004). The lower expression of Wisp1 in mature cochlea may reflect changes in the otic capsule as well as a potential role as a connective tissue growth factor (see review in Brigstock 2003).

Nephronectin, a component of the basement membrane of several embryonic tissues (Brandenberger et al. 2001), interacts with integrin α8β1 in vitro and in embryonic kidney in vivo (Brandenberger et al. 2001; Morimura et al. 2001). Mice carrying targeted disruption of Itga8 for integrin α8β1 have stereocilia defects, suggesting that integrin α8β1 plays a role in regulating hair-cell differentiation and stereocilia maturation (Littlewood Evans and Muller 2000). Early in development (E14.5), nephronectin is found in the basal lamina of the otocyst, but later mainly on the apical epithelial surface outside the developing sensory region (Brandenberger et al. 2001). Integrin α8β1 is found on the apical surface of auditory and vestibular hair cells from E16 to P0 (Littlewood Evans and Muller 2000). Interactions between nephronectin and integrin α8β1 in the sensory epithelium at postnatal stages, particularly at the onset of hearing at 2–3 weeks of age, should be investigated to define the role of these proteins in stereocilia maintenance.

In conclusion, we have demonstrated in this study that Affymetrix microarray analysis, in combination with RMA, is a feasible tool. Previous studies on global gene profiling successfully illustrated how such an approach enables identification of genes that are crucial during ear development and those associated with hereditary deafness (Chen and Corey 2002a, b; Hertzano et al. 2004; Sage et al. 2005). In this study, we extended gene profiling analysis to later stages and showed continuous changes in gene expression beyond 3 weeks of age. These age-related changes in gene expression are likely to reflect the continuing functional maturation of the cochlea, previously observed to reach full maturity at 4 months. At this age, mice also become more resistant to noise-induced hearing loss (Henry 1983). In addition, our analysis of sh2 mice has uncovered pathways and potential interacting genes that are affected by defective myosin XV. This study marks an essential initial step in further characterization of these molecules and pathways, not only in sh2 cochlear pathology but also for proper development and maintenance of auditory function.