Introduction

The majority of genes related to hearing loss can be classified into three major groups. In general, these are based on their functional roles, which are (1) stereocilia-based mechanoelectrical transduction, (2) K+ recirculation and conductances, and (3) the compositional integrity and function of basement membranes. These genes are mainly expressed in the sensory epithelium, their surrounding supporting cells, or the cellular components of the cochlear lateral wall (composed primarily of the stria vascularis and spiral ligament). The set of genes that are involved in stereociliary development and homeostasis is suggested to comprise a network of interactive proteins, which include myosin VIIa, harmonin, cadherin 23, protocadherin 15, whirlin, and Sans (Boeda et al. 2002; Ahmed et al. 2003; Kikkawa et al. 2003; Mburu et al. 2003; Weil et al. 2003; Siemens et al. 2004). Many of the deafness genes initially appeared to be uniquely expressed in the inner ear, but subsequent data showed that their expression is not restricted to the inner ear. These genes are nonetheless pertinent for inner ear function (Beisel et al. 2004). Hearing loss genes are grouped into syndromic and nonsyndromic forms based on the clinical manifestations of gene mutations. It is estimated that over 50% of the several hundred deafness loci are still unknown or uncharacterized. In the mouse, there exist at least 175 spontaneous or genetically manipulated hearing-impaired mutant lines carrying defective genes, with more being identified from the large-scale ENU-mutagenesis screen (Nolan et al. 2000; Kiernan et al. 2002; Rhodes et al. 2003). Studies of these genes and their effects require techniques spanning from behavior to the analysis of unique cell types of the ear and their structure–function aspects. An essential strategy for such an analysis requires the examination of all genes expressed in the ear that are potentially affected and is exemplified by microarray technologies.

DNA microarray analysis does provide a means to study genes at a whole-genome level and can be used to (1) estimate expression levels, (2) begin characterizing previously unknown genes, (3) analyze differential gene expression among tissues, cells types, or specific treatments, (4) analyze mutational variations of a gene, and (5) identify gene interactions through clustering. Although limited in number, microarray studies of the inner ear have ranged from being more general discussions of utilization of these techniques and approaches for the investigation of the inner ear (Lomax et al. 2000; Chen and Corey 2002a,b) to more specific applications involving gene discovery, tissue comparison, and clustering analyses for gene network identification (Chen and Corey 2002a,b; Cho et al. 2002; Abe et al. 2003; Hawkins et al. 2003; Hertzano et al. 2004; Iijima et al. 2004). These studies demonstrate the general use and benefit of microarray analysis as an analytical tool for such analyses.

A common problem of all these studies is the lack of cDNAs or oligonucleotides exclusively representing inner ear transcripts. This impacts on the identification of genes involved in the development and function of the inner ear in both overabundance of uninformative genes as well as the absence of relevant genes. Another major difficulty in microarray assessment is the necessity of obtaining sufficient quantities of starting tissue for the generation of microarray probes. This requirement has mandated the use of whole inner ear/cochlear preparations, which can effectively “dilute out” the identification and quantification of pertinent transcripts. Recent technical advances in amplification of transcripts (Baugh et al. 2001; Badiee et al. 2003) and probe signal (Capaldi et al. 2000) have drastically reduced the required amount of starting material.

Recently, various cDNA libraries specific to the ear were constructed and analyzed (Beisel et al. 2004; Pompeia et al. 2004) as well as the mouse cDNA encyclopedia (Okazaki et al. 2002). We constructed a cDNA microarray chip derived from cDNA libraries produced from the inner ear. Herein, we report the validation and utilization of this “inner-ear-pertinent” cDNA microarray chip to identify the relative expression levels of known deafness genes and potential inner-ear-pertinent transcripts. Through the use of dissected fractions from the mouse cochlea representing the organ of Corti (OC), the lateral wall containing the stria vascularis (LW), and the spiral ganglion neurons (SG), differentially expressed genes were identified and their relationship to deafness loci was assessed.

Materials and methods

DNA preparation for printing

Bacteria cultures in 96-well plates were allowed to thaw at room temperature for approximately 30 min. Five microliters of each culture was added to a 96-well thin-wall PCR plate containing TE buffer (10 mM Tris, 1 mM EDTA). The plate was then heated to 95°C for 10 min, and an aliquot, 5 l, of each supernatant was transferred to a fresh 96-well PCR plate and used to amplify the cDNA inserts.

Clone inserts were amplified in a Peltier thermal cycler (MJ Research, Inc., Waltham, MA) using a modified M13 primer set as described (Hegde et al. 2000). Amplification was performed in 26 cycles with an annealing temperature of 55°C. The extension time was 6 min for the RIKEN inserts and 3 min for the NIH inserts. Following amplification, 5 l of each sample was run out on a 1% agarose gel, which was used to determine the percentage of inserts that produced a product and the number of products with multiple PCR bands. The remainder of each sample was precipitated and then stored at 4°C until printing.

Slide preparation and printing

Dry PCR products were resuspended in sterile water (Sigma-Aldrich, Corp., St. Louis, MO), and 5 l was transferred using a Biomek FX robot (Beckman Coulter, Inc., Fullerton, CA) into 384-well plates containing 5 l DMSO (Sigma). The samples were printed in triplicate on Corning UltraGAPS slides with a Cartesian ProSys printer (Genomic Solutions, Ann Arbor, MI) and TeleChem Stealth pins (TeleChem International, Inc., Sunnyvale, CA). The relative humidity during the print was controlled at 40%. A total of 21,888 spots were printed (7296 unique spots) with a center to center distance of 200 m. The printed arrays were then UV cross-linked at 250 mJ using a Stratalinker (Strategene) and stored in a desiccator at room temperature until used. An expiration date of 6 months was used for all microarray chips.

Experimental design

Three sets of comparisons were made: OC vs. LW, OC vs. SG, and LW vs. SG. Table 1 depicts each comparison and the usage of the tissue samples from four individual animals, designated as mice A–D. The experiments were designed to provide both biological and technical replications and to be compatible with various methods of data analysis (Churchill 2002).

Table 1 Experimental microarray design: sample comparisons
Full size table

RNA extraction

Tympanic bullae from four normal hearing “outbred” CF-1 male mice (4–5 weeks) (Harlan, Indianapolis, IA) were dissected and immediately placed in cold Minimum Essential Medium (Invitrogen Corp., Carlsbad, CA) for fine dissection. Following removal of the temporal bone, the cochleae were dissected at 4°C to obtain large portions of the lateral wall (composed mostly of the stria vascularis and spiral ligament) and spiral ganglia and the whole OC, as represented in Figure 1. The samples were placed in tubes containing RNA Later (Ambion, Inc., Austin, TX) and stored at −20°C. Note that tissue samples from different mice and from different cochlear regions were not combined. Rather, each sample was processed separately in all of the succeeding steps (RNA isolation, amplification, and probe generation).

Fig. 1
figure 1

Dissection of the mouse cochlea. A histological representation of the dissected fractions of the mouse inner ear is depicted. Three functionally distinct regions of the cochlea were chosen for isolation and represent the lateral wall (LW) comprising the stria vascularis, spiral ligament and spiral prominence, the organ of Corti (OC), and the spiral ganglion (SG) neurons. All samples are comprised of tissue from all three turns including the basal hook region.

Full size image

For RNA isolation, tissue samples were transferred to lysis buffer, disrupted, and then homogenized using a QIAshredder column (QIAGEN Sciences, Inc., Germantown, MD). Total RNA was isolated using the Qiagen RNeasy Mini kit following the manufacturer’s standard protocol. DNase treatment was performed on the column as described in the manufacturer’s protocol. RNA integrity was verified using the Agilent 2100 Bioanalyzer with an RNA Pico Chip (Agilent Technologies, Palo Alto, CA) (see Figs. 2A and B). RNA concentration was determined with a UV spectrophotometer and also estimated with the Bioanalyzer.

Fig. 2
figure 2

Assessment of total RNA and amplified RNA samples. Quality assessment of total and amplified RNA. Bioanalyzer gel image (A) of total RNAs is represented by duplicated samples from each of the three dissected cochlear regions. In each sample, the 28S, 18S, and 5S ribosomal bands are visible, and this is depicted by the bioanalyzer electropherogram (B) of total RNA isolated from the OC. (C) Bioanalyzer electropherogram of 1× amplified RNA (aRNA) from the OC overlaid on the corresponding electropherogram of a 6-kb RNA ladder (Ambion, Inc., Austin, TX). For all aRNA samples, the size range was about 100–6000 bases, with a midpoint of about 1000 bp.

Full size image

RNA amplification

Amplified RNA (aRNA) was synthesized using one round of a T7-based linear amplification protocol as described (Van Gelder et al. 1990; Baugh et al. 2001). Specifically, approximately 50 ng of total RNA was combined with 100 pmol of a T7 Oligo(dT) primer (Ambion), and the sample was denatured at 70°C for 10 min. First-strand cDNA synthesis was performed by adding 5× first-strand buffer (250 mM Tris–HCl, pH 8.3; 375 mM KCl; 15 mM MgCl2), 1 l of 10 mM dNTPs, 20 U of RNasin (Promega Corp., Madison, WI), 2 l of 100 mM DTT, and 200 U of Superscript II (Invitrogen). The reaction was incubated at 42°C for 2 h with the addition of another 200 U of Superscriptase II after 1 h. The reaction was then heat inactivated at 60°C for 15 min. Second-strand synthesis was performed using 5× second-strand buffer [100 mM Tris–HCl, pH 6.9; 450 mM KCl; 23 mM MgCl2; 0.75 mM beta-nicotinamide adenine dinucleotide (β-NAD+); 50 mM (NH4)2SO4], 4 l of 10 mM dNTPs, 20 U Escherichia coli DNA Polymerase I (Invitrogen), 10 U E. coli DNA Ligase (Invitrogen), and 2 U ribonuclease H (Invitrogen). The reaction was incubated at 16°C for 2 h. Afterwards, the cDNA was purified and concentrated using the DNAclear kit (Ambion).

In vitro transcription was performed using the MEGAscript T7 kit (Ambion) by adding NTPs, 10× reaction buffer, and T7 enzyme to the purified cDNA. The resulting solution was incubated at 37°C for 14 h. Afterwards, the aRNA was purified using the RNeasy Mini Kit (QIAGEN). RNA integrity was verified, and the concentration was determined as described above (see Fig. 2C). We were able to achieve greater than a thousand-fold amplification of the mRNA. Linear amplification of these samples was validated by their global expression data and comparing genes that are comparably expressed, such as housekeeping and ribosomal protein genes.

Probe synthesis, hybridization, and laser scanning

A probe was generated for each of the 12 aRNA samples. A series of hybridizations was performed, with each probe being used in exactly two hybridization experiments as indicated in Table 1. For probe generation and hybridizations, we used the 3DNA 350 RP Kit (Genisphere Inc., Hatfield, PA), which utilizes a dendrimer-based strategy to enhance the signal (Capaldi et al. 2000). We followed the standard protocol for probe generation for a dual channel experiment, beginning with 200 ng of each aRNA sample. Briefly, RNA from each tissue was reverse-transcribed into cDNA using unlabeled dNTPs. A capture sequence, which serves as the binding site for the fluorescent dendrimer, was ligated to the cDNA. The probes were combined and successive hybridizations were performed, first of the cDNA to the glass slide, then of the dendrimer mix to the cDNA. In both hybridization mixes, 3 l Mouse Cot1 DNA (Invitrogen) was added, and 16 g poly d(A)40–60 (Amersham Bioscience, Piscataway, NJ) was substituted for the dT blocker supplied in the kit. Both hybridizations were performed using the 2× formamide-based hybridization buffer at 50°C. The hybridization times and posthybridization washes followed the recommendations in the protocol. Following hybridization, the slides were scanned with a GenePix 4000B laser scanner (Axon Instruments, Union City, CA) to determine the amount of each fluorescent cDNA that hybridized. Initial spot analysis and background subtraction were performed using the GenePix Pro 5.0 software (Axon Instruments).

Normalization and background subtraction

Background was calculated using a local feature background median, which was subtracted from each spot. Using this method, a background value is calculated for each feature individually based on the median intensity of pixels immediately surrounding the feature. A positive signal was defined as those spots with intensity levels greater than two standard deviations above background. The background-subtracted ratio of medians (635/532) was calculated for each spot. These data were then normalized using locally weighted print-tip normalization (Yang et al. 2002) to transform each ratio into a log value.

Data analysis

Log-transformed values from locally weighted normalization were input into the Significance Analysis of Microarrays (SAM) program (Tusher et al. 2001). Each comparison was input as a one class experiment, utilizing block permutations for the replicate spots on each chip. The delta value, which is used to set the cutoff point for significance, was adjusted to maximize the number of genes deemed significant while minimizing the number of false positives determined by the program. We found that there was a very narrow range for the delta value if we wanted to keep the number of false positives to a minimum.

Results

Microarray characterization

Our mouse inner ear-pertinent custom microarray chip (designated CMA-IE1) is derived from the subtracted version of the RIKEN adult inner ear set (Unigene Library ID: 9974) and the NIH organ of Corti library (Unigene Library ID: 10920). As summarized in Table 2, the array is composed of 5758 of the RIKEN clones and 1288 NIH clones, all spotted in triplicate. The majority of the NIH clones on the array consist of unknown or uncharacterized ESTs to facilitate gene discovery. Clones were also added, both from the NIH library and from our own collections that are involved in important auditory pathways. In addition, the array contains a number of controls (buffer, mouse Cot1, polyA) and spikes for the purposes of normalization and data analysis. We define spikes as genes added to the array from various species (e.g., Xenopus laevis, Anopheles gambiae, Schistosoma mansoni, and Apis mellifera) that have been shown not to hybridize to any known mouse transcript. Approximately 60 different spikes were added to the microarray chip. Spikes can serve as negative controls or included in the hybridization mix to quantify the sensitivity of the assay.

Table 2 Microarray statistical analysis
Full size table

Depending on the comparison, around 22–25% of the clones represented on the array showed positive signals greater than two standard deviations above background. This varied based on the cDNA source. Specifically, about 27–31% of the clones from the NIH library were detected at this level compared with about 19–24% of the RIKEN clones. For each comparison, the ratio of NIH to RIKEN clones detected was nearly the same. For the most part, we expected to see a fairly low level of detection. This is because the RIKEN inner ear set is highly subtracted and therefore consists of many genes expressed at low levels and/or only in specific tissues or cell types. The NIH organ of Corti unknown and/or uncharacterized cDNA clones ranged from being singletons to those which are moderately expressed. Singletons are defined as cDNAs that were isolated only once in a given cDNA library. In addition, we expect many genes from the NIH library to be detectable in the immature ear of postnatal mice (postnatal days 5–13) and not in the adult cochlea because these cDNAs originate from mice at various developmental stages.

In general, low-frequency transcripts (10–25 copies/cell) were detected, whereas rare transcripts (<10 copies/cell) were not significantly above signal threshold to be classified as detectable. The limits of detection as defined by cellular transcript frequencies were ascertained by comparisons with the relative expression levels of housekeeping genes observed in reference control samples along with the clone frequency observed in the organ of Corti library (Pompeia et al. 2004). Because the organ of Corti library was not subtracted, the relative frequency of housekeeping and ribosomal protein transcripts within this library could also be inferred and compared with the corresponding cDNAs and their relative expression levels. One example of our microarray assay limit of detection is the observed expression of the nicotinic acetylcholine receptor α9 (Chrna9), which can be difficult to detect by in situ hybridization. Neither prestin nor Chrna10 was detected, and these genes are expressed at very low levels in the organ of Corti (Judice et al. 2002). This is consistent with microarray experiments in general. There is usually a cutoff point for detection that is dependent on many factors, including the number of copies of a gene present in the probe and the methods used for labeling and hybridization. The cutoff point can be estimated using spike controls (Badiee et al. 2003), which requires a great deal of planning, effort, and validation. Instead, we propose utilizing genes for ribosomal protein cDNAs present on the array as an internal standard. These genes were chosen because there are a relatively large number of them, and, for the most part, they are ubiquitously expressed. In addition, the mitochondrial ribosomal protein transcripts are considered to be rare transcripts. We found that the ribosomal protein gene expression levels could be used as an additional internal standard for estimating relative transcript frequency levels (Fig. 3). Interestingly, the relative expression levels observed in the microarray corresponded, in general, to the number of sequences of each of the ribosomal protein EST identified in the GenBank database. It must be recognized that their expression levels may vary among the different cell and tissue types, but their relative expression levels should not vary when comparing similar cellular or tissue samples. These factors also enable one to set up a QPCR standard curve to quantify expression levels of genes on the array based on their intensity levels. Genes for ribosomal proteins can also provide a good measure of overall assay sensitivity by comparing the intensity levels observed in any microarray experiment (Fig. 3). However, it is important that the lack of significant hybridization signal in our studies should not be taken to mean that there is no gene expression because the expression levels may be below the level of detection relative to other low-frequency transcripts.

Fig. 3
figure 3

Utilization of genes specifying ribosomal proteins in normalization assessment. Intensity levels of a large set of genes specifying ribosomal proteins present on the array were examined. Data are for one of the organ of Corti versus spiral ganglion comparisons, but similar results were observed across all experiments. The ribosomal protein genes were categorized into high (>1000 ESTs), moderate (>500 ESTs), and low (<250 ESTs) expression levels based on the number of ESTs in the entire mouse GenBank database. Genes with moderate and low expression levels were cytoplasmic ribosomal proteins and are designated. Those with background signal levels were mitochondrial ribosomal protein genes and are Mrpl10, Mrpl51, Mrps9, Mrps16, and Mrps19a.

Full size image

Differential expression analyses

For each comparison, locally weighted normalization produced at least 12 log-transformed values (4 hybridizations multiplied by 3 spots per slide) for every cDNA represented (see Fig. 4). These values were input into the SAM program and used to generate plots that determined which genes were significantly differentially expressed. Figure 5 shows SAM plots for each of the three comparisons. The delta values were adjusted to minimize the number of false positives while maximizing biological significance. The median false significant values were less than 0.5, indicating that our data are reliable. In each case, slightly more than 200 genes were characterized as differentially expressed. However, this included genes that were represented on the array more than once. The number of unique genes characterized as differentially expressed was around 150 for each comparison. The OC showed the highest number of unique upregulated genes, followed by the LW, and then the SG. This makes sense considering the source of the cDNA spotted onto the array. In addition, the OC has a larger diversity of cell types and functions compared with the other two tissues.

Fig. 4
figure 4

Locally weighted (lowess) log ratio normalization of microarray data. A microarray comparison between SG (red signal) and LW (green signal) is shown. (A) A typical M versus A scatter plot of unnormalized microarray data. The x-axis is 0.5log2 (red intensity × green intensity) or A. The y-axis is log intensity ratio or M. The log ratio values are skewed to the negative, especially at low intensities. (B) M versus A scatter plot of normalized microarray data. The distribution is now symmetrical about zero at all intensity values. Data are typical for all comparisons performed.

Full size image
Fig. 5
figure 5

SAM plots for cochlear dissected fraction expression comparisons. Comparisons were made between LW vs. OC (A), SG vs. OC (B), and SG vs. LW (C). For each comparison, the total number significant, median number false significant, and delta value are shown. Significant differentially expressed genes are represented by either green (tissue 1) or red (tissue 2) data points shown outside of the delta cutoff lines. The delta value is set to maximize the significant data points and minimize the false significant.

Full size image

Genes were identified that showed significantly higher expression in one tissue type compared with the other two. For example, genes upregulated in the OC for both the OC vs. LW and OC vs. SG comparisons were put into a group. We surmised that these genes may play an important functional role in their respective tissues. Table 3 shows the results of these data. Included are normalized SAM scores to compare the relative levels of differential expression. The OC group consists of several genes that do or may play crucial roles in hearing. These include otoancorin (Otoa) and transmembrane serine protease 3 (Tmprss3), which are both implicated in autosomal recessive deafness (Scott et al. 2001; Zwaenepoel et al. 2002). Also included are otoraplin (Otor), a possible deafness gene associated with the otic capsule (Cohen-Salmon et al. 2000), and otospiralin (Otos), which leads to hair cell degeneration and deafness in guinea pigs when downregulated (Delprat et al. 2002). The remaining genes in this group play diverse functional roles in the OC, and many have been shown to be expressed at relatively high levels in the OC compared with surrounding regions. One example is Skp1a, also known as organ of Corti protein 2 (Ocp2), which is part of a novel SCF ubiquitin ligase complex involved in cochlear homeostasis (Liang et al. 1997).

Table 3 Organ of Corti differentially expressed genes
Full size table

The SG group consists of a large proportion of myelin-related genes, including Plp, Cnp1, Mal, Mpz, and Fth. There are also a number of genes with other neuronal functions. For example, Ptgds is responsible for synthesis of brain-type prostaglandin D2, which is an abundant prostanoid in the central nervous system and a potent sleep-promoting substance in mammals (Onoe et al. 1988). Another example is Nrp, a receptor that mediates sensory pathfinding in the developing nervous system (He and Tessier-Lavigne 1997). This selection of myelin-related and neuronal genes is consistent with our expectations for the SG, which is composed of nerve fibers innervating the hair cells of the OC. Based on expression analyses of Mpz, Plp, and Erbb3 (Knipper et al. 1998; Lee et al. 2001; Hume et al. 2003), these genes should be highly expressed in SGs compared with the other dissected fractions. Similar to the immunostaining results, Erbb3 was present at higher levels in the SG compared with the OC fraction and was not detectable in the LW. A similar expression pattern was observed with Plp, which had significantly lower levels in the OC than the SG and was just above background in the LW. In situ hybridization studies demonstrated Plp mRNA in the cochlear oligodendrocytes but not in OC or LW (Knipper et al. 1998). Interestingly, a novel splice variant of collagen 1a1 (Col1a1), which uses an alternative polyadenylation signal, was identified and appears to be specific for the SG. It is listed in Table 4 under RIKEN clone ID F930112H09. The “typical” 3NCR of Col1a1 is present in all three dissected fragments at about the same levels. The functional importance, if any, of this novel splice variant is unknown.

Table 4 Spiral ganglion differentially expressed genes
Full size table

In the LW set, two major functional groups of genes were identified. The first group is involved in maintaining ion concentration and regulation of fluids within the inner ear. These include Car2, Car14, Slc4a11, Ppp1r1b, and Atp1b3. The second group of genes is involved in cell growth, proliferation, and differentiation. These genes include Tgfbr2, Qscn6, and Ndr3. Again, these groups of genes are consistent with our expectations for the LW, which is involved in homeostasis of cochlear fluids and endolymph secretion, among other things.

Discussion

To expand utilization of global expression analyses, baseline data of the normal relative levels and patterns of gene expression still need to be established for developmental stages and specific regions or cell types of the inner ear. Our custom microarray chip, CMA-IE1, represents an initial attempt at using a microarray designed specifically for the global expression of inner ear genes. cDNAs were primarily obtained from cDNA libraries with 20,000 clones being analyzed (Beisel et al. 2004; Pompeia et al. 2004), where approximately 2000 transcripts were identified only in the inner ear and likely representing rare to low-frequency inner ear transcripts. These unique ESTs represent 2–3% of the current mouse transcriptome encyclopedia (Okazaki et al. 2002; Carninci et al. 2003) and are not currently represented by any commercial microarray chip. This microarray increases our chances of finding new or novel genes involved in hearing or in development of the ear. As demonstrated in this study, four differentially expressed genes, Otoa, Otor, Otos, and Tmprss3, were identified that are associated with hearing impairment (Cohen-Salmon et al. 2000; Robertson et al. 2000; Ben-Yosef et al. 2001; Masmoudi et al. 2001; Scott et al. 2001; Delprat et al. 2002; Zwaenepoel et al. 2002). Although a number of unidentified singletons exhibited differential expression in the LW, OC, or SG, none were found to colocalize with a deafness locus (Beisel et al. 2004; Pompeia et al. 2004). Identification of functionally pertinent genes demonstrates the overall utility of the CMA-IE1 chip for further studies in spatiotemporal expression differences in normal and diseased inner-ear end organs, especially considering the representation of a large number of unidentified singletons and clusters. Reduction of the cellular complexity of the compared samples should also facilitate global expression analyses. Our goal is to develop a comprehensive set of transcripts pertaining to the inner ear on the array for identification of gene networks and arrangement of genes into functional clusters. Characterizing such pathways under normal or altered conditions requires the ability to screen many candidate genes simultaneously at different time points and under different experimental conditions.

A major hurdle for global expression analyses is the threshold of detection and the sensitivity of the microarray system to discriminate significant changes in expression of rare to low-frequency transcripts. Although a large number of low-frequency genes (10–25 transcripts/cell) can be detected, a number of inner-ear-pertinent genes are at or below this level of detection. The identification of both rare transcripts at <10 copies per cell and transcripts with limited but restricted expression within the different inner-ear cell types can be masked by assay noise generated by the general population. Limiting the transcript repertoire to a single cell population is a valid approach. However, obtaining a pristine cell population of at least 500 cells represents a technical challenge (Glowatzki et al. 1995; Zheng et al. 2000) and also requires linear amplification of the transcript repertoire. Linear amplification from a small number of cells does result in loss of specificity because of the production of shorter and shorter probes with each round of amplification (Scherer et al. 2003; Wilson et al. 2004). Biologically significant changes may still be undetectable by global analyses in the expression levels of inner-ear-pertinent genes. Nevertheless, quantitative PCR analyses for individual genes would be the method of choice to quantify these transcripts (Liberman et al. 2002; Stankovic and Corfas 2003).

Currently, only limited data are available for comparing our study with other inner-ear microarray studies, which primarily utilize commercially available chips. Expression analyses have been performed with an Atlas expression array (Clontech, Palo Alto, CA) containing 588 cDNAs, using 8- to 12-week-old rat whole cochlea and dissected fractions representing the modulus (MOD) and a combination of organ of Corti and lateral wall (OC–LW) (Cho et al. 2002). Sixteen differentially expressed genes were observed between rat MOD and OC–LW, and of these, only five genes (P2rx2, Plp, Pmp22, Limk1, and Nnat) were shared in common with our microarray. P2x purinogenic receptor 2 (P2rx2) was expressed at significantly higher levels in the mouse OC and rat OC–LW than in the SG, LW, or MOD. As suggested by immunohistochemical data (Housley et al. 1999; Jarlebark et al. 2002), this channel is expressed in OHCs, Deiters, and SG type I cells and is upregulated in response to noise (Wang et al. 2003). The remaining genes are primarily associated with neuronal tissue. We found myelin proteolipid protein (Plp) and neuronatin (Nnat) differentially expressed in the SG, whereas both Pmp22 and Limk1 transcripts were not significantly above background in all three dissected fractions.

Another study used the mouse Mu30K chip (Affymetrix, Santa Clara, CA) to examine differential expression between P2 and P32 mouse whole cochlea (Chen and Corey 2002a,b). There was little overlap between the Mu30 array and our CMA-IE1 chip. Only three genes could be compared that were developmentally upregulated in adult mice (Chen and Corey 2002a,b). These were connexin43 (Gja1), a twin-pore potassium ion channel (Kcnk1), and the Na+, K+-activated ATP phosphohydrolase α1 subunit (Atp1a1). These genes appear to play a role, to varying degrees, in K+ recirculation in the inner ear (Estevez et al. 2001; Liu et al. 2001; Wangemann 2002; Nicolas et al. 2003). We found higher levels of Gja1 transcripts in the SG and LW compared with the OC. This reflects the tissue distribution of connexin43 within the cochlea, in which these channels are mainly localized in the stria vascularis and spiral limbus (a component of the spiral ganglion fraction) (Suzuki et al. 2003; Cohen-Salmon et al. 2004). Both ATP1a1 and Kcnk1 (also designated as Twik-1) are ubiquitously expressed in all tissues (Orlowski and Lingrel 1988; Lesage et al. 1997) and are found in the ganglion, vestibular end organs, and the cochlea of the inner ear. However, immunohistochemistry studies have demonstrated that ATP1a1 and Kcnk1 are highly expressed in the marginal and vestibular dark cells. Both these proteins were shown at lower levels in the organ of Corti and spiral ganglion neurons (Zuo et al. 1995; Nicolas et al. 2003). We found higher levels of both these transcripts in the SG and OC compared with the LW, which does not appear to fit the protein expression patterns. Whole mount in situ hybridization and immunohistochemistry in combination with quantitative RT-PCR should help to resolve these discrepancies between transcript and protein expression levels as well as their associated cellular distribution patterns. Ideally, SAGE analyses (Gunnersen et al. 2002) of dissected cochlear fractions from adult mice should provide a more accurate depiction of the global transcript frequencies in these three fractions. Although these different semiquantitative expression platforms may provide limited comparative data, they still have demonstrated the usefulness of global expression analysis in the inner ear.

Our data demonstrate that differentially expressed genes can be identified among the three cochlear fragments and, in general, reflect their cellular, structural, cytoarchitectural, and functional disparities. Both the SG and LW fractions have a more limited cellular diversity compared with the OC. As noted in differential expression analyses, the cellular diversity is further reflected in the three groups of differentially expressed genes presented in Tables 3 4–5. We categorized these genes from their general functional descriptors associated with their “gene ontology (GO) terms,” which is based on function, process, and component characteristics of the gene products (Deng et al. 2004; Harris et al. 2004). The majority (11/16; 68.8%) of differentially expressed genes in the SG are associated with neuronal tissues with three additional clones, secreted frizzled-related protein 4 (Sfrp4), chemokine ligand 12 (Cycl12), and breast cancer metastasis-suppressor 1-like (Brms1l), representing molecules involved in cellular development and proliferation. In the LW fraction, genes related to ion regulation and transport (8/23; 32%) and those involved in cellular development and proliferation (6/23; 24%) constitute the majority of differentially expressed genes. Both carbonic anhydrase genes (Car2 and Car14) and Na+, K+ ATPase (Atp1b3) are all expressed in the lateral wall spiral ligament fibrocytes and are thought to play a role in cochlear fluid and ion homeostasis (Ichimiya et al. 1994; Spicer et al. 1997). K+ homeostasis function of the marginal cells is mediated through the action of membrane proteins, such as Na+, K+ ATPases, protein phosphatase inhibitors, and Na+, K+, Cl co-transporters (Zhao et al. 1994; Agrup et al. 1997; Spicer et al. 1997; Delpire et al. 1999; Dixon et al. 1999; Wangemann 2002). In the LW, we found a differentially expressed subunit of the protein phosphatase 1 regulatory (inhibitor) protein, Ppp1r1b. This gene is likely associated with marginal cells. Another ion transporter, a sodium bicarbonate transporter (Slc4a11), was also identified in the lateral wall. One member of this family of genes, Slc4a7, was found to result in blindness and auditory impairment in mice that carry the targeted disrupted Slc4a7 gene (Bok et al. 2003) and was localized to fibroblasts in the spiral ligament, spiral prominence, and the suprastrial region. The functional properties of the differentially expressed genes associated with the OC were fairly diverse and likely are a reflection of the number and disparity of different cell types within the OC. As such, no one group of genes could be readily identified as being prevalent. To approximate the transcript frequency of the OC differentially expressed genes, we used the frequency of clones derived from the NIH organ of Corti cDNA library. By and large, gene expression levels ranged from moderate [<10% of all the clones: otospiralin (Otos), otoraplin (Otor), and S-phase associated protein 1A (Skp1)], low to moderate [10–30%: claudin 9 (Cldn9), dermatan sulfate proteoglycan 3 (Gspg3), and gelsolin (Gsn)], and the remainder, which were at a low level of occurrence. OC differentially expressed genes were also represented by three apparently low-frequency transcripts: nonspecific dipeptidase (Cn2) and two hypothetical proteins, MGC58343 and E130318F19Rik. These three genes were identified in the RIKEN subtracted library cDNAs but not in the organ of Corti cDNA clusters (Pompeia et al. 2004). Thus discrimination of differentially expressed transcripts could be detected regardless of the corresponding frequency.

Table 5 Lateral wall differentially expressed genes
Full size table

Although the array was validated through differential expression analyses, detection of cell-unique and rare transcripts is still a major issue and is dependent on the complexity of the sample as well as quantity of starting material necessary for the generation of microarray probes. The small quantities of material obtained from either cells or dissected cochlear tissues necessitate a twofold approach: the amplification of total RNA (Van Gelder et al. 1990; Baugh et al. 2001; Xiang et al. 2003) and a signal amplification of fluorescence signals using dendrimer technology (Stears et al. 2000). Both these strategies were incorporated in our studies. Because of the large number of different cell types and the low percentage of target cell types (e.g., hair cells) within the cochlea, it must be recognized that important rare to low-frequency inner-ear-pertinent transcripts are consequently diluted out in the sample. Methodologies for analysis of small sample sizes, representing single cell to small populations of a single cell type, are still evolving. Because of sampling error, the use of pristine populations of specific cell types is more desirable than single cell samples. Similarly, the statistical methods and approaches for normalization of the raw data and subsequent evaluation of microarray generated data (Tusher et al. 2001; Yang et al. 2002; Dudoit et al. 2003) require sufficient replication, which is difficult with single cell samples.

Microarrays can be used to identify genes that are expressed as well as upregulated or downregulated in the cochlea or its subsections to identify potential candidate deafness genes (Lomax et al. 2000; Chen and Corey 2002a,b; Iijima et al. 2004). Studies could also be designed to examine the expression profiles of mutants versus their wild-type counterparts to facilitate the discovery of genes involved in a hearing-related disorder. There are also a wide variety of developmental pathways and gene networks in the ear that could be delineated with microarray studies. Some examples are primary neuron formation and differentiation (Fritzsch and Beisel 2003) and the identification of downstream targets of transcription factors involved in hair cell formation and differentiation (Bryant et al. 2002; Hertzano et al. 2004). We will work to continually identify and add genes to the array representing unknown and/or uncharacterized transcripts from cochlear libraries to facilitate studies such as these.