A Simple Method for Purification of Vestibular Hair Cells and Non-Sensory Cells, and Application for Proteomic Analysis

Mechanosensitive hair cells and supporting cells comprise the sensory epithelia of the inner ear. The paucity of both cell types has hampered molecular and cell biological studies, which often require large quantities of purified cells. Here, we report a strategy allowing the enrichment of relatively pure populations of vestibular hair cells and non-sensory cells including supporting cells. We utilized specific uptake of fluorescent styryl dyes for labeling of hair cells. Enzymatic isolation and flow cytometry was used to generate pure populations of sensory hair cells and non-sensory cells. We applied mass spectrometry to perform a qualitative high-resolution analysis of the proteomic makeup of both the hair cell and non-sensory cell populations. Our conservative analysis identified more than 600 proteins with a false discovery rate of <3% at the protein level and <1% at the peptide level. Analysis of proteins exclusively detected in either population revealed 64 proteins that were specific to hair cells and 103 proteins that were only detectable in non-sensory cells. Statistical analyses extended these groups by 53 proteins that are strongly upregulated in hair cells versus non-sensory cells and vice versa by 68 proteins. Our results demonstrate that enzymatic dissociation of styryl dye-labeled sensory hair cells and non-sensory cells is a valid method to generate pure enough cell populations for flow cytometry and subsequent molecular analyses.


Introduction
Molecular analyses of the inner ear's specialized cell types are hindered by the paucity of these cells. This fact might be one of the reasons why hearing and balance are among the senses that are still only partially elucidated at the molecular level. Although a single inner ear contains several thousand sensory hair cells, the cells are scattered into five vestibular sensory patches plus a sixth auditory sensory epithelium located in the cochlea. This spatial dispersion combined with the circumstance that the inner ear is shielded by one of the hardest bones of the body makes it difficult to obtain sufficient quantities of sensory hair cells and their associated supporting cells for molecular analysis. Obviously, sensory hair cells are interesting because present-day research seeks to understand the process of mechanoelectrical transduction, or pursues the specific proteins that contribute to the unique features of the hair cells' afferent ribbon synapses, among a battery of other interesting topics surrounding hair cell biology [1,2]. Supporting cells, on the other hand, are interesting because in non-mammalian vertebrates they appear to serve as somatic stem cells, able to reverse vestibular and cochlear hair cell loss and restore function [3]. In mammals, only a few supporting cells of the adult vestibular sensory epithelia display stem cell character-istics [4], whereas cochlear supporting cells lose this feature during the first neonatal weeks [5][6][7].
Creative use of transgenic mice in combination with flow cytometry is a recently utilized strategy for purification of hair cells [7], supporting cells [6,8,9], and other otic cell types [10,11] for molecular and other cell biological analyses. Likewise, fluorescently labeled antibodies to cell surface proteins have also been used for purification of various cell populations from the inner ear [7,12]. Despite many advantages of these two strategies, they have the disadvantage of requiring either a transgenic reporter or the expression of a specific cell surface marker on the cell type of interest. We sought to develop a strategy that eliminates these requirements by utilizing a functional feature of mature sensory hair cells -their ability to rapidly take up certain styryl dyes [13,14]. In addition, we used the avian inner ear utricle and saccule, two vestibular organs whose sensory maculae can be enzymatically detached and peeled away from underlying cells, allowing the harvest of sensory epithelia that consist solely of hair cells, and non-sensory cells including supporting cells. We chose to analyse the purified cell populations by mass spectrometry, which unveiled a snapshot of the proteomic profiles of vestibular hair cells and non-sensory cells. We utilized a statistical data analysis strategy that was valuable in dealing with potential cross-contamination, which we identified as a potential limitation of the technology. Our overall strategy led to the identification of more than one hundred proteins each specific for hair cells and nonsensory cells demonstrating the applicability of styryl dye labeling and flow cytometry for inner ear research.

Dissociation of vestibular sensory epithelia into single cells
We used chicken embryos at their 18 th day of incubation for isolation of hair cells, non-sensory and supporting cells. We focused on the vestibular maculae of the utricle and saccule for three reasons: i) they comprise the largest hair cell-bearing organs of the inner ear containing more than 20,000 hair cells per utricular macula, ii) the hair cells are functional at this late embryonic age [15], and iii) utricles and saccules can be dissected relatively quickly in larger numbers. After dissection and removal of otolithic membranes, the tissues were exposed to the styryl dye AM1-43 or FM1-43FX (Fig. 1A,D). Brief exposure to either of these dyes intensely labels living hair cells, whereas supporting and other non-sensory cells remain unlabeled [13] (Fig. 1C,F). Differentially labeling of hair cells and non-sensory cells is the basis for subsequent separation of hair cells from residual unlabeled cells of the sensory epithelia by flow cytometry. Specificity of the dye-uptake was confirmed by immunocytochemistry with antibodies to the hair cell marker myosin VIIA (Fig. 1C, F). After hair cell labeling, the sensory epithelia of the utricles and saccules were enzymatically detached from underlying stromal cells, mechanically separated from the stromal layer, and the living epithelia consisting of labeled hair cells and unlabeled non-sensory cells including supporting cells were collected in fresh media (Fig.  1B, E).
We optimized the dissociation method for vestibular sensory epithelia to ensure thorough cell separation and minimal cell aggregation but also high viability judged by cell shape and exclusion of propidium iodide, a dye that is generally unable to enter viable cells. Representative results obtained with different dissociation strategies are shown in Figure 2. The predominantly enzymatic digestion conditions included 0.25% trypsin, Accutase cell detachment mixture, an enzyme-free formulation of chelating reagents, 0.05% trypsin, Accumax cell dissociation mixture, and 50% Accumax plus 0.025% trypsin. We found that neither trypsin alone nor the commercially available enzyme cocktails Accutase or Accumax were satisfactory to quantitatively dissociate the tissue. These tests were systematically done by varying incubation times from a few minutes to up-to 30 minutes, followed by mild trituration, and resulted in either insufficient cell separation or starkly reduced viability ( Fig. 2A-D). Combining Accumax cell dissociation solution at half strength with a low concentration of trypsin, however, for a total incubation time of 7 minutes resulted in the optimal separation of viable individual cells (Fig. 2F). Hair cells separated with this procedure displayed at least some rudimentary preservation of cytomorphology. The enzyme-free formulation of chelating reagents alone was also highly efficient for cell separation (Fig. 2E); however, hair cell morphology was not well preserved in this condition, presumably caused by chelating divalent cations such as Ca 2+ , which are important for hair bundle integrity [16,17]. Moreover, cell viability was reduced when compared with the Accumax and trypsin combination.
Flow cytometric purification of AM1-43 labeled hair cells and unlabeled non-sensory cells After cell dissociation, intense AM1-43 labeling of hair cells persisted (Fig. 2F), which we utilized to separate the AM1-43positive cells from unstained cells. The flow cytometric gating strategy disregarded propidium iodide-labeled dead cells, which ranged between 7-15%, as well as cell debris (Fig. 3A). Doublets were identified and excluded based on non-proportionate forward scatter for height and area parameters (Fig. 3B). Of the viable single cells, AM1-43-high and AM1-43-low cell populations were gated for collection (Fig. 3C). We expected that the AM1-43-high population consist of labeled hair cells, whereas the AM1-43-low population should comprise mainly supporting cells, contaminating mesenchymal cells from the underlying stroma, undifferentiated/progenitor cells, and perhaps some immature or damaged hair cells that did not take up the styryl dye. These potential contaminants are not a limitation of the embryonic age of the tissue because undifferentiated and immature cells are also detectable in posthatch chickens [15]. Approximately 20-25% cells displayed AM1-43 fluorescence intensities between the low and high gates and were not collected. To demonstrate specificity, we used vestibular sensory epithelia not exposed to AM1-43 dye as Figure 2. Dissociation of vestibular sensory epithelia into single hair cells and non-sensory cells. AM1-43-stained sensory epithelia underwent different enzymatic and non-enzymatic treatments to test for optimal conditions to separate hair cells and non-sensory cells and to preserve hair cell morphology. Conditions were: 0.25% trypsin (A), 0.05% trypsin (B), accutase (C), accumax (D), enzyme-free (E) and 50% accumax + 0.025% trypsin (F). Shown are representative images of cells after mild trituration following 7 min incubations at 37uC. doi:10.1371/journal.pone.0066026.g002 a negative control and subjected the dissociated cells to flow cytometric analysis. With this control, we only detected a single population of viable single cells that displayed background levels of fluorescence in the AM1-43 detection channel (Fig. 3D). In 6 experiments using approximately 120630 utricles and saccules per independent experiment, we collected on average 31.468.8% AM1-43-high, presumed hair cells, and on average 43.369.5% AM1-43-low, presumed non-sensory cells. In numbers, this corresponds to 172,200660,000 hair cells and 261,7006100,400 non-sensory cells per experiment. When re-sorted with the same parameters, each of the populations displayed at least 90% purity (Fig. 3E,F).

Mass spectrometry and proteomic analyses revealed distinct proteomic signatures of hair cells and nonsensory cells
The AM1-43-high and AM1-43-low populations of each sorting experiment were collected into lysis buffer, concentrated and the proteins of each population were electrophoretically separated. Eight polyacrylamide gel pieces representing eight electrophoretically fractionated groups of proteins were excised for each cell population, in-gel digested with trypsin, and analyzed by liquid chromatography-tandem mass spectrometry (LC-MS/MS) for protein identification and quantification. In total, we obtained three independent datasets for AM1-43-high and AM1-43-low cell populations, respectively.
For the interpretation of the resulting mass spectrometry based datasets, refined statistical search strategies that mitigate and control incorrect peptide identifications were used. Specifically, we utilized concatenated target-decoy database searches, which employ a strategy using composite protein target sequence databases and decoy sequences that comprise the reversed target sequences to estimate false positive identification rates for individual peptide-spectral matches (PSMs) [18]. With this method, it is possible to distinguish between correctly identified spectra, which should be derived solely or mainly from target sequences, and incorrectly identified PSMs, which should be traceable more or less in equal proportions to target and decoy sequences. Based on the resulting hits, a false discovery rate (FDR) was generated for each PSM dataset, which we used to filter matching PSMs (Fig. 4A).
We further used spectral counting as a quantitative tool to assemble expression profiles of proteins detectable in hair cell and supporting cell fractions. Spectral counting relies on the identification of peptide spectra at the tandem mass spectrometry fragment ion level and sums the number of spectra identified for a given protein (Fig. 4B). The results for all detected peptides are integrated and reported as a single count value for a particular protein. One of the major drawbacks in relying on spectral counting data is that its quantitative power at a low number of counts can be unreliable. To address this issue to the best of our ability, and in dealing with data derived from very low cell counts, a minimum of two peptides identified for a particular protein was set as a prerequisite. When we applied these criteria, we were able to identify and provide semi-quantitative abundance values for 634 proteins. These hits were based on 18,224 PSMs with a protein FDR of 2.7% and a peptide FDR of 0.8%, respectively (Fig. 4A). We acknowledge that more exact quantitative measurements would be needed for more precise data analysis, for example summed dissociation-product ion-current intensities [19] or isobaric tags [20] combined with sophisticated normalization and standardization [21]. tringent analyses revealed proteins that are specifically detectable in hair cells and non-sensory cells  (SCAMP1). Myosin VIIA is a commonly used hair cell marker [22,23], which confirms that the AM1-43-high cell population indeed contained hair cells.
Because AM1-43 is specifically taken up by hair cells, we were surprised that some previously described specific hair cell marker proteins, such as otoferlin, calretinin or parvalbumin were not only observed in the presumptive hair cell fraction, but were also detected in the AM1-43-low presumed non-sensory cell samples. A possible explanation for this potential contamination is that the mechanoelectrical transduction apparatus of some hair cells might have become damaged during the dissection, thereby leading to a fraction of unlabeled hair cells that would have been sorted into the AM1-43-low cell fraction.
To address this issue, we quantitatively assessed the expression profiles in the two different cell types with spectral counting and statistical testing of the contingencies of individual protein classifications into the two groups. We categorized the samples based on the assumption that a protein would be specific for either the hair cell or supporting cell population, which can be tested with Fisher's exact analysis, generating a p-value for each identified protein. By far, the most abundant protein in the hair cell fraction was otoferlin (OTOF), a protein defective in a human deafness form DFNB9, and involved in the exocytosis and replenishment of synaptic vesicles to specialized ribbon synapses in hair cells [26][27][28]. Based on the significantly higher abundance of OTOF in the AM1-43-high cell population (321 spectral counts versus 17 spectral counts in the AM1-43-low population), we were confident that otoferlin is specific for the AM1-43-high cell population and that its expression in AM1-43-low cells is either caused by contaminating unlabeled hair cells or by low-level expression of otoferlin in non-sensory cells. Other proteins that were re-categorized to the hair cell population using Fisher's exact test included adenylate kinase isozyme 1 (AK1), an enzyme involved in energy metabolism [29], the synaptic vesicle protein Vtype proton ATPase subunit B (ATP6V1B2), as well as the hair cell markers calretinin [30,31] and parvalbumin [32,33]. Besides the lack of dye uptake by certain hair cells and the resulting potential contamination of the AM1-43-low population, we also reason that a single round of flow cytometry, even with discarding 20-25% of cells that were neither AM1-43 high nor low (Fig. 3C) results in .90% enrichment, but not absolute purity. Double sorting, on the other hand, would have dramatically reduced the cell number and thereby affected the overall detection sensitivity. Based on the results of the Fisher's exact analysis, however, with a cutoff at a pvalue of ,0.05, we were able to reassign 53 additional proteins to the hair cell fraction (Table 3). Our interpretation of these  assignments is that these 53 proteins are either hair cell specific or strongly upregulated in hair cells compared to non-sensory cells. Conversely, we also found proteins that are strongly upregulated or specifically detectable in the non-sensory cell fraction, based on the same assumptions and Fisher's exact test with a cutoff of p,0.05, we reassigned 68 proteins to the non-sensory cell fraction (Table 4).
Our results revealed the limitation of our strategy that even minor cross-contaminations of the different cell types can mask the specific categorization of proteins. For proteins that are abundantly expressed in one cell type and not or only at a low level in the other, the spectral counting in combination with the Fisher's exact test turned out to be a valuable tool to re-assign proteins to a single cell type.

Categorization of the hair cells' and non-sensory cells' proteomes
To characterize the actual proteins that we identified with our proteomic approach in more detail, we subcategorized all identified proteins according to their annotated subcellular localization (Fig. 5A) and function (Fig. 5B). With respect to all proteins that we detected in hair cells and all proteins detected in non-sensory cells, we found no significant differences in proteome composition (Fig. 5A, before quantification). This was not very surprising because the majority of identified proteins (467) were observed in both cell types (see Fig. 4C). Nearly 50% of all identified proteins were cytoplasmic, 16% were nuclear and 13% were of mitochondrial origin. The residual 21% localized to vesicles, plasma membrane, Golgi apparatus, lysozymes or are secreted proteins. A small portion of proteins was not annotated and could not be assigned to a subcellular localization. Regarding function, the largest fraction (16-18%) of proteins identified in both cell types was found to be involved in energy metabolism, followed by trafficking, signal transduction, protein synthesis and degradation (Fig. 5B, before quantification). 1% of all identified proteins of each cell type function as extracellular matrix proteins.
Next, we conducted the same analysis but implied the spectral counts of each identified protein in order to compare relative expression levels. The subcellular distribution of all proteins was still comparable between hair cells and supporting cells (Fig. 5A, after quantification), however the ratios between subcellular compartments changed. Whereas an increase of 6% and 9% points was detected for the cytoplasmic localization, as well as vesicular and secreted proteins, respectively, a slight decrease was observed for the other subcellular compartments compared to before quantification.
Similar results were obtained for the analysis of the cellular functions. Here, a major increase was observed for cytoskeletal proteins that maintain the cellular structure for both hair cells and non-sensory cells (Fig. 5B, after quantification). A substantial difference between hair cell and non-sensory cell proteins appeared for the category trafficking. Whereas the percentage of non-sensory cell proteins involved in trafficking nearly stayed constant compared to before quantification, an increase of 9% points was noted for hair cells proteins. This result might be indicative of a potential higher need for protein trafficking in hair cells compared with non-sensory cells. Besides trafficking of stereociliary bundle proteins, hair cells also maintain substantial trafficking of proteins to the basolateral wall and synaptic sites. Interestingly, the most abundant protein we identified in the presumed hair cell fraction was otoferlin, which plays a key role in replenishment of synaptic vesicles in hair cells [26]. For nonsensory cells, an upregulation of extracellular matrix (ECM) proteins was noted after quantification (1% versus 0% points in Table 1 hair cells), which could be an indication that these cells such as the supporting cells that are included in this population are closer associated with the basilar membrane compared to hair cells, or reflective of a possible contamination of this cell population with mesenchymal cells from the underlying stroma. In summary, our analyses revealed differences in the proteomic compositions of chicken vestibular hair cells and non-sensory cells, which is not surprising given the specific function associated with sensory hair cells compared to non-sensory cell types. Our quantitative assessment of the data and the comparison is further limited by the fact that only the measureable portions of the proteomes are being considered, which creates a bias for abundant proteins detectable with the methods used.
Based on these considerations, we hypothesized that potential differences between the two populations would be even more obvious if we focus our analysis on proteins that are either exclusively detectable in each group (Tables 1 and 2) and proteins that are highly enriched or specific for each group (Tables 3 and  4). Comparisons with respect to subcellular localization revealed that of the specific hair cell and non-sensory cell proteomes 40% to 50% of all unique proteins were of cytoplasmic origin (Fig. 6A, before quantification). A higher percentage of unique non-sensory cell proteins over unique hair cell proteins was assigned to the ER (12% compared to 3% for hair cell specific proteins), or were not annotated. In contrast, slightly higher percentages of unique or upregulated hair cell proteins were found to be of mitochondrial, vesicular, Golgi and lysosomal origin. After quantification, a main difference arose for the vesicle proteins where an increase of 26% points was revealed for unique or upregulated hair cell proteins compared to 1% of unique or upregulated non-sensory cell proteins (Fig. 6A, after quantification). This increase mainly arose from the high number of spectral counts of the two proteins otoferlin and clathrin, both shown to be involved in hair cell vesicle trafficking [26,34]. Accordingly, for the cellular function, a notably strong upregulation was observed for hair cell specific proteins involved in trafficking to 50% of all hair cell specific/ enriched proteins versus 4% of all non-sensory cell specific/ enriched proteins (Fig. 6B, after quantification). These quantitative assessments demonstrate that in comparison to non-sensory cells, protein trafficking is strongly reflected in the hair cell proteome. As discussed earlier, this might reflect the high turnover of synaptic vesicles due to sustained exocytosis at the ribbon synapses, with otoferlin as a key player in vesicle recycling and replenishment, as well trafficking of proteins to the stereociliary hair bundle. Conversely, based on the quantification, the non-sensory cells' proteome appears to be enriched for proteins involved in synthesis, degradation, folding and particularly cytoskeletal proteins, which could be an indication for a higher protein turnover and cytoskeletal specializations in these cells, despite the well-known cytoskeletal structures of hair cells.

Validation of the proteomic analyses with immunohistochemistry
Not surprisingly, we identified a number of proteins in hair cells and non-sensory cells that previously were known markers for these cell types. Otoferlin for example, is a known hair cell protein [26,35] that was identified in our analysis as highly enriched in hair cells after Fisher exact analysis. Monoclonal antibody staining confirmed that otoferlin is detectable in E18 by chicken utricular hair cells, co-labeled with antibodies to myosin VIIA, and that otoferlin immunolabeling is absent from non-sensory cells (Fig.  7A,B). We used Sox2 immunostaining to distinguish sensory epithelium cells from mesenchymal stromal cells. In the E18 chicken utricle, Sox2 protein is expressed by supporting cells and  hair cells (Fig. 7A,B). We also confirmed hair cell expression of the mitochondrial protein apoptosis-inducing factor 1 (AIFM1), which was identified by our mass spectrometry analysis as hair cell only protein (Fig. 7C,D). The protein was, however, also detectable albeit with lower intensity in non-sensory cells. This result revealed, as previously discussed, a limitation of the comparative analyses that is a general lack of sensitivity for proteins that are not highly abundant. AIFM1, for example appears to be strongly enriched in hair cells and was identified via two independent peptides in one of the mass spectrometry experiments ( Table 1). The protein was not detected by mass spectrometry in the nonhair cell fraction. Immunolabeling revealed a clear difference in staining intensity between hair cells and non-sensory cells, highlighting the differential expression of AIFM1 in these two cell types, but it also demonstrated expression of AIFM1 in nonsensory and supporting cells. This result shows that absence of detection of a protein in mass spectrometry does not mean that the protein is not present. Mass spectrometry has detection limits, which has been elegantly shown and discussed in a recent quantitative study of hair bundle proteins [21]. Overall, as reported in these recent results, we also observed that the detection limit for spectra is limited, leveling out at about 10 4 per mass spectrometry run in the best cases. Particularly for abundant and large proteins, such a detection limit is not a big problem because the statistical likelihood that these proteins are represented by multiple peptides in a single run is quite high. For smaller proteins that are less abundant, the limit of detection might not be reached in a single run. In addition, it is reasonable to presume that simple biochemical features also limit the representation of certain groups of proteins -for example globular cytoplasmic versus membrane-spanning proteins, or detergent solubility, charge, protein degradation sensitivity, etc. For better representation and less variability, a substantial increase of the detection limit and methods for exclusion of abundant proteins would probably be the most efficient means. For non-sensory cells, collagen XVIII alpha 1 and talin were identified by mass spectrometry, and monoclonal antibodies to these two chicken proteins detected them in association with nonsensory and supporting cells and not hair cells in the E18 utricle (Fig. 8). Antibodies to the extracellular matrix protein collagen XVIII [36], which was identified in the non-sensory cell only fraction (Table 2), labeled the basal lamina directly underneath the non-sensory cell layer. No immunoreactivity was detectable in hair cells. This localization, combined with the mass spectrometry data suggests that non-sensory and presumably supporting cells secrete collagen XVIII, but it cannot exclude a possible contribution of mesenchymal stromal cells to the mass spectrometry data because these cells were also labeled with collagen XVIII antibodies (Fig.  8A). The cytoskeletal protein talin, which is found concentrated at focal adhesion points and at points of cell-substratum contact [37], was identified via Fisher exact analysis as highly enriched by nonsensory cells compared to hair cells. Monoclonal antibodies detected the protein in supporting cells as well as in mesenchymal stromal cells, but not in hair cells (Fig. 8B,C).

Concluding thoughts
We report a simple method to purify vestibular hair cells and non-sensory cells from the chicken inner ear. The approach generates cell populations of .90% purity that can be used for molecular studies, such as proteomic analyses. Our comprehensive evaluation of the individual datasets revealed certain limitations, such as presumptive inefficient labeling of damaged hair cells and the potential for cross-contamination during the single-pass flow cytometric sorting. On the other hand, we also showed that statistical analyses of proteomic data are a powerful tool to extract categorical information of protein distribution in experiments where minor cross-contamination affects the results. Our proteomic analyses identified proteins and protein categories that are enriched in vestibular hair cells and non-sensory cells. Some of these proteins were previously not considered in the context of inner ear sensory biology, and our datasets consequently are of relevance to researchers interested in hair cell and non-sensory or supporting cell function and development.

Dissociation of vestibular sensory epithelia into single hair cells and non-sensory cells
Embryonic day 18 (E18) chicken embryos were sacrificed by rapid decapitation. Utricles and saccules were dissected from the head in ice-cold HBSS with calcium and magnesium (Gibco) and otolithic membranes were removed without any enzymatic treatment. Next, utricles and saccules were carefully transferred with a micro spoon into AM1-43 dye solution (10 mM AM1-43 (Biotium) in Medium 199 (M199, Cellgro)) for 30 seconds at room temperature in a standard petri dish. Under these conditions AM1-43 preferentially enters hair cells via the mechanotransduction channel [13,14]. After staining, utricles and saccules were transferred into fresh Medium M199 to wash off residual dye. A control sample of 5 utricles underwent the same staining procedure in AM1-43 dye free M199 medium.
Stained tissues were incubated in thermolysin (0.5 mg/ml; Sigma) in M199 for 30 minutes at 37uC and subsequently  inactivated in M199 medium containing 5% FBS. The sensory epithelium was carefully dissected from the underlying stromal cells using a 1 ml 26-Gauge 1/2 needle syringe. We tested several dissociation conditions using various concentrations of trypsin, commercially available Accutase and Accumax (Innovative Cell Technologies), as well as an enzyme-free cell dissociation solution (Millipore, S-004-C). The most optimal conditions were determined by systematic pilot experiments. For enzymatic dissociation of sensory epithelia for flow cytometry, 120 sensory epithelia were incubated in 100 ml enzyme cocktail containing 50% Accumax and 0.025% trypsin/EDTA (Sigma) in M199 for 7 minutes at 37uC. After 3 minutes incubation, the tissue was gently disrupted by trituration for three times using a 1 ml syringe equipped with a 20-Gauge needle. After additional incubation for 4 minutes at 37uC, single cells were carefully dissociated by gently triturating the sample three to four times using a 26-Gauge needle attached to a 1 ml syringe. To stop trypsin activity, 0.25% trypsin inhibitor (Worthington) was added.
For subsequent fluorescent activated cell sorting, the cell suspension was passed through a 40 mm cell strainer (BD Falcon) to remove cell clumps.

Immunocytochemistry
E18 chicken utricles and saccules were harvested in M199 (Gibco). For styryl dye uptake, the utricles were bathed in 10 mM FM1-43FX in M199, which is a fixable derivative of AM1-43 (Biotium) for 30 seconds and washed with M199 media (Invitrogen). The tissues were then transferred to 0.5 mg/ml thermolysin (Sigma) in M199 for 30 minutes at 37uC and then incubated in 5% serum for inactivation. Utricles and saccules were fixed with 4% paraformaldehyde in PBS for 30 minutes. The sensory epithelia were removed in 4% paraformaldehyde in PBS using a 1 ml 26-Gauge 1/2 needle syringe.

Fluorescence-activated cell sorting
Cells were analyzed using a BD Aria FACS cytometer. Control cells without AM1-43 staining that underwent the same conditions were first analyzed for background fluorescence activity. AM1-43 stained cells were then analyzed. Our gating strategy was as follows: 1) Exclude cells that take up the DNA intercalating dye propidium iodide, 2) Remove debris based on side scatter (SSC) and forward scatter (FSC) parameters, 3) Discard doublets based on FSC-height and FSC-area, and 4) Gate AM1-43-high cells (35.060.027%) and AM1-43-low cells (42.060.034%) for collection (see Fig. 3). A 100 mm flow cytometer nozzle size was used for all sorts in an effort to reduce cell damage. AM1-43-high and AM1-43-low populations were collected in lysis buffer (125 mM Tris-HCl, 50% glycerol, 4% SDS) in preparation for further analysis.

Mass spectrometry
After separation of the AM1-43-high and AM1-43-low populations by FACS sorting, proteins were enriched by trichloroacetic acid (TCA, Sigma) precipitation. Briefly, one volume of ice-cold 50% trichloroacetic acid was mixed with one volume of cell suspension after FACS sorting and incubated over night on ice. Precipitated protein was collected by centrifugation at 38,000 x g (Sorvall SS34) for 30 minutes at 4uC. The supernatant was aspirated and the remaining pellet was washed with two volume of ice cold 90% acetone while incubating at -20uC for 45 minutes. After a second centrifugation at 38,000 x g (Sorvall SS34) for 30 min at 4uC, the supernatant was again aspirated and the pellet was air-dried. To dissolve the pellet 30 ml 8 M urea containing 100 mM DTT was added and incubated at 95uC for 5 minutes. After addition of 30 ml 2x Laemmli Buffer (62.5 mM Tris-HCl, Figure 5. Categorization of subcellular localization and predicted cellular function of identified hair cell and non-sensory cell proteins. (A) All 634 identified hair cell and non-sensory cell proteins were classified according to gene ontology annotations and classified according to their subcellular localizations (upper bar graphs, before quantification). Taking into account the spectral counts of each identified protein resulted in slight changes of the distribution (lower bar graphs, after quantification). (B) Display of the gene ontology classifications of predicted cellular functions (upper bar graphs, before quantification) and after taking into account the spectral counts of each identified protein (lower bar graphs, after quantification). doi:10.1371/journal.pone.0066026.g005 25% glycerol, 2% SDS, 0.01% Bromphenol Blue, pH 6.8), to the TCA precipitated protein sample, followed by 10 minutes incubation at 95uC, the samples were immediately applied completely on one gel lane of a 4-20% Bis/Tris Gradient Gel (Invitrogen). The gel was coomassie stained and each gel lane fractionated into 8 gel bands. The gel bands were in-gel digested using trypsin (Promega) and protease max (Promega) as previously described [39]. Three independent experiments were conducted resulting in 6 gel lanes analyzed. The extracted peptides were dried to completion using a speed vac, after which they were reconstituted using a buffer of 0.2% formic acid, 2% acetonitrile, 97.8% water. The HPLC was an Eksigent nano2D (Eksigent), in which a self-packed 150 mM ID C18 column was used. The electrospray source was either a Michrom Advance operated at 600 nL/min or an Advion Nanomate which was operated at 450 nL/min. Two mass spectrometers were used, an LCQ Deca XP+ and a LTQ Orbitrap Velos (Thermo Fisher). Data acquisition was done in a data dependent fashion in which the top 3 (Deca) or the top 10 (Velos) most intense multiply charged peptide ions were selected for MSMS fragmentation. In total three independent data sets were generated (N = 3) on each cell type, therefore 48 LC-MS/MS runs were interrogated.
The data were extracted using a msconvert script to mzXML format after which was searched using a Sorcerer (SAGE-N) search station employing the Sequest algorithm. Both, the NCBI Gallus gallus as well as the ipi chicken databases were searched. The  (Table 1) compared to 170 proteins exclusively identified or enriched in non-sensory cells ( Table 2) (upper bar graphs, before quantification). Taking into account the spectral counts of each identified protein resulted in changes of the distribution (lower bar graphs, after quantification). (B) Display of the gene ontology classifications of predicted cellular functions of the proteins unique to hair cells and supporting cells (upper bar graphs, before quantification) and after taking into account the spectral counts of each identified protein (lower bar graphs, after quantification). doi:10.1371/journal.pone.0066026.g006 LCQ Deca XP+ data was searched with a 1.2 Da mass window and the LTQ Orbitrap Velos data were searched using a 50 ppm mass window. We allowed for the static modification of propionamide on Cysteine and variable modifications of Methionine oxidation, Lysine acetylation, Serine, Threonine and Tyrosine phosphorylation as well as Lysine ubiquitination.