In this study we developed a protocol to isolate the major cell types of the auditory and vestibular organs of newborn wild type mouse inner ears. We applied this approach to characterize the tissue- and cell type–specific transcriptome of the newborn mouse inner ear, followed by a computational analysis to identify co-expressed genes and regulators of cell fate. The cis-regulatory motif analysis was carried out using a de novo
motif discovery tool with no bias to any pre-selected TFs 
. Searching all possible DNA motifs, the analysis defined only two statistically significant motifs, corresponding to ZEB1/2 (enriched in promoters of genes whose expression was reduced in sensory cells) and c-Ets-1/2 (enriched in promoters of genes whose expression was elevated in blood cells). Most of the reported successful analyses using similar approaches were achieved in lower organisms (primarily yeast) in which transcriptional regulation is much simpler 
. Additional regulators are likely to be identified by sorting the cell types we studied to even more homogeneous subpopulations (e.g. separating hair cells from supporting cells).
The strength of this approach is further demonstrated by the ability to detect not only compartment specific regulators of cell fate (Zeb1/2, c-Ets1/2, miR-128, miR-9 and miR200b) but also miR-96, a miRNA which is expressed only in a subset of the sensory epithelial cells. A recent study by Lewis et al. identified putative targets of miR-96 by comparing expression profiles of whole auditory sensory tissues dissected from wild type and miR-96 mutant mice 
. Interestingly, the miR-96 putative targets identified by Lewis et al. and by our study represent two non-overlapping groups. Further analysis of the cell type–specific expression patterns of these putative targets reveals that while most of the targets identified by Lewis et al. are elevated and co-expressed with miR-96 in the same cell population (sensory epithelium), our analysis identified targets whose expression is reduced in the sensory epithelium (). This is consistent with the two functional effects of repression of targets by miRNAs: repression of leaky transcription, and buffering of transcriptional noise 
. In the first case, the miRNA and its targets are expressed in mutually exclusive cell types. In the second case, the miRNA functions to reduce fluctuation in the expression of its target due to transcriptional noise, and therefore it is co-expressed with its targets in the same cells. Since miR-96 is expressed in the inner-ear specifically in hair cells and as the hair cells consist of only a small fraction of the total number of cells in this organ, only targets that are hair cell-specific or hair cell-enriched (and hence targets that belong to the second group) are likely to be detected when extracting RNA from intact sensory tissues (for example, tissues that consists of a mix of epithelial and non-epithelial cells, or a mix of hair cells and supporting cells) as was done by Lewis et al. For the other group of targets, the effect of miR-96 is expected to be diluted by their expression in the other inner-ear compartments. On the other hand, in our study, using a cell type–specific analysis of wild-type inner ears, we could only identify targets of miR-96 that are expressed also outside the sensory epithelial cells (and hence targets that belong to the first group). We hypothesize that repeating the experiment described by Lewis et al. using a cell type–specific approach would likely identify both groups of complementary target genes.
Cell type–specific targets of miR-96.
We applied our method to analyze ears from Twirler mutant mice to determine if the observed inner ear malformations were associated with a disruption of wild-type cell type–specific gene expression profiles. The results revealed misregulation of epithelial and mesenchymal specific genes in the non-epithelial compartment. This finding is consistent with a misregulation of Zeb1-regulated gene expression in the Twirler mutant mice. Furthermore, a group of genes that encodes the extracellular matrices of the mouse inner ear – Otog, Tecta, Tectb
- was upregulated in the non-epithelial cells of the homozygous Twirler inner ears and harbored a binding site for Zeb1 in their promoter, suggesting that they are direct targets of the Zeb1 transcription factor in the developing inner ear mesenchyme. Of note, immunohistochemical analysis of inner ears from Zeb1
-null mice 
and their heterozygous or wild type littermate controls showed very subtle, if any, structural abnormalities in the ears of the Zeb1
-null mice (Data not shown). It is possible that Zeb2
may compensate for the loss of Zeb1
-null but not Tw/Tw
inner ear mesenchyme. This is consistent with the observation that Tw
is not a loss-of-function allele of Zeb1 
. The expression profiles of Twirler ears suggest that a pathologic disruption of epithelial and mesenchymal cell identities underlies the inner ear malformations observed in the Twirler mouse mutant, consistent with misregulation of the ZEB1 pathway. This could arise from a loss of mesenchymal cell identity leading to mesenchymal-epithelial transition (MET), a loss of epithelial cell identity leading to epithelial-mesenchymal transition (EMT), or a combination of these mechanisms. The exact mechanism leading to the misregulation of the ZEB1 pathway is still unknown, but likely results from loss of binding of one or more Myb proteins to a binding site disrupted by the Twirler Zeb1
. Therefore, further work is required in order to elucidate the exact mechanism of the Twirler mutation. A limitation of our approach is the utilization of CD326 to separate the epithelial from the non-epithelial compartment in the Twirler mutant mice, as CD326 itself could be regulated by Zeb1. Nevertheless, we were unable to identify other epithelial markers with a stable pattern of expression to separate epithelial and non-epithelial cells.
Ultimately, to prove direct regulation of gene expression by transcription factors, one has to demonstrate protein-DNA interactions. This is specifically challenging when working with tissues or small organs such as the inner ear sensory epithelia. Future development of techniques to perform large-scale whole transcriptome analysis of protein-DNA interactions using small numbers of cells will enable combining cell type–specific approaches with techniques such as chromatin immunoprecipitation-sequencing (ChIP-Seq) 
. Finally, implementation of our approach to study other mouse models for hearing loss will likely shed light on the molecular mechanisms downstream and upstream of many of the deafness genes.