Novel cell types and developmental lineages revealed by single-cell RNA-seq analysis of the mouse crista ampullaris

This study provides transcriptomic characterization of the cells of the crista ampullaris, sensory structures at the base of the semicircular canals that are critical for vestibular function. We performed single-cell RNA-seq on ampullae microdissected from E16, E18, P3, and P7 mice. Cluster analysis identified the hair cells, support cells and glia of the crista as well as dark cells and other nonsensory epithelial cells of the ampulla, mesenchymal cells, vascular cells, macrophages, and melanocytes. Cluster-specific expression of genes predicted their spatially restricted domains of gene expression in the crista and ampulla. Analysis of cellular proportions across developmental time showed dynamics in cellular composition. The new cell types revealed by single-cell RNA-seq could be important for understanding crista function and the markers identified in this study will enable the examination of their dynamics during development and disease.


Introduction
The vertebrate inner ear contains mechanosensory organs that sense sound and balance. In mammals, the cochlea senses sound, the saccule and utricle sense horizontal and vertical acceleration, respectively, and the cristae in each of the semicircular canals sense angular head movements, which is critical for maintaining balance and the vestibulo-ocular reflex (Highstein and Holstein, 2012). Cristae exhibit significant age-related degeneration of hair cells, particularly type I hair cells (Rauch et al., 2001;Lopez et al., 2005). Dysfunction of crista is therefore implicated in balance disorders and falls in the elderly, as well as in the pathology of Meniere's, benign paroxysmal positional vertigo and other balance disorders.
Whereas other sensory organs in the inner ear have been profiled by single-cell RNA-seq (scRNAseq), no studies to date have analyzed the crista ampullaris. Analysis of scRNA-seq data from the utricle, cochlea, and endolymphatic sac have revealed developmental transitions and rare cell types of possible importance to the etiology of hearing and vestibular disorders (Burns et al., 2015;Honda et al., 2017;Yamashita et al., 2018;Korrapati et al., 2019;Hoa et al., 2020;Ranum et al., 2019;Petitpré et al., 2018;Shrestha et al., 2018;Sun et al., 2018). The sensory epithelium of the crista includes type I and type II hair cells each having distinct morphology and synapses, support cells and glia (Desai et al., 2005). Five nonsensory cell types of the ampulla have been described morphologically: (1) transitional epithelial cells bordering the sensory epithelium, (2) periotic mesenchyme surrounding the ampulla epithelium, (3) dark cells of the ampulla, the osmiophilic and endolymph-producing cells, (4) melanocytes underlying the dark cells and (5) light cells, the osmiophobic cells of the planum semilunatum and the ampulla roof and wall (Kö ppl et al., 2018;Kimura et al., 1964;McLamb and Park, 1992;Tachibana et al., 1987;Nakai and Hilding, 1968;Bairati and Iurato, 1960;Ten Cate and Rarey, 1992). Understanding the cellular composition of the crista and the gene expression in each cell type is a critical step towards understanding inner ear development, function, and vestibulopathies.
In mammals, regeneration in the vestibular organs is slow, highly variable and limited to Type IIlike hair cells with immature hair bundle morphology (Kinoshita et al., 2019;Sayyid et al., 2019;Forge et al., 1993). An unexplained phenomenon of importance to the goal of therapeutic regeneration of hair cells in the vestibular system is the perinatal window of greater regeneration-competence in support cells (Slowik and Bermingham-McDonogh, 2013a). In adult cristae, a subset of support cells maintains competence to convert into hair cells in response to Notch inhibition (Slowik and Bermingham-McDonogh, 2013b), which implies heterogeneity of support cells in the crista.
To gain insights into the cellular composition of the crista during the perinatal window of greater regeneration-competence of support cells, we performed single-cell RNA-seq in crista ampullaris microdissected from E16, E18, P3, and P7 mice. We identified and validated many novel cell types, cell-specific markers, and developmental changes in cellular composition of the crista that are in addition to the support cells we initially aimed to study. Integrated analysis of the data provides novel insights into the pathways and transcription factors that may regulate cell-specific gene expression and functions in the crista as well as the cell types involved in vestibular disease.

Cellular diversity in the crista ampullaris
Cells were dissociated from crista ampullaris of mice at E16, E18, P3, and P7 and their gene expression was analyzed by scRNA-seq ( Figure 1a). Dimensional reduction was performed using UMAP to visualize transcriptional differences in cells (Figure 1b). Based on enrichment for known markers (Figure 1c and d), we identified major cell types as shown in Figure 1b. Clustering detects the major cell types at all four ages examined, although the proportions of some types change between E16 and P7. Cluster analysis identifies distinct cell groups enriched with known markers for hair cells, support cells, dark cells of the crista as well as nonsensory epithelial cells of the ampulla roof and wall, glia, otic mesenchymal cells, macrophages, and vascular cells (Figure 1b-d). Clustering also identifies transcriptionally distinct subtypes of cells. Analysis of the cellular subtypes, markers and dynamics for each major cluster is shown separately in Figures 2, 3, 4, 5, 6 and 7 and described below. The specific tissue for dissociation into single cells included the SE, transitional epithelium, the epithelial cells making up the walls of the ampulla, ampullar associated mesenchyme and the glial cells associated with the peripheral processes of the vestibular neurons. The cell bodies of the neurons were not included.

Heterogeneity and developmental changes in the sensory epithelium and glia
Clustering identifies six transcriptionally distinct clusters of support cells and hair cells (Figure 2a-c). Two clusters of hair cells show specific expression for known markers of vestibular type I and type II hair cell subtypes, Ocm (Simmons et al., 2010) and Anxa4 (McInturff et al., 2018), respectively (Figure 2c and g). Similarly, Ocm IF in P3 B6 crista identifies cells in the hair cell layer having calyxtype synapse and lacking Sox2nuclei ( Figure 2-figure supplement 1). Clustering also identified two support cell clusters that both express the known support cell markers Zpld1 (Vijayakumar et al., 2019) and Otog (El-Amraoui et al., 2001;Figure 2b-c); however, the clusterspecific expression of unique markers such as Id1 indicates heterogeneity in the support cell population (Figure 2c and g). The cells we label 'SC-HC transition' express a combination of support cell markers and hair cell markers (Figure 2c). The cells we label "NS-SC transition express some support cell markers such as Zpld1 but not others such as Otog (Figure 2c). RNA ISH data from the Allen Developing Mouse Brain Atlas (Miller et al., 2014) independently validates the localization of several markers in support cells and hair cells of the E15.5 B6 crista (   . Additionally, proportions of the hair cell and support cell clusters change between E18 and P7 ( Figure 2f). For example, hair cells and Srxn1 HI support cells increase as a percentage of total cells in the crista at P7 relative to E18, while Id1 HI support cells and transitional cells decrease (Figure 2f). Taken together, we surmised that developmental dynamics in the postnatal crista likely include hair cell recruitment from the immature support cell layer and corresponding changes in support cell composition.
To determine whether support cell clusters indeed represent distinct cell populations, we examined the localization of the cluster-specific markers Id1, Sox2, and Atoh1 in situ. Anti-Id1 IF demonstrates 20-40% of support cells in the crista (Figure 2i and l, Figure 2-figure supplement 5, Video 1, Video 2). Atoh1 IF detects~30 cells in the support cell layer in addition to some Myo7a + and Myo7acells in the hair cell layer in the P0 crista (see arrowheads in Figure 2j, counts in Figure 2m). Similar to the cellular dynamics shown by scRNA-seq (Figure 2f), manual cell counts in IF labeled cristae show the addition of~500 hair cells by P14 (Figure 2k),~40% decrease in Id1 + support cells by P22 ( Figure 2l) and~50% decrease in Atoh1 + cells in the support cell layer by P7 relative to P0 (Figure 2m).
Based on the known role of Notch-signaling in crista sensory development (Kiernan et al., 2001), expression of ligands, receptors and targets of the Notch-signaling pathway was of particular interest. We found enrichment of the Notch targets Hes1, Hey1, Jag1, Igfbp3, and Slitrk6 in the support cells and the SC-HC transition cells and decline in Notch targets in maturing hair cells (Figure 2c,d and h). Additionally, Hes1 and Jag1 decline in support cells by P3 (Figure 2d and h). To identify novel pathways and transcription factor activity associated with support cell maturation, we performed GSEA on transcriptional changes in support cells across developmental stages. By P7, support cells show increased enrichment for heat stress response pathway and the transcriptional targets of Hsf1/2 ( Figure 2h Hsp70 localizes to an apical structure extending from some support cells adjacent to hair cell stereocilia bundles (Figure 2-figure supplement 6b).
The FGF pathway has been shown by our lab and others to be important in the development of the sensory epithelium in the cochlea (Hayashi et al., 2008;Yang et al., 2019;Chang et al., 2004;Pirvola et al., 2000) and is implicated in crista development (Chang et al., 2004;Pirvola et al., 2000) but the Fgf ligands involved in canal and crista formation are unclear. We therefore looked at hair cells trajectory in cristae. Note that the trajectory between support cells and hair cells is continuous. (f) The relative contributions of support cell and hair cell subtypes to the composition of the support cell and hair cell clusters at the indicated developmental stages. Note that the contribution of the Id1 HI support cell subtype decreases as that of Srxn1 HI support cell subtype and hair cells increases. (g) 'Blendplots' show the overlap of marker specificity in UMAP. (h) shows differential expression of the Notch target Hes1 and the Hsp70 component Hspa1a across developmental stages in the support cell clusters. (i) Id1 (magenta) localization in a subset of Sox2 + (green) support cells in P0 B6 crista relative to phalloidin (cyan). (j) Atoh1 (green), Myo7a (blue) and Sox2 (green) IF in optical sections of P0 B6 mouse crista. Arrowheads indicate Atoh1 + /Sox2 + nuclei in the support cell layer. (k) Hair cell counts made manually in whole B6 cristae at P0 (n = 15), P14-21 (n = 7), 3 months (n = 30), and 6 months (n = 6). Note the addition of~500 hair cells. Asterisks indicate p<1Â10 À7 . (l) The Id1 + percentage of Sox2 + support cells at P0 (n = 6) and P22 (n = 5). (m) The number Atoh1 + cells in the Sox2 + support cell layer per crista at P0 (n = 5) and P7 (n = 5). Scale bars = 50 mm. The online version of this article includes the following figure supplement(s) for figure 2:       the expression of all 4 FGF receptors and all ligands in our single cell combined dataset. We find that Fgfs 1, 3, 7, 8, 9, 10, 16, and 21 are expressed in either the support cells, SC-HC transition cells or the hair cells ( Figure 2-figure supplement 7a and b). In the crista, Fgf21 marks the precursors of Type II hair cells, whereas Fgf7 marks the transition to the Type I hair cells and is also expressed by a subset of these cells. Fgf8 shows greater expression in Type I hair cells than Type II hair cells ( Figure 2-figure supplement 7a), similar to the pattern of Fgf8-CreER activity (Ratzan et al., 2020), and increasing expression along the SC-Type I transition. Fgfr1 is the predominant receptor with only sparse expression of Fgfr2 and no expression of Fgfr3 or Fgfr4 within the sensory/prosensory cells. Nonsensory epithelial cells express Fgfr2 (Figure 2-figure supplement 7a-b), as previously described (Pirvola et al., 2000), as well as Fgfr1. Fgfr4 is expressed in the mesenchyme cells ( Figure 2-figure supplement 7b), as previously described (Hayashi et al., 2010).
Ampulla mesenchyme also includes blood vessels and macrophages; many macrophages in the ampulla localize to blood vessels (Figure 5a). Clustering detects endothelial cells, pericytes and distinct subtypes of macrophages in the crista ampullaris ( Figure 1, Figure 7). Tlr2 HI and Adrb2 HI macrophage clusters are both present at all stages examined (Figure 7c), whereas macrophages enriched in Mki67 and other cell cycle markers were only present at P3 and P7. Tlr2 HI and Adrb2 HI macrophage subtypes may represent inflammatory states. For example, Tlr2 HI macrophages express Tlr2, Ccl3, Il1b and other cytokine-signaling ligands whereas Adrb2 HI macrophages are enriched in immediate-early targets including Egr1, Ier2, Jun, and Fos. RNA ISH images from the Allen Inst. independently validate the presence of a sparse population of Egr1 + cells in the loose mesenchyme surrounding the E15.5 ampulla (Figure 7-figure supplement 1). Expression patterns in endothelial cells versus pericytes (Figure 7d) are consistent with those in the Human Protein Atlas (Uhlén et al., 2015) and in scRNA-seq data from brain . Analysis of the proportions of macrophage and vascular cell clusters suggests macrophage dynamics in the E16-P7 window (Figure 7b

Comparative analysis of cochlear and crista hair cell and support cell expression
The hair and support cells of the cochlea and the crista have similarities but are specialized functionally, which suggests that similarities and differences can be found at the transcriptional level. To Video 2. Id1 immunofluorescence in P4 crista. Video shows a series of 1 micron optical sections per second imaged in P4 Hes5-GFP cristae from a colleagues' colony showing anti-Sox2 (white), anti-Id1 (magenta), Hes5-GFP (yellow), and Hoechst 33342 (cyan). https://elifesciences.org/articles/60108#video2 compare the gene expression profile of crista hair cells and support cells to those in the cochlea, we aligned a publicly available dataset of P7 cochlea cells from Kolla et al., 2020 to the P7 crista cells of the present study, then clustered the cells, identified cell type clusters based on known markers and performed differential expression analysis. UMAP and clustering suggest that crista cells including support cells and hair cells resemble cochlear cells (Figure 8a). Heatmaps show shared and specific markers for crista vs. cochlea support cells and hair cells identified by differential expression analysis (Figure 8b). Differential expression analysis identifies known cochlea-specific support cell markers Tecta and Tectb, the known crista-specific support cell marker Zpld1, additional markers identified in previous studies comparing cochlear hair cells and support cells to vestibular hair cells and support cells (Elkan-Miller et al., 2011;Scheffer et al., 2015;Giffen et al., 2019), as well as several novel crista-specific markers such as Agr3, Meis2 and Bricd5 in support cells and some immature hair cells and Pcdh20 and Cib3 in hair cells.

Expression of genes associated with vestibular dysfunction and disease in the crista ampullaris
For insights into the cellular players in vestibulopathies, in Figure 9-source data 1 we show cluster analysis of expression in crista ampullaris cells of genes associated with vestibular diseases and disorders (i.e. Meniere's disease, vertigo, motion sickness, circling, and hyperactivity in mice and relevant gene ontology terms). Clustering shows that many of the genes associated with vestibular disorders are expressed predominantly in support cells and hair cells (Figure 9 shows a subset, Figure 9source data 1 shows all vestibular disorder genes). However, some of these genes implicated in   Figure 9-source data 1 shows the full list of vestibular disorder-associated genes. Column color bars indicate the major cell types and subtypes. Row color bars denote the following: IMPC, genes mutated in mice that exhibit circling (cornflour blue) and hyperactivity (teal) or both (purple) phenotypes. Disorders, gene associations with Vestibular Figure 9 continued on next page vestibular dysfunction are also expressed in other cell types of the ampulla. For example, macrophages show enrichment for Cstb, Cxcl10,Atg5,Cyba,Hexb,Nfkb1,Tlr2,Pepd,and

Discussion
scRNA-seq is a powerful technology for studying cellular diversity and changes in developing tissues. Our results provide several new insights into the composition and development of the crista and identify markers for novel subtypes of support cells, nonsensory epithelial cells, macrophages, glia, and mesenchymal cells. Most notably, we found cell cluster-specific markers that predicted stereotypic gene expression domains not previously described in the ampulla epithelium and mesenchyme. Second, this analysis resolves novel developmental transitions at the transcriptional level including the separate lineages of type I and type II hair cells, the maturation of Scwann cells as well as evidence for nonsensory-support cell conversion and macrophage diversity and local proliferation. Third, we showed this dataset can be leveraged to identify the stage-and lineage-dependent transcriptional changes during the formation of hair cells in the crista. We have uploaded this dataset to GEO (GSE168901) for independent investigation and reanalysis as the informatics tools and research questions continue to evolve. . Our analysis provides evidence that type I and type II hair cells differentiate from a common immature hair cell population and we have identified differential gene expression in the type I and type II lineage including Sox21 and Insm1, respectively. Insm1 is a transcriptional repressor required for outer hair cell differentiation (Wiwatpanit et al., 2018) but its function in the vestibular system has not been examined to our knowledge. Our finding raises the question of whether Insm1 suppresses type I hair cell differentiation in developing type II hair cells of the crista similar to its role in the cochlea, where it represses inner hair cell-enriched genes in nascent outer hair cells.

Support cell subtypes and postnatal hair cell addition in the crista
Between E18 and P7, clusters of type I hair cells, type II hair cells and Srxn1 HI support cells increase as a percentage of sensory epithelial cells while Id1 HI support cells and transitional cells decrease. One possible explanation for the hair cell increase is an increase in hair cell differentiation during this period. We examined the first possibility by immunolabeling and manually counting hair cell numbers to confirm that hair cells indeed increase postnatally. These findings are consistent with Figure 9 continued dysfunction (forest green), susceptibility to ototoxicity (magenta) and Meniere's disease (maroon) were compiled from MSigDB, OMIM and Malacards databases. GWAS, hits in genome-wide association studies of motion sickness (green) and vestibular migraine (orange). Balance, genes in the GO term 'neuromuscular process controlling balance' (sky blue). IE morph, genes in the GO term 'inner ear morphogenesis' (khaki). GO devo., genes in GO terms 'stereocilium organization' (midnight blue) and 'otolith development' (lime green). The online version of this article includes the following source data for figure 9: Source data 1. Expression of genes associated with vestibular disease and dysfunction in the crista ampullaris. postnatal decline in immature hair cells numbers reported in the utricle (Rüsch et al., 1998). It follows that support cells should decrease. scRNA-seq showed decline in Id1 HI support cells, which we confirmed independently by manual counts of Id1 + nuclei in situ. An alternative explanation for these changes is that the Id1 HI support cells and Srxn1 HI support cells represent immature and mature support cells but RNA velocity does not indicate such a transition. These findings support that some support cells remain competent to undergo differentiation into hair cells and thus represent a potential cellular target for promoting hair cell regeneration through cellular reprogramming. In mammals, regeneration in the vestibular organs is slow, highly variable and limited to Type II-like hair cells with immature hair bundle morphology (Kinoshita et al., 2019;Sayyid et al., 2019;Forge et al., 1993). Mechanistically, support cell-hair cell conversion may relate to transient gene expression in the transitional cells. The transitional cell-enriched genes Isl1, Sox2, and Sox21 are known to regulate support cell-hair cell conversion in the cochlea (Yamashita et al., 2018;Hosoya et al., 2011;Dabdoub et al., 2008). Additionally, Notch targets Hes1, Hey1, Jag1, Igfbp3, and Slitrk6 (Campbell et al., 2016) were elevated in early postnatal support cells and transitional cells, then declined in maturing hair cells. The transient expression of Notch is consistent with the contrasting positive early role and later negative role of Notch-signaling in hair cell differentiation (Campbell et al., 2016;Hartman et al., 2010), the perinatal window of greater regeneration-competence in support cells (Slowik and Bermingham-McDonogh, 2013a) and earlier results demonstrating competence in a subset of support cells to convert into hair cells in response to Notch inhibition (Slowik and Bermingham-McDonogh, 2013b).
We have identified specific expression patterns of Fgf ligands in the crista and found that Fgf7 and Fgf21 show specificity to the two types of hair cells. Furthermore, we find Fgf7 in the transition to type I hair cells and Fgf21 in transition to type II cells. The roles of Fgf7 and Fgf21 in inner ear development and function are unknown as far as we know. Findings from other systems implicate Fgf7 in presynaptic organization. For example, Fgf7 promotes clustering of anti-synapsin + vesicles and neurite branching in cultured motorneurons (Umemori et al., 2004). In vivo, Fgf7 deletion in mice is associated with normal motor function but altered nociceptive responses in sensory neurons of the dorsal root ganglion (Liu et al., 2015). Furthermore, Fgf7 knockout mice have increased susceptibility to epilepsy associated with increased hippocampal neurogenesis and mossy fiber sprouting and decreased inhibitory vesicles in pyramidal neurons (Terauchi et al., 2010;Lee et al., 2012). In contrast to Fgf7, Fgf21-signaling through its receptor Fgr1 and coreceptor Klb promotes insulin sensitivity and glucose uptake (Kliewer and Mangelsdorf, 2019), which are broadly important for neurodevelopment and function. Given Fgf7 expression in crista type I hair cells and the evidence for its specific role in neural function, we are testing whether signaling of Fgf7 through its receptor Fgfr2 influences the vestibular afferents. Currently, we are also utilizing this dataset to identify candidate regulatory genes and pathways for cellular reprogramming to promote hair cell regeneration.
To our surprise, inducible Hsp70 IF is present in apical processes extending from a subset of support cells at P7 but not at P3. The anti-Hsp70 foci are adjacent both to hair bundles and the apex of a subset of support cells. It is unclear from which cells these foci originate. When we initially found upregulation of heat-shock response genes in support cells by P7, we hypothesized that heat-shock response was likely artifactual and caused by stress from dissociation. However, when we acutely dissect cristae, fix and stain, we find that Hsp70 indeed increases, suggesting that Hsp70 is upregulated by P7 in the crista. Upregulation of expression of heat shock response genes was observed at P7 in hair cells, support cells, nonsensory epithelial cells by scRNA-seq. Exosomal secretion of Hsp70 from support cells promotes hair cell survival in the utricle and cochlea (Breglio et al., 2020;May et al., 2013;Taleb et al., 2009). Additionally, there is an association of a single-nucleotide polymorphism in HSPA1A/HSP70-1 with Meniere's disease (Kawaguchi et al., 2008). The role of Hsp70 in inner ear development is an open question that could be addressed in future studies.
Novel cell types in the ampulla roof and possible functions scRNA-seq analysis resolved detailed expression profiles for transitional epithelial cells and dark cells and for novel cell types in the roof of the nonsensory ampulla epithelium. Further study will be needed to determine the precise relationships between the cell-specific gene expression pattern identified by cluster analysis and the morphological variations (e.g., microvilli, granules, cell shape, and osmiophilicity) identified by earlier histological analysis in ampulla epithelium. Regardless, these results provide insights into the development and function of the ampulla epithelium. Nonsensory ampulla epithelial cells and support cells express the peptidase Spint2, which inhibits the proteolytic activation of Hgf (Kawaguchi et al., 1997). Hgf-signaling is required for melanocyte invasion of the nonsensory cochlear epithelium (Shibata et al., 2016). By extension, Spint2 may act to inhibit melanocyte invasion of the ampulla epithelium through inhibition of Hgf/c-met-signaling in melanocytes.
The cellular heterogeneity found in the ampulla epithelium may reflect differences in developmental lineage. Cristae derive from the dorsal otocyst (Li et al., 1978), with sensory specification evident by E11.5-12 (Morsli et al., 1998). At E12-E13, outpockets of the dorsal otocyst termed the canal pouches remodel to form the tube-like epithelium of semicircular canals and ampullae. This process involves apical fusions of the central walls of the pouches to form bilayered epithelium termed the canal fusion plates, then basal detachment and resorptions of the epithelial cells in the canal fusion plates into the rims of the pouches to form the canals (Morsli et al., 1998;Martin and Swanson, 1993;Rakowiecki and Epstein, 2013;Salminen et al., 2000). These cellular movements that remodel the canal fusion plate into the semicircular canals are regulated through morphogenetic gradients in signaling of Wnt, Notch, Bmp and Netrin 1 (Ntn1) within the rims and plates (Kiernan et al., 2001;Rakowiecki and Epstein, 2013;Salminen et al., 2000;Chang et al., 2008). For example, Ntn1 expression in the canal fusion plates is required for basal detachment and resorption (Salminen et al., 2000). Expression of Ntn1 persists in the inner rim of the canals and in the roof of the ampulla. Thus, transcriptional heterogeneity within the canal and ampulla may relate to regional restrictions in signaling and expression in the canal rims and fusion plates. Single cell transcriptomic analysis of earlier embryonic cristae would likely resolve these important developmental transitions. scRNA-seq analysis also identified a Wnt3-specific expression in a cluster of cells in the roof of the ampulla. Wnt-signaling is present throughout the dorsal otocyst by E11.5 (Rakowiecki and Epstein, 2013;Noda et al., 2012), then persists in the vestibular epithelium and increases in response to injury . Whereas Wnt1 and Wnt3a expression in the hindbrain was demonstrated to be required for induction of otocyst development, other functional redundancies have hampered the identification of critical sources of Wnt for later Wnt-dependent processes  such as orientation of hair cell stereocilia bundles (Hua et al., 2014), the formation of the canal fusion plate (Rakowiecki and Epstein, 2013;Noda et al., 2012) and otic mesenchyme differentiation (Bohnenpoll et al., 2014). Thus, the Wnt3 + rooftop cells may act as a morphogen source for Frizzled-signaling. Regarding canal formation, Wnt3 is expressed coincidently with Nr4a3 and Ntn1 in the dorso-lateral E11.5 otocyst (Appendix 1-figure 1). Nr4a3/ Nor1 (Ponnio et al., 2002) and Ntn1 (Salminen et al., 2000) activity in the canal plate/canal inner rim may relate to Wnt3-signaling. Also of relevance to the question of regional Wnt activity in the crista is that Wnt-signaling regulators showed specificity for cell types in the ampulla (e.g. see in Appendix 1-figure 2 the Wnt inhibitor Wif1 in ampulla-proximal dense mesenchyme, Wnt7b in nonsensory epithelium, the Wnt activator Rspo3 in dark cells). The role of rooftop cells in cupula function and in vestibular disorders is unknown.

Schwann cell development and implications for the study of demyelinating disease
Glia in the cochleovestibular nerve are derived from the neural crest (Sandell et al., 2014) and otic placode (Xu et al., 2017). In mouse, glia are known to proliferate within the cochleovestibular nerve up to birth (Ruben, 1967) and in response to injury (Lang et al., 2011). Consistent with this, many glia express G2/M markers at E18 and these glial progenitors decline by P7. Furthermore, myelination markers such as Mbp and Mpz, Egr2 (Bosio et al., 1996;Jessen and Mirsky, 2008) increase along the trajectory extending from glial progenitors. The transcriptional dynamics in Schwann cell differentiation resolved by scRNA-seq include genes associated with congenital demyelinating diseases such as Pmp22 (Charcot Marie Tooth and Deafness Syndrome) and Plp1 (Pelizaeus-Merzbacher disease). Future studies could leverage the transcriptomic profile of Schwann cell development for insights into pathological processes such as Schwann cell dedifferentiation after injury and Schwann cell transformation in vestibular schwannoma.

Implications of mesenchymal cell diversity in the ampulla
Our analysis resolved detailed expression profiles for several previously unknown subtypes of mesenchymal cells. Specifically, we identified three types of mesenchyme: dense mesenchyme associated with the ampulla epithelium, dense mesenchyme near the boney otic capsule and, in between, loose mesenchyme containing vasculature. Cluster specificity correlated corresponded well to spatial specificity in ISH (Allen Atlas) of cluster-specific markers revealing stereotypic patterns of localization of the subtypes both along a spatial axis between ampulla epithelium and the cartilaginous otic capsule and in loose (Car3 HI ) versus dense (Car3 LO ) mesenchyme. The identification of spatially distinct mesenchymal cell subtypes has potential implications for understanding how mesenchyme supports crista development and function. Whether mesenchymal cell subtypes of the ampulla are altered in vestibular disease as the mesenchymal cell subtypes in the cochlea are in hearing loss (Hequembourg and Liberman, 2001) is another open question.
Macrophages show enrichment for several genes associated with vestibular disorders including Meniere's disease, which raises the question of whether these cells have important roles in vestibular disease. Furthermore, cluster analysis resolved two transcriptionally distinct subtypes of macrophages: Tlr2 HI /Ccl3 HI /Ccl4 HI /Nfkbib HI /Cxcl2 HI macrophages and Adrb2 HI /Dusp6 HI /Egr1 HI /Folr2 HI macrophages. Enrichment of immediately-early target genes was recently reported to be an experimental artifact in studies of microglia (Li et al., 2019), however, RNA ISH images from the Allen Institute corroborate the presence of a sparse Egr1 + cell population in the ampulla mesenchyme. Whether the expression signatures of ampulla macrophage subtypes relate to localization, morphology, inflammatory activation, and developmental origins will be interesting questions to explore. We also found macrophages enriched for cell cycle markers, raising the possibility that some macrophages form locally within the developing ampulla, in contrast those recruited after hair cell injury (Kaur et al., 2015).
In our dataset, we did not have sufficient numbers endothelial cells and pericytes to make fine distinctions. Clustering can make tenuous divisions in small groups of cells but these 'over-clustered' groups had no convincing specific markers. Enrichment for these cells and other rare cells prior to scRNA-seq may resolve novel subtypes as has been shown for endothelial cells and pericytes in other systems .

Conclusions
Our findings add to the fundamental understanding of the cellular diversity of the ampulla. Several novel subtypes of epithelial and mesenchymal cells were identified using scRNA-seq: Id1 + and Srxn1 HI support cells, support cells actively transitioning into hair cells, wall, roof and Wnt3 + rooftop cells in the nonsensory ampulla epithelium, a spectrum of mesenchymal cell subtypes showing stereotypic localization within the otic capsule, Tlr2 HI and Adrb2 HI macrophages and progenitors of both glial cells and mesenchymal cells. The roles of these new cells in vestibular development, function and disease have yet to be elucidated. The specific markers of cell subtypes will now allow the study of cellular dynamics in ampulla development and vestibular disease.

Mice
For scRNA-seq, timed pregnant C57BL/6J mice (B6; Jackson stock: 000664, RRID:IMSR_JAX: 000664) were purchased and aged to E18, P3, and P7. For the E16 sample, Sox2-GFP mice (Jackson stock: 07592) were bred to generate a timed pregnant litter. For most immunofluorescence (IF) studies, C57BL/6J mice were bred to generate timed pregnant litters. Two supplementary images show tissue collected opportunistically from Slc1a3-CreER:LNL-tTA:tetO-mAscl1-ires-GFP mice (Todd et al., 2020) and Hes5-GFP (Nelson et al., 2011) lines used in our colleagues' published research. To induce CreER, tamoxifen (1.5 mg in 100 mL of corn oil) was administered to adult mice daily for 5 days. Stages were verified by Theiler's criteria. Mice were housed in the University of Washington Department of Comparative Medicine. All procedures were reviewed and approved by the Institutional Animal Care and Use Committee of the University of Washington and performed in accordance with NIH guidelines.

Dissociation of cells
All three cristae, including the ampullae, were dissected from E16 mice (n = 6), E18 mice (n = 5), P3 mice (n = 8), and P7 mice (n = 12) in ice cold Hank's buffered salt solution (HBSS;Cat. No. 14025-92;Thermo Fisher Scientific;Waltham, MA). For the E18, P3, and P7 samples, half of the cristae were dissected away from the ampulla to increase the proportion of sensory cells. Scarpa's ganglion was not included. Dissected cristae and cristae with ampulla intact were then pooled (1:1) and dissociated by treatments with collagenase and papain as follows. First, cristae were digested in 2% collagenase IV (Worthington Biochemical; Lakewood, NJ) at 37˚C for 5 min. A 10% vol of FBS was added and cristae were washed in HBSS three times. Second, cristae were digested in 2% collagenase II (Worthington) at 37˚C for 30 min. As before, a 10% vol of FBS was added and cristae were washed in HBSS three times. Third, cristae were dissociated to single cells using the papain dissociation kit (Worthington, #LK003150) for approximately 1 hr with trituration every 10 min. then stopped with the addition of ovomucoid and spun per the manufacturer's instructions. Cells were strained through the cell strainer and counted with a hemocytometer. A total of 7000 cells were then input into the 10X protocol.

Tissue isolation and immunofluorescence and 3D imaging
Temporal bones were fixed overnight at 4˚C in 4% paraformaldehyde (Cat. No. 15710; Electron Microscopy Sciences; Hatfield, PA) diluted in phosphate-buffered saline (PBS; Cat. No. BP399; Thermo Fisher Scientific). Tissues were washed 3 Â~30 min in PBS. Temporal bones in some experiments were decalcified (167 mM/5% EDTA/PBS for five days at 4˚C with daily exchanges), embedded in 4% agarose and sectioned at 200 mm with a vibratome. For whole mounts, ampullae were microdissected from the temporal bones. To uncover the crista sensory epithelium, the roof and wall epithelium of the ampulla were dissected away from cristae. To remove otoliths that stuck to cristae during dissection, cristae were decalcified in 167 mM (~5%) EDTA/PBS overnight at 37˚C. For experiments using goat anti-Wnt3 or mouse anti-Myo7a, antigen retrieval was performed in sodium citrate buffer (10 mM Sodium citrate, 0.05% Tween 20, pH 6.0) for 45 min in a vegetable steamer. Cristae and vibratome sections were blocked (10% donkey serum/0.5% Triton-X-100/PBS) overnight at 4˚C. Tissues were then incubated overnight in primary antibodies ( Table 1) diluted in block solution at room temperature. After 3 Â 1 hr washes in 0.5% Triton-X-100/PBS, tissues were incubated at least 1 hr in block, then overnight in secondary antibodies diluted in block at room temperature. After 3 Â 1 hr washes in 0.5% Triton-X-100/PBS, tissues were mounted (Fluoromount-G; Cat. No. BP399; Southern Biotech; Birmingham, AL) under No. one cover slips and imaged using a Zeiss LSM880 with Airyscan. Linear adjustments to micrographs were made using 'Levels' in Adobe Photoshop v21.1.1 to increase signal:noise.

RNA in situ hybridization
Digoxigenin (DIG)-labeled Wnt3 probe spanning 62-1504 bp of the cDNA (NM_009521.2) was prepared according to the manufacturer's manual for DIG-11-UTP (Cat. No. 11209256910; Sigma) and the hybridization was performed on paraffin sections of inner ear as previously described (Hayashi et al., 2007) with differences in washes post-hybridization. Here, the following washes were performed at 65˚C after the hybridization: 60 min in 0.5X SSC/50% formamide/1% SDS, 90 min in 0.5X SSC/50% formamide/0.1% Tween 20 and 2 Â 30 min washes in 0.5X SSC/0.1% Tween 20. ISH was then visualized using alkaline phosphatase-conjugated sheep anti-DIG Fab fragments (Cat. No. 11093274910;Sigma) and NBT/BCIP (Cat. No. B1911; Sigma). The RNA ISH images shown in the figure supplements are from the Allen Developing Mouse Brain Atlas (Miller et al., 2014). Supplementary file 2 lists the URLs of the source images.

Morphometric quantitation and manual cell counting
For support cell and hair cell counts, total nuclei expressing specific markers were manually counted using Fiji/ImageJ v2 (Schindelin et al., 2012) as described previously (Wilkerson et al., 2018). Total hair cell counts made in P0 B6 cristae ( Figure 2) are compared to our previously published >P14 B6 dataset (Wilkerson et al., 2018). For hair cell counts, anterior and posterior cristae were grouped together and counts represent total hair cell nuclei per crista. For pigmented cell area measurements in Fiji, 'Threshold' was used to isolate the pigmented cell signal and the area above threshold measured. To test for dependence of pigmented area, Id1 + support cell counts and Atoh1 + cells in the support cell layer on age, two-sided Student's t-tests (alpha = 0.05) were performed using 't.test' in 'stats v3.6.2' in RStudio v1.2.5033. To test for crista hair cell count dependence on age and cristatype, two-way ANOVA and multiple linear regression analysis were performed using 'aov' and 'TukeyHSD' in 'stats v3.6.2' in RStudio v1.2.5033. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication. . Supplementary file 2. RNA ISH images from the Allen Developing Mouse Brain Atlas.

Author contributions
. Transparent reporting form

Data availability
All RNA-Seq data has been ave been deposited in NCBI's Gene Expression Omnibus (Edgar et al., 2002) and are accessible through GEO Series accession number# GSE168901.
The following dataset was generated: