Sox2 and FGF20 interact to regulate organ of Corti hair cell and supporting cell development in a spatially-graded manner

The mouse organ of Corti, housed inside the cochlea, contains hair cells and supporting cells that transduce sound into electrical signals. These cells develop in two main steps: progenitor specification followed by differentiation. Fibroblast Growth Factor (FGF) signaling is important in this developmental pathway, as deletion of FGF receptor 1 (Fgfr1) or its ligand, Fgf20, leads to the loss of hair cells and supporting cells from the organ of Corti. However, whether FGF20-FGFR1 signaling is required during specification or differentiation, and how it interacts with the transcription factor Sox2, also important for hair cell and supporting cell development, has been a topic of debate. Here, we show that while FGF20-FGFR1 signaling functions during progenitor differentiation, FGFR1 has an FGF20-independent, Sox2-dependent role in specification. We also show that a combination of reduction in Sox2 expression and Fgf20 deletion recapitulates the Fgfr1-deletion phenotype. Furthermore, we uncovered a strong genetic interaction between Sox2 and Fgf20, especially in regulating the development of hair cells and supporting cells towards the basal end and the outer compartment of the cochlea. To explain this genetic interaction and its effects on the basal end of the cochlea, we provide evidence that decreased Sox2 expression delays specification, which begins at the apex of the cochlea and progresses towards the base, while Fgf20-deletion results in premature onset of differentiation, which begins near the base of the cochlea and progresses towards the apex. Thereby, Sox2 and Fgf20 interact to ensure that specification occurs before differentiation towards the cochlear base. These findings reveal an intricate developmental program regulating organ of Corti development along the basal-apical axis of the cochlea.


Author summary
The mammalian cochlea contains the organ of Corti, a specialized sensory epithelium populated by hair cells and supporting cells that detect sound. Hair cells are susceptible to injury by noise, toxins, and other insults. In mammals, hair cells cannot be regenerated a1111111111 a1111111111 a1111111111 a1111111111 a1111111111

Introduction
The inner ear contains six sensory organs required for the senses of hearing and balance. The cochlea, a snail-like coiled duct, is the auditory organ. It contains specialized sensory epithelia, called the organ of Corti, composed of hair cells (HCs) and supporting cells (SCs). In mammals, this sensory epithelium is elegantly patterned, with one row of inner hair cells (IHCs) and three rows of outer hair cells (OHCs), separated by two rows of pillar cells forming the tunnel of Corti. Each row of OHCs is associated with a row of supporting cells called Deiters' cells. Here, we refer to pillar cells and Deiters' cells collectively as SCs.
Organ of Corti development has been described as occurring in two main steps: prosensory specification and differentiation [1]. During prosensory specification, proliferative progenitors at the floor of the developing cochlear duct are specified and then exit the cell cycle to form the postmitotic prosensory domain. Here, we define specification to be a process that makes progenitors competent to differentiate. We also use cell cycle exit as a marker for specified cells in the prosensory domain (prosensory cells). During differentiation, prosensory cells differentiate into both HCs and SCs [2]. Interestingly, cell cycle exit, marking the completion of specification, and initiation of differentiation occur in waves that travel in opposite directions along the length of the cochlear duct. At around embryonic day 12.5 (E12.5) in the mouse, progenitors begin to exit the cell cycle and express the cyclin-dependent kinase inhibitor CDKN1B (p27 Kip1 ) in a wave that begins at the apex of the cochlea (the cochlear tip) and reaches the base of the cochlea by around E14.5 [3,4]. Afterwards, the specified prosensory cells begin differentiating into HCs and SCs in a second wave that begins at the mid-base at around E13.5, and spreads quickly to the rest of the base and to the apex over the next few days [1]. Thus, while prosensory specification occurs in an apical-to-basal gradient, differentiation occurs in a basal-to-apical gradient. Notably, while the basal end of the cochlear duct differentiates immediately after prosensory specification, the apical end has a longer time between specification and differentiation, providing a larger "temporal buffer" for apical development. The spiral ganglion, containing neurons that synapse with HCs, has been shown to be important for this delay in apical differentiation, via inhibitory Sonic Hedgehog (SHH) signaling [5][6][7][8].
The transcription factor Sox2 is one of the earliest markers of prosensory cells [9,10]. Mice with specific Sox2 hypomorphic mutations that affect inner ear expression have hearing impairment due to decreased HC and SC number, while mice with inner ear-specific Sox2 null mutations are completely deaf and have no HCs or SCs [11,12]. Genetic experiments show that Sox2 is both necessary and sufficient for prosensory specification. Absence of Sox2 expression leads to the loss of Cdkn1b expression at E14, a marker for the prosensory domain [12], while ectopic Sox2 expression in cochlear nonsensory epithelium can induce ectopic sensory patches [13][14][15].
The Fibroblast Growth Factor (FGF) signaling pathway also plays vital roles in organ of Corti development [16]. Studies utilizing cochlear explants showed that inhibition of FGF signaling prior to and during stages of HC and SC differentiation results in decreased HC and SC number [17]. Signaling through FGF receptor 1 (FGFR1), in particular, is essential during this process. Conditional deletion of Fgfr1 (Fgfr1-CKO) in the developing cochlear epithelium resulted in dramatically reduced HC and SC number [18][19][20]. This has been attributed to decreased Sox2 expression in the prosensory domain of Fgfr1-CKO mice, leading to a defect in prosensory specification [19].
FGF20 has been hypothesized to be the FGFR1 ligand during organ of Corti development. Both in vitro inhibition of FGF20 with an anti-FGF20 antibody [17] and in vivo knockout of Fgf20 (Fgf20-KO) [21] led to decreased HC and SC number, similar to the Fgfr1-CKO phenotype. However, the Fgf20-KO phenotype is clearly not as severe as that of Fgfr1-CKO. Almost all OHCs and some IHCs are missing in Fgfr1-CKO mice [19], while only 2/3 of OHCs are missing in Fgf20-KO mice, without any loss of IHCs [21]. This suggests that another FGF ligand may be redundant with and compensating for the loss of FGF20, the identity of which is currently unknown.
Another difference between Fgfr1-CKO and Fgf20-KO mice is the proposed mechanism accounting for the decrease in HCs and SCs. Interestingly, unlike in Fgfr1-CKO mice, Sox2 expression in the prosensory domain is not disrupted in Fgf20-KO mice [19,21]. Rather, FGF20 seems to function during HC and SC differentiation. These differences between the Fgfr1-CKO and Fgf20-KO phenotypes and their relationship with Sox2 suggest that FGF20/ FGFR1 signaling has a more complex and as yet unexplained role during organ of Corti development.
Here, we hypothesize that FGFR1 signaling has functions in both steps of organ of Corti development: an earlier role in prosensory specification that involves Sox2, and a later role in the initiation of differentiation. We provide evidence that FGF20 regulates differentiation but not specification. Moreover, while Fgfr1 functions upstream of Sox2, Fgf20 is downstream of Sox2. We further show that Sox2 and Fgf20 genetically interact during organ of Corti development. Interestingly, downregulation of both genes leads to the loss of HCs and SCs preferentially towards the outer compartment and the basal end of the cochlear duct. To explain the more severe basal phenotype, we provide evidence that Sox2 regulates the timing of prosensory specification, while Fgf20 regulates the timing of differentiation. As these two steps occur along a developmental pathway, we hypothesize that prosensory specification must occur prior to differentiation. In Sox2 hypomorphic mice, prosensory specification is delayed, while in Fgf20-KO mice, the onset of differentiation occurs prematurely. When combined, these two defects led to differentiation attempting to initiate prior to the completion of specification towards the basal end of the cochlear duct. These results define unique functions of and complex interactions among FGF20, FGFR1, and Sox2 during organ of Corti development and highlight the potential importance of the timing of specification and differentiation along different regions of the cochlear duct.
We examined the cochleae at P0 (Fig 1A and 1B) and quantified the length of the cochlear duct and the total number of IHCs, OHCs, and SCs (Fig 1C-1F), as well as the number of cells along the basal, middle, and apical turns of the cochlear duct (S1A-S1C Fig). Refer to Fig 1G for the positions of basal, middle, and apical turns along the cochlear duct. We identified HCs based on Phalloidin labeling and SCs based on Prox1/Sox2 labeling. IHCs and OHCs were distinguished based on location relative to p75NTR-labeled inner pillar cells (IHCs are neural, or towards the center of the coiled duct; OHCs are abneural).
In both Fgf20-KO and Fgfr1-CKO cochleae, there were gaps in the sensory epithelium that lacked HCs and SCs along the entire cochlear duct. Quantitatively, Fgf20-KO cochleae had a 6% reduction in cochlear length compared to control (Fgf20 +/-) cochleae, while Fgfr1-CKO cochleae had a 28% reduction compared to control (Fgfr1 Cre/+ ;Fgfr1 flox/+ ). Fgf20-KO did not have a significant reduction in the number of IHCs, while Fgfr1-CKO cochleae had a 40% reduction. Fgf20-KO cochleae only had a 76% reduction in the number of OHCs, while Fgfr1-CKO cochleae had almost a complete lack of OHCs, a 97% reduction. For SCs, Fgf20-KO cochleae had a 59% reduction, while Fgfr1-CKO cochleae had an 84% reduction. These patterns persisted when HC and SC numbers were normalized to cochlear length. These results were all consistent with previous studies [19,21] and showed that the Fgfr1-CKO phenotype is more severe than the Fgf20-KO phenotype in cochlear length and in the number of HCs and SCs. We hypothesize that during organ of Corti development, there is an additional FGFR1 ligand that is partially redundant with FGF20.
Notably, while the total number of IHCs was decreased in Fgfr1-CKO cochleae, the decrease was only observed in the basal and middle turns of the cochlea, not in the apical turn (S1A Fig). In addition, the number of IHCs normalized to cochlear length was slightly increased in Fgf20-KO cochleae (Fig 1D), and this increase was only prominent in the middle and apical turns of the cochlea, but not in the basal turn (S1A Fig). The increase in IHCs could be explained by the shortened cochlear duct length in Fgf20-KO mice. No such basal/middle/ apical turn discrepancies existed in the number of OHCs or SCs in either genotype (S1B and S1C Fig).
Our previous studies also noted that the apical tip of Fgf20-KO cochleae has delayed differentiation relative to control at E16.5 and P0, but catches up by P7 [21]. We confirmed this result, finding that at P0 in control cochleae, sensory epithelium at the apical tip has begun to differentiate, based on phalloidin and p75NTR expression, while in Fgf20-KO cochleae, there was no sign of differentiation at the apical tip. There was a similar delay in differentiation at the apical tip of Fgfr1-CKO cochleae relative to control (S1E Fig). Refer to S1D Fig for the location of the apical tip.

FGFR1 but not FGF20 regulates Sox2 expression
Next, we examined Sox2 expression in Fgf20-KO and Fgfr1-CKO cochleae at E14.5 by RNA in situ hybridization and immunofluorescence. In control cochleae, Sox2 mRNA and protein were highly expressed in the prosensory domain (Fig 2A, refer to Fig 2C). The expression of Sox2 was not changed in Fgf20-KO cochleae compared to control; however, it was noticeably decreased in Fgfr1-CKO cochleae (Fig 2A), in agreement with previous findings [19][20][21]. This indicates that FGFR1 has an additional role, independent of FGF20, in regulating Sox2, which is required for prosensory specification [12]. Similar to Sox2, CDKN1B expression in the prosensory domain is also regulated by FGFR1, but not by FGF20 [18,19,21]. We confirmed these results, finding that while CDKN1B expression was not changed in Fgf20-KO cochleae at E14.5 relative to control, it was dramatically downregulated in Fgfr1-CKO cochleae (Fig 2B). This is consistent with the role of Sox2 in regulating CDKN1B expression [12]. We hypothesize that a yet unidentified FGF ligand (in addition to or independent of FGF20) signaling via FGFR1 regulates Sox2 expression (and therefore CDKN1B expression) during prosensory specification, while FGF20 signaling via FGFR1 regulates differentiation ( Fig 2D).
We also wanted to confirm that FGF20 signals to epithelial FGFR1 at around the initiation of differentiation. To do so, we examined the expression of Etv4 (also known as Pea3) and Etv5 (also known as Erm), two well-established downstream effectors of FGF signaling [23], by in situ hybridization. The expression of these two genes are downregulated with FGF signaling inhibition in E14 cochlear explants [17]. At E14.5, there were two domains of Etv4 and Etv5 expression in control cochleae: the prosensory domain and the outer sulcus (S2A Fig, brackets). The outer sulcus is the region of the cochlear epithelium abneural to the prosensory domain at E14.5. In Fgf20-KO cochleae, expression of both genes was not detected in the prosensory domain. In Fgfr1-CKO cochleae, expression of both genes was similarly not detected in the prosensory domain. Expression of Etv4 and Etv5 in the outer sulcus was not affected in Fgf20-KO and Fgfr1-CKO cochleae (S2A Fig). These results confirm that FGF20 signals through epithelial FGFR1 in the prosensory domain.
Previous studies have also reported a decrease in proliferation in Kölliker's organ (neural to the prosensory domain, S2B Fig) in Fgfr1-CKO cochleae [20]. We replicated this result by examining EdU (5-ethynyl-2'-deoxyuridine) incorporation at E14.5. Fgfr1-CKO mice had a complete lack of EdU-incorporating Kölliker's organ cells, while Fgf20-KO mice did not show a decrease in EdU incorporation (S2B Fig). This finding is also consistent with an additional FGF ligand signaling via FGFR1, likely at an earlier stage. We do not know whether the proliferation defect in Kölliker's organ contributes to the reduction in HC and SC number in Fgfr1-CKO mice.

Genetic rescue of the Fgf20-KO phenotype suggests that FGF20 is required for differentiation
We have previously shown that recombinant FGF9, which is biochemically similar to FGF20 with similar receptor binding specificity [23,24], is able to rescue the loss of HCs and SCs in Fgf20-KO explant cochleae [21]. Interestingly, while treatment with FGF9 at E13.5 and E14.5 was able to rescue the Fgf20-KO phenotype, treatment at E15.5 was not. This temporal rescue specificity suggests that FGF20 signaling is required for the initiation of HC and SC differentiation.
Initially, pregnant dams were fed a Dox diet from E13.5 to E15.5 and pups were harvested at P0 to examine HC and SC development. As expected, Dox treatment itself did not appear to affect HC or SC development in Fgf20-het and Fgf20-null cochleae, both of which showed the expected phenotypes (Fig 3A and 3B). Ectopic expression of FGF9 during these stages also did not affect HC or SC development in Fgf9-OA cochleae, showing that excess FGF20/FGF9 was not sufficient to produce ectopic HCs and SCs. Importantly, ectopic expression of FGF9 resulted in a full rescue of the number and patterning of HCs and SCs in Fgf9-rescue pups. The organ of Corti in these rescue pups had one row of IHCs, three rows of OHCs, and five rows of SCs throughout the entire length of the cochlear duct, without any gaps (Fig 3A and  3B). This shows that FGF20/FGF9 signaling at E13.5-E15.5 is sufficient for HC and SC differentiation. The quantified results from all of the rescue experiments are summarized in Fig 3C, where the number of OHCs and SCs are represented as a percentage of that of Fgf20-het mice treated with the same Dox regimen. All of the quantified data are presented in S3 Fig. To more precisely determine the timing of rescue sufficiency, we fed pregnant dams Dox for a period of 24 hours starting at E13.5, E14.5, or E15.5 (see S3 Fig for schematic of Dox regimens). With E13.5 Dox, patterning and OHC number in the basal turn of the cochlea were completely rescued in Fgf9-rescue mice ( Fig 3A). However, OHC number in the middle and particularly the apical turns were only partially rescued, resulting in regions with two rows of OHCs instead of three. For instance, in the apical turn, OHC number was restored to 81% of Fgf20-het mice, which is statistically significantly increased compared to Fgf20-null, but also statistically significantly decreased compared to Fgf20-het, indicating partial rescue ( Fig 3C). With E14.5 Dox, patterning and OHC number in the middle and apical turns were completely rescued. However, OHC number in the basal turn was not completely rescued, with regions of one or two rows of OHCs, instead of three. With E15.5 Dox, patterning and OHC number was not rescued in the basal and middle turns, as gaps still formed between islands of HCs ( Fig   Fig 2. FGFR1 but not FGF20 regulates Sox2 expression. (A) Sections through the middle turn of E14.5 cochlear ducts from Fgf20 +/-, Fgf20 -/-, and Foxg1 Cre/+ ;Fgfr1 flox/+ , Foxg1 Cre/+ ;Fgfr1 flox/mice. RNA in situ hybridization (top) and immunofluorescence for Sox2 (red, bottom), which is expressed in the prosensory domain at this stage. Refer to schematic in (C). Samples are representative of n = (top) 3, 3, 3, 3; (bottom) 4, 4, 3, 3. (B) Immunofluorescence for CDKN1B (green) in sections through the basal, middle, and apical turns of E14.5 Fgf20 +/-, Fgf20 -/-, and Foxg1 Cre/+ ;Fgfr1 flox/+ , Foxg1 Cre/+ ;Fgfr1 flox/cochleae. Samples are representative of n = 5, 5, 3, 3. (C) Schematic of a cross section through the middle turn of the E14.5 cochlear duct, showing the location of the prosensory domain (PD). Neural indicates the side of the duct towards the spiral ganglion cells; abneural indicates away. (D) A model of genetic pathways during organ of Corti development. Ligand X/FGFR1 signaling regulates Sox2 expression during prosensory specification; FGF20/ FGFR1 signaling regulates differentiation. Ligand X may include FGF20, along with another functionally redundant ligand. DAPI, nuclei (blue). Scale bar, 100 μm. See also S2 Fig. https://doi.org/10.1371/journal.pgen.1008254.g002 3A). However, OHC number in the apical turn was partially rescued, with two or three rows of OHCs not separated by gaps towards the tip of the apex. In all of these experiments, the rescue of SCs followed the same pattern as that of OHCs ( Fig 3B).
These rescue results show that FGF20/FGF9 is sufficient for OHC and SC differentiation in the basal turn of the cochlea at E13.5, in the middle and apical turns at E14.5-E15.5, and in the tip of the apical turn at E15.5. Since the initiation of HC and SC differentiation occurs in the base/mid-base of the cochlea at E13.5 and progresses apically over the next few days, these results strongly imply that FGF20 functions during the initiation of differentiation, rather than prosensory specification, consistent with our model (Fig 2D).

Decrease in Sox2 expression results in similar phenotypes as disruptions to FGFR1 signaling
Our results and previous findings suggest that FGFR1 regulates prosensory specification via Sox2 [19]. Mice with an inner ear-specific Sox2 hypomorphic mutation (Sox2 Ysb/Ysb , see below) have defects in prosensory specification, accounting for a small loss of HCs and SCs, whereas mice with inner-ear specific Sox2 null mutations have a complete lack of prosensory specification and a complete absence of sensory epithelium [12]. To examine how much the reduction in Sox2 expression in Fgfr1-CKO cochlea contributes to the phenotype at P0, we combined the Sox2 -(Sox2 constitutive null) and Sox2 Ysb alleles to closely examine the effects of reduction in Sox2 expression on organ of Corti development, on a similar genetic background as our Fgf20-KO and Fgfr1-CKO mice. We hypothesized that if Fgfr1 acts upstream of Sox2, then reducing Sox2 expression should at least partially recapitulate the Fgfr1-CKO cochlea phenotype. The Sox2 Ysb allele is a regulatory mutant in which transgene insertion in chromosome 3 disrupts some otic enhancers, resulting in hypomorphic Sox2 expression in the inner ear [11,12].
We generated a Sox2 allelic series of mice with the following genotypes, in order of highest to lowest levels of Sox2 expression: Sox2 +/+ (wildtype), Sox2 Ysb/+ , Sox2 Ysb/Ysb , and Sox2 Ysb/-. In this allelic series, decrease in Sox2 expression had a dose-dependent effect on cochlea length at P0 (Fig 4A-4C). Sox2 Ysb/+ cochleae had a 6% reduction in length compared to wildtype (although not statistically significant), Sox2 Ysb/Ysb cochleae had a 24% reduction, and Sox2 Ysb/had a 46% reduction. Sox2 Ysb/+ organ of Corti developed relatively normally, with three rows of OHCs and one row of IHCs ( Fig 4A). Interestingly, there were occasional ectopic IHCs neural (inner) to the normal row of IHCs, especially in the middle and apical turns of the Sox2 Ysb/ + cochlea (Fig 4A, arrowheads). However, there was no significant increase in IHC number (total or normalized to length) compared to wildtype cochleae ( Fig 4D). The Sox2 Ysb/Ysb cochlea appeared much more abnormal, with gaps in the sensory epithelium that lacked HCs and SCs in the basal turn ( Fig 4A and 4B), similar to what was observed previously [12]. Moreover, at the base, in the sensory islands between the gaps, there were often four rows of OHCs and six rows of SCs. In the middle and apical turns, there were the normal three rows of OHCs and five rows of SCs. There were also numerous ectopic IHCs throughout the middle and apical turns, sometimes forming an entire second row of cells (Fig 4A), resulting in increased number of IHCs in the middle turn compared to wildtype (S4A Fig). However, the total and length-normalized number of IHCs in Sox2 Ysb/Ysb cochleae did not significantly differ from that of wildtype cochleae ( Fig 4D). In terms of OHCs, Sox2 Ysb/Ysb cochleae exhibited a 40% decrease in total number compared to wildtype cochleae ( Fig 4E). This decrease was not quite as severe when normalized to cochlear length (21% decrease). Strikingly, Sox2 Ysb/cochleae lacked almost all HCs and SCs, except in the apical turn (Fig 4A and 4B). The decrease in OHC number (93%) in Sox2 Ysb/cochleae compared to wildtype was more severe than the decrease in IHC number (75%). Notably, IHC number was significantly decreased in the basal and middle turns, but not in the apical turn (S4A Fig). OHC number was significantly decreased throughout all three turns (S4B Fig). In all of these genotypes, the number of SCs followed the pattern of loss of OHCs ( Fig 4F and S4C Fig). Interestingly, while Sox2 Ysb/cochleae almost completely lacked HCs and SCs in the basal and middle turns, in 7 of 11 Sox2-Ysb/cochleae examined, one or two small islands of HCs or SCs were found at the basal tip Overall, these results showed that the basal end of the cochlea is more sensitive to the loss of Sox2 expression than the apical end. Furthermore, while both IHCs and OHCs were affected, OHCs were more sensitive to decrease in Sox2 expression than IHCs. Importantly, both of these features were found in Fgfr1-CKO cochleae, where the decrease in IHCs was only found in the basal and middle turns and there were almost no OHCs along the entire cochlear duct (S1A and S1B Fig). Therefore, we conclude that decrease in Sox2 expression, leading to defects in prosensory specification, could account for the Fgfr1-CKO phenotype. Furthermore, the decrease in Sox2 expression could also account for the difference in severity between the Fgf20-KO and Fgfr1-CKO phenotypes, since Fgf20-KO cochleae, which had normal Sox2 expression, did not have a decrease in the number of IHCs, unlike Fgfr1-CKO and Sox2 Ysb/cochleae.

Decrease in levels of Sox2 expression delays prosensory specification
We sought to determine why a decrease in Sox2 expression more severely affected the basal end of the cochlear duct. Initially, we examined Sox2 expression at E14.5. As expected, Sox2 This decrease in expression was not more severe at the basal turn of the cochlear duct, relative to the middle and apical turns, suggesting that the more severe basal phenotype in Sox2 Ysb/cochleae cannot be explained by differential Sox2 expression. Similarly, CDKN1B expression was downregulated in the prosensory domain of Sox2 Ysb/cochleae, consistent with previous studies [12]. Interestingly, the decrease in expression was also not more severe at the basal turn relative to the middle and apical turns (S5B Fig). Using CDKN1B as a marker of prosensory specification, this suggests that the more severe basal phenotype also cannot be explained by differential regulation of prosensory specification along the length of the cochlea.
As described in the introduction, the wave of cell cycle exit (marking the completion of prosensory specification) and the wave of differentiation travel in opposite directions along the cochlear duct during development, resulting in the basal end of the cochlear duct differentiating immediately after specification. The apical end, meanwhile, exhibits a delay in differentiation, resulting in a longer temporal buffer between specification and differentiation. In this developmental pathway, specification must be completed prior to the initiation of differentiation. We reasoned, therefore, that disruptions to the timing of prosensory specification will preferentially interfere with basal sensory epithelia development, potentially accounting for the more severe basal phenotype in Sox2 hypomorphs. Notably, Sox2 expression in the prosensory domain has recently been shown to follow an apical-to-basal pattern, suggesting that Sox2 may play a role in the apical-to-basal progression of cell cycle exit and the completion of prosensory specification [27].
To test this hypothesis, we examined cell cycle exit in the prosensory domain via Ki67 expression, as a marker of the status of prosensory specification. Ki67 is expressed by cycling cells, but not cells in the G 0 phase of the cell cycle [28]. In the developing cochlea at around E12.5 to E15.5, cells of the prosensory domain, sometimes referred to as the zone of non-proliferation, have turned off or are beginning to turn off Ki67 expression as they exit the cell cycle [3]. At E14.5 in Sox2 Ysb This indicates that the wave of cell cycle exit, which starts at the apex, has reached the very base of the cochlear duct. However, in Sox2 Ysb/cochleae, only the prosensory domain at the apical turn of the cochlear duct (serial section 6) has turned off Ki67, not at the mid-basal or basal turns (serial sections 1-3); the middle turns (serial sections 4 and 5), meanwhile, were just starting to turn off Ki67 (Fig 5A, brackets). In all, in the 5 Sox2 Ysb/+ cochleae examined, the most basal section that has not yet turned off Ki67 are 1, 1, 1, 1, and 2; in the 6 Sox2 Ysb/cochleae examined, they are 4, 4, 4, 5, 5, and 5 (p = 0.008, Mann-Whitney U test). This indicates that in Sox2 Ysb/cochleae, cell cycle exit is delayed relative to Sox2 Ysb/+ cochleae, suggesting a delay in the completion of prosensory specification. In addition, the nuclei of prosensory domain cells shift away from the luminal surface of the cochlear epithelium upon specification [29]. This basal shift of nuclei localization within the cell leaves a blank space between DAPI-stained nuclei and the luminal surface of the cochlear duct, which can be visualized in all six serial sections in Sox2 Ysb/+ cochleae at E14.5 (Fig 5A, asterisks). However, in Sox2 Ysb/cochleae, cells of the prosensory domain mostly did not exhibit this nuclei shift at E14.5.
At E15.5, the prosensory domain along the entire length of the cochlear duct has turned off Ki67 expression in both Sox2 Ysb/+ and Sox2 Ysb/cochleae, in all samples examined (Fig 5A,  brackets). This suggests that cell cycle exit and prosensory specification in Sox2 Ysb/cochleae has caught up by this stage. Prosensory nuclei localization has also begun to catch up at E15.5 in Sox2 Ysb/cochleae (Fig 5A, asterisks). Overall, these results suggest that prosensory specification is delayed in Sox2 Ysb/cochleae, but not permanently disrupted. Prosensory specification must occur prior to differentiation to generate HCs and SCs. Therefore, the period of time in between cell cycle exit and the initiation of differentiation represents a temporal buffer (Fig 5B, green shading) preventing differentiation from initiating prior to specification. As differentiation begins in the basal/mid-basal cochlear turns shortly after specification, the delay in specification in Sox2 Ysb/cochleae leads to progenitors not having been specified in time for differentiation at the basal end of the cochlear duct (Fig 5B,  crosshatch pattern). We propose that this at least partially explains why the basal end of the cochlea is more sensitive to decreases in the level of Sox2 expression. Moreover, since differentiation begins in the mid-base and spreads to the rest of the base, progenitors at the basal tip in Sox2 Ysb/cochleae may still have time to undergo specification prior to differentiation. This may explain why small islands of HCs and SCs are sometimes seen in the basal tip of Sox2 Ysb/cochleae (S4D Fig).
Notably, while we detected a difference in Ki67 expression between Sox2 Ysb/+ and Sox2 Ysb/cochleae at E14.5 (Fig 5A), we did not detect a difference in EdU incorporation after 1 hour of EdU injection. In the basal turn of Sox2 Ysb/cochleae at E14.5, cells of the presumptive prosensory domain still expressed Ki67, but did not incorporate EdU (S6A Fig, brackets). This indicates that while these presumptive prosensory cells have not exited the cell cycle, they are no longer actively undergoing DNA synthesis. We predict that these cells are either cycling slowly or are transiently "stuck" at a particular stage of the cell cycle.

Sox2 is upstream of Fgf20
While the delay in prosensory specification can explain the preferential loss of sensory epithelium from the basal end of Sox2 hypomorph cochleae, it does not readily explain the preferential loss of OHCs, relative to IHCs. Since this preference for OHC loss is reminiscent of the Fgf20/Fgfr1 deletion phenotypes, we investigated the possibility that Sox2 may be upstream of FGF20-FGFR1 signaling. Interestingly, both Etv4 and Etv5 were dramatically downregulated in the prosensory domain of Sox2 Ysb/cochleae compared to control (Fig 6A). This shows that FGF20-FGFR1 signaling was disrupted in the Sox2 hypomorph cochleae. Examination of Fgfr1 and Fgf20 expression by in situ hybridization revealed that while Fgfr1 expression did not appear to be affected in Sox2 Ysb/cochleae at E14.5, Fgf20 expression was absent ( Fig 6B). This suggests that while Fgfr1 functions upstream of Sox2 (Fig 2A), Fgf20 is downstream of Sox2. This model predicts that Fgf20 expression would be downregulated in Fgfr1-CKO cochleae, which was confirmed by in situ hybridization ( Fig 6C).
The above results indicate that the loss of Fgf20 could partially account for the Sox2 Ysb/phenotype. Therefore, to determine whether loss of Fgf20 also causes delayed prosensory specification, we examined Ki67 expression in Fgf20-KO cochleae. At E14.5, there was no detectable delay in cell cycle exit in Fgf20-KO cochleae, as loss of Ki67 expression reached the base (serial section 1) in all 6 control and all 6 Fgf20-KO cochleae examined (S6B Fig, brackets). There was also no detectable delay in prosensory basal nuclei shift in Fgf20-KO cochleae (S6B Fig, asterisks). These results were expected as the Fgf20-KO phenotype is not more severe at the basal end of the cochlear duct. This is also consistent with Fgf20 being required during differentiation rather than prosensory specification ( Fig 2D). However, these results do not answer whether and how the loss of Fgf20 contributes to the Sox2 hypomorph phenotype.
Since Fgfr1 is upstream of Sox2, we next asked whether Fgfr1 deletion also results in a delay in prosensory cell cycle exit via decrease in Sox2 expression. Interestingly, similar to the Fgf20-KO, there was no detectable delay in cell cycle exit or basal nuclei shift in Fgfr1-CKO cochleae compared to control at E14.5 (S6B Fig, brackets, asterisks). We hypothesized that this lack of a detectable difference may be due to the relatively small reduction in Sox2 expression in Fgfr1-CKO cochleae compared to Sox2 Ysb/cochleae. To examine cell cycle exit more closely for subtle changes, we used Sox2 as a marker of prosensory cells and quantified the number of Sox2-expressing cells that also expressed Ki67 in serial sections through the cochlear duct. Because at E14.5 Sox2 also labels the Kölliker's organ, which is neural (inner) to the prosensory domain, we quantified cells at the abneural (outer) border of the prosensory domain. In E14.5 Fgf20-KO cochleae, a few Ki67+/Sox2+ cells could be found at the abneural border, mostly towards the basal end of the cochlear duct, similar to control cochleae (Fig 6D and 6E). However, there were significantly more Ki67+/Sox2+ cells at the abneural border in Fgfr1-CKO cochleae compared to control (Fig 6D and 6E, arrowheads). This shows that, as expected, deletion of Fgfr1, but not Fgf20, does lead to a quantifiable defect in prosensory cell cycle exit.
We also asked whether a decrease in Sox2 expression can account for the absence of proliferation in Kölliker's organ in Fgfr1-CKO cochleae. Interestingly, EdU-incorporation was decreased in Kölliker's organ in Sox2 Ysb/cochleae at E14.5, especially in the region adjacent to the prosensory domain (S6C Fig, bracket). However, EdU-incorporation was not completely absent from Kölliker's organ, unlike in Fgfr1-CKO cochleae. This suggests that loss of Sox2 in combination with other factors contributes to the Kölliker's organ phenotype in Fgfr1-CKO cochleae.
Interestingly, while the total number of IHCs was decreased in Fgf20 -/-;Sox2 Ysb/Ysb cochleae relative to all other genotypes, this decrease was only found in the basal and middle turns, but not the apical turn (S7B and S8 Figs). No such basal/middle/apical turn discrepancies existed in the number of OHCs or SCs (S7C, S7D and S8 Figs). This is reminiscent of the Fgfr1-CKO and Sox2 Ysb/phenotypes.
To ensure that the Fgf20and Sox2 Ysb interaction is not purely an artifact of the Sox2 Ysb allele, we generated Fgf20 +/+ ;Sox2 +/+ (wildtype), Fgf20 +/-;Sox2 +/+ (Fgf20-het), Fgf20 +/+ ;Sox2 +/-(Sox2-het), and Fgf20 +/-;Sox2 +/-(double het) mice to look for an interaction between the Fgf20and Sox2alleles (S9A Fig). At P0, cochlear length did not significantly differ among the four genotypes (S9B Fig). HC quantification showed that neither Fgf20 nor Sox2 exhibited haploinsufficiency for total or length-normalized number of IHCs or OHCs (S9C and S9D Fig). However, in Fgf20-het and much more so in Sox2-het cochleae, occasional ectopic IHCs can be found in the middle and apical turns of the cochlear duct (S9A Fig, arrowheads). Interestingly, in double het cochleae, many more ectopic IHCs were found, even in the basal turn. These ectopic IHCs led to an increase in the total and length-normalized number of IHCs in double het cochleae, compared to wildtype (S9C Fig). Notably, a significant increase in IHCs was only found in the basal turn, not the middle or apical turns (S9E Fig). In the basal turn, IHC number was significantly increased in double het cochleae compared to wildtype, Fgf20-het, and Sox2-het cochleae. Double het cochleae also had a significant decrease in total and length-normalized number of OHCs compared to wildtype (S9D Fig). Again, a significant decrease in OHCs was only found in the basal turn, not the middle or apical turns (S9F Fig). These results confirm a genetic interaction between Fgf20 and Sox2.

Loss of Fgf20 does not further delay prosensory specification in Sox2 hypomorph cochleae
We propose that the Fgf20 -/-;Sox2 Ysb/Ysb phenotype lies in between that of Fgfr1-CKO and Sox2 Ysb/in terms of severity of reductions in cochlear length and in the number of HCs and SCs. We further hypothesize that these three phenotypes form a continuum with the Fgf20-KO phenotype (Fig 9A). Along this continuum, all four genotypes lack FGF20 signaling, but vary in the level of Sox2 expression and phenotype severity in the basal end of the cochlear duct and the outer compartment (outer rows of OHCs and SCs). From this, and from the Fgf20and Sox2 Ysb series of alleles, we conclude that the basal end of the cochlear duct and the outer compartment are more sensitive to the loss of Fgf20 and Sox2, relative to the apical end and inner compartment, respectively.
To determine the mechanism underlying the Sox2 and Fgf20 interaction, we asked whether the loss of Fgf20 further reduces Sox2 expression on a Sox2 hypomorphic background. In other words, we asked whether Fgf20 has a role in regulating Sox2 expression on a sensitized background. Examination of prosensory domain Sox2 expression at E14.5 revealed, as expected, that Fgf20 -/-;Sox2 Ysb/+ cochleae did not have a decrease in Sox2 expression compared to   The two factors analyzed are Fgf20 (Fgf20 +/+ , Fgf20 +/-, Fgf20 -/-) and Sox2 (Sox2 +/+ , Sox2 Ysb/+ , Sox2 Ysb/Ysb ) gene dosage. A p value < 0.05 (yellow highlight) for Fgf20 or Sox2 indicates that the particular factor (independent variable) has a statistically significant effect on the measurement (dependent variable). Whereas a p value < 0.05 (orange highlight) for Interaction indicates a statistically Loss of Fgf20 did not contribute to a further decrease in CDKN1B expression on a Sox2 Ysb/Ysb background, nor was there a basal-apical difference in CDKN1B expression in Fgf20 -/-;Sox2 Ysb/ Ysb cochleae at E14.5. These results are consistent with our hypothesis that Fgf20 does not regulate Sox2 expression or prosensory specification.
https://doi.org/10.1371/journal.pgen.1008254.g008 at the expression of the two transcription factors via in situ hybridization. In Fgf20-KO cochleae at E14.5, Hey1 expression is downregulated while Hey2 is almost completely absent compared to control (Fig 10A). To test whether FGF20 loss leads to premature differentiation, we examined myosin VI (Myo6) expression, a marker of differentiated HCs [30]. At E14.5, the cochleae of 3 of 12 control embryos examined contained Myo6-expressing HCs, while the cochleae of 18 of 19 littermate Fgf20-KO embryos contained Myo6-expressing HCs (p < 0.001, Fisher's exact test; Fig 10B). If present, the Myo6-expressing HCs at this stage were always found in the basal and mid-basal turns of the cochlea. These results show that there is premature onset of differentiation in Fgf20-KO cochleae, which begins in the basal/mid-basal turns. This result is surprising given our previous finding of delayed differentiation in the apical end of Fgf20-KO cochleae at later stages, which we confirm here (S1E Fig). These findings suggest that while initiation of differentiation occurs earlier in Fgf20-KO cochleae, apical progression of differentiation may be slower.
Similar to a delay in prosensory specification, premature onset of differentiation narrows the temporal buffer between the completion of specification and initiation of differentiation. In the context of a slight delay in specification due to decreased Sox2 levels, premature differentiation from the loss of Fgf20 can lead to an attempt at differentiation before specification in the basal end of the cochlea. We propose that Sox2 and Fgf20 interact to regulate the boundaries of the temporal buffer, helping to ensure that differentiation begins after the completion of specification (Fig 11).

Fgfr1 is involved in prosensory specification and differentiation, while
Fgf20 is only involved in differentiation Fgf20 and Fgfr1 are required for HC and SC development. Based on similarities in the phenotype caused by the loss of FGF20 and loss of FGFR1 signaling, FGF20 has been hypothesized as the FGFR1 ligand during organ of Corti development [17][18][19][20][21]. However, the exact role of FGF20/FGFR1 during organ of Corti development has been a topic of debate. We previously reported that Fgf20-KO mice do not have defects in prosensory specification, and have a normally formed prosensory domain [21]. We further showed that FGF20 signaling is important during the initiation stage of differentiation, and that Fgf20-KO cochleae have gaps in the differentiated sensory epithelium filled with undifferentiated prosensory progenitors. However, other studies have shown in vitro that FGF20 regulates prosensory specification via Sox2 [31] and in vivo that FGFR1 is required for prosensory specification via Sox2 [19]. Here, we show, using an in vivo rescue model, that ectopic FGF9 signaling is sufficient to rescue the Fgf20-KO phenotype in a spatiotemporal pattern that matched the timing of initiation of differentiation along the length of the cochlear duct. We conclude, therefore, that FGF20 is involved in differentiation and is not necessary for prosensory specification.
Notably, the Fgf20-KO phenotype, in which two-thirds of OHCs fail to develop, is not as severe as the Fgfr1-CKO phenotype, which lacks almost all OHCs as well as half of IHCs. Potential explanations for this include differences in mouse genetic background, and the existence of a redundant FGF ligand(s). To rule out the former, we examine here Fgf20-KO and Fgfr1-CKO mice on a similar genetic background, and replicated the difference in phenotype severity. We also replicated the decrease in Sox2 expression in the prosensory domain previously reported in Fgfr1-CKO mice [19]. We further reaffirmed that Sox2 expression in the prosensory domain is not affected by the loss of Fgf20. This suggests that another FGF ligand signaling through FGFR1 is required to maintain Sox2 expression during prosensory specification. The identity of this ligand is currently unknown.
We hypothesized that the severity of the Fgfr1-CKO phenotype is due to decreased Sox2 expression causing disrupted prosensory specification and the loss of FGF20 signaling during Sox2 and Fgf20 interact to modulate a temporal buffer between specification and differentiation. Model of the roles of Sox2 and Fgf20 in organ of Corti development, which occurs in two stages: unspecified progenitors (tan shading) undergo specification and cell cycle exit to become prosensory cells (green shading), which then differentiate into hair cells and supporting cells (HCs & SCs; red shading). In wildtype cochleae, cell cycle exit (indicating completion of specification) begins at the apex of the cochlea and proceeds basally. Afterwards, differentiation initiates at the mid-base of the cochlea and proceeds basally and apically. The prosensory cells exist within a temporal buffer (green shading), defined as the time between cell cycle exit and initiation of differentiation. In Sox2/Fgf20 mutant cochleae, decrease in levels of Sox2 expression in the developing cochlea leads to delayed prosensory specification and cell cycle exit (arrow 1), while loss of Fgf20 leads to premature onset of differentiation at the basal and mid-basal cochlear turns (arrow 2) as well as delayed differentiation at the apical turn (arrow 3). Loss of both Sox2 and Fgf20 leads to loss of the temporal buffer between specification and differentiation towards the base of the cochlear duct, disrupting the development of HCs and SCs in the basal region (crosshatch pattern). differentiation. Consistent with this hypothesis, the combination of Fgf20 -/and Sox2 Ysb/Ysb mutations phenocopied Fgfr1-CKO cochleae. The similarities in phenotype include approximately a 30% reduction in cochlear length and almost a complete loss of OHCs and SCs and approximately a 50% loss of IHCs. Interestingly, the Fgf20 -/-;Sox2 Ysb/Ysb phenotype is also similar to the Sox2 Ysb/phenotype. We conclude that the Fgfr1-CKO, Fgf20 -/-;Sox2 Ysb/Ysb , and Sox2-Ysb/phenotypes likely lie along the same continuum, as these three genotypes all exhibited a lack of Fgf20 expression or signaling and varying levels of Sox2 expression (Fig 9A). Fgf20-KO cochleae, in which Sox2 expression was not affected, lies at the mild end of this continuum. Interestingly, this continuum shows that in the absence of Fgf20 expression or signaling, reductions in the level of Sox2 most severely affected sensory epithelium development of the cochlear base and the outer compartment. Moving from the Fgf20-KO (mild) end of the spectrum towards the Sox2 Ysb/-(severe) end, increasing numbers of HCs and SCs are lost, preferentially form the cochlear base and the outer compartment.
Importantly, while these results seem to suggest that the main function of FGFR1 signaling during early stages of organ of Corti development is to regulate Sox2 expression, we have not ruled out the potential for other functions of FGFR1 signaling. There are also notable differences between the Fgfr1-CKO and Fgf20 -/-;Sox2 Ysb/Ysb phenotypes at P0. For instance, Fgfr1-CKO cochleae have a slightly less severe OHC phenotype at the cochlear base relative to the apex, while Fgf20 -/-;Sox2 Ysb/Ysb cochleae have a slightly more severe OHC phenotype at the base. We attribute these phenotype differences to differences in the level of Sox2 expression and the timing of decrease in Sox2 expression. However, they may be instead attributable to additional functions of FGFR1 signaling not captured by the combination of Fgf20 -/and Sox2 Ysb/Ysb mutations.
Foxg1 Cre has been used in several studies to target the otic epithelium, including to conditionally delete Fgfr1 [18][19][20]. One concern with Foxg1 Cre is that it is a null allele [22]. Foxg1null mice have shortened cochlear length, although HC and SC differentiation did not appear to be directly affected [32]. Previous work [33] and our results here showed that Foxg1 is not haploinsufficient during cochlea development, as Foxg1 Cre/+ ;Fgfr1 flox/+ cochleae had very similar phenotypes to Fgfr1 flox/cochleae. Moreover, the use of the Six1enh21-Cre transgene, which targets the otic epithelium in a similar spatiotemporal pattern as Foxg1 Cre , to conditionally delete Fgfr1 resulted in the same phenotype as Foxg1 Cre/+ ;Fgfr1 flox/cochleae [19]. This included the loss of almost all OHCs, loss of IHCs, and decreased prosensory Sox2 expression. Therefore, the increased severity of Foxg1 Cre/+ ;Fgfr1 flox/cochleae relative to Fgf20 -/cochleae is likely not attributable Foxg1 haploinsufficiency.

Sox2 and Fgf20 interact to affect development towards the basal end of the cochlea
We show here conclusive evidence that Sox2 and Fgf20 genetically interact during cochlea development. Interestingly, HC and SC development towards the basal end of the cochlea is more severely affected by the loss of Sox2 and Fgf20 and their interaction. While we hypothesize that Sox2 and Fgf20 are involved in distinct steps during organ of Corti development (prosensory specification and differentiation, respectively), there is nevertheless potential for a strong interaction. We propose that the timing of specification and differentiation define a temporal buffer that normally prevents differentiation from initiating prior to the completion of specification, and that Sox2 and Fgf20 modulate the borders of this buffer. In a developmental pathway, the upstream event (specification) must occur prior to the downstream event (differentiation). Therefore, loss of Sox2 and Fgf20 leading to delayed specification and premature differentiation onset, respectively, disrupts the temporal buffer, especially towards the cochlear base (Fig 11).
Here, we use cell cycle exit in the prosensory domain (also known as the zone of non-proliferation) as a marker for the completion of specification [3]. We hypothesize that prosensory cells become specified and primed for differentiation upon withdrawal from the cell cycle. Previous studies showed that prosensory cells are indeed capable of differentiating into HCs and SCs directly after cell cycle exit, even in the apex. When Shh was deleted from the spiral ganglion, differentiation began in the apex shortly after cell cycle exit and progressed towards the base [5]. This suggests that specification occurs in an apex-to-base direction. It also suggests that normally, SHH signaling prevents the apex from differentiating immediately after specification. We cannot rule out, however, that specification occurs in the same direction as differentiation (base-to-apex), independently of cell cycle exit. Such a scenario would still be consistent with our model that a combination of delayed specification and premature onset of differentiation accounts for the more severe basal phenotype in Fgf20/Sox2 mutants.
The effect of loss of Fgf20 on the timing of differentiation is small. We estimate that the onset of differentiation in Fgf20-KO cochleae is advanced by only around 0.5 days. By itself, this effect does not lead to a more severe mid-basal or basal phenotype in Fgf20-KO cochleae. However, we present evidence that on a sensitized genetic background causing delayed specification, this small change in the timing of differentiation leads to a large defect in HC and SC production towards the basal end of the cochleae. We propose that this at least partially explains the interaction between Sox2 and Fgf20. Furthermore, the relative sparing of development towards the apical end of Sox2 Ysb/Ysb ;Fgf20 -/cochleae, especially of IHCs, can be further explained by a delay in differentiation at the apical end due to the loss of Fgf20. We do not know why an apical-basal difference in timing of differentiation exists in Fgf20-KO cochleae. Perhaps there is a delay in the apical progression of differentiation, or perhaps other factors contribute to the differentiation of the apical end of the cochlea. Consistent with the latter, by P7 in Fgf20-KO cochleae, the apical tip contains a full complement of IHCs and OHCs, unlike the rest of the cochlea [21]. Importantly, the model that Sox2 and Fgf20 regulate the distinct processes of specification and differentiation, respectively, is a simplified take on a complex developmental pathway for the sake of addressing our specific question. While we show the potential for a Sox2 and Fgf20 interaction in modulating the temporal buffer between specification and differentiation, Sox2 also has known roles during HC and SC differentiation [13,15,34,35]. Therefore, the genetic interaction may occur during differentiation as well. While interaction at this stage may explain the preferential loss of outer compartment cells in Sox2 and Fgf20 mutants, it does not explain the selective loss of basal cochlear HCs and SCs. Therefore, we conclude that the Sox2 and Fgf20 interaction regulates the temporal buffer, with potential further interactions during differentiation.
The Notch ligand Jagged1 (Jag1) is thought to be important for cochlear prosensory specification via lateral induction [36][37][38][39][40][41][42]. Interestingly, Notch signaling has also been shown to be upstream of both Fgf20 and Sox2 in the developing cochlea [31]. Conditional deletion of Jag1 or Rbpj, the major transcriptional effector of canonical Notch signaling, resulted in the loss of HCs and SCs, particularly from the basal end of the cochlear duct, similar to Fgf20/Sox2 mutants. Unlike Fgf20/Sox2 mutants, however, deletion of Jag1 or Rbpj led to preferential loss of Sox2 and CDKN1B expression from the prosensory domain at the basal end of the cochlea [37,39,43]. This suggests that Jag1-Notch signaling is required for prosensory specification, especially towards the cochlear base. This likely accounts for the more severe basal phenotype of Jag1 or Rbpj mutants. This same mechanism likely does not explain the more severe basal phenotype of Fgf20/Sox2 mutants, as Sox2 and CDKN1B expression was not more severely reduced or absent in the cochlear base in these mice. Notably, not all studies agree that Jag1 or Rbpj is required for Sox2 and CDKN1B expression or for prosensory specification [44]. More studies are required to further elucidate the functional relationship between Jag1/Notch, Fgf20, and Sox2 during cochlea development.
Other genes that potentially interact with Fgf20 and Sox2 during cochlea development include Mycn (N-Myc) and Mycl (L-Myc). Interestingly, deletion of Mycn and Mycl from the cochlear epithelium results in accelerated cell cycle exit and delayed initiation of differentiation [45], opposite to the effects of loss of Sox2 and Fgf20. Addressing potential interactions between Sox2, Fgf20, Mycn, and Mycl is another topic for future studies.
Outer compartment of the cochlear sensory epithelium is more sensitive to the loss of Fgfr1, Fgf20, and Sox2 than the inner compartment In all of the genotypes we observed in this study, the loss of outer compartment cells (i.e. OHCs) was predominant. Only in the most severe cases in which almost all OHCs were missing, as seen in Fgfr1-CKO, Fgf20 -/-;Sox2 Ysb/Ysb , and Sox2 Ysb/cochleae, were IHCs also lost. Similarly, reduction in SC number always preferentially affected the outermost cells. This suggests that the organ of Corti outer compartment is more sensitive to the loss of Fgfr1, Fgf20, and Sox2 than the inner compartment. The combination of Fgf20and Sox2 Ysb alleles elegantly demonstrates this: as the number of Fgf20and Sox2 Ysb alleles increased, the number of OHCs progressively decreased. In the double homozygous mutants, the number of IHCs decreased as well. We also show here that in Fgfr1-CKO and Fgf20 -/-;Sox2 Ysb/Ysb cochleae, there is a delay in cell cycle exit and completion of specification, especially towards the abneural or outer side of the prosensory domain, which may contribute to the more severe outer compartment phenotype. However, Fgf20-KO cochleae, which also exhibit a more severe outer compartment phenotype, did not show a delay in specification of the outer side of the prosensory domain, at least at the stage we examined. Further studies are needed to elucidate differences in the timing of cell cycle exit and specification in the outer and inner compartments, and how they may affect differentiation.
Previous studies noted that the dosage of Fgfr1 affects the degree of organ of Corti outer compartment loss. In Fgfr1 hypomorphs with 80% reduction in transcription, only the third row of OHCs were missing, while hypomorphs with 90% reduction had a slightly more severe phenotype [20]. Therefore, Fgfr1 loss preferentially affects the outermost HCs. Other studies suggested that the timing of Fgfr1 deletion is important in determining the degree of outer compartment loss and level of Sox2 expression. When an earlier-expressed Cre driver (Six1-enh21-Cre) was used to conditionally delete Fgfr1, almost all OHCs and some IHCs were lost, with a 66% reduction in Sox2 expression at E14.5 [19]. When a later-expressed Cre driver (Emx2 Cre ) was used, many more OHCs and IHCs remained, with only a 12% reduction in Sox2 expression. Our results are consistent with both of these studies. We show that FGF20-independent FGFR1 signaling and Sox2 are required early, affecting both IHC and OHC development, while FGF20-FGFR1 signaling is important during later stages, affecting only OHC development.
Differentiation in the organ of Corti not only occurs in a basal-to-apical gradient, but also in an orthogonal inner-to-outer gradient. That is, IHCs differentiate first, followed by each sequential row of OHCs [46]. This wave of differentiation suggests that perhaps outer compartment HCs and SCs require a longer temporal buffer between specification and differentiation. The genetic interaction between Sox2 and Fgf20 in modulating this temporal buffer, therefore, could also account for the preferential loss of outer compartment HCs and SCs. We hypothesize that the requirement for a longer temporal buffer may also be involved in determining OHC fate. In Fgf20 +/-;Sox2 +/cochleae, there was a slight decrease in OHCs that was compensated for by ectopic IHCs, suggesting a fate switch from OHCs into IHCs. Here, we confirmed previous suggestions that Fgf20 regulates Hey1 and Hey2 to prevent premature differentiation in the developing organ of Corti [8,30]. Interestingly, in Hey1/Hey2 double knockout cochleae, there was a similar slight decrease in OHCs compensated for by ectopic IHCs [30]. Furthermore, inner ear-specific deletion of either Smoothened or Neurod1, which led to premature differentiation in the apical cochlear turn, also led to loss of OHCs and the presence of ectopic IHCs at the apex [6,8]. These findings further support a model where timing of specification and differentiation affect IHC versus OHC fate, an interesting and important topic for future studies.
Previously, we hypothesized that Fgf20 is strictly required for the differentiation of an outer compartment progenitor [21]. However, data we present here show that Fgf20, on a sensitized, Sox2 hypomorphic background, is also required for inner compartment differentiation. We conclude that inner and outer compartment progenitors likely are not distinct populations. Rather, all prosensory progenitors giving rise to the organ of Corti exist on an inner-to-outer continuum. FGF20 signaling, in combination with other factors including Sox2, are required for the proper development of all of these cells, though with varying sensitivities.

The relationship between Fgf20 and Hey1/Hey2 in regulating differentiation is complex
We show in vivo that Fgf20 is upstream of Hey1 and Hey2. Supporting this result, Fgfr1 has also been shown in vivo to be upstream of Hey2 [19]. Interestingly, in explant studies, inhibition of FGF signaling alone did not result in decreased Hey1/Hey2 expression or premature differentiation [30]. However, FGF inhibition has been shown to rescue the overexpression of Hey1/Hey2 and the delay in differentiation induced by SHH signaling overactivation [8,30]. Notably, these studies suggest that SHH signaling from the spiral ganglion regulates Fgf20 expression, which in turn regulates Hey1 and Hey2 expression to prevent premature differentiation in the organ of Corti [5,8,30]. Our results here showing that Fgf20 regulates Hey1 and Hey2 expression and timing of differentiation are mostly consistent with these studies. However, Hey1/Hey2 double knockout cochleae do not exhibit a loss of OHCs to the extent of Fgf20-KO cochleae, suggesting that other genes downstream of Fgf20 are important in prosensory cell differentiation (Fig 11). Moreover, deletion of Fgf20 only led to premature differentiation at the basal and mid-basal turns. Fgf20 deletion actually delayed differentiation in the apical end of the cochlea. Deletion of Hey1/Hey2, contrarily, led to premature differentiation along the entire length of the cochlear duct, although it is unclear how Hey1/Hey2 loss affects the timing of apical differentiation beyond E15.0 [30]. This suggests that other factors downstream of Fgf20 interact with Hey1/ Hey2 to regulate the timing of differentiation. Perhaps these same genes contribute to the loss of OHCs in Fgf20-KO cochleae. Mekk4, which has been shown to be downstream of Fgf20 and necessary for OHC differentiation [47] could be one of these genes. Identifying other factors downstream of Fgf20 will be a topic of future studies.

Ethics statement
All studies performed were in accordance with the Institutional Animal Care and Use Committee at Washington University in St. Louis (protocol #20160113) and University of Nebraska Medical Center (protocol #16-004-02 and 16-005-02).

Mice
Mice were group housed with littermates, in breeding pairs, or in a breeding harem (2 females to 1 male), with food and water provided ad libitum.
For timed-pregnancy experiments, embryonic day 0.5 (E0.5) was assigned as noon of the day the vaginal plug was found. For postnatal experiments, postnatal day 0 (P0) was determined as the day of birth.
Mice were of mixed sexes and maintained on a mixed C57BL/6J x 129X1/SvJ genetic background. All mice were backcrossed at least three generations onto this background. The following mouse lines were used: • Fgf20 Cre (Fgf20 -; MGI:5751785): knockin allele containing a sequence encoding a GFP-Cre fusion protein replacing exon 1 of Fgf20, resulting in a null mutation [18].

Doxycycline induction
Pregnant dams were starved overnight the night before initiation of Dox induction and fed Dox Diet, Grain-Based Doxycycline, 200 mg/kg (S3888, Bio-Serv, Flemington, NJ) ad libitum starting at noon on the start date of Dox induction. On the stop date of Dox induction, Dox Diet was replaced with regular mouse chow at noon.

Sample preparation and sectioning
For whole mount cochleae, inner ears were dissected out of P0 pups and fixed in 4% PFA in PBS overnight at 4˚C with gentle agitation. Samples were then washed x3 in PBS. Cochleae were dissected away from the vestibule, otic capsule, and periotic mesenchyme with Dumont #55 Forceps (RS-5010, Roboz, Gaithersburg, MD). The roof of the cochlear duct was opened up by dissecting away the stria vascularis and Reissner's membrane; tectorial membrane was removed to expose hair and supporting cells. For sectioning, heads from E14.5 embryos were fixed in 4% PFA in PBS overnight at 4˚C with gentle agitation. Samples were then washed x3 in PBS and cryoprotected in 15% sucrose in PBS overnight and then in 30% sucrose in PBS overnight. Samples were embedded in Tissue-Tek O.C.T. compound (4583, VWR International, Radnor, PA) and frozen on dry ice. Serial horizontal sections through base of the head were cut at 12 μm with a cryostat, dried at room temperature, and stored at -80˚C until use.

RNA in situ hybridization
Probe preparation: mouse cDNA plasmids containing the following inserts were used to make RNA in situ probes, and were cut and transcribed with the indicated restriction enzyme ( Frozen section in situ hybridization: frozen slides were warmed for 20 min at room temperature and then 5 min at 50˚C on a slide warmer. Sections were fixed in 4% PFA in PBS for 20 min at room temperature, washed x2 in PBS and treated with pre-warmed 10 μg/ml Proteinase K (03115828001, Sigma-Aldrich, St. Louis, MO) in PBS for 7 min at 37˚C. Sections were then fixed in 4% PFA in PBS for 15 min at room temperature, washed x2 in PBS, acetylated in 0.25% acetic anhydrate in 0.1M Triethanolamine, pH 8.0, for 10 min, and washed again in PBS. Sections were then placed in pre-warmed hybridization buffer (50% formamide, 5x SSC buffer, 5 mM EDTA, 50 μg/ml yeast tRNA) for 3 h at 60˚C in humidified chamber for prehybridization. Sections were then hybridized in 10 μg/ml probe/hybridization buffer overnight (12-16 h) at 60˚C. The next day, sections were washed in 1x SSC for 10 min at 60˚C, followed by 1.5x SSC for 10 min at 60˚C, 2x SSC for 20 min at 37˚C x2, and 0.2x SSC for 30  Finally, sections were dehydrated in 30% and then 70% methanol, 5 min each, followed by 100% methanol for 15 min. Sections were then rehydrated in 70% and 30% methanol and then PBS, 5 min each, and mounted in 95% glycerol.

Immunofluorescence
Whole mount: cochleae were incubated in PBS + 0.5% Tween-20 (PBSTw) for 1 h to permeabilize. Cochleae were then blocked using PBSTw + 5% donkey serum for 1 h and then incubated in PBSTw + 1% donkey serum with the primary antibody overnight at 4˚C. Cochleae were then washed x3 in PBS and incubated in PBS + 1% Tween-20 with the secondary antibody. After wash in PBS x3, cochleae were mounted in 95% glycerol with the sensory epithelium facing up.
Frozen slides were warmed for 30 min at room temperature and washed in PBS before incubating in PBS + 0.5% Triton X-100 (PBST) for 1 h to permeabilize the tissue. Sections were then blocked using in PBST + 5% donkey serum for 1 h and then incubated in PBST + 1% donkey serum with the primary antibody overnight at 4˚C in a humidified chamber. Sections were then washed x3 in PBS and incubated in PBS + 1% Triton X-100 with the secondary antibody. After wash in PBS x3, slides were mounted in VectaShield antifade mounting medium with DAPI (H-1200, Vector Labs, Burlingame, CA).

Cell proliferation assay
EdU (E10187, Thermo-Fisher Scientific, Waltham, MA) was injected i.p. into pregnant dams at 100 μg per gram body weight. Embryos were harvested at 1 h after injection. EdU was detected using the Click-iT EdU Alexa Fluor 488 kit (C10337, Thermo-Fisher Scientific, Waltham, MA) according to manufacturer's instructions.

Imaging
Brightfield microscopy was done using a Hamamatsu NanoZoomer slide scanning system with a 20x objective. Images were processed with the NanoZoomer Digital Pathology (NDP. view2) software.
Fluorescent microscopy was done using a Zeiss LSM 700 confocal or Zeiss Axio Imager Z1 with Apotome 2, with z-stack step-size determined based on objective lens type (10x or 20x), as recommended by the ZEN software (around 1 μm). Fluorescent images shown are maximum projections. Low magnification fluorescent images shown of the whole cochlear duct required stitching together, by hand, several images. Images were processed with ImageJ (imagej.nih.gov).

Quantification
Measurements and cell quantification (using the Cell Counter plugin by Kurt De Vos) were done using ImageJ. Total cochlear duct length was defined as the length from the very base of the cochlea to the very tip of the apex, along the tunnel of Corti. Hair cells were identified via Phalloidin, which binds to F-actin [51]. Supporting cells (SCs, including pillar cells and Deiters' cells) were identified based on positive labeling with both Prox1 [52] and Sox2 [10]. Inner hair cells (IHCs) were differentiated from outer hair cells (OHCs) based on their neural/ abneural location, respectively, relative to p75NTR-expressing inner pillar cells [53]. For total cell counts, IHCs, OHCs, and SCs were counted along the entire length of the cochlea. Total cell counts were also normalized to cochlear length and presented as cell count per 100 μm of cochlea (e.g. IHCs/100 μm). For cell quantification at the basal, middle, and apical turns of the cochlea, the cochlear duct was evenly divided into thirds, and total IHCs, OHCs, and SCs were quantified for each third and normalized to length. For the Fgf9-rescue experiments in Fig 3, IHCs, OHCs, and SCs from at least 300 μm regions of the basal (10%), middle (40%), and apical (70%) turns of the cochleae were counted and normalized to 100 μm along the length of the cochlear duct.
In Sox2 Ysb/cochleae, p75NTR expression was mostly absent, resulting in sensory islands without p75NTR-expressing inner pillar cells. In these cochleae, HCs not associated with inner pillar cells were presumed to be IHCs during quantification. When a curved line was drawn connecting the p75NTR islands along the organ of Corti, these presumed IHCs were always neural (inner) to that line.