Introduction

Housed in the mammalian cochlea, the auditory epithelium, also referred to as the Organ of Corti (OC), contains the mechanosensory receptors hair cells (HCs) that detect sound information ^{1, 2}. Two types of HCs exist: inner HCs (IHCs) and outer HCs (OHCs), with one row of IHCs and three rows of OHCs ³. OHCs and IHCs derive from the cochlear progenitors experiencing high level of Atoh1, a master transcription factor (TF) in HC development ^4–6. No HCs are formed in Atoh1 ^-/- cochleae, highlighting the essential roles of Atoh1 in specifying the general HC fates in the undifferentiated progenitors ^{7, 8}. OHCs uniquely express motor protein Prestin (encoded by Slc26a5) and serve as sound amplifiers ^{9, 10}. In contrast, IHCs specifically express Fgf8, vGlut3 (encoded by Slc17a8) and Otoferlin ^11–15, acting as the primary sensory cells that form synapses with different subtypes of type I spiral (auditory) neurons ^16–19. Insm1 is necessary in stabilizing OHC fate, and ∼50% of Insm1^-/- OHCs tend to transdifferentiate into IHC-like cells ^{20, 21}. Ikzf2 is not needed in early OHC development but is required in consolidating the OHC fate, and Ikzf2 ^cello/cello deficient OHCs are dysfunctional and misexpress IHC genes ²². Insm1 is an epistatic but indirect positive regulator of Ikzf2, and restoration of Ikzf2 can partially mitigate the phenotypes of Insm1^-/- OHCs ²¹. In contrast to Insm1 and Ikzf2, Tbx2 is required in IHC fate specification, differentiation, and maintenance, and IHC to OHC conversion occurs in the absence of Tbx2 ^23–25.

It is necessary to maintain the survival of cochlear HCs after their production. Pou4f3, a POU-domain family TF and known hearing loss gene (DFNA15), is essential for HCs to survive ^26–30. Different from Atoh1, Pou4f3 is not involved in fate determination of HCs ³¹. As one target gene regulated by Atoh1 ^{27, 32}, Pou4f3 is turned on in HCs between E14.5 and E16 in a basal to apical gradient, and subsequently Pou4f3 activates Gfi1, another zinc finger TF ^{33, 34}. Gfi1 becomes undetectable in Pou4f3-deficient HCs ³⁴. HCs, especially the OHCs, degenerate in Pou4f3 and Gfi1-mutants by birth ³⁴. Besides Pou4f3 and Gfi1, RNA binding protein Rbm24 is also specifically expressed in cochlear HCs ³⁵. Our recent study shows that Rbm24 is indispensable in maintaining the survival of OHCs, but the OHC death happens much later than the case in Pou4f3 and Gfi1-mutants, with OHCs develop normally by birth but the majority of them undergo cell death by postnatal day 19 (P19) in Rbm24^-/- mice ³⁶. It remains elusive about how Rbm24 is regulated during cochlear HC development.

The similar but delayed phenotype of OHC death in Rbm24 mutants relative to Pou4f3 and Gfi1-deficient mice prompted us to dissect the potential genetic interactions among them by utilizing in vivo genetic approach. Our data clearly showed that onset of Rbm24 expression was completely repressed in Pou4f3^-/- HCs. Surprisingly, Rbm24 expression was normal in Gfi1^-/- HCs. Moreover, neither Pou4f3 nor Gfi1 expression was altered in the Rbm24^-/- HCs. Thus, our data showed that Pou4f3, but not Gfi1, is required in Rbm24 expression. Furthermore, we identified three Rbm24 enhancers that were sufficient to drive specific EGFP reporter expression in HCs, which are likely bound by Pou4f3. Finally, restoration of Rbm24 alone cannot alleviate the degeneration of Pou4f3^-/- HCs, highlighting that additional targeted genes of Pou4f3 are needed to be restored before keeping Pou4f4^-/-HCs able to survive. In sum, our study provides new insights into the genetic interactions among Pou4f3, Gfi1 and Rbm24, with potential applications into HC protection.

Results

Rbm24 expression is completely repressed in Pou4f3 ^-/- cochlear HCs

HC degeneration happens in both Pou4f3^-/- and Rbm24^-/- mice, with the phenotypes occur earlier and are more severe in Pou4f3^-/- mice ^{34, 36}. It prompted us to speculate that genetic interaction between Pou4f3 and Rbm24 exists. Given Pou4f3 is the upstream positive regulator of Rbm24, Rbm24 expression should be blocked in the absence of Pou4f3. To rapidly determine whether it is the case, we took advantage of our previously established CRISPR-stop approach ³⁷. As illustrated in Supplemental Figure 1A, co-injecting one sgRNA-1 against Pou4f3 and base editor components in one-cell stage mouse zygotes yielded founder 0 (F0) mice whose tail DNA was subjected to sanger sequencing. Relative to wild type (WT) mice (Supplemental Figure 1B), the F0 mice carrying the homozygous premature emergence of stop codon (TAG) were defined as Pou4f3^-/-mice that were immediately ready for phenotypic analysis (Supplemental Figure 1C).

Triple staining of Pou4f3, Rbm24 and Insm1 demonstrated that OHCs were Pou4f3+/Rbm24+/Insm1+, and IHCs expressed Pou4f3 and Rbm24, but not Insm1 in basal turn of WT mice at E16.5 (n=3, Supplemental Figure 1D-D’’’). Pou4f3 expression was completely absent in the Pou4f3 ^-/- mice (n=9, Supplemental Figure 1E-E’’’), confirming that Pou4f3 translation was blocked. Different from WT mice, Rbm24 expression completely disappeared in the Pou4f3 ^-/- mice at E16.5 (Supplemental Figure 1E-E’’’). Notably, Insm1 expression seemed normal in Pou4f3 ^-/- mice (arrows in Supplemental Figure 1E-E’’’), although we noticed that the alignment of Insm1+ OHCs was not as regular as the WT OHCs (Supplemental Figure 1D-D’’’). Moreover, the presence of Insm1+ OHCs in the Pou4f3 ^-/- mice ruled out the possibility that the absence of Rbm24 was due to a secondary effect of OHC death or delayed differentiation, and also agreed with previous report that initial HC differentiation can occur without Pou4f3 ^{31, 34}. Thus, the rapid Pou4f3 loss-of-function analysis by CRISPR-stop supports our hypothesis that Pou4f3 is required in turning on Rbm24 expression.

Figure 1.

To further validate the observation above, we also constructed a germ line stable Pou4f3 mutants in which the entire Pou4f3 genome between sgRNA-2 and sgRNA-3 was deleted (Figure 1A). The three genotypes of Pou4f3^+/+, Pou4f3^+/- and Pou4f3^-/- were readily distinguished by tail DNA PCR (Figure 1B). Relative to WT mice (n=3, Figure 1C), Rbm24 again was completely absent in Pou4f3^-/- mice at E16.5 (n=3, Figure 1D). Like Insm1, Bcl11b is another OHC marker ²⁰. Triple staining of Rbm24, Pou4f3 and Bcl11b also demonstrated that, relative to WT HCs (Figure 1E-E’’’), Rbm24 was completely absent in Pou4f3^-/- HCs (Figure 1F-F’’’). Presence of Bcl11b+ OHCs also confirmed that production of nascent OHCs were normal in the absence of Pou4f3. Moreover, in contrast to WT mice at P1 (n=3, Supplemental Figure 2A-A’’), HC degeneration was manifest in the germ line Pou4f3^-/- mutants at P0 (n=4, Supplemental Figure 2B-B’’) and P1 (n=3, Supplemental Figure 2C-C’’). By normalizing numbers of the remaining HCs in Pou4f3^-/-mice at P0 or P1 to those in WT mice at P1, we found that in P0 Pou4f3^-/- mice, 35.94% ± 4.52%, 38.16% ± 2.79% and 63.8% ± 5.28% HCs survived in basal, middle and apical turns, respectively. However, in P1 Pou4f3^-/-mice, 5.54% ± 3.11%, 15.79% ± 3.65% and 65.64% ± 7.95% of the HCs respectively survived in basal, middle and apical turns. Thus, HC degeneration at middle and basal turns was more severe (** p<0.01) at P1 than that at P0, but there was no significant difference in apical turns (Supplemental Figure 2D).

Figure 2.

Different from WT cochleae in which all Myo7a+ HCs expressed Rbm24 (Supplemental Figure 2E-E’’), in Pou4f3^-/- cochleae the remaining HCs lost Rbm24 expression (arrows in Supplemental Figure 2F-F’’). Collectively, our data from two different Pou4f3-deficient models support that Pou4f3 is indispensable for turning on Rbm24. Nonetheless, whether Gfi1 is also essential for Rbm24 expression remains to be determined.

Construction of a Gfi1-3× HA-P2A-Cre/+ knockin mouse strain

We determined whether the regulation of Rbm24 by Pou4f3 through Gfi1 or independently of Gfi1. Due to lack of a good commercial Gfi1 antibody suitable for immunostaining, we decided to construct a Gfi1-3× HA-P2A-Cre/+ (Gfi1^HA-Cre/+ in brief) knockin mouse strain via our routine CRISPR/Cas9 approach (Supplemental Figure 3A-C). In principle, the C-terminus of Gfi1 was tagged with three HA fragments, and the Cre is under the endogenous cis-regulatory elements of Gfi1. The WT (Gfi1^+/+), heterozygous (Gfi1^HA-Cre/+) and homozygous (Gfi1^{HA-Cre/HA-Cre}) mice were readily distinguished by tail PCR (Supplemental Figure 3D). In addition to the Gfi1 locus, Southern blot assay showed that random insertion of the targeting vector (Supplemental Figure 3B) happened in the unknown genomic region. Nonetheless, it was likely that the random insertion happened in a silent genomic region, which was evidenced by the analysis below.

Figure 3.

Dual staining of HA (Gfi1) and pan-HC marker Myo7a demonstrated that, similar to Gfi1^+/+ mice (n=3, Supplemental Figure 3E-E’’), HC development in Gfi1^{HA-Cre/HA-Cre} was normal at P1 (n=3, Supplemental Figure 3F-F’’). Moreover, HA (Gfi1) was highly expressed in all Myo7a+ IHCs and OHCs in Gfi1^{HA-Cre/HA-Cre} mice (arrows in Supplemental Figure 3F-F’’). It confirmed that Gfi1 endogenous expression is not affected in the Gfi1^{HA-Cre/HA-Cre} mice, representing an advantage over the previous Gfi1^Cre/+ strain where one copy of Gfi1 is lost and early onset of hearing loss happens in Gfi1^{Cre/+ 38, 39}.

We additionally performed the fate mapping analysis of Gfi1^HA-Cre/+; Rosa26-loxp-stop-loxp-tdTomato (Ai9)/+ (Gfi1^HA-Cre/+; Ai9/+) at P2. The Gfi1^HA-Cre/+; Ai9/+ model allowed us to both visualize temporal Gfi1 protein expression pattern and permanently mark cells experiencing Gfi1 expression with tdTomato. Neither tdTomato nor HA was detected in the Myo7a+ HCs of the control Ai9/+ mice (n=5, Supplemental Figure 3G-G’’’ and I-I’’). In contrast, majority of the Myo7a+ HCs expressed tdTomato and HA in the Gfi1^HA-Cre/+; Ai9/+ mice (n=5, Supplemental Figure 3H-H’’’ and J-J’’). We noticed that the general HA (Gfi1) expression level in IHCs seemed higher than that in OHCs (Supplemental Figure 3J-J’’). Moreover, HA (Gfi1) expression levels exhibited manifest heterogeneities among OHCs. The OHCs expressing the highest (#1), intermediate (#2) and lowest (#3) levels of HA (Gfi1) were respectively marked (arrows in Supplemental Figure 3J-J’’). Notably, such a heterogenous Gfi1 expression in neonatal HCs was not observed in Gfi1^{HA-Cre/HA-Cre} mice (Supplemental Figure 3F-F’’), likely due to the higher level of HA tagged Gfi1 in Gfi1^{HA-Cre/HA-Cre} than in Gfi1^HA-Cre/+ mice. Besides HCs, tdTomato+ cells were present in the non-sensory regions (arrows in Supplemental Figure 3H), consistent with the previous report ³⁸. Collectively, Gfi1^HA-Cre/+ strain was suitable to visualize Gfi1 protein by HA antibody, albeit random insertion of the targeting vector existed. Alternatively, Gfi1^HA-Cre/+ at least can be treated as a pseudo transgenic mouse strain, by which we can reliably visualize Gfi1 and trace cells experiencing Gfi1 expression.

Gfi1 expression is prohibited in the Pou4f3^-/- cochlear HCs

Given HA can faithfully represent the Gfi1 expression in the Gfi1^HA-Cre/+, HA expression was expected to be repressed in the Pou4f3^-/- HCs, as reported by previous studies ^{33, 34}. We confirmed this predication by our CRISPR-stop approach ³⁷. Basically, the pipeline was identical to the production of Pou4f3 mutants described above (Supplemental Figure 1), except that here the zygotes were derived from male Gfi1^HA-Cre/+ mice (Supplemental Figure 4A). We obtained the mosaic (or chimeric) Pou4f3^-/- mice, which was confirmed by Sanger sequencing of tail PCR (Supplemental Figure 4B and C), partly due to the injection time was the late stage of one-cell stage zygotes. In control Gfi1^HA-Cre/+ mice, there was a pure ‘C’ base (blue arrow in Supplemental Figure 4B), however, a mixture of ‘C’ and ‘T’ double peaks existed in Gfi1^HA-Cre/+; Pou4f3^-/- (mosaic) mice (red arrow in Supplemental Figure 4C). Nonetheless, the advantage of the mosaic Pou4f3^-/- mice was that it allowed us to have internal control Pou4f3^+/+ and Pou4f3^-/- HCs in the same cochlea, and to determine the phenotype occurs in a cell autonomous or non-cell autonomous manner.

Figure 4.

In control Gfi1^HA-Cre/+ mice (n=3) at E16.5, all HCs expressed Pou4f3, Rbm24 and HA(Gfi1), albeit HA (Gfi1) levels again seemed heterogenous among the OHCs (Supplemental Figure 4D-D’’’). In contrast, in the Gfi1^HA-Cre/+; Pou4f3^-/- (mosaic) mice (n=3), we captured HCs that lost or maintained Pou4f3 expression (Supplemental Figure 4E-E’’’). Notably, Pou4f3+ HCs expressed HA(Gfi1) and Rbm24 (blue arrows in Supplemental Figure 4E-E’’’), in contrast, both HA(Gfi1) and Rbm24 disappeared in HCs that lost Pou4f3 expression (orange arrows in Supplemental Figure 4E-E’’’). Thus, our data clearly confirmed that HA indeed was a reliable readout for Gfi1 expression and was sensitive to loss of Pou4f3. Moreover, it supported that Pou4f3 regulates Rbm24 in a cell autonomous manner.

Gfi1 is dispensable for Rbm24 expression

Next, we produced Gfi1^-/- mutants by the same CRISPR-stop approach ³⁷. One-cell stage zygotes derived from male Gfi1^HA-Cre/+ mice were injected with base editor components and 4 different sgRNAs located in different exons of Gfi1 (Figure 2A). The reason why 4 sgRNAs were combined was elaborated in the discussion part. With sgRNA-6 as an example, Sanger sequencing of tail DNA demonstrated that, relative to control Gfi1^HA-Cre/+ mice (black arrow in Figure 2B), premature emergence of TAG stop codon happened in F0 mice, resulting in homozygous Gfi1 inactivation (red arrow in Figure 2C). All HCs expressed HA (Gfi1) and Rbm24 in control Gfi1^HA-Cre/+ mice (n=3, Figure 2D-D’’). Notably, Rbm24 expression was maintained in the Gfi1^-/- HCs that lost HA (Gfi1) at E16.5 (n=4, Figure 2E-E’’). It suggested that Rbm24 expression is not dependent on Gfi1.

The advantage of the above Gfi1^-/- model was rapid and direct confirming the absence of Gfi1 expression in the F0 mice. However, we cannot guarantee that all HCs completely lost Gfi1 expression at the single cell resolution via HA staining, because only one allele of Gfi1 was Gfi1^HA-Cre. Thus, to rule out the possibility that only the HA tagged Gfi1 allele was mutated, we further generated the germ line stable Gfi1 mutants in which its majority DNA fragments between sgRNA-4 and sgRNA-9 were deleted (Supplemental Figure 5A). The WT (Gfi1^+/+), Gfi1^+/- and Gfi1^-/- mice were readily to be identified by tail PCR (Supplemental Figure 5B). Consistent with previous reports ⁴⁰, dual staining of Rbm24 and Pou4f3 showed that, relative to WT mice (n=3, Supplemental Figure 5C-C’’), severe degeneration of HCs, especially of the OHCs, was present in the Gfi1^-/- mice (n=3, Supplemental Figure 5D-D’’). It validated the success of generating the Gfi1^-/- mice. Notably, the surviving Pou4f3+ HCs kept the expression of Rbm24 (orange arrows in Supplemental Figure 5D-D’’). In contrast, no significant difference was observed between WT and Gfi1^-/- mice at E16.5, as both WT and Gfi1^-/-HCs expressed Pou4f3 and Rbm24, and both WT and Gfi1^-/- OHCs expressed Bcl11b (Supplemental Figure 5E-F’’’). It suggests that degeneration of Gfi1^-/- HCs does not begin by E16.5. Presence of Pou4f3 in Gfi1^-/-HCs agreed with previous notion that Pou4f3 is epistatic to Gfi1. Collectively, analysis of both Gfi1 mutant models supports that Gfi1 is dispensable for Rbm24 expression.

Figure 5.

Pou4f3 and Gfi1 expressions are normal in Rbm24^-/- cochlear HCs

Given that Pou4f3 is upstream and positive regulator of Rbm24, we predicted that Pou4f3 expression should be unaffected in the Rbm24^-/- HCs. To determine whether the predication was correct or not, we established the Rbm24^-/- mouse strain by injecting one Rbm24 sgRNA (sgRNA-10) into one-cell stage zygotes derived from male Gfi1^HA-Cre/+ mice (Figure 3A). We chose Gfi1^HA-Cre/+ zygotes in order to simultaneously assess Gfi1 and Pou4f3 expression patterns in Rbm24^-/- HCs. Relative to the control Gfi1^HA-Cre/+; Rbm24^+/+ (Figure 3B), premature emerge of the TAG stop codon appeared in the Gfi1^HA-Cre/+; Rbm24^-/- (mosaic) mice (Figure 3C). Notably, Sanger sequencing of tail DNA showed that the Rbm24 mutation was mosaic, as mixed ‘T’ and ‘C’ peak existed (red arrow in Figure 3C).

The mosaic inactivation of Rbm24 was further confirmed by triple staining of Rbm24, HA and Pou4f3 (Figure 3D-E’’’). Relative to control Gfi1^HA-Cre/+; Rbm24^+/+ (n=3, Figure 3D-D’’’), Rbm24 expression disappeared in a fraction of cochlear HCs (orange arrows in Figure 3E-E’’’) but remained normal in other HCs (blue arrows in Figure 3E-E’’’) of the Gfi1^HA-Cre/+; Rbm24^-/- (mosaic) mice (n=3) at E17. Notably, regardless of whether Rbm24 was inactivated or not in the cochlear HCs, both HA (Gfi1) and Pou4f3 expression patterns were intact. It supports that Pou4f3 is epistatic to Rbm24 and deletion of Rbm24 does not affect Pou4f3 expression. Moreover, our data suggests that Gfi1 and Rbm24 do not have genetic interactions. Gfi1 inactivation did not affect Rbm24 (Figure 2 and Supplemental Figure 5), and vice versa (Figure 3E-E’’’).

Three Rbm24 enhancers can drive specific EGFP expression in cochlear HCs

After showing that Pou4f3, but not Gfi1, is indispensable in mediating Rbm24 expression, we next determined the mechanism underlying how Pou4f3 controls Rbm24 expression, in particular, the cis-regulation elements (CREs) of Rbm24. According to our previous Transposase-Accessible Chromatin with high-throughput sequencing (ATAC-seq) of neonatal cochlear HCs ⁸, four CREs of Rbm24 were identified: one proximal promoter (arrow in Figure 4A) and three distal ones that were potential enhancers and defined as Eh1, Eh2 and Eh3, respectively (dotted square in Figure 4A). Eh1 and Eh2 were distributed upstream, whereas Eh3 was downstream of Rbm24 coding region. Moreover, we reanalyzed the Pou4f3 Cut&Run assay from one previous study ³², and found that Pou4f3 binds to Eh1, Eh2 and Eh3, but not to the Rbm24 promoter (Figure 4A).

Whether Eh1, Eh2 and Eh3 are bona-fide Rbm24 enhancers remain to be determined. We reasoned that, if Eh1, Eh2 and Eh3 were Rbm24 enhancers, either of them, together with the mini promoter of mouse heat shock protein 68 (Hsp68) ^{8, 41–43}, would be sufficient to drive specific reporter expression in cochlear HCs. To address this question, we established three transgenic mouse strains: Eh1-EGFP+ (n=7, Figure 4B-C’’), Eh2-EGFP+ (n=6, Figure 4D-E’’) and Eh3-EGFP+ (n=4, Figure 4F-G’’). Briefly, in each strain the EGFP expression was driven by the mini promoter of Hsp68 and Eh1 (or Eh2 or Eh3). It is known that the mini promoter of Hsp68 alone is not able to drive EGFP expression ⁴⁴. In contrast, strong EGFP expression was observed by whole mount analysis in all three transgenic lines (Figure 4B, D and F). Moreover, dual staining of EGFP and Myo7a in cryosectioned cochleae showed that EGFP was specifically expressed in IHCs and OHCs in all three strains at P1 (Figure 4C-C’’, E-E’’ and G-G’’). Collectively, our transgenic assay supports that Eh1, Eh2 and Eh3 are Rbm24 enhancers. Moreover, Pou4f3 mediates the Rbm24 expression primarily by binding to Rbm24 enhancers.

Forced Rbm24 expression fails to alleviate the degeneration of Pou4f3^-/-HCs

Whether ectopic Rbm24 expression can alleviate the degeneration of Pou4f3^-/- HCs remains unknown, albeit we showed that Rbm24 is certainly regulated by Pou4f3. Thus, we established a new conditional Rosa26-Loxp-stop-Loxp-Rbm24*3xHA/+ (Rosa26^Rbm24/+ in short) mouse model (Figure 5A and B). To turn on ectopic Rbm24 in cochlear HCs, Rosa26^Rbm24/+ was further crossed with the Atoh1^Cre/+ that targets majority of cochlear HCs and a fraction of SCs ³⁹. Notably, in the Rosa26^Rbm24/+strain, Rbm24 was tagged with three HA fragments, which facilitated us to distinguish the endogenous and the ectopic Rbm24. In the control Pou4f3^+/+ (Figure 5C-C’’) and the germ line stable Pou4f3^-/- (Figure 5D-D’’) mice, the Myo7a+ HCs did not express HA (Rbm24), however, almost of all Myo7a+ HCs expressed HA (Rbm24) in the Pou4f3^+/+; Atoh1^Cre/+; Rosa26^Rbm24/+mice at P1 (white arrows in Figure 5E-E’’). As expected, besides HCs, we captured a few SCs expressing HA (Rbm24) but not Myo7a (orange arrows in Figure 5E-E’’). Those cochlear HCs expressing HA (Rbm24) seemed normal, suggesting that they can tolerate extra Rbm24 expression by P1. Collectively, it supports that the Rosa26^Rbm24/+model was successfully generated and ectopic Rbm24 was induced in cochlear HCs in Pou4f3^+/+; Atoh1^Cre/+; Rosa26^Rbm24/+.

In contrast to our expectation, ectopic Rbm24 expression failed to mitigate the degeneration of Pou4f3^-/- HCs. Severe degeneration of HCs (both IHCs and OHCs) occurred in Pou4f3^-/-; Atoh1^Cre/+; Rosa26^Rbm24/+(Figure 5F-F’’), indistinguishable from the Pou4f3^-/-mice (Figure 5D-D’’), except that the remaining HCs expressed Rbm24 in the Pou4f3^-/-; Atoh1^Cre/+; Rosa26^Rbm24/+ (arrows in Figure 5F-F’’). It suggests that restoration of Rbm24 alone is not sufficient to prevent Pou4f3^-/- cochlear HCs from undergoing degeneration.

Discussion

Molecular mechanisms underlying cochlear HC survival

Our sound receptor HCs are vulnerable to various genetic mutations, environmental ototoxic factors and ageing. The degeneration of HCs is one of the primary reasons accounting for human sensorineural hearing impairment ⁴⁵. Previous studies have revealed many genes whose mutations lead to HC degeneration starting at different ages. Severe HC developmental defect happens in Atoh1 ^-/- mutants ^{7, 46}. It is known that Atoh1^-/- cochlear sensory cells undergo apoptosis. Whether the absence of HCs in Atoh1^-/- mice is due to that HCs are never produced, or immediately die after initial emergence remains unclear, partly due to the difficulty of unambiguously defining the nascent HCs by molecular markers.

Differently from Atoh1^-/- mutants, in both Pou4f3^-/-and Gfi1^-/- mice ^{31, 33, 34}, the initial cochlear HC development is normal but become defective at perinatal ages, consistent with the notion that Pou4f3 and Gfi1 are not required in HC fate specification but needed in subsequent differentiation and survival. It is known that caspase 3 is active during the hair cell death in the Pou4f3^-/- mutants, and the anti-apoptotic factor z-VAD-fmk has a protective effect on the Pou4f3^-/-HCs between 14.5 and E16.5 ⁴⁷. Moreover, we noticed that the overall HC degeneration in Gfi1^-/- was milder than in Pou4f3^-/- mice at P1, because only OHC degeneration happened in the Gfi1^-/-, but both IHC and OHC were degenerated in Pou4f3^-/- mice (Supplemental Figures 2 and 5).

It is known that Gfi1 is the target of Pou4f3 ³⁴. Whether forced Gfi1 expression can alleviate the degeneration of Pou4f3^-/- HCs remains unknown, too. Likewise, Pou4f3 is one of the target genes regulated by Atoh1 ³², however, whether forced expression of either Pou4f3 or Gfi1, or both Pou4f3 and Gfi1, is able to mitigate the developmental defects of Atoh1^-/- HCs remains to be determined. Future studies are warranted to resolve those remaining questions before further deeply understanding the mechanisms underlying how to keep cochlear HCs to survive.

Roles of Rbm24 and how Rbm24 is regulated during cochlear HC development

Shortly after cochlear HCs emerge, Rbm24 is expressed and permanently maintained, evidenced by transcriptomic analyses and antibody staining ^{23, 35, 48–50}. Rbm24 is not necessary in early phase of cochlear HC development, because Rbm24^-/-HC development is normal by P1, however, Rbm24 is required for OHC to survive after birth ³⁶. Rbm24^-/- OHCs, but not IHCs, are degenerated by P19 ³⁶. Rbm24 is also involved in mRNA stability and pre-mRNA alternative splicing of genes including Cdh23 and Pcdh15 that are crucial for HC stereocilia development ^51–53. Nonetheless, the single-cell transcriptomic analysis of Rbm24^-/- IHCs or OHCs is not yet available and the detailed molecular mechanism underlying Rbm24^-/-HC degeneration remains to be clarified in future studies.

What are the trans-acting factors that are involved in regulating Rbm24 expression? First, our current study provided strong genetic evidence to support that Pou4f3 is needed in turning on Rbm24. In the absence of Pou4f3, Rbm24 expression cannot be triggered. The normal expression of Pou4f3 in Rbm24^-/- HCs further confirmed that Pou4f3 is epistatic to Rbm24 and that Rbm24 is dispensable for Pou4f3 expression. Besides Pou4f3, the engagement of Atoh1 in Rbm24 regulation is supported by two lines of evidence: one is that Rbm24 expression disappears in the Atoh1^-/- cochlear HCs ⁵⁴; the second is that Rbm24 belongs to the Atoh1 binding targets revealed by Atoh1 Cut&Run assay ^{8, 32}. Notably, Atoh1 binds both Rbm24 promoter and its three enhancers Eh1, Eh2 and Eh3 ^{8, 32}, which differs with Pou4f3 which only binds Rbm24 enhancers (Figure 4). Thus, it is likely that Pou4f3 and Atoh1 cooperate to regulate Rbm24, and either Pou4f3 or Atoh1 mutation leads to repression of Rbm24 expression. Whether Rbm24 expression in adult HCs needs Pou4f3 remains to be determined by future conditional Pou4f3 loss-of-function studies.

Gfi1 and Rbm24 expression are independent to each other

We initially hypothesized that Pou4f3 mediated Rbm24 expression through Gfi1. However, this hypothesis is not supported by the observation that Rbm24 expression is normal in the Gfi1^-/- HCs (Figure 2 and Supplemental Figure 5). Moreover, in the absence of Rbm24, Gfi1 expression is not altered in cochlear HCs, either. Thus, Rbm24 and Gfi1 expressions seem independent to each other. It might be due to the functional difference between Pou4f3 and Gfi1 during cochlear HC development. Both Pou4f3 and Gfi1 are needed in promoting expression of genes involved in HC differentiation ^{28, 30, 34, 40}, however, Gfi1, but not Pou4f3, also represses the preceding neural genes expressed in nascent HCs ⁴⁰.

Furthermore, we have another observation relevant to Gfi1 isoforms that is worthy of discussion. Our previous report shows that one efficient sgRNA is enough to induce homozygous gene inactivation by CRISPR-stop ³⁶. Notably, we used 4 different Gfi1 sgRNAs that are distributed in different exons (Figure 2A). We indeed successfully established one Gfi1 mutant by using the sgRNA-5 alone which in principle would effectively pre-stop the Gfi1 translation in exon 1 coding SNAG repressor domain. This Gfi1 mutant (sgRNA-5 alone) was able to yield the HC degeneration phenotype, but the HA (Gfi1) was still detectable in HCs. It is consistent with the notion that Gfi1 mutant lacing the SNAG domain is equivalent to Gfi1-null model ⁵⁵. Thus, it is likely that Gfi1 has multiple unknown isoforms, many of which might be recognized by the HA tag antibody because HA is tagged in the last exon 6. When 4 Gfi1 sgRNAs were combined, we were able to both reproduce the HC degeneration phenotype and completely block the HA (Gfi1) expression.

Why does ectopic Rbm24 fail to alleviate the degeneration of Pou4f3^-/- HCs?

After confirming the epistatic genetic interaction between Pou4f3 and Rbm24, we predicted that ectopic Rbm24 expression should be able to, at least partly, alleviate the degeneration of Pou4f3^-/- HCs, similar to the case between Insm1 and Ikzf2 in cochlear OHCs in our recent study ²¹. However, the actual observation is that Rbm24 failed to rescue the HC degeneration in the Pou4f3^-/-mice. There are two possible interpretations: 1) Despite it is regulated by Pou4f3, Rbm24 is not directly involved in the pathway regulating HC survival. Indeed, the cell death in Rbm24^-/- HCs happens later than in Pou4f3^-/- HCs; 2) Although Rbm24 is a key Pou4f3 downstream target, its forced expression alone cannot compensate loss of Pou4f3 or other known Pou4f3 targets such as orphan thyroid nuclear receptor Nr2f2 and Caprin-1 ^{56, 57}. Caprin-1 is recruited to stress granules in cochlear HCs exposed to ototoxic trauma ⁵⁷.

The potential application of the three Rbm24 enhancers in cochlear HC gene therapy

Gene therapy is a promising strategy to restore the hearing capacity in human inheriting gene mutations ⁴⁵. A few examples have been reported, including Otoferlin and vGlut3 gene replacement ^58–61. Currently, the therapeutic cDNAs were primarily delivered into HCs with an adeno-associated virus (AAV) vector. Despite many AAVs are reported ^{62, 63}, their transfecting cochlear cells is non-selective and HC specific AAV is not available yet. One solution is that, instead of the CAG/CMV ubiquitous promoter widely used in current AAVs ^{64, 65}, either of the three Rbm24 enhancers together with Hsp68 mini promoter (in brief Rbm24-Hsp68) should be able to produce an AAV that would specifically transfect HCs. Moreover, Rbm24 is permanently expressed in cochlear HCs. Given the Rbm24-Hsp68 AAV works as expected, it would be a powerful tool for future HC specific gene therapy at all postnatal ages, with potential applications in treating clinical human deafness.

Methods and materials

Mice

The Atoh1^Cre/+ model was kindly provided by Dr Lin Gan (Augusta University, USA)³⁹. The Rosa26-loxp-stop-loxp-tdTomato (Ai9)/+ strain (Jax#: 007905) was from the Jackson Laboratory. All mice were bred and raised in a SPF-level animal room and all animal procedures were performed according to the guidelines (NA-032-2022) of the IACUC of the Institute of Neuroscience (ION), Center for Excellence in Brain Science and Intelligence Technology, Chinese Academy of Sciences.

One-step generation of homozygous Pou4f3 or Gfi1 or Rbm24 mutants by CRISPR-stop

The detailed protocol of how to use CRISPR-stop to generate homozygous gene mutants is described in our previous report ^{36, 37}. Briefly, efficient pre-tested sgRNAs (Table S1) and hA3ABE3 were co-injected into one-cell stage mouse zygotes that were subsequently transplanted into pseudopregnant female mice giving birth to the founder 0 (F0) mice. With Sanger sequencing of tail PCR, the F0 mice with homozygous mutation (pre-emergence of protein translation stop codon) were identified and immediately ready for analysis.

Generating Pou4f3 or Gfi1-null mutants with large DNA fragment deletion by CRISPR/Cas9

To construct germ line stable null mutants of either Pou4f3 or Gfi1, Cas9 mRNA, two efficient pre-tested sgRNAs respectively located at the proximal and distal end of the targeted gene, together with a single strand DNA donor (120 bp) (Table S3), were co-injected into the one-cell stage wild type zygotes. Notably, the left half (60 bp) of the single strand DNA donor was homologous to the 5’ end, and the right half (60 bp) was to the 3’end of the targeted gene. The post-injected zygotes were transplanted into pseudopregnant females that would give birth to the F0 mice subjected to tail PCR screening (primers were listed in Table S2). The F0 mice harboring the designed large DNA deletion between the two sgRNAs were identified and further bred with wild type mice to establish the germ-line stable mutants (F1 or afterward).

Construction of the Gfi1*3×HA-P2A-Cre knockin mouse strain

The sgRNA against Gfi1 (5ʹ-ATGGACTCAAATGAGTACCC-3ʹ), Cas9 mRNA and the targeting vector (Supplemental Figure 3B) were co-injected into the one-cell stage wild type mouse zygotes. The targeting vector contained three portions, 5’ homologous arm (800 bp), 3’ homologous arm (800 bp) and the part in the middle of 5’ and 3’ arms that contained three HA fragments followed by 2A-Cre. The F0 mice with the potential gene targeting (Supplemental Figure 3C) were screened by tail PCR, followed by crossing with wild type mice to produce F1 mice. Those F1 mice were confirmed by tail PCR again, and further subjected to Southern blot. The detailed protocol of Southern blot is described in our previous report ⁶⁶. Tail DNA PCR was used for routine genotyping and primers used to distinguish between KI (Gfi1^HA-Cre/+) and WT alleles were F4, R5 and R6 (Table S2).

Construction of Eh1-EGFP+, Eh2-EGFP+ and Eh3-EGFP+ transgenic reporter lines

All three transgenic reporter mouse lines, Eh1-EGFP+, Eh2-EGFP+ and Eh3-EGFP+ were produced by the same experimental procedures. The core DNA sequences of each enhancer (Eh1 or Eh2 or Eh3 in Figure 4A), mRNA of PiggyBac transposase, and the PiggyBac vector (Figure 4B, D and F) were co-injected into the one-cell stage of wild type mouse zygotes. The PiggyBac vector contained Eh1/Eh2/Eh3 core DNA sequences (Table S3), mini promoter of mouse heat shock protein 68 (Hsp68) and the EGFP coding sequence. The PiggyBac vector was randomly integrated into mouse genome by the transposase. The F0 mice containing the PiggyBac vector were screened by tail PCR and further bred with wild type mice to produce germ line stable F1 transgenic reporter strains. The samples analyzed in Figure 4B-G’’ were from F1 or afterwards. Primers used for genotyping transgenic reporter strain were F8, F9 and F10 and a common primer R11 and primer sequences were listed in Table S2.

Generation of Rosa26^{CAG-lsl-Rbm24*3×HA/+} mouse models

The Rosa26^{CAG-lsl-Rbm24*3×HA/+} (Rosa26^Rbm24/+) knockin mouse strain was constructed by homologous recombination mediated by CRISPR/Cas9. The pre-tested Rosa26 sgRNA (5ʹ-ACTCCAGTCTTTCTAGAAGA-3ʹ), targeting vector (Figure 5A), together with the mRNA of Cas9 were co-injected into the one-cell stage wild type mouse zygotes. Similarly, F0 mice with potentially correct gene targeting were screened by tail PCR and further bred with wild type mice to establish germ line stable F1 mice. The primers used to distinguish wild type and the KI (Rosa26^Rbm24/+) alleles were F12, R13 and R14 (Table S2).

Sample processing and immunofluorescence assay

Inner ears were dissected and fixed in 4% paraformaldehyde (PFA) in PBS (E607016-0500, Sangon Biotech) at 4℃ overnight. For cochlear cryosections, inner ears were dehydrated in 30% sucrose (V900116, Sigma) at 4℃ before embedding into the optimal cutting temperature (OCT) compound (4583, SAKURA) and slices were cut with 14 μm thickness. The detailed protocol of immunofluorescent staining is described in our previous report ⁶⁷. The following primary antibodies were used: anti-Rbm24 (rabbit, 1:500, 18178-1-AP, Proteintech); anti-Pou4f3 (mouse, 1:500, sc-81980, Santa Cruz); anti-Pou4f3 (rabbit, 1:500, NBP1-88349, Novus Biologicals); anti-HA (rat, 1:500, 11867423001, Roche); anti-Myo7a (rabbit, 1:500, 25-6790, Proteus Biosciences); anti-Myo7a (mouse, 1:500, MYO7A 138-1, Developmental Studies Hybridoma Bank); anti-Ctip2 (Bcl11b) (rat, 1:500, ab18465, Abcam); anti-Insm1 (guinea pig, 1:6000, a kind gift from Dr. Carmen Birchmeier from Max Delbrueck Center for Molecular Medicine, Germany); anti-GFP (chicken, 1:500, ab13970, Abcam). Different corresponding Alexa Fluor-conjugated secondary antibodies were used for detecting the primary antibodies above. Hoechst 33342 (1:1000, H3570, Thermo Fisher Scientific) was used for nuclear DNA staining.

Lastly, either whole-mount or cryosection samples were mounted with Prolong Gold antifade medium (P36930, Thermo Fisher Scientific) at room temperature for 12 h. Samples were scanned by the Nikon C2 or Nikon NiE-A1 Plus confocal microscope. ImageJ software was used to process the confocal images.

Cell quantification and statistical analysis

Before immunofluorescent staining, each cochlear sample was grossly separated into three portions but with different length. Each portion of the same cochlea was initially scanned with confocal microscope but at a low magnification (10◊lens). After calculating the total length of each cochlea, it was precisely divided into basal, middle and apical turn with equal length. Then, for cell counting in the experiments (Supplemental Figure 2D), in each turn ∼200μm length of the sensory epithelium was scanned with confocal microscope but at a high magnification (60◊lens) by which the number of HCs was counted. In the Pou4f3^-/- cochleae, the percentage of the surviving HCs was calculated by normalizing the number of the remaining HCs to the counterparts in the wild type mice. All cell numbers were presented as means ± SEM. For statistical analyses, we used GraphPad Prism 6.0 software and performed Student’s t tests with Bonferroni correction.

Supplemental Figure Legends

Supplemental Figure 1.

Supplemental Figure 2.

Supplemental Figure 3.

Supplemental Figure 4.

Supplemental Figure 5.