In situ regeneration of inner hair cells in the damaged cochlea by temporally regulated co-expression of Atoh1 and Tbx2

Sound detection is achieved through hair cells (HCs) in the cochlea, the auditory organ located in the ventral part of the inner ear. The cochlear auditory epithelium, also known as the organ of Corti (OC), harbors one row of inner HCs (IHCs) on its medial side and three rows of outer HCs (OHCs) on the lateral side (Groves et al., 2013; Luo et al., 2021; Montcouquiol and Kelley, 2020; Wu and Kelley, 2012). OHCs are sound amplifiers and specifically express Prestin, a motor protein encoded by Slc26a5 (Sun et al., 2022b; Zheng et al., 2000); Slc26a5^−/− mice display severe hearing impairment (Liberman et al., 2002). Conversely, IHCs are primary sensory cells that specifically express vGlut3, which is encoded by Slc17a8; Slc17a8^−/− mice are completely deaf (Ruel et al., 2008; Seal et al., 2008) because vGlut3 is essential for the release of the excitatory neurotransmitter glutamate through the unique ribbon synapse structure formed between IHCs and auditory spiral ganglion neurons (SGNs) (Li et al., 2018a). Notably, both OHCs and IHCs are crucial for sound detection and their damage causes permanent deafness in mammals, which have lost HC regeneration capacity (Atkinson et al., 2015; Janesick and Heller, 2019; Stone and Cotanche, 2007). Intriguingly, the bHLH transcription factor Atoh1 is necessary for cochlear pan-HC development: whereas all HCs disappear in Atoh1^−/− mice (Bermingham et al., 1999; Li et al., 2022; Luo et al., 2022), Atoh1 overexpression induces supernumerary new HCs (Gubbels et al., 2008; Iyer et al., 2022; Kelly et al., 2012; Liu et al., 2012a; Zheng and Gao, 2000). Furthermore, Insm1 and Ikzf2 play crucial roles in specifying or stabilizing the OHC fate (Chessum et al., 2018; Li et al., 2023; Sun and Liu, 2023; Wiwatpanit et al., 2018), and our group and two other groups have independently shown that Tbx2 is required in IHC fate specification, differentiation and maintenance (Bi et al., 2022; García-Añoveros et al., 2022; Kaiser et al., 2022).

Adjacent to HCs, the following distinct subtypes of supporting cells (SCs) are located from the medial to lateral side: inner border cells (IBCs), inner phalangeal cells (IPhs), pillar cells (PCs) and Deiters' cells (DCs) (Luo et al., 2021). IBCs and IPhs belong to the medial SC subgroup and their gene expression profiles and developmental origins are distinct from those of lateral PCs and DCs (Basch et al., 2016; Kolla et al., 2020). Whereas Prox1 is transiently and Fgfr3 is permanently expressed in PCs and DCs (Bermingham-McDonogh et al., 2006; Fritzsch et al., 2010; Kolla et al., 2020; Liu et al., 2012a,b), proteolipid protein 1 (encoded by Plp1) and solute carrier family 1 member 3 (Slc1a3) are enriched in IBCs and IPhs (Glowatzki et al., 2006; Gómez-Casati et al., 2010). PCs and/or DCs and OHCs have been suggested to be born from the same lateral progenitors (Kolla et al., 2020), and, accordingly, we have previously shown that PCs and DCs can be converted into OHC-like cells through concurrent expression of Ikzf2 and Atoh1 (Sun et al., 2021). Similarly, IBCs and/or IPhs and IHCs share identical medial progenitors (Basch et al., 2016; Kolla et al., 2020), and we have reported that neonatal IBCs and/or IPhs are reprogrammed into IHCs expressing the early marker Myo6 but not vGlut3 through the ectopic expression of Atoh1 alone (Liu et al., 2014), and into Myo7a+ and/or vGlut3+ IHCs through the dual expression of Tbx2 and Atoh1 in wild-type cochleae (Bi et al., 2022). However, because a suitable genetic model is lacking, no study has thus far reported the transcriptomic, electrophysiological and morphological characteristics of new IHCs generated in the damaged cochlea, mimicking the HC regeneration that naturally occurs in nonmammalian vertebrates such as chicken and fish (Atkinson et al., 2015; Corwin and Cotanche, 1988). Moreover, whether new IHCs can enable hearing improvement is unknown; if they cannot, it is crucial to identify the main barrier preventing hearing recovery.

By exploiting a well-designed genetic mouse model, we damaged endogenous neonatal IHCs and then conditionally induced Atoh1 and Tbx2 expression specifically in neonatal IBCs and/or IPhs. On average, 477 new IHCs were generated per cochlea, and both single-cell transcriptomic and electrophysiological assays revealed that the new IHCs resembled wild-type IHCs. However, mechanoelectrical transduction (MET) was defective in the new IHCs, and this is at least one of the main barriers that prevented us from achieving hearing improvement. Our results highlight future directions for methods to regenerate functional IHCs and realize hearing restoration in vivo.

To achieve specific IHC damage in neonatal cochleae, we generated a new knock-in mouse model, Fgf8-P2A-DTR-T2A-DTR/+ (briefly, Fgf8-DTR/+), in which diphtheria toxin (DT) receptor (DTR) expression is strictly controlled by Fgf8 promoters and enhancers (Fig. S1A-C). Fgf8 is highly and exclusively expressed in IHCs within perinatal cochleae (Jacques et al., 2007; Pan et al., 2023; Quadros et al., 2017; Ratzan et al., 2020), and transient DT administration would cause the death of DTR-expressing HCs (Cox et al., 2014; Golub et al., 2012; Hu et al., 2016; Tong et al., 2015). Southern blotting confirmed the absence of random insertion of the targeting vector in the mouse genome (Fig. S1D,E), and tail-DNA PCR readily distinguished the wild-type and knock-in alleles (Fig. S1F). Both heterozygous and homozygous mice were fertile and healthy, and did not exhibit any abnormalities. These findings also indicated that Fgf8 expression itself was intact because Fgf8^−/− mice die at ∼E9.5 (Sun et al., 1999). Thus, we predicted that IHCs and OHCs would be normal in Fgf8-DTR/+ mice in the absence of DT treatment (no DT), and that IHCs, but not OHCs, would be lost upon administration of DT (15 ng/g body weight) at P2.

Our prediction of IHC loss was experimentally confirmed at P42 (Fig. 1A-B′) and IHC quantification across entire cochleae revealed that only 16.3±6.7 (n=4) IHCs were present in DT-treated Fgf8-DTR/+ mice at P42, significantly fewer than the 732.0±15.4 (n=3) IHCs in no-DT littermates (Fig. 1C); thus, ∼97.8% of IHCs were lost after a single administration of DT at P2. Moreover, considerable IHC loss occurred in all cochlear turns (Fig. 1D). By contrast, three rows of well-aligned OHCs were present (Fig. 1A-B′), and OHC numbers did not differ between no-DT and DT-treated groups (Fig. 1E). Finally, auditory brainstem response (ABR) assays revealed that hearing thresholds at all frequencies in DT-treated mice (n=7) were drastically higher than those in no-DT mice (n=8) at P42 (Fig. 1F). Collectively, our results indicate that Fgf8-DTR/+ is a powerful genetic model that is suitable to specifically damage IHCs in the neonatal mouse cochlea.

We have previously shown that 29.5±1.2% of neonatal cochlear IBCs and/or IPhs are converted into vGlut3-expressing IHCs in the undamaged cochlea upon dual expression of Atoh1 and Tbx2 (Bi et al., 2022), which is achieved using Plp1-CreER+; Rosa26-CAG-Loxp-Stop-Loxp-Tbx2*3×V5-P2A-DHFR-Atoh1*3×HA-DHFR-T2A-tdTomato/+ (briefly, Plp1-TAT) (Fig. S1G). Plp1-CreER+ is a recognized IBC- and/or IPh-specific Cre driver within the OC (Gómez-Casati et al., 2010), and TAT denotes Tbx2 (fused with V5 tag), Atoh1 [fused with HA tag and protein unstable domain derived from dihydrofolate reductase (DHFR)] and tdTomato. If a protein of interest is fused with DHFR containing the destabilizing domains (DDs), the protein will undergo rapid proteasomal degradation (Iwamoto et al., 2010); however, treatment with trimethoprim (TMP), a cell-permeable small molecule that can bind to and stabilize the DDs, will result in the protein degradation being stopped in a rapid, reversible and TMP dose-dependent manner (Iwamoto et al., 2010). Thus, in the current study, permanent Tbx2 expression and transient Atoh1 expression (hereafter ‘Tbx2 and Atoh1 expression’) and permanent tdTomato expression were conditionally and specifically induced in neonatal IBCs and/or IPhs upon exposure to tamoxifen (TMX) at P0 and P1, and then to TMP (for stabilizing Atoh1 transiently) at P3 and P4 (Fig. S1G). The tdTomato expression helped to distinguish new IHCs from endogenous IHCs, which do not express tdTomato. The success of the transient Atoh1 expression was supported by two key lines of evidence described in our previous study (Bi et al., 2022): (1) Atoh1 protein was very faint or undetectable when Plp1-TAT mice were administered TMX alone; and (2) high Atoh1 protein expression was detected when Plp1-TAT mice were administered both TMP and TMX. Notably, Atoh1 protein expression was high when cochlear samples were analyzed 3 h after the last TMP treatment, but became very weak or undetectable 3 days later. Therefore, a single dose of TMP can stabilize Atoh1 protein for up to 3 days at most.

We determined here whether specific damage of endogenous IHCs further boosts the reprogramming efficiency of Atoh1 and Tbx2 (Fig. 2A). Plp1-TAT; Fgf8-DTR/+ (Plp1-TAT-DTR, in short) mice were either (1) administered DT only and used as the control group (Fig. 2B-B″); or (2) administered TMX at P0 and P1, DT at P2, and TMP at P3 and P4, and defined as the TMX/DT/TMP-treated group (Fig. 2C-C″). Importantly, to accurately label IBCs and/or IPhs for subsequent fate-mapping analysis, TMX was administered before DT treatment (Fig. 2A), because the opposite order could potentially alter the Cre expression pattern in Plp1-CreER+.

As in DT-treated Fgf8-DTR/+ mice (Fig. 1B,B′), only 17.3±4.0 (n=4) IHCs were present in DT-treated Plp1-TAT-DTR mice at P42 (arrows in Fig. 2B-B″ and Fig. S2A-A‴). By contrast, 477.0±48.0 (n=4) new IHCs, which were tdTomato+/vGlut3+/Prestin− (Fig. 2C-C″ and Fig. S2B-B‴), were detected in entire cochleae of TMX/DT/TMP-treated Plp1-TAT-DTR mice at P42. Intriguingly, the number of vGlut3+ new IHCs here (477.0±48.0) was ∼2.2-fold higher than the corresponding number (221.8±19.1) in Plp1-TAT mice in which endogenous IHCs were not damaged (Bi et al., 2022). Accordingly, normalizing the new IHC numbers against the total number of tdTomato+ cells (close to new IHCs) revealed that 73.5±2.3% of the tdTomato+ cells (which were originally IBCs and/or IPhs) were reprogrammed into vGlut3+ new IHCs in TMX/DT/TMP-treated Plp1-TAT-DTR mice, which was also markedly higher than the 29.5±1.2% measured in Plp1-TAT mice (Fig. 2D). Based on these results, we concluded that the reprogramming efficiency of Atoh1 and Tbx2 in neonatal IBCs and/or IPhs was drastically increased when endogenous IHCs were damaged.

Next, we determined whether ribbon synapses were formed between new IHCs and SGNs. Ctbp2+ ribbon synapses were quantified based on co-staining for Ctbp2 and Myo7a in wild-type mice (Fig. 2E,E′) and co-labeling for Ctbp2 and tdTomato (Fig. 2F,F′) in TMX/DT/TMP-treated Plp1-TAT-DTR mice at P42. The average numbers of synapses per new IHC were similar (at 8 kHz) or slightly lower (at 16 and 30 kHz) relative to those per endogenous IHC (Fig. 2G).

Besides Ctbp2 staining, we performed triple labeling for neurofilament (NF)-200, tdTomato and Myo7a, which revealed that the NF-200 signal was lower in DT-treated Plp1-TAT-DTR mice (Fig. S2C-C″) than in TMX/DT/TMP-treated Plp1-TAT-DTR mice (Fig. S2D-D″) at P42. However, the NF-200 signal was the strongest in the control Plp1CreER+; Rosa26-LSL-tdTomato (Ai9)/+ (Plp1-Ai9) mice that were also administered TMX and TMP (Fig. S2E-E″). Furthermore, in DT-treated Plp1-TAT-DTR mice, the NF-200 signal appeared weaker in the regions where IHCs were completely lost (yellow arrows in Fig. S2C-C″) than in regions near the remaining IHCs (red arrows in Fig. S2C-C″). Collectively, our results support the conclusion that ribbon synapses are formed between IHCs and SGNs in Plp1-TAT-DTR mice at P42, although the synapse numbers are lower than in the control Plp1-Ai9 mice.

In nonmammalian vertebrates such as chicken and fish, hearing ability can be restored after damage (Corwin and Cotanche, 1988; Janesick and Heller, 2019; Stone and Cotanche, 2007). This occurs through two processes: (1) transdifferentiation, during which SCs located near damaged HCs directly switch their SC fate to HC fate without proliferation; and (2) mitotic regeneration, during which the SCs first proliferate and expand their populations before starting to transdifferentiate into auditory HCs. To ascertain which process is used in our model, we analyzed Plp1-TAT-DTR mice that, besides being treated with TMX at P0 and P1, DT at P2, and TMP at P3 and P4, were administered 5-ethynyl-2′-deoxyuridine (EdU). Moreover, we used two paradigms to cover four distinct time-points: (1) EdU administration at P5 and P7, and analysis at P9 (n=3, Fig. S3A-B‴); and (2) EdU administration at P7, P9 and P11, and analysis at P12 (n=3, Fig. S3C-D‴).

In both experimental paradigms, no EdU incorporation was detected in the tdTomato+/vGlut3+ new IHCs (arrows in Fig. S3B-B‴ and D-D‴). Notably, however, successful EdU administration was confirmed by the presence of EdU+ cells in the cochlear SGN area (Fig. S3A,C). These results suggest that the vast majority of the new IHCs are derived from direct transdifferentiation rather than mitotic regeneration.

We first used scanning electron microscopy (SEM) to determine whether hair bundles or stereocilia were present on the surface of new IHCs at P42. In control Plp1-Ai9 mice, the typical stereocilia of endogenous OHCs exhibited a ‘V’ or ‘W’ shape (pink), and the morphology of stereocilia in IHCs resembled ‘bird-wing’ (purple) (Fig. 3A,A′). In Plp1-TAT-DTR mice subject to no treatment, we detected a single row of IHCs (Fig. 3B,B′), whereas, in accordance with the observation in Plp1-TAT mice (Bi et al., 2022), we detected extra IHCs (blue) when TMX and TMP, but not DT, were administered to Plp1-TAT-DTR mice (Fig. 3C,C′). As expected, ‘bird-wing’-like IHC stereocilia were absent in DT-treated Plp1-TAT-DTR mice (Fig. 3D,D′) but were recovered in TMX/DT/TMP-treated Plp1-TAT-DTR mice at P42 (Fig. 3E,E′).

We next used transmission electron microscopy (TEM) to further confirm the presence of ribbon synapses in new IHCs (Fig. 3F-G′). Consistent with our immunostaining results (Fig. 2 and Fig. S2), ribbon synapses featuring high electron density were captured in both control Plp1-Ai9 and TMX/DT/TMP-treated Plp1-TAT-DTR mice (red dotted circles in Fig. 3F′,G′). However, ABR measurements did not reveal significant hearing improvement in TMX/DT/TMP-treated Plp1-TAT-DTR mice (n=30) relative to DT-treated Plp1-TAT-DTR mice (n=10) (Fig. 3H). Notably, the hearing thresholds at all frequencies in the 30 Plp1-TAT-DTR were also significantly higher than those in control Plp1-Ai9 mice (blue line in Fig. 3H) at P42. Thus, we concluded that in the 477.0±48.0 new IHCs, further manipulations are necessary to achieve improved hearing recovery.

We performed whole-cell patch-clamp recording of individual new IHCs or wild-type IHCs. Both endogenous IHCs in wild-type mice and new IHCs in Plp1-TAT-DTR mice were recorded at P42. First, both wild-type IHCs (Fig. 4A) and new IHCs (Fig. 4B) exhibited voltage-dependent Ca²⁺ currents, although current-amplitude variations were larger in the new IHCs (n=10, cell number) than in wild-type IHCs (n=5) (Fig. 4C). Second, linear capacitance, which is indicative of cell size, of new IHCs (n=9) was smaller than that of wild-type IHCs (n=9) (Fig. 4D), which is consistent with the new IHCs being derived from IBCs and/or IPhs, both originally smaller than endogenous IHCs. Third, exocytosis-triggered change in membrane capacitance, a readout for the release of glutamate-containing vesicles, was measured when depolarization was induced in both wild-type and new IHCs (Fig. 4E), and, moreover, surface-area increments in wild-type and new IHCs were comparable (Fig. 4F), which indicates that normal sustained vesicle-release functions were established in the new IHCs.

We also measured MET currents in the new IHCs (Fig. 4G). At P42, IHCs (n=5, cell number) from wild-type mice exhibited fluidjet-dependent MET currents (blue color in Fig. 4H,I), whereas only very weak (low signal-to-noise ratio) or background MET currents were induced in the new IHCs (n=6) (red color in Fig. 4H,I). This could account for the failure to achieve measurable hearing improvement in the Plp1-TAT-DTR mice (Fig. 3H). Nonetheless, because the new IHCs exhibited voltage-dependent Ca²⁺ currents and neurotransmitter release, we hypothesized that the global gene expression profiles of new IHCs would diverge from those of both neonatal IBCs and/or IPhs (origin of new IHCs) and adult IBCs and/or IPhs (final fate of neonatal IBCs and/or IPhs if ectopic Atoh1 and Tbx2 are not expressed), and, conversely, that the new IHCs would resemble adult endogenous IHCs. To test this, we performed single-cell RNA-seq analysis (below).

We aimed to globally quantify the number of IHC genes upregulated and IBC and/or IPh genes downregulated in new IHCs relative to IBCs and/or IPhs. Achieving this goal required us to: (1) determine the transcriptomic profiles of new IHCs and IBCs and/or IPhs by using the same transcriptomic analysis approach; and (2) define the IHC and IBC and/or IPh genes, respectively. We obtained the transcriptomic profiles of adult endogenous IBCs and/or IPhs at P42 (P42_IBCs/IPhs, for short) by applying 10× Genomics single-cell RNA-seq to the cochlear tissues of Plp1-Ai9 mice. A total of 4051 qualified cells were obtained and divided into 18 clusters by using t-distributed stochastic neighbor embedding (t-SNE) analysis (Fig. S4A); an average of 3133 genes were detected per cell; tdTomato was enriched in the cells of clusters 0, 7, 10 and 12, confirming that these were cells of the Plp1+ lineage (Fig. S4B), and the cells of clusters 0, 7 and 12 were cochlear glial cells. Here, we focused on only cluster 10 cells, in which Epcam and Slc1a3 were enriched (Fig. S4C,D): Epcam is a marker specifically expressed in the cochlear epithelium (Hertzano et al., 2011; Roccio et al., 2018; Sun et al., 2022a); Slc1a3 encodes a glutamate-aspartate transporter (Glast), a known marker of IBCs and/or IPhs (Glowatzki et al., 2006; Liu et al., 2014). Ultimately, we selected 55 cells in cluster 10 and defined these as P42_IBCs/IPhs. Furthermore, we defined 423 IBC and/or IPh genes that showed significant enrichment in IBCs and/or IPhs relative to that in all other non-IBC and/or IPh cell populations. The gene list is included in Table S1.

Similarly, the transcriptomic profiles of new IHCs were obtained by applying the 10× Genomics single-cell RNA-seq to cochlear samples from TMX/DT/TMP-treated Plp1-TAT-DTR mice at P42. A total of 4661 qualified cells were obtained and further divided into 19 clusters (Fig. S4E); an average of 2862 genes were detected per cell. Briefly, the 41 cells in cluster 15, which were tdTomato+/Myo7a+/vGlut3 (or Slc17a8)+ (arrows in Fig. S4F-H), were defined as new IHCs (P42_new IHCs, for short). Furthermore, 432 genes were classified as IHC genes by integrating three available datasets (Bi et al., 2022; Li et al., 2018b; Sun et al., 2021); these genes were highly expressed [transcripts per million (TPM) >16] in IHCs, but not expressed in cochlear PCs, DCs and IBCs and/or IPhs. The 432 IHC genes are listed in Table S1.

Expression levels of 626 genes were significantly higher (log₂FC>0.5 and P<0.05) in P42_new IHCs than in P42_IBCs/IPhs (Fig. S5A), and gene ontology (GO) analysis revealed that these genes were enriched in functions involved in, for example, sensory perception of sound, auditory receptor cell differentiation and vesicle-mediated transport (Fig. S5B); the genes enriched in each GO category are included in Table S2. Furthermore, 186/626 (29.7%) genes were among the 432 IHC genes and included the IHC-specific genes Slc17a8, Otof and Slc7a14, and the pan-HC genes Cib2, Lhfpl5, Espn, Myo6 and Myo7a (pink arrows in Fig. S5A). All 626 and 186 genes are listed in Table S3. Thus, we conclude that 43.1% (186/432) of the IHC genes are substantially upregulated in P42_new IHCs.

Conversely, the expression levels of 337 genes were significantly lower (log₂FC>0.5 and P<0.05) in P42_new IHCs than in P42_IBCs/IPhs (Fig. S5A), and GO analysis revealed that these genes were over-represented in positive regulation of cell proliferation and negative regulation of the Notch pathway (Fig. S5C); the genes in each GO category are included in Table S2. More importantly, 191/337 (56.7%) genes were included among the 423 IBC and/or IPh genes, such as Slc1a3, Gjb2, S100a1, Sox2 and Sox10 (blue arrows in Fig. S5A); these 337 and 191 genes are all listed in Table S3. Therefore, we conclude that 45.2% (191/423) of the IBC and/or IPh genes, which otherwise would be expressed in P42_IBCs/IPhs, are significantly downregulated in P42_new IHCs.

To determine the differentiation status of P42_new IHCs, we mixed the P42_new IHCs (n=41) with E16_IHCs (n=106), P1_IHCs (n=103), P7_IHCs (n=79) and P30_IHCs (n=50). The E16_IHCs, P1_IHCs and P7_IHCs were from the dataset of one previous single-cell cochlear transcriptomic study (Kolla et al., 2020), and the P30_IHCs were from our previous report (Bi et al., 2022). Briefly, six cell clusters were formed, with 98% (49/50) of the P30_IHCs belonging to cluster 3 and 100% (41/41) of the P42_new IHCs sorting into cluster 4 (Fig. 5A); notably, the P42_new IHCs and P30_IHCs aggregated close to each other (dotted circle in Fig. 5B). Thus, we concluded that the global gene-expression profile of P42_new IHCs most closely resembles that of P30_IHCs, even though these cells were assigned to two distinct clusters, suggesting the existence of molecular differences between P42_new IHCs and P30_IHCs. Moreover, trajectory analysis of all IHCs together revealed the maturation process from E16_IHCs to P30_IHCs (black arrows in Fig. 5C), and the calculated pseudotime matched the actual ages of the IHC lineage (Fig. 5D). Collectively, our results supported the conclusion that P42_new IHCs most closely resemble endogenous mature P30_IHCs.

Finally, we characterized the disparity between P42_new IHCs and P30_IHCs. Because the gene-expression profiles of the cells were obtained using two distinct methods (10× Genomics and smart-seq), the profiles cannot be compared directly. Thus, we instead focused on the 432 IHC genes and 423 IBC and/or IPh genes (Table S1), and assessed whether a given IHC or IBC and/or IPh gene was expressed in P42_new IHCs. A gene featuring a normalized expression value of >0.5 was regarded as being highly expressed.

In P42_new IHCs, 249/432 (57.6%) IHC genes were highly expressed and included IHC-specific genes such as Otof, Slc17a8 and Slc7a14, and pan-HC genes such as Lmo7 and Pou4f3 (green arrows in top half of Fig. S6A); conversely, the remaining 183/432 (42.4%) genes were defined as being expressed at low levels in P42_new IHCs (bottom half of Fig. S6A). All 249 and 183 genes are listed in Table S4. Moreover, we specifically selected 10 MET and related tip-link component genes (Liu et al., 2021; Qiu and Müller, 2022) and determined their expression level in P42_new IHCs by using IBCs and/or IPhs as the reference. All the genes selected here, except Tmc2, are expressed in adult IHCs (Bi et al., 2022). We found that five genes – Cib2, Lhfpl5, Tmc1, Tomt and Myo7a – were significantly (log₂FC>2, P<0.05) upregulated in P42_new IHCs, relative to P42_IBCs/IPhs (Fig. S6C). The detailed statistic comparison of the 10 genes between P42_new IHCs and P42_IBCs/IPhs is listed in Table S4.

We further hypothesized that if P42_new IHCs do not fully resemble P30_IHCs, then new IHCs would continue to express a fraction of IBC and/or IPh genes. Accordingly, among the 423 IBC and/or IPh genes, 168 (39.7%) were expressed at a high level, including S100a1, Id1, and Id4 (green arrows in Fig. S6B), whereas the remaining 255/423 (60.3%) genes were expressed at undetectable or low levels in P42_new IHCs and included genes such as Sox10 and Hes1 (gray arrows in Fig. S6B). These 168 and 255 genes are also listed in Table S4.

In Fig. 6, we summarize the upregulation and downregulation of IHC genes and IBC and/or IPh genes during cell-fate conversion, as well as the molecular difference between P42_new IHCs and adult endogenous IHCs. Currently, the overall transcriptomic similarity between P42_new IHCs and P30_IHCs is ∼60%. Future efforts aimed at diminishing the disparities between these cells are required to optimize the regenerated new IHCs.

To the best of our knowledge, this is the first study that has attempted to regenerate new IHCs in the damaged cochlea in which endogenous wild-type IHCs were pre-ablated. Our findings provide compelling evidence that persistent Tbx2 expression coupled with transient Atoh1 expression efficiently transform neonatal IBCs and/or IPhs into IHCs expressing the functional marker vGlut3. Notably, the new IHCs resemble endogenous IHCs in various molecular, ultrastructural and electrophysiological aspects, but the MET of the new IHCs is defective, which precludes the restoration of hearing capacity in the damaged cochlea. Although the defective MET might not represent the only barrier preventing hearing impairment from being alleviated, this defect must be overcome in future IHC regeneration studies.

Fgf8-DTR/+ mice survive well after a single DT treatment, even though Fgf8 is also expressed in other organs besides the ear. In terms of damaging neonatal IHCs, Fgf8-DTR/+ offers advantages over the vGlut3-DTR/+ strain (Xia et al., 2021), because Fgf8 shows IHC-specific expression whereas vGlut3 is expressed in both IHCs and glial cells in neonatal cochleae (Li et al., 2018a; Quadros et al., 2017). However, vGlut3-DTR/+ is a valuable model for damaging adult IHCs, because the expression of Fgf8 is turned off but that of vGlut3 is maintained in adult IHCs (Li et al., 2018a; Wiwatpanit et al., 2018). Including our previous Slc26a5-DTR/+ model designed to specifically damage postnatal cochlear OHCs (Sun et al., 2021), the three knock-in mouse strains are sufficient to allow the damage of cochlear IHCs or OHCs at all postnatal ages.

Reprogramming of SCs into HCs is an effective process for hearing recovery after trauma in nonmammalian vertebrates (Atkinson et al., 2018; Chen et al., 2021). Our previous work has shown that persistent Atoh1 overexpression alone can convert IBCs and/or IPhs into IHCs expressing Myo6 and/or Myo7a, but not vGlut3 (Liu et al., 2014). By comparison, persistent Tbx2 expression coupled with transient Atoh1 expression in IBCs and/or IPhs not only induced IHCs more effectively, but also yielded new IHCs in a more differentiated state (vGlut3+), either in the presence of IHC pre-damage (current study) or absence IHC pre-damage (Bi et al., 2022). This raises the interesting question of why additional Tbx2 expression is required alongside that of Atoh1, because Tbx2 is expressed in endogenous IBCs and/or IPhs (Bi et al., 2022; García-Añoveros et al., 2022; Kaiser et al., 2022). Although a clear answer is currently unavailable, we suspect that Tbx2 could produce a dose-dependent effect and that Tbx2 present at higher levels might more effectively cooperate with Atoh1 to trigger the cell-fate switch from IBCs and/or IPhs to IHCs. Notably, vGlut3+ IHC-like cells are yielded from the SCs overexpressing Atoh1 (moderate level) driven from the Rosa26 locus with the Sox9-CreER driver (Iyer et al., 2022). Thus, it suggests that the Atoh1 dose is also a key factor that determines the differentiation status of the new IHCs. In other words, it is possible that our previous transgenic model has multiple copies of Atoh1 (Liu et al., 2012a), and that extremely high levels of Atoh1 prevent new IHCs from expressing vGlut3 (Liu et al., 2014). Nonetheless, a new Rosa26-CAG-Loxp-Stop-Loxp-DHFR-Atoh1-DHFR/+ model is warranted to clarify the key factors that mediate the differentiation status of the new IHC-like cells.

Whether damaging endogenous IHCs can augment the reprogramming efficiency of Tbx2 and Atoh1 has remained unknown. By using the new Fgf8-DTR/+ strain, we obtained results clearly showing that the reprogramming efficiency (73.5%±2.3%) of Tbx2 and Atoh1 expression in the presence of IHC damage was markedly higher than that in the absence of HC damage (29.5%±1.2%), and, as expected, larger numbers of new IHCs were generated in the presence of IHC damage than in its absence. How might IHC damage enhance the reprogramming efficiency of Tbx2 and Atoh1? We propose that: (1) certain unknown components released from the dying IHCs might stimulate the response of IBCs and/or IPhs to Tbx2 and Atoh1 manipulation; or (2) intercellular signaling, such as Notch signaling (Iyer et al., 2022), from IHCs to IBCs and/or IPhs might prevent the IBCs and/or IPhs that express Atoh1 and Tbx2 from adopting the IHC fate, with these inhibitory effects being lost upon IHC ablation. It has also been recently reported that absence of HCs might change the epigenetic status and establish a permissive state for HC genes to be reactivated in cochlear SCs (Nguyen et al., 2023).

Marked hearing improvement at all frequency regions was not achieved using our current experimental paradigm, which raises this question: to what extent do the new IHCs resemble endogenous IHCs? We grossly addressed this by quantifying the upregulation and downregulation of IHC genes and IBC and/or IPh genes, respectively, in the new IHCs (Fig. 6A,B): 43.1% (186/432) of the IHC genes were significantly upregulated and 45.2% (191/423) of the IBC and/or IPh genes were significantly downregulated in P42_new IHCs relative to P42_IBCs/IPhs, which suggested a cell-fate conversion of ∼45%; however, in the P42_new IHCs, 42.4% (183/432) of the IHC genes were still expressed at low levels and, conversely, 39.7% (168/423) of the IBC and/or IPh genes were expressed at high levels (Fig. 6C), suggesting a cell-fate conversion of ∼60%. Considering both calculations, we conclude that the cell-fate conversion rate lies between 45% and 60%. Notably, this estimation of cell-fate conversion is simplified and solely based on transcriptomic analysis.

Intriguingly, five genes overlapped between the 183 and 186 IHC genes mentioned above (Fig. 6D), suggesting that these five genes were not sufficiently upregulated to match their expression level in mature endogenous IHCs; conversely, 60 genes overlapped between the 168 and 191 IBC and/or IPh genes (Fig. 6E), indicating that these 60 genes were not sufficiently downregulated. A similar phenomenon has been recorded in the cell-fate conversion from adult PCs and/or DCs into Prestin+ OHC-like cells (Sun et al., 2021). Together, these findings suggest that further upregulating IHC genes and downregulating IBC and/or IPh genes in new IHCs would facilitate the differentiation of the new IHCs.

Cochlear IHCs perform two main functions in the hearing process: signal detection, through MET channels in stereocilia at their top surface (Chan and Hudspeth, 2005; Qiu and Müller, 2022; Richardson et al., 2011); and signal transduction, mediated by ribbon synapses at their bottom (Fettiplace, 2017; Özçete and Moser, 2021; Safieddine et al., 2012). Signal detection and transduction are both indispensable for normal hearing (Fettiplace and Hackney, 2006). Ribbon synapses were formed here between new IHCs and SGN fibers (Fig. 2), and vesicle release was observed when the new IHCs were depolarized (Fig. 4), which suggested normal signal transduction at the bottom part of the new IHCs. By contrast, two lines of evidence indicate that defective MET in the new IHCs represents one key barrier towards achieving marked hearing improvement in our current IHC damage model: (1) stereocilia on the apical surface of P42_new IHCs were irregularly aligned (Fig. 3) and (2) the MET current in P42_new IHCs was very weak (Fig. 4G-I).

The weak MET current in new IHCs is likely not entirely due to a lack of expression of MET components, because genes encoding key proteins involved in MET function, including Tmc1 and Lhfpl5 (Kurima et al., 2002; Vreugde et al., 2002; Xiong et al., 2012), are expressed in the new IHCs (Table S3). This is not an unexpected finding because MET-related proteins are expressed independently of the formation of stereocilia (Cai et al., 2013). Because stereocilia were not as regularly organized in new IHCs as in endogenous IHCs (Fig. 3E,E′), MET impairment is more likely caused by defective MET channel assembly. Emx2 has been reported to be required in hair-bundle organization (Jiang et al., 2017), and, recently, intrinsic spontaneous calcium action potential activity was found to be important in maintaining the integrity of IHC stereociliary bundles (Carlton et al., 2023). However, genes encoding MET-related or tip-link-related components, such as Cdh23, Tmie and Pcdh15 were not markedly enriched in P42_new IHCs (Fig. S6C). Whether induction of Cdh23, Tmie and Pcdh15 will substantially enhance the MET current in the new IHCs warrants further investigation.

In future studies, we will also identify key candidate genes that regulate the morphogenesis of IHC stereocilia by using our in vivo high-throughput genetic-screening approach (Wang et al., 2021; Zhang et al., 2018). If mutation of any these candidate genes results in the formation of stereocilia resembling those in P42_new IHCs, we will include the genes to the combination of Atoh1 and Tbx2. Thus, combined ectopic expression of key IHC developmental genes in IBCs and/or IPhs could serve as a promising future approach to regenerate IHCs in individuals with hearing impairment.

Fgf8-DTR/+ mice were created by co-injecting one sgRNA against the Fgf8 locus, donor DNA and Cas9 mRNA into one-cell-stage mouse zygotes. The sgRNA was 5′-AGCTGGGCGAGCGCCTATCG-3′, and the donor DNA was designed as described in Fig. S1. F0 mice born from pseudopregnant mothers were screened, and mice with potentially successful gene targeting were identified using junction PCR. All potentially correct F0 mice were crossed with wild-type C57BL/6 mice, and the F1 mice offspring that were germline stable were subject to further junction-PCR and Southern blotting analyses to confirm the absence of random insertion of donor DNA in the F1 genome. Southern blotting was performed according to our protocol described previously (Li et al., 2018a). F2 and subsequent mice were genotyped using regular tail-DNA PCR. The detailed primer sequences are provided in Table S5. The Fgf8-DTR/+ mice will be deposited with the Jackson laboratory.

Tamoxifen (TMX; T5648, Sigma-Aldrich) was dissolved in corn oil (C8267, Sigma-Aldrich). Diphtheria toxin (DT; D0564, Sigma-Aldrich) was dissolved in 0.9% NaCl solution. Trimethoprim lactate salt (TMP; T0667, Sigma-Aldrich) was resolved in 1× phosphate-buffered saline (PBS). In all experiments, mice were administered TMX (3 mg/40 g body weight) at P0 and P1, DT (15 ng/g body weight) at P2, and TMP (300 μg/g body weight) at P3 and P4. An EdU labeling kit (C10637, Thermo Scientific) was used as per the protocol provided with the kit.

Plp1-CreER mice (005975) and Rosa26-LSL-tdTomato (Ai9/+) mice (007909) were from The Jackson Laboratory. The Rosa26-LSL-TAT/+ mouse strain has been reported in our previous study (Bi et al., 2022). All mice were bred and raised in SPF-level animal rooms, and animal procedures were performed according to the guidelines (NA-032-2019) of the IACUC of the Institute of Neuroscience (ION), CAS Center for Excellence in Brain Science and Intelligence Technology, Chinese Academy of Sciences (Shanghai, China). Both sexes of mice were used in our study.

Mice were anesthetized and sacrificed, during which the heart was perfused with fresh 1× PBS and then with 4% paraformaldehyde (PFA) to remove blood from the inner ear. Subsequently, inner ear tissues were dissected, post-fixed with 4% PFA overnight at 4°C and washed thrice with 1× PBS, after which the inner ear samples were decalcified with 120 mM EDTA (ST066, Beyotime) for 2 days at 4°C (until they were adequately soft) before whole-mount preparation. For whole-mount analysis, each cochlea was divided into three parts and first scanned at 10× magnification under a confocal microscope (Nikon C2, TiE-A1 or NiE-A1 Plus). In each acquired image, a line was drawn passing through the center of the IHCs and OHCs to measure the entire length of each cochlear duct, and the cochlear sample was then precisely divided into basal, middle and apical regions of equal length.

The following primary antibodies were used for immunolabeling: anti-Prestin (goat, 1:1000, sc-22692, Santa Cruz), anti-vGlut3 (rabbit, 1:500, 135203, Synaptic Systems), anti-Myo7a (rabbit, 1:500, 25-6790, Proteus Bioscience), anti-Ctbp2 (mouse, 1:200, 612044, BD Biosciences), anti-Fabp7 (rabbit, 1:500, ab32423, Abcam) and anti-NF-200 (mouse, 1:500, N0142, Sigma). Cochlear tissues were counterstained with Hoechst 33342 solution (1:1000, H3570, Thermo Scientific) to visualize nuclei. Finally, cochlear samples were mounted with Prolong Gold antifade medium (P36930, Thermo Scientific). Immunofluorescence labeling was performed as described in our previous protocol (Liu et al., 2010).

First, Myo7a⁺ cells were quantified in cochleae of control (no DT) and DT-treated Fgf8-DTR/+ mice at P42 (Fig. 1) by scanning entire cochlear turns under a confocal microscope at 10× magnification. Second, numbers of new IHCs (tdTomato+/vGlut3+) or tdTomato+ cells were estimated through confocal scanning at 60× magnification and 1 μm interval. For each turn, three areas were selected and a final average cell number was obtained per sample, after which a second average was calculated among all mice analyzed. Cell-fate conversion rate was calculated by normalizing tdTomato+/vGlut3+ cell numbers against total tdTomato+ cell numbers in the same confocal scanning regions. Notably, we only quantified tdTomato+ cells that were near vGlut3+ IHCs, which allowed us to confidently define these cells as derivatives of IBCs and/or IPhs, regardless of their cell fates.

To precisely quantify Ctbp2+ ribbon synapses, three frequency areas (8, 16 and 30 kHz) were selected according to a previous method (Müller et al., 2005). Confocal scanning with 2.0× digital zoom and 0.41 μm interval was performed on samples from wild-type or Plp1-TAT-DTR mice at P42 (Fig. 2E-F′). Ctbp2+ puncta were manually counted, and the Myo7a or tdTomato signal helped mark the boundary of each IHC. All cell-counting data are presented as mean±s.e.m., and statistical analyses were performed using one-way ANOVA, followed by a two-tailed and unpaired Student's t-test with Bonferroni correction.

For 10× Genomics single-cell RNA-seq, fresh cochleae of Plp1-Ai9 mice (for P42_IBCs/IPhs) and Plp1-TAT-DTR mice (for P42_new IHCs) were dissected and first digested in a choline chloride solution containing 20 U/ml papain (LK003178, Worthington) and 100 U/ml DNase I (LK003172, Worthington) for 20 min at 37°C, then digested again with an additional protease (P5147, Sigma; 1 mg/ml) and dispase (LS02104, Worthington; 1 mg/ml) for 20 min at 25°C. The choline chloride solution contained 92 mM choline chloride, 2.5 mM KCl, 1.2 mM NaH₂PO₄, 30 mM NaHCO₃, 20 mM HEPES, 25 mM glucose, 5 mM sodium ascorbate, 2 mM thiourea, 3 mM sodium pyruvate, 10 mM MgSO₄.7H₂O, 0.5 mM CaCl₂.2H₂O and 12 mM N-acetyl-L-cysteine. Next, the post-digested tissues were gently triturated using fire-polished glass Pasteur pipettes (13-678-20b, Fisher) featuring four distinct pore sizes (largest size used first, smallest last). Finally, the single-cell suspensions obtained were filtered using a 30 μm cell strainer (130-098-458, Miltenyi) and any cell debris was removed using a debris-removing buffer (130-109-398, Miltenyi). All procedures were completed within 2.5 h of the animals being euthanized. The quality and survival rate of single cells were estimated using a Trypan Blue solution (15250061, Thermo Fisher Scientific) before the cells were subject to 10× Genomics single-cell preparation and library procedures (Chromium Single Cell 3′ v3). Paired-end sequencing was performed on the final libraries on an Illumina Novaseq platform; the average sequencing depth of Plp1-Ai9 and Plp1-TAT-DTR libraries was 111,228 and 84,165 reads, respectively.

For single-cell RNA-seq data produced using the 10× Genomics platform, establishment of the mouse reference genome and alignment were performed using Cell Ranger (v3.0.2) (Zheng et al., 2017). Downstream analysis was performed using the R package Seurat (v3.2.3) (Stuart et al., 2019). Cells with <500 or >7500 unique genes and 20% mitochondrial genes were excluded from the analysis. The expression data were normalized using the ‘NormalizeData’ function. Mitochondrial content and the effects of cell cycle heterogeneity were regressed when scaling data in the ‘ScaleData’ function. Principal component analysis was conducted using ‘RunPCA’ and clustering was performed using ‘FindNeighbors’ and ‘FindClusters’. Two types of dimensional reduction, uniform manifold approximation and projection (UMAP) and t-distributed stochastic neighbor embedding (t-SNE), were conducted using ‘RunUMAP’ and ‘RunTSNE’, respectively. To integrate data from the 10× Genomics platform and the smart-seq platform (from our previous two studies, GEO accession numbers GSE161156 and GSE199369), ‘FindIntegrationAnchors’ and ‘IntegrateData’ were applied for integration analysis.

P42_new IHCs were defined as cells in which the expression levels of tdTomato (WPRE sequence in Ai9 mouse strain), Myo6, Myo7a and Slc17a8 were above zero. Moreover, Slc26a5 and Ikzf2 were examined and their expression levels were found to be nearly zero in P42_new IHCs. Slc1a3+/tdTomato+/Celf4−/Otof−/Fgfr3− cells were defined as P42_IBCs/IPhs. We also selected E16/P1/P7_IHCs from a previous study (GSE137299) (Kolla et al., 2020). E16_IHCs and P7_IHCs were defined as Myo6+/Fgf8+ cells, and Myo6+/Fgf8+/Myo7a+ cells were classified as P1_IHCs.

To identify cell-type-specific genes, distinct criteria were set for cells sequenced using the 10× Genomics and smart-seq platforms. Differentially expressed genes (DEGs) among P42_IBCs/IPhs and all other non-IBC and/or IPh cell populations were regarded as the initial P42_IBC/IPh genes, which were calculated by employing ‘FindAllMarkers’ with a log₂ fold-change (FC) of ≥0.5 and adjusted P-value of ≤0.05 by using the Wilcoxon rank-sum test. P30_IHC genes were calculated by overlapping our previous single-cell smart-seq dataset (GSE199369) and another bulk-seq database (GSE111348) (Li et al., 2018b). Likewise, adult PC and/or DC genes were calculated from our previous dataset (GSE161156). The IHC or PC and/or DC genes were selected with an average TPM of >16, and at least half the cells expressed these genes with a TPM value of >16. The PC and/or DC genes were used as references to remove genes shared by IHCs and PCs and/or DCs. Thus, 432 P30_IHC genes and 423 P42_ IBC/IPh genes were identified (Table S1). Finally, for the P42_new IHCs, a gene was regarded as being highly expressed if (1) its average normalized expression value>0.5, and (2) >50% of the cells expressed the gene with a normalized expression value>0.5.

DEGs between P42_new IHCs and P42_IBC/IPhs were calculated using ‘FindMarkers’ (log₂FC>0.5, P<0.05). Based on the identified DEGs, gene ontology (GO) terms of biological process enrichment were evaluated (FDR<0.05) using DAVID (Database for Annotation, Visualization, and Integrated Discovery). By using the pre-processed matrix in the Seurat object, trajectory analysis was performed using Monocle (v2.14.0) (Trapnell et al., 2014). The top 2000 most variable genes calculated using the ‘FindVariableFeatures’ function in Seurat were used to construct a pseudotime trajectory, and the ‘orderCells’ function in Monocle was used to arrange cells along the pseudotime axis.

ABR values were measured at 4, 5.6, 8, 11.3, 16, 22.6, 32 and 45 kHz at P42, following our previously published protocol (Li et al., 2018a). A two-tailed and unpaird Student's t-test was applied to determine the statistical significance regarding the hearing thresholds at the same frequency among different mice (Figs 1F and 3H).

We followed the scanning electron microscopy (SEM) protocol detailed in our previous study (Sun et al., 2021). In transmission electron microscopy (TEM) analysis of IHCs, similar sample-treatment steps were used to those for SEM, except that the samples were subsequently embedded in an epoxy resin monomer (90529-77-4, SPI Supplies). Finally, 70 nm ultrathin transverse sections of the OC were obtained using an ultramicrotome (Leica EM UC6) and examined using a JEOL-1230 transmission electron microscope (Nippon Tekno).

The apical turns of cochlear sensory regions were dissected out and pinned on a glass coverslip. All recordings were performed at room temperature. Recording pipettes were pulled from borosilicate glass capillaries (TW150-4F, World Precision Instruments) and coated with dental wax. The typical pipette-resistance range was 4-7 MΩ. IHCs were voltage-clamped using an Axon 200B amplifier (Axon Instruments-Molecular Devices) interfaced with a Digidata 1440B (Molecular Devices). All recordings were performed using jClamp software (for details, please refer to http://www.scisoftco.com/jclamp.html). Liquid junction potential was corrected offline.

Membrane capacitance changes were recorded to monitor the fusion of synaptic vesicles during exocytosis. IHCs were held at −80 mV, and exocytosis was induced by applying a depolarizing pulse of 0 mV. Sinewaves of 1 kHz and 70 mV peak-to-peak amplitude were superimposed on the holding potential before and after stimulation.

Whole-cell patch-clamp was used to record the MET current of IHCs, according to protocols described previously (Liu et al., 2019). Adult cochlear sensory regions were dissected in a solution containing (in mM) 141.7 NaCl, 5.36 KCl, 0.1 CaCl₂, 1 MgCl₂, 0.5 MgSO₄, 3.4 L-glutamine, 10 glucose and 10 H-HEPES (pH 7.4), and then transferred into a clean recording chamber bathed with a recording solution containing (in mM) 144 NaCl, 0.7 NaH₂PO₄, 5.8 KCl, 1.3 CaCl₂, 0.9 MgCl₂, 5.6 glucose and 10 H-HEPES (pH 7.4). Patch pipettes featuring a tip diameter of 2-3 μm were prepared using a pipette puller (Sutter, P-1000). The intracellular solution contained (in mM) 140 KCl, 1 MgCl₂, 0.1 EGTA, 2 Mg-ATP, 0.3 Na-GTP and 10 H-HEPES (pH 7.2). The MET currents of IHCs were sampled at 100 kHz with a patch-clamp amplifier (Axon, 700B) and the holding potential was set at −70 mV. To evoke MET currents, a fluidjet pipette featuring a tip diameter of 6-10 μm was positioned near a hair bundle (at a distance of ∼5 μm) and 40 Hz sine waves were delivered through a piezoelectric disc to stimulate IHC hair bundles.

We thank Dr Min Zhang and Ms. Zhenning Zhou from the Molecular Biology Facility of the Institute of Neuroscience (ION), Chinese Academy of Sciences for 10× Genomics single-cell library preparation; Drs Xu Wang and Yu Kong from the Electronic Microscope Facility of the ION for SEM and TEM assistance; Dr Qian Hu from the Optical Imaging Facility of the ION for support with confocal image analysis; and Ms Qian Liu from the Department of Embryology of the ION rodent animal center for helping us in transplanting zygotes into pseudopregnant female mice for generating the Fgf8-DTR/+ knock-in mouse strain.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.201888.reviewer-comments.pdf