Development and transdifferentiation into inner hair cells require Tbx2

Abstract Atoh1 is essential for the development of both outer hair cells (OHCs) and inner hair cells (IHCs) in the mammalian cochlea. Whereas Ikzf2 is necessary for OHC development, the key gene required for IHC development remains unknown. We found that deletion of Tbx2 in neonatal IHCs led to their transdifferentiation into OHCs by repressing 26.7% of IHC genes and inducing 56.3% of OHC genes, including Ikzf2. More importantly, persistent expression of Tbx2 coupled with transient Atoh1 expression effectively reprogrammed non-sensory supporting cells into new IHCs expressing the functional IHC marker vGlut3. The differentiation status of these new IHCs was considerably more advanced than that previously reported. Thus, Tbx2 is essential for IHC development and co-upregulation of Tbx2 with Atoh1 in supporting cells represents a new approach for treating deafness related to IHC degeneration.


INTRODUCTION
Precise specification of distinct cell fates is crucial for organogenesis and to define the molecular mechanisms underlying this fundamental developmental event, a fascinating model that has been widely used is the mouse cochlea. In the cochlea, two subtypes of sound receptor cells-the inner hair cells (IHCs) and the outer hair cells (OHCs)-are located in the auditory epithelium, which is also referred to as the organ of Corti [1][2][3]. IHCs and OHCs are derived from the same progenitors expressing Atoh1, a key b-HLH transcription factor (TF) necessary for generating both IHCs and OHCs [4,5]. Adjacent to HCs are distinct subtypes of non-sensory cochlear supporting cells (SCs) that are arranged from the medial to lateral portion and named inner border cells (IBCs), inner phalangeal cells (IPhs), pillar cells (PCs) and Deiters' cells (DCs) [1,2,6]. Whereas the cochlear SCs of non-mammalian vertebrates, including birds and fish, can regenerate HCs upon damage, the SCs in mammals lack this regenerative capacity [7,8]. Therefore, damage or degener-ation of either OHCs or IHCs results in permanent hearing impairment in mammals, including humans.
The IHCs and OHCs share various pan-HC markers such as Myo6 and Myo7a, but the cells also differ in several aspects. OHCs are sound amplifiers and specifically express Prestin, a motor protein encoded by Slc26a5 [9,10] and Slc26a5 -/mice display severe hearing impairment [11]. Conversely, IHCs are primary sensory cells and specifically express vGlut3 (encoded by Slc17a8), Otoferlin and Slc7a14 [12][13][14][15]. The Slc17a8 -/and Otoferlin -/mice exhibit profound deafness because both vGlut3 and Otoferlin are heavily involved in the packaging and exocytosis of ribbon synapse vesicles containing the excitatory neurotransmitter glutamate in the IHCs, respectively. Recently, Insm1 and Ikzf2 were reported as key regulators of OHC development and, accordingly, OHCs tend to transdifferentiate into IHCs or IHC-like cells in Insm1 or Ikzf2 mutant mice [16,17].
In contrast to the aforementioned identification of genes for OHC development, it remains poorly understood what gene is needed for normal IHC
Notably, Tbx2 was found to be the top TF gene highly expressed in IHCs but not OHCs (red arrow in Fig. 1C). However, the Tbx2 protein expression pattern in the cochlea remains poorly characterized. Next, we generated a new knockin mouse strain, Tbx2 * 3 × HA-P2A-iCreER-T2A-EGFP/+ (abbreviated as Tbx2-HA/+), wherein three HA tags were fused to the C-terminus of Tbx2 ( Fig. 1D and Supplementary Fig. S1A-F) and the Tbx2 expression pattern was primarily characterized using an anti-HA antibody. In agreement with previous reports [22,23], the HA (Tbx2) expression domain was found to overlap with that of Sox2 in the otocyst, but not in the hindbrain, at Embryonic Day 9.5 (E9.5) (Supplementary Fig. S1G-G"). Tbx2 was broadly expressed in cochlear duct cells at E13.5. However, at E15.5, although Tbx2 expression was maintained in the medial portion including Myo6+ IHCs, the expression was undetectable in lateral progenitors (LPs) that eventually become OHCs, PCs and DCs in the basal turn ( Supplementary Fig. S1H-H"). By contrast, Tbx2 continued to be broadly expressed in the apical turn at E15.5 ( Supplementary Fig.  S1I-I"). The wave of Tbx2 downregulation was largely completed by E17.5, when Tbx2 was undetectable in basal OHCs, PCs and DCs, and apical LPs ( Fig. 1E-F"). As expected, Tbx2 was highly expressed in IHCs, but not OHCs, at P1 ( Fig. 1G and H), P15 and P30 (Fig. 1I). Collectively, these results showed that Tbx2 is persistently and specifically expressed in both differentiating and mature IHCs, prompting us to hypothesize that Tbx2 is essential for IHC development.

Neonatal IHCs transdifferentiate into OHCs when Tbx2 is conditionally deleted
Our aforementioned hypothesis was supported by genetic evidence obtained from two distinct Tbx2 mutant strains. The first was a germ-line Tbx2 knockout mouse strain (Tbx2 +/-) in which the entire Tbx2 locus (∼9.6 kbp) was deleted ( Supplementary  Fig. S2A-C). Tbx2 -/mice die at early embryonic ages due to cardiac defect [24]. Importantly, the OHC-specific marker Prestin was weakly detected in Tbx2 +/but not Tbx2 +/+ IHCs at P7, P14 and P42. Because the IHC phenotypes in Tbx2 +/were mild, detailed analyses are not presented and we focused our study on the second mutant strain that we generated: Tbx2 flox/+ , wherein the second exon of Tbx2 was flanked by two loxp sequences (Supplementary Fig. S2D-I). Here, WT mice (control group) and Slc17a8 iCreER/+ ; Tbx2 flox/conditional knockout mice (abbreviated as Tbx2 cko mice; experimental group) were administered TMX at P2 and P3 and examined at different ages.
At P7, IHCs were vGlut3+/Prestin-in WT mice ( Fig. 2A-A"), whereas ectopic Prestin expression at heterogeneous levels was detected in IHCs in Tbx2 cko mice (arrows in Fig. 2B-B"). Notably, relative to Prestin-IHCs (asterisks in Fig. 2B Fig. 2F-F"). Thus, our data supported the notion that IHCs tend to transdifferentiate into OHCs when Tbx2 is absent. Hereafter, to differentiate between the IHCs that did or did not undergo cell-fate change (asterisks in Fig. 2), the Prestin+ IHCs were defined as induced-OHCs derived from IHCs (abbreviated as iOHCs) in which vGlut3 expression was either decreased or completely lost.

Cell-fate conversion of iOHCs is largely completed by P14
We next estimated the progression of cell-fate conversion in iOHCs between P7 and P42. First, we quantified the cells in entire cochlear turns that expressed high levels of Prestin (Prestin High ) but low (or undetectable) levels of vGlut3 (vGlut3 Low ). The percentage of these cells-defined as Prestin High /vGlut3 Low iOHCs (yellow arrows in Fig. 2B-B", yellow and blue arrows in Fig. 2D-D", and yellow arrows in Fig. 2F-F")-was the lowest (30.5% ± 3.7%) at P7 (n = 3) and was significantly increased to 90.5% ± 3.1% at P14 (n = 3) and 87.4% ± 2.7% at P42 (n = 3) (Fig. 2G). Although a Tbx2 antibody was not available for validating the absence of Tbx2, the Prestin High /vGlut3 Low iOHCs are likely to correspond to the endogenous IHCs in which Tbx2 was successfully deleted. Second, we quantified cells that expressed high levels of vGlut3 (vGlut3 High ) but low (or undetectable) levels of Prestin (Prestin Low ); the percentage of these cellsdefined as vGlut3 High /Prestin Low cells-was the highest (69.5% ± 3.7%) at P7 and was drastically decreased to 9.5% ± 3.1% at P14 and 12.6% ± 2.7% at P42 (Fig. 2H).
Notably, we expected the vGlut3 High /Prestin Low cells to include two subpopulations: (i) the iOHCs that were in the early process of cell-fate conversion (white arrows in Fig. 2B-B" and Fig. 2D-D"); and (ii) the endogenous IHCs that did not undergo cell-fate change (asterisks in Fig. 2), either due to their lack of responsiveness to Tbx2 deletion or due to unsuccessful deletion of Tbx2. Thus, our results suggested that although the cell-fate conversion rate was not synchronized in distinct iOHCs, the conversion was largely completed at P14, because no significant difference was observed between P14 and P42 ( Fig. 2G and H). Because vGlut3 High /Prestin Low cells accounted for only ∼12.6% of the cells in the IHC region at P42 (Fig. 2H), the hearing thresholds of Tbx2 cko mice (n = 8) at all tested frequencies were significantly higher than those of WT mice (n = 8) (Fig. 2I).
We performed double-labeling for Prestin with two additional IHC markers-Otoferlin (Supplementary Considering these results collectively, we propose that Tbx2 plays an essential role in IHC development by repressing the expression of OHC genes or preventing the transdifferentiation of IHCs into OHCs (Fig. 2J); thus, following the loss of Tbx2, neonatal IHCs gradually differentiate into iOHCs (Fig. 2K).
To determine the differentiation status of endogenous WT OHCs that was most similar to that of P14 iOHCs, we pooled the 46 P14 iOHCs from this study with the 87 WT OHCs at E16 (E16 WT OHCs), 170 WT OHCs at P1 (P1 WT OHCs) and 39 WT OHCs at P7 (P7 WT OHCs) from one previous single-cell RNA-seq study [25] and the 17 P30 WT OHCs from another study [19]. In total, eight clusters were revealed through uniform manifold approximation and projection (UMAP) analysis (Fig. 3F). Notably, 7/46 (15.2%) of the P14 iOHCs belonged to Cluster 8, which mainly included the P30 WT OHCs, whereas the remaining 39/46 (84.8%) were assigned to other clusters (Fig. 3G). Moreover, trajectory analysis by using Monocle revealed that P14 iOHCs aggregated with P30 WT OHCs ( Fig. 3H and I), supporting the notion that P14 iOHCs are more similar to P30 WT OHCs than to WT OHCs at other ages.
A similar pattern was obtained when we included additional IHCs obtained from the same single-cell RNA-seq study (Supplementary Fig. S4B and C). Together, these results suggested that upon Tbx2 loss, the IHCs became iOHCs that were most closely related to P30 WT OHCs.

Ikzf2 protein is also expressed in iOHCs
The similarity between iOHCs and endogenous OHCs led us to predict that Ikzf2 protein (also known as Helios), a key regulator of OHC development [16,19], would be expressed in iOHCs. Because a suitable commercial Ikzf2 antibody for immunostaining was unavailable, we constructed a new knockin mouse strain, Ikzf2 * 3 × V5-P2A-Tdtomato/+ (abbreviated as Ikzf2 V5/+ ), wherein three V5 tags were fused to the C-terminus of Ikzf2 ( Fig. 3J and Supplementary Fig. S5A-C). The correct gene targeting of Ikzf2 V5/+ was confirmed through Southern blotting and tail-DNA PCR (Supplementary Fig. S5D and E). Thus, a V5 antibody could be used to detect Ikzf2 protein. Here, we did not use Tdtomato as a reporter of Ikzf2 mRNA expression because the signal-to-noise ratio of Tdtomato was lower than that of the V5 antibody.

Tbx2 is necessary for maintaining cell fate of adult cochlear IHCs
We next investigated whether Tbx2 is also required for maintaining the cell fate of fully mature IHCs at adult ages. WT (n = 3) and Tbx2 cko (n = 3) mice were administered TMX at P60 and P61 and then analysed at P120. Three rows of Prestin+ OHCs and one row of vGlut3+ IHCs were well Natl Sci Rev, 2022, Vol. 9, nwac156   aligned in WT cochleae (Fig. 4A-A"), whereas additional but discontinuous Prestin High /vGlut3 Low cells existed in the IHC region of Tbx2 cko cochleae (yellow arrows in Fig. 4B-B"). According to our aforementioned criterion, Prestin High /vGlut3 Low cells were defined as iOHCs, and the nearby vGlut3 High /Prestin Low IHCs were suspected of being endogenous IHCs in which Tbx2 was not successfully deleted (asterisks in Fig. 4B-B"). When we grouped the iOHCs in all the cochlear turns together and quantified the results, the calculated percentage of Prestin High /vGlut3 Low iOHCs at P120 was 60.0% ± 2.7% (Fig. 4C). Moreover, the middle turn was found to contain the fewest and the basal turn the most iOHCs when the cells in the three turns were counted separately (Fig. 4D).
Notably, vGlut3 was undetectable in iOHCs at P42 when Tbx2 was deleted at P2 and P3 (arrows in Fig. 2F-F"), but vGlut3 was detectable, albeit faintly, in iOHCs at P120 (2 months after Tbx2 deletion at P60/P61) (arrows in Fig. 4B-B"). Collectively, our data support the view that Tbx2 is required in maintaining or stabilizing the IHC fate at both neonatal (P2/P3) and adult (P60/P61) ages. In the absence of Tbx2, IHCs tend to transdifferentiate into OHCs and become iOHCs. However, the cell-fate conversion rate might be lower at adult ages since fewer IHCs transdifferentiated into iOHCs in adult (Fig. 4C) than neonatal ( Fig. 2A-G)

A new genetic model to induce temporal Atoh1 but permanent Tbx2 expression
The functional importance of Tbx2 in IHC development observed here prompted us to further hypothesize that Tbx2 together with Atoh1 should be capable of converting cochlear IBCs/IPhs, which localize near IHCs, into more differentiated new IHCs (i.e. into vGlut3+ cells) than what was obtained with Atoh1 alone in our previous study [26]. The IBCs/IPhs were effectively targeted using Plp1-CreER+ (Fig. 5A-A"'), as reported in other studies [26][27][28]. Moreover, we aimed to only induce temporal Atoh1 expression, mimicking the Atoh1 expression observed during endogenous IHC development [5,[29][30][31], and thus used the approach detailed below.
When the destabilizing domains (DDs) derived from Escherichia coli dihydrofolate reductase (DHFR) are fused with a protein of interest, rapid proteasomal degradation of the protein is triggered [32]; however, the cell-permeable small molecule trimethoprim (TMP) can bind to and stabilize the DDs, and thus TMP treatment results in the protein degradation being prevented in a rapid, reversible and TMP dose-dependent manner [32]. Furthermore, fusion of DHFR DDs to both N-and Ctermini of a protein yields superior temporal control than does fusion to one terminus alone [33]. Therefore, for our analyses, we constructed a new strain,   (abbreviated as Rosa26-LSL-TAT/+) ( Fig. 5B and Supplementary Fig. S6A-F), in which Atoh1 protein was tagged with HA and DHFR; here, as expected, Atoh1 is unstable and undergoes rapid degradation under control conditions (Supplementary Fig.  S6G) but becomes temporally stable in the pres-ence of TMP ( Supplementary Fig. S6H) [32,34]. Moreover, upon Cre-mediated recombination, Atoh1 (also tagged with three HA fragments), Tbx2 (tagged with three V5 fragments) and Tdtomato are transcribed from the same polycistronic mRNA in this strain.

Rosa26-CAG-Loxp-Stop-Loxp-Tbx2
We confirmed the transient and persistent expression of Atoh1 and Tbx2, respectively, in our model by using the four assays described here; in all cases, TMX was administered at P0 and P1 and TMP at P3 and P4, unless specified otherwise. First, in Plp1-CreER+; Rosa26-LSL-TAT/+ (abbreviated as Plp1-TAT) mice in the absence of TMX administration, no Tdtomato+, V5 (Tbx2)+ or HA (Atoh1)+ cells were observed. Second, Plp1-TAT mice that were administered TMX were divided into two groups-No-TMP and TMP-treated-and both groups were analysed at P4 (3 h after last TMP treatment). In cochleae from no-TMP mice, we detected Tdtomato+ cells in which V5 (Tbx2) expression was high but HA (Atoh1) expression was faint or undetectable (arrows in Fig. 5C-C"'); by contrast, in TMP-treated mice, Tdtomato+ cells expressing high levels of Tbx2 and Atoh1 were present (arrows in Fig. 5D-D"'). The faint Atoh1 expression in Fig. 5C might be due to occasional incomplete degradation of Atoh1 protein. Third, when we examined the TMP-treated mice at P7, we found that high Tbx2 and Tdtomato expression was maintained but Atoh1 expression was reversed to a faint or undetectable level in the same cells (arrows in Fig. 5E-E"'). Fourth, Atoh1 expression was restored to a high level if an additional (third) dose of TMP was administered 3 h before sacrifice at P7 (arrows in Fig. 5F-F"'). Collectively, these results supported our conclusions that whereas the expression of Tbx2 and Tdtomato solely relied on TMX, Atoh1 expression depended on both TMX and TMP, and that the Atoh1 protein level was reversible and TMP treatment stabilized Atoh1 only transiently (for <3 days).

Transient Atoh1 and permanent Tbx2 expression together successfully convert neonatal IBCs/IPhs into vGlut3+ new IHCs
We first confirmed that Tbx2 alone cannot convert neonatal IBCs/IPhs into IHCs or general HCs by characterizing Plp1-TAT mice that were administered only TMX and then analysed at P42. All Tdtomato+ cells, which were derived from IBCs/IPhs, failed to express Myo7a (Fig. 5G-G"'). We also did not observe Tdtomato+/vGlut3+ cells. Second, we ascertained whether simultaneously high levels of Atoh1 and Tbx2 can convert neonatal IBCs/IPhs into vGlut3+ new IHCs (Fig. 6A). In the cochleae of control Plp1-Ai9 mice at P42 (n = 3), neither the IHC marker vGlut3 nor the OHC marker Prestin was expressed in the Tdtomato+ cells that were IBCs/IPhs (arrows in Fig. 6B-B"').
By contrast, Tdtomato+ cells expressing vGlut3, but not Prestin, were observed in the cochleae of Plp1-TAT mice at P42 (arrows in Fig. 6C-C"'). Because these Tdtomato+/vGlut3+ cells were derived from IBCs/IPhs, they were defined as new IHCs (or conservatively named as IHC-like cells) and the new IHCs were found to be primarily adjacent to the endogenous IHCs that were vGlut3+/ Tdtomato-. Among the 563 new IHCs identified in the middle and apical turns, 561 (99.6%) were located at the HC layer and likely to lose contact with the basement membrane ( Supplementary Fig.  S7A-A"' and Supplementary Video S1); by contrast, only 2/563 cells (0.4%) appeared to maintain contact with the basement membrane (Supplementary Fig. S7B-B"'). Below and adjacent to these new IHCs were Tdtomato+ cells expressing neither vGlut3 nor Prestin (arrows in Fig. 6D-D"'), which we defined as IBCs/IPhs that failed to become IHCs and were primarily located in the SC layer.
Next, we quantified the Tdtomato+/vGlut3+ new IHCs localized close to endogenous IHCs. Per 200-μm of cochlear duct (n = 5), 8.0 ± 0.8, 9.3 ± 0.7 and 5.2 ± 0.8 new IHCs were present in basal, middle and apical turns, respectively. Furthermore, we calculated the cell-fate conversion rate by normalizing the number of new IHCs to the total number of Tdtomato+ cells that were close to the endogenous IHCs within the same region: 34.0% ± 3.1%, 30.2% ± 2.2% and 23.9% ± 1.1% of the Tdtomato+ IBCs/IPhs were respectively converted into vGlut3+ new IHCs in the basal, middle and apical turns (Fig. 6E). However, the average cell-fate conversion rate differed by <1.5fold among the three turns and thus, to simplify the subsequent analysis, we grouped the three turns together. Our results revealed 221.8 ± 19.1 new IHCs in the entire cochleae of Plp1-TAT mice at P42 (Fig. 6F), with an average cell-fate conversion rate of 29.5% ± 1.2%; the remaining ∼70.5% of the cells were IBCs/IPhs that failed to become new IHCs. Notably, the vGlut3+ new IHCs appeared at P7, but their numbers at P7 were significantly lower than at P14 and P42 (Fig. 6F). All the vGlut3+ new IHCs also expressed another IHC marker, Otoferlin, in Plp1-TAT mice, but not in Plp1-Ai9 mice, at P42 ( Supplementary Fig. S8A-B"'). Furthermore, when Myo7a was used as a marker to define new HCs in general, 239.3 ± 17.2 Tdtomato+/Myo7a+ new HCs were detected in the entire cochleae of Plp1-TAT mice at P42, but these cells were not present in Plp1-Ai9 mice (arrows in Supplementary  Fig. S8C-D"). new HCs, we deduced that most, if not all, of the Myo7a+ new HCs adopted the IHC fate. Lastly, auditory brainstem response (ABR) assay showed that hearing thresholds did not differ significantly between Plp1-Ai9 and Plp1-TAT mice at P42, except at 45 kHz (Fig. 6G). This result suggested that, except at high frequency (45 kHz), the additional IHCs did not affect the function of endogenous IHCs and this agreed with the finding in Huwe1 mutant mice, which also harbor extra IHCs but display normal hearing ability [35]. Moreover, scanning electron microscopy (SEM) analysis at P42 revealed that, relative to control Plp1-Ai9 mice that contained a single row of endogenous IHCs (purple color in Fig. 6H and H'), Plp1-TAT mice harbored an additional but discontinuous row of IHCs (blue color) whose stereocilia showed a 'bird-wing' pattern ( Fig. 6I and I'); 12.1 ± 1.2 (n = 3) extra IHCs with stereocilia were counted per 200 μm and 68.1% ± 4.8% of these cells featured relatively well organized stereocilia (white arrowheads in Fig. 6I). The numbers were higher than those determined through immunostaining analysis because we used SEM to scan areas with larger numbers of new IHCs. Nonetheless, compared to the endogenous IHCs, the stereocilia in these new IHCs were immature, variable in numbers and of a poorer quality. Considering these results collectively, we conclude that transient Atoh1 and permanent Tbx2 expression reprogrammed neonatal IBCs/IPhs into new IHCs that expressed the early pan-HC marker Myo7a, and the IHC-specific markers vGlut3 and Otoferlin, and possessed IHC-like stereocilia.

DISCUSSION
The proliferation of undifferentiated cochlear sensory progenitors continues until E14.5 and is followed by a differentiation wave that moves in a basal-to-apical gradient [36,37]. Atoh1 is essential for specifying the general HC fate, because both IHCs and OHCs are lost in Atoh1 -/mice [4] and Insm1 and Ikzf2 have recently been reported to be critical for OHC development [16,17]. However, the gene specifically necessary for IHC development has remained unknown. Our study has revealed that Tbx2 plays an essential role in maintaining the cell fate of differentiating and mature IHCs by preventing these cells from transdifferentiating into OHCs. Approximately 56.3% of OHC genes, including Slc26a5 and Ikzf2, are upregulated and 26.7% of IHC genes, including Slc17a8 and Otof, are downregulated in the Tbx2 -/-IHCs that we defined here as iOHCs at P14. We also confirmed that Ikzf2 protein is highly expressed in these iOHCs (Fig. 3), further indicating that ectopic Ikzf2 expression by itself may be sufficient to transdifferentiate IHCs to OHCs [16,19]. Thus, we speculate that Tbx2 stabilizes the IHC fate partially by repressing Ikzf2 expression. However, we cannot exclude the possibility that this is a secondary effect following cell-fate changes. Whether Tbx2 directly binds to Ikzf2 cisregulatory elements warrants future investigation by using in vivo Tbx2 CUT&RUN assays in cochlear tissues.
The non-sensory SCs including IBCs/IPhs that localize in the medial cochlear portion and close to IHCs are plastic and can replenish themselves after damage in neonatal cochleae [28,38]. Determining how functional IHCs can be regenerated from these cochlear non-sensory SCs is a long-term goal in the inner-ear biology field. New IHCs that were previously derived from neonatal IBCs/IPhs through Atoh1 ectopic expression alone express the nascent HC marker Myo6 or Myo7a, but fail to turn on the expression of the late IHC marker vGlut3 [26]. We propose two possible interpretations: (i) Atoh1 is transiently expressed in WT IHCs but is persistently induced in the immature new IHCs [26]; and (ii) other key genes are required to further drive the differentiation of the new IHCs. In this study, we designed a new genetic model that enables transient Atoh1 and persistent Tbx2 expression to be effectively induced in neonatal IBCs/IPhs. Excitingly, we found that the differentiation status of the new IHCs is considerably more advanced than what was previously reported [26]. Furthermore, the reprogramming efficiency achieved using Atoh1 and Tbx2 was 29.5%, which is higher than the 17.8% measured with Atoh1 alone [26]. Thus, we propose that synergistic interactions exist between Atoh1 and Tbx2 during the cell-fate conversion process, much the same as between Atoh1 and Ikzf2 described in our OHC regeneration study [19] or between Atoh1 and Pou4f3 during endogenous HC development [39]. However, it remains unclear why the neonatal IBCs/IPhs that express endogenous Tbx2 fail to become vGlut3+ new IHCs when ectopic Atoh1 is induced [26]. One possibility is that Tbx2 exerts a dose-dependent effect on cell-fate determination. Moreover, the precise roles played by Tbx2 in the medial non-sensory SCs including IBCs/IPhs remain unknown.
While we were in the process of submitting this manuscript, a similar Tbx2 conditional loss-offunction study in IHCs was reported in which Tbx2 was deleted in either embryonic or neonatal differentiating IHCs [40]. Our present data confirm that Tbx2 is required for preventing differentiating IHCs from transdifferentiation into OHCs. Moreover, we extended Tbx2 ablation in IHCs to P60/P61 and showed that Tbx2 also plays an essential role in stabilizing or maintaining the cell fate of fully mature IHCs. Notably, we also extended the characterization of iOHCs by performing single-cell transcriptomic analysis (Fig. 3), which enabled us to obtain a global gene-expression profile of the iOHCs. Because only 26.7% of the IHC genes were significantly decreased in iOHCs, our data raise the possibility that iOHCs and endogenous OHCs are not as similar as previously suggested, based on immunostaining with a couple of known IHC and OHC markers [40]. Nonetheless, both studies clearly support the notion that iOHCs resemble endogenous OHCs in several aspects. Lastly, our study showed that transient Atoh1 and permanent Tbx2 expression can convert neonatal IBCs/IPhs into vGlut3+ IHCs. Importantly, besides targeting cochlear IBCs/IPhs, Plp1-CreER+ also targets vestibular SCs and innerear glial cells [23,41]. Thus, a better option would be the future development of a new Cre or CreER mouse strain specific to IBCs/IPhs. The effect of damaging the endogenous IHCs on the transdifferentiating IBCs/IPhs into IHCs also warrants future investigation. In summary, the key findings in both studies will facilitate future treatment of IHC degeneration-associated hearing loss in the clinic.

Mouse models
The mouse strains Plp1-CreER+ (Stock# 005 975) and Rosa26-LSL-Tdtomato/+ (Ai9; Stock# 007 909) were from The Jackson Laboratory. Slc17a8 iCreER/+ , which is also known as vGlut3-P2A-iCreER/+, was reported and described in detail in our previous study [18]. Tbx2-HA/+, Tbx2 +/-, Tbx2 flox/+ , Ikzf2 V5/+ and Rosa26-LSL-TAT/+ mouse strains were generated by CRISPR/Cas9mediated homologous recombination in one-cellstage mouse zygotes. The PCR primers used for genotyping each strain and their amplicon sizes are described in Supplementary Table S3. All mice were 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.

Sample processing, immunofluorescence assays and quantification
The dissected cochlear ducts were divided into three pieces: basal, middle and apical turns. The detailed immunostaining protocols have been described previously [42]. The iOHC percentage was calculated by normalizing the numbers of all iOHCs to the total number of IHCs. The 16-kHz frequency region was selected to quantify the numbers of Ctbp2+ puncta. The percentages of vGlut3+ new IHCs or Myo7a+ nascent HCs were calculated by normalizing the numbers of vGlut3+/Tdtomato+ or Myo7a+/Tdtomato+ cells to the total Tdtomato+ cells (but only those close to IHCs). Statistical analyses were performed using one-way ANOVA and Student's t-test with Bonferroni corrections.

ABR measurement, SEM preparation and analysis
ABR measurements were performed at 4, 5.6, 8, 11.3, 16, 22.6, 32 and 45 kHz on P42 mice, following our previously published protocol [18]. Student's t-tests were used to determine the statistical significance of differences in hearing thresholds at the same frequency among distinct mice (Figs 2I and 6G). For SEM, we used the protocol described in detail in our previous study [19].

Preparation of cell suspensions, smart-seq single-cell RNA-seq and bioinformatics analysis
Cochlear samples were dissected out from Slc17a8-Ai9 mice at P14 or P30 and from Slc17a8-Tbx2cko-Ai9 mice at P14. The dissociated Tdtomato+ cells were manually picked. The picked endogenous IHCs and iOHCs were immediately subject to reverse-transcription and cDNA amplification by using a Smart-Seq HT kit (Cat# 634 437, Takara). The final libraries were subject to paired-end sequencing, which yielded ∼4 G of raw data per library. The FASTQ files of the smart-seq data were aligned to the mouse genome (GRCm38 mm10) by using Hisat2 alignment package (v2.1.0) [43]. Raw count matrices were generated using HTseq (v0.10.0) [44] and the TPM values were calculated using StringTie (v1.3.5) [45]. Trajectory analysis was performed in Monocle (v2.14.0) [46]. All the raw data of our single-cell RNA-seq analyses have been deposited in the GEO (Gene Expression Omnibus) under accession number: GSE199369.