ybx1 acts upstream of atoh1a to promote the rapid regeneration of hair cells in zebrafish lateral-line neuromasts

Pre-processing of single-cell RNA-sequencing data and pseudotime inference

We used scRNA-seq data from (Baek et al., 2022) of neomycin-treated NMs at 0, 0.5, 1, 3, 5, and 10 h after HC ablation and employed the same pre-processing steps in Seurat (version 5.0.1) as in the original study (Hao et al., 2024). In brief, cells were filtered according to previous quality control thresholds (300 <number of expressed genes <6,000; mitochondrial counts <5%), followed by count- and log-normalization (Seurat::NormalizeData), and variable feature selection (Seurat::FindVariableFeatures; number of features = 5,000). Samples from individual timepoints were then integrated using reciprocal PCA (RPCA; number of integration features = 3,500). Dimensionality reduction was performed on the integrated data using PCA (number of dimensions = 30) followed by t-SNE (number of basis PCs = 10). Cells were then labeled as either HC progenitors, young HCs, mature HCs, central SCs, dorsoventral SCs, anteroposterior SCs, amplifying SCs or mantle cells according to their expression of known marker genes in the computed Louvain clusters (number of basis PCs = 10; resolution = 1.7).

Prior to pseudotime inference, we subset separate regeneration trajectories for central SCs (0–10 hpa) and HCs (central SCs, 0–3 hpa; HC progenitors, 1–5 hpa; young and mature HCs, 3–10 hpa). Due to their strong similarities across all timepoints, we used the original, unintegrated data to compute a new t-SNE embedding for the central SCs. For the HC trajectory, we instead used the RPCA-integrated data to compute the updated t-SNE embedding. We used Slingshot (version 2.8.0) with clusters from the original timepoints or assigned cell types to infer cells’ unique pseudotime values across the trajectories for SCs and HCs, respectively (Street et al., 2018).

Transfer learning and GRN inference with DELAY

Following scRNA-seq pre-processing and pseudotime inference, we used DELAY (version 0.2.0) to infer the HC- and SC-specific GRNs during early regeneration (Reagor, Velez-Angel & Hudspeth, 2023). We trained DELAY’s RNA-specific model jointly on ground-truth interactions for both HC and SC trajectories using published targets of sox2, sox21a, six1a, and six1b from scATAC-seq of inner-ear HC regeneration (Jimenez et al., 2022). For both trajectories, we used targets of sox2 and six1b to train the model. For validation, we held out the targets of six1a and sox21a for the HCs and SCs, respectively. The complete GRNs for 303 TFs expressed in either lineage were then inferred independently for each trajectory. To construct a single GRN for early regeneration, we combined DELAY’s maximum predictions per TF-target pair and selected the top twenty regulators for each gene. This approach produced a realistic network containing an approximately scale-free outdegree distribution while limiting genes’ in-degrees to reasonable values for future in silico simulations.

De novo discovery and relative enrichment of Ybx1’s bipartite motif

To identify enriched motifs in ybx1’s predicted targets, we used twoBitToFa to gather the enhancer sequences spanning 50 kb upstream and downstream of the transcription start sites for each TF in the combined GRN for early regeneration (UCSC Command Line Utilities; UCSC zebrafish genome GRCz11). We then divided the sequences into 100 bp fragments, which we classified as either controls or predicted ybx1 targets. To identify enriched motifs in the predicted targets’ enhancers relative to those of the controls, we performed discriminative motif analysis using STREME (version 5.5.5) (Bailey, 2021). We then used FIMO (version 5.5.5) to identify individual occurrences of Ybx1’s bipartite motif across the enhancers of the predicted targets and controls (Grant, Bailey & Noble, 2011). We computed the relative enrichment of Ybx1’s bipartite motif occurrences across the promoters of the predicted targets versus the controls by first mean-normalizing then smoothing (rolling window = 50 bp) the binned motif occurrences (width = 10 bp) across the enhancers. We then performed min-max scaling of the empirical probability distributions and defined the relative enrichment as the gain in the normalized frequency of motif occurrences per bin, i.e., f (ybx1 target occurrences)/f (control occurrences).

Conservation of atoh1a’s candidate regeneration-responsive promoter element

The conservation of atoh1a’s candidate regeneration-responsive promoter element was determined using a simple alignment of that gene’s orthologues across Danio rerio (GRCz11 [ensembl]), Cyprinus carpio (Cypcar_WagV4.0 [ensembl]), Carassius auratus (ASM336829v1 [ensembl]), Labeo rohita (IGBB_LRoh.1.0 [NCBI]), Lepisosteus oculatus (LepOcu1 [ensembl]), Mus musculus (GRCm39 [ensembl]), and Homo sapiens (GRCh38.p14 [ensembl]).

Zebrafish husbandry, strains, genotyping, and HC ablations

Experiments were performed on zebrafish larvae 3–5 dpf in accordance with the standards of Rockefeller University’s Institutional Animal Care and Use Committee (protocol number 22,081 H [PRV 19,076 H]). Eggs were collected and maintained at 28.5 °C in egg water containing 1 mg/L methylene blue. Heterozygous ybx1^sa34489 zebrafish larvae were obtained from the Zebrafish International Resource Center. The following primers were used for PCR and Sanger sequencing of ybx1 mutant offspring: 5′-GTGGAGAGATGTGACAGAATATCG-3′ (forward); 5′-CATAACTGAAATAAACCCT GGAGCG-3′ (reverse). We also used Tg(atoh1a:dTomato) larvae to visualize newly regenerated HCs by fluorescence microscopy.

To ablate the first primordium-derived HCs, we treated 3 dpf larvae twice for 1 h each with 300 µM neomycin sulfate (Sigma-Aldrich, St. Louis, MO, USA) with 6 h of recovery between treatments. We performed the second treatment to kill all non-regenerating HCs that survived or matured after the first treatment. Though their HCs are somewhat less sensitive to ablation, we used 3 dpf larvae because their lateral-line NMs contained fewer HCs and were therefore more susceptible to synchronization of HC regeneration timing post-ablation.

Fixation, immunolabeling, and fluorescence imaging of zebrafish larvae

To label lateral-line HCs, we fixed 3–5 dpf zebrafish trunks with 4% formaldehyde in phosphate-buffered saline (PBS) solution with 0.1% Tween-20 (0.1% PBST) for either 1 h at room temperature or overnight at 4 °C. For end-point experiments with inbred ybx1 mutant offspring, we washed the trunks for 30 m in fresh 0.1% PBST, followed by a 1 h incubation at room temperature in 0.05% PBST with DAPI (1:200) and Alexa Fluor Plus 405 phalloidin (1:40; Invitrogen, Eugene, USA). For time-course experiments with Tg(atoh1a:dTomato) larvae, we washed the trunks in 0.5% PBST and incubated them for 1 h at room temperature in 0.05% PBST blocking solution with 1% BSA and again for 2 h in fresh blocking solution with DAPI (1:200), phalloidin (1:40) and goat anti-tdTomato DyLight550-conjugated antibodies (1:50; MyBioSource, San Diego, CA, USA). We washed the trunks once more for 20–30 m in 1X PBS before mounting and imaging at 100X on an Olympus IX83 inverted confocal microscope with a microlens-based super-resolution imaging system (VT-iSIM; VisiTech International).

For the immunolabeling of Ybx1 in wild-type, heterozygous, and homozygous ybx1 mutant larvae, we fixed 5 dpf zebrafish trunks in 4% formaldehyde for 3–4 h at room temperature, then rinsed with methanol prior to storage at −20 °C. Before immunostaining, we gradually rehydrated the trunks in fresh PBS, permeabilized them for 5 m with cold acetone (−20 °C), then washed them several times in 0.1% PBST. We incubated the trunks at room temperature for 2–4 h or overnight at 4 °C in a 0.1% PBST blocking solution containing 5% normal donkey serum and 2% DMSO, then again overnight at 4 °C in fresh blocking solution with rabbit anti-YB-1 (C-terminal) antibodies (1:100; Sigma-Aldrich, St. Louis, USA). After several washes in 0.1% PBST, we again incubated the trunks for 3 h at room temperature in fresh blocking solution with donkey Alexa Fluor 488 anti-rabbit antibodies (1:500; Invitrogen). After further washes in 0.1% PBST, we incubated the trunks for 10 m in DAPI (1:500) and stored them at 4 °C until imaging them at 60X magnification. When immunostaining Tg(atoh1a:dTomato) larvae, we additionally used monoclonal mouse anti-RFP antibodies (1:1000; Invitrogen) and donkey Alexa Fluor 561 anti-mouse antibodies (1:500; Invitrogen) during the primary and secondary incubations, respectively.

Quantification and statistical analyses

For the quantification of HC and nonsensory cell counts across endpoint and time-course experiments, each data point n represented a single NM and replicates N represented individual larvae. In general, these counts corresponded to the first-primordium derived anteroposterior NMs of the posterior lateral line from the second to the fifth NM, though sometimes including the first, sixth, or terminal NMs. For the quantification of Ybx1 immunofluorescence in atoh1a⁺ cells, data points n represented single cells. Wherever present, the reported average values indicated the arithmetic mean over the pooled data points n and the error bars indicated the standard error of the mean. The statistical significance for the mean cell-count differences—as well as for Ybx1 immunofluorescence—was computed using one-sided or two-sided T-tests for HCs and atoh1a⁺ cells or nonsensory cells, respectively, and the statistical power was computed using Cohen’s d as the effect size. For the quantification of NMs with ≥1 new HC, the statistical significance was computed using a one-sided Fisher’s exact test for proportions.

Supervised inference of the early GRN in regenerating central SCs and HCs

To reconstruct lateral-line NMs’ dynamic GRNs, we fine-tuned DELAY on published scRNA-seq data of HCs and SCs immediately following injury, from zero to ten hours post ablation (hpa) of HCs (Baek et al., 2022). As separate ground truth, we also used published targets of sox2, sox21a, six1a, and six1b from single-cell ATAC-seq of inner-ear HC regeneration (Jimenez et al., 2022). To gather the original scRNA-seq of early NM regeneration, (Baek et al., 2022) sequenced the transcriptomes of single cells belonging to several sensory and non-sensory-cell types, including HC progenitors, young HCs, mature HCs, central SCs, dorsoventral (DV) SCs, anteroposterior (AP) SCs, amplifying SCs and mantle cells (Fig. 1A). From these cell types, we subsequently subset two trajectories that corresponded to central SC regeneration (0–10 hpa) and HC regeneration—consisting of central SCs (0–3 hpa), HC progenitors (1–5 hpa), and young and mature HCs (3–10 hpa)—which we used to infer trajectory-specific pseudotime values (Fig. 1B).

We trained DELAY on these trajectories and inferred the unique HC- and SC-specific GRNs for 303 TFs expressed in either lineage. To highlight the most important TFs across both GRNs, we combined DELAY’s maximum predictions per TF-target pair and selected the top twenty regulators for each gene, which resulted in a combined GRN showcasing the key TFs and interactions coordinating NMs’ early injury response and HC and SC specification (Figs. 1C, 1D). The central hub in the combined network, Y-box binding protein 1 (ybx1), is a predicted regulator of 255 TFs involved in the regeneration of HCs and central SCs. Although no HC- or NM-specific roles are currently known, Ybx1 is a broadly expressed DNA- and RNA-binding protein that regulates transcription, translation, and cellular stress responses in other tissues (Zasedateleva et al., 2002; Sun et al., 2018; Guarino et al., 2019). Baek et al. (2022) detected ybx1 mRNA in at least 50% of all NM cells, and especially in HC progenitors and young HCs, making it one of the highest-expressed TFs across the two trajectories (Fig. 1E).

Ybx1 uses proximal bipartite motifs to regulate target genes’ expression

Because DELAY relies primarily on highly expressing cells, strong mRNA expression provided a robust signal to identify ybx1’s putative targets (Reagor, Velez-Angel & Hudspeth, 2023). The algorithm’s low false-positive rate also facilitated de novo discovery of enriched motifs across the enhancers of ybx1’s predicted targets. For each TF in the combined GRN, we gathered the enhancer sequences spanning 50 kb upstream and downstream of their transcription start sites (TSSs) and divided the sequences into 100 bp fragments, which we classified either as controls or as predicted ybx1 targets. We then used STREME to perform discriminative motif analysis and identify enriched motifs in the predicted targets’ enhancers that were similar to Ybx1’s known DNA-binding motifs (Bailey, 2021). We previously used the same approach to identify enriched motifs whose de novo sequences closely resembled known DNA-binding motifs for TFs such as Gfi1 and Hey1 during murine inner-ear HC development (Reagor, Velez-Angel & Hudspeth, 2023).

Using its highly conserved cold-shock domain (CSD), Ybx1 can interact with a broad range of sequences commonly comprising the consensus motif 5′-CNNC-3′or related sequences (Zasedateleva et al., 2002). Ybx1 therefore possesses immense flexibility to recognize and regulate the expression of target genes whose regulatory sequences contain variants of this short motif. Moreover, Ybx1 binds to single-stranded DNA (ssDNA)—often near other ssDNA-binding factors, or itself—to melt double-stranded DNA and regulate the transcription of its targets (Lasham et al., 2000; Zasedateleva et al., 2002; Lyabin, Eliseeva & Ovchinnikov, 2014). Proximally to its predicted targets’ TSSs in the combined HC and SC GRN, we discovered an enriched, bipartite motif that contained two sequences that resembled Ybx1’s forward and reverse-complement DNA-binding sites (Fig. 1F). We subsequently used FIMO to identify 239 TFs in the combined GRN whose enhancers or promoters contained at least one occurrence of the de novo motif, 199 of which were also predicted targets of ybx1 in the combined GRN (Fig. 1G) (Grant, Bailey & Noble, 2011). Together, these results suggest that Ybx1 can regulate diverse targets in regenerating HCs and central SCs by recognizing bipartite motifs that contain both forward (5′-CNNC-3′) and reverse-complement (5′-GNNG-3′) DNA-binding motifs.

A candidate regeneration-responsive promoter element for atoh1a

We noticed that one Ybx1 bipartite motif overlapped with the start codon of atoh1a, suggesting that ybx1 operates directly upstream of HC specification. Alignment of the surrounding nucleotides revealed a candidate regeneration-responsive promoter element (cRRPE) in cyprinids—carps and minnows—that contained two variants of Ybx1’s forward (5′-CAAC-3′) and reverse-complement (5′-GATG-3′) DNA-binding sites (Fig. 1H). The orthologous alignment of atoh1a across cyprinids including Danio rerio, Cyprinus carpio (common carp), Carassius auratus (goldfish), and Labeo rohita (rohu) as well as other species such as Lepisosteus oculatus (spotted gar), Mus musculus (mouse), and Homo sapiens (human) suggested an origin for the cRRPE no later than the last common ancestor of the Neopterygii (i.e., carps, minnows, and gars), resulting from the emergence of a forward Ybx1 DNA-binding site near a preexisting reverse-complement site.

We also identified in the 5′-untranslated region (UTR) of atoh1a a conserved element with a Ybx1 RNA-binding site—the so-called “dorsal localization element” (DLE), that includes the sequence AGCAC followed by a short hairpin or stem-loop (CCA-N₆-TGG) (Zaucker et al., 2018). This observation suggests that the cRRPE might also have evolved to regulate atoh1a’s translation. Ybx1 regulates the expression of multiple Nodal pathway genes in four-cell-stage embryos using DLEs to sequester transcripts and co-regulate their translation (Zaucker et al., 2018). Our identification of an identical element in the 5′UTR of atoh1a suggests that Ybx1 similarly regulates atoh1a’s translation, especially during periods of cellular stress when Ybx1’s expression is at its highest. We further identified an internal ribosome entry site (IRES) adjacent to atoh1a’s DLE-like element, which is unexpected for a monocistronic mRNA and might also support atoh1a’s translation during periods of cellular stress (Lyabin, Eliseeva & Ovchinnikov, 2014; Weingarten-Gabbay et al., 2016). Overall, our observations of multiple, conserved DNA- and RNA-binding sites suggests that—in addition to its putative role in global transcriptional regulation—Ybx1 specifically regulates atoh1a’s expression at multiple levels.

ybx1 haploinsufficiency impairs the regeneration of lateral-line HCs

To investigate the impact of ybx1 on HC development and regeneration, we obtained ybx1^sa34489 mutant zebrafish larvae that possessed an early stop in the ninth codon of ybx1. Because ybx1 does not possess an alternative start codon prior to the CSD (i.e., amino acids 56–104), homozygous ybx1^sa34489 larvae should not contain functional Ybx1 proteins with nucleic acid-binding capabilities (Kumari et al., 2013). We inbred heterozygous ybx1 mutant adults to obtain wild-type, heterozygous, and homozygous mutant larvae, which we then treated with neomycin to ablate their lateral-line HCs (Fig. 2A). At five days post-fertilization (5 dpf), we did not observe a difference in the number of HCs between the NMs of untreated ybx1 mutant larvae and those of their wild-type siblings (p > 0.2; Fig. 2B; Supplementary Tables). However, the NMs of both the heterozygous and homozygous ybx1 mutant larvae regenerated ∼1.5 fewer HCs by two days post ablation (2 dpa) than their wild-type siblings (p < 0.02; Fig. 2B; Supplementary Tables). Moreover, we did not observe a statistically significant difference in the number of DAPI⁺ nonsensory cells between regenerating NMs from wild-type and ybx1 mutant larvae (p > 0.9; Fig. 2C). These results suggest that ybx1^sa34489 is a dominant mutation that specifically disrupts the regeneration of lateral-line HCs in both heterozygous and homozygous ybx1 mutant larvae.

To confirm that our mutants possessed no or reduced expression of Ybx1, we used commercially available anti-Ybx1 antibodies to perform immunolabeling of NMs from wild-type, heterozygous, and homozygous ybx1 mutant larvae. As expected, we did not observe any Ybx1 expression in NMs of homozygous ybx1 mutants (Fig. 2D). For immunolabeling of heterozygous and wild-type larvae, we additionally bred our heterozygous ybx1 adults with Tg(atoh1a:dTomato) zebrafish to differentiate between Ybx1 expression in nonsensory cells versus in HC progenitors and young HCs. These experiments revealed that Ybx1 was broadly expressed across the cytoplasm of HCs, nonsensory cells, and peridermal cells from both the wild-type and heterozygous ybx1 mutant larvae (Fig. 2E). Strong expression in the SCs and peridermal cells resembled previous immunostaining in caudal fins of adult fish, where Ybx1 contributes to stress granule formation following injury (Guarino et al., 2019). However, our quantification of Ybx1’s immunofluorescence in atoh1a⁺ cells confirmed that the HC progenitors and young HCs from NMs of heterozygous mutants expressed significantly less Ybx1 than their wild-type siblings (p < 0.001; Fig. 2F). This cell type-specific decrease in Ybx1 expression may therefore explain the observed haploinsufficiency of heterozygous ybx1 mutant larvae during regeneration.

ybx1 promotes the rapid regeneration of lateral-line HCs following ablation

Because ybx1 is the central hub in the combined GRN for early regeneration, we wondered whether mutants’ small but consistent deficits in HC number indicated an early delay in regeneration. To test this hypothesis, we again bred our heterozygous ybx1 mutants with Tg(atoh1a:dTomato) zebrafish to yield equal numbers of wild-type sibling and heterozygous ybx1 mutant offspring that expressed dTomato in their newly regenerated HCs (Fig. 2G). Because atoh1a expression decreased after ∼12 h in mature HCs, the expression of atoh1a was a convenient marker for identifying new HCs as well as the downstream impacts of perturbations to that gene’s putative regulators such as ybx1. We therefore used these larvae to perform time-course observations and quantified new HCs in regenerating NMs at 12, 16, 20, and 24 hpa (Fig. 2H). These observations, we reasoned, might reveal whether wild-type and heterozygous ybx1 mutant larvae possessed different rates of HC addition or induction timing post-ablation.

As expected, we observed that heterozygous ybx1 mutants exhibited delayed onset of HC regeneration following damage. Across mutant larvae, the number of new HCs per NM remained ∼1 between 12–16 hpa despite increasing to ∼1.75 in wild-type siblings (p < 0.01 at 16 hpa; Fig. 2H; Supplementary Tables). From 16–20 hpa, NMs from the heterozygous mutants belatedly upregulated their regeneration of new HCs, lagging their wild-type siblings by roughly 4 h (p < 0.01; Fig. 2H; Supplementary Tables). The proportion of wild-type NMs with ≥1 newly regenerated HC also led heterozygous mutants by approximately 4 h, reaching ∼80% by 16 hpa versus 20 hpa in mutants (Fig. 2I; Supplementary Tables). Interestingly, NMs from both wild-type and heterozygous larvae failed to recover the same rate of HC addition by 24 hpa as their untreated controls, indicating that normal development alone does not account for the delayed regeneration. Overall, these results demonstrated that ybx1 lies upstream of atoh1a during early HC regeneration and promotes ∼20% faster regeneration in wild-type animals.