Comprehensive long-read cDNA sequencing of cerebellar granule cell progenitors reveals extensive isoform diversity

To comprehensively investigate the isoform landscape in cerebellar GCPs, we performed long-read cDNA sequencing on GCP samples using a Nanopore sequencer [17] (Fig 1A). Our analysis identified a total of 99,010 transcripts (Fig 1B). Among the identified transcripts, known isoforms accounted for a total of 58.3%, comprising 47.7% Full Splice Match (FSM) isoforms that perfectly matched existing annotations, and 10.6% Incomplete Splice Match (ISM) isoforms (Fig 1B). We also detected a substantial number of novel isoforms, with 14.8% being Novel In Catalog (NIC) and 17.8% being Novel Not In Catalog (NNIC) (Fig 1B). The remaining transcripts included categories such as Genic Genomic, Antisense, Fusion, Intergenic, and Genic Intron. This broad range of isoform categories highlights the extensive complexity of the GCP transcriptome captured by long-read sequencing.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Comprehensive long-read cDNA sequencing unveils extensive isoform diversity and key transcription factors in cerebellar GCPs.

A. Schematic diagram of the cDNA long-read sequencing analysis workflow using cerebellar granule cell progenitors (GCPs). The analysis is based on data from n = 3 independent biological replicates. Created in BioRender. Hoshino, M. (2026) https://BioRender.com/22hzij1. B. Classification of all identified transcripts (99,010 in total) based on Sqanti3 [59] categories, displaying the proportion of Full Splice Match (FSM), Incomplete Splice Match (ISM), Novel In Catalog (NIC), Novel Not In Catalog (NNIC), and other categories. C. Distribution of isoform counts per gene among genes with at least one transcript meeting the TPM > 1 screening criterion in all three replicates. The graph shows the percentage of genes with 1, 2, 3, 4, or ≥5 detected isoforms (11,836 genes in total). D. Gene Ontology (GO) enrichment analysis for genes with high isoform diversity (defined as ≥5 isoforms with TPM > 1 in all three replicates; 749 genes total), performed using Enrichr [63–65] (GO Biological Process 2025). E. Venn diagram showing the intersection of three gene lists to identify key transcription factors relevant to cerebellar development: (1) genes with high isoform diversity, (2) known transcription factors (from Mus musculus TFDB v4.0), and (3) genes associated with “Brain Development” term (GO:0007420) from the Enrichr. The analysis identified three common transcription factors: Meis1, Pax6, and Nfib. The complete list of high-diversity genes is provided in the S1 Data. F. Structures and expression levels of identified Meis1 transcript isoforms. Top: Isoform structures visualized with ggtranscript [66]. Bottom: Corresponding expression levels in Transcripts Per Million (TPM). G. Predicted protein domain structures of coding Meis1 isoforms. At the top, the full-length protein-coding exons are aligned, with numbers indicating the corresponding amino acid residues encoded by each exon to clarify the relationship between exons and protein domains. Protein sequences were obtained via Sqanti3, with domains predicted by InterPro [67] and visualized using drawProteins [68]. Key conserved domains are shown. Locations of the putative Nuclear Export Signal (NES; blue) and Nuclear Localization Signal (NLS; red) were manually annotated. Some isoforms lack the C-terminal homeodomain containing the NLS. The data underlying this figure can be found in S5 Data.

https://doi.org/10.1371/journal.pbio.3003897.g001

To define a transcript set for downstream analysis based on expression level, we next assessed their expression levels. A histogram of Transcript Per Million (TPM) values revealed that a substantial fraction of transcripts, particularly the novel isoforms, were expressed at low levels (S1C Fig). Since low-abundance transcripts can represent either rare, functional isoforms or transcriptional noise, we applied an expression threshold to define a working dataset for downstream analysis.

To understand the extent of isoform diversity per gene, we analyzed the number of isoforms produced by each gene. Initially, without filtering by expression level, we found that a substantial proportion of genes (30.86%) expressed five or more isoforms (S1D Fig). To examine isoform diversity using a consistent expression criterion, we applied a screening criterion, retaining only those transcripts with a TPM value greater than 1 in each of our three replicates. This analysis identified 11,836 genes with at least one transcript meeting this criterion. Of these, 48.5% expressed a single isoform, while a notable 6.33% of genes (749 genes) expressed five or more isoforms, highlighting a subset of genes with particularly high isoform diversity (Fig 1C and S1 Data).

We then performed Gene Ontology (GO) enrichment analysis for the 749 genes exhibiting high isoform diversity (≥5 isoforms with TPM > 1), focusing on Biological Process terms (Fig 1D). The top enriched GO terms were predominantly related to mRNA splicing (e.g., “mRNA Splicing via Spliceosome,” “RNA Processing,” “Regulation of mRNA splicing”) and protein–RNA complex assembly. This enrichment suggests that genes with high isoform diversity are themselves enriched in functions related to the regulation and execution of transcript diversification, indicating a self-reinforcing mechanism of complexity in the transcriptome.

Identification of transcription factors with extensive isoform diversity in GCPs

To understand how this extensive isoform diversity regulates GCP differentiation, we focused on transcription factors, as they are the master regulators that orchestrate the gene expression programs driving this process. We therefore performed a Venn diagram analysis (Fig 1E). This analysis combined the 749 high isoform diversity genes with a list of transcription factors and 171 genes annotated with the “Brain Development” GO term (GO:0007420). This intersection revealed three common transcription factors: Meis1, Pax6, and Nfib. All of these transcription factors have been implicated in cerebellar development, with their disruption leading to structural or developmental abnormalities [16,18–20].

For Meis1, we detected seven transcripts (five novel), predicted to encode several protein variants. Most notably, these included a canonical full-length isoform and a major novel isoform lacking the DNA-binding homeodomain (Fig 1F and 1G; S2 Data) [21]. The absolute read counts for each Meis1 transcript variant identified in Fig 1F are provided in S2 Data. For Pax6, we identified 10 transcripts (four novel). Our data captured its well-established diversity, including known isoforms like Pax6(5a) that arise from subtle variations within the paired domain to alter DNA-binding specificity [22,23]. In addition to these, our analysis revealed more drastically altered novel variants, including the one that lacks the paired domain entirely while retaining the homeodomain, and another predicted to be a short protein lacking both canonical domains (S1E and S1F Fig). For Nfib, we detected 14 transcripts (eight novel). This aligns with the known strategy for Nfib, where functional diversity often arises from variations in the C-terminal transactivation domain while the N-terminal DNA-binding domain is conserved, generating proteins such as dominant-negative isoforms [24,25]. Our analysis expanded this repertoire, uncovering novel variants with more substantial changes, including one completely lacking the CTF/NFI C-terminal domain itself, and another lacking all predicted domains (S1G and S1H Fig). Among these three factors, we chose to focus on Meis1 for our in-depth functional analysis. This decision was based on two key points. First, in contrast to the isoforms of Pax6 and Nfib that primarily suggest changes in transcriptional activity, the diversity of Meis1 isoforms pointed to a novel regulatory mechanism based on subcellular localization. Specifically, we found that a canonical full-length isoform containing a nuclear localization signal (NLS) and a major homeodomain-less variant that lacks the NLS but retains domains associated with a nuclear export signal (NES) were detected in GCPs [12]. Second, Meis1’s established role as an upstream regulator of Pax6 placed this potential mechanism in a critical biological context. Therefore, we proceeded to use Meis1 as a model to understand how the structural diversity of its isoforms orchestrates the process of neural progenitor differentiation.

Meis1 gene produces two major, spatially separated protein isoforms in GCPs

To examine the protein expression of MEIS1 isoforms in GCPs, we performed Western blotting with a pan-MEIS1 antibody (Fig 2A, left panel). We observed a major band around 50 kDa, consistent with the predicted size of the full-length protein, together with a lower-molecular-weight band. As these were the only major protein products robustly detected, our subsequent functional analysis focused on these forms. The lower-molecular-weight band was consistent with a previously described homeodomain-less isoform, “MEIS1D,” reported in colorectal cancer [21]. Based on this, we hypothesized that the lower-molecular-weight band in GCPs represents homeodomain-less isoform. To identify this lower-molecular-weight band as the homeodomain-less isoform, we generated a specific antibody against its unique 5-amino acid C-terminus (S2A Fig). When the same membrane was re-probed with this newly generated antibody, the upper band corresponding to MEIS1-FL was no longer detected, whereas the lower-molecular-weight band was specifically recognized (Fig 2A, middle panel). To further exclude the possibility of background reactivity, we performed immunoprecipitation followed by Western blotting. Immunoprecipitation of GCP lysates with the MEIS1-HdL-specific antibody followed by immunoblotting with a pan-MEIS1 antibody detected a single band that matched the mobility of the lower-molecular-weight band observed in whole-cell lysates (S2B Fig). Based on this protein evidence, we hereafter refer to the full-length protein as MEIS1-FL (NCBI: NP_001180200.1; Ensembl: ENSMUST00000068264.14) and the homeodomain-less isoform as MEIS1-HdL (transcript10374.11.nic in Fig 1G). Thus, western blot and RT-PCR analyses collectively demonstrate that GCPs co-express two major Meis1 isoforms, FL and HdL, at both the protein and transcript levels (Fig 2A and 2B).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Meis1 isoforms exhibit distinct expression patterns and subcellular localizations in GCPs.

A. Immunoblotting of P7 GCP lysates using the indicated antibodies. A major band at ~50 kDa corresponds to MEIS1-FL. A lower-molecular-weight band is detected by the pan-MEIS1 antibody and is specifically recognized by the MEIS1-HdL-specific antibody upon re-probing of the same membrane. B. Meis1 PCR amplification using a negative control (RT⁻), plasmids coding for Meis1-FL or Meis1-HdL, or P6 GCP cDNA (RT⁺) templates. Upper panel: RT-PCR performed with a primer pair that amplifies only Meis1-FL (primers recognizing exon8 and exon11). Lower panel: RT-PCR was performed using primer pairs that amplify only Meis1-HdL (primers recognizing the start codon and the junction of exon7-9). Both Meis1 isoforms are detected by RT-PCR. C. Representative immunofluorescence images showing the subcellular localization of MEIS1-FL-EGFP and MEIS1-HdL-EGFP in P9 cerebellar GCPs electroporated in vivo at P8. Sections were immunostained for GFP (green, MEIS1 isoforms), N-cadherin (magenta, cell boundaries), and counterstained with DAPI (blue, nuclei). D. Quantification of the cytosolic-to-total GFP signal ratio in individual electroporated cells from (C). The cytosolic region was defined as the N-cadherin-positive cellular area excluding the DAPI-stained nucleus. (n = 25 cells from 3 mice for MEIS1-FL; n = 25 cells from 4 mice for MEIS1-HdL). E. Schematic of GCP separation via FACS, using Atoh1^AtGFP/⁺; Ki67^KiRFP/⁺ mice. This method isolates three distinct populations used for subsequent analysis: ATOH1-expressing GCPs (GFP+RFP+), ATOH1-nonexpressing GCPs (GFP−RFP+), and postmitotic granule cells (GCs; GFP−RFP−). F. Relative mRNA expression of Meis1-FL and Meis1-HdL (norm. to 18s rRNA) in the cell populations defined in (E), as quantified by qRT-PCR. (n = 3 samples from 3 mice). The data underlying this figure can be found in S5 Data.

https://doi.org/10.1371/journal.pbio.3003897.g002

Next, we investigated the subcellular localization of MEIS1 isoforms in GCPs. Subcellular fractionation of GCP lysates followed by Western blotting with a pan-MEIS1 antibody revealed that the ~50 kDa band (MEIS1-FL) was predominantly localized in the nuclear fraction, while the lower band (MEIS1-HdL) was found primarily in the cytoplasmic fraction (S2C Fig). This observation aligns with our structural prediction that MEIS1-FL contains an NLS within its homeodomain, whereas MEIS1-HdL, lacking the homeodomain, retains an NES (Fig 1G). To investigate this differential localization in the developing cerebellum, we introduced plasmids encoding MEIS1-FL-EGFP and MEIS1-HdL-EGFP fusion proteins into GCPs in the postnatal day 8 (P8) cerebellum via in vivo electroporation, and analyzed them at P9. Immunofluorescence staining showed that MEIS1-FL-EGFP localized strongly to the nucleus, while MEIS1-HdL-EGFP exhibited robust cytoplasmic localization (Fig 2C and 2D). We obtained consistent results in N2a cells. Subcellular fractionation of N2a cells overexpressing untagged constructs revealed that MEIS1-FL was largely nuclear but also present in the cytoplasm, whereas MEIS1-HdL was predominantly cytoplasmic (S2D Fig). Consistent with this, immunofluorescence staining of N2a cells overexpressing EGFP-tagged constructs (S2E Fig) further confirmed that MEIS1-FL-EGFP localized strongly to the nucleus, while MEIS1-HdL-EGFP showed robust cytoplasmic localization. Taken together, these results demonstrate that MEIS1 isoforms occupy distinct subcellular compartments in GCPs, and this physical separation strongly suggests they have distinct functional roles.

Differential expression profiles of Meis1-FL and Meis1-HdL during GCP differentiation

We next examined the developmental stage-specific expression of Meis1-FL and Meis1-HdL in GCPs. Our previous work has shown a two-step amplification model for GCP differentiation, where highly proliferative, ATOH1-expressing GCPs give rise to an intermediate, transit-amplifying population that is ATOH1-negative and NEUROD1-positive, before finally exiting the cell cycle to become GCs [9]. To distinguish and isolate these distinct progenitor populations, we utilized Atoh1^AtGFP/+; Ki67^KiRFP/+ knock-in mice [26,27]. We validated these reporter mice by demonstrating that GFP signal largely overlapped with Atoh1 immunostaining (S3A Fig), and RFP signal overlapped with KI67 immunostaining (S3B Fig). The combined GFP and RFP expression patterns showed that GFP was more restricted to the oEGL, while RFP was expressed more broadly but still predominantly in the oEGL (S3C Fig). This finding is consistent with the known restricted localization of endogenous ATOH1 to the oEGL [8], further validating our reporter mice. We used FACS to isolate three populations based on our model: ATOH1-expressing GCPs (GFP+RFP+), ATOH1-nonexpressing GCPs (GFP−RFP+), and postmitotic GCs (GFP−RFP−) (Figs 2E and S3D). The specificity of our sorting strategy was further validated by qRT-PCR for Atoh1, which showed Atoh1 expression exclusively in GFP+RFP+ sorted cells (S3E Fig). Quantitative RT-PCR analysis of Meis1 isoforms across these sorted populations revealed that Meis1-HdL expression was highly enriched in GFP+RFP+ sorted cells (ATOH1-expressing GCPs), with minimal expression in GFP−RFP+ sorted GCPs (ATOH1-nonexpressing, NEUROD1-expressing GCPs) and GFP−RFP− sorted cells (GCs) (Fig 2F, lower graph). In contrast, Meis1-FL transcript levels remained relatively constant across ATOH1-expressing GCPs, ATOH1-nonexpressing GCPs, and postmitotic GCs (Fig 2F, upper graph). Taken together with our localization data, these findings reveal a clear spatio-temporal partitioning of the two major MEIS1 protein products. MEIS1-HdL is confined to the cytoplasm, and its transcript levels are highest at the earliest, most proliferative stage of GCP development. This physical and developmental segregation suggests they have distinct functional roles.

Opposing functions of MEIS1-FL and MEIS1-HdL in regulating GCP fate in vivo

To examine the functions of MEIS1-FL and MEIS1-HdL in cerebellar granule cell development, we performed in vivo overexpression experiments in GCPs. We introduced expression vectors encoding either Meis1-FL or Meis1-HdL into ATOH1-expressing GCPs via electroporation into the P8 cerebella. At 72 hours post-electroporation (P11), the cerebella were collected, fixed, and immunostained. Electroporated cells were identified by co-electroporated histone H3.1-GFP.

In control animals, the GFP+ cells comprised approximately 50% GCPs (Ki67+ cells) and 50% differentiating/differentiated GCs (p27+ cells) (Figs 3A, 3D, 3E, and S4A). Overexpression of Meis1-FL significantly decreased the proportion of GCPs (Ki67+ cells) and increased the proportion of GCs (p27 + cells) (Figs 3B, 3D, 3E, and S4B). Conversely, Meis1-HdL overexpression led to an increase in the proportion of GCPs (Ki67+ cells) and a decrease in the proportion of GCs (p27+ cells), indicative of maintaining an undifferentiated, proliferative state (Figs 3C, 3D, 3E, and S4C). Similarly, MEIS1-FL overexpression decreased the proportion of ATOH1-expressing GCPs, whereas MEIS1-HdL overexpression increased ATOH1-expressing GCPs (Fig 3A–3C and 3F). These results indicate that MEIS1-FL and MEIS1-HdL have opposing effects on GCP differentiation. This promotion of GCP differentiation by MEIS1-FL is consistent with our previous observation that conditional deletion of Meis1, an allele that disrupts all homeodomain-containing isoforms (most notably the full-length MEIS1-FL), delayed GCP differentiation [16].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. MEIS1-HdL maintains the undifferentiated state of GCPs, while MEIS1-FL promotes GC differentiation.

A–C. Representative immunofluorescence images of P11 cerebella following in vivo electroporation at P8. Cerebellar sections were immunostained for ATOH1 (magenta) and KI67 (cyan). Electroporated cells, identified by co-electroporated H3.1-EGFP (yellow), were transfected with a control vector (A), Meis1-FL (B), or Meis1-HdL (C). D–F. Quantification of electroporated cell differentiation status. The percentage of GFP+ cells expressing KI67 (D), p27 (E), or ATOH1 (F) was quantified from sections described in (A–C). (n = 4 mice per group). The data underlying this figure can be found in S5 Data.

https://doi.org/10.1371/journal.pbio.3003897.g003

To further investigate the physiological function of MEIS1-HdL, we performed loss-of-function experiments using shRNA. We first generated a Meis1-HdL-specific knockdown vector (sh-Meis1-HdL), which effectively suppressed Meis1-HdL expression without affecting Meis1-FL (S5A Fig, left panel). For comparison, we also utilized a previously established pan-Meis1 knockdown vector (sh-Meis1-all) [16], which suppressed the expression of both Meis1-FL and Meis1-HdL (S5A Fig, right panel).

These shRNA vectors, along with co-electroporated histone H3.1-GFP to identify electroporated cells, were introduced into GCPs in the P8 cerebella via in vivo electroporation. Cerebellar sections were analyzed by immunostaining 72 hours post-electroporation (P11). Specific knockdown of Meis1-HdL significantly reduced the proportion of proliferative GCPs (Ki67+ cells) and increased the proportion of postmitotic GCs (p27+ cells) (Figs 4A, 4D, 4E, and S5B). A similar trend was observed for ATOH1-expressing GCP population (Fig 4A and 4F). These findings suggest that the endogenous function of MEIS1-HdL is to suppress GCP differentiation. Crucially, the phenotypes induced by sh-Meis1-HdL were rescued by co-electroporation with a knockdown-resistant Meis1-HdL expression vector (Res-Meis1-HdL), restoring the proportion of KI67+ (Fig 4B and 4D) and ATOH1+ cells (Fig 4B and 4F), and decreasing p27+ cells (Figs 4E and S5C). This rescue demonstrates that the observed effects are due to the loss of MEIS1-HdL, thereby establishing its critical role in suppressing GCP differentiation.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Endogenous MEIS1-HdL participates in maintaining GCPs in an immature and proliferative state.

A–C. Representative immunofluorescence images of P11 cerebella following in vivo electroporation at P8. Cerebellar sections were immunostained for ATOH1 (magenta) and KI67 (cyan). Electroporated cells, identified by co-electroporated H3.1-EGFP (yellow), were transfected with sh-Meis1-HdL (A), sh-Meis1-HdL plus a knockdown-resistant rescue construct (Res-Meis1-HdL) (B), or sh-Meis1-all (C). D–F. Quantification of electroporated cell differentiation status. The percentage of GFP+ cells expressing KI67 (D), p27 (E), or ATOH1 (F) was quantified from sections described in (A–C). (n = 4 mice for Control, sh-Meis1-HdL, and Rescue; n = 3 mice for sh-Meis1-all). The data underlying this figure can be found in S5 Data.

https://doi.org/10.1371/journal.pbio.3003897.g004

Notably, knocking down both isoforms with sh-Meis1-all did not significantly alter the overall proportion of KI67+ or p27+ cells (Figs 4C, 4D, 4E, and S5D). We interpret this lack of phenotype as resulting from mutual cancellation; the pro-differentiative effect of MEIS1-HdL loss is likely balanced by the anti-differentiative effect of MEIS1-FL loss. However, the specific reduction in the ATOH1-expressing GCP population (Fig 4C and 4F) may suggest a dominant role for MEIS1-HdL in maintaining this progenitor state, an effect that is not fully compensated for by the simultaneous loss of MEIS1-FL.

MEIS1-FL and MEIS1-HdL stabilize ATOH1 protein levels

The precise control of ATOH1 protein abundance is critical for regulating the transition of GCPs from proliferation to differentiation [8,10,28–30]. Our previous work has shown that MEIS1-FL promotes transcription of Pax6, activates BMP signaling, and acts to promote ATOH1 degradation [16]. However, we found that MEIS1-HdL lacking homeodomain also affects the differentiation of GCPs to GCs (Figs 3 and 4), suggesting that MEIS1 proteins may have additional functions to regulate the amount of ATOH1. To test this, we first examined the effect of Meis1 knockdown on ATOH1 protein levels in vivo by measuring ATOH1 fluorescence intensities. In cerebella electroporated with sh-Meis1-HdL at P8 and fixed at P11, ATOH1 fluorescence intensities in electroporated cells (GFP+) were lower compared to surrounding non-electroporated cells (S6A, S6B, and S6E Fig). In contrast, KI67 fluorescence intensities were not significantly affected by sh-Meis1-HdL (S6A, S6B, and S6D Fig). This suggests that MEIS1-HdL may participate in maintaining ATOH1 protein levels in ATOH1-expressing GCPs (S6A, S6B, and S6D Fig). Similar tendencies were also observed for sh-Meis1-all electroporated cerebella (S6A and S6C–S6E Fig). However, because sh-Meis1-all was designed to suppress both MEIS1-HdL and MEIS1-FL, it was unclear whether MEIS1-FL is also capable of maintaining ATOH1 protein levels at this point.

To further investigate the machinery of MEIS1 isoforms to regulate ATOH1 protein levels, we transfected ATOH1 and MEIS1 expression vectors into N2a cells and performed immunoblotting. We used a GST-tagged ATOH1 (GST-ATOH1) expression vector to discriminate between endogenous and exogenous ATOH1 proteins, although it is known that N2a cells do not express endogenous ATOH1 under normal conditions [31]. Administration of MG132, an inhibitor of proteasome-dependent protein degradation, enhanced the ATOH1 signals (S7A Fig, lane 5) relative to the control (S7A Fig, lane 1), consistent with the previous findings that ATOH1 is degraded via proteasome-dependent protein degradation [10,11,32,33]. To determine the specific domains of MEIS1 required for ATOH1 stabilization, we compared the effects of MEIS1-FL, MEIS1-HdL, and a further truncated N-terminal MEIS1 fragment (1–130 aa). MEIS1 possesses a highly conserved Homeobox protein PKNOX/Meis N-terminal domain (71–184 aa), which encompasses the MEIS Homology domains (MH-A, 72–111 aa; and MH-B, 136–180 aa) [34], and a C-terminal Homeodomain (HD, 290–329 aa). While MEIS1-HdL retains the intact Homeobox protein PKNOX/Meis N-terminal domain but is devoid of the HD, the MEIS1(1–130) truncation lacks both the MH-B domain and the HD. Interestingly, ATOH1 signals were also higher following co-transfection with MEIS1-HdL (lane 3 in S7A Fig) or MEIS1-FL (S7A Fig, lane 4), whereas the N-terminal MEIS1 fragment (1–130 aa) did not exhibit such an effect (S7A Fig, lane 2). To further analyze ATOH1 stability, we immunoblotted transfected N2a cells treated with cycloheximide (CHX, a protein synthesis inhibitor) for 0, 4, and 8 h. This revealed that MEIS1-HdL and MEIS1-FL both strongly suppressed ATOH1 degradation to the same extent as MG132 administration (S7B and S7C Fig). These observations suggest that both MEIS1-HdL and MEIS1-FL have the capability to suppress the proteasome-dependent degradation of ATOH1.

MEIS1 inhibits CUL3-mediated ubiquitination of ATOH1

Because the proteasome-dependent degradation is triggered by ubiquitination of proteins, we attempted to assess ATOH1 ubiquitination in cultured cells. N2a cells were co-transfected with GST-ATOH1 and HA-Ub. Then, GST-ATOH1 was pulled down by glutathione beads and immunoblotted with an anti-HA antibody. Consistent with previous findings [10,32], ATOH1 was strongly polyubiquitinated (Fig 5A, lane 1). However, co-transfection with MEIS1-FL or MEIS1-HdL significantly reduced GST-ATOH1 polyubiquitination (Fig 5A, lanes 2 and 3; Fig 5B). These findings suggest that both MEIS1-FL and MEIS1-HdL suppress the functions of molecules related to ATOH1 polyubiquitination, eventually blocking ATOH1 degradation. The fact that not only MEIS1-FL but also MEIS1-HdL does this suggests that this function is not elicited by MEIS1-induced transcriptional regulation.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. MEIS1 inhibits CUL3-mediated ATOH1 ubiquitination, leading to ATOH1 stabilization.

A. Immunoblotting of GST pull-down fractions from N2a cells co-transfected with GST-ATOH1, HA-Ub, and either MEIS1-FL or MEIS1-HdL. Polyubiquitinated ATOH1 (upper panel) was detected using an anti-HA antibody, while total GST-ATOH1 levels (middle panel) were detected using an anti-GST antibody. MEIS1 expression in input lysates was confirmed by immunoblotting with an anti-MEIS1 antibody (lower panel). B. Quantification of polyubiquitinated ATOH1 levels, normalized to total GST-ATOH1 levels from (A). C. Schematic illustrating the proteomic analysis strategy to identify ATOH1-binding molecules in the presence or absence of MEIS1 in N2a cells. D. List of ubiquitin-modifying enzymes and related proteins (identified by PANTHER GO analysis) that showed reduced binding to ATOH1 in the presence of MEIS1-FL, as identified by proteomic analysis (as described in C). E. Immunoblotting of GST pull-down fractions from N2a cells transfected with GST-ATOH1 alone or co-transfected with GST-ATOH1 and either MEIS1-FL or MEIS1-HdL. Interaction with endogenous CUL3 was determined by immunoblotting with an anti-CUL3 antibody. The interaction is disrupted by the co-expression of either MEIS1-FL or MEIS1-HdL. MEIS1 expression in input lysates was confirmed by immunoblotting with an anti-MEIS1 antibody (lower panel). F. Immunoblotting of GST pull-down fractions from N2a cells co-transfected with GST-ATOH1, HA-Ub, and either a control scramble shRNA (sh-Scramble) or a Cul3 knockdown shRNA (sh-Cul3). Polyubiquitinated ATOH1 (upper panel) was detected with an anti-HA antibody, and total GST-ATOH1 levels (middle panel) were detected with an anti-GST antibody. CUL3 expression in input lysates was confirmed by immunoblotting with an anti-CUL3 antibody (lower panel). G. Immunoblotting analysis of N2a cell lysates following cycloheximide (CHX) chase assay. N2a cells transfected with GST-ATOH1 and either sh-Scramble or sh-Cul3 were treated with CHX for 0, 4, and 8 hours. ATOH1 protein levels were detected using an anti-ATOH1 antibody, with GAPDH as a loading control. H. Quantification of ATOH1 protein levels from (G), normalized to GAPDH, showing that knockdown of Cul3 inhibits ATOH1 degradation. (n = 3 independent experiments for (B) and n = 4 independent experiments for (H)). The data underlying this figure can be found in S5 Data.

https://doi.org/10.1371/journal.pbio.3003897.g005

Based on our findings suggesting that MEIS1 isoform proteins affect the function of molecules involved in ATOH1 polyubiquitination, we searched for ATOH1-binding molecules whose binding affinities were altered by the presence of MEIS1 in cultured cells. After transfecting GST-ATOH1 or GST-ATOH1 plus MEIS1-FL into N2a cells, GST-ATOH1 was pulled down, and its binding molecules were subjected to LC–MS/MS (Fig 5C). We obtained many candidate molecules that bind ATOH1 (see S3 Data), identifying several ATOH1-binding molecules whose interaction with ATOH1 was disrupted or much reduced in the presence of MEIS1-FL (Fig 5D). Notably, these included the E3 ubiquitin ligase Cullin-3 (CUL3) and its known partners UBE2M and CAND1 (Fig 5D and S3 Data) [35–37], suggesting that CUL3 may act as an E3-ligase for ATOH1. To date, HUWE1 is the only E3-ligase reported to participate in ATOH1 degradation [10,32]. Although we found that HUWE1 binds to GST-ATOH1, its binding ability was not affected by MEIS1-FL, which further focused our attention on CUL3.

To confirm the proteomic finding that MEIS1 disrupts the ATOH1-CUL3 interaction, we performed co-immunoprecipitation experiments. We co-transfected GST-ATOH1 and MEIS1-FL or MEIS1-HdL into N2a cells, followed by pull-down with GST-ATOH1. Similar to the LC–MS/MS results, immunoblotting showed that CUL3 bound ATOH1 in the absence of MEIS1 (Fig 5E, lane 1), but not in the presence of MEIS1-FL or MEIS1-HdL (Fig 5E, lanes 2, 3).

Next, we examined the involvement of CUL3 in ATOH1 polyubiquitination. GST-ATOH1 and HA-Ub were co-transfected with a control or Cul3-knockdown vector (sh-Cul3) into N2a cells, which were subjected to pulldown and immunoblotting with the indicated antibodies (Fig 5F), in the absence of MG132. GST-ATOH1 polyubiquitination was strongly suppressed by Cul3-knockdown (Fig 5F), suggesting that CUL3 is involved in ATOH1 polyubiquitination. To further analyze the effect of Cul3 knockdown on ATOH1 stability, we immunoblotted N2a cells transfected with GST-ATOH1 and control or sh-Cul3 and administered CHX for 0, 4, and 8 h (Fig 5G). This revealed that Cul3-knockdown markedly suppressed ATOH1 degradation in N2a cells (Fig 5H). These observations suggest that CUL3 acts as a key E3 ubiquitin ligase for ATOH1 and that MEIS1-FL and MEIS1-HdL inhibit the interaction between CUL3 and ATOH1, eventually suppressing the polyubiquitination and degradation of ATOH1.

MEIS1 inhibits the degradation of ATOH1 induced by S328 phosphorylation

Previous studies have shown that phosphorylation of ATOH1 regulates its degradation [10,32,38], which led us to the notion that MEIS1 may affect ATOH1 phosphorylation as a part of its stabilization mechanism. To test this, we initially investigated ATOH1 phosphorylation sites by LC–MS/MS using N2a cells transfected with ATOH1 alone or ATOH1 and MEIS1-FL (S8A Fig). We observed that ATOH1 phosphorylation at some sites was increased in the presence of MEIS1-FL (S8B Fig and S4 Data). Particularly, analysis of MS signal intensities indicated that S328 phosphorylation was markedly increased. Consistent with the proteomics data, immunoblot analysis using an antibody against ATOH1 p-S328 indicated that phosphorylation at S328 was much increased in the presence of either MEIS1-FL or MEIS1-HdL (S8C Fig). It has been reported that S328 phosphorylation is involved in the degradation of ATOH1 [10,32]. Accordingly, the CHX chase assay showed that ATOH1-S328A (Ser 328 of ATOH1 was replaced with Ala; an ATOH1-S328 non-phosphorylated form) was stable, whereas ATOH1-S328D (Ser 328 of ATOH1 was replaced with Asp; an ATOH1-S328 phosphomimic form) was unstable (S8D–S8H Fig). Consistent with its constitutive degradation, the steady-state baseline protein level of ATOH1-S328D was markedly lower than that of wild-type ATOH1 (S8E and S8F Fig). However, the stability of ATOH1-S328D increased in the presence of MEIS1-FL or MEIS1-HdL (S8G and S8H Fig). These observations suggest that the MEIS1 isoforms suppress ATOH1 degradation induced by phosphorylation at S328 without inhibiting phosphorylation of S328 ATOH1.

Next, GST-ATOH1 or GST-ATOH1-S328A and HA-Ub were co-transfected into N2a cells; pull-down and immunoblotting were subsequently performed with the indicated antibodies (S8I Fig). The polyubiquitination efficiency of S328A was lower than that of the wild-type ATOH1 (S8I Fig upper panel), which is consistent with the observation that ATOH1-S328A is stable (S8D and S8H Fig). Furthermore, the interaction of GST-ATOH1-S328A with CUL3 was much weaker than that of GST-ATOH1 with CUL3 (S8I Fig second panel from the top). These findings indicate that S328 phosphorylation is involved in CUL3 binding and polyubiquitination of the ATOH1 protein. MEIS1 isoforms do not suppress S328 phosphorylation but instead act to inhibit the interaction between S328-phosphorylated ATOH1 and CUL3, leading to the stabilization of ATOH1.

MEIS1 inactivates CUL3 by promoting COP9 signalosome (CSN) and CUL3 complex formation

To understand how MEIS1 inhibits CUL3 activity, we next investigated MEIS1-binding molecules in N2a cells. GST-tagged MEIS1-FL (GST-MEIS1-FL) was transfected into N2a cells, precipitated using glutathione beads, and subjected to LC–MS/MS (S9A Fig). We obtained 1,417 MEIS1-binding candidate molecules (S3 Data). Interestingly, these included several components of the COP9 signalosome (CSN) complex, such as COPS2, COPS3, COPS4, COPS6, COPS7a, and COPS8 (Fig 6A and S3 Data), suggesting that MEIS1-FL interacts with the CSN complex. It is known that the CSN complex, consisting of eight components, suppresses substrate polyubiquitination by the Cullin-family E3 protein ligases. By competitively binding Cullin proteins, the CSN complex masks the substrate-binding regions of Cullin proteins, thus inhibiting their interactions with substrates [37]. The CSN complex also suppresses protein neddylation, thereby inhibiting Cullin protein function.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. MEIS1 isoforms promote CUL3–COP9 signalosome complex formation to inactivate CUL3.

A. List of components of the COP9 signalosome (CSN) complex identified as MEIS1-binding partners via proteomic analysis (experimental scheme is shown in S9A Fig). B. Immunoblotting of GST pull-down fractions from N2a cells transfected with GST, GST-MEIS1-FL, or GST-MEIS1-HdL. Interactions with endogenously expressed COPS2 and COPS5 were determined using anti-COPS2 and anti-COPS5 antibodies. Total GST-tagged proteins were detected with an anti-GST antibody. C. Sequential co-immunoprecipitation assay demonstrating a CUL3–MEIS1–COPS6 tripartite complex. N2a cells were co-transfected with Flag-HA-COPS6 and GST-MEIS1-FL or GST-MEIS1-HdL (or GST as control). Lysates were first subjected to immunoprecipitation with an anti-HA antibody, followed by a GST pull-down. The final immunoblot for endogenous CUL3 confirmed its inclusion in a tripartite complex with MEIS1 isoforms and COPS6. D. Co-immunoprecipitation of endogenous proteins from P7 GCP lysates. Endogenous MEIS1 was immunoprecipitated using an anti-MEIS1 antibody, with IgG serving as a negative control. The resulting precipitates were then immunoblotted for the CSN components COPS4 and COPS5 to detect an interaction. E. Co-immunoprecipitation assay to assess the effect of MEIS1 isoforms on the COPS6-CUL3 interaction. Flag-HA-COPS6 was immunoprecipitated using an anti-HA antibody from N2a cells co-expressing a control vector, MEIS1-FL, or MEIS1-HdL. The immunoblots show that co-expression of either MEIS1 isoform enhances the binding of endogenous CUL3 to COPS6. F. Quantification of CUL3 signal intensity from (E). G. In vitro ubiquitination assay demonstrating that COPS6 is required for MEIS1 to inhibit ATOH1 polyubiquitination. N2a cells were co-transfected with GST-ATOH1, HA-Ub, and the indicated constructs (MEIS1-FL, sh-Cops6, or controls). Following a GST pull-down, blots were probed for HA (polyubiquitination, top panel) and GST (total ATOH1, middle panel). MEIS1 expression in input lysates was confirmed by immunoblotting with an anti-MEIS1 antibody (lower panel). The ability of MEIS1-FL to reduce ubiquitination is reversed by the knockdown of Cops6 (see S9D Fig for knockdown efficiency). H. Dose-dependent stabilization of GST-ATOH1 by MEIS1 isoforms. N2a cells were co-transfected with a constant amount of GST-ATOH1 plasmid and increasing amounts (0–5 µg) of plasmids expressing MEIS1-FL or MEIS1-HdL. Cell lysates were immunoblotted for the indicated proteins, with GAPDH as a loading control. I. Quantification of ATOH1 protein levels from (H), normalized to GAPDH and plotted against the concentration of the transfected MEIS1 expression vector. The graph illustrates the apparent difference in dose-dependent ATOH1 stabilization in this assay. Statistical significance between the overall dose-dependent stabilization effects of MEIS1-HdL and MEIS1-FL was determined using two-way ANOVA. J. Immunoblot of cytoplasmic and nuclear fractions from N2a cells co-expressing ATOH1 and HA-Ub. The blot shows that while polyubiquitinated ATOH1 is detected in both fractions, CUL3 is predominantly localized to the cytoplasm. βIII-tubulin and Lamin A/C serve as markers for the cytoplasmic and nuclear fractions, respectively. Note that the cytoplasmic ATOH1 signal is prominent due to overexpression and the long exposure time required to detect polyubiquitinated species. (n = 3 independent experiments). The data underlying this figure can be found in S5 Data.

https://doi.org/10.1371/journal.pbio.3003897.g006

To confirm whether MEIS1 interacts with the intact, functional CSN complex, we performed GST pull-down assays targeting both COPS2 (identified by LC–MS/MS) and COPS5. We specifically selected COPS5 because it acts as the essential catalytic center of the CSN complex, despite its absence from our initial MS screening due to potential technical limitations. Immunoblotting revealed that not only MEIS1-FL but also MEIS1-HdL interacted with endogenous CSN components, COPS2 and COPS5 (Fig 6B). We conducted similar pull-down and immunoblotting with HA-tagged COPS6 (HA-COPS6) and GST-MEIS1-truncated mutant (1–130 aa) or GST-MEIS1 isoforms in N2a cells. As expected, HA-COPS6 was found to bind GST-MEIS1-FL and GST-MEIS1-HdL (S9B Fig). In contrast, GST-MEIS1-1–130aa did not bind HA-COPS6 (S9B Fig). Given that this N-terminal MEIS1 fragment (MEIS1-1–130aa) does not have the ATOH1-stabilizing ability (S7 Fig), these observations suggest that binding to the CSN complex is crucial for MEIS1 proteins to stabilize ATOH1.

To test whether MEIS1, CSN complex and CUL3 form a tripartite complex, we performed a sequential immunoprecipitation experiment. HA-COPS6 and GST-MEIS1-FL or GST-MEIS1-HdL were transfected into N2a cells, followed by immunoprecipitation with HA and successive pull-down with GST. Immunoblotting revealed a significant signal for endogenous CUL3 in the presence of either MEIS1-FL or MEIS1-HdL (Fig 6C), suggesting that the CSN complex, MEIS1 isoforms, and CUL3 form a tripartite complex in cells. To test the interaction of endogenously expressed MEIS1 and CSN complex, immunoprecipitation was performed on the GCP lysates with the MEIS1 antibody recognizing both MEIS1 isoforms. Immunoblotting with COPS4 or COPS5 suggested that endogenous MEIS1 and the CSN complex interact with each other (Fig 6D). Given that MEIS1-HdL interacts with components of the CSN complex in N2a cells (Figs 6B and S9B), it was suggested that not only MEIS1-FL but also MEIS1-HdL interacts with the CSN complex in GCPs.

We next tested whether the presence of MEIS1 isoforms affects the CUL3-CSN interaction. N2a cells were transfected with HA-COPS6 and MEIS1-FL or MEIS1-HdL, followed by immunoprecipitation with the HA antibody. Immunoblotting showed that the amount of co-precipitated CUL3 was significantly greater in the presence of either MEIS1-FL or MEIS1-HdL, compared to control (Fig 6E and 6F). This suggests that MEIS1 isoforms inactivate CUL3 by enhancing its association with the CUL3-inhibitory complex, CSN. We further tested the possibility that CUL3 is required for the binding between MEIS1 isoforms and the CSN complex. However, it was not likely because the introduction of sh-Cul3 into N2a cells did not affect the binding between HA-COPS6 and MEIS1-FL or MEIS1-HdL (S9C Fig). To investigate whether the CSN complex is involved in the function of MEIS1 to suppress ATOH1 polyubiquitination, GST-ATOH1 and HA-Ub were transfected into N2a cells in the presence of MG132, followed by GST pull-down. Immunoblotting with HA revealed that MEIS1-FL introduction reduced ATOH1 polyubiquitination (Fig 6G, lanes 1 and 2). However, co-introduction of the knockdown vector for Cops6 (sh-Cops6, S9D Fig) ameliorated the effects of MEIS1-FL (Fig 6G, lane 3). These findings suggest that MEIS1 suppresses ATOH1 polyubiquitination and degradation by promoting the CSN complex–CUL3 interaction and by inhibiting the CUL3–ATOH1 interaction. When transfected into N2a cells, the expression of NRF2, NCOA3, and PP2A/C was much higher in the presence of MG132 (S9E Fig, lane 4) than in the control (S9E Fig, lane 1), consistent with previous findings that these proteins are degraded by CUL3-dependent polyubiquitination [39–41]. Co-introduction of MEIS1-FL or MEIS1-HdL strongly increased NRF2 and PP2A/C expression and moderately increased that of NCOA3 (S9E Fig, lanes 2 and 3), relative to the control (S9E Fig, lane 1). This suggests that these MEIS1 isoforms may broadly suppress CUL3-dependent degradation of various proteins.

Transfected tagged COPS6 was expressed in both the cytoplasm and nucleus in N2a cells (S9F Fig), consistent with a previous report that the CSN complex localizes in both the cell nucleus and cytoplasm. Similar COPS6 subcellular localization was observed following co-transfection with MEIS1-FL or MEIS1-HdL (S9G and S9H Fig), suggesting that neither MEIS1-FL nor MEIS1-HdL affects CSN-complex subcellular localization.

Thus, MEIS1-FL and MEIS1-HdL have similar functions in that they both inhibit the degradation of ATOH1. However, there may be differences in the strength of their effects. To test this, GST-ATOH1 was co-transfected with MEIS1-FL or MEIS1-HdL, using the indicated quantities of DNA (0–5 μg, Fig 6H). While large amounts of MEIS1-FL vector were required to suppress ATOH1 degradation, this was achieved using very small amounts of MEIS1-HdL vector (Fig 6H and 6I). In this overexpression assay, MEIS1-HdL produced detectable ATOH1 stabilization at lower plasmid inputs than MEIS1-FL. Because construct-dependent technical factors cannot be fully excluded, we interpret this as an apparent difference in dose responsiveness rather than definitive evidence of intrinsically stronger molecular activity.

To validate this spatial requirement, we generated a nuclear-targeted MEIS1-HdL by fusing it with a nuclear localization signal (NLS-Meis1-HdL). Subcellular fractionation confirmed its strong localization to the nucleus, although a minor fraction remained in the cytoplasm (S10A Fig). Crucially, this forced nuclear localization significantly attenuated the dose-dependent ATOH1-stabilizing ability of MEIS1-HdL (S10B and S10C Fig). The residual ATOH1 stabilization observed at higher plasmid dosages is likely attributable to this minor cytoplasmic fraction of NLS-MEIS1-HdL. These results support the importance of cytoplasmic localization for the full ATOH1-stabilizing activity of MEIS1-HdL. Because a minor fraction of NLS-MEIS1-HdL remained detectable in the cytoplasmic fraction, however, minor nuclear contributions cannot be formally excluded. Furthermore, because MEIS1 typically functions as a transcription factor in the nucleus through association with PBX proteins, we examined whether PBX3, one of the MEIS-interacting PBX family members expressed in this system, is detectable in the MEIS1-HdL/CSN complex. Co-immunoprecipitation assays revealed no detectable interaction between PBX3 and the MEIS1-HdL/CSN complex (S10D Fig). Because only PBX3 was directly examined in this assay, these results support a PBX3-independent mechanism for MEIS1-HdL-mediated ATOH1 stabilization, but do not exclude possible contributions from other PBX family members. Taken together, these results support a predominantly cytoplasmic mechanism for MEIS1-HdL-mediated ATOH1 stabilization, while not formally excluding additional nuclear contributions.

Fractionation using N2a cells transfected with ATOH1 and HA-Ub revealed that polyubiquitinated ATOH1 was localized in both the cytoplasm and nucleus (Fig 6J). In contrast, CUL3 was localized abundantly in the cytoplasm and negligibly in the nuclei (Fig 6J), consistent with prior suggestions [42]. These data are therefore consistent with a major role for cytoplasmic CUL3 in ATOH1 polyubiquitination under our assay conditions. Because MEIS1-HdL was localized predominantly in the cytoplasm, whereas MEIS1-FL was localized predominantly in the nucleus (S2C Fig), we interpret the difference between the two isoforms primarily in terms of differential access to the cytoplasmic CUL3/CSN machinery.

The MEIS1-HdL–CSN pathway maintains GCPs in an immature state

We showed that MEIS1-HdL introduction into the EGL suppressed the differentiation from GCPs to GCs: 3 d after electroporation, the proportions of Ki67+ and ATOH1+ GCPs were higher, while those of p27+ GCs were lower (Fig 3C–3F). On the other hand, our in vitro experiments revealed that ATOH1 stabilization by MEIS1 was mediated by the CSN complex (Fig 6G). To link these in vivo and in vitro findings, we investigated the role of the CSN complex in GCPs. First, we introduced sh-Cops6 alone into the P8 EGL. This resulted in a reduction of Ki67+ and ATOH1+ GCPs and an increase in p27+ GCs (Figs 7C–7E, S11A, and S11B), indicating that the CSN complex is essential for maintaining GCPs in an undifferentiated state. Next, we tested whether the CSN complex is required for the in vivo function of MEIS1-HdL. We co-electroporated sh-Cops6 together with the MEIS1-HdL into the P8 EGL under the experimental conditions shown in Fig 3C. The effects of MEIS1-HdL were cancelled by co-introduction with sh-Cops6 (Fig 7A–7E). This suggests that CSN is required for MEIS1-HdL to function in GCPs, and that the CSN complex works with MEIS1-HdL to suppress the GCP differentiation into GCs, during cerebellar development (Fig 7F).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. The COP9 signalosome is required for MEIS1-HdL to maintain GCPs in an undifferentiated state.

A. Representative immunofluorescence images of P11 cerebella following in vivo electroporation at P8. Sections were immunostained for ATOH1 (magenta) and KI67 (cyan). Electroporated cells are identified by co-electroporated H3.1-EGFP (yellow). These images show GCPs electroporated with MEIS1-HdL and sh-Cops6. B. Representative immunofluorescence images of P11 cerebella following in vivo electroporation at P8. Sections were immunostained for p27 (magenta). Electroporated cells are identified by co-electroporated H3.1-EGFP (green). These images show GCPs electroporated with MEIS1-HdL and sh-Cops6. C–E. Quantification of electroporated cell differentiation status. The percentage of GFP+ cells expressing KI67 (C), p27 (D), or ATOH1 (E) was quantified from cerebellar sections electroporated with the indicated vectors. (n = 4 mice for Control, MEIS1-HdL and sh-Cops6; n = 3 mice for MEIS1-HdL + sh-Cops6). F. Schematic diagram of the MEIS1-HdL function. Created in BioRender. Hoshino, M. (2026) https://BioRender.com/qvr6glt; https://BioRender.com/ceqrc9v. The data underlying this figure can be found in S5 Data.

https://doi.org/10.1371/journal.pbio.3003897.g007

Animals

All animal experiments in this study were approved by the Animal Care and Use Committee of the National Institute of Neuroscience, National Center of Neurology and Psychiatry (Tokyo, Japan; project 2019028R2), and were conducted in accordance with relevant Japanese national regulations and guidelines, including the Act on Welfare and Management of Animals, the Standards Relating to the Care and Management of Laboratory Animals and Relief of Pain, and the Guidelines for Proper Conduct of Animal Experiments issued by the Science Council of Japan. Mice were housed under specific pathogen-free conditions with food and water ad libitum. ICR pups were obtained from SLC (Japan). Atoh1^AtGFP/+ mice (The Jackson Laboratory: B6.129S-Atoh1tm4.1Hzo/J, Stock No. 013593) and Ki67^KiRFP/+ mice (The Jackson Laboratory: Mki67tm1.1Cle/J, Stock No. 029802) were obtained from The Jackson Laboratory.

Long-read cDNA sequencing sample preparation

GCP isolation was performed as previously described [55,56]. For the long-read sequencing samples, isolated P7 GCPs were plated and cultured for 48 h in the presence of 200 nM smoothened agonist. Following the culture period, total RNA was extracted using the Direct-zol RNA MiniPrep Plus kit (Zymo Research). For Nanopore sequencing, cDNA libraries were prepared from 50 ng of total RNA using the SMART-Seq v4 Ultra Low Input RNA Kit for Sequencing (Takara Bio USA) according to the manufacturer’s protocol, with 16 PCR cycles for cDNA amplification. Custom primers were utilized for cDNA synthesis:

1. 3′ SMART-Seq CDS Primer II A: AAGCAGTGGTATCAACGCAGAGTACTTTTTTTTTTTTTTTTTTTTTTTTTTTTTTVN (HPLC-purified)
2. PCR Primer II A: AAGCAGTGGTATCAACGCAGAGT (HPLC-purified)

The synthesized cDNA was then sequenced on a Nanopore PromethION platform.

Long-read cDNA sequencing data analysis

Raw FASTQ files obtained from Nanopore PromethION sequencing were processed using the IsoQuant [57] pipeline for alignment, transcript quantification, and novel isoform identification. Alignment was performed within the IsoQuant pipeline using Minimap [58]. The quality of the resulting sequencing data was assessed prior to analysis. The reads exhibited a high average quality score of approximately 30 (S1A Fig), and the read length distribution peaked around 1000 base pairs (S1B Fig), confirming the high quality and robust coverage of the dataset. The resulting GTF files, containing quantified and identified transcripts, were then subjected to curation using Sqanti3 [59].

RT-PCR

RNA was extracted from isolated GCP using Direct-zolTM RNA MiniPrep w/TriReagent (ZYMO RESEARCH). The ReverTra Ace qPCR RT Kit (TOYOBO) was used to generate cDNA. The primer sequences used are as following:

1. Meis1-FL: CACAAAAAGCGTGGCATCTT and TGATGCCCATGTGCTGCTGA,
2. Meis1-HdL: ATGGCGCAAAGGTACGACGA and TCAGAAGGGTAAGGGTGCTTGC.

Quantitative PCR (qPCR)

Isolated GCPs were suspended in cold DPBS containing 1% BSA. Cell sorting was performed in FACSAria Fusion. Gating of GFP and RFP negative populations was performed using GCPs from WT mice. Fluorescence compensation of RFP and GFP was performed using GCP of Ki67^KiRFP/+ or Atoh1^AtGFP/+ mice. Cells were sorted into cold DPBS containing 1% BSA and centrifuged at 800g for 3 min.

cDNA was generated with ReverTra Ace qPCR RT Kit (TOYOBO). Relative gene expression was compared with the geometric mean of 18S rRNA. The primer sequences used are as following:

1. 18S rRNA: CCCGAAGCGTTTACTTTGAA and CCCTCTTAATCATGGCCTCA,
2. Meis1-FL: ACAGCAGTGAGCAAGGTG and CAGAAGGGTAAGGGTGTGTT,
3. Meis1-HdL: TGAGCAAGCACCCTTACCCTTCTGAA and GACTGCTCGGTTGGACTG,
4. Atoh1: AGCTTCCTCTGGGGGTTACT and TTCTGTGCCATCATCGCTGT.

Plasmids

The expression vectors of Atoh1, Meis1 1–130aa and Meis1-FL were cloned from cDNAs from P7 C57/BL6 mouse cerebellum. PCR amplification was performed using the following primers: Atoh1-Fw: 5′-AATTGGATCCATGTCCCGCCTGCTGCATGC-3′ Atoh1-Rv: 5′-AATTGGATCCTTACATGTAGTGCCACTGCC-3′ Meis1 1–130aa-Rv: 5′-AATTGGATCCTTATGCGCGAATCTGTTTGGCGA-3′. For Meis1-FL, we used the forward primer described previously [16] and a specific reverse primer. Meis1-HdL was cloned using the same forward primer as Meis1-FL and the specific reverse primer: 5′-AATTGGATCCGCAAGCACCCTTACCCTTCTGA-3′. Cloned fragments were inserted into a pCAGGS vector or pEF-BOS-GST vectors (gifts from K. Kaibuchi, Fujita Health University, Nagoya, Japan) and cloned sequences were confirmed by sequencing. pCAG-H3.1-EGFP vector was a gift from Dr. N. Masuyama. HA-Ub vector was kindly provided by Dr. S. Wakatsuki (NCNP, Tokyo, Japan). Flag-HA-COPS6 was a gift from Wade Harper (Addgene plasmid #22542; http://n2t.net/addgene:22542; RRID:Addgene_22542) [60]. To generate the NLS-Meis1-HdL vector, a nucleotide sequence encoding a 3x SV40 nuclear localization signal (NLS) with spacers (5′-ATGGGATCAGATCCAAAAAAGAAGAGAAAGGTAGACCCGAAGAAAAAGCGTAAGGTCGATCCCAAAAAGAAACGGAAAGTGGGTAGCGGC-3′) was inserted in-frame immediately upstream of the start codon of the Meis1-HdL sequence. Res-Meis1-HdL and mutated forms of Atoh1 S328A and S328D were generated following the mutagenesis protocol of PrimeSTAR Max (Takara). The primers we used to create Res-Meis1-HdL were GAGCAGGCCCCTTTACCCTTCTGAAGAACA and TAAAGGGGCCTGCTCACTGCTGTTATCCCC. shRNAs were generated by inserting the double-stranded oligonucleotides into a mU6 pro vector. The targeting sequence of each shRNAs was designed by siDirect 2.0 [61] and the target sequences are indicated below.

1. sh-Meis1-HdL: 5′- CAGTGAGCAAGCACCCTTA-3′,
2. sh-Meis1-all: 5′- GCACAAGATACAGGACTTACC -3′ [16],
3. sh-Cul3: 5′- AGCTGCTATAGTGCGAATAAT -3′,
4. sh-Cops6-1: 5′-ACCAAGGAGGAGCAGTTTAAA -3′,
5. sh-Cops6-2: 5′-TTGAGTCTGTCATCGATATAA -3′,
6. sh-Scramble: 5′-TACGCGCATAAGATTAGGG-3′

Generation of the MEIS1-HdL-specific antibody

The custom polyclonal antibody against MEIS1-HdL was generated in rabbits by Eurofins Genomics (Tokyo, Japan). A synthetic peptide corresponding to the frameshift-derived unique C-terminus of MEIS1-HdL (NH2-C-(Ahx)-EQAPLPF-COOH) was conjugated to a carrier protein and used for immunization. To obtain a highly specific antibody, the crude antiserum was purified utilizing a two-step affinity chromatography procedure. First, the antiserum was passed through a TOYOPEARL AF-Tresyl-650 column (Tosoh) coupled with recombinant GST protein to deplete non-specific and anti-GST antibodies (negative selection). The flow-through fraction was subsequently applied to a TOYOPEARL AF-Tresyl-650 column coupled with recombinant GST-MEIS1-HdL protein (positive selection). After extensive washing, the specifically bound antibodies were eluted with 100 mM Glycine-HCl buffer (pH 2.5) and immediately neutralized to pH 7.0–7.5 using 1.0 M Tris-HCl (pH 9.0).

Cell culture, transfection, and drug treatment

N2a cells were obtained from ATCC (Manassas, VA, USA) and cultured in Dulbecco’s modified Eagle medium (DMEM) containing 10% fetal bovine serum (FBS) and 100 U/ml penicillin–streptomycin. Transfection of N2a cells was performed using transfection reagent (Bio-Rad, Hercules, CA, USA).

Cycloheximide (100 μM) was diluted in dimethyl sulfoxide (DMSO) and added to N2a cells, as indicated. MG132 diluted in DMSO was administered to the cultured cells indicated, after which cells were harvested at the indicated times.

Immunohistochemistry

Tissues were fixed with 4% paraformaldehyde (PFA) in PBS and cryoprotected with 30% sucrose in PBS. After tissues were embedded in optimal cutting temperature (OCT) compound, cryosections were made at 16 μm. Sections were incubated in blocking buffer containing 1% BSA and 0.1% Triton X-100 in PBS at room temperature (RT between 20 and 25 °C) for 1 h and subsequently immunolabeled using the following primary antibodies in blocking buffer at 4 °C overnight. Specimens were subsequently rinsed with PBS and incubated with secondary antibodies conjugated with Alexa Fluor 488, Alexa Fluor 568, Alexa Fluor 594, or Alexa Fluor 647 (1:400; Invitrogen) and DAPI (1:3,000; Invitrogen) in blocking buffer containing 1% BSA and 0.2% Triton X-100 in PBS at RT for 2 h. Fluorescence images were acquired using a Zeiss LSM 780 confocal microscope system (Carl Zeiss) and ZEN 2009 software. Antibodies used in this study were as follows: anti-ATOH1 antibody (homemade, rabbit), anti-KI67 antibody (eBioscience, 14-5698-82, rat), anti-p27 antibody (MBL, 554, rabbit), anti-HA(C29F4) (CST, 3724S, rabbit), anti-GFP antibody (kind gift from Dr. A. Imura, Foundation for Biomedical Research and Innovation, Kobe, Japan rat), anti-GFP (Aves Labs, GFP-1010, chicken). Fluorescence intensity quantification was performed using ImageJ software (NIH). The nuclear region of interest (ROI) for each cell was manually defined based on the DAPI staining channel. The mean fluorescence intensities of the target proteins (ATOH1, KI67) and DAPI were measured within each ROI. To correct for variations in nuclear volume, chromatin density, and staining efficiency, the intensity of the target protein was normalized to the DAPI intensity of the corresponding nucleus. For in vivo electroporation experiments, the normalized intensity of electroporated cells (GFP-positive) was compared to that of neighboring non-electroporated cells (GFP-negative) within the same field of view. To clearly label the electroporated cells, an H3.1-GFP expression plasmid was co-transfected in all in vivo electroporation experiments. To quantify the proportion of cells expressing specific differentiation or proliferation markers (such as KI67 or p27) within the electroporated population, the number of cells double-positive for nuclear GFP and the target protein was counted using ImageJ software. This value was then divided by the total number of GFP-positive electroporated cells within the analyzed cerebellar layers to calculate the percentage of positive cells. Multiple sections from at least three independent animals were analyzed for each condition.

In vivo electroporation

In vivo electroporation in neonatal mice has been described previously [16]. Expression plasmids were diluted to 1 μg/μl, shRNAs to 2 μg/μl, and pCAG-H2BGFP to 0.5 μg/μl in Milli-Q water (Millipore). Fast Green was added to visualize the plasmid solutions, which were injected into P8 ICR cerebella over the skull. Mice received electric pulses (80 V for 50 ms, with 150 ms intervals) using forceps-type electrodes (NEPA Gene, Chiba, Japan). The pups were kept warm at 37 °C during recuperation and returned to the litter after fully recovering. Pups were fixed with 4% PFA for 3 days after electroporation.

Biochemical analyses

For immunoblotting, cells were harvested and lysed in Cell Lysis Buffer (22352-04, Nacalai Tesque) containing protease inhibitors (11836153,001 Roche). Proteins were transferred to polyvinylidene fluoride membranes, which were incubated with primary antibodies at 4 °C overnight. After the membranes were incubated with secondary antibodies at room temperature for 2 h, horseradish peroxidase substrate (Millipore) was applied, and immuno-signals were detected using a LAS4000 system (Fujifilm) and FUSION SOLO S (Vilber). Antibodies used in this study were as follows: anti-GAPDH (CST, #2118, rabbit), anti-GST (CST, #2622, rabbit), anti-HA (CST, #3724, rabbit), anti- ATOH1 (homemade, rabbit), anti-MEIS1-HdL antibody (homemade, rabbit), anti-MEIS1 (abcam, ab19867, rabbit), anti-MEIS1(MyBioSource, MBS9403588, rabbit), anti-CUL3 (CST, #2759, rabbit), anti-COPS2 (Bethyl Laboratories, A300-028A, rabbit), anti-COPS5 (Bethyl Laboratories, A300-014A, rabbit), anti-SRC-3 (NCOA3; CST, #2126, rabbit), anti-PP2A/C subunit (CST, #2038, rabbit), anti-NRF2 (Novusbio, NBP1-32822, rabbit), anti-Lamin A/C (CST, #4777, mouse), anti-βIII-tubulin (Millipore, MAB1637, mouse), and anti-ATOH1 p-S328 (homemade, rabbit).

For Immunoprecipitation (IP), cells were lysed in Pierce IP Lysis Buffer (Thermo Scientific) containing protease inhibitors. Lysates were incubated with antibodies and captured using the Dynabeads Protein G Immunoprecipitation Kit (Invitrogen) according to the manufacturer’s instructions. To prevent the masking of target protein signals by the heavy and light chains of the immunoprecipitating antibodies during subsequent immunoblotting, TrueBlot ULTRA: Anti-Rabbit IgG HRP (Rockland Immunochemicals, 18-8816-31) or Clean-Blot IP Detection Reagent (HRP) (Thermo Scientific, 21230) was utilized for detection in specific experiments.

For the GST pull-down assay, GST-tagged vectors expressing Atoh1, Atoh1 mutants, Meis1-FL, and Meis1-HdL were transfected into N2a cells. After 48 h, cells were lysed in Cell Lysis Buffer (Nacalai Tesque), and supernatants were incubated with glutathione-conjugated Sepharose beads (GE Healthcare) at 4 °C overnight. Beads were washed twice and resuspended in sample buffer. For protein purification (Fig 6), washed Sepharose beads were incubated with TED buffer (50 mM Tris-HCl pH 8.0, 10 mM reduced glutathione, 1 mM DTT, 1 mM EDTA) at 4 °C for 30 min to elute the GST-fusion proteins. For the sequential precipitation assay (HA-IP followed by GST-PD), lysates were first incubated with anti-HA antibody. The antibody-antigen complexes were precipitated using Protein G Sepharose beads. The immunoprecipitated protein complexes were eluted with HA peptide (Sigma) and subsequently incubated with glutathione-Sepharose beads to pull down GST-tagged interacting partners. All immunoblotting and immunoprecipitation experiments were repeated at least three times independently, and representative images are shown.

Quantification of ubiquitination

For the quantification of ATOH1 polyubiquitination, the signal intensity of the high-molecular-weight smear appearing above the unmodified ATOH1 band was measured using ImageJ software. The background signal was subtracted to determine the final intensity values.

LC–MS/MS analysis

Phospho-MS and LC–MS/MS analysis was performed as previously described [62]. The proteomic screening was performed on a single biological replicate (n = 1), and candidate interactions were subsequently validated by independent co-immunoprecipitation and immunoblotting assays. The complete list of identified proteins is provided in Supplementary Data (S3 Data). Candidate molecules involved in the ubiquitin-proteasome system were selected based on PANTHER GO analysis.

Statistical analyses

Pairwise comparisons between the means of different groups were performed using a Student t test (two-tailed, unpaired). To evaluate statistical significance across multiple groups or variables, such as comparing overall dose-dependent effects between different MEIS1 isoforms, a two-way analysis of variance (ANOVA) was employed. The difference between data sets was considered statistically significant at *p < 0.05, **p < 0.01, or ***p < 0.001. The data are reported as the mean ± SEM.