ASCL1 promotes nuclear shrinkage in transdifferentiation by suppressing NUP37

Summary

Despite the central role of the nucleus in the cell, it remains unclear what determines nuclear size. Previous studies suggest a correlation between nuclear and cell size. However, many neurons have massive axon arbors, while maintaining a typical nuclear size. Here, we show a marked reduction of nuclear, but not cell, size during the direct conversion of human fibroblasts to induced neurons by ASCL1, miR124-9-9^∗, and p53 short hairpin RNA (shRNA) (AMp). Similar nuclear shrinkage was observed in the maturation of induced pluripotent stem cell-derived human cortical neurons. ASCL1 suppressed the transcription of the nucleoporin NUP37 by direct binding to its promoter. NUP37 knockdown enhanced AMp-mediated transdifferentiation and nuclear shrinkage, while NUP37 overexpression achieved the opposite. It appears that ASCL1 promotes nuclear shrinkage by suppressing NUP37 and reducing the nuclear pore complex (NPC), which gates nuclear transport. The study suggests a critical role of NPC in controlling nuclear size to match cell state.

Graphical abstract

Highlights

• •ASCL1 reduces nuclear size in the direct conversion of human fibroblasts to neurons
• •ASCL1 suppresses expression of the nucleoporin NUP37 by binding to its promoter
• •ASCL1 decreases the number and density of nuclear pore complexes in the conversion
• •NUP37 knockdown enhances ASCL1-mediated conversion and nuclear shrinkage

It is unclear what determines nuclear size despite its central role in the cell. Feng and colleagues have observed nuclear size reduction in the direct conversion of human fibroblasts to neurons, as ASCL1 decreases the nuclear pore complex and its critical component NUP37. Knockdown of the nucleoporin NUP37 enhances ASCL1-mediated transdifferentiation and nuclear shrinkage, while NUP37 overexpression achieves the opposite.

Introduction

Since the Scottish botanist Robert Brown named the cell nucleus in 1831, we still do not have a clear answer on how nuclear size is regulated. The prevailing thought for more than a century has been that cells maintain a relatively constant nucleocytoplasmic ratio (Cantwell and Nurse, 2019a, 2019b; Webster et al., 2009; Boveri, 1905). The diversity of cells on earth makes it challenging to formulate a general theory on nuclear size control. Postmitotic neurons provide some of the most extreme counterexamples to the idea of a constant nucleocytoplasmic ratio. Careful 3D reconstruction shows that a single nigral dopaminergic neuron in the rat brain has an axon arbor with an average total length of 45 cm (Matsuda et al., 2009), while a single cholinergic neuron in the mouse basal forebrain has a total axon length of 50 cm (Wu et al., 2014). Motor and sensory neurons in the human sciatic nerve have axons longer than 1 m (Standring, 2015). All these neurons have a soma with a typical size, generally several tens of micrometers in diameter. The nucleocytoplasmic ratio of these neurons (<0.001) is much smaller than that of many mitotic cells (generally >0.2) (Swanson et al., 1991; Sebastian et al., 2021).

The direct conversion of human fibroblasts to induced neurons (iNs) provides an excellent opportunity to study nuclear size control, as epigenetic reprogramming of the same genome generates a completely different cell type, without the potentially confounding influence of cell division (Masserdotti et al., 2016). The pioneer factor ASCL1 (A) (Wapinski et al., 2013) is critical in driving cell cycle exit within 1–2 days (Jiang et al., 2015) and rapid transdifferentiation in 14 days (Zhu et al., 2023; Pang et al., 2011). The microRNA cluster miR124-9/9^∗ (M) (Yoo et al., 2011) significantly enhances ASCL1-mediated reprogramming (Zhu et al., 2023) by repressing REST, switching key chromatin remodelers, and suppressing the RNA splicing regulators PTB and nPTB (Lu and Yoo, 2018). Our previous studies have shown that p53 short hairpin RNA (shRNA) (p) significantly enhances ASCL1-mediated neuronal transdifferentiation by inducing the TET family of DNA hydroxymethylases (Jiang et al., 2015; Xu et al., 2016; Li et al., 2019; Zhu et al., 2023). Using the combination of AMp (uppercase for overexpression [OE] and lowercase for knockdown), we can efficiently convert human fibroblasts to iNs (Jiang et al., 2015; Xu et al., 2016; Li et al., 2019; Zhu et al., 2023).

The transdifferentiation of human fibroblasts to neurons entails marked changes in cell morphology to support new functions, such as permanent cell cycle exit, which should require substantial nuclear remodeling. However, much is unknown about these cellular changes and their mechanisms in the generation of neurons from mitotic cells, either in development or during transdifferentiation. The nuclear envelope plays critical roles in the regulation of global gene expression patterns (Stancheva and Schirmer, 2014), which is drastically changed during neuronal transdifferentiation in a highly coordinated fashion (Wapinski et al., 2013, 2017; Treutlein et al., 2016). Embedded in the nuclear envelope, the ∼110-MDa nuclear pore complex (NPC), composed of approximately 34 different nucleoporins, gates the transport of macromolecules between the nucleus and the cytoplasm (Petrovic et al., 2022). We found that the expression of NUP37, a critical NPC component that facilitates the proliferation of cancer cells (Luo et al., 2017; Huang et al., 2020; Xiong et al., 2023), was significantly suppressed by ASCL1 in the transdifferentiation, because ASCL1 directly bound to the promoter of NUP37 and recruited suppressive histones and removed permissive histones. NUP37 knockdown markedly enhanced ASCL1-mediated transdifferentiation and nuclear shrinkage, while NUP37 OE achieved the opposite. The neuronal transdifferentiation induced significant reductions in the number and density of NPCs. The reductions were enhanced by NUP37 knockdown.

Results

Nuclear size reduces when human fibroblasts are converted to neurons

Human fetal lung fibroblasts, MRC5 cells, were reprogrammed to iNs with doxycycline (DOX)-inducible lentiviruses expressing AMp (Jiang et al., 2015; Zhu et al., 2023). The virus-transduced fibroblasts were arrested at the G1/S checkpoint by serum withdrawal (−fetal bovine serum [FBS]), which significantly enhances the transdifferentiation (Jiang et al., 2015). The synchronized cells were treated with DOX to initiate transdifferentiation in neural induction medium at day 0 (Figure 1A). We have shown that all fibroblasts are infected by all reprogramming factors at multiplicity of infection (MOI) of 10 (Jiang et al., 2015; Zhu et al., 2023). ASCL1 drives all cells out of cell cycle in 1–2 days (Jiang et al., 2015; Zhu et al., 2023). We found a progressive and marked reduction of nuclear volume (from 1,329 ± 35.3 μm³ to 398 ± 11.01 μm³) (Figure 1B) and nuclear area (from 235 ± 10.4 μm² to 95 ± 3.7 μm²) from day −2 to day 14, as the cells were reprogrammed from fibroblasts to neurons (Figures 1C and 1D). Since the correlation between nuclear volume and nuclear area was very high (r = 0.99) (Figure S1) and substantial computational resources and time were needed to perform 3D reconstructions for nuclear volume measurements, we quantified nuclear area to represent nuclear size in all subsequent experiments. Consistent nuclear shrinkage was observed when AG22056 human newborn foreskin fibroblasts (207 ± 9.9 μm² to 125 ± 3.5 μm²) or adult human skin fibroblasts GM09918 (221 ± 7.6 μm² to 99 ± 4.3 μm²) were reprogrammed to iNs with the same method (Figure 1E). We measured the average cell area in binarized images by dividing the total area occupied the cells by the number of cells (Figure S2) and found no significant change in cell area from the starting fibroblasts at day −2 to the resulting TUJ1⁺ neuronal cells or MAP2⁺ mature neurons at day 14 (Figure 1F). The lack of change in cell area focused our attention on nuclear shrinkage in neuronal conversion. When we measured nuclear area in MAP2⁺ cortical neurons differentiated from normal human induced pluripotent stem cells (iPSCs) (Jiang et al., 2025), we found significant nuclear shrinkage as the neurons matured from days 30 to 40 to 80 (75 ± 1.8 μm², 68 ± 2.5 μm², and 35 ± 0.7 μm², respectively) (Figure 1G). A summary of the shrinkage of the nucleus across different transdifferentiation and differentiation experiments is shown in Figure 1H. All raw data and statistical analyses are in Table S1.

Figure 1.

Reduction of nuclear size during the direct conversion of human fibroblasts to neurons

(A) Schematic for the transdifferentiation of human fibroblasts to iNs by lentiviruses expressing ASCL1, miR124-9-9^∗-BclxL, and p53 shRNA (AMp, uppercase for overexpression and lowercase for knockdown). –FBS, serum withdrawal to synchronize cell cycle at the G1/S checkpoint. Scale bar, 100 μm.

(B) Phase contrast images of MRC5 cells under conversion at the indicated time points. Scale bar, 100 μm. Insets, super-resolution images of DAPI-stained nuclei. Scale bar, 10 μm.

(C) Nuclear volume quantification for MRC5 cells throughout reprogramming. n = 50 frames from 3 independent experiments for each time point, unpaired t test vs. day −2. ^∗p < 0.01.

(D) Quantification of nuclear area for MRC5 cells throughout reprogramming. n = 50 frames from 3 independent experiments for each time point, unpaired t test vs. day −2, ^∗p < 0.01.

(E) Quantification of nuclear area at the indicated time points as MRC5, AG22056 newborn foreskin fibroblasts, or GM09918 (78 years) skin fibroblasts were being converted to iNs. n = 50 frames from 3 independent experiments for each time point, unpaired t test vs. day −2, ^∗p < 0.01.

(F) The average area of MRC5, AG22056, or GM09918 cells as fibroblasts at day −2 (Fib) or TUJ1⁺ or MAP2⁺ iNs. ns, no significance. n = 50 frames from three independent experiments.

(G) iPSC-derived cortical neurons were co-stained for MAP2 and DAPI at days 30, 40, and 80 of differentiation. Scale bar, 50 μm. Inset, super-resolution images of neuronal nuclei; scale bar, 10 μm.

(H) Quantification of nuclear area of the indicated samples. ^∗p < 0.01, vs. the preceding bar (or D30 for iPSC-derived neurons), n = 50 frames from 3 independent experiments.

ASCL1 suppresses the expression of NUP37 during the transdifferentiation

We plotted cell conversion trajectories based on single-cell RNA sequencing (scRNA-seq) data from Ascl1-mediated direct conversion of mouse embryonic fibroblasts (MEFs) to iNs (Treutlein et al., 2016). Force-directed graph was computed on uniform manifold approximation and projection (UMAP) positions of the cells and visualized by scanpy (Wolf et al., 2018). A branched principal tree represented as a dendrogram was generated on multiscale diffusion components calculated by Palantir (Setty et al., 2019; Faure et al., 2023) with cell types annotated based on the original data (Treutlein et al., 2016) (Figure 2A). MEFs undergoing conversion could develop along a myocyte or neuron trajectory, as shown by two arms stemming from cells transduced by Ascl1. The neuron arm expressed many neuron-specific genes, such as Syt1 (synaptotagmin 1) and Rbfox3 (NeuN), confirming the identity of converted cells. Expression of the nucleoporin Nup37 showed an opposite pattern, decreasing along the trajectory from MEF to neuron, suggesting that it could be a target for further investigation. As NUP210 is involved in neuronal differentiation (D'Angelo et al., 2012), we also included this nucleoporin as a control in our analyses. Quantitative reverse-transcription PCR (RT-qPCR) (Table S2 for primers) showed that expression of both NUP37 and NUP210 decreased significantly between days −2 and 14 (Figure 2B).

Figure 2.

ASCL1 suppressed NUP37 transcription by binding to its promoter

(A) Using single-cell RNA-seq data from Ascl1-mediated direct conversion of MEF cells to iNs (Treutlein et al., 2016), dendrogram and Force Atlas plots showed cell types, developing trajectory, and expression levels of Syt1, Rbfox3, Nup37, and Nup210.

(B) NUP37 and NUP210 expression levels at the indicated time points in the conversion of human fibroblasts to iNs. ^∗p < 0.01, vs. day −2, unpaired t test, n = 6 replicates from three independent experiments for each time point.

(C) ATAC-seq peaks at the promoter of NUP37 from datasets that have either ASCL1 knockout (KO) or OE. Studies (1) Aydin et al., 2019, (2) Woods et al., 2022, (3) Paun et al., 2023, (4) Parkinson et al., 2022, and (5) Park et al., 2017.

(D) Quantification of peak height for the ATAC-seq datasets, expressed as fold-change.

(E) Bulk RNA-seq comparison of NUP37 expression levels expressed in fold-change from the same studies as in (D).

(F) Alignment of ASCL1 ChIP-seq datasets surrounding the promoter of NUP37 to show the location of ChIP-qPCR primers (blue box).

(G) ChIP-qPCR measurement of ASCL1 binding on NUP37 at the indicated time points of the transdifferentiation. ^∗p < 0.05, vs. day −2, unpaired t test, n = 6 replicates from three independent experiments for each time point.

(H) Alignment of ASCL1 ChIP-seq datasets surrounding the gene body of NUP210 to show the location of ChIP-qPCR primers (green box).

(I) ChIP-qPCR measurement of ASCL1 binding on NUP210 at the indicated time points of the transdifferentiation. ^∗p < 0.05, vs. day −2, unpaired t test, n = 6 replicates from three independent experiments for each time point.

(J) Luciferase reporter assay for the ASCL1-binding site in the NUP37 promoter in the absence or presence of ASCL1 lentivirus (LV) in HEK293 and MRC5 cells. pRL-SV40, Renilla luciferase plasmid as a control to normalize transfection efficiency; pGL3-NUP37, firefly luciferase reporter driven by the ASCL1-binding site on the NUP37 promoter. ^∗ and #, p < 0.0001, one-way ANOVA, n = 8 samples from 4 independent experiments.

(K) Alignment of various histone modification peaks (permissive or repressive, marked by the green or red bar on the left, respectively) surrounding the NUP37 promoter (the same blue box as in F) from other studies listed in Table S4.

(L and M) NUP37 qPCR of H3K27Me3 ChIP (repressive) (L) or H3K4Me3 ChIP (permissive) (M) at the indicated time points. ^∗p < 0.01, vs. day −2, unpaired t test, n = 8 replicates from 4 independent experiments for each time point.

To examine how ASCL1 affects the promoter of NUP37 and NUP210, we analyzed ATAC-seq datasets with ASCL1 OE or knockout (KO) in previous studies 1 through 5 (Woods et al., 2022; Păun et al., 2023; Aydin et al., 2019; Parkinson et al., 2022; Park et al., 2017) (Table S3) (Figure 2C). Quantification of chromatin accessibility peak height showed that accessibility to the NUP37 promoter was reduced with ASCL1 OE and increased with ASCL1 KO (Figure 2D). Analysis of RNA-seq datasets from the same five studies showed that ASCL1 OE indeed suppressed NUP37 transcription, while ASCL1 KO increased it (Figure 2E). In contrast, we found no consistent changes in accessibility to the NUP210 promoter or NUP210 expression levels with ASCL1 OE or KO in the same datasets (Figure S3).

Existing human ASCL1 chromatin immunoprecipitation sequencing (ChIP-seq) datasets (Păun et al., 2023; Park et al., 2017; Woods et al., 2022; Borromeo et al., 2016) showed a potential ASCL1-binding region in the promoter of NUP37 (blue dashed box) (Figure 2F) and major peaks in the intron between exons 3 and 4 of NUP210 (green dashed box) (Figure 2H), but no significant ASCL1 peak in the promoter of NUP210. Since this was the most likely potential binding site of ASCL1 in the NUP210 gene, we decided to focus our ChIP-qPCR experiments on this region, rather than the promoter region of NUP210. ASCL1 ChIP-qPCR with primers flanking the blue or green box showed a progressive and marked increase of ASCL1 binding on the NUP37 promoter from days −2 to 14 (Figure 2G). In contrast, no consistent change was observed in ASCL1 binding on the NUP210 gene as fibroblasts were reprogrammed to neurons (Figure 2I). We cloned the NUP37 promoter region into the pGL3 luciferase reporter construct and transfected the reporter in HEK293 or MRC5 cells. Marked expression of firefly luciferase was produced by the NUP37 promoter. The activity was strongly suppressed by the OE of ASCL1 (Figure 2J).

We also analyzed the binding of modified histones surrounding the NUP37 promoter (blue box) in human postmortem brain tissue and cultured neurons from other studies listed in Table S4 and highlighted permissive (green bar) or repressive histone marks (red bar) (Figure 2K). ChIP-qPCR experiments showed a significant and progressive increase in the binding of the repressive histone mark H3K27Me3 on the NUP37 promoter (Figure 2L) and a corresponding significant decrease in the binding of the permissive histone mark H3K4Me3 (Figure 2M). These data suggest that ASCL1 binds to the NUP37 promoter and recruits appropriately modified histones to suppress the transcription of NUP37.

Transdifferentiation-induced reduction in NPC is enhanced by NUP37 knockdown

As NUP37 is an important NPC subunit (Luo et al., 2017; Huang et al., 2020; Xiong et al., 2023), we examined the NPC by staining the cells with the mAb414 antibody, which recognizes the FG-repeat regions on several core constituents of NPC (Bayliss et al., 2000). MRC5 cells transduced without (at day −2) or with AMp or AMpu (u: NUP37 shRNA) for different days were co-stained with mAb414, DAPI, and MAP2 to quantify the number and density of NPC in MAP2^- cells and MAP2⁺ neurons (Figure 3A). AMp-mediated transdifferentiation induced rapid and significant reductions in the number of NPC per nucleus (Figure 3B) and the density of NPC per unit nuclear area (Figure 3C). The reductions were much more pronounced in MAP2⁺ neurons than MAP2⁻ cells (Figure 3). NUP37 knockdown on top of AMp significantly decreased NPC number in both MAP2⁻ cells and MAP2⁺ neurons (Figure 3B) and reduced NPC density in MAP2⁺ neurons (Figure 3C). The reduction in NPC density (Figure 3C) was less pronounced than the reduction in NPC number (Figure 3B) due to the concomitant decrease of nuclear size.

Figure 3.

Reduced number and density of nuclear pore complexes in the fibroblast-to-neuron conversion

(A) MRC5 cells transduced without or with AMp or AMpu were co-stained for DAPI, mAb414 (NPC), and MAP2 at the indicated days. Scale bars: 50 μM and 10 μM for inset. Magenta (MAP2⁻ cells) or green (MAP2⁺ neurons) boxes were enlarged in color-matched panels. Green dotted circles, MAP2⁺ neurons.

(B) The number of mAb414⁺ puncta (NPC) per nucleus in MAP2⁻ cells and MAP2⁺ neurons was quantified at different time points for the AMp or AMpu condition. #p < 0.0001, one-way ANOVA, day −2 vs. all other days; ^∗p < 0.0001, unpaired t test between the indicated; ˆp < 0.01 unpaired t test between AMp and AMpu on the same day for MAP2⁻ cells; $p < 0.01 unpaired t test between AMp and AMpu on the same day for MAP2⁺ neurons; n = 50 nuclei from 3 independent experiments.

(C) NPC density (mAb414⁺ puncta/nuclear size) in MAP2⁻ cells and MAP2⁺ neurons was quantified at different time points for the AMp or AMpu condition. #p < 0.0001, one-way ANOVA, day −2 vs. all other days; ^∗p < 0.05, unpaired t test between the indicated; $p < 0.01 unpaired t test between AMp and AMpu on the same day for MAP2⁺ neurons; n = 50 nuclei from 3 independent experiments.

NUP37 knockdown enhances neuronal conversion and nuclear shrinkage

Since ASCL1 suppressed NUP37 expression during the conversion, we examined whether decreasing NUP37 expression affected the transdifferentiation. Indeed, NUP37 shRNA (u) significantly enhanced the effect of AMp in the generation of TUJ1⁺ neurons and MAP2⁺ mature neurons, while NUP37 OE (U) greatly attenuated the conversion, as quantified by reprogramming efficiency and yield (Figures 4A–4F). NUP37 knockdown decreased the total number of cells, while NUP37 OE increased it (Figure 4G). Live/Dead staining showed that NUP37 knockdown increased cell death, while NUP37 OE decreased cell death (Figure S4). Furthermore, NUP37 knockdown reduced nuclear area, while NUP37 OE increased it (Figure 4H). Starting from day 3, there were more MAP2⁺ neurons in the AMpu condition, compared to the AMp condition (Figures 4I–4R), although AMpu induced higher cell loss from day 7 onward (Figure 4S) (Figure S5). NUP37 knockdown increased the expression of mature neuronal markers, such as MAP2, SYT1, RBFOX3, SNAP25, and GRIA1 at day 14 (Figure 4T), suggesting an enhancing effect on neuronal maturation.

Figure 4.

NUP37 knockdown significantly enhanced AMp-mediated transdifferentiation and nuclear shrinkage

(A) Western blot of NUP37 in MRC5 cells transduced without (−) or with the indicated reprogramming factors.

(B–D) MRC5 human fibroblasts reprogrammed with ASCL1, MIR124-9-9^∗-BclxL, and p53 shRNA (AMp) (B), AMp and NUP37 shRNA (AMpu) (C), or AMp and NUP37 overexpression (AMpU) (D) were co-stained as indicated on day 14. Scale bar, 100 μm.

(E–G) Reprogramming efficiency (E) as measured by the percentages of TUJ1⁺ or MAP2⁺ cells among all DAPI⁺ cells, reprogramming yield of MAP2⁺ cells per frame (F), and the number of DAPI⁺ cells per frame (G) at day 14. # and ^∗, p < 0.05, n = 15 (3 experiments, 5 frames each), vs. AMp for the indicated cell type, unpaired t test.

(H) Nuclear area of MAP2⁺ neurons for each condition. ^∗p < 0.001, n = 50 frames from 3 independent experiments, vs. AMp, unpaired t test.

(I–P) MRC5 cells reprogrammed with AMp (I–L) or AMpu (M–P) were co-stained as indicated at different time points. Scale bar, 100 μm.

(Q‒S) (Q) Reprogramming efficiency of MAP2⁺-generated neurons per DAPI⁺ nuclei. (R) Yield of MAP2⁺ neurons. (S) Number of DAPI⁺ cells per frame. ^∗p < 0.01, n = 15 (3 experiments, 5 frames each), vs. AMp at the same time point, unpaired t test.

(T) RT-qPCR measurement of mature neuronal markers in AMp- or AMpu-induced neurons at D14. ^∗p < 0.05, n = 6 (3 experiments, duplicate for each), vs. AMp, unpaired t test.

NUP210 knockdown on top of AMp (AMp210) slightly reduced the percentage of TUJ1⁺ neurons without affecting the percentage of MAP2⁺ neurons. It decreased the yield of MAP2⁺ neurons and the total number of cells. Consistent with the lack of a transdifferentiation-enhancing effect, NUP210 knockdown did not appreciably change nuclear area (Figure S6). NUP37 is a subunit of the outer ring Y complex, which anchors NPCs within the nuclear membrane (Bilokapic and Schwartz, 2012). Alterations to the outer ring components, including NUP37, changes NPC number and spatial organization (Raices and D'Angelo, 2022). In contrast, NUP210 is a transmembrane nucleoporin dispensable for NPC assembly in many cell types with its loss often buffered by other nucleoporins (Stavru et al., 2006).

The ability of NUP37 knockdown to enhance AMp-mediated neuronal conversion and reduce nuclear area was confirmed in AG22056 human newborn foreskin fibroblasts and GM09918 human adult (78 years) skin fibroblasts (Figure S7). Most of the iNs converted by AMp or AMpu were GABAergic neurons expressing parvalbumin; a very small percentage expressed somatostatin or serotonin receptor 5HT3a. They were negative for markers of other types of neurons. We confirmed this by RT-qPCR measurement of the expression of the corresponding genes. Both AMp- or AMpu-iNs exhibited very similar electrophysiological properties (Figure S8).

Cooperation of ASCL1 overexpression and NUP37 knockdown in neuronal conversion and nuclear shrinkage

To understand how each component of AMpu contributes to the transdifferentiation, we converted MRC5 cells with A, M, p, u, and their combinations (Figures 5A–5P). As shown before (Jiang et al., 2015; Xu et al., 2016; Li et al., 2019; Zhu et al., 2023), the conversion required A, but A alone was very ineffective (Figures 5B–5Q and 5R). By itself, NUP37 knockdown (u) did not generate any neurons (Figure 5E), but markedly enhanced neuronal conversion induced by A (Figure 5J) or any combinations of factors that included A (Figures 5M, 5O, and 5P). The significant effect of u is highlighted in red in the statistical analyses of reprogramming efficiency (Figure 5Q) and the yield of MAP2⁺ neurons (Figure 5R). NUP37 knockdown (u) reduced the number of DAPI⁺ cells in comparison to the control condition or in combination with any other factors (Figure 5S). Very similar cell loss was induced by u or Au (Figure 5S); yet Au converted cells (Figure 5J), while u did not (Figures 5E–5Q and 5R). Additional pairwise comparisons (pu vs. Apu, Mu vs. Amu, Mpu vs. AMpu) also validate this (Figure 5). Thus, the effects of u on cell loss and reprogramming are independent; the latter requires A. Either A or u reduced nuclear area (Figures 4A, 5B, 5E, and 5T); their combination (Au) reduced nuclear area further (Figure 5T). The presence of u decreased nuclear area beyond the effect induced by any other combinations of factors, with or without A (Figure 5T). The converted neurons were live and functional; they fired action potentials regardless of whether they were converted by AMp or AMpu (Figures S8P and S8P′). This further corroborates that nuclear size reduction and cell loss are independent. We also conducted the opposite experiments by overexpressing NUP37 (U) in combination with A, M, and p and found that U decreased reprogramming efficiency and the yield of MAP2⁺ neurons, while reversing the nuclear shrinkage induced by AMp (Figure S9). To further confirm that NUP37 expression level affects AMp-induced transdifferentiation, we overexpressed NUP37 (U) and NUP37 shRNA (u) together with AMp and found that U abrogated the effects of u in enhancing reprogramming and reducing nuclear area (Figure S10).

Figure 5.

Cooperation of ASCL1 and NUP37 shRNA in reprogramming and nuclear shrinkage

(A–P) MRC5 human fibroblasts were reprogrammed without or with the indicated combinations of ASCL1 (A), miR124-9-9^∗-BclxL (M), p53 shRNA (p) and NUP37 shRNA (u), and co-stained as indicated at day 14. Bar, 100 μm.

(Q–S) Reprogramming efficiency (Q) as measured by the percentages of TUJ1⁺ or MAP2⁺ cells among all DAPI⁺ cells, reprogramming yield of MAP2⁺ cells per frame (R), and the number of DAPI⁺ cells per frame (S) at day 14. # and ^∗, p < 0.001, n = 15 (3 experiments, 5 frames each), vs. the corresponding condition without u for the indicated cell type, unpaired t test. $p < 0.001, n = 15 (3 experiments, 5 frames each), vs. no virus (−V), unpaired t test.

(T) Nuclear area for each condition. ^∗p < 0.05, n = 50 frames from 3 independent experiments, vs. the corresponding condition without u; unpaired t test. $p < 0.005, n = 50 frames from 3 independent experiments, vs. no virus (−V), unpaired t test.

Discussion

A fundamental question in biology is what determines the size of a cell and its various organelles. The diversity of cell sizes in organisms on earth makes it intractable at present to formulate a general theory. For the vast majority of eukaryotic cells, the nucleus is the indispensable organelle where genomic information is stored, transcribed, and replicated when necessary. Thus, the question of nuclear size determination has fascinated biologists since Robert Brown named the nucleus almost 200 years ago. Many studies over the past century have supported the idea that proliferative cells from yeast to those of humans maintain a relatively constant nucleocytoplasmic ratio (Cantwell and Nurse, 2019a, 2019b; Webster et al., 2009; Boveri, 1905). However, some neurons, such as nigral dopaminergic neurons (Matsuda et al., 2009), basal forebrain cholinergic neurons (Wu et al., 2014), motor and sensory neurons in the sciatic nerve (Standring, 2015), have a massive axon with a total length of tens to hundreds of centimeters while maintaining a nuclear size typical of neurons. It is unclear what happens to nuclear size when a proliferative cell becomes a postmitotic neuron. In the direct conversion of human fibroblasts to iNs, we found a progressive and significant reduction of nuclear size without a significant change in cell size. Thus, the embodiment of the same genome in the form of a neuron vs. a fibroblast does not involve cell size change, even though the morphologies are profoundly different. This study provides a concrete counterexample to the idea of constant nucleocytoplasmic ratio. Nuclear shrinkage in neuronal transdifferentiation was corroborated by the reduction in nuclear size during the maturation of iPSC-derived cortical neurons. Quantification of nuclear sizes suggests that the more mature a neuron is, the smaller its nucleus. Indeed, nuclear area did not change significantly from days 0–2, when ASCL1 drives cell cycle exit (Jiang et al., 2015). The significant decrease of nuclear area on day 3 paralleled the first appearance of MAP2⁺ neurons. Subsequent reduction of nuclear area coincided with the generation of more MAP2⁺ neurons and the extension of increasingly complex neuronal processes. These results suggest that nuclear shrinkage is part of the neuronal maturation process, perhaps controlled by the same mechanism that is likely to be complex. Future studies that comprehensively measure nuclear size of neurons vs. non-neurons at different developmental stages may reveal more interesting insights on nuclear size determination.

We have found that the nucleoporin NUP37 plays a critical role in the nuclear shrinkage induced by neuronal transdifferentiation. As NUP37 was progressively and significantly suppressed by ASCL1 in the direct conversion, NUP37 knockdown significantly enhanced the reprogramming and further reduced nuclear area. Consistent with these, OE of NUP37 attenuated the transdifferentiation and nuclear shrinkage. Although NUP37 knockdown did not convert fibroblasts to neurons, it markedly reduced nuclear area. In cooperation with ASCL1, NUP37 knockdown enhanced reprogramming and reduced nuclear area further. Previous studies have shown that NUP37 knockdown suppresses the proliferation of cancer cells by inducing G1 arrest and apoptosis, while its OE promotes the growth, migration and invasion of carcinomas (Luo et al., 2017; Huang et al., 2020; Xiong et al., 2023). We found that NUP37 knockdown reduced the total number of DAPI⁺ cells, while increasing the number of MAP2⁺ neurons. The results suggest that NUP37 knockdown may have eliminated unconverted non-neuronal cells, while enhancing AMp-induced transdifferentiation.

In ASCL1-induced transdifferentiation, ASCL1 is causal, in the sense that it is sufficient and necessary, albeit at a very low efficiency (Figure 5). There are many barriers that prevent the efficient manifestation of the causal ability of ASCL1. One of these barriers is nuclear size, which appears to be regulated by the NPC. ASCL1 directly binds to the NUP37 promoter to suppress its expression. Reduced expression of NUP37, a critical a subunit that anchors the core, decreased NPC number and density (Figure 3). Nuclear shrinkage is necessary but not sufficient for the conversion of fibroblasts to neurons, as NUP37 knockdown alone reduced nuclear size without making neurons (Figure 5). ASCL1 induced the conversion and nuclear shrinkage, which were both attenuated by NUP37 OE (Figure S9). Consistent results were seen in Figures 4A–4H, where NUP37 OE attenuated the transdifferentiation and nuclear shrinkage induced by AMp. Furthermore, NUP37 OE abrogated the enhancing effect of NUP37 knockdown in the transdifferentiation and nuclear shrinkage (Figure S10).

As a component of the ∼110-MDa human NPC, NUP37 interacts with other nucleoporins in the coat nucleoporin complex of the outer rings, which function as structural support and docking sites for other nucleoporins (Bilokapic and Schwartz, 2012; Lin and Hoelz, 2019). The NPC controls the nuclear export and import of macromolecules including mRNA, proteins, and their complexes. As neurons neither replicate the genome nor divide the nucleus, it is conceivable that NPCs of neurons may have different components and/or properties compared with mitotic cells (Cristi et al., 2023; Fare and Rothstein, 2024). The NPC is a dynamic structure that accommodates remodeling during cell fate transition, as well as variation in composition and conformation depending on cell type and functional state (D'Angelo et al., 2012; Raices and D'Angelo, 2017). The reduced number and density of NPCs during transdifferentiation (Figure 3) suggest a concerted nuclear reprogramming process induced by ASCL1 that includes NPC reduction and nuclear shrinkage. Because ASCL1 suppressed NUP37, which enhanced NPC reduction, nuclear shrinkage, and the transdifferentiation, these data support a causal role for NUP37 repression in these processes. Consistent with this interpretation, OE of NUP37 attenuated the transdifferentiation and increased nuclear size. Substantial efforts by the field are required to understand how the NPC controls nucleocytoplasmic transport of macromolecules differently in neurons vs. mitotic cells. The present study offers the first evidence that NUP37 plays a pivotal role in regulating nuclear shrinkage and NPC reduction in the conversion of mitotic cells to postmitotic neurons.

Methods

Materials

We purchased the following plasmids from Addgene: pLKO.1/p53shRNA (#19119), pLKO.1/scrambled shRNA (#1864), pMD2.G (#12259), psPAX2 (#12260), pTight-9-124-Bclx (miR9/9^∗-124, #60857), pRL-SV40P (#27163), and pGL3 enhancer vector (#212938). The BAC clone containing the NUP37 genomic DNA was purchased from BACPAC resources (RP11-282A19). FUW-tet-LoxP-Ascl1 and FUW-LoxP-M2rtTA were generated previously (Jiang et al., 2015). Human NUP37 pORF was purchased from Genomics-Online (ABIN5735196) and cloned into the FUW-tetO-LoxP lentiviral vector. shRNA for NUP37 (Forward:

CCGGCATTGCCTCCAGTAATCAAATCTCGAGATTTGATTACTGGAGGCAATGTTTTTG, reverse:

AATTCAAAAACATTGCCTCCAGTAATCAAATCTCGAGATTTGATTACTGGAGGCAATG) was cloned into pLKO.1. All constructs were verified by sequencing. Our previous publications have shown the effectiveness of the reprogramming factors in causing the OE of ASCL1 (A) and miRNA9/9^∗-124 (M) and p53 knockdown (p) by western blotting and RT-qPCR (Jiang et al., 2015; Zhu et al., 2023).

Lentivirus production

High-titer lentiviruses are prepared as described previously (Jiang et al., 2015; Xu et al., 2016; Li et al., 2019; Zhu et al., 2023). When MRC5 cells were infected with AMpu and M2rtTA at MOI of 10 for each kind of virus, there should be no cell death in the presence of puromycin (Jiang et al., 2015; Xu et al., 2016; Li et al., 2019; Zhu et al., 2023), due to the presence of puromycin resistance gene in pLKO.1/p53 shRNA.

Cell culture and transdifferentiation

Human primary cell lines included MRC5 (fetal lung fibroblast, 14 gestational weeks, male, American Type Culture Collection [ATCC] [CCL-171]), GM09918 (skin fibroblast, 78 years old, male, Coriell Institute), and AG22056 (newborn foreskin fibroblast, male, Coriell Institute). Fibroblasts were cultured in DMEM containing 10% heat-inactivated FBS, 1× L-glutamine, 1× non-essential amino acids, and 1× penicillin/streptomycin. They were seeded at 3 × 10⁴/cm² on Matrigel-coated multiwall plates and infected with the indicated combination of lentiviruses at MOI of 10 for each virus in the presence of polybrene (8 μg/mL) for 16 h. After washing the cells in FBS-free media three times, they were cultured in FBS-free media for 24 h to arrest cell cycle at the G1/S checkpoint (Jiang et al., 2015). Transdifferentiation was initiated at day 0 by changing the media to neural induction media with DOX (DMEM/F12, 1×N2 (Gibco, 17502048), 1×B27 (Gibco, 12587010), 0.1 mM NEAA, 0.5 μM dorsomorphin (Tocris, 3093), 2.5 μM SB431542 (Tocris, 1614), 1 μM PD0332991 (Tocris, 4786), 3 μM CHIR99021 (Tocris, 4423), 0.2 mM vitamin C (Sigma, A4544), 10 μM Y27632 (Tocris, 1254), 0.5 mM db-cAMP (ApexBio, B9001), 20 ng/mL NGF (PeproTech, 450-01), 20 ng/mL GDNF (PeproTech, 450-10), 20 ng/mL BDNF (PeproTech, 450-02), and 1 μg/mL DOX (Tocris, 4090)). Media was changed completely every 2 days, with DOX treatment for the first 7 days only. Cells were maintained at 37°C and 5% CO₂. Face masks were worn in the cell culture room to minimize mycoplasma infection of cells, which routinely tested negative for mycoplasma by PCR (Zhang et al., 2021).

Nuclear volume measurement

Nuclear volume was determined by 3D reconstruction of a z stack of images of DAPI-stained nuclei (Figures S1A and S1B). The number of Z-steps is determined by the Nyquist step size, which is λn/(NA)², where λ is the wavelength, n is the refractive index of the medium, and NA is the numerical aperture of the objective lenses. Volume measurement was performed using the 3D Objects Counter plugin in ImageJ.

Differentiation of human iPSCs to cortical neurons

Human iPSCs derived from a healthy individual with non-integrating episomal vectors (CW70344) were purchased from the California Institute for Regenerative Medicine (CIRM) Human Pluripotent Stem Cell Line Repository. Differentiation of human iPSCs to cortical neurons was described previously (Jiang et al., 2025).

RT-qPCR

Total cellular RNA extracted using TRIzol (Invitrogen, 15596018) was reverse transcribed to cDNA using the iScript cDNA synthesis kit (Bio-Rad 1708891). Using the primers shown in Table S2, qPCR was performed on cDNA derived from 1 μg RNA on a CFX Duet Real-Time PCR System (Bio-Rad) with the program consisting of an initial denaturation step of 4 min at 95°C, followed by 40 cycles of 30 s at 95°C, 30s at 60°C, and 30 s at 72°C. Average threshold cycle (Ct) values were normalized against the average Ct value of GAPDH from the same sample.

Luciferase reporter assay

We cloned the promoter region of the NUP37 gene by PCR amplification of a BAC clone containing the genomic DNA of the human NUP37 gene (RP11-282A19) using the forward primer GTCCTAGGTACCGAGAAGAAGGGAGGGAACAAGA and the reverse primer TTTCGGAAGCTTCTCTTCAGTGTGCGGAAGGA. The underlined parts of the primers correspond to the NUP37 promoter region, while the italicized parts are KpnI and HindIII sites for cloning, respectively. The bolded parts are extra sequence to help restriction enzyme digestion. After sequencing verification, the 408 bp promoter was subcloned into pGL3 firefly luciferase reporter vector. The pGL3-NUP37 promoter construct was co-transfected into HEK293FT or MRC5 cells, together with a Renilla luciferase transfection control vector (pRL-SV40). At 24 h (HEK293FT) or 48 h (MRC5) after transfection, cells were lysed for luminescence measurements using the Promega Dual-Luciferase Reporter Assay kit (Promega, E1910), according to the manufacturer’s protocol. The first dispensation (Luciferase Assay reagent II) measured firefly luciferase activity, representing the transcriptional output from the NUP37 promoter. Dispensation 2 (Stop and Glo reagent) quenched firefly activity and measured Renilla luciferase activity, enabling normalization for transfection efficiency.

ChIP-qPCR

Using four independent ChIP-seq datasets assessing ASCL1 binding in human cells (Woods et al., 2022; Borromeo et al., 2016; Park et al., 2017; Păun et al., 2023) visualized in the Integrated Genome Browser (Robinson et al., 2011; Thorvaldsdóttir et al., 2013), we designed primers for ChIP-qPCR measurements of ASCL1 binding on NUP37 (CAGGCTGGTCTTCCTTGGG and CGCCGCTGAATACAGTTCAA) or NUP210 (TCTGTAAGTTACCTGCTCGGG and AACGAGGGTCCCTGTGTTCT) based on the ASCL1 peaks in Figure 2F. MRC5 cells cultured in 6-well plates (2 plates per condition) were reprogrammed with the method in Figure 1A. Cells were cross-linked for 10 min with formaldehyde (final concentration 0.75%), which was quenched with 125 mM glycine. After cells were lysed in ChIP lysis buffer (50 mM HEPES-KOH pH 7.5, 140 mM NaCl, 1 mM EDTA pH8, 1% Triton X-100, 0.1% sodium deoxycholate, 0.1% SDS, protease inhibitors [Roche, 04693116001]), genomic DNA was sonicated using a Fisher Scientific Sonic Dismembrator (Model 500). After determining DNA concentration, 25 μg DNA was used for each immunoprecipitation. Each sample was diluted 1:10 in RIPA buffer (50 mM Tris-HCl pH 8, 150 mM NaCl, 2 mM EDTA pH 8, 0.5% sodium deoxycholate, 0.1% SDS, protease inhibitors) for immunoprecipitation with the indicated antibody (against ASCL1, H3K4Me3, or H3K27Me3) or no antibody (control). After incubation in Protein A agarose and salmon sperm DNA beads (Millipore, 16-157), chromatin was washed with the low salt wash buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl pH 8, 150 mM NaCl), the high salt wash buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl pH 8, 500 mM NaCl), and LiCl wash buffer (0.25 M LiCl, 1% NP-40, 1% sodium deoxycholate, 1 mM EDTA, 10 mM Tris-HCl pH 8) (Lee et al., 2017; Palomo-Irigoyen et al., 2020). DNA was eluted and purified (QIAquick PCR purification kit, Qiagen 28104), followed by qPCR determination of binding, relative to the input.

Immunofluorescence staining and imaging

Immunocytochemistry was performed as previously described (Jiang et al., 2015). Briefly, cells at day 14 were fixed in 4% paraformaldehyde, followed by permeabilization in 0.1% Triton X-100, and incubation in a primary antibody (Table S5) overnight at 4°C. Following washing and further incubation in secondary antibody (Table S5) for 1 h at RT, the cells were treated with DAPI nuclear stain (2.5 mg/mL) for 10 min. Cells were then imaged using a Leica AF6000 inverted microscope or a Nikon N-STORM/N-SIM Super Resolution Microscope. Three independent experiments were completed for each condition described. MAP2⁺, TUJ1⁺, and DAPI⁺ cells were counted manually. Small and bright DAPI⁺ dots were dead cells and were not counted (Figure S11). We have used this standard consistently in the present study and our previous publications (Jiang et al., 2015; Xu et al., 2016; Li et al., 2019; Zhu et al., 2023). For clarification, Figure S11 details the counting standard we used for all immunostaining quantification.

Quantification of NPC number and density

We quantified NPC in Fiji/ImageJ from maximum-intensity projections of mAb414 immunostaining. Images were background-corrected (Subtract Background with rolling-ball radius of ∼15 pixels) and lightly smoothed (Gaussian blur σ ≈ 0.8 pixels); then puncta were detected with Find Maxima using a fixed prominence threshold (noise tolerance tuned on pilot images and held constant thereafter). Detected peaks were converted to a binary mask and counted via Analyze Particles with a single, pre-specified size window. Pixel size was calibrated prior to analysis, and all parameters were applied identically to every image.

Electrophysiology

MRC5 cells plated on Matrigel-coated plastic coverslips (Thermo Fisher Scientific, 174969) were transduced with lentiviruses expressing AMp or AMpu. On days 18–22, whole-cell patch-clamp electrophysiological recordings were performed. The bath solution contained 130 mM NaCl, 3 mM KCl, 1.2 mM CaCl₂, 1 mM MgCl₂, 10 mM HEPES-NaOH, and 10 mM glucose, pH 7.3, osmolarity 310 mOsm. The pipette solution contained 125 mM K-gluconate, 2 mM MgCl₂, 10 mM HEPES, 1 mM EGTA, 2 mM Na₂ATP, 0.4 mM Na₂GTP, and 5 mM Na₂-phosphocreatine, pH 7.3, osmolarity 300 mOsm. Recordings on cells with neuronal morphology were obtained with an Axon Instruments 200B amplifier that was controlled and monitored by an IBM PC running pClamp 10.0 with a DigiData 1440A series interface (Axon instruments). For the recording of voltage-gated sodium (Na_v) and potassium (K_v) currents, cells (held at −60 mV) were depolarized by brief pulses of voltage steps (−50 mV to +50 mV). For the recording of evoked action potentials, membrane potentials were kept at −60 mV, and a series of hyperpolarizing and depolarizing step currents were injected. Data were analyzed in Clampfit (Molecular Devices) and Igor software (RRID: SCR_000325, WaveMetrics, Lake Oswego, OR).

Bioinformatic analysis of single-cell RNA-seq, bulk RNA-seq, and ATAC-seq data

For analysis of the scRNA-seq dataset, pre-processed and annotated gene expression matrix was downloaded from GSE67310 (Treutlein et al., 2016) and reformatted and analyzed in scanpy (v.1.9.2) (Wolf et al., 2018). Principal components (PCs) were computed on highly variable genes detected by default parameters with the gene expression matrix. The first 16 PCs were used for constructing nearest neighbor graph and UMAP positions. For visualizing with force-directed graph, ForceAtlas2 (Jacomy et al., 2014) was used on the first 2 UMAPs with 10 nearest neighbors. For trajectory tracing, Palantir (Setty et al., 2019) was used to compute embedding in multiscale diffusion space on UMAP, and tree branches were traced by simple principal tree wrapped in scFates package (Faure et al., 2023). Pseudotime was then computed setting MEF as root, and dendrogram of the trajectories was generated with scFates (Faure et al., 2023). Expression value of single gene was visualized based on the original data. For bulk RNA-seq datasets, quality control of raw sequencing reads by FastQC was used, following by pre-processing by removal of low-quality bases, adapter sequences, and short reads by the Trimmomatic tool. FastQC was re-run to confirm improvements. HISAT2 was used to align reads to the reference genome, followed by transcript assembly by StringTie and subsequent gene quantification by featureCounts. Normalization and differential gene expression analysis was conducted by DESeq2, using a median of ratios method. For analysis of ATAC-seq datasets, FastQC quality control and Trimmomatic pre-processing were completed, followed by Bowtie2 alignment designed for short-read sequencing data. MACS2 Callpeak, a peak-calling algorithm, was then used to identify open chromatin regions by detecting enrichment of sequencing reads.

Live/Dead staining

Cells were stained and imaged at different time points in transdifferentiation with a commercially available LIVE/DEAD cell imaging kit (Thermo, R37601). Staining was conducted according to the manufacturer’s protocol. In brief, cells were incubated for 15 min in 1× working solution of 50% cell culture medium and 50% Calcein AM live cell indicator and BOBO-3 iodide dead cell indicator. Cells were then incubated for 12 min in 1 μm/mL Hoechst 33342 nuclear stain.

Statistical analysis

All data from this study were analyzed using GraphPad Prism 8 (GraphPad Software). Unless specified, mean ± SEM has been used throughout. Differences between two groups were assessed via Student’s t test.

Resource availability

Lead contact

Further information and requests should be directed to the lead contact Dr. Jian Feng (jianfeng@buffalo.edu).

Materials availability

Produced materials are available upon request.

Data and code availability

Codes are deposited at https://github.com/fenglabbuf/2026_Fisher_et_al.

Acknowledgments

The work is supported by the National Institutes of Health grants NS113763 (J.F.), AG079797 (J.F. and Z.Y.), and NS127728 (J.F. and Z.Y.) and US Department of Veterans Affairs Merit Review award I01BX002452 (J.F.).

Author contributions

E.F. and J.F. conceived the project and wrote the manuscript with input from all other authors. E.F. performed most of the experiments and analyses. Z.J. performed electrophysiological recording and analyses under the guidance of Z.Y. L.L. performed bioinformatics analyses. G.C. performed western blot analysis. K.S. conducted the differentiation of human iPSCs to cortical neurons.

Declaration of interests

The authors declare no competing interests.

Published: February 26, 2026