Introduction

Neurons in the central nervous system (CNS) possess limited regenerative capacity, resulting in permanent neuronal loss after injury or damage. In contrast, glia especially microglia exhibit a higher regenerative capacity(1–3). Reprogramming glial cells with regenerative potential into neurons lacking regenerative capacity offers a promising strategy for repairing damage in the CNS. The glia-to-neuron conversion process is achieved primarily by the overexpression of specific fate-determining factors or the knockdown of certain genes. The manipulation of single genes such as NeuroD1 (4), PAX6 (5), SOX2 (6–9), ASCL1 (10) and PTBP1 (11, 12) has been reported to induce inter-neuroectodermal lineage conversion, transforming astrocytes, oligodendrocyte precursor cells (OPCs) or Muller glia into neurons. While this approach may hold significant potential for clinical therapies, it still requires validation through more stringent and robust approaches. Well-controlled lineage tracing and unambiguous live imaging are gold standards for demonstrating the glia-to-neuron conversion (13, 14). However, controversial studies employing rigorous lineage tracing and live imaging techniques have failed to observe some inter-lineage glia-to-neuron conversions under single-gene manipulation (15–23), including NeuroD1 (21–23). In addition to inter-lineage conversion, Nakashima et al. have reported a NeuroD1-induced microglia-to-neuron cross-lineage conversion (24). Using lentiviral overexpression of NeuroD1, we conducted lineage tracing, in vitro live-cell imaging, and microglia ablation, demonstrating that the previously reported cross-lineage conversion of microglia into neurons is a result of experimental artifacts, e.g., off-target transduction from virus (13). Although Nakashima et al. argued that the phenomenon we observed was due to insufficient NeuroD1 expression in microglia (25), our comparison of the data from both studies (13, 24) indicates that our NeuroD1 expression levels are comparable to, or even higher than, those reported in their study (26).

Surprisingly, Nakashima et al. recently reported that NeuroD1 can not only can achieve the cross-lineage conversion of microglia into neurons at brain injury sites in vivo, but also promotes recovery from brain injury (27). The implementation of novel yet contentious strategies in clinical trials or practice requires extreme caution. Such conflicting reports are of particular concern because they directly influence the potential translation of glia-to-neuron reprogramming strategies into therapeutic applications. To address this controversy, we have employed a comprehensive combination of methodologies, including genetic lineage tracing (virus-independent to prevent potential leakage) and two-photon in vivo live imaging (enabling direct observation of NeuroD1-expressing microglial cell fate). Our findings unequivocally demonstrate that microglia do not undergo conversion into neurons in vivo under either physiological or pathological conditions. These results underscore the critical importance of employing rigorous lineage tracing and cell fate mapping techniques to validate glial reprogramming approaches.

Results

Microglia-specific NeuroD1 expression using a virus-free genetic lineage tracing system

In our previous study, lineage tracing revealed that lentivirus-driven ectopic expression of NeuroD1 does not induce the conversion of microglia into neurons but rather triggers microglial cell death through BCL2-dependent apoptosis (28). However, the viral tools used demonstrated limited efficiency in specifically targeting microglia in vivo and resulted in off-target neuronal expression, regardless of whether lentivirus or adeno-associated virus (AAV) (with different serotypes 6, 8 and 9) was employed, and irrespective of the use of CX3CR1 or CD68 promoters (Figure S1). Similar observations have also been reported by our prior study and those of other groups (21, 28, 29).

Direct evidence for glia-to-neuron conversion by NeuroD1 via a virus-free ectopic expression system genetic lineage tracing is lacking, by which avoids experimental artifacts from virus leakage. Thus, we utilized transgenic animals to ectopically express NeuroD1 in microglia and track the cell fate of microglia expressing NeuroD1. We crossed a microglia-specific mouse line CX3CR1-CreER (30) with an inducible NeuroD1-expressing mouse line Rosa26-LoxP-Stop-LoxP-NeuroD1-IRES-GFP (LSL-NeuroD1-GFP) (31), to obtain CX3CR1-CreER::LSL-NeuroD1-GFP mice(NeuroD1^MG for short). After tamoxifen treatment, NeuroD1 was ectopically expressed in microglia, accompanied by GFP for lineage tracing (Figure 1A). After tamoxifen administration, the majority of GFP⁺ cells faithfully express NeuroD1 in both the cortex and the striatum across time (Figure 1B-D, Figure S2). Results showed that GFP was specifically expressed in microglia avoiding off-target expression in neurons (Figure 1E-H).

Figure 1.

NeuroD1 fails to induce microglia-to-neuron conversion in vivo

Based on this stringent lineage tracing tool, all GFP⁺ cells remained IBA1⁺ microglia across all examined time points (4, 10, 18, 30, and 45 days post–tamoxifen induction; Figure 1E, F, Figure 2B,D). Consistently, no GFP⁺NeuN⁺ double-positive cells were detected, indicating that ectopic expression of NeuroD1 did not induce microglia-to-neuron conversion (Figure 1G,H, Figure 2B,D). Besides, GFP⁺ cells were also not co-labeled with MAP2 or TUJ1(Figure S3). Therefore, the genetic lineage tracing experiment reveals that NeuroD1 cannot drive microglia-to-neuron conversion in vivo.

Figure 2.

NeuroD1 expression induces microglial loss but not conversion in vivo

Since immunostaining might yield false-positive signals in the absence of true expression, we further examined microglial states in vivo using 2-photon microscopy to track NeuroD1-expressing, GFP-labeled microglia. Imaging of GFP⁺ cells in the cortex of NeuroD1^MG mice was performed every 4 days starting from 18 days post–tamoxifen induction (Figure 3A). Through 2-photon imaging, we did not observe any morphological evidence of microglia-to-neuron conversion. Instead, the GFP-expressing cells exhibited highly ramified morphologies with long processes characteristic of microglia, rather than typical morphology of neuronal cells (Figure 3B, Movie 1). Notably, some GFP-labeled microglia disappeared over time(Figure 3B, yellow arrow), suggesting that NeuroD1 expression does not induce microglia to neuron conversion but instead promotes microglial cell death in vivo, consistent with our previous in vitro findings(32).To further assess this effect, we quantified GFP⁺ microglia over time after tamoxifen induction. The number of GFP⁺ cells initially increased and peaked at D30, but subsequently declined by D45 (Figure 3C, D), supporting the conclusion that NeuroD1 expression contribute to microglia loss in vivo. To further investigate the mechanisms underlying microglial loss following NeuroD1 overexpression, we examined markers of autophagy in cortical sections at D18. In NeuroD1^MG mice, reporter⁺ cells exhibited about 4-fold higher LC3 and P62 levels than controls, as shown by immunostaining (Figure 3E-G), indicating accumulation of autophagosomal markers and suggesting impaired autophagic flux. Consistent with increased cell death, TUNEL staining revealed the increase of TUNEL+ cells among reporter⁺ cells in NeuroD1^MG mice relative to controls (Figure S4). Taken together, these data indicate that NeuroD1 overexpression triggers cellular stress and death pathways (including autophagy dysfunction and apoptosis), leading to microglial loss in vivo rather than microglia-to-neuron conversion.

Figure 3.

NeuroD1 fails to reprogram microglia into neurons even after traumatic brain injury

Previous studies have suggested that injury-induced environments might facilitate glial cell fate reprogramming. For example, Heinrich et al. demonstrated that a stab wound prior to virus delivery is required for SOX2-mediated conversion of NG2 glia(33). Irie et al. showed that NeuroD1 can directly reprogram microglia/macrophages into neurons at the lesion site following transient middle cerebral artery occlusion(27). Likewise, Chen et al. reported that NeuroD1 converts astrocytes into neuron after spinal cord injury (SCI). These findings suggest that the failure of NeuroD1-mediated microglia-to-neuron conversion under physiological conditions may result from an unfavorable microenvironment.

To test the possibility, we subjected NeuroD1^MG mice to traumatic brain injury (TBI) in the motor cortex, followed by tamoxifen induction, and utilized Cx3cr1-creER::Ai14 (Ai14^MG) mice as controls (Figure 4A). Firstly, we performed grip strength task and rotarod tests to evaluate whether NeuroD1 expression in microglia could ameliorate motor deficits in TBI (Figure 4A). In grip strength task, the NeuroD1-overexpression group showed comparable performance to controls, with no significant differences in the strength across groups (Figure 4B). Similarly, no improvements were observed in rotarod task, reflected by the latency to fall (Figure 4C), indicating that NeuroD1 expression in microglia does not rescue TBI-induced motor dysfunction.

Figure 4.

To further investigate whether NeuroD1 expression could induce microglia-to-neuron conversion, we performed immunostaining on these brains. In both lesion core and distal regions, the GFP-labeled NeuroD1-expressing microglia did not co-localize with the neuronal marker NeuN (Figure 4D, E) and MAP2 (Figure 4F, G), similar to what was observed in control group (data not shown). Besides, we collected reporter (tdTomato or GFP) positive cells in both physiological and TBI condition for scRNA-seq, On the basis of cell type-specific annotations, we identified the cell types of the sorted cells and found no evidence of neuronal clusters in the scRNA-seq results (Figure 4H). Additionally, the microglial markers P2ry12, Hexb, and Tmem119 remained highly expressed in reporter⁺ microglia (Figure 4I, J), whereas neuronal markers such as Rbfox3, Tubb3, and Map2 were absent in both control and NeuroD1-overexpressing microglia (Figure 4I,J). Overall, these findings demonstrate that NeuroD1 fails to induce microglia-to-neuron conversion, even in the context of TBI.

Discussion

Our study provides compelling evidence that ectopic expression of NeuroD1 does not induce microglia-to-neuron conversion in the adult mouse brain. By employing virus-independent genetic lineage tracing and two-photon in vivo imaging, we systematically assessed the cell fate of NeuroD1-expressing microglia under both physiological and disease conditions. Across all approaches, NeuroD1-expressing microglia consistently retained their microglial identity and failed to acquire neuronal markers or morphology. These findings stand in contrast to previous reports claiming NeuroD1-mediated microglia-to-neuron conversion(25, 34), and instead support the notion that earlier observations may have resulted from experimental artifacts such as viral off-target expression(22, 26, 32, 35).

Previous studies have highlighted the challenges in validating direct glia-to-neuron reprogramming(26, 32). While initial reports demonstrated that the overexpression of single transcription factors such as SOX2, ASCL1, or NeuroD1 could convert glial cells into neurons, subsequent investigations using rigorous lineage tracing or live imaging failed to reproduce some of these inter-lineage conversions(21, 33). Our results reinforce these concerns by showing that even in a brain injury environment, NeuroD1 is insufficient to drive microglial fate conversion. In fact, our long-time observation revealed the disappearance of NeuroD1-expressing microglia over time, consistent with our prior finding that NeuroD1 overexpression induces BCL2-dependent apoptosis in microglia(32). Thus, rather than promoting neuronal reprogramming, NeuroD1 expression appears to compromise microglial survival. The combined evidence of LC3/P62 accumulation and increased TUNEL positivity, together with longitudinal in vivo imaging and prior in vitro data, indicates that NeuroD1 overexpression compromises microglial homeostasis and promotes cell death rather than inducing microglia-to-neuron conversion. This may also explain why the reporter efficiency of NeuroD1^MG is lower than that of Ai14^MG (quantification data not shown). These results suggest that NeuroD1 promotes microglial loss through multiple stress pathways rather than facilitating neuronal reprogramming.

Importantly, our data also demonstrate that NeuroD1 expression in microglia does not provide functional benefits in the TBI model. Behavioral assessments, including grip strength, and rotarod tests, revealed no improvement in motor function, which correlates with the absence of microglia-to-neuron conversion observed histologically. This discrepancy with previous claims that NeuroD1 promotes functional recovery following stroke or injury likely reflects differences in experimental design, particularly the reliance on viral delivery systems prone to off-target effects(34, 36). Our use of a virus-free genetic strategy ensures that NeuroD1 expression is restricted to microglia, thereby providing a more definitive evaluation of its reprogramming capacity. Although microglia may not represent a suitable cellular source for neuronal conversion, our team has developed an alternative therapeutic strategy based on microglia replacement(37, 38). Notably, this approach has been successfully translated into clinical practice and was shown to halt disease progression in patients with adult-onset leukoencephalopathy with axonal spheroids and pigmented glia (ALSP), supporting microglia as a promising target for cell-based therapies in neurological diseases(39).

Together, these findings underscore the necessity of applying stringent lineage-tracing approaches and multimodal validation strategies when evaluating cell fate conversion. While glia-to-neuron reprogramming remains a promising avenue for regenerative medicine, our results caution against premature translation of NeuroD1-based approaches into clinical settings. The lack of microglia-to-neuron conversion observed here does not diminish the therapeutic potential of targeting microglia in brain disorders; rather, it highlights the importance of exploring alternative strategies that preserve or modulate microglial functions without attempting to force lineage conversion. We agree the potential that NeuroD1 is able to induce stem cells or stem-like cells conversion(40), but direct microglia-to-neuron conversion is not applicable. Future studies should focus on identifying context-dependent mechanisms that may enable or restrict glial plasticity, and on developing approaches that combine genetic, epigenetic, and environmental cues to achieve reliable and functional neuronal reprogramming.

Materials and Methods

Animals

Cx3cr1-CreER[B6.129P2(C)-Cx3cr1^{tm2.1(cre/ERT2)Jung}/J, Stock No: 020940], Ai14 [B6.Cg-Gt(ROSA)26Sor^{tm14(CAG-tdTomato)Hze}/J, Stock No: 007914], LSL-Neurod1-GFP [Gt(ROSA)26Sor^{tm1(Neurod1-EGFP)Able}/BfriJ, Stock No: 034697] mice were purchased from The Jackson Laboratory. CX3CR1-CreER mice were crossed with LSL-Neurod1-EGFP mice to obtain Cx3cr1-CreER^+/-::LSL-Neurod1-EGFP^+/+ (NeuroD1^MG) mice. Similarly, Cx3cr1-CreER mice were crossed with Ai14 mice to obtain Cx3cr1-CreER::Ai14 (Ai14^MG) mice. All animals were housed in the Animal Facility of the Department of Laboratory Animal Science at Fudan University on a 12-h light‒dark cycle with food and water provided ad libitum. No significant sex differences were observed in this study.

Drug administration

For the NeuroD1^MG and Ai14^MG mice, tamoxifen at 150 mg/kg body weight was administered via intragastric gavage (i.g.) for four consecutive days to induce loxP-dependent recombination.

Tissue preparation for immunohistochemistry

The animals were deeply anesthetized with 1% pentobarbital sodium (60 mg/kg) via intraperitoneal injection and transcranially perfused with cold 0.01 M PBS and cold 4% paraformaldehyde (PFA). The brains were immediately separated from the skulls and postfixed in 4% PFA for 24h at 4°C. For vibratome sectioning, brains were directly cut into coronal sections at the desired thickness using a Leica vibratome after post-fixation. For cryo-sectioning, brains were dehydrated in 30% sucrose in 0.01 M PBS at 4 °C for 2–3 days, followed by immersion in fresh 30% sucrose for an additional day. Tissue sections containing the regions of interest were then cut at 35 or 40 μm thickness using a Leica CM1950 cryostat.

Immunohistochemistry

The brain sections were rinsed 3 times in 0.01 M PBS for 10 min. The samples were blocked with 0.01 M PBS containing 0.3% Triton X-100 and 4% normal donkey serum (NDS, Jackson ImmunoResearch, Cat#: 017-000-121) at room temperature (RT) for 2 h. The samples were subsequently incubated with primary antibodies in 0.3% PBST containing 1% NDS at RT overnight. The samples were rinsed 3 times with 0.03% PBST for 10 min and then incubated with fluorescent dye-conjugated secondary antibodies in 0.03% PBST containing 1% NDS at RT for 2 h, together with 1:500 4’,6-diamidino-2-phenylindole (DAPI, Sigma‒Aldrich, Cat#: D9542-10MG). Finally, the samples were rinsed with 0.03% PBST for 10 min 3 times, and the slides were mounted with antifade mounting medium. The edges of the cover glass were then sealed with nail enamel.

The primary antibodies used in this study were as follows: rabbit anti-Iba1 (Wako, Cat#: 019-19741, 1:500), goat anti-IBA1 (Abcam, Cat#: ab5076, 1:500), goat anti-GFP (Abcam, Cat#: AB6673, 1:500), goat anti-mCherry (Biorbyt, cat#: orb11618, 1:500), rabbit anti-GFP (Invitrogen, Cat#: A-11122, 1:1000), mouse anti-NEUN (Abcam, Cat#: ab104224, 1:500), rabbit anti-NeuroD1 (Abcam, Cat#: ab109224, 1:500), rabbit anti-RFP (Abcam, Cat#: ab62341, 1:1000), rabbit anti-DCX (Abcam, Cat#: ab18723, 1:200), mouse anti-MAP2 (Sigma, Cat#M9942: a, 1:2000), and rabbit anti-TUJ1 (Sigma, Cat#: T3952, 1:2000), mouse anti-LC3B (Cell Signaling Technology, Cat#: 83506T, 1:200); rabbit anti-p62 (Cell Signaling Technology, Cat#: 23214T, 1:500). The secondary antibodies used were as follows: AF488 donkey anti-goat IgG (Jackson ImmunoResearch, Cat#: 705-545-003, 1:2000), AF488 donkey anti-rabbit IgG (Jackson ImmunoResearch, Cat#: 711-545-152, 1:2000), AF568 donkey anti-rabbit IgG (Invitrogen by Thermo Fisher Scientific, Cat#: A10042, 1:2000), AF568 donkey anti-goat IgG (Invitrogen by Thermo Fisher Scientific, Cat#: A11057, 1:2000), AF647 donkey anti-rabbit IgG (Jackson ImmunoResearch, Cat#: 711-605-152, 1:2000), AF647 donkey anti-mouse IgG (Jackson ImmunoResearch, Cat#: 715-605-151, 1:2000), and AF647 donkey anti-goat IgG (Jackson ImmunoResearch, Cat#: 705-605-003, 1:2000).

In situ apoptosis assay

A commercial Tunnel Cell Apoptosis Detection Kit (Service bio, Cat#: G1502-50T; Elab science, Cat#: E-CK-A321) was used to detect in situ cell apoptosis. Briefly, brain sections were treated with Proteinase K (20 ug/mL) for 10 mins at RT. Then, brain sections were incubated with equilibration buffer for 10 minutes at RT. Next, apoptotic cells were labeled by TMR-5-dUTP or FITC-dUTP labeling mix at RT and counterstained with DAPI. Positive controls were treated with DNase I and negative controls were labeled by buffer without recombinant TdT enzyme.

Confocal microscopy

Confocal images of the fluorescent samples were acquired via 20x (NA 0.75), 40x (NA 1.30) or 60x (NA 1.40) objectives of a Nikon A2 confocal microscope. The Z-stacked focal planes were taken and maximally projected by ImageJ, whereas brightness and contrast were adjusted in ImageJ if necessary.

Tile scans of whole coronal brain sections were captured with an Olympus VS120 or VS200 system and exported from OlyVIA Software in TIFF format. Further data analysis was carried out in ImageJ.

Two-photon surgery and imaging

Two-photon cranial window surgery was performed on mice anesthetized with isoflurane (3% induction, 1.2%-1.5% maintenance). A 3.5 mm-diameter circular cranial window was created at the skull above the sensory cortex, and a same-diameter circular D263T glass slice was implanted. The cranial window and a head bar were bonded using glue and dental cement to maintain stability. In addition to performing routine postoperative care, the mice were injected intraperitoneally with antibiotics and anti-inflammatory drugs (dexamethasone sodium phosphate injection and ceftiofur sodium for injection) for five consecutive days.

Two-photon imaging was performed on days 18, 22, 26, and 30 post-tamoxifen induction (as shown in the timeline in Fig. 2). During imaging, the mice were anesthetized with isoflurane (3% induction, 1.2%-1.5% maintenance), and body temperature was maintained by a heating pad. Imaging was performed via a self-developed two-photon microscopy system (developed by the Institute for Translational Brain Research, Fudan University) with one resonant and one galvo scan mirror. The excitation was delivered with a Coherent Chameleon Ti-Sapphire pulsed laser. The fluorescence emission was filtered with a Semrork FF01-520/35-25, FF01-593/46 filter and detected with a Hamamatsu photomultiplier tube (H7422-40P). The EGFP signal in Cx3cr1-CreER::LSL-Neurod1-EGFP mice was excited by a 920 nm laser and imaged via a 16x, 0.8 NA water-dipping objective (Nikon). Images were acquired at a depth of approximately 100 μm below the dura using ScanImage(41) at step intervals of 2 μm. Each Z-axis plane image (512×512 pixels) was imaged at a consistent zoom factor (∼300 × 300 μm area) and acquired at a 1.06 Hz frame rate.

Image processing was performed via ImageJ, and the Z-stack image containing the target cells was selected. Brightness and contrast were adjusted if needed, and the maximal projection was performed. Video processing was performed via Adobe Premiere software.

AAV vectors, lentivirus preparation, and stereotaxic injection

The AAV vectors were purchased from Obio Technology, Ltd., and PackGene Biotech, Inc., and were as follows: AAV1/2-CMV-EGFP, AAV5/2-CMV-EGFP, AAV6/2-CMV-EGFP, AAV8/2-CMV-EGFP, AAV9/2-CMV-EGFP, AAV2-DJ-CMV-EGFP, AAV2-retro-CMV-EGFP, AAV-PHP.eB-CMV-EGFP, AAV6-CX3CR1-GFP, AAV8-CX3CR1-GFP, AAV9-CX3CR1-GFP, AAV6-CD68-GFP, AAV8-CD68-GFP, and AAV9-CD68-GFP.

All lentiviruses were packaged by Obio Technology, Ltd., according to the author’s design and requirements. All lentiviruses were packaged in HEK293 cells, and the packaged plasmids consisted of pMDL, VSV-G and pREV. The following lentiviral vectors were used: pLenti-CAG-flag-NeuorD1, pLenti-CAG-flag-MCS, pLenti-CMV-GFP, pLenti-CX3CR1-GFP and pLenti-CD68-GFP.

For intracranial injection, the mouse was anesthetized with isoflurane. The stereotaxic injection coordinates were +1.0 mm anterior/posterior, ± 1.7 mm medial/lateral, and −0.75 mm dorsal/ventral.

Traumatic brain injury (TBI)

TBI was induced using the controlled cortical impact (CCI) model as previously described(42). Briefly, mice were anesthetized with ketamine (100 mg/kg) and xylazine (10 mg/kg) and placed in a stereotaxic frame (Stoelting, Wood Dale, IL, USA). Body temperature was maintained at 37.0 ± 0.5 °C using a heating pad throughout the procedure. After scalp disinfection, a midline incision (1.5–2 cm) was made to expose the skull. A 4 mm craniotomy was performed over the right parietal cortex (between bregma and lambda, 1 mm lateral to the midline) using a portable drill, taking care to preserve the dura mater. Mice with damaged dura were excluded from further analysis. Cortical impact was delivered using a 3 mm rounded steel impactor (PinPoint PCI3000, Hatteras Instruments Inc., USA) at a velocity of 1.5 m/s, impact depth of 1.5 mm, and dwell time of 100 ms. After impact, the bone flap was sealed with sterile bone wax and the scalp was sutured. Sham-operated mice underwent the same surgical procedures without cortical impact. Following surgery, mice were placed in a 37 °C recovery chamber until fully ambulatory.

Rotarod test

The mice were acclimated in the behavioral room for half an hour before testing. Briefly, during the training phase, the mice were trained to walk on a rotating rod at 5 RPM, allowing them to maintain balance by walking forward. Each mouse was placed on the rod for 60 seconds, after which it was returned to its home cage. This procedure was repeated 3 times with a 10-min interval. In the testing phase, the mice were placed in separate lanes on the rod, which was set to accelerate from 4 to 40 RPM over a 300-second period. This procedure was repeated three times at approximately 30-min intervals. The latency to fall was recorded for statistical analysis.

Grip strength test

The mice were housed in the behavioral room half an hour before the behavioral testing. Briefly, the grip strength meter was positioned on the table. Each mouse was gently held by the tail and placed with all the paws on the grip strength grid. The mouse was then gently pulled back until its grasp was released, after which the maximum holding force was recorded. This procedure was repeated 6 times, and the maximum force was used for statistical analysis.

Single-cell preparation and fluorescence-activated cell sorting (FACS)

For scRNA-seq, FACS was used to harvest GFP⁺ or tdTomato⁺ cells from tamoxifen-treated Cx3cr1-creER::LSL-Neurod1-GFP and Cx3cr1-CreER::Ai14 mice. Briefly, brains were dissected and cut into 1-mm³ pieces using a mouse stainless steel brain matrix (RWD). The tissue pieces were digested in 3 ml of DMEM containing 8 U/ml papain (Sangon Biotech, cat: A501612), 125 U/ml DNase I (Sangon Biotech, cat: A510099), and 1.5 mg/ml trypsin inhibitor (Aladdin, cat: A274384) via a gentleMACS Octo Dissociator (Miltenyi) with the 37C_ABDK program.

After 30 min of digestion, the mixture was gently pipetted up and down and then filtered through a 70 µm cell strainer. The dissociated cells were centrifuged at 370 × g at RT for 10 min, and the supernatant was discarded. The cell pellets were resuspended in 4 ml of 30% Percoll (Cytiva, cat: 17089101) and centrifuged again at 370 × g and 17°C for 10 min to remove debris. After centrifugation, the supernatant was discarded, and the cells were washed once with DPBS containing 0.5% BSA. Before the cells were loaded onto the SONY MA900, they were stained with 7-AAD (BD Pharmingen, cat: 559925, 1:700) to label dead cells. Finally, the tdTomato⁺ 7-AAD^- or GFP⁺7-AAD^- cells were collected for scRNA-seq immediately.

scRNA-seq data analysis

scRNA-seq was conducted via the MobiCube High-throughput Single Cell 3′ Transcriptome Set V2.1 (PN-S050200301) and the MobiNova-100 microfluidic platform. The single-cell suspension was adjusted to a concentration of 700–1200 cells/μl and immediately loaded onto a chip for microdroplet formation via the MobiNova-100 system. Reverse transcription, cDNA amplification, and DNA library construction were carried out according to the manufacturer’s protocol. High-throughput sequencing was performed in PE-150 mode.

FASTQ files were processed and aligned to the mouse reference genome (mm10) via MobiVision version 3.2, with unique molecular identifier (UMI) counts aggregated for each barcode. The resulting UMI count matrix was analyzed in R version 4.2.2 via Seurat version 4.3.0. To ensure data quality, we excluded low-quality cells and potential multiplets on the basis of the following criteria: cells with fewer than 200 genes or more than 5000 genes and cells with more than 10% UMIs mapped to mitochondrial genes. scRNA-seq data from the four groups were integrated via IntegratedData functions according to the manufacturer’s instructions. The remaining data were normalized via the NormalizeData function and scaled with ScaleData. For scRNA-seq analysis, principal component analysis (PCA) was conducted on variable features, with the top 30 principal components (PCs) selected for clustering and visualization via tSNE and uniform manifold approximation and projection (UMAP). Cell clustering was performed with a resolution of 1 to annotate the cell types. Differentially expressed genes (DEGs) among cell groups were identified via the FindMarkers function with parameters set to min.pct = 0.1 and a log fold change (FC) threshold of 0.25 (Wilcoxon rank-sum test). Genes with |log2FC| > 0.25 and P < 0.05 were considered significant DEGs and subsequently used for Gene Ontology (GO) analysis. Visualization of the results was performed via the VlnPlot and FeaturePlot functions.

Statistical analysis

Statistical analyses were performed with Prism 8.4.0 (GraphPad). Each data point represents the average statistical result of more than three brain sections in different bregma regions of one mouse. The results were evaluated independently in a double-blind manner. All the results are presented as the means ± standard deviations (SDs). One-way analysis of variance (ANOVA) with Holm‒Sidak’s multiple comparisons test (post hoc) was performed for multiple comparisons, whereas a two-tailed independent t test was used to compare the differences between two groups. No outliers were excluded.

Supporting Information

Fig. S1.

Fig. S2.

Fig. S3.

Fig. S4.

Data availability

The sequencing data generated in this study have been deposited in the Gene Expression Omnibus under accession number GSE277222. All other data supporting the findings of this study are available from the corresponding author upon reasonable request.

Acknowledgements

This work was supported by Brain Science and Brain-like Intelligence Technology–National Science and Technology Major Project (2022ZD0207200, 2022ZD0204700); National Natural Science Foundation of China (32571128, 323B2030, 325B2039); the Fellowship of China National Postdoctoral Program for Innovative Talents (BX20250132); the Fellowship of China Postdoctoral Science Foundation (2025M782569); Shanghai Pilot Program for Basic Research (21TQ014), Changping Laboratory (2025B-07-18) and Lin Gang Laboratory (LGL-8998-02). In addition, authors also express their gratitude and respect to all animals sacrificed in this study. During the preparation of this manuscript, the authors, as the non-native English speakers, used ChatGPT 4o and 5 to improve the language and enhance its readability.

Additional files

Video 1. This video shows that GFP⁺ cells did not convert to neurons in vivo via 2-photon imaging at 18, 22, 26 and 30 days post-tamoxifen administration. Refers to Figure 3.

Additional information

Funding

MOST | National Natural Science Foundation of China (NSFC) (32571128)

• Bo Peng

MOST | National Natural Science Foundation of China (NSFC) (323B2030)

• Xiaoyu Li

MOST | National Natural Science Foundation of China (NSFC) (325B2039)

• Yanxia Rao

China Postdoctoral Science Foundation (2025M782569)

• Xiaoyu Li

China Postdoctoral Science Foundation (BX20250132)

• Xiaoyu Li

Shanghai Pilot Program for Basic Research (21TQ014)

• Bo Peng

Brain Science and Brain-like Intelligence Technology-National Science and Technology Major Project (2022ZD0207200)

• Bo Peng

Brain Science and Brain-like Intelligence Technology-National Science and Technology Major Project (2022ZD0204700)

• Yanxia Rao