Abstract

   In vivo conversion of nonneuronal cells into neurons is a proposed
   strategy to replace neurons lost to CNS injury or disease.
   Glia-to-neuron trans-differentiation by viral vector–mediated GFAP
   mini-promoter–driven NeuroD1 remains hotly debated. Developing
   inducible, lineage-traceable transgenic mice, we find that
   astrocyte-to-neuron conversion is restricted to a specific time window
   within the lesion core of injured spinal cord and brain. Spatiotemporal
   lineage-mapping combined with single-cell transcriptomics reveals that
   ectopic NeuroD1 induces astrocyte-to-neuron conversion specifically in
   lesion cores via transit-amplifying OLIG2^+ progenitors during early
   injury phase, but not in late phases or in nonreactive astrocytes.
   Neither a loss-of-function NeuroD1 mutant nor stemness-reprogramming
   factor SOX2 induces astrocyte-to-neuron conversion. However, contrary
   to previous reports, the neuronal-like cells generated by NeuroD1 lack
   mature neuroelectrical properties, limiting their functional
   integration into neural circuits. Together, our findings establish a
   spatiotemporal framework for NeuroD1-driven glia-to-neuron conversion,
   revealing a mechanistic shift from direct astrocyte conversion toward
   transit-amplifying intermediates and highlighting the functional
   immaturity of NeuroD1-converted neurons.
     __________________________________________________________________

   NeuroD1 converts astrocytes to neurons via transit-amplifying
   progenitors, but the resulting neurons lack functional maturity.

INTRODUCTION

   Most neurodegenerative diseases of the central nervous system (CNS) are
   characterized by the progressive loss of neurons. The adult mammalian
   brain and spinal cord have limited regenerative capacity ([46]1–[47]3).
   Although diverse therapeutic strategies for CNS injuries have been
   investigated, including stimulation of endogenous neural stem cell
   (NSC) differentiation, administration of neuroprotective compounds, and
   transplantation of exogenous NSCs or neurons ([48]4–[49]8), efficacies
   in replacing lost neurons and generating functional neuronal activity
   and networks are limited. The direct in situ conversion of endogenous
   nonneuronal cells into functional neurons through fate conversion in
   the brain and spinal cord following injury has been proposed as a
   potential therapeutic approach ([50]9–[51]12).

   Neurons, astrocytes, and oligodendrocytes, the three fundamental neural
   cell types in the CNS, share the same NSC or progenitor cell origins
   during development ([52]13). Astrocytes are highly abundant and
   heterogeneous in the CNS and can be transient amplified in response to
   injury, representing a potential source to replace neuronal loss after
   injury or disease ([53]13–[54]16). It has been reported that neurogenic
   factors, which are active in radial glial stem cells during embryonic
   neurogenesis but down-regulated during differentiation, can be
   reactivated in postnatal and adult astrocytes to promote neuronal
   formation ([55]10, [56]17). Diseases and injuries can trigger astrocyte
   remodeling and reactivity including non-proliferative or proliferative
   responses ([57]18, [58]19). The plasticity of reactive astrocytes in
   response to injury positions them as a potential candidate for induced
   neurogenesis or lineage conversion ([59]20–[60]22).

   NeuroD1, a member of the basic helix-loop-helix transcription factor
   family, regulates neuronal differentiation during embryonic development
   ([61]23) and is vital for the survival and maturation of adult-born
   neurons ([62]24, [63]25) through reprogramming of chromatin and
   transcriptional landscapes ([64]26, [65]27). Previous studies reported
   that adeno-associated virus (AAV) vector–mediated expression of NeuroD1
   induces glial fibrillary acidic protein (GFAP)–expressing astrocytes to
   trans-differentiate into mature neurons ([66]28, [67]29), proposing a
   therapeutic potential of NeuroD1 in animal models of spinal cord
   injuries, stroke, and epilepsy as well as other neurological disorders
   ([68]30–[69]36). However, recent studies have raised questions about
   the conclusions of these studies as to authentic conversion of
   astroglia into neurons ([70]37–[71]40). Mis- or leaky expression of the
   neurogenic factor and reporter in already existing neurons may occur
   when using AAVs carrying a GFAP mini-promoter to drive NeuroD1
   expression ([72]37–[73]40). Thus, reporters used in this context do not
   faithfully indicate the glial origin of labeled neurons. Furthermore,
   the permissive state of astrocytes for NeuroD1-induced
   trans-differentiation remains controversial ([74]17, [75]41). While
   some reports suggest that NeuroD1 can drive neuronal conversion even in
   nonreactive astrocytes ([76]42), this remains a subject of debate.
   Additionally, the specific time window during which NeuroD1 induces
   neuronal formation from astrocytes following CNS injury remains poorly
   understood. Now, in vivo tools with transgenic animal models for cell
   lineage– and temporal-specific NeuroD1 induction are lacking, hindering
   robust in vivo genetic fate mapping needed to precisely define the cell
   state and temporal dynamics of astroglia-to-neuron conversion.

   To address the potential issues of mis-expression and leakage of
   reporters in viral vectors in endogenous neurons, we developed an
   inducible, lineage-traceable transgenic knock-in mouse line for NeuroD1
   expression and used these mice to examine resident astrocyte-to-neuron
   conversion in different regions following spinal cord or brain injury.
   We found that ectopic NeuroD1 expression in reactive astrocytes in the
   lesion core via a transit-amplifying intermediate, but not in distal
   injury region, drove the formation of neuronal cells in the injured
   spinal cord and brain. Trans-differentiation is observed when NeuroD1
   expression is induced during the early post-injury phase but not at
   later stages. However, contrary to the previous reports, we find that
   the newly converted neuron-like cells induced by NeuroD1 did not
   exhibit the electrophysiological characteristics of typical mature
   neurons. Thus, our robust, reproducible genetic models resolve key
   controversies by demonstrating NeuroD1 induction of authentic
   astrocyte-to-neuron conversion following CNS injury. In addition, our
   studies highlight that reprogramming through transit-amplifying
   intermediates, rather than direct astrocyte conversion, represents a
   unified mechanism for in vivo neuronal conversion while uncovering that
   the NeuroD1-converted neurons fail to achieve functional
   characteristics of mature neurons. These findings underscore the need
   to develop therapies that promote fully functional neurons in CNS
   diseases and injuries.

RESULTS

Generation of inducible and lineage-traceable NeuroD1 knock-in mice

   To determine the effect of NeuroD1 on astrocyte-to-neuron conversion
   after CNS injury, we generated an inducible transgenic knock-in line
   for NeuroD1 expression. The Rosa26 locus was targeted with a transgene
   containing a loxP-flanked STOP cassette, a mouse Neurod1 cDNA, and a
   P2A–green fluorescent protein (GFP) reporter. The NeuroD1-P2A^GFP
   fusion allows simultaneous expression of NeuroD1 and GFP, with the GFP
   signal used to trace NeuroD1-expressing cells ([77]Fig. 1A). To
   activate transgene expression in astrocytes, we cross-bred a
   well-validated, astrocyte-specific Aldh1l1-CreERT2 line ([78]43) with a
   Rosa-tdTomato (Ai9) reporter line (control mice) ([79]44) or with the
   Rosa-NeuroD1-P2A^GFP transgenics, hereafter referred to as iNeuroD1^GFP
   mice ([80]Fig. 1A).

Fig. 1. Ectopic NeuroD1 expression results in reactive astrocyte-to-neuronal
cell conversion in the lesion core after spinal cord injury.

   [81]Fig. 1.
   [82]Open in a new tab

   (A) Schematic of Rosa-CAG-loxP-STOP-loxP-tdTomato reporter mice,
   activated with Aldh1ll-CreERT2 to generate tdTomato-expressing
   astrocytes in control mice (left). Schematic of
   CAG-loxP-STOP-loxP-iNeuroD1^GFP transgene inserted into the Rosa26
   locus, activated with Aldh1ll-CreERT2 to generate NeuroD1-GFP
   expression in astrocytes in iNeuroD1^GFP mice (right). (B) Experimental
   scheme showing spinal cord injury (SCI) on day 0, following by TAM
   injection (dpi 5 to 8), in control and iNeuroD1^GFP mice at dpi 10. (C)
   Representative images showing the co-labeling of GFP (green) and
   NeuroD1 (red). Scale bars, 25 μm. (D) Western blot analysis for NeuroD1
   in iNeuroD1^GFP and control mice with or without injury, normalized to
   β-actin. (E) Representative image of GFAP-labeled spinal cord showing
   the lesion core and distal region after SCI in control mice. Blue,
   4′,6-diamidino-2-phenylindole (DAPI) stain. Scale bar, 250 μm. (F)
   Immunostaining for NeuN and tdTomato (tdT) or NeuroD1^GFP in the lesion
   core of control and iNeuroD1^GFP mice, respectively, at dpi 10. Scale
   bars, 50 μm. Right: Quantification of the percentage of
   reporter-positive cells expressing NeuN in the lesion core. n = 3 mice
   per group; means ± SEM. (G) Representative immunolabeling images for
   NeuN and tdTomato (tdT) or GFP in the distal region of control and
   iNeuroD1^GFP mice, respectively, at dpi 10. Scale bars, 50 μm. Right:
   Quantification of the percentage of reporter positive cells also
   positive for NeuN in the distal region. n = 3 mice per group. n.d., not
   detectable.

   To validate astrocyte-specificity of exogenous gene expression in
   control and iNeuroD1^GFP mice, spinal cord tissue was harvested after 3
   days of tamoxifen (TAM) administration (100 mg/kg/day) to induce gene
   expression (fig. S1A) ([83]45). After post-induction at day 5, tdTomato
   and GFP signals were predominantly detected in GFAP^+ astrocytes in
   both control and iNeuroD1^GFP mice, and these signals were not detected
   in neurons (NeuN^+), oligodendrocyte progenitors (PDGFRα^+), or
   microglia (Iba1^+) (fig. S1, B and C). These data indicate that ectopic
   NeuroD1 expression does not convert normal astrocytes to NeuN^+ neurons
   in the uninjured spinal cord

   To next investigate the fate of ectopic NeuroD1-expressing cells after
   injury, control and iNeuroD1^GFP mice were subjected to dorsal
   hemisection spinal cord injury ([84]46). TAM was administered from 5 to
   8 days post-injury (dpi) to induce NeuroD1 or tdTomato (control)
   expression ([85]Fig. 1B) ([86]45). GFP and NeuroD1 were co-expressed in
   the spinal cords of iNeuroD1^GFP mice, either with or without an
   injury, at 2 days after TAM administration to activate transgene
   expression in Aldh1ll^+ astrocytes ([87]Fig. 1C). GFP expression thus
   serves as a surrogate readout of TAM-inducible NeuroD1 expression under
   both physiological and pathological conditions. The induction of
   NeuroD1 expression by TAM-treatment was confirmed by Western blot
   analysis of spinal cord tissues in iNeuroD1^GFP mice at 10 days
   post-induction ([88]Fig. 1D). Notably, spinal cord injury alone did not
   induce endogenous NeuroD1 expression in the spinal cord of control mice
   ([89]Fig. 1D). Together, these data indicate that TAM induces ectopic
   NeuroD1 expression in the Aldh1ll^+ astrocytic lineage in iNeuroD1^GFP
   mice.

Ectopic NeuroD1 expression at early injury phase induces
astrocyte-to-neuronal cell conversion

   To investigate the effect of NeuroD1 expression on the fate of
   Aldh1l1-CreERT2–expressing astrocytes, we performed spinal cord injury
   in control and iNeuroD1^GFP adult mice at postnatal day 60 (P60) and
   administered TAM (100 mg/kg/day) ([90]45) from dpi 5 to 8. During this
   post-injury period, reactive gliosis with proliferative GFAP^+
   astrocytes peaks at dpi 5 (fig. S2, A and B), compared to the acute
   phase at dpi 3 or the late phase at dpi 10, suggesting a transient
   amplifying phase of reactive astrocytes during the early phase of
   injury around dpi 5.

   Because astrocyte limitans borders around lesions extend ~0.5 mm from
   the injury site after spinal cord injury ([91]47, [92]48), we divided
   injured spinal cord tissues into the lesion core and peri-injury distal
   regions ([93]Fig. 1E), which are enriched with proliferative and
   non-proliferative reactive astrocytes, respectively (fig. S2, A and B).
   TAM administration induced expression of tdTomato in control mice and
   NeuroD1/GFP in iNeuroD1^GFP mice ([94]Fig. 1F). At dpi 10, most of
   GFP-expressing cells in the lesion core of iNeuroD1^GFP mice were
   NeuN^+ neuronal cells (77.6 ± 1.9%), whereas, in control mice, none of
   the tdTomato^+ astrocytes were co-labeled with NeuN ([95]Fig. 1F).
   These data suggest that ectopic NeuroD1 induces robust NeuN^+ neuronal
   cell formation from reactive astrocytes in the lesion core. In
   contrast, in the distal injury regions of control and iNeuroD1^GFP
   mice, none of the reporter-labeled cells expressed NeuN ([96]Fig. 1G).
   This indicates that, in non-proliferative reactive astrocytes at the
   distal injury region, NeuroD1 expression did not convert astrocytes
   into NeuN^+ neuronal cells. Thus, during the transient amplifying
   period of reactive astrocytes after spinal cord injury, NeuroD1 can
   only induce the robust expression of NeuN in neuron-like cells in
   reactive astrocytes in the lesion core and not in distal regions.

NeuroD1 ectopic expression before injury promote astrocyte-to-neuron
conversion

   An injury could potentially activate the Aldh1l1 promoter in
   preexisting neuronal cells, a confound of the AAV-mediated delivery
   approach ([97]49). To ensure that NeuroD1 was only expressed in
   preexisting resident astrocytes directed by Aldh1l1-CreERT2 in adult
   mice, we administered TAM to iNeuroD1^GFP mice for 3 days before injury
   to induce Cre expression in resident astrocytes and then performed
   spinal cord injury ([98]Fig. 2A). Immunostaining for the astrocytic
   markers GFAP and S100β confirmed the presence of GFP reporter–labeled
   astrocytes within the lesion core ([99]Fig. 2B). At dpi 21, ~54.4% of
   NeuroD1^GFP+ astrocytes became NeuN^+ neuronal-like cells in the lesion
   core, whereas no NeuroD1^GFP+/NeuN^+ cells were observed in the distal
   injury region ([100]Fig. 2, C and D). The NeuN induction rate compared
   favorably with the 77.6% of NeuN^+GFP^+ cells detected in iNeuroD1^GFP
   mice when the injury preceded the TAM-induction ([101]Fig. 1F). These
   data suggest that the NeuN^+ neuronal-like cells formed post-injury are
   derived from preexisting astrocytes with the forced NeuroD1 expression
   before the injury as opposed to ectopic transgene expression in
   endogenous neurons.

Fig. 2. NeuroD1-induced astrocyte-to-neuron conversion in a time- and
cell-state–dependent manner after spinal cord injury.

   [102]Fig. 2.
   [103]Open in a new tab

   (A) Pre-injury experimental scheme to induce NeuroD1^GFP expression in
   Aldh1l1^+ astrocytes, followed by SCI and analysis at dpi 21. (B)
   Representative immunolabeling images of S100β and GFAP in the lesion
   core at 21 days after SCI. Scale bar, 50 μm. (C) Representative
   immunolabeling images of NeuroD1^GFP and NeuN in the lesion core and at
   distal regions in iNeuroD1^GFP mice at dpi 21. Scale bars, 50 μm. (D)
   Quantification of the percentage of NeuN^+ NeuroD1^GFP+ cells. n = 3
   mice per group; means ± SEM. (E) Post-injury experimental scheme to
   induce SCI (day 0), followed by daily BrdU injection from days 2 to 5,
   TAM from days 5 to 8 to induce NeuroD1^GFP expression in Aldh1l1^+
   astrocytes, and analysis at dpi 10. (F) Representative immunolabeling
   images of NeuroD1^GFP, BrdU, and NeuN in the lesion core at dpi 10.
   Scale bars, 25 μm and (zoomed-in region) 10 μm. (G) Quantification of
   the percentage of BrdU^+NeuroD1^GFP+NeuN^+ cells n = 3 mice per group;
   means ± SEM. (H) Post-acute experimental scheme to induce NeuroD1^GFP
   expression in Aldh1l1^+ astrocytes, followed by SCI and analysis at dpi
   18. (I) Representative immunolabeling images of S100β and GFAP in the
   lesion core at dpi 18. Scale bar, 50 μm. (J) Immunostaining for NeuN
   and NeuroD1^GFP in the lesion core of iNeuroD1^GFP mice at dpi 18.
   Scale bar, 50 μm. (K) Quantification of the percentage of
   NeuroD1^GFP+cells expressing NeuN in the lesion core. n = 3 mice per
   group; means ± SEM. (L) Quantification of the percentage of
   Ki67^+NeuroD1^GFP+cells in the lesion core. n = 3 mice per group;
   means ± SEM. n.d., not detectable.

Conversion is restricted to the injury core via transient-amplifying
intermediates

   After injury, astrocytes in proliferative and non-proliferative states
   are detected in the lesion core and peri-injury distal region,
   respectively ([104]18). To trace the fate of proliferative reactive
   astrocyte subpopulations in iNeuroD1^GFP mice, we administrated
   5-bromo-2′-deoxyuridine (BrdU) from dpi 2 to 5, the peak period of
   reactive astrocyte self-amplification ([105]48), to label
   transit-amplifying cells and followed by TAM administration from dpi 5
   to 8 to induce NeuroD1 expression ([106]Fig. 2E). We detected
   BrdU^+/NeuroD1^GFP+/NeuN^+ intermediates in the lesion core but not in
   distal regions ([107]Fig. 2F), and most of BrdU-labeled NeuroD1^GFP+
   cells expressed NeuN (91.6 ± 4.4%, [108]Fig. 2G). Thus,
   transient-amplifying intermediates with NeuroD1 overexpression are
   converted into NeuN-expressing neuronal-like cells in the lesion core
   after spinal cord injury, whereas non-proliferative astrocytes in the
   peri-injury distal region are not.

NeuroD1 does not induce astrocyte-to-neuron conversion at late injury phase

   To determine whether the temporal window of reactive gliosis affects
   NeuroD1-induced neuronal conversion, we delayed TAM administration to
   induce NeuroD1 expression at dpi 14 to 16 ([109]Fig. 2H), corresponding
   to a late phase of gliosis when astrocytes exhibit reduced
   proliferative activity (fig. S2, A and B). At dpi 18, in the lesion
   core of iNeuroD1^GFP mice, where reactive astrocytes marked by GFAP and
   S100β were abundant ([110]Fig. 2I), we did not detect GFP-expressing
   cells co-labeled with NeuN ([111]Fig. 2, J and K) or Ki67^+
   proliferative astrocytes ([112]Fig. 2L). Compared to neuronal
   conversion by NeuroD1 induction in the injury core during the early
   post-injury phase (dpi 5 to 8), the failure of neuronal conversion at
   later stages suggests that NeuroD1-induced astrocyte-to-neuron
   conversion is spatiotemporally restricted, occurring primarily during
   the transit-amplifying phase of reactive astrocytes within the lesion
   core.

Immature neuroblasts are formed during astrocyte-to-neuron conversion

   To track the transition of reactive astrocytes into neurons, we
   analyzed iNeuroD1^GFP mice that had been subjected to spinal cord
   injury and treated with TAM from dpi 5 to 8 ([113]Fig. 3A). At the
   early phase of injury, most of NeuroD1^GFP+/NeuN^+ cells (~98% at dpi
   10 and 78% at dpi 30) co-expressed GFAP and exhibited nonneuronal or
   glia-like morphology, with bipolar and multipolar branches ([114]Fig.
   3, B and C). In contrast, at 17 weeks post-injury (wpi), only a small
   population of NeuroD1^GFP+/NeuN^+ cells (13.5 ± 5.6%) continued to
   express GFAP. These data suggest that, between dpi 10 and 30,
   NeuroD1^GFP+/NeuN^+ cells are at an intermediate GFAP^+ stage of their
   differentiation into more neuronal-like cells.

Fig. 3. Reactive astrocytes undergo NeuroD1-induced conversion through
intermediate states to neuron-like cells.

   [115]Fig. 3.
   [116]Open in a new tab

   (A) Experimental plan to identify intermediate states of gene
   expression after injury. SCI was induced on day 0, followed by
   TAM-induction of NeuroD1^GFP expression in Aldh1l1^+ astrocytes at day
   7, and analysis at dpi 10 and 14 and at wpi 17. (B) Representative
   immunolabeling images of NeuroD1^GFP, GFAP, and NeuN in the lesion core
   at dpi 10 and 30 and at wpi 17. Scale bars, 50 μm and (zoomed-in
   region) 10 μm. (C) Quantification of the percentage of
   GFAP^+NeuN^+NeuroD1^GFP+ (upper graph) or GFAP^−/NeuN^+NeuroD1^GFP+
   (lower graph) cells. n = 3 mice per group; means ± SEM. ***P < 0.001,
   one-way analysis of variance (ANOVA) followed by Holm-Šídák’s multiple
   comparisons test. (D) Representative immunolabeling images of
   NeuroD1^GFP or the tdTomato reporter with DCX in the lesion core of
   iNeuroD1^GFP and control mice, respectively, at dpi 10, 14, and 24.
   Scale bars, 50 μm and (zoomed-in region) 10 μm. (E) Quantification of
   number of DCX^+ cells in lesion field of spinal cord injury. n = 3 mice
   per group; means ± SEM. n.d., not detectable. (F) Quantification of the
   percentage of DCX^+ reporter^+ cells. n = 3 mice per group;
   means ± SEM. n.d., not detectable. (G) Representative immunolabeling
   images of NeuroD1^GFP with OLIG2 and NeuN in the lesion cores from
   three animals at dpi 14. Scale bars, 50 μm and (zoomed-in region) 10
   μm. (H) Summary schematic showing that NeuroD1-induced conversion of
   reactive astrocytes to neuron-like cells occurs only in the lesion core
   and passes through an intermediate state.

   To further characterize early-stage NeuN^+/NeuroD1^GFP+ cells, we
   analyzed expression of doublecortin (DCX), a marker of transient
   immature neuroblasts during early adult neurogenesis ([117]50). At dpi
   10, ~36% of NeuroD1^GFP+ cells in the lesion core were DCX^+, whereas
   no DCX expression was detected in tdTomato^+ cells in control mice
   ([118]Fig. 3, D to F). Between dpi 14 and 24, the expression of DCX in
   NeuroD1^GFP+ cells gradually declined ([119]Fig. 3, D to F).

   Following spinal cord injury, a substantial proportion of reactive
   astrocytes within the lesion core at the early stages exhibited
   up-regulated expression of oligodendrocyte transcription factor 2
   (OLIG2) (fig. S2, C and D), which was also detected in proliferative
   reactive astrocytes (fig. S2E) ([120]51). The OLIG2 transcription
   factor is essential for the proliferation of reactive astrocytes
   ([121]51) and serves as a marker for transit-amplifying progenitors
   with neurogenic potential in the adult brain ([122]52, [123]53).
   Notably, most of NeuroD1^GFP+ cells (~63.5%) expressed OLIG2 in the
   lesion core ([124]Fig. 3G). These results suggest that
   NeuroD1^GFP+/NeuN^+ cells may arise from an OLIG2^+ amplifying
   progenitor state and that NeuroD1-mediated astrocyte-to-neuron
   conversion occurs through immature neuroblast intermediates ([125]Fig.
   3H).

NeuroD1 induces reactive astrocyte-to-neuron conversion in the brain injury
core

   To assess the effects of NeuroD1 in different CNS regions, we
   investigated the NeuroD1-mediated astrocyte fate switch in the brains
   of iNeuroD1^GFP mice by using two injury models: focal cerebral
   ischemia and stab wound injury. In the focal cerebral ischemia brain
   injury model, extensive GFAP^+ reactive astrocytes were induced
   surrounding the lesion, accompanied by cortical neuronal loss
   ([126]Fig. 4, A and B) ([127]54). When TAM was administered from dpi 5
   to 8 to induce NeuroD1^GFP expression ([128]Fig. 4A), ~62.8% of
   NeuroD1^GFP+/GFAP^+ cortical astrocytes expressed the neuronal marker
   NeuN within the lesion core by dpi 10 ([129]Fig. 4, C and D). GFAP^+
   was not co-expressed with NeuN in the distal region in the injured
   brain of iNeuroD1^GFP mice as well as both lesion core and distal
   regions of control mice ([130]Fig. 4D). Moreover, DCX was expressed by
   ~54.3% of NeuroD1^GFP+ cells, indicative of an immature neuronal state
   ([131]Fig. 4, E and F), while, among NeuroD1^GFP+/NeuN^+ cells, ~72.3%
   expressed OLIG2 ([132]Fig. 4, G and H). These data suggest that
   NeuroD1-mediated astrocyte-to-neuron conversion also occurs through
   immature intermediates in the focal cerebral ischemia injury model.

Fig. 4. NeuroD1 converts reactive astrocytes to neuron-like cells in focal
stroke and stab wound brain injury models.

   [133]Fig. 4.
   [134]Open in a new tab

   (A) Experimental scheme to induce focal stroke injury. (B) GFAP
   immunolabeling of lesion. Scale bar, 250 μm. (C) Representative images
   of GFAP, NeuroD1^GFP, and NeuN in the lesion core of iNeuroD1^GFP mice
   at dpi 10. Scale bar, 50 μm. (D) Quantification of the percentage of
   NeuN^+NeuroD1^GFP+ cells in the distal region and lesion core. n = 3
   mice per group; means ± SEM. (E) Representative images of reporter and
   DCX in control and iNeuroD1^GFP mice at dpi 10. Scale bars, 50 μm and
   (zoomed-in region) 10 μm. (F) Quantification of number of DCX^+ cells
   (top) and percentage of reporter^+ DCX^+ cells (bottom) in lesion
   field. n = 3 mice per group; means ± SEM. (G) Representative images of
   NeuroD1^GFP and OLIG2 at dpi 10. Scale bars, 50 μm and (zoomed-in
   region) 10 μm. (H) Quantification of OLIG2^+ /NeuroD1^GFP+NeuN^+ cells.
   n = 3 mice per group; means ± SEM. (I) Experimental scheme of stab
   wound (SW) injury. (J) GFAP immunolabeling of lesion core. Scale bar,
   100 μm. (K) Representative images of GFAP, NeuroD1^GFP, and NeuN in the
   lesion core of iNeuroD1^GFP mice. Scale bar, 50 μm. (L) Quantification
   of the percentage of NeuN^+NeuroD1^GFP+ cells in different region.
   n = 3 mice per group; means ± SEM. n.d., not detectable. (M)
   Representative images of NeuroD1^GFP and DCX in iNeuroD1^GFP mice.
   Scale bars, 50 μm and (zoomed-in region) 10 μm. (N) Quantification of
   number of DCX^+ cells (left) and percentage of NeuroD1^GFP+DCX^+ cells
   (right) in lesion. n = 3 mice per group; means ± SEM. (O)
   Representative images of NeuroD1^GFP and OLIG2 at dpi 10. Scale bars,
   50 μm and (zoomed-in region) 10 μm. (P) Quantification of OLIG2^+
   /NeuroD1^GFP+NeuN^+ cells. n = 3 mice per group; means ± SEM.

   The stab wound brain injury results in minimal neuron loss ([135]51,
   [136]55). In this model, we examined the NeuroD1 effects on
   glia-to-neuron conversion by administrating TAM from dpi 5 to 8 to
   induce NeuroD1^GFP expression ([137]Fig. 4, I and J). Of the
   NeuroD1^GFP+ cells in the lesion core, ~22.6% expressed NeuN ([138]Fig.
   4, K and L). In contrast, GFP and NeuN did not co-localize in regions
   distal to the stab wound ([139]Fig. 4, K and L). Additionally, we also
   observed that ~9.3% NeuroD1^GFP+ cells expressed DCX ([140]Fig. 4, M
   and N) and ~72.2% of NeuroD1^GFP+/NeuN^+ cells expressed OLIG2
   ([141]Fig. 4, O and P). These data from two additional models of brain
   injury demonstrate that ectopic expression of NeuroD1 is able to
   convert proliferating or progenitor-like reactive astrocytes in the
   lesion core into neuron-like cells in the injured brain.

Single-cell transcriptomics reveals NeuroD1-mediated neuronal expansion
within lesion core

   To comprehensively determine the cellular identities of
   NeuroD1^GFP-expressing cells following spinal cord injury, we performed
   single-nucleus RNA sequencing (snRNA-seq) of the lesion core and the
   peri-injury regions in iNeuroD1^GFP mice at dpi 10. Single cells
   dissociated from spinal cord segments including the lesion core and
   distal-injury regions from three mice were subjected to snRNA-seq using
   the 10X Genomics Chromium platform. We focused on the EGFP-expressing
   nuclei identified by the EGFP-WPRE pre-RNA transcripts after filtering
   out low-quality cells (see Materials and Methods; [142]Fig. 5A).
   Uniform Manifold Approximation and Projection (UMAP) revealed distinct
   differences in the transcriptomic features of EGFP-expressing nuclei
   isolated from the lesion core and those from the distal region
   ([143]Fig. 5, B and C). Using unbiased clustering, 176 and 288
   EGFP-expressing nuclei from the lesion core and distal-injury site
   samples, respectively, were grouped into four clusters ([144]Fig. 5, B
   and C). Cell clusters were annotated using a harmonized atlas of known
   cell types in spinal cord ([145]56). Three neural cell types were
   represented: neurons, astrocytes, and oligodendrocyte lineage cells
   ([146]Fig. 5D). Analysis of cell proportions across these clusters
   revealed notable compositional differences between the lesion core and
   distal region, with a predominance of neuronal populations in the
   lesion core ([147]Fig. 5, B and E). Gene Ontology (GO) terms of
   differentially expressed genes overrepresented in the lesion core
   compared to distal regions included terms involved in nervous system
   development, neuron projection, and neurogenesis ([148]Fig. 5F). UMAP
   plots showed that EGFP-expressing cells with neuronal gene expression
   were substantially enriched in the population unique to the lesion core
   ([149]Fig. 5G). Thus, these single-cell profiling analyses demonstrate
   that ectopic expression of NeuroD1 in reactive astrocytes in the lesion
   core results in an increase in neuronal cell populations, consistent
   with the formation of NeuroD1^GFP+/NeuN^+ neuronal cells at the lesion
   core in iNeuroD1^GFP mice.

Fig. 5. snRNA-seq data reveal NeuroD1-induced transcriptomic changes.

   [150]Fig. 5.
   [151]Open in a new tab

   (A) Illustration of tissue sampling and snRNA-seq experimental and
   sequence-based cell sorting workflow. (B) UMAP plots showing cells
   identified in the lesion core and distal region of iNeuroD1^GFP mice
   obtained at dpi 10. (C) UMAP visualization of 176 and 288 spinal cord
   nuclei sequenced from lesion core and peri-injury samples,
   respectively, color coded based on signature gene expression. (D) UMAP
   plots showing annotations of cells identified in lesion core and distal
   region. (E) Cell proportions across different clusters in the lesion
   core and distal region. (F) GO analysis of differentially expressed
   genes in the EGFP-WPRE sequence-sorted cells from lesion core compared
   with those from the distal region. (G) Gene expression patterns of
   neuronal genes and astrocyte genes notably changed in the lesion core
   relative to distal region.

NeuroD1 loss-of-function mutant cannot induce astrocyte-to-neuron conversion

   To validate that NeuroD1 neurogenic activity is necessary for the
   conversion of astrocytes into neuron-like cells in our model, we
   generated mice that inducible-express a previously described NeuroD1
   mutant that lacks the transactivation domain ([152]Fig. 6, A and B)
   ([153]57). Using an anti-NeuroD1 antibody targeting the N-terminal
   epitope for Western blot, we confirmed the expression of the truncated
   NeuroD1 protein in Aldh1l1-CreERT2; Rosa-NeuroD1-mut (iNeuroD1-mut^GFP)
   mice 2 days after TAM administration ([154]Fig. 6C). Following TAM
   treatment, brains were immunolabeled for GFAP and NeuroD1, which
   confirmed that the NeuroD1-mutant protein was ectopically expressed in
   GFAP^+ astrocytes in iNeuroD1-mut^GFP mice ([155]Fig. 6D).

Fig. 6. Astrocyte-to-neuronal cell conversion requires the neurogenetic
activity of NeuroD1.

   [156]Fig. 6.
   [157]Open in a new tab

   (A) Schematic of NeuroD1 and mutant NeuroD1(1 to 155 amino acids)
   [NeuroD1(1-55aa)]. AD1, activation domain 1; AD2, activation domain 2;
   bHLH, basic helix-loop-helix. (B) Diagram of
   CAG-loxP-STOP-loxP-NeuroD1-mutant ^GFP transgene targeted to the Rosa26
   locus to generate iNeuroD1-mut^GFP mice. (C) Western blot analysis for
   NeuroD1 and NeuroD1 mutant protein in tissues harvested 2 days after
   last TAM injection. Nd1, iNeuroD1^GFP mice; Nd1Mut, NeuroD1-mut^GFP
   mice. (D) Representative images of brain tissue stained for GFP
   (green), GFAP (magenta), and NeuroD1 (red). Scale bars, 10 μm. (E)
   Representative immunolabeling images of NeuroD1-mut^GFP+ cells at
   lesion core for NeuN and GFAP at dpi 10 and 28. Scale bars, 50 μm. (F)
   Quantification of NeuroD1-mut^GFP+NeuN^+ cells in lesion core at dpi 10
   and 28 (n = 3 mice per group; means ± SEM). n.d., not detectable. (G)
   Representative immunolabeling images for NeuroD1^GFP+NeuN^+ converted
   neuronal cells (arrows) and NeuroD1^GFP-NeuN^+ preexisting neurons
   (arrowheads) of iNeuroD1^GFP mice at dpi 10. Soma area was labeled by
   dashed circle. Scale bars, 50 μm and (zoomed-in region) 25 μm. (H) Box
   plot showing the range and median of somas for NeuroD1-induced
   converted neurons and preexisting neurons at dpi 10 and 30 and at wpi
   17. n = 60 cells per group from 3 animals. ***P < 0.001, Student’s t
   test.

   We next used the same spinal cord injury model to determine the effects
   of NeuroD1-mut on the fate of reactive astrocytes. We did not detect
   any NeuN expression in GFAP^+ reactive astrocytes at the lesion core of
   iNeuroD1-mut^GFP mice at dpi 10 after TAM administration from dpi 5 to
   8, whereas the iNeuroD1-mut^GFP+/GFAP^+ reactive astrocytes were
   abundant ([158]Fig. 6, E and F). To assess whether the NeuroD1 mutant
   has weak neurogenic activity rather than a complete loss of function,
   we examined iNeuroD1-mut^GFP mice at a later stage, at dpi 28
   ([159]Fig. 6E). No NeuroD1-mut^GFP+/NeuN^+ neuron-like cells were
   detected ([160]Fig. 6F). These results indicate the crucial role of
   NeuroD1 neurogenic activity in the conversion of proliferating or
   dedifferentiated reactive astrocytes into neuron-like cells.

Stemness reprogramming by SOX2 alone is ineffective in astrocyte-to-neuron
conversion

   Expression of a stemness-reprogramming factor, SOX2, was reported to
   facilitate resident astrocytes into DCX^+ neuroblasts that subsequently
   mature into synapse-forming neurons in the injured adult spinal cord
   ([161]58). To investigate whether ectopic SOX2 expression could induce
   neuronal conversion from reactive astrocytes, we generated a
   TAM-inducible transgenic knock-in line for SOX2 expression carrying an
   mCherry reporter (fig. S3A). We cross-bred the inducible
   SOX2-expressing transgenic mice with Aldh1l1-CreERT2 mice to produce
   Aldh1l1-CreERT2; Rosa-Sox2 mice, hereafter called iSox2^mCherry mice,
   and validated that mCherry^+ cells co-expressed SOX2 in the spinal cord
   followed by TAM administration (fig. S3B). Next, we performed the same
   dorsal hemisection spinal cord injury (day 0) and injected TAM from dpi
   5 to 8 in iSox2^mCherry mice (fig. S3C). In contrast to iNeuroD1^GFP
   mice, in the iSox2^mCherry mice, only a small fraction (less than 3%)
   of mCherry^+NeuN^+ cells were detectable in the lesion core region at
   dpi 10, with no mCherry^+ cells showing NeuN^+ expression at dpi 23
   (fig. S3, D and E). Similarly, very few mCherry^+DCX^+ cells were
   observed in the lesion core region at dpi 11 (fig. S3F). Thus, compared
   to NeuroD1, ectopic expression of SOX2 does not effectively induce
   NeuN^+ neuronal cell trans-differentiation from proliferating or
   dedifferentiated reactive astrocytes in the mouse model after spinal
   cord injury.

Neuronal cells resulting from NeuroD1 induction are not physiologically
functional

   Even when using NeuroD1 with neurogenic activity, the induced neuronal
   cells displayed smaller soma sizes compared to preexisting neurons
   ([162]Fig. 6, G and H), suggesting a potential deficit in electrical
   activity and integration into neural circuits ([163]59). This
   morphological feature prompts us to investigate the
   electrophysiological characteristics of these induced neurons.

   To evaluate the electrophysiological properties of NeuroD1^GFP+/NeuN^+
   cells, we performed whole-cell patch-clamp recordings of these cells in
   the lesion cores of acute ex vivo spinal cord slices and filled the
   recorded cells biocytin ([164]Fig. 7A). Together with avidin staining
   of biocytin-filled NeuroD1^GFP cells, immunostaining confirmed that
   they were positive to neuron marker NeuN ([165]Fig. 7B). Only GFP^+
   cells positive to avidin and NeuN were analyzed. We evaluated the
   generation of action potentials at three time points: (i) wpi 3, the
   early conversion stage; (ii) wpi 9, the stage in previous studies when
   mature neurons were detected ([166]30); and (iii) wpi 17, a later
   stage.

Fig. 7. NeuroD1-induced neuronal cells are not functionally mature.

   [167]Fig. 7.
   [168]Open in a new tab

   (A) Schema of patch-clamp recording in spinal cord slices, along with a
   differential interference contrast image of a NeuroD1-induced cell
   filled with avidin. Scale bar, 25 μm. (B) Representative immunolabeling
   images of recorded (iNeuroD1^GFP+/avidin^+) cells for NeuN. Scale bar,
   25 μm. (C) Representative voltage responses to step current injections
   of NeuroD1^GFP+ NeuN^+ cells at wpi 3, 9, and 17. (D) Phase plots (top,
   voltage; bottom, slope) of NeuroD1^GFP+ NeuN^+ cells at wpi 3, 9, and
   17. dv/dt, the rate of change of membrane voltage over time. (E)
   Representative repetitive action potentials in response to depolarizing
   current injection of a preexisting neuron. (F) Pie chart of the
   percentage of recorded NeuroD1^GFP+NeuN^+ neuron-like cells (n = 27
   cells) with different firing patterns at wpi 3. (G) Representative
   whole-cell currents of recorded cells in the absence and presence of 1
   μM tetrodotoxin (TTX) at wpi 3. Note the blockade of the inward Na^+
   currents in the presence of TTX. (H) Representative immunolabeling
   images of NeuroD1^GFP+ cells with vGAT and Map2 at wpi 17. Scale bars,
   10 μm. (I) Quantification of synaptic puncta in each NeuroD1-induced
   neuron-like cells and preexisting neurons. n = 3 mice per group; ≥60
   cells were counted; means ± SEM. *P < 0.05, Student’s t test.

   To characterize the electrophysiological profiles, we generated phase
   plots of the waveforms to quantify the action potential depolarizing
   and repolarizing slope ([169]Fig. 7, C and D). At the early stage wpi
   3, the NeuroD1^GFP+/NeuN^+ cells exhibited heterogeneous
   electrophysiological properties. Approximately 84% of the
   NeuroD1^GFP+cells retained glia-like membrane properties without action
   potential generation during stepwise current injections, while 12% of
   cells generated only single spikelet when subjected to strong current
   stimulation (>500 pA, 500 ms in duration), and 4% exhibited a smaller
   spikelet followed by sustained depolarization ([170]Fig. 7, C, D, and
   F). This is in contrast to the preexisting neurons that had repetitive
   action potentials evoked by depolarizing current steps in current-clamp
   mode ([171]Fig. 7E). These neuron-like cells exhibited a resting
   membrane potential ~−70 mV with no signs of instability or abnormal
   depolarization, suggesting that they are healthy cells. Consistently,
   sodium currents, which are essential for generating action potentials,
   were detected in the NeuroD1^GFP+/NeuN^+ cells and verified through
   tetrodotoxin blockade ([172]Fig. 7G). Given that sodium currents are
   absent in normal astrocytes ([173]60), the electrophysiological
   properties of the NeuroD1^GFP+/NeuN^+ cells resemble immature neurons.

   Similarly, at wpi 9, we failed to detect repetitive action potentials
   in NeuroD1^GFP+/NeuN^+ cells, and these cells exhibited a weak single
   spikelet ([174]Fig. 7, C and D). Even at the later stage wpi 17,
   NeuroD1^GFP+/NeuN^+ cells had mainly nonneuronal electrophysiology
   ([175]Fig. 7, C and D). Although a set of NeuroD1^GFP+/NeuN^+ cells
   generated a smaller spikelet, the depolarizing responses were
   substantially greater than that of typical neurons ([176]Fig. 7C).
   These observations indicate that the neuronal-like cells converted from
   astrocytes by expression of NeuroD1 do not have electrophysiological
   properties of mature neurons.

   Last, we further compared the circuit integration between
   NeuroD1-converted neuron-like cells and preexisting neurons. For the
   converted NeuroD1^GFP+ neuron-like cells in the lesion core, we
   observed that MAP2^+ cells with large somas were surrounded by synaptic
   markers vGAT (vesicular γ-aminobutyric acid transporter) and vGlut2
   (vesicular glutamate transporter 2) ([177]Fig. 7H). However, the
   numbers of synapses were much lower in the converted neurons than in
   preexisting neurons ([178]Fig. 7, H and I). Together, although
   transient amplifying reactive astrocytes in the lesion core can be
   converted into neuronal-like cells, these converted neurons have
   limited synaptic connectivity and neuronal activity.

DISCUSSION

   Neurodegenerative diseases and CNS injuries are challenging to treat
   due to the limitation of regenerative capacity of neurons in the adult
   brain and spinal cord ([179]1, [180]2). A potential therapeutic
   approach is direct in situ cellular reprogramming, in which endogenous
   nonneuronal cells, such as astrocytes or microglia, are converted into
   neurons in vivo ([181]61, [182]62). Astrocytes are the most common and
   abundant cells in the CNS, which makes them a potential cell source for
   reprogramming in neuronal replacement therapy ([183]14–[184]16,
   [185]63). Previous studies with viral vectors as delivery vehicles
   reported that ectopic expression of NeuroD1 can convert reactive
   astrocytes into mature and functional neurons under various CNS injury
   and disease conditions ([186]28–[187]35). However, recent studies have
   raised concerns regarding the authenticity of the astrocytic origins of
   converted neurons, particularly made using AAV or lentiviral vectors
   carrying a GFAP mini-promoter for the delivery of NeuroD1
   ([188]37–[189]40). Evidence suggests that the human GFAP mini-promoter
   used in viral vectors exhibits mis-expression in preexisting neurons
   when used to drive the expression of neurogenic transcription factors
   like NeuroD1 ([190]39, [191]64, [192]65). This results in the
   expression of NeuroD1 and its reporter not only in astrocytes but also
   in endogenous neurons, confounding conclusions about the authentic
   neuronal conversion of resident astrocytes. In this study, we aimed to
   address these controversies by developing inducible, lineage-traceable
   transgenic mouse models to precisely evaluate astrocyte-to-neuron
   conversion in vivo. Our findings from multiple inducible transgenic
   mouse models provide compelling evidence that ectopic expression of
   NeuroD1 in resident reactive astrocytes following acute spinal cord or
   brain injury induces the formation of neuron-like cells specifically
   within the lesion core. However, in contrast to previous reports, our
   results also uncover previously unrecognized functional limitations in
   the converted neurons.

Inducible genetic models demonstrate the authenticity of astroglia-to-neuron
conversion

   The authenticity for the astroglia-to-neuron is hotly debated
   ([193]37–[194]40). Our inducible genetic animal models offer definitive
   spatiotemporal clarity and reproducibility that viral systems lack.
   With spatiotemporal cell lineage mapping at pre- and post-injury stages
   using the inducible transgenic knock-in mouse models, our findings
   resolve the controversies regarding NeuroD1-induced astrocyte-to-neuron
   conversion. Our data provide compelling evidence that the ectopic
   expression of NeuroD1 in astrocyte lineage cells after injury to the
   spinal cord or brain leads to the formation of neuron-like cells
   specifically in the lesion core. By integrating single-cell
   transcriptomics, loss-of-function mutants, and spinal cord and brain
   injury models, we provide molecular and cellular evidence to clarify
   the astrocytic identity of NeuroD1-mediated neuronal conversion. TAM
   administration may influence estrogen receptor–mediated gene
   expression, potentially affecting neural progenitor cell fate during
   spinal cord injury ([195]66, [196]67). To control for this, we used
   Ai9-tdT; Aldh1l1-CreER mice and administered TAM in the same manner as
   in NeuroD1; Aldh1l1-CreER mice. Notably, no NeuN expression was
   observed in these control animals following TAM treatment, indicating
   that TAM alone does not induce neuronal conversion from
   reporter-positive astrocytes ([197]Fig. 1F and fig. S1), unlike ectopic
   expression of NeuroD1. Thus, TAM treatment alone does not promote
   astrocyte-to-neuron conversion in the spinal cord injury model,
   although we cannot fully exclude the possibility that TAM affects
   estrogen receptor–mediated gene expression in ways that may influence
   NeuroD1-driven neuronal fate conversion.

Astrocyte-to-neuron conversion is spatiotemporally restricted

   In contract to the previous report of the neuronal conversion from
   nonreactive astrocytes ([198]42), we found that NeuroD1-mediated
   neuronal conversion does not occur in physiologically normal,
   nonreactive astrocytes in the intact spinal cord and brain. In
   addition, ectopic expression of NeuroD1 is unable to
   trans-differentiate the non-proliferative reactive astrocytes into
   neurons in the peri-injury region. In contrast, within the lesion core,
   most of NeuroD1-expressing cells adopt neuron-like characteristics,
   suggesting that NeuroD1-mediated neuronal conversion from reactive
   astrocytes is region specific. Notably, this conversion does not occur
   beyond the early injury phase, underscoring a specific time window
   after injury for NeuroD1-induced astrocyte reprogramming. Thus, our
   findings suggest that the astrocyte-to-neuron conversion is restricted
   to a specific time window after injury and to the core region of the
   injured spinal cord and brain, providing previously unrecognized
   insights into the contextual dependencies of NeuroD1-mediated
   glia-to-neuron reprogramming.

NeuroD1-mediated astrocyte-to-neuronal conversion is cell-state dependent

   Our findings demonstrate that NeuroD1 promotes neuronal
   trans-differentiation only from a distinct reactive astrocyte
   transit-amplifying state in the injury core, in contrast to the
   previously reported all states of astrocytes including nonreactive
   astrocytes ([199]42). We found that Aldh1l1^+ mature astrocytes can be
   dedifferentiated or transitioned into OLIG2-expressing
   transit-amplifying intermediates after acute injury. The
   NeuroD1-mediated fate conversion mainly targets these proliferative or
   transit-amplifying intermediates such as those expressing OLIG2.
   Notably, OLIG2 is critical for reactive astrocyte proliferation
   ([200]51) and is expressed in the transit-amplifying (type C)
   progenitors for neurogenesis in the adult brain ([201]52, [202]53).
   This suggests that transit-amplifying reactive astrocytes formed in the
   core after acute injury may impart reprogramming competence to the
   NeuroD1-mediated neuronal cell conversion process. In contrast, in
   regions distal to the injury or during the late phase of injury, where
   reactive astrocytes remain in a non-proliferative state, NeuroD1
   expression fails to induce neuronal conversion. Thus, our findings
   reveal a previously unrecognized spatiotemporal and cell-state–specific
   conversion of resident reactive astrocytes to neurons by ectopic
   NeuroD1 post-injury in both spinal cord and brain.

Transit-amplifying intermediates represent a unified mechanism for neuronal
conversion

   Our spatiotemporally specific lineage analyses identify the
   intermediate states such as BrdU^+NeuN^+ or OLIG2^+NeuN^+ transitional
   cells, suggesting that the conversion of astrocytes into neuron-like
   cells is contingent on the transit-amplifying state of reactive
   astrocytes within the lesion core. Consistently, retroviral vectors
   carrying pro-neurogenic factors, which selectively transduce mitotic
   cells, are sufficient to promote glia-to-neuron conversion after injury
   ([203]33, [204]68), although these retroviruses may potentially target
   injury induced, dividing NSCs or microglia for neuronal reprogramming.
   In addition, alteration of cell extrinsic signaling such as Notch
   signaling can program astrocytes to a proliferative ground state to
   facilitate glia-to-neuronal fate conversion ([205]20–[206]22, [207]68).
   We showed that mature astrocytes transition to the transit-amplifying
   intermediate progenitor state in the injury core, rendering them
   permissive to NeuroD1-mediated neuronal trans-differentiation.
   Consistently, previous studies have shown that mature astrocytes
   proliferating after injury undergo transcriptional dedifferentiation
   and reprogramming into a transient progenitor-like state ([208]48) and
   that astrocytes isolated from injured adult cerebral cortex during this
   time period can give rise to neurospheres with pluripotent potential
   that are able to generate neurons in vitro, whereas mature astrocytes
   from adult uninjured cortex do not have this potential ([209]69,
   [210]70). These findings further support the conclusion that it is not
   surprising that the forced ectopic over-expression of a powerful
   fate-directing transcription factor like NeuroD1 in the
   dedifferentiated, proliferating, or progenitor-like reactive astrocytes
   in the immediate vicinity of injuries would drive them to differentiate
   into immature neuronal-like cells. Together, these observations suggest
   transit-amplifying intermediates as a convergent mechanism of
   cell-state plasticity underlying in vivo glia-to-neuron conversion.
   Thus, by shifting the mechanistic framework from direct conversion of
   astrocytes toward transit-amplifying intermediates, our study reveals a
   unified mechanism for in vivo reprogramming and provides a
   translational potential by leveraging transit-amplifying intermediates
   to enhance neuronal reprogramming for CNS repair.

Stem-cell reprogramming alone is not sufficient to promote neuronal
trans-differentiation

   To further validate the specificity and necessity of NeuroD1 neurogenic
   activity, we explored the effects of a transcription-inactive NeuroD1
   mutant. We found that a transcription-inactive NeuroD1 mutant failed to
   induce astrocyte-to-neuron conversion, suggesting the specificity and
   necessity of NeuroD1 neurogenic activity in the process. This contrasts
   with AAV-GFAP mini-promoter–mediated mis-expression of a NeuroD1
   mutant, which induces reporter expression in preexisting endogenous
   neurons ([211]39), further indicating the technical issues introduced
   with this mode of delivery. These observations underscore the
   importance of using functionally active NeuroD1 for the glia-to-neuron
   conversion process.

   Furthermore, to investigate the effects of the stem cell-reprogramming
   factor SOX2, previously reported to induce glia-to-neuron
   trans-differentiation with viral vectors ([212]58), we found that
   transgenic expression of SOX2 could not effectively convert reactive
   astrocytes into NeuN^+ neurons in the lesion core in our transgenic
   animal systems. These observations suggest that expression of a
   stemness-reprogramming factor SOX2 alone is insufficient to promote
   neuronal trans-differentiation, but it may facilitate or cooperate with
   other neurogenic factors to induce neurogenesis ([213]71).

Neurons derived from NeuroD1-expressing astrocytes are not functionally
mature

   Contrary to previous reports ([214]28–[215]36), our
   electrophysiological analyses unexpectedly revealed that neuronal-like
   cells derived from NeuroD1-induced astrocyte conversion lacked the
   electrophysiological properties of mature neurons. Notably, the soma
   size of NeuroD1-induced neurons derived from astrocytes was smaller
   than that of normal neurons ([216]Fig. 6, G and H). These data suggest
   that, while NeuroD1 can initiate the reprogramming of reactive
   astrocytes into neuron-like cells, these cells do not reach full
   maturity and functionality. Moreover, their integration into existing
   neural circuits was limited, as they formed fewer synaptic connections
   compared to preexisting neurons. Although the impact of NeuroD1
   expression levels and duration on the functionality of the converted
   neurons remains to be determined, we observed that NeuroD1-converted
   cells did not develop the functional properties of mature neurons even
   after 17 weeks post-NeuroD1 induction. Thus, optimizing the expression
   timing, regional targeting, and cell-state specificity of neurogenic
   factors is critical for promoting functional maturity and integration
   of these converted neurons, ultimately enhancing their therapeutic
   efficacy for CNS injuries and neurodegenerative diseases.

MATERIALS AND METHODS

Animals

   NeuroD1, NeuroD1(1 to 155 amino acids), and Sox2-inducible knock-in
   mice were crossed with Aldh1l1-CreERT2 mice (JAX strain no. 031008) to
   generate iNeuroD1, iNeuroD1-mut, and iSox2 mice. Rosa-tdTomato mice
   (Ai9, JAX strain no. 007909) were crossed with Aldh1l1-CreERT2 mice to
   generate Control (Ai9; Aldhl1l-CreERT2) mice. Both conditional single
   and double transgenic mice were used in our experiments. Both male and
   female mice were used in this study, with no sex-specific phenotypic
   differences observed. All mice used in experiments were maintained in a
   pathogen-free vivarium with a 12-hour light/dark cycle with free access
   to normal chow food and water. All animal experiments were approved by
   the Institutional Animal Care and Use Committee of Children’s Hospital
   of Fudan University and Cincinnati Children’s Hospital Medical Center
   ([2025]62).

Generation of inducible transgenic knock-in mice by CRISPR-Cas9 system

   The CAG-loxP-stop-loxP-NeuroD1 or NeuroD1(1 to 155 amino
   acids)–EGFP–WPRE (Woodchuck Hepatitis Virus Posttranscriptional
   Regulatory Element) were inserted into the Rosa locus. Biocytogen’s
   CRISPR-Cas9–based EGE system was used to prepare gene knock-in mice.
   Single guide RNAs (sgRNAs) are designed near the genomic insertion
   site, with targeting vectors’ 5′ and 3′ homology arms being 1812 base
   pairs (bp) and 1779 bp each. Guide RNAs were screened for on-target
   activity use UCA (Universal CRISPR Activity Assay), an sgRNA activity
   detection system developed by Biocytogen, and validated sgRNAs and Cas9
   were transcribed in vitro. Both the Cas9 mRNA and the sgRNAs were
   microinjected, mixed, and co-injected into the cytoplasm of
   one-cell-stage fertilized eggs. After injection, surviving zygotes were
   transferred into oviducts of Institute of Cancer Research (ICR) albino
   pseudopregnant females. To generate inducible iSox2^mCherry knock-in
   mice, a donor vector was constructed in which the mCherry-Sox2-SV40 pA
   was in the reverse orientation and flanked by Lox66 and Lox71, allowing
   the one-way inversion upon expression of Cre. The transgene was put
   under the CAG promoter and protected by an insulator from the unwanted
   promoter activity on the 3′-side. The vector was targeted to the Col1a1
   safe harbor locus with the aid of Cas9-gRNA that cleaves the target
   site. Pups were born and genotyped by PCR screening. Progeny from three
   founder mice carrying individual transgene gave rise to the same
   phenotype. The data presented are derived from the progeny of a single
   transgenic knock-in line.

Spinal cord injury

   In our experiments, we performed the dorsal hemisection SCI model in
   adult mice (~8 to 10 weeks). All mice were anesthetized with 4%
   isoflurane until unconscious followed by 2% isoflurane during surgery.
   After exposing the spinal vertebrae at the level of T9-T12, meningeal
   tissue in the intervertebral space was cleared to minimize fibroblast
   invasion into the lesion core. Notably, we did not perform the
   laminectomy but only separated the intervertebral space for exposure to
   reduce post-injury tissue adhesion. Then, the hemisection was created
   from midline to lateral side by using a Sharpoint Stab Knife (15°,
   straight), extending toward the ventral surface of the cord. After the
   surgery, all mice were placed back to cages equipped with clean bedding
   and 37°C heating blanket to await recovery. The date of injury was
   denoted as dpi 0.

Stab wound injury

   In our experiments, we performed the stab wound model in adult mice (~8
   to 10 weeks). All mice were anesthetized with 4% isoflurane until
   unconscious followed by 2% isoflurane during surgery. An incision was
   made along the midline of the scalp, the periosteum was cleared, and
   bregma was exposed. A 1-mm hole was drilled into the skull through a
   1-mm aperture centered 1 mm lateral from bregma. Stab wounds were
   placed into the somatosensory cortex (bregma from −0.9 to −2.7 mm,
   latero-lateral 1 mm) using Sharpoint Stab Knife. After the surgery, all
   mice were placed back to cages equipped with clean bedding and 37°C
   heating blanket to await recovery. The date of injury was denoted as
   dpi 0.

Photothrombotic stroke

   Photothrombotic stroke injury was performed in adult mice (~8 to 10
   weeks). All mice were anesthetized with 4% isoflurane until unconscious
   followed by 2% isoflurane during surgery. An incision was made along
   the midline of the scalp, the periosteum was cleared, and bregma was
   exposed. The skull was illuminated with 532-nm laser for 12 min through
   a 3-mm aperture centered 2 mm lateral from bregma beginning 5 min after
   intraperitoneal injection of rose bengal (15 mg/ml) into mice at 150
   mg/g. After the surgery, all mice were placed back to cages equipped
   with clean bedding and 37°C heating blanket to await recovery. The date
   of injury was denoted as dpi 0.

TAM and BrdU administration

   TAM (Sigma-Aldrich, T5648) was dissolved in 100% ethanol and then
   diluted in Core oil (MedChemExpress, HY-Y1888) to a final concentration
   of 20 mg/ml. TAM was intraperitoneally injected into mice at 100 mg/kg
   (~100 μl) once per day. BrdU (Sigma-Aldrich, B5002) was dissolved in
   double-distilled water in 10 mg/ml and preserved at −20°C from light.
   Consist BrdU labeling was performed by intraperitoneally injecting BrdU
   (50 mg/kg) for labeling reactive proliferative glia before the TAM
   injection. The time points of injections for each experiment are
   described in Results, illustration in figures, or figure legends.

Tissue processing and immunohistochemistry

   Mice were anesthetized with Avertin and perfused with
   phosphate-buffered saline (PBS) followed by 4% paraformaldehyde (PFA).
   Brains and spinal cords were isolated and post-fixed for 2 hours with
   4% PFA at 4°C and dehydrated in 25% sucrose at 4°C until they were
   completely settled. Samples were frozen in Tissue-Tek optimum cutting
   temperature compound (Sakura, catalog no. 4583) with dry ice and
   cryosectioned 10 μm thick using a sliding microtome (Leica). For
   immunostaining, cryosections were blocked with blocking solution (5%
   normal donkey serum and 0.2% Triton X-100) for 1 hour at room
   temperature (RT). The sections were then incubated with primary
   antibodies in the same blocking solution at 4°C overnight. For BrdU
   staining, cryosections were preprocessed with heat-induced antigen
   retrieval for 10 min at 95°C in sodium citrate buffer (10 mM, pH 6) and
   cooled for 15 min at RT. The primary antibodies used for
   immunofluorescence were goat anti-GFP (Novus Biologicals, catalog no.
   NB100-1770; RRID: AB_523903; 1:1000), chicken anti-GFAP (Millipore,
   catalog no. AB5541; RRID: AB_177521; 1:1000), rabbit anti-NeuN
   (Millipore, catalog no. ABN78; RRID: AB_10807945; 1:400), rabbit
   anti-NeuroD1 (Abcam, catalog no. ab109224; RRID: AB_10861489; 1:200),
   rabbit anti-OLIG2 (Abcam, catalog no. AB9610; RRID:AB_570666; 1:1000),
   goat anti-OLIG2 (NOVUS, catalog no. AF2418; RRID:AB_2157554; 1:40), rat
   anti-PDGFRα (BD Biosciences, catalog no. 558774; RRID: AB_397117;
   1:100), rabbit anti-Iba1 (FUJIFILM Wako Shibayagi, catalog no.
   019-19741; RRID: AB_839504; 1:1000), rat anti-BrdU (Abcam, catalog no.
   ab6326; RRID: AB_305426; 1:500), rabbit anti-S100β (Proteintech,
   catalog no. 15146-1-AP; 1:200), rabbit anti-DCX (Cell Signaling
   Technology, catalog no. 4604; RRID: AB_2276960; 1:800), rabbit
   anti-MAP2 (Abcam, catalog no. ab35454; RRID: AB_776174; 1:1000), mouse
   anti-vGlut2 (Abcam, catalog no. ab79157; RRID: AB_1603114; 1:500), and
   chicken anti-vGAT (Synaptic Systems, catalog no. 131006; RRID:
   AB_887872, 1:500). After brief washing with PBS (for 5 min, three
   times), the sections were incubated with secondary antibodies for 2
   hours at RT. The secondary antibodies used in this study were Alexa
   Fluor Cy2-, Cy3-, or Cy5- conjugated donkey immunoglobulin G (IgG)
   against rat, rabbit, mouse, goat, chicken (Jackson ImmunoResearch;
   1:500). Nuclei were co-stained with 4′,6-diamidino-2-phenylindole.
   Images were captured using confocal microscopy (Leica TSC SP8).

Western blot analysis

   Tissues were isolated and homogenized in radioimmunoprecipitation assay
   buffer (Cell Signaling Technology, catalog no. 9806) supplemented with
   protease inhibitor cocktail (Roche, catalog no. 11873580001),
   phosphatase inhibitor (Roche, catalog no. 4906837001), and
   phenylmethylsulfonyl fluoride (Sangon Biotech, catalog no. 1861274).
   Lysates were quantified with the Beyotime’s BCA Protein Assay Kit.
   Equal amounts of these lysates (25 mg per lane) were resolved with a
   12% bis-tris gel, transferred to a 0.45-μm polyvinylidene difluoride
   membrane (EMD Millipore, catalog no. HATF00010) by the electrophoretic
   transfer method. The membranes were blocked in 5% nonfat powdered milk
   [in 0.2% TBST (Tris-buffered saline with Tween 20)] for 1 hour and
   incubated with primary antibody at 4°C overnight. The primary
   antibodies used in this study were rabbit anti-NeuroD1 (Abcam, catalog
   no. ab109224; 1:1000) and mouse anti–β-actin (ORIGENE, catalog no.
   TA811000; 1:2000). After washing the blot membrane with 0.2%TBST (10
   min, three times), the membranes were incubated with horseradish
   peroxidase–conjugated anti-rabbit or anti-mouse IgG (Jackson
   ImmunoResearch; 1:20,000) in 0.2%TBST for 1 hour at RT. After washing
   with 0.2%TBST (10 min, three times), the membranes were developed with
   chemiluminescence reaction using an ECL kit (Millipore, catalog no.
   WBKLS0500) and signal detection using ChemiDoc XRS+ imaging system
   (Bio-Rad).

Acute brain slice processing

   The acute brain slices were prepared as previously described ([217]72).
   Mice were anesthetized with pentobarbital sodium (100 mg/kg,
   intraperitoneally) and perfused transcardially with ice-cold slicing
   solution, and the spinal cords were rapidly dissected and chilled in
   ice-cold sucrose–artificial cerebrospinal fluid (ACSF) solution. The
   tissues were placed on a previously trimmed 3% agar block and attached
   the ventral side with superglue to the agar in a vertical direction.
   Parasagittal slices (200 μm) were sliced with a speed of 0.12 mm/s
   using a vibratome (VT 1200 s, Leica). The slices were transferred to
   warm (34.5°C) ACSF composed of 126 mM NaCl, 2.5 mM KCl, 26 mM NaHCO[3],
   1.25 mM NaH[2]PO[4], 2 mM MgSO[4], 2 mM CaCl[2], and 25 mM dextrose
   (290 mosmol, pH 7.2) and were bubbled with 95% O[2] and 5% CO[2].

   For whole-cell patch-clamp recordings, slices were transferred to a
   submerged chamber on the infrared-differential interference contrast
   microscope (BX-51WI, Olympus). We used a multiclamp 700B amplifier
   (Molecular Devices) and a Power1401 for whole-cell recording and data
   acquisition. GFP^+ cells were identified using transmitted light and
   green fluorescence. During the recordings, borosilicate glass pipettes
   (4 to 7 megohms) were filled with an internal solution containing 130
   mM K-gluconate, 10 mM KCl, 4 mM MgATP, 0.3 mM Na[2]GTP, 10 mM Hepes, 10
   mM Na[2]-phosphocreatine, and 0.2% biocytin (288 mosmol, pH 7.29) or
   140 mM K-gluconate, 3 mM KCl, 2 mM MgCl[2], 10 mM Hepes, 0.2 mM EGTA, 2
   mM Na[2]ATP, and 0.2% biocytin (285 to 295 mosmol, pH 7.2).

   To examine the electrophysiological properties, positive current pulses
   (10 pA per step, 500 ms in duration) were injected to the cell recorded
   when it was at the resting membrane potential. To obtain the Na^+
   currents, we obtained whole-cell recording with patch pipettes filled
   with Cs^+-based internal solution containing 140 mM CsCl, 2 mM MgCl[2],
   2 mM Na[2]ATP, 10 mM Hepes, and 10 mM EGTA (287 mosmol, pH 7.2 with
   CsOH). CdCl[2] (100 μM) was included in the bath solution to block
   voltage-gated Ca^2+ currents. Cells were held at −100 mV for 80 ms and
   then depolarized to different membrane potential levels (from −90 to 60
   mV, 10 mV per step) for 50 ms. P/N subtractions were performed offline
   by using Signal software. Spike2 version 10 and Signal software version
   8 (Cambridge Electronic Design) together with MATLAB (MathWorks R2021a)
   were used for data analysis. In this study, spikelet was defined by
   three criteria: (i) the rising phase must reach a slope of at least 20
   V/s (that defines the voltage threshold), (ii) the voltage difference
   between the peak and threshold must exceed 30 mV, and (iii) the
   absolute peak voltage must surpass 0 mV. The smaller spikelet,
   representing a more immature action potential variant, retained the
   same threshold slope and peak voltage requirements but differed in two
   aspects: the peak-to-threshold voltage difference was reduced to 10 to
   30 mV, and threshold current (rheobase) exceeded 4500 pA when using
   500-ms current injection.

Single-nucleus isolation and sequencing

   Spinal cords were rapidly extruded from iNeuroD1^GFP mice (n = 3) after
   anesthesia. Each cord was separated into lesion core and peri-injury
   regions. snRNA-seq analysis was performed by the commercial service of
   Genergy Biotechnology Co. Ltd. (Shanghai, China). Nuclei were loaded
   onto the 10X Chromium Single Cell Platform (10X Genomics) at a
   concentration of 700 to 1200 nuclei per microliter (Single Cell 3′
   library and Gel Bead Kit v.3) as described in the manufacturer’s
   protocol. Generation of gel beads in emulsion (GEMs), barcoding, GEM-RT
   clean-up, complementary DNA amplification, and library construction
   were all performed as per the manufacturer’s protocol. Qubit was used
   for library quantification before pooling. The final library pool was
   sequenced on the Illumina NovaSeq X plus instrument using 150-bp
   paired-end reads.

Single-nucleus RNA-seq analysis

   Mapping to mouse mm10 genome, quality control, and read counting of
   Ensembl genes were performed by cellranger software with default
   parameter (v7.1.0). Unsupervised clustering and visualization were
   performed with R (Seurat package version 2.2). Genes expressed in fewer
   than two cells were filtered out. Cells with >200 genes and <10%
   mitochondrial genes were further processed. Then, variation coefficient
   of genes was calculated with Seurat. Dimensionality reduction of data
   was performed by using component analysis based on the first 2000
   highest variable genes. A k-nearest neighbor graph was constructed from
   Euclidean distances in the space of the first 10 significant principal
   components. Louvain modularity optimization algorithm was used to
   cluster the cells in the graph, and clustering results were visualized
   by using UMAP. Cells expressing high levels of genes encoding
   hemoglobin were discarded.

   Differential expression of each cluster was calculated using the
   “bimod” test as implemented in Seurat FindMarkers function.
   Seurat-Bimod statistical test was used to find differentially expressed
   genes between each group of cells and other groups of cells [False
   Discovery Rate (FDR) ≤ 0.05 and |log2 fold change| ≥ 1.5]. GO
   enrichment analysis for these significant differentially expressed
   genes was performed by TopGO R package, and the Kyoto Encyclopedia of
   Genes and Genomes (KEGG) pathway enrichment analysis was performed
   using the Hypergeometric test in R. Significantly enriched GO terms and
   KEGG pathways were selected by a threshold FDR (adjusted P
   value) ≤ 0.05. Cell clusters were annotated using canonical markers of
   known cell types.

Quantification and statistical analysis

   All analyses were conducted using GraphPad Prism v10 (San Diego,
   California, [218]www.graphpad.com). Data are presented as means ± SEM.
   Statistical significance between two datasets was determined using a
   two-tailed Student’s t test. For comparisons involving multiple groups,
   one-way analysis of variance (ANOVA) followed by a post hoc multiple
   comparisons test was performed. Statistical significant thresholds were
   set at *P < 0.05, **P < 0.01, and ***P < 0.001. While no randomization
   method was applied for data collection, all data were quantified in a
   blinded manner.

Acknowledgments