In Vivo Glial Trans-differentiation for Neuronal Replacement and Functional Recovery in Central Nervous System

Abstract

The adult mammalian central nervous system (CNS) is deficient in intrinsic machineries to replace neurons lost in injuries or progressive degeneration. Various types of these neurons constitute neural circuitries wired to support vital sensory, motor and cognitive functions. Based on the pioneer studies in cell lineage conversion, one promising strategy is to convert in vivo glial cells into neural progenitors or directly into neurons that can be eventually rewired for functional recovery. We first briefly summarize the well-studied regeneration-capable CNS in the zebrafish, focusing on their post-injury spontaneous reprogramming of the retinal Müller glia (MG). We then compare the signaling transductions, transcriptional and epigenetic regulations in the zebrafish MGs with their mammalian counterparts, which perpetuate certain barriers against proliferation and neurogenesis and thus fail in MG-to-progenitor conversion. Next, we discuss emerging evidence from mouse studies, in which the in vivo glia-to-neuron conversion could be achieved with sequential or one-step genetic manipulations, such as the conversions from retinal MGs to interneurons, photoreceptors or retinal ganglion cells (RGCs), as well as the conversions from midbrain astrocytes to dopaminergic or GABAergic neurons. Some of these in vivo studies showed considerable coverage of subtypes in the newly induced neurons and partial reestablishment in neural circuits and functions. Importantly, we would like to point out some crucial technical concerns that need to be addressed to convincingly show successful glia-to-neuron conversion. Finally, we present challenges and future directions in the field for better neural function recovery.

Introduction

Terminally differentiated neurons cannot proliferate, and neurogenesis in mature mammals has been suggested to be restricted to the subventricular zone in lateral ventricle and subgranular zone in the dentate gyrus of hippocampus. The mammalian CNS is thus incapable of regeneration. Although stressed neurons may interfere with normal functions, neuronal loss is the irreversible process that dysfunctions complex sensory, motor and cognitive pathways. In vivo glia-to-neuron conversion [1] represents a theoretical strategy to replace lost neurons in the late stage of CNS injuries and progressive neurodegenerative diseases. These include retinopathies such as glaucoma, macular degeneration and diabetic retinal impairments, optic nerve or spinal cord injuries, as well as some age-dependent degeneration in other CNS regions such as the Alzheimer’s disease, the Parkinson’s disease and the Huntington’s disease. In this review we first summarized the current understandings in spontaneous glia-to-neuron conversion in injured retinas of zebrafish and the involved signaling pathways. We then discussed some recent studies that achieved in vivo glia-to-neuron trans-differentiation artificially to generate new neurons in mouse models, in which the differentiated glial cells transformed into neurons directly without undergoing pluripotent or progenitor states. Importantly, the review pointed out some potential technical concerns that needed to be addressed to convincingly show the generation of new neurons through bona fide glia-to-neuron trans-differentiation, rather than some alternative experimental artifacts. Finally, we presented current challenges and future directions in this field to help us achieve favorable neural function recovery.

Cell components and histological structure of the vertebrate retina - ideal model for studying CNS injury and repair

The vertebrate retina is well-known for highly organized anatomical and histological structures. Moreover, the relatively easy accessibility makes the retina an ideal model system for studying CNS neuron injury and repair. The vertebrate retina is composed of six types of neurons and three types of glial cells [2], being laminated into three histological nuclear layers. The rod and cone photoreceptors, located in the outer nuclear layer (ONL), function to transform light to electrochemical signals, which are next transduced via the bipolar cells, located in the inner nuclear layer (INL), toward the retinal ganglion cells (RGCs). The horizontal cells and the amacrine cells play essential roles in signal integration among photoreceptors and RGCs respectively. The RGCs, within the ganglion cells layer (GCL), form long projecting axons which are bundled into the optic nerves that exit the left and right retinas, laterally crossing each other at the optic chiasm and eventually connecting with cerebral post-synaptic targets at the dorsal lateral geniculate nucleus (dLGN) or the midbrain superior colliculus (SC). Müller glia (MG) is the most predominant glial cell type that establishes scaffolding architecture for the entirety of the retina. Their somas reside in the INL and have two main processes that stretch across all layers reaching both the ONL and GCL. From the main radial processes, MGs develop horizontal branches to interact with all types of neurons, supporting neurovascular exchange, metabolic homeostasis and neuroprotection [3].

Distinctive progress and regulation of MG-dependent retinal self-repair in zebrafish versus mouse

Reprogrammed zebrafish MGs self-repair the whole retina and show state transitions in morphologies and functions

Distinct from its regeneration-incapable mammalian counterpart, zebrafish can self-repair the injured retina wholly, a process contributed by the intrinsic regenerative abilities of Müller glial cells (MGs) in response to damages. In other words, MGs can be reprogrammed to a limited number of proliferative multipotent cells with similar properties to the retinal progenitors, which then differentiate to generate all types of neurons in the retina. Specifically, instead of remaining in persistent gliosis to form tissue scar, zebrafish MGs first undergo a process called interkinetic nuclear migration, in which the nuclei of damaged MGs drift from their INL residence to the ONL, dividing asymmetrically to generate one progenitor, and then return to the INL. The newly generated retinal progenitor achieves several rounds of the interkinetic nuclear migrations and cell division back and forth across the ONL, INL and GCL to establish a small reservoir of retinal progenitors, which eventually re-differentiate into corresponding types of neurons within each retinal nuclear layer [4].

Extracellular cues regulating the MG-mediated retinal regeneration in zebrafish

Relying on the stretching processes interacting with all retinal cell types in different nuclear layers and their functions in homeostasis and phagocytosis, MGs readily respond to retinal injury after receiving signals from disrupted cell-surface connections, such as the Notch signaling [5, 6]. Zebrafish MGs in injured retinas have been shown to be both the donors and the receivers of some extracellular diffusible factors, which transduce injury signals downstream to converge on a few main signaling pathways, and mediate proliferative and neurogenic effects [3, 4, 7]. To date the donor origins have not been fully identified. One possibility is that the injury-induced growth factors and cytokines may act in a paracrine fashion, being secreted or exposed from reactive microglia or dying retinal neurons. In addition, the zebrafish MGs may act in an autocrine fashion to enhance their own reprogramming capacities. Previous studies have identified many injury-induced growth factors, including insulin [8, 9], heparin-binding epidermal-like growth factor (HB-EGF) [5, 9], insulin-like growth factor 1 (IGF-1) and fibroblast growth factor 2 (FGF-2) [9], and cytokines, such as leptin, interleukin 6 (IL-6) and interleukin 11 (IL-11) [10], ciliary neurotrophic factor (CNTF) [11, 12], tumor necrosis factor ɑ (TNF-ɑ) [6, 13]. These extracellular cues commonly act synergistically to activate MG reprogramming [6, 9, 10], suggesting that stressed or dead neurons, together with active microglia, coordinate injured zebrafish retina to cross a certain threshold to dramatically activate MG reprogramming. These injury-induced growth factors and cytokines converge downstream on some main intracellular signaling pathways, such as MAPK/pERK, PI3K, JAK/pSTAT3, Wnt/β-catenin and Hippo pathways (see recent detailed review articles [3, 4, 7]).

Intracellular epigenetic and transcriptional regulations of the zebrafish MG reprogramming

How do retina injury-induced cues orchestrate zebrafish MG reprogramming? Previous studies have shown that histone deacetylase HDAC1 is required [14, 15], suggesting that H3K27 acetylation-mediated chromatin remodeling regulate core set of genes associated with zebrafish MG proliferation and neurogenesis. Indeed, after HDAC1 knockdown in MGs the levels of many gene expression regulators, such as Ascl1a, Lin28a and let-7, have been found to be altered in zebrafish. Moreover, some DNA cytosine methylase genes have been shown to be upregulated in MGs after zebrafish retinal injury [16, 17]. These studies of global epigenetic regulators suggest that zebrafish MGs undergo a transition in the epigenome to favor the expression of a core set of genes for reprogramming and neurogenesis.

At the transcriptional level, Stat3 is the key nuclear effector downstream to the JAK/STAT cytokine or growth factor sensing pathways that is activated upon injuries [12]. Phosphorylation of Stat3 increased ubiquitously in all MGs after retinal injuries, including the subset of MGs which actively proliferates and reprograms as well as the subset of MGs maintained at resting state. Although being seemingly distributed in all subsets of MGs, Stat3 has been shown to be required for injury-activated MG reprogramming [6, 10, 13]. In addition to Stat3 activation, upregulation of another transcription factor Ascl1 is necessary for MGs to initiate the injury-stimulated spontaneous identity reprogramming and proliferation to generate neural progenitor-like cells [10, 13]. Indeed, knocking out Ascl1 in retinas of zebrafish impaired spontaneous reprogramming of MGs upon retinal injury [18]. The molecular mechanisms by which Ascl1 acts to regulate MG proliferation and neurogenesis are not fully deciphered yet. Previous studies have suggested that it might partially relate to the Ascl1-Lin28-let7 regulatory loop [4, 19], all of which are associated with stemness maintenance and cell fate differentiation (see recent detailed review [3]).

Using genetic manipulation learned from machineries in zebrafish to induce interneurons and photoreceptors in mouse retina

In significant contrast to MGs in zebrafish, MGs in mouse retina completely lose their ability to be reprogrammed to reproduce retinal cells after the injury. Based on knowledge gained from the zebrafish, great efforts have been devoted to reprogramming mouse MGs for retina repair and functional recovery. In an earlier in vivo study, overexpression of Ascl1 in mouse MGs (Table 1), together with retinal injury, successfully reprogrammed a limited number of MGs to amacrine-, bipolar- or photoreceptor-like neurons in young mice [20], suggesting that Ascl1 is also a key regulator for MG reprogramming in mice. However, during development and maturation the chromatin structure in mouse MGs may change to impair the accessibility of Ascl1 [20]. In a following study, Ascl1 overexpression was combined with histone deacetylase (HDAC) inhibitor treatment [21], which aimed at increasing the overall level of chromatin accessibility (Table 1). In order to restrict Ascl1 expression specifically in MGs without leaking to the resident old neurons, two MG-specific Cre mouse lines - solute carrier family 1 member 3 (Slc1a3) and retinaldehyde binding protein 1 (Rlbp1) were combined. NMDA treatment was selected as the injury model, which should have primarily damaged the INL neurons and eliminated the ganglion neurons. In the injured mouse retina, Ascl1 expressing MGs underwent significant morphological transitions after the injection of TSA, a HDAC inhibitor. Such combinatory treatment was named with the abbreviation ANT (Ascl1, NMDA, TSA). For MGs overexpressing Ascl1, the ANT treatment significantly increased the expression of neuronal marker orthodenticle homeobox 2 (Otx2) and reduced the expression of glial marker SRY-Box transcription factor 9 (Sox9). Interestingly, unlike the studies in zebrafish, some experimental evidence in this study suggested that the MG-to-neuron transition was direct trans-differentiation that did not go through stem cell-like states as the intermediates [21]. First, morphologically some transient cells showed the hybrid appearance in between MG glia and bipolar neurons. Second, most of the ANT-induced new-born neurons were EdU-negative, suggesting a direct lineage trans-differentiation occurred without the proliferation of MG-derived progenitors, the hallmark of zebrafish MG reprogramming. Third, in addition to the morphology and biochemical assays, single-cell RNA-seq, a more comprehensive method to profile the genetic state of a population of cells, was exploited to reveal the cellular identities of reprogrammed MGs. The results demonstrated that ANT treatment converted MGs into two clear clusters of cells away from the original glial state. The relatively large cluster of induced cells showed a transcriptome resembling the neural progenitor cells, as well as a relatively small cluster of induced cells were bipolar- or amacrine-like [21, 22]. Since the study did not sample the induced cells from different transient time points physically, neither applying pseudo-time or RNA velocity trajectory analyses, it will not be evident to conclude if the progenitor-state might be the transient intermediate antecedent to the fully conversion to neurons, or the progenitor-like state represents a distinct branch away from the neurogenic branch.

Table 1.

Studies achieved in vivo glia-to-neuron trans-differentiation in mouse models to generate new neurons

Resident glial cells	Manipulation step #1	Manipulation step #2	Induced neurons	subtypes	Partial functional recovery	Ref.1
Müller glia	Ascl1 OE2	HDAC inhibition	Bipolar neurons, amacrine cells	N.A.3	N.A.	[20-22]
Müller glia	β-Catenin OE	Otx2, Crx, Nrl OE	Rods	N.A.	Yes	[23]
Müller glia	PTB KD4		Cones	N.A.	Yes	[24]
Müller glia	PTB KD		RGCs, amacrine	Yes	Yes	[25]
Astrocytes	PTB KD		DA neurons	Yes	Yes	[26]
Astrocytes	NeuroD1 Dlx2		GABA neurons	Yes	Yes	[27]

^1.Ref.: Reference

^2.OE: overexpression

^3.N.A.: not applied

^4.KD: knockdown

Another important study showed the feasibility to convert mouse MGs to generate new rod photoreceptors in vivo with two-step gene manipulation [23] (Table 1). Aiming at manipulating MGs specifically, this study used GFAP-driven transgene expression packaged in AAVs. The first step was to overexpress β-catenin in MGs to activate the proliferative programs [23, 28]. The second step was combinatory expression of Otx2, Crx and Nrl to induce β-catenin-activated cells to differentiate into rod photoreceptors. AAV-GFAP-EGFP and AAV-Rhodopsin-tdTamato were co-injected to facilitate the visualization of MG-to-Rod transition. Three transient states were captured across the cell identity conversion process. Some manipulated MGs started by expressing rod marker rhodopsin. Next, these MG-derived cells asymmetrically divided into two daughter cells – one staying in the INL and the other one migrating to the ONL. Lastly, the cell residing in ONL could differentiate into functional rod photoreceptors, whereas the other MG-derived cell in INL returned to a progenitor-like cell state and stopped expressing rhodopsin.

In vivo glia-to-neuron trans-differentiation to generate new RGCs and long projection neurons in other CNS regions

In these aforementioned in vivo studies in mice, the feasibility to convert mouse MGs to some types of interneurons such as the bipolar and amacrine cells, as well as photoreceptors, has been shown. However, the induction of long projection neuronal types, such as RGCs, by glia-to-neuron conversion remained a difficult task until some recently published studies [25]. These long-projection neuronal types are usually crucial for reconstitution of the broken neural circuits. Compared to the induction of interneurons, it is highly likely that more regulatory barriers need to be surpassed in order to obtain not only the correct cell identities of these projection neurons, but also the biochemical momentums to develop soma-axon polarity, axonal morphogenesis and/or pathfinding. A recent study, using GFAP-driven CRISPR-CasRx system to knockdown the PTB gene in mouse MGs, has shown direct trans-differentiation from MGs to functional RGCs, which were able to project new axons back to synaptic targets and partially recovered some lost visual functions (Table 1). Besides RGCs, this study also found a small portion of MGs that converted to amacrine cells after knocking down PTB. Although the fine mechanisms have not been fully uncovered, PTB serving as a RNA-binding protein [26, 29, 30] has been shown to regulate some key microRNAs associated with the stemness maintenance and neurogenesis. Interestingly, by knocking down the same PTB gene in mouse retinas, the other study (Table 1) showed the MGs were converted to cone photoreceptors [24], rather than generating new RGCs. Besides using shRNA to knockdown PTB in the second study, these two studies seem to be highly similar in experimental designs. Thus, there remains a gap to determine if other potential differences in experimental operations caused this divergence in the outcomes of cell identity conversion after manipulating exactly the same gene. Additionally, PTB knockdown has also been shown to convert to striatal astrocyte into several subtypes of dopaminergic neurons (Table 1), which then developed long projection axons re-wiring to the striatum from the substantia nigra [26]. Obviously, independent studies from different groups are necessary to confirm if deleting PTB is indeed able to produce new RGCs by reprogramming MGs.

Cross-species multiomics predicts hubs in MG reprogramming for functional validations

It is well known that the ability of MGs to be reprogrammed to reproduce neurons lost after retinal injury is gradually reduced from zebrafish compared to chicken and completely lost in mice. As described above, after 6-week ANT treatment, most of the affected MGs were converted to bipolar-like neurons, meanwhile a small portion of the amacrine-like neurons were generated. However, the ANT treatment did not show evidence for successful conversion from MGs to RGCs or photoreceptors. What are the genetic/epigenetic barriers that prevent mouse MGs to be reprogrammed to generate RGCs? In a latest study [31] comparing the distinct transcriptomes and epigenomes across zebrafish, chicken and mouse MGs, integrative bioinformatics analyses, including bulk RNA-seq, bulk ATAC-seq and single-cell RNA-seq have been exploited to prolife the transcriptomic transitions and epigenetic barriers associated with the sharp difference between zebrafish and mouse in terms of the MG-dependent spontaneous retinal repair. The study also predicted the potential gene regulatory machineries that might drive the naturally failed transition in mouse MGs from the inert state to regeneration-capable states. With such list of candidates, Hoang et al. [31] has shown that manipulations on some targets indeed exhibited promising effects with the in vivo functional validations.

After NMDA (INL injury) or light damage (ONL injury), Hoang et al. [31] first applied bulk RNA-seq on FACS-sorted MG cells, which revealed that either zebrafish (GFAP+) or mouse MGs (GLAST+) experienced rapid and significant changes to their transcriptomes rapidly in the initial 4-10 hours after either types of injures. However, 36 hours after the injuries, MGs from injured mouse retina progressively reverted their transcriptomic states back to the quiescent state, which nearly resembled the uninjured resting mouse MGs on the PCA plot. With sharp difference, the transcriptomic states of zebrafish MGs from injured retinas consistently progressed further away from the quiescent state after the initial period (Figure 1). Single-cell RNA-seq (scRNA-seq) analysis [31], covering all types of retinal cells, was exploited on mouse or zebrafish retinas after INL or ONL injuries. The scRNA-seq has exhibited cells identified with MG markers rapidly altered their transcriptomes during the initial period after injuries in both zebrafish and mouse. Afterwards, the transition of mouse MGs failed to maintain such momentum and trajected back to the quiescence-like states. In contrast, zebrafish MGs gradually and consistently diverged away from quiescence (Figure 1). These results suggested the existence of some key regulators facilitating the injury-to-regeneration transition in zebrafish MGs, as well as some barriers blocking similar transitions in mouse MGs.

Figure 1. Comparison between zebrafish and mouse MGs in post-injury state transition and some unraveled mechanisms.

The diagram was modified from the trajectory analyses of bulk RNA-seq and scRNA-seq recently published in a cross-species study by Hoang et al. [31]. Based on transcriptomic profiles, post-injury zebrafish MGs can pass the reactive gliosis and proceed to the neural progenitor-like proliferative state and even further to the neurogenic state. On the other hand, mouse MGs can rapidly enter the reactive gliosis after injury. However, mouse MGs fail to advance to proliferation and neurogenesis, but reverse backward to the quiescent state. Some previous studies focused on specific single genes as well as some recent studies exploiting multi-omic approaches have revealed several promoting or inhibitory factors involved in the MG-to-neuron transition.

Interestingly, observed only in zebrafish and chicken MGs but not in mouse MGs, the scRNA-seq trajectory and RNA velocity analyses showed that following the entrance into a transient gliotic state characterized by genes such as protein coding gene mesencephalic astrocyte derived neurotrophic factor (manf), a smaller portion of the reactive MGs returned to the quiescence, whereas a relatively larger portion proceeded to the proliferative and potentially neurogenic branch, which was characterized by the expression of proliferating cell nuclear antigen (pcna). Although a few shared differential gene sets did exist among all three species, species-specific gene sets within upregulating patterns in zebrafish and chicken MGs were enriched for gene ontology terms (GO terms), such as DNA replication and cell cycle. However, mouse-specific GO terms were apoptosis, TNF and NF-kappa B signaling pathway, etc. Furthermore, Hoang et al. [31] have indicated that injured zebrafish MGs passed the transient gliotic state and continuingly entered the proliferative and neurogenic state, which indeed significantly overlapped with the transcriptomes of the multipotent retinal progenitor cells (RPCs) during early development. Both scRNA-seq of the developing zebrafish retinas [31] and mouse retinas [32] were exploited to characterize the core sets of genes in developing RPCs. As a result, two subtypes of RPCs have been identified - neurogenic RPCs and all other RPCs known as primary RPCs. When all subtypes of RPCs were integrated with the resting and injured adult zebrafish MGs, data has shown that either NMDA or light-damage was able to drive zebrafish MGs toward RPC-like states. Such transcriptomic rejuvenation from MGs backward to RPCs is consistent with their morphological transitions, during which the activated zebrafish MGs undergo interkinetic nuclear migration, a signature characteristic of neural progenitors [33].

In addition to the transcriptomic trajectories, Hoang et al. [31] utilized bulk ATAC-seq on injured zebrafish MGs or mouse MGs, showing that the promoter regions of some typical pro-neuronal genes, such as ascl1a and lin28a, maintained heterochromatin form during resting state in both zebrafish and mouse. However, once injured the zebrafish MGs, but not the mouse MGs, exhibited epigenetic transitions toward euchromatin and increased accessibility at ascl1a and lin28a promoter regions. This suggests some epigenetic barriers prevent mouse MGs from spontaneously reprogramming and regenerating various types of neurons as zebrafish retain can do (Figure 1). Besides illustrating the changes in chromatin landscapes, the ATAC-seq is also useful for predicting which higher hierarchical regulatory elements, especially TFs, can specifically promote the spontaneous regeneration in zebrafish retina, as well as predicting which specific machineries potentially prevent mouse MGs to act similarly. Next, Hoang et al. [31] established an advanced algorithm to integrate the differentially expressed TFs before and after injuries in zebrafish or mouse MGs, with the specific TF-binding motifs enriched from the differentially accessible chromatin regions, specifically the gene promoter regions. A list of TF candidates was then obtained, which comprehensively regulated the set of genes that specifically facilitates zebrafish MGs proliferation and neurogenesis (Figure 1). On the other hand, a list of TF candidates which specifically regulated inhibitory genes (Figure 1) during the failure of mouse MGs reprogramming was also obtained. Eventually, many targets from these candidates lists showed promising effects in the functional validations [31]. For example, the knockdown of hmga1a, yap1, myb1 or smarca5 genes, all of which were predictive candidates in the Integrated Regulatory Network Analysis (IReNA), have shown significant reduction in proliferative MGs in the spontaneous reprogramming in the zebrafish. More importantly, the conditional knockout of nuclear factor I (NFI) family of TFs in adult mouse retina, which was predicted (IReNA) to be a hub maintaining the quiescent state or reverting active MGs back to the quiescence, successfully reduced the genes expressed in resting MGs such as Glul, Rlbp1, Aqp4 and Apoe genes. Also, conditional deletion of NFI family in the adult mouse upregulated the cell cycle-associated genes Ccnd1 and Ccnd3, and most critically increasing the neurogenic pioneering TF Ascl1. Upon NMDA injury, more MGs showed increased proliferation (Edu+) in the NFI-deficient mice than in the control. An observable number of MG-derived new-born cells could be stained with the bipolar cell and photoreceptor marker Crx, as well as other neuronal markers such as NeuN and HuC/D. Preciously, in addition to the immunostaining based on random sampling, Hoang et al. [31] conducted additional scRNA-seq of the FACS-sorted MGs and MG-derived new-born cells from the NFI-deficient mice. Unlike the WT mice (control MGs), MGs from NFI-deficient mice could be activated by NMDA injury, forming a transcriptomic cluster distinct from the control MGs. A small portion of the MG-to-neuron cells exhibited amacrine or bipolar markers, whereas the relatively large portion of the activated NFI-deficient MGs were non-neurogenic and only exhibited proliferative cell-cycle regulatory marker genes.

Potential technical concerns for providing convincing evidence supporting in vivo glia-to-neuron conversion

The first concern is whether we exclusively target glial cells.

When interpreting experimental outcomes in glia-to-neuron conversion-related studies, the prevailing standard is to track and quantify to what extent the originally fluorescently labeled glial cells been converted to express specific neuronal markers. Thus, some types of misinterpretation could possibly occur by co-labeling the original resident neurons in parallel with the glial cells. This may lead to counting the resident neurons as “newborns”. In other words, have the fluorescent glial cells turned to new neurons, or have the old neurons turned fluorescent? Moreover, if the genes encoding the fluorescent protein and the reprogramming factors were co-leaked into the resident old neurons, it may increase the chance of misinterpretation. For instance, some reprogramming factors used for promoting glial-neuron transition, such as the Sox family, Lin28a/b, Ascl1 and microRNAs miR9/124 and their associated signaling pathways (e.g., MAPK/ERK, Akt, JAK/pSTAT3) also have protective effects on neurons [34, 35]. As a result, unexpected leakage of these factors into resident neurons might result in increased survival of old resident neurons rather than the production of new neurons from glial cells.

One likely cause of wrongfully targeting resident neurons is the genetic approaches used for specifically targeting glial cells. AAVs expressing certain GFAP mini promoter-driven expression vector has been mostly used to achieve exclusive glial targeting. For example, in one study mention above, AAV-GFAP-vector was used to overexpress β-catenin, some pro-rod factors, as well as the EGFP marker to label MGs [23]. In another study, AAV-GFAP-vector was used to express the CasRx editor when knocking down PTB to convert MGs to RGCs [25]. Moreover, AAV-GFAP-vector was also used to express shPTB to convert MGs to cones photoreceptors [24]. Neither AAVs nor the GFAP mini promoter are guaranteed to exclusively target glial cells. Different types of AAVs have been widely used as genetic tools to infect neurons for genetic manipulation. Therefore, the GFAP mini promoter in the aforementioned studies was the only restriction for glial specificity, which has been shown to be problematic in cell specificity, especially when driving certain fluorescent proteins [36, 37]. Moreover, GFAP has been reported to have relatively lower level of expression in neurons. The detectable expression of GFAP in several types of neurons can be further upregulated several folds upon neural injury (see public RNA-seq datasets GSE137400, GSE87046, GSE93674). Furthermore, in some circumstances the genetic manipulation(s) applied to drive glial-neuron conversion may have themselves enhanced the expression of GFAP in neurons. For example, in one previous study, conditional PTB knockout in neural stem cells led to upregulated GFAP in the cortex [38]. Although not directly tested, it is possible that some increased GFAP expression could come from neurons. If this were true, deleting PTB in neurons might initiate a positive feedback loop, leading to more shPTB and fluorescent protein expression in wrongfully targeted old resident neurons [26]. In order to minimize the misinterpretation caused by improperly labeling old resident neurons, some necessary control experiments are needed, such as examining if AAV-GFAP-driven gene can be expressed in neurons with or without injuries at different time course after infection. Nevertheless, in our opinion using AAV-GFAP overexpression as a tool to specifically target glial cells is not a reliable approach.

A better strategy is to exploit multiple mouse lines, in which the Cre recombinase is driven by glial specific marker genes promoters. Under such condition, the Cre expression is relatively lower than that with AAV-GFAP overexpression, further reducing the possibility of wrongfully expression of genes encoding fluorescence protein and/or reprogramming factors in neurons. One good example of such approach was that used by the study from Thomas A. Reh’s group [21], in which two MG-specific Cre mouse lines (Slc1a3, Rlbp1) other than the GFAP-Cre line were used. If the results obtained from both Cre lines are consistent, it would greatly strengthen the conclusion regarding glia-to-neuron conversion.

The second concern is the existence of cytoplasmic material exchange between glial cells and resident neurons.

In many previous studies (see review article in detail [39]) focusing on retinal cell transplantation studies, it was reported that transplanted stem cells could differentiate into specific retinal cells and be integrated into the injured retina. However, later studies [40-42] found out that cytoplasmic material exchange occurred between transplanted stem cells and original resident retinal cells. The materials transferred included not only fluorescence proteins, but also mRNAs, shRNAs, plasmids, peptides and proteins as large as the Cre recombinase. As a result, the observed fluorescence labeled as “new” photoreceptors were actually not from the transplanted donor stem cells, but due to the cytoplasmic material exchange of fluorescent proteins into the residual “old” resident neurons [40-42]. If the cytoplasmic material exchange is an inevitable, ubiquitous phenomenon between neighboring cells, the glia-to-neuron conversion studies may easily be misinterpreted no matter how strict the glial specific promoters were selected for expressing Cre, fluorescent proteins and/or shRNAs, etc. To minimize the misinterpretation due to the presence of cytoplasmic material exchange, additional control experiments should be required. For instance, to control the transfer of fluorescence proteins from targeted glial cells to resident neurons, specifically infecting the glial cells with only the gene encoding the fluorescence protein, without the reprogramming factors driving the glia-to-neuron conversion, should be performed, followed by examining the presence of the fluorescence protein in resident neurons at different time points.

A potential combinatory strategy to confirm glia-to-neuron conversion results

Due to the aforementioned concerns and potential resolutions, we think that no single approach could completely rule out the possibility that old resident neurons are misinterpreted to be new neurons converted from the glial cells. Therefore, we propose a combinatory strategy (Figure 2) to more convincingly validate the obtained results of glia-to-neuron trans-differentiation.

Figure 2. Diagram of transcriptomic and morphological transitions occurring across glia-to-neuron conversion.

Along with the cell identity conversion during glia-to-neuron conversion, the transcriptomes transit across distinguishable states, which can be validated with single-cell RNA-sequencing (scRNA-seq). Transient populations of cells sampled across the time course can be analyzed with trajectory analysis. In parallel with transcriptomic transition, the cellular morphology also progressively transits from a glial cell to a mature and functional neuron. To validate the generation of a glia-derived neuron, interkinetic nuclear migration, glia-neuron hybrid appearance, as well as the growing axons and growth cones can all easily be visualized across the relatively long period of time. Smarca5: protein coding gene SWI/SNF related, matrix associated, actin dependent regulator of chromatin, subfamily A, member 5. Yap1: protein coding gene Yes1 associated transcriptional regulator. Insm1a: protein coding gene INSM transcriptional repressor 1. NFI: nuclear factor one transcription factor family.

In addition to the aforementioned resolutions for wrongfully targeting resident neurons or cytoplasmic material transfer, the bona fide glia-to-neuron conversion should be a cellular process containing multiple intermediate/transient states, in morphology, cell position, and transcriptomic states (Figure 2). Therefore, capturing these sequential transient states across multiple time points should be key evidence in supporting glia-to-neuron conversion. For instance, in the retina, MG cell bodies reside in the different layers than RGCs or photoreceptors. To convert MGs into photoreceptors or RGCs, they have to migrate from the MG layer to the specific layer where the final type of retina cell is located. Imaging and capturing this MG migration would be a strong indicator. Second, during direct trans-differentiation from glial cells to neurons, the transcriptomic profile also undergoes a gradual cell type-specific change. Thus, sampling the originally targeted glial cells at different time points during the glia-to-neuron conversion for single-cell RNA-seq analysis is another ideal validation approach. The results should provide solid evidence revealing the gradual transcriptomic transitions during the glia-to-neuron conversion process (Figure 2).

Third, once the targeted glial cells convert to new neurons, the new neurons need to extend axons toward their intended synaptic targets. There is no doubt that this process is the prerequisite for re-wiring and functional recovery. In addition, this process usually lasts for a much longer time than that of the glia-to-neuron conversion process. Therefore, we think that the strongest evidence supporting the generation of new neurons from glial cells is direct observation of the process of such axon growth and pathfinding towards their targets (Figure 2), especially for long projection neurons, such as those innervating the optic nerve or the nigrostriatal pathway. The current available whole brain tissue clearing [43] and deep 3D imaging [43] techniques are well established and suitable for accomplishing this observation. Unfortunately, to our knowledge, no previous studies in the glia-to-neuron conversion field have ever tried to capture the axon growth process, a process lasting at least a month. For instance, in the study where MGs were reprogrammed into RGCs by deleting PTB [25], the results showed that axons from newly generated RGCs were able to grow from the ONL all the way toward mid-brain superior colliculus within two weeks (~7.5 mm/wk). However, in most axon regeneration studies [34, 35], the optimal multi-gene combinatory manipulations could only promote RGC axons to regrowth and reach the optic chiasm after four weeks (~1 mm/wk). Moreover, most regenerating axons reaching the chiasm could not find their way back to their targeted brain nuclei. In another study following cell transplantation strategy [44], within the transplanted highly regeneration capable P0 mouse RGCs, only a few cells could achieve effective axonogenesis. For the transplanted RGCs that did extend axons, three weeks after transplantation into adult receivers’ GCL, a very limited number of RGCs projected axons that only approached the optic nerve exit in the retina, far from reaching the chiasm or the superior colliculus. Such a great discrepancy of RGC axon regrowth between that of surviving RGCs or transplanted young RGCs and that of MG-derived new RGCs raises reasonable concerns of the aforementioned study [25]. In another recent study [26], in which deleting PTB resulted in astrocyte-to-neuron conversion, retrograde labeling virus was injected into projection terminals to show new axon growth and synaptogenesis from newly converted neurons. However, it was still possible that the retrogradely labeled neurons were not all astrocytes-derived but contained some survived old resident neurons.

Collectively, we think that to draw a convincing conclusion regarding glia-to-neuron conversion, some or all of the following validations should be completed, including 1) better controlling the glial cell specific genetic targeting, 2) capturing the intermediate cell morphologies or localizations during the trans-differentiation process, 3) mapping the progenies of glial cells undergoing fate reprogramming with classical biochemical assays against specific lineage marker genes, 4) capturing the intermediate transcriptomic profiles at different time points during the trans-differentiation process with single-cell RNA-seq, and 5) capturing the axon growth process from the newly glial cells-derived new neurons with tissue clearing and 3D imaging.

Future perspectives

The glia-to-neuron trans-differentiation provides a promising pipeline to induce new neurons within the residential region. The generation of an adequate number and the correct subtypes of new neurons definitely stands as the prerequisite for the functional recovery of neural circuitry and activities. In one recent study [26], it was reported that targeting glial cells at different brain regions led to the generation of different types of neurons specific for that region, indicating some unknown local environmental cues might function to direct the neuronal types. Therefore, understanding the cellular and molecular mechanisms by which local environment directs glia-to-neuron trans-differentiation should be an obvious next step.

Moreover, circuitry re-wiring and optimal functional recovery also requires axon growth, accurate pathfinding, correct synaptogenesis, and remyelination. Using the visual pathway as an ideal model, newly induced RGCs need to extend axons radially within the GCL and find the optic nerve exit region. Once entering the optic nerve, the myelin debris from degenerated axons makes the optic nerve a hostile environment for axon growth. Based on studies (see Review article in details [35]) from the optic nerve regeneration field, the optic chiasm especially in the adult mouse is the next growth resistant region. Data from optic nerve axon regeneration commonly showed when reaching the chiasm, the growth cones may be reversed or guided toward the opposite eye rather than passing the chiasm to reach their upper-level targets. In the developmental phase of CNS in mice, the attraction and repulsion of growth cone guidance, for example at the optic chiasm, is tightly regulated by many guidance cues, which may not be maintained in the adult mouse brain [45]. More studies will be required to identify optimal guidance cues and machineries to guide the newly induced neuronal axons in the adult brain to their synaptic targets.

Furthermore, it is not a spontaneous process for the trans-differentiation-induced new neuronal somas to develop and extend axons. First, it remains unknown if the manipulations for converting glial cells to neurons would still support or even inhibit axon growth. Second, based on neurodevelopment studies, the synaptogenesis and axon pruning usually require the cessation of active axonal growth [45, 46]. Therefore, in order to achieve cell identity conversion and then the functional circuit re-wiring, precise temporal control of gene expression is needed. Considering the multiplex of cellular programs directing glial-neuronal trans-differentiation, a precise, cue-directed genome editing toolbox is necessary. Advances in the CRISPR/Cas system have opened the path to engineering an inducible editing toolkit. Current inducible CRISPR/Cas systems control either the expression of Cas effectors and/or sgRNAs (pre-transcriptional) or the activation of dormant Cas effectors (post-transcriptional) along the temporal axis. These systems can be activated through a myriad of mechanisms, such as various drugs [47], blue light [48, 49], and far red light [50]. Recently, Chylinski et al. have devised the CRISPR-Switch system, based on loxP-sgRNA scaffolds that can either be induced (CRISPR-Switch-ON) or restricted (CRISPR-Switch-OFF) by Cre recombinase activity [51]. Importantly, the CRISPR-Switch system can be applied to consecutively target multiple loci by combining the CRISPR-Switch-ON and Switch-OFF systems with shared loxP sites (CRISPR-Switch-OVER), allowing for an efficient method of controlling genome editing across the temporal axis. We think that inducible CRISPR/Cas editing systems can be applied to the paradigm of glial-neuronal trans-differentiation. With multiple methods to precisely activate and terminate genome editing, the cellular and molecular processes driving glial-neuronal trans-differentiation can be replicated. For instance, considering the complex genes at separate loci necessary for trans-differentiation, expression of certain loci required earlier in the trans-differentiation timeline can be mediated through traditional toolkits (e.g. AAV-siRNA/OE), and later steps of the timeline can be activated by inducible CRISPR/Cas cues (e.g. light, ligands).

Abbreviations:

MG

Müller glia

RGC

retinal ganglion cell

ONL

outer nuclear layer

INL

inner nuclear layer

GCL

ganglion cell layer

MAPK

mitogen-activated protein kinase

ERK

extracellular signal-regulated kinase

PI3K

phosphoinositide 3-kinase

JAK

Janus kinase

STAT3

signal transducer and activator of transcription 3

Ascl1

Achaete-Scute family BHLH transcription factor 1

Lin28

LIN-28 family RNA-binding protein

PTB

polypyrimidine tract binding protein 1

NMDA

N-Methyl-D-aspartic acid

EdU

5-Ethynyl-2´-deoxyuridine

RNA-seq

next-generation sequencing for messenger RNA

ATAC-seq

assay for transposase-accessible chromatin using sequencing

GFAP

glial fibrillary acidic protein

AAV

adeno-associated virus

CRISPR

clustered regularly interspaced short palindromic repeats