The in vitro generation of neuronal subtypes provides an invaluable resource for the study of many neuronal diseases. Figure 1 summarized the transcription factor combinations that were used for in vitro direct reprogramming into neuronal subtypes. One such promising effort is in the reprogramming of glutamatergic neurons. Glutamate is the major neurotransmitter of the central nervous system and an imbalance in its production can lead to major neuronal defects[20,21]. Degradation of glutamatergic neurons is linked to disorders such as schizophrenia[22]. On the other hand, overexpression of glutamate can lead to excitotoxicity and glutamatergic cell death, which are associated with Alzheimer's disease and amyotrophic lateral sclerosis[23,24].

As mentioned previously, ASCL1 alone is capable of reprogramming fibroblasts into excitatory glutamatergic neurons. However, ASCL1 is often used in a combinatorial transcription factor approach for direct reprogramming to generate neurons, such as the widely used BAM combination (ASCL1, BRN2 and MYT1L) discovered by Vierbuchen et al[25]. In this study, reprogramming using ASCL1 alone resulted in generation of Tuj1-positive neurons. However, the morphological complexity, maturity and action potentials of the cells could be further improved by the addition of BRN2 and MYT1L. BAM were able to reprogram fibroblasts into predominantly a mix of both glutamatergic and GABAergic neurons; while no other major neuronal subtypes were detected in significant numbers[26]. Unlike glutamate, gamma aminobutyric acid (GABA) is the major inhibitory neurotransmitter produced by GABAergic neurons[27]. A further study has discovered that BAM expression is only required in the initial stages of the reprogramming process and removing gene expression 3 days after transduction has no effect on the neuronal reprogramming process[26].

The incorporation of BRN2 and MYT1L into neuronal reprogramming is due to their importance in neural development. In the mouse hypothalamus, BRN2 enhances neural differentiation during development and promotes activation of other events including the maturation and survival of paraventricular nuclei, as well as supraoptic nuclei [28,29]. MYT1L is a gene expressing a zinc finger protein which can be found in early differentiating neurons but is absent from glial cell populations, suggesting a role in early neuronal differentiation[30]. Although these factors are important for the reprogramming of neurons, neither BRN2 or MYT1L alone is sufficient to induce neurons, as they require ASCL1 to initiate the specification process[25].

Dopaminergic (DA) neurons are the main source of dopamine production in the mammalian nervous system and its degeneration could lead to devastating neurological disorders, such as Parkinson’s disease. Restoration of DA neurons in Parkinson's disease could provide a treatment for the disease as demonstrated in rat models[31]. In order to achieve this level of cell type specification, it is necessary to incorporate fate specification factors during the reprogramming process.

By incorporating the expression of DA specification factors LMX1A and FOXA4 along with BAM factors, approximately 10% of the human iN reprogrammed from human embryonic fibroblasts were DA neurons. Interestingly, the same study found that LMX1A and FOXA4 drove more human iN to a DA fate but did not increase the overall neuronal conversion rate[26]. The differentiation of DA neurons has been shown to be influenced by a positive feedback system in which FOXA1 and FOXA2 promote the expression of LMX1A and LMX1B, which subsequently leads to the development of mesodiencephalic DA neurons[32].

There are also other studies that utilized alternative combination of factors with similar results. Caiazzo et al[33] were able to induce DA neuron reprogramming using fibroblasts with a combination of Ascl1, Nurr1 and Lmx1a. This 3-factor cocktail was able to induce approximately 85% iN that are positive for the DA marker, TH. In this regard, Nurr1 is a DA specific receptor, essential for the formation and survival of DA neurons[34]. It is not activated by ligands but rather forms a heterodimer complex with retinoid X receptor (RXR), and together this complex is able to bind RXR ligands that produce signalling essential for the survival of DA neurons[34].

A subsequent study identified another set of 5 transcription factors which successfully converted human fibroblasts into human-induced DA neurons: ASCL1, NGN2, SOX2, NURR1 and PITX3. Importantly, ASCL1, NGN2 and SOX2 are required for this reprogramming process. On the other hand, exclusion of NURR1 and PITX3 did not have an effect in the early reprogramming process, rather these factors increased the dendrite network and facilitated maturation of DA neurons[35].

Diverse subtypes of sensory neurons are responsible for pain and itch perception. Mutations in sensory neuron-specific proteins result in development of a wide range of sensory disorders, such as Friedreich’s ataxia[36]. Due to limited availability of human sensory neurons, research in this field is largely dependent on animal models, especially rodents. Therefore, many mechanisms involved in human pain and itch perception remain uncharacterized. Studies from human primary dorsal root ganglion (DRG) neurons have identified differences between human and mouse nociceptors in function of individual channels, receptors and their response to chemical stimuli[37]. Hence, the development of protocols for the generation of bona-fide human sensory neurons in sufficient numbers is vital for accurate modeling of processes like pain and itch.

A previous study has demonstrated the feasibility of reprogramming fibroblasts into sensory neurons using a combination of Brn3a either with Ngn1 or Ngn2[38]. The cells displayed pseudounipolar morphology, and selectively responded to chemical mimics of pain, temperature, and itch. Both combinations with Ngn1 or Ngn2 equally induced the differentiation into three functional classes of sensory neurons, expressing one of the tropomyosin receptor kinase TrkA, TrkB, or TrkC[38]. Ngn1 and Ngn2 are alternative neurogenic bHLH factors that are expressed within progenitors of the developing DRG[39]. Both factors promote cell cycle exit through the induction of NeuroD1 and NeuroD4, and regulate two waves of neurogenesis[40-42]. Ngn1 and Ngn2 are co-expressed during much part of the early phase of DRG neurogenesis, and are both required for the generation of TrkB+ and TrkC+ sensory neurons. On the other hand, in the later phase Ngn1 is exclusively expressed in neural precursors and is largely responsible for the development of TrkA+ class sensory neurons[39]. At the time of cell cycle exit, the sensory neurons express the “pan-sensory” factors Brn3a and Islet1, which terminate the expression of the bHLH neurogenic factors and initiate the expression of definitive sensory markers to complete neurogenesis[42].

Similarly, functional nociceptor neurons can be generated through the transgenic expression of Ascl1, Myt1l, Ngn1, Isl2, and Klf7 in mouse and human fibroblasts[43]. The resultant iN expressed functional receptors for the noxious compounds menthol (TrpM8), mustard oil (TrpA1) and capsaicin (TrpV1). They also displayed the nociceptor-specific TTX-resistant Nav1.8 sodium channel. Notably, these iN were able to replicate inflammatory and chemotherapy-induced sensitization, which form the basis of pathological pain. Furthermore, this reprogramming method was successful in the generation of patient-derived neurons from familial dysautonomia, thus representing a promising approach for disease modeling[43]. For the five factors employed, Ascl1 and Myt1l are members of the three-factor combination for neuronal lineage reprogramming BAM[25]; Klf7 is a factor involved in TrkA expression maintenance[44]; and Isl2 was selected based on its expression profile in DRG[45]. The extent of Isl2 contribution to the lineage reprogramming remained unknown and further studies are needed to clarify this.

Retinal neurons are another promising target for therapeutic in situ reprogramming strategies. The retina is a highly organized structure bearing several major types of neurons, including rod and cone photoreceptors, bipolar, amacrine, horizontal and retinal ganglion cells (RGC)[46]. Rod and cone photoreceptors are responsible for detecting light stimuli and converting it to electrical signals that are later sent to the brain for visual perception. The loss of photoreceptors is a key hallmark of many blinding diseases and currently there are no effective treatments to cure blindness once photoreceptors are lost. Therefore, cellular reprogramming is a powerful technique for developing novel approaches for photoreceptor regeneration.

Akihiro Umezawa’s group used a promising combination of factors (CRX, RAX and NEUROD1) to induce the conversion of both iris cells[47] and fibroblasts into photoreceptor-like cells[48]. These factors induced the expression of photoreceptor-related genes and the reprogrammed cells became positive for the rod marker rhodopsin and the cone marker blue-opsin. Notably, some of the reprogrammed cells were photoresponsive. Interestingly, the authors showed that although OTX2 enhanced the upregulation of retinal genes, it was not essential for the reprogramming[48]. The same combination of factors was used to reprogram peripheral blood mononuclear cells[49], which facilitate clinical applications of this reprogramming strategy as blood samples are easier to collect from patients than fibroblasts. The generated photoreceptor-like cells expressed blue and red/green opsin, and some of the cells were able to respond to light stimuli. However, the reprogrammed cells expressed low levels of rod marker rhodopsin and some photoreceptor-related genes were not detected, which indicated that additional factors might be required to produce mature and functional photoreceptors[49].

In the developing retina, OTX2 is expressed in progenitors and early precursors that become committed to a photoreceptor fate[50,51]. NEUROD and CRX are factors expressed in developing photoreceptors and their expression is maintained in photoreceptors of mature retina, therefore they are proposed to participate in cell fate specification, as well as in maturation and survival processes[41,51]. Additionally, NEUROD is known to promote neuron specification in retinal progenitors and regulates interneuron differentiation to direct the amacrine cells fate[41]. On the other hand, CRX is downstream of OTX2 activity, it enhances the expression of photoreceptor-specific genes and is important for terminal differentiation into rod and cone photoreceptors[51,52]. RAX is a factor expressed in retinal and hypothalamic progenitor cells and is a key transcription factor for eye development in vertebrates. In the developing retina, it is involved in retinal progenitor proliferation and photoreceptor fate specification[53-55].

RGC are another interesting therapeutic target within the retina. These specialised retinal neurons form the optic nerve which convey the visual cues to the brain[46,51]. Degeneration of RGC is a major hallmark in glaucoma, a major blinding disease affecting the aging population[56]. Thus, in vitro generation of RGC offers an exciting avenue to develop regenerative therapy for this disease.

A study by Guimarães et al[56] demonstrated that RGC can be reprogrammed from postnatal Müller glia through the overexpression of Ngn2. The forced expression of Ngn2 produced a pool of iN which express genes associated with photoreceptors, amacrine cells and RGCs. However, this was only possible with Müller glia from young mice, as P(21) Müller glia cells failed to reprogram into iN. They also showed that the presence of mitogenic factors like EGF or FGF2 during the expansion of Müller glia enhanced the efficiency of the reprogramming. This finding supports the theory that the starting cell types for reprogramming have a strong influence on the neuronal subtypes in the resultant iN. For instance, ASCL1 or NGN2 alone can reprogram astroglia from the cerebellum and neocortex into neuron subtypes in the brain[57]. On the other hand, these factors can drive reprogramming of Müller glia into iN subtypes with retinal neuronal identities[56,58]. Thus, careful consideration should be taken in choosing the starting cell type in reprogramming to generate specialized neuronal subtypes.

Most of the reprogramming studies mentioned above have taken a candidate elimination approach to identify the optimal set of transcription factor(s) required for direct reprogramming, which is a tedious and laborious screening process. Recent advances in computational biology provide an alternative approach to predict the required transcription factor for cellular reprogramming in a faster and more efficient manner. One early example is CellNet, a computational algorithm that analyzes, categozises and predicts the function of transcription factors[59]. Subsequently, several other programs have been described to predict the transcription factors for cellular reprogramming, including Mogrify[60], BART[61], MAGICACT[62] and CellRouter[63]. Most of these programs work by comparing the quantitative amount of transcription factor in one cell type to another cell type to create a specificity score, which is used to create a ranking in which transcription factor are sorted and categorized depending on the cell type they are most prominent in[59,64]. Another exciting approach to predict transcription factors for cellular reprogramming is by creating computer models of cells. These programs, such as DeepNEU, can be used to create a simulation model of the cell using deep learning and provides a simulation of the events after the introduction of selected transcription factors[65]. Although these advanced algorithmic models are still in their infancy, further development will improve our understanding of the fundamental properties of cells and the molecular interplay of transcription factors that promote cellular reprogramming.

The potential to apply direct reprogramming in vivo represents an exciting direction for regenerative medicine. Therapeutic approaches using in vivo reprogramming have the potential to re-purpose local cells into the cells lost following injury or disease, thus providing an alternative regenerative approach to transplantation[66]. This has already been demonstrated for many neuronal subtypes, including glutamatergic and GABAergic neurons[67]. It was shown that overexpression of NeuroD1 is sufficient to convert astrocytes to glutamatergic neurons in rodents in vivo, as characterized by vGlut1 expression and glutamate-mediated synapses. Compared to astrocytes, interestingly the same study found that the NG2 glial cells have a larger reprogramming capacity into multiple neuronal subtypes, as the presence of both glutamatergic and GABAergic neurons was detected after NeuroD1 overexpression[67].

There has also been an extensive effort to re-purpose local cells for retinal regeneration. Many studies target the Müller glia within the retina for in vivo reprogramming, which are cells responsible for maintaining the integrity and homeostasis of the retina[68]. Interestingly, the Müller glia exhibit some progenitor properties. In the teleost fish and chicken, upon retinal injury the Müller glia can dedifferentiate into multipotent progenitors and give rise to all retinal neural subtypes[69]. This remarkable trait makes Müller glia an excellent candidate for reprogramming studies.

In several elegant studies by Tom Reh’s group, the author has demonstrated successful reprogramming of mouse Müller glia into a range of retinal neurons both ex vivo and in vivo, by the forced expression of the pro-neural factor Ascl1[58,70]. When Ascl1 was overexpressed in retinal explants and in Müller glia dissociated cultures, the cells re-entered the mitotic cell cycle and expressed neural progenitor genes[58]. This reprogramming was accompanied with chromatin remodeling, acquisition of neural morphology and the ability to respond to neurotransmitters[58]. In addition to its role as a pioneer proneural factor for direct reprogramming and retinal development, Ascl1 induction in response to injury is required for retinal regeneration in the fish[71-73]. Conversely, organisms with limited regenerative capacity in the retina do not upregulate Ascl1 following retinal damage[58,74]. Subsequently, in a landmark study the same group extended the application of Ascl1 to reprogram Müller glia into retinal neurons in vivo following retinal injury[70]. Marker analysis showed that the reprogrammed retinal neurons were able to functionally integrate with the existing retinal circuit. Interestingly, this reprogramming approach was less effective in adult mice compared to young mice[75]. The restrictive regenerative capacity of adult mice Müller glia is thought to be caused by reduced epigenetic accessibility of progenitor genes in the cell, as the addition of histone deacetylase tricostatin A is able to overcome this epigenetic hurdle in adult Müller glia[70]. In support of this, epigenetic profiling using ATAC-seq demonstrated that the treatment with trichostatin A favored the accessibility of genes associated with neural development and differentiation, such as Otx2[70]. In this case, Otx2 is known to regulate genes associated with bipolar and amacrine cells. However, it should be noted that most of the reprogrammed retinal neurons are bipolar cells, suggesting that additional factors are required to induce photoreceptors in vivo.

A subsequent study from Bo Chen’s group demonstrated that adult mice Müller glia can be reprogrammed to rod photoreceptors without the necessity of retinal injury[76]. In this study, the authors first stimulated Müller glia proliferation by forced expression of β-catenin, followed by overexpression of Otx2, Crx and Nrl after two weeks. This approach allowed successful generation of rod photoreceptors in vivo that functionally integrate into the retinal and visual cortex circuits. In a remarkable experiment, the authors were able to use this in vivo reprogramming approach to restore light response in a mouse model of photoreceptor degeneration. Collectively, these studies demonstrate the potential of in vivo reprogramming as a novel approach for neural regeneration.

Beside transcription factors, alternative direct reprogramming strategies using small chemicals and microRNAs are also two exciting research directions to improve the usability of the technology. For instance, chemical reprogramming using small chemicals has been shown to induce functional neurons without the need for transcription factor. A cocktail of 7 small molecules (Valproic acid, CHIR99021, Repsox, Forskolin, SP600125, GO6983 and Y-27632) was successful at reprogramming human fibroblasts into neurons with functional electrophysiology representative of glutamatergic and GABAergic cells[77]. Furthermore, the use of a combinational approach of small molecules together with transcription factors can improve direct reprogramming. In a study by Liu et al[15], NGN2 along with small molecules, forskolin and dorsomorphin, directly reprogrammed human fetal lung fibroblasts to cholinergic neurons with functional electrophysiology. In this regard, forskolin is a cAMP activator in the PKA signalling pathway and dorsomorphin acts as an inhibitor of the BMP signalling pathway, both of which are signaling pathways involved in neurogenesis.

Similarly, microRNAs have also been used in combinational approaches with transcription factors to induce neuronal reprogramming. microRNAs are known to play important roles in post transcriptional regulation, neural differentiation, morphological and phenotypic development[78-82]. In a study by Yoo et al[83], miR-9/9* and miR-124 were used in combination with NEUROD1 to convert human fibroblasts into neurons, however these cells would not always demonstrate repetitive action potential, signifying immature neurons. In order to tackle this issue the same group also introduced two other factors, ASCL1 and MYT1L, and in turn were able to induce neurons with higher maturity which demonstrated repetitive action potentials. These studies highlighted alternative approaches to transcription factors that promote neuronal reprogramming.