Small molecules reprogram reactive astrocytes into neuronal cells in the injured adult spinal cord

Graphical abstract

Highlights

• •Chemical screening identifies a small molecule cocktail LC for reprogramming.
• •Mature neurons can be induced from cultured astrocytes with LC alone.
• •LC induces neuronal reprogramming in injured but not intact adult spinal cord.
• •Astrocyte-converted neurons can also be induced in injured aged spinal cord.

Abstract

Introduction

Ectopic expression of transcription factor-mediated in vivo neuronal reprogramming provides promising strategy to compensate for neuronal loss, while its further clinical application may be hindered by delivery and safety concerns. As a novel and attractive alternative, small molecules may offer a non-viral and non-integrative chemical approach for reprogramming cell fates. Recent definitive evidences have shown that small molecules can convert non-neuronal cells into neurons in vitro. However, whether small molecules alone can induce neuronal reprogramming in vivo remains largely unknown.

Objectives

To identify chemical compounds that can induce in vivo neuronal reprogramming in the adult spinal cord.

Methods

Immunocytochemistry, immunohistochemistry, qRT-PCR and fate-mapping are performed to analyze the role of small molecules in reprogramming astrocytes into neuronal cells in vitro and in vivo.

Results

By screening, we identify a chemical cocktail with only two chemical compounds that can directly and rapidly reprogram cultured astrocytes into neuronal cells. Importantly, this chemical cocktail can also successfully trigger neuronal reprogramming in the injured adult spinal cord without introducing exogenous genetic factors. These chemically induced cells showed typical neuronal morphologies and neuron-specific marker expression and could become mature and survive for more than 12 months. Lineage tracing indicated that the chemical compound-converted neuronal cells mainly originated from post-injury spinal reactive astrocytes.

Conclusion

Our proof-of-principle study demonstrates that in vivo glia-to-neuron conversion can be manipulated in a chemical compound-based manner. Albeit our current chemical cocktail has a low reprogramming efficiency, it will bring in vivo cell fate reprogramming closer to clinical application in brain and spinal cord repair. Future studies should focus on further refining our chemical cocktail and reprogramming approach to boost the reprogramming efficiency.

Introduction

The central nervous system (CNS) of adult mammals does not have the intrinsic ability to regenerate and replace damaged neurons. Therefore, irreversible neuronal loss is linked to both acute CNS injuries and chronic neurodegenerative diseases and ultimately results in a wide spectrum of persistent functional deficits and neurological disability. Instructing non-neuronal CNS-resident cells to become neurons is a promising strategy for CNS repair. Encouraged by the advent of induced pluripotent stem cells (iPSCs) [1], direct neuronal reprogramming has made significant progress in recent years, making it possible to produce functional neurons that are specific to a patient without having to go through a pluripotent intermediate. The direct neuronal reprogramming was firstly achieved by ectopically expressing neural transcription factors Brn2, Ascl1, and Myt1l (BAM) in non-neuronal cells [2]. Since then, different combinations of fate-determining factors have been screened in vitro and applied to produce subtype-specific neurons for cell transplantation-based neuroregenerative medicine, disease modeling and drug identification [2], [3], [4], [5], [6], [7], [8], [9], [10], [11]. Importantly, given the plasticity of certain resident glial cells in CNS, transcription factors-mediated in vivo direct neuronal reprogramming has gained momentum, providing a perspective for in situ cell-based regenerative therapy [12], [13], [14], [15], [16], [17], [18]. In addition to transcription factors, non-neuronal somatic cells have also been reprogrammed into neurons by neurogenic microRNAs (miRNAs) [19], [20], [21], [22]. In transcription factors or miRNAs-mediated neuronal reprogramming, however, the genetic manipulation and the technical challenges in delivering exogenous genes in vivo have raised safety concerns and limited their potential for future clinical applications.

Recently, small molecules as an attractive alternative have substantial appeal in the field of lineage reprogramming [23], [24], [25], [26], [27]. Compared to transcription factors and miRNAs, small molecules are easily synthesized, preserved and inexpensive. Furthermore, they are cell permeable, non-immunogenic, easy-to-control, and reversible. Importantly, small molecule-mediated reprogramming does not result in the dangers of inserting exogenous gene sequences into the genome. Based on these advantages, pioneering studies by several groups have demonstrated that small molecules can be used for cell-fate reprogramming. For instance, small molecules have been indicated to boost the effectiveness of cell reprogramming mediated by transcription factors [28], [29]. Small molecules can also replace some transcription factors to induce cell lineage reprogramming [30], [31], [32], [33]. Of note, recent studies demonstrate that chemical compounds alone are equipped to convert somatic cells into functional neurons in vitro [34], [35], [36]. Without exogenous genetic factors, however, it remains largely unknown whether neuronal reprogramming can be induced by small molecules in vivo.

In this study, immunocytochemistry, immunohistochemistry, qRT-PCR and lineage tracing were used to analyze the role of small molecules in reprogramming astrocytes into neuronal cells in vitro and in vivo. Initially, we carried out a series of chemical screening and identified a chemical cocktail with only two compounds that could directly and rapidly convert reactive astrocytes into neuronal cells in the damaged spinal cord of adult mice, even of aged mice. Typical neuronal morphologies and neuron-specific marker expression were observed in these chemically induced cells. Furthermore, they could become mature and survive for more than 12 months. Lineage tracing indicated that the chemical compound-converted neuronal cells mainly originated from post-injury spinal reactive astrocytes. Together, this proof-of-principle study demonstrates the feasibility of chemical compound-induced in vivo neuronal reprogramming for in situ repair of spinal cord injury (SCI), suggestive of a promising potential for clinical applications.

Materials and methods

Animals

For primary astrocyte culture or in vivo chemical reprogramming, wild-type newborn (P0), adult (2–3 months old, 22–26 g) and aged (13–16 months old) C57BL/6J mice were obtained from Shanghai Ling Chang Biotech Co., Ltd. Newborn mice were used to culture primary astrocytes. Adult and aged mice were used to analyzed small molecule-induced in vivo neuronal reprogramming. For lineage tracing, FVB/N-Tg(GFAP-GFP)14Mes/J transgenic mice expressing green fluorescent protein (GFP) under the GFAP promoter were obtained from Jackson Laboratory. In all, about 100 animals were used in our study. Mice were housed under conventional conditions with controlled temperature (22 ± 3 °C), humidity (60 ± 10 %) and light (12-h light–dark cycle) with unrestricted access to food (standard pellet feed, 150–350 g/kg body weight per day) and water.

Ethics statement

All experiments involving animals were conducted according to the ethical policies and procedures approved by the Naval Medical University Animal Experimentation Ethics Committee (NMUMREC-2021-033).

Primary cell culture

Following the described protocol [14], [37], highly purified primary cortical and spinal astrocytes that were isolated from P0 mouse brains and cultured at 37 °C with 5 % CO₂ in DMEM/F12 (1:1) medium containing 10 % fetal bovine serum (FBS) (Gibco) and 1 % penicillin–streptomycin. The culture medium was changed the following day with new culture medium, and it was changed every three days. When cultured cells reached confluence, loosely attached microglia and oligodendrocyte precursor cells were removed from the cell monolayer by vigorous shaking in order to obtain enriched astrocytes.

In vitro astrocyte-to-neuron conversion

For the purpose of reprogramming neurons, cultured astrocytes (4 × 10⁴/ml) were passaged and seeded on glass coverslips or culture vessels that had already been coated with gelatin and matrigel. Chemicals were applied to astrocytes 24 h after plating to induce neuronal reprogramming. Three days later, the neuronal induction medium (NI), DMEM:F12:neurobasal (2:2:1) with 0.8 % N-2 and 0.4 % B-27, was substituted for the astrocyte medium (AM). 7 days after treatment with compound (D7), the culture medium was replaced with neuronal maturation medium (NM), NI supplemented with GDNF and BDNF (10 ng/ml each, Peprotech). Every other day, half of the culture medium was changed during the cell fate reprogramming process. The astrocyte-converted neuronal cells were identified by immunostaining and morphological analysis.

SCI model

In vivo reprogramming was tested in a crush-injured SCI models. To establish a clinically relevant model of SCI, adult or aged mice received a ∼2 cm incision along the midline of the back and laminectomy at T8 after anesthetized with 2 % pentobarbital (30 mg/kg). Using a pair of forceps with a 0.4 mm spacer, lateral compression was performed on the T8 spinal cord for 15  s to induce crush injury [38]. The animals underwent manual bladder expression twice daily following surgery until reflexive bladder control had fully recovered. In order to conduct an immunohistological analysis, spinal cords were collected at the appropriate time.

Chemical compounds administration and BrdU labelling

Small molecules (Supplementary Table 1) were obtained from MedChemExpress (MCE). They were prepared for use at the specified concentrations after being dissolved in DMSO in accordance with the manufacturer's instructions. When handling these compounds, direct skin contact or inhalation should be avoided. For in vitro neuronal conversion, small molecules were added directly to cell cultures. For in vivo neuronal reprogramming, small molecules were administrated by intraperitoneal injection.

As previously described protocol [39], animals received intraperitoneal injections of BrdU (5-bromo-2-deoxyuridine, Sigma) at a dose of 100 mg/kg body weight twice daily for the specified time in order to label the proliferating cells in the spinal cord. Using an anti-BrdU antibody, fluorescent staining was used to identify BrdU incorporation.

Viral injection

To fluorescently trace astrocytes, an adeno-associated virus system (AAV2/9-GFAP-mCherry) was used in our in vivo chemical reprogramming. After diluted in sterile saline solution, the high-titer viral solution containing AAV2/9-GFAP-mCherry (greater than 1 × 10¹³ vg/ml, 100 μL) was slowly injected into mice via tail vein with an insulin syringe.

Immunocytochemistry and immunohistochemistry

Cultures were fixed with 4 % paraformaldehyde (PFA) for 20 min at room temperature in preparation for immunocytochemistry. The cells were blocked and permeabilized for one hour in PBS with 3 % BSA and 0.2 % Triton X-100 after being washed three times with PBS. For immunohistochemistry, mice were transcardially perfused with ice-cold PBS and then 4 % PFA after receiving an excessive dose of anesthesia. Dissected spinal cords were post-fixed overnight at 4 °C in 4 % PFA following perfusion. Cryoprotection was achieved by submerging PFA-fixed tissues in 30 % sucrose at 4 °C for 48 h. The spinal sections were prepared by cutting them on a Leica cryostat at a thickness of 14 μm. Cells or spinal sections were incubated with the primary antibodies (Supplementary Table 2) overnight at 4 °C in order to perform immunostaining. Indirect fluorescence was performed using the proper secondary antibodies (Jackson ImmunoResearch) conjugated to Alexa Fluor 488, 594, or 647. Hoechst 33342 (Hst) was used as a counterstain for nuclei. Images were taken using a Leica SP5 confocal microscope or a Nikon E600FN microscope.

Quantitative real-time reverse-transcription polymerase chain reaction (qRT-PCR)

qRT-PCR was used to analyze gene expression as previously described [40]. RNase-free DNase (Thermo Scientific Fermentas) was used to remove contaminating DNA after extracting the total RNA from cultured cells using Trizol reagent (Invitrogen). After reverse-transcribing RNA into cDNA with a RevertAid First Strand cDNA Synthesis Kit from Thermo Scientific Fermentas, qRT-PCR was carried out using a Bio-Rad MyiQ™ with SYBR Green Realtime PCR Master Mix (TOYOBO Biotech). The detailed procedure of PCR was as follows: 1 min at 95 °C, followed by 40 cycles of 15 s at 95 °C, 15 s at 60 °C and 45 s at 72 °C. After normalizing to the expression of Gapdh, the 2^− ΔΔCt method was used to calculate and quantify gene expression. Supplementary Table 3 contains a comprehensive list of primers.

Western blot analysis

Primary cell cultures were homogenized in RIPA buffer that was supplemented with protease cocktail inhibitors (Beyotime, Shanghai, China). Cell lysates were subjected to Western blot analysis with antibodies listed in Supplementary Table 4. The protein bands were analyzed and quantified using Image Lab (ODYSSEY CLX, LI-COR, America), normalizing target proteins to GAPDH or β-ACTIN bands.

Quantitative analysis in vivo

According to previous protocol [41], [42], standard histological procedures were performed to quantitatively analyze the induced neuronal cells in spinal cord. For each group, 3–5 spinal cord-crushed mice were utilized. the purpose of cell number counts, every 10th serial 14-μm-thick spinal cord section and a total of 8–10 sections per animal were used for immunostaining. Around the lesion site, there are four to six confocal images taken at random with a 20 × objective and the Image pro-Plus 6.0 was applied for cell counts.

Data and statistical analysis

At least three experimental animal or culture batches were used for each result. Mean ± s.d was used to express the quantitative data. Student’s t-tests or one-way ANOVA with Tukey’s post hoc test were used for statistical analysis, and significance was defined as *P < 0.05, **P < 0.01, or ***P < 0.001. Experiments that were blind to the treatment conditions were used to collect and analyze all data. Under a 20x magnification, the image fields were chosen at random.

Results

Identification of small molecules for reprogramming cultured astrocytes into neuronal cells

In previous studies, we have reported that the endogenous differentiated astrocytes could be converted into neuronal cells in the adult spinal cord by ectopically expressing a single transcription factor SOX2 [14], [43]. To examine whether transcription factors can be replaced by small molecules to induce neuronal fate in vivo, we started out with a primary compound screening for chemical reprogramming astrocytes into neuronal cells in vitro. Based on three major criteria including inhibiting glial signaling pathways, activating neuronal signaling pathways or targeting epigenetic modulators to make neuronal genes open for transcription, 6 small molecules (6F) including LDN193189 (L), CHIR99021 (C), i-BET151 (B), SAG (G), SB431542 (S), and ISX9 (I) were selected as our starting candidate compound pool (Supplementary Table 1). Using an in vitro astrocyte-to-neuron conversion system that we previously set up [37], these potential candidate compounds were applied for further functional screening (Fig. 1A). From the P0 mouse's cortex, primary astrocytes for neuronal reprogramming were isolated. Before being used for reprogramming, the cultured cells were passed through at least three times to prevent neuronal cells from contaminating primary astrocytes. Immunocytochemical analysis indicated that consistently expressed the astrocyte markers GFAP and ALDH1L1, but not the neuronal markers DCX, TUBB3, MA P2 and NeuN (Supplementary Fig. 1). Moreover, no neural stem cells (NESTIN⁺) and oligodendrocyte lineage cells (OLIG2⁺) but only a few microglia (IBA⁺, <3 %) were detected in the cultures (Supplementary Fig. 1). Astrocytes cultured under our experimental conditions had a purity of more than 95 %, as determined by their expression of GFAP or ALDH1L1. (Supplementary Fig. 1).

Fig. 1.

Identification of neuronal fate-inducing small molecules in astrocyte culture. (A) Diagram of the small-molecule screening for candidate compounds. AM, astrocyte medium; NI, neuronal induction medium. (B and C) DCX immunostaining showed that the 6F pool (L, LDN193189; C, CHIR99021; B, i-BET151; G, SAG; S, SB431542; I, ISX9) was found to substantially potentiate astrocyte-to-neuron conversion. Nuclei were counterstained with Hoechst 33,342 (Hst). (D) Quantification of DCX-positive cells induced by 6F or by withdrawing individual chemicals from 6F. (E) Representative micrographs of DCX and TUBB3-positive cells induced by LC (LDN193189, CHIR99021). ***P < 0.001 by Student’s t-test (C) or one-way ANOVA Tukey’s post hoc test (D) (n = 3; cells = 700–1,500 for each condition). The scale bars represent 20 µm.

The cultured astrocytes maintained in a 10 % FBS-containing DMEM/F12 medium were treated with the 6F compound pool, DMSO treatment as a control (Fig. 1A). The medium was changed to an inducing neuron medium three days later (Fig. 1A). Induced neuronal cells were initially identified by staining for doublecortin (DCX, an immature neuronal marker) expression. Compared to the control group, we found that 6F treatment reprogrammed astrocytes into DCX-positive cells with a bipolar neuron-like cell morphology at day 7 (D7) (Fig. 1 B and C). To tease out the precise master molecules that contribute to neuronal reprogramming, further screening was performed by withdrawing individual chemicals from the 6F pool. As shown in Fig. 1D, B (i-BET151), G (SAG), S (SB431542) and I (ISX9) were dispensable for generating DCX-positive cells. Immunocytochemical analysis indicated that a significant fraction of cultured astrocytes was reprogrammed into DCX-positive cells at 7 days after exposing them to LC (LDN193189 and CHIR99021) but not L (LDN193189) or C (CHIR99021) (Fig. 1E; Supplementary Fig. 2). suggestive of a synergistic action of LDN193189 and CHIR99021 in cell fate reprogramming. Of note, LC-treated astrocytes gradually lost their flat morphology and TUBB3 (a pan-neuronal marker) was detected at D12 (Fig. 1E). Importantly, LC was also shown to convert spinal cord-derived astrocytes into DCX- and TUBB3-positive cells (Supplementary Fig. 3). Together, these findings suggest that the compound combination LC can successfully reprogram cultured astrocytes into neuronal cells without introducing exogenous transcriptional factors.

Characterization of chemically induced neuronal cells and lineage reprogramming process

Although LC treatment clearly reprogrammed cultured astrocytes into induced cells that expressed early neuronal markers DCX and TUBB3 in the screening condition (Fig. 1), we found that only a few mature neuronal cells immunopositive for MAP2 and NeuN were detected afterwards (data not shown). Therefore, we optimized the reprogramming condition to promote the survival and maturation of the chemically induced neuronal cells (CiNs). After 14 days of exposure to LC, astrocytes switched from neuronal induction medium to neuronal maturation medium at D5, D7, D9 and D11, respectively (Supplementary Fig. 4A). The neurotropic factors GDNF and BDNF were supplemented in neuronal maturation medium to improve neuronal cell survival and maturation. Immunostaining showed that the most TUBB3-positive neuronal cells were induced by LC treatment when the neuronal induction medium was changed to neuronal maturation medium at D7 (Supplementary Fig. 4B and C). It is interesting to note that the astrocyte marker GFAP was co-labeled on a small percentage of TUBB3-positive cells, suggesting that they were immature induced neuronal cells and derived from astrocytes (Supplementary Fig. 4B). According to quantitative analysis, the percentage of TUBB3⁺/GFAP⁺ cells in TUBB3⁺ cells was significantly low in the LC-mediated reprogramming system when the neuronal induction medium was changed to neuronal maturation medium at D7 but not at other time points (Supplementary Fig. 4B and D). Importantly, changing the neuronal induction medium to neuronal maturation medium at D7 resulted in the largest number of LC-induced MAP2⁺ mature neuronal cells (Supplementary Fig. 4E–G), suggestive of CiNs survival and maturation being significantly improved.

The optimized chemical induction protocol was then applied to LC-mediated conversion of astrocytes into neurons (Fig. 2A), and the maturation of CiNs was investigated by immunocytochemical analysis. At D7, the CiNs were immunopositive for DCX and TUBB3 and showed a typical morphology similar to that of a neuron, with a small soma-like cell body and long bipolar or multipolar processes, indicating that they were immature neuronal cells (Fig. 2B). At D15, the CiNs changed in their appearance and gradually acquired the shape of a mature neuron with multiple and intricate processes, expressing mature neuron-specific markers, MAP2 and NeuN (Fig. 2C and D). In sharp contrast, no MAP2- or NeuN-positive cells were observed in the DMSO-treated astrocyte cultures (Fig. 2E). At D28, the neurites of CiNs became longer and their morphology became more complex (Fig. 2F). Mature neurons form dense neural circuits by connecting to each other via synapses. Therefore, the synapse formation is an important indicator of neuronal function. To determine whether small molecule-converted neurons are functionally connected, the synapse formation was examined with the presynaptic terminal marker synapsin-1 (SYN1). As shown in Fig. 2G, we observed SYN1-labeled robust synaptic puncta on the soma and along the processes of CiNs, suggestive of the formation of functional synapses between the chemically reprogrammed neurons.

Fig. 2.

Maturation of LC-induced neuronal cells. (A) Diagram of astrocyte-to-neuron conversion mediated by LC. AM, astrocyte medium; NI, neuronal induction medium; NM, neuronal maturation medium. (B-E) In LC-treated astrocyte cultures, DCX, MAP2 and NeuN positive cells were observed, which were co-labelled by TUBB3. (F) No MAP2 or NeuN positive cells were detected in DMSO-treated astrocyte cultures. (G) Confocal images show that LC-induced neurons express SYN1 on the soma and processes, indicative of synapse-formation. The scale bars represent 20 µm.

To further globally understand the process of cell fate reprogramming, time-course analysis was performed. At the early reprogramming stage, most of LC-induced TUBB3-positive immature neuronal cells were also immunopositive for GFAP, suggesting that the CiNs were derived from astrocytes (Fig. 3A). Of note, Fig. 3 A and B showed that the percentage of TUBB3⁺/GFAP⁺ cells in TUBB3⁺ cells was gradually decreased over time, suggestive of a shift from astrocytes to neuronal cells. In addition, we also carried out qRT-PCR and Western blot to investigate the changes of gene and protein expression profiling in a time-course manner after treatment of cells with LC. Interestingly, we found a dramatic rise of some factors that determine neural fate, including Ngn2, NeuroD1/2, and Myt1l was induced within 48 hr, while the other neural-fate mastering genes including Ascl1, Brn2, Foxg1, and Lmx1a were not significantly activated early in the reprogramming process (Fig. 3C, Supplementary Fig. 5). The transcription of Ngn2, NeuroD1/2, and Myt1l peaked at D6-D8 (Fig. 3C). Among these neural transcription factors, the expression of Ngn2, NeuroD1 and Ascl1 that are most frequently used to induce neuronal fate conversion was further confirmed by immunocytochemistry. As shown in Supplementary Fig. 6, the expression of endogenous transcription factors Ngn2 and NeuroD1 were detected at D5 after chemical induction, while no Ascl1 expression was detected. In tandem with the activation of the neural transcription factor, the immature neuronal gene DCX increased sharply, peaking at D6 (up to 10000-fold) and descending afterwards (Fig. 3D, Supplementary Fig. 5). In the meantime, the mature neuronal markers including NeuN, and Map2 was gradually increased and peaked at D12 (up to 7.5-fold), indicative of a shift of CiNs from immature to mature state (Fig. 3D, Supplementary Fig. 5). Of note, the gene and protein expression of synaptic marker synapsin (syn) was induced and increased in the CiNs in a time-course manner, suggestive of synapse formation (Fig. 3D, Supplementary Fig. 5). To determine whether LC treatment might reprogram astrocytes into neural pregenitor cells (NPCs), the multipotent genes Sox2, Pax6, and Blbp were also examined during LC-induced reprogramming process from D0 to D12. We found that no significant activation of these genes was detected after LC treatment (Fig. 3E). Collectively, these data suggest that the defined compound combination LC has successfully induced astrocyte-to-neuron conversion by eliciting endogenous neurogenic program, which is a direct process without going through an intermediate progenitor.

Fig. 3.

Biological analysis of LC-induced astrocyte-to-neuron conversion. (A and B) During the LC-mediated chemical reprogramming, a fraction of induced TUBB3 positive cells were co-labelled by GFAP, and the ratio of TUBB3⁺GFAP⁺/TUBB3⁺ was gradually decreased. *P < 0.05, **P < 0.01, ***P < 0.001 by one-way ANOVA Tukey’s post hoc test (versus D5 group; n = 3; cells = 700–1,000 for each condition). (C-E) qRT-PCR analysis of the expression of representative genes for neurogenic factors (C), neuronal lineage markers (D), and multipotent stem cell markers (E), in LC-treated astrocytes at the indicated time. *P < 0.05, **P < 0.01, ***P < 0.001 by one-way ANOVA Tukey’s post hoc test (versus D0 group). The scale bars represent 50 µm.

LC treatment fails to induce neurogenesis in the intact adult spinal cord

Based on the successful of reprogramming astrocytes into neuronal cells in vitro, we next asked whether LC could also induce astrocyte-to-neuron conversion in vivo. Using a ADMETlab 2.0 server [44], the detailed ADMET properties of LC were evaluated before application in vivo (Supplementary Table 5). As shown in Fig. 4A and B, the healthy adult mice were treated with LC by intraperitoneal injection (10 mg/kg LDN193189, 20 mg/kg CHIR99021) once a day, and the induced neurogenesis was investigated by detecting DCX expression at 2 and 4 weeks post injection (wpi). The healthy adult mice intraperitoneally injected with DMSO served as a control. Immunohistochemical analysis revealed that there were no DCX-positive cells in the spinal cords of LC- or DMSO-treated mice (Fig. 4 C and D). Virtually, the spinal astrocytes still retained their quiescent shape as validated by staining for GFAP (Fig. 4C and D). These results suggest that LC treatment does not induce neurogenesis in the intact adult spinal cord without introducing exogenous transcriptional factors.

Fig. 4.

LC treatment does not induce neurogenesis in the intact adult spinal cord. (A and B) Schematic diagrams of experimental procedures. i.p., intraperitoneal injection. (C and D) Immunohistochemistry showed that DCX positive cells were not detectable in the spinal cords of animals injected with LC or DMSO (Ctrl) at 2 and 4 wpi. There were 5 mice in each group. The scale bars represent 200 µm (C) and 25 µm (D).

LC treatment induces neurogenesis in the injured adult spinal cord

It was reported that CNS injury-caused alterations in the cellular milieu were essential for transcription factor SOX2-induced reprogramming of glial cells into neurons conversion in the cerebral cortex [45]. After injury, the quiescent glial cells were switched to a reactive state that might make them susceptible to lineage reprogramming. Therefore, we next investigated whether LC treatment could trigger neuronal reprogramming in the adult spinal cord in the context of acute invasive injury. We utilized a clinically relevant SCI model in which trauma damage was caused by a crush wound inflicted to the spinal cord. In our previous studies, this crush-injured SCI model had been shown to result in massive activation of glial cells including astrocytes and microglia [41], [46]. To investigate whether neurogenesis can be chemically induced in the injured spinal cord during on-going reactive gliosis, LC was intraperitoneally injected into the mice once a day immediately after spinal crush-wound injury (Fig. 5A and B). At the same time, BrdU, a thymidine analog, was administered for 14 consecutive days to trace the newly-generated CiNs (Fig. 5B). DCX expression has been shown to be mostly linked to adult neurogenesis, but not to reactive gliosis or regenerative axonal growth [47]. At 5, 7, and 14 days post injection (dpi), no significant DCX expression was detected in the damaged spinal cord of the control mice treated with DMSO (Fig. 5 C and D, Supplementary Fig. 7A and B), consistent with the fact that the spinal cord of adult mammalian has no inherent capacity for neurogenesis. In the spinal cord of LC-treated animal, in contrast, DCX-positive cells were induced as early as 5 dpi, peaking at 1 wpi (Fig. 5 C and D, Supplementary Fig. 7A and B). Of note, the majority of DCX⁺ cells induced by LC were concentrated around the lesion region and exhibited an immature neuronal morphology (Fig. 5C, Supplementary Fig. 7A and B). Interestingly, the LC-induced DCX-positive cells were observed to incorporate BrdU, while they were rarely immunoreactive for PCNA (Supplementary Fig. 7C and D). Additionally, the LC-triggered neurogenesis was further confirmed by identification of the BrdU-incorporated TUBB3 positive cells at 2 and 4 wpi (Fig. 5E). Because LC was administered by intraperitoneal injection, we checked whether the compounds affected other tissues. As shown in Supplementary Fig. 8, LC injection had no significant effects on the neurogenesis in the subventricular zone (SVZ). Moreover, LC did not trigger glial activation or affect SCI-induced gliosis and scar-forming (Supplementary Fig. 9). Collectively, our data suggest that LC treatment successfully induces neurogenesis in the damaged adult spinal cord without introducing exogenous transcriptional factors.

Fig. 5.

LC treatment induces neurogenesis in the adult spinal cord after crush injury. (A and B) Schematic diagram of experimental procedures. i.p., intraperitoneal injection. (C and D) Immunohistochemistry showed that DCX positive cells were detected in the injured adult spinal cord of animal injected with LC but not DMSO (Ctrl). n.d., no detected. ***P < 0.001 by Student’s t-test (n = 4 mice per group). (E) Representative images of newly generated neuronal cells indicated by BrdU-traced TUBB3 positive cells in the injured adult spinal cord of animal injected with LC at 2 and 4 wpi. The scale bars represent 200 µm (C) and 25 µm (E).

In vitro, the time-course analysis of the cellular morphology (Fig. 2 B-D) and the expression of neuronal markers (Fig. 3D) highlighted a progressive maturation of CiNs. The cellular fate of the LC-induced CiNs was next assessed by labeling them with BrdU and the mature neuron-specific markers, MAP2 and NeuN, to determine whether they can develop into mature neuronal cells in the damaged spinal cord. Remarkably, mature BrdU⁺MAP2⁺ and BrdU⁺NeuN⁺ neuronal cells were observed in the spinal cord 4 weeks after animal was treated with LC, suggestive of maturation of CiNs (Fig. 6 A-C). In the control mice, conversely, MAP2 and NeuN were almost undetectable in BrdU-traced cells (Fig. 6C). The induced mature neuronal cells were further confirmed by triple immunostaining of BrdU, TUBB3 and NeuN. As shown in Fig. 6D, three types of cells were identified in the LC-treated damaged spinal cord with this triple staining. The NeuN⁺/TUBB3⁺/BrdU⁺ and NeuN^-/TUBB3⁺/BrdU⁺ cells indicated LC-induced mature and immature neuronal cells, respectively (Fig. 6D). The NeuN⁺/TUBB3⁺/BrdU^- cells were primarily endogenous spinal mature neuronal cells, albeit a few of them might also be LC-induced mature neuron (Fig. 6D). Immunohistochemistry revealed that the induced mature NeuN⁺/TUBB3⁺/BrdU⁺ neuronal cells were observed in LC-treated mice injured spinal cord as early as 2 wpi, peaking at 4 wpi (Fig. 6 E-G). Of note, the majority of these induced mature neuronal cells were found close to the lesion site (Supplementary Figure 10A). Importantly, these LC-induced mature neuronal cells could be detected even 8 and 12 months post injury (mpi), highlighting that they can survive for a long time (Supplementary Figure 11). On the contrary, few NeuN⁺/TUBB3⁺/BrdU⁺ neuronal cells were oberved in the damaged spinal cord of control mouse (Fig. 6G, Supplementary Figure 10B). Taken together, these results indicate that the LC-induced immature CiNs can differentiate into mature neuronal cells in the crushed adult spinal cord.

Fig. 6.

Maturation of LC-induced new neuronal cells in the injured adult spinal cord. (A and B) Expression of the mature neuronal markers MAP2 (A) or NeuN (B) in BrdU-traced cells in the injured adult spinal cord of animals injected with LC at 4 wpi. MAP2⁺/BrdU⁺ and NeuN⁺/BrdU⁺ cells were indicated by arrowheads. Orthogonal view of cells with expression of the MAP2 and BrdU was also shown. (C) Quantification of LC-induced MAP2⁺/BrdU⁺ and NeuN⁺/BrdU⁺ cells in the injured adult spinal cord at 4 wpi (n = 5 mice per group; ***P < 0.001 by Student’s t-test; n.d., no detected.). (D) Immunohistochemical analysis of the neuronal identity in the injured adult spinal cord of animals injected with LC at 4 wpi. The orthogonal view was also shown in the right lane. The NeuN⁺/TUBB3⁺/BrdU⁺ (D1) and NeuN^-/TUBB3⁺/BrdU⁺ (D3) cells indicated LC-induced mature and immature neuronal cells, respectively. The NeuN⁺/TUBB3⁺/BrdU^- cells (D2) were primarily endogenous spinal mature neuron, but could also be LC-induced mature neuron. (E and F) Representative images of NeuN⁺/TUBB3⁺/BrdU⁺ cells in the injured adult spinal cord of animals injected with LC at 4 and 8 wpi. NeuN⁺/TUBB3⁺/BrdU⁺ cells were indicated by arrowheads. Orthogonal view of cells with expression of the NeuN, TUBB3 and BrdU was also shown. (G) Quantification of LC-induced NeuN⁺/TUBB3⁺/BrdU⁺ cells in the injured adult spinal cord at 2, 4 and 8 wpi. ***P < 0.001 by Student’s t-test (n = 5 mice per group; n.d., no detected). The scale bars represent 50 µm.

Next, we performed immunohistochemistry to analyze the cellular identity of LC-induced CiNs in the injured spinal cord. The CiNs were identified by co-staining with BrdU and TUBB3 (Fig. 7). Immunostaining showed that Neither choline acetyltransferase (ChAT), a marker for cholinergic motor neurons, nor vesicular glutamate transporter 1 (VGLUT-1), a marker for excitatory neurons, were found to be expressed in CiNs. (data not shown). Interestingly, as shown in Fig. 7 A and B, The expression of Gamma aminobutyric acid (GABA) and Glutamic acid decarboxylase 65/67 (GAD6), two kinds of inhibitory neuronal markers, were detected in CiNs. Importantly, some CiNs were found to be co-labelled by presynaptic terminal marker SYN1. Confocal analysis revealed that the dense SYN1 labeled bouton-like terminals were juxtaposed to the soma and processes of CiNs, suggesting that they have integrated into the local neural circuits (Fig. 7C and D). Together, these results suggest that LC can reprogram resident reactive astrocytes in the damaged adult spinal cord into synapse-forming GABAergic interneurons.

Fig. 7.

LC-induced neuronal cells resemble GABAergic interneurons in the adult spinal cord. (A and B) Confocal images showing LC-induced newborn neuronal cells (indicated by BrdU⁺/TUBB3⁺) express inhibitory neuronal markers, GABA and GAD6. (C and D) Confocal images showing synapse-formation on the soma (C) and processes (D) of LC-induced newborn neuronal cells. The scale bars represent 10 µm.

To determine the relative contributions of these two compounds to the in vivo chemical reprogramming, we further performed experiments by injection of LDN193189 or CHIR99021 alone. When the mice were treated with a single molecule, LDN193189 or CHIR99021, no DCX⁺ cells were detectable in the damaged spinal cord (data now shown). Similarly, albeit LC treatment resulted in induced mature NeuN⁺/TUBB3⁺/BrdU⁺ neuronal cells, they were not observed in the damaged spinal cord of mouse that received either CHIR99021 or LDN193189 injections alone (Supplementary Figure 12). Together, these data suggest that LDN193189 and CHIR99021 are both essentially required for the in vivo chemical compound-based neuronal reprogramming.

LC-induced neurogenesis originates from spinal astrocytes

The next goal was to identify the cellular source of CiNs generated in the injured adult spinal cord. Initially, we performed immunohistochemical analysis early in the LC-mediated reprogramming process. As shown in Fig. 8A, DCX⁺ cells were produced at 5 dpi upon LC treatment, and some of them were also immunoreactive for ALDH1L1, a marker specific to astrocytes, highlighting a potential astroglial origin of the induced neuronal cells. In comparison, none of them were co-labelled by oligodendrocyte lineage marker OLIG2 or microglia marker IBA1, excluding the potential derivation from these two cell types (Fig. 8B and C). Therefore, the most likely cellular source for the LC-induced new neuronal cells is spinal astrocytes.

Fig. 8.

LC-induced new neuronal cells originate from spinal astrocytes. (A-C) At the early stage of chemical reprogramming, a fraction of LC-induced DCX + cells were co-labelled by ALDH1L1 (astrocyte marker), whereas no LC-induced DCX⁺ cell was co-labelled by OLIG2 (oligodendrocyte lineage marker) or IBA1 (microglia marker). (D-F) Tracing LC-induced new neuronal cells by tail intravenous injection of AAV2/9-GFAP-mCherry virus. (D) Experimental scheme. (E and F) Representative images showing that LC-induced DCX⁺ and NeuN⁺/TUBB3⁺ cells originated from virus-infected astrocytes (indicated by mCherry⁺). (G and H) Fate mapping analysis of LC-induced new neuronal cells. FVB/N-Tg(GFAP-GFP)14Mes/J transgenic mice were treated with LC. LC-induced DCX⁺ and NeuN⁺/TUBB3⁺ cells were derived from GFP-traced spinal astrocytes. The scale bars represent 50 µm (A-C, E), 30 µm (G and H) and 20 µm (F).

Then, viral tracing and genetic fate mapping were used to further confirm the astrocytic origin of CiNs. We used an adeno-associated virus system (AAV2/9-GFAP-mCherry) to fluorescently label astrocytes. The viral solution containing AAV2/9-GFAP-mCherry was injected into tail vein 3 days before SCI, and immunohistochemical analysis of the source of CiNs was performed at 1 and 4 wpi (Fig. 8D). Supplementary Figure 13 showed that spinal astrocytes were specifically targeted by the AAV2/9-GFAP-mCherry virus. At 1 wpi, the majority of induced DCX⁺ cells were also mCherry immunoreactive, indicative of astrocyte-converted immature neuronal cells (Fig. 8E). At 4 wpi, a fraction of mCherry-traced cells was found to express TUBB3 and NeuN, suggestive of spinal astrocyte-derived mature neuronal cells (Fig. 8F). In sharp contrast, these neuronal markers were undetectable in mCherry-labeled cells in the spinal cord of DMSO-treated control mouse (Supplementary Figure 14). Additionally, astrocytes were also genetically traced using FVB/N-Tg(GFAP-GFP)14Mes/J transgenic mouse, in which an active GFAP promoter specifically drives GFP reporter expression. In the spinal cord of this tracing mouse, the astrocyte markers GFAP and ALDH1L1 were expressed by the vast majority of GFP⁺ cells, while few of these cells were immunopositive for oligodendrocyte lineage marker OLIG1 (Supplementary Figure 15). Neither neuronal marker NeuN nor microglia marker IBA1 were detectable in GFP-positive cells (Supplementary Figure 15). These findings suggest that spinal astrocytes are exclusively traced in the mice. Adult FVB/N-Tg(GFAP-GFP)14Mes/J transgenic mouse was then subjected to SCI and treated with LC. When examined at 5 dpi, we observed that GFP-traced DCX⁺ cells were induced in spinal cord (Fig. 8G). Quantitatively, there were 92.19 ± 3.33 % and 91.48 ± 2.95 % of DCX⁺ cells labelled by GFP at 5 and 7 dpi, respectively. At 4 wpi, we found that a portion of GFP- labelled cells was co-stained by neuronal markers, NeuN and TUBB3, in the spinal cord of mouse that received LC treatment (Fig. 8H), whereas these neuronal markers were not detectable in the spinal cord of mouse that received DMSO treatment (data not shown). Collectively, these results strongly suggest that resident spinal astrocytes are the primary source of LC-induced CiNs.

LC treatment induces neurogenesis in the injured aged spinal cord

It is common knowledge that the ability of neural regeneration declines in aging mammals. Although age does not significantly affect the total number of astrocytes in the CNS of rodents, primates, or humans, astrocyte dystrophy is observed in aged brain [48]. Importantly, our previous study showed that age-dependent astrocytes exhibited heterogeneous susceptibility to neuronal reprogramming in vitro [37]. Thus, we next asked whether LC treatment could also drive astrocyte-to-neuron conversion in the damaged spinal cord of aged mouse. To address this issue, aged mouse (greater than 12 months) with SCI received LC via intraperitoneal injection, and BrdU incorporation was performed to trace the induced new neuronal cells (Fig. 9A and B). At 4 and 8 wpi, immunohistochemistry showed that NeuN⁺/TUBB3⁺/BrdU⁺ neuronal cells were observed in the injured spinal cord of mouse that received LC treatment (Fig. 9C and D). Of note, quantitative analysis revealed much less neuronal induction efficiency of astrocytes in aged spinal cord than that in adult spinal cord (Fig. 6G, Fig. 9E). Therefore, our findings suggest that the LC-mediated astrocyte-to-neuron conversion tends to be a little difficult, albeit CiNs are induced.

Fig. 9.

LC induces neurogenesis in the injured spinal cord of aged mouse. (A and B) Schematic diagram of experimental procedures. (C and D) Representative images of LC-induced NeuN⁺/TUBB3⁺/BrdU⁺ cells in the injured spinal cord of aged mouse at 4 and 8 wpi. (E) Quantification of LC-induced NeuN⁺/TUBB3⁺/BrdU⁺ cells in the injured spinal cord of aged mouse at 4 and 8 wpi. ***P < 0.001 by Student’s t-test (n = 3 mice per group). The scale bars represent 50 µm.

Discussion

In this study, we identified a chemical cocktail consisting of just two compounds that successfully converted reactive astrocytes into neuronal cells in the dmaged spinal cord of adult mice. Typical neuronal morphologies and neuron-specific marker expression were observed in these CiNs. Importantly, they can develop into mature neuronal cells and survive more than 12 months in the lesioned spinal cord. Collectively, these findings suggest that the lineage fates of resident glial cells can be converted by only small molecules without introducing exogenous cell-fate determiners, such as transcriptional factors or microRNAs. Although it is still a proof of concept, the transgene-free approach to chemical reprogramming in vivo offers a novel strategy for CNS repair.

Using neural transcription factors, several groups, including our own, have successfully converted endogenous glial cells into neurons in spinal cord or brain over the last decade, attracting attention of neurobiologists [12], [13], [14], [15], [16], [42], [43], [45], [49], [50]. However, genetic reprogramming needs to face many challenges including efficiency, safety, and gene delivery. Therefore, many efforts are being focused on small molecule-based cell fate reprogramming, a non-viral and non-integrating approach. In fact, pioneering work by several laboratories has shown that fibroblasts and astrocytes can be converted into neurons by chemical compounds in vitro [34], [35], [36], [51], [52]. It's interesting to note that Ma et al. showed that endogenous astrocytes can be reprogrammed into neurons in the intact adult brain, including the striatum and cortex, by a chemical cocktail with five compounds (Forskolin, ISX9, CHIR99021, I-BET151, and Y-27632) [53]. However, it remains unknown whether chemical compounds can drive neuronal reprogramming in the adult spinal cord, even under injured conditions. Here, we tested a variety of chemical compounds that target signaling pathways and epigenetic modifications involved in neurodevelopment and neurodifferentiation. After screening, we identified a compound combination of LDN193189 and CHIR99021 that could chemically reprogram reactive astrocytes into neuronal cells in the crushed spinal cord. LDN193189 is an inhibitor of the bone morphogenetic protein (BMP) signaling pathway that plays a key role in astroglia lineage commitment [54]. CHIR99021 can act as a WNT agonist and stabilize β-catenin by selectively inhibiting glycogen synthase kinase 3β (GSK3β), which is essential for inducing neuronal differentiation [55]. Previous studies showed that LDN193189 and CHIR99021 were core small molecules widely used for chemical reprogramming in vitro [34], [35], [36], [51], [52]. Of note, these two small molecules combined with other compounds were shown to induce cultured human stem cells (hSCs) differentiating into spinal cord neurons [56]. By a series of screening, surprisingly, we found that the chemical combination of LDN193189 and CHIR99021 not only converted cultured astrocytes into neuronal cells but also induced neurogenesis in damaged adult spinal cord. However, Neither LDN193189 nor CHIR99021 on their own could cause neuronal reprogramming in vivo or in vitro, suggestive of a synergistic effect of these two small molecules on the chemical conversion. It is reasonable to assume that LDN193189 may first disrupt the cell fate maintenance of astrocytes to exit their lineage by inhibiting BMP pathway, and then CHIR99021 induces them changing their cell fate toward neuronal cells by repressing the WNT/GSK3β/β-catenin pathway.

Despite the fact that the precise molecular mechanisms underlying compound-induced reprogramming remain unknown, we discovered that endogenous neural transcription factors such as Ngn2, NeuroD1/2, and Myt1l were activated during the chemical conversion. Among these cell-fate-determining factors, the most dramatic upregulation at the transcriptional level was Ngn2 and NeuroD1. Interestingly, the transcriptional activation of Ngn2 and NeuroD1 was also observed in the process of neuronal reprogramming mediated by other combination of small molecules [34], [35], [36], [51], [52], highlighting that they may play critical roles in the chemical conversion. As a neuron-fate-determining proneural gene, Ngn2 overexpression in postnatal cortical astroglia can convert them into glutamatergic neurons [11]. Additionally, Ngn2 has also been demonstrated to successfully induce the fate of neuronal cells with additional transcription factors or chemical compounds [57], [58]. As a member of the basic helix-loop-helix (bHLH) transcription factor family, NeuroD1 has an impact on the fate of specific neuronal cells, which is essential for the adult neurogenesis [59], [60], [61]. Both in vitro and in vivo, it has been demonstrated that astrocytes can be converted into neurons by forced expression of NeuroD1 [12]. Another bHLH transcription factor, NeuroD2, has been identified as an early-onset neuronal transcript required for neuronal development and survival in CNS [62]. Myt1l is a pan neuron-specific transcription factor that actively represses multiple non-neuronal fates to maintain neuronal identity and has widely used for neuronal reprogramming [2], [63], [64], [65]. When these neural transcription factors were activated, the transcriptional level of the immature neuronal marker DCX increased significantly, followed by a gradual upregulation of mature neuronal markers such as NeuN, Map2, and Syn. It is of note that the transcriptional changes of neuronal markers are after that of neural transcription factors, suggestive a hierarchical transcriptional activation process for the small molecules-mediated neuronal fate patterning. Importantly, this staged gene activation is companied by the typical morphological changes from astrocytes to neuron and the expression of neuronal markers. Further studies are needed to understand how the small molecules elicit the sequential activation of the endogenous genes for neuronal reprogramming. It is of note that no significant activation of the enriched genes in pluripotent and neural stem cells, Sox2, Pax6, and Blbp, was detected during the chemical conversion. We also found that the chemically induced DCX⁺ cells did not form clusters and were rarely immunoreactive for PCNA in early stages in spinal cord. These results indicate that the chemical conversion is a direct cell fate reprogramming without passing a highly proliferative progenitor intermediate.

After the chemical cocktail was identified in vitro, they were further used for inducing cell-fate conversion in vivo. Firstly, the small molecules were intraperitoneally injected into the healthy adult mice. However, no significant neuronal reprogramming was observed in the intact spinal cords. Of note, a recent study reported that intraperitoneal injection of LDN193189 and CHIR99021 combined with other compounds also failed to induce appreciable chemical conversion inside healthy mouse brain [52]. We then intraperitoneally injected the small molecules into the adult mice with crush-injured SCI. To our surprise, we found that the chemical combination successfully induced appreciable neurogenesis in the damaged spinal cord. Continuous BrdU labeling after spinal cord injury revealed that reactive glial cells contributed to CiNs. Using the AAV2/9-GFAP-mCherry viral tracing system, we demonstrated that the CiNs are derived from resident spinal reactive astrocytes but not neural progenitor cells or other cell types. Importantly, the genetic fate mapping with FVB/N-Tg(GFAP-GFP)14Mes/J mice also provided direct genetic proof for the resident reactive astrocyte origin of CiNs. However, future studies need to clarify why and how this type of cells were specifically targeted by the small molecules. Given the fact that the small molecule-driven neuronal conversion was observed in injured but not noninjured spinal cord, we think that the injury-induced cellular milieu may facilitate fate reprogramming. Consistent with our study, strong environmental effects on in vivo direct neuronal reprogramming have also been reported by several groups [45], [66]. Astrocytes are a major glial cell type in the CNS and play integral trophic, structural, and metabolic roles to maintain are responsible for diverse functions to maintain neuronal homeostasis in the developing brain and the adult brain. Under physiological conditions, astrocytes are quiescent and largely stop proliferation in adult rodents. They may be refractory to chemical reprogramming. However, quiescent astrocytes become activated and begin to proliferate again after CNS injury or neurodegenerative disorders. Reactive astrocytes undergo morphological, molecular, and functional remodeling and even dedifferentiate to take on some progenitor cell characteristics [67], [68], [69]. It has been well documented that reactive astrocytes fail to generate neurons in vivo and genetically remain within glial lineages [70]. However, the injury-induced reactive state may render astrocytes more prone to adopting favorable fates, providing a potential target cell population our small molecule-mediated fate conversion. In fact, we found that the majority of CiNs were found in close proximity to the lesion site in spinal cord. Importantly, because of failure in reprogramming quiescent astrocytes into neuronal cells in intact spinal cord, it seems that our chemical cocktail has little effect on normal astrocytes and may not affect their functions. Of note, the robust synaptic puncta were observed on the soma and along the processes of CiNs, suggestive of the establishment of a functional synaptic network. However, our current chemical approach has a low reprogramming efficiency and a small number of CiNs. In order to produce more neuronal cells and even to induce subtype-specific neuronal cells that are necessary for functional improvement following SCI, further refinement of our chemical cocktail and reprogramming approach is required.

Conclusion

In this study, we identify a chemical cocktail with only two chemical compounds that can reprogram astrocytes into neuronal cells. Our proof-of-principle study provides a transgene-free method for the chemical generation of induced neuronal cells in the damaged adult spinal cord. Albeit our current chemical cocktail has a low reprogramming efficiency, it suggests a new path to regenerate neurons for CNS repair. Taken together, our study raises the possibility of in vivo chemical reprogramming strategies for therapeutic application in neuroregenerative medicine. Future studies should focus on further refining our chemical cocktail and reprogramming approach to boost the reprogramming efficiency, which will enhance its functional repair potential.

Compliance with Ethics Requirements

All Institutional and National Guidelines for the care and use of animals (fisheries) were followed.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary material

The following are the Supplementary data to this article: