Human retinal ganglion cell axon regeneration by recapitulating developmental mechanisms: effects of recruitment of the mTOR pathway

Axons regenerate poorly in the central nervous system (CNS), leading to functional deficits, often permanent, following brain disease or injury (Curcio and Bradke, 2018; Liu et al., 2011). Therefore, therapeutic facilitation of axonal regeneration is key to functional recovery. The optic nerve, which progressively degenerates in glaucoma leading to irreversible blindness (Almasieh et al., 2012), provides an accessible and facile model of axonal regeneration. The optic nerve contains the axons of retinal ganglion cells (RGCs), which connect the retina to higher centers in the brain for visual perception and therefore their degeneration leads to permanent vision loss. There is no effective treatment for RGC degeneration. Given this intractable situation, one of the approaches to negotiate the pathology of the disease is to understand the mechanism underlying RGC development, recapitulation of which may facilitate optic nerve regeneration, thus rescuing vision loss (Hilton and Bradke, 2017). Recently, several intrinsic factors have been identified that regulate axon regeneration (Curcio and Bradke, 2018; Liu et al., 2011; Moore and Goldberg, 2011). For example, deletion of negative regulators of the mTOR pathway, such as phosphate and tensin homolog (PTEN) or tuberous sclerosis 1 (TSC1), has been observed to promote robust axon regeneration in an optic nerve crush model (Park et al., 2008). Conversely, inhibition of the mTOR pathway by rapamycin abrogated PTEN/TSC1-mediated axon regeneration (Park et al., 2008). Subsequently, evidence has emerged supporting the influence of diverse factors, including the JAK-STAT3 pathway inhibitor SOCS3 (Smith et al., 2009), the Wnt pathway ligand Wnt10B (Tassew et al., 2017), the heterochronic gene Lin28a (Wang et al., 2018) and ubiquitin-like containing PHD ring finger (UHRF1) (Oh et al., 2018), on optic nerve regeneration. In each of these cases, it was observed that the mTOR pathway might have played some role in the regenerative process. Thus, current cumulative evidence suggests that the mTOR pathway is a key regulator of optic nerve regeneration and therefore a valid target for therapeutic RGC regeneration in glaucoma. However, these studies have been carried out in rodents; therefore, their clinical relevance in the progression of the disease and therapeutic regeneration remains poorly understood. Here, we have tested the hypothesis that the mTOR pathway regulates the development of human RGCs (hRGCs) and that its recruitment upon injury facilitates axon regeneration.

The mTOR pathway is a ubiquitous nutrient-sensing intracellular pathway that promotes growth and survival of cells mainly by regulating protein synthesis (Costa-Mattioli and Monteggia, 2013). It is suppressed in response to cellular stress and is associated with abnormalities in the developing nervous system. mTOR pathway activation involves inhibition of the TSC1/2 complex via the receptor-mediated activation of AKT kinases. The disinhibited mTOR complex 1 [raptor (RPTOR)-containing and rapamycin-sensitive] stimulates mRNA translation by activating ribosomal protein S6 and the elongation factor eIF4E. Inhibition of TSC1/2 activates the mTOR pathway and is, therefore, a practical approach to examine the influence of the mTOR pathway in development and injury (Choi et al., 2018; Huang et al., 2008; Park et al., 2008; Tee et al., 2016; Maiese et al., 2013). The mTOR pathway, which remains active during development, is downregulated in the adult brain and is further decreased after injury (Costa-Mattioli and Monteggia, 2013; Garza-Lombó and Gonsebatt, 2016; Lee, 2015; Morgan-Warren et al., 2013). Here, we demonstrate that the mTOR pathway constitutes an evolutionarily conserved regulatory mechanism underlying the development and regeneration of hRGCs. For example, the mTOR pathway was active during RGC differentiation and its experimental activation facilitated the differentiation of hRGCs from induced pluripotent stem cells (iPSCs). The positive influence of the mTOR pathway on differentiation included enhanced neuritogenesis and axon guidance and growth; hRGCs expressed a battery of axon guidance molecules and the response of their axons to netrin-DCC interactions was facilitated by the activated mTOR pathway. In contrast, inhibition of the pathway adversely affected hRGC differentiation, including neurite complexities and axon growth. The biochemical and morphological changes extended into hRGC physiology; the spontaneous action potentials and local networking and synaptogenesis facilitated by activating the mTOR pathway were abrogated when it was inhibited. Furthermore, we demonstrate that hRGCs possess the ability to regenerate their axons following chemical axotomy and that the mTOR pathway regulates the efficiency of axon regeneration. Together, our findings demonstrate the involvement of the mTOR pathway in RGC development and posit it as a valid target for examining the mechanism(s) underlying hRGC degeneration and the potential of therapeutic regeneration of the optic nerve in glaucoma. This information will also improve our understanding of axon regeneration in the CNS in general.

To determine the involvement of the mTOR pathway in RGC differentiation, we first examined its status during retinal development. Because of the unavailability of human fetal tissues, initial experiments were carried out in rats with the rationale that results may be extrapolated across species because of the evolutionarily conserved developmental mechanisms of retinal histogenesis (Hoshino et al., 2017). Our primary readout for activated mTOR signaling was the detection of immunoreactivities corresponding to phospho-S6 (Ser240/244); activated mTOR phosphorylates p70 ribosomal S6 kinase (p70S6K), which in turn phosphorylates and activates S6, a ribosomal protein that regulates protein translation (Saxton and Sabatini, 2017). Our secondary readout was the expression of transcripts and immunoreactivities corresponding to the TSC2, which inhibits the mTOR1 complex and correlates negatively with activation of the mTOR pathway (Saxton and Sabatini, 2017). We observed that phospho-S6 (p-S6) staining was robust at stages representing both early [embryonic day (E) 14] and late (E18) histogenesis but was undetectable in adult retina, suggesting active mTOR during retinal development and its cessation upon maturity (Fig. 1A-C). In both E14 and E18 retina, p-S6 staining was confined to the inner retina, where nascent RGCs, characterized by the expression of BRN3A (POU4F1), were localized (Fig. 1A-C). Quantification of immunoreactive cells revealed that the majority of p-S6-positive cells (p-S6⁺) co-expressed BRN3A in E14 and E18 retina and that p-S6⁺ BRN3A⁺ cells were undetectable in the adult retina (Fig. 1D), indicating a correlation between mTOR activity and RGC development. Such correlation was further shown by the temporal expression patterns of TSC2 protein (Fig. 1E) and transcripts (Fig. 1F), which were inverse of those for the transcriptional regulators of RGC differentiation, Atoh7, and Brn3B (Fig. 1G,H). Next, to determine if this correlation was conserved during human retinal development, we curated RNA sequencing data obtained from different stages of fetal and adult human retina (Hoshino et al., 2017). We found that the expression patterns of TSC2 and of ATOH7 and BRN3B in the developing human retina were similar to those observed in rats (Fig. 1I-K).

Together, these observations suggest that the conserved decline in mTOR pathway activity, indicated by the temporal activation of TSC1/TSC2 expression, might be functionally correlated with hRGC differentiation. This notion was tested in the iPSC model of human neurogenesis. We have recently demonstrated that both rodent and human pluripotent cells can be directly differentiated along the RGC lineage by chemical recruitment of RGC developmental mechanisms (Teotia et al., 2017a). The efficiency of RGC generation was ∼40% (Fig. S1A,B). Human iPSCs were differentiated into RGCs when the mTOR pathway was activated by lentivirus-mediated transduction of TSC2-shRNA (hRGC^TSC2-shRNA) or repressed by the exposure of cells to rapamycin, an inhibitor of the mTOR1 complex (Saxton and Sabatini, 2017) (hRGC^Rapamycin+; Fig. 2A). Controls included cells transduced with the empty lentivirus (hRGC^Control). The transduction efficiency was ∼80% (Fig. S1C). To address the concern that the results might be due to the function of pluripotent cells, experiments were carried out in parallel with retinal progenitor cells (RPCs) enriched from the developing rat retina (Fig. S2A). The transduction efficiency of TSC2-shRNA lentivirus in rat RPCs was ∼70% (Fig. S1C).

Transduction of TSC2-shRNA lentivirus led to significant silencing of both TSC2 protein and transcript expression in both human (h)iPSC- (Fig. 2B, Fig. S3) and rat RPC- (Figs S3B and S4) derived RGCs, compared with those transduced with control lentivirus. This result was accompanied by a significant increase in the levels of p-S6 in hRGC^TSC2-shRNA groups versus controls (Fig. 2C). In contrast, exposure to rapamycin led to a significant decrease in p-S6 levels in hRGC^Rapamycin groups, versus controls (Fig. 2C). Levels of S6 remained unchanged across the groups (Fig. 2D). That the effect of rapamycin was not due to its cytotoxicity was revealed by the absence of significant changes between all groups in a cell survival assay (Fig. S5). Together, these results demonstrated that TSC2-shRNA and rapamycin facilitated and inhibited the mTOR pathway, respectively, in RGCs generated from hiPSCs and rat RPCs. Next, we examined the consequence of perturbation of mTOR signaling on RGC generation. Quantification of hRGCs (GFP⁺β-tubulin⁺ cells) revealed that, in comparison to controls, there was a significant increase (87.55±1.97% versus 76.00±0.57%, P=0.0049) and decrease (30.55±2.77% versus 76.00±0.57%, P≤0.0001) in the proportions of these cells in hRGC^TSC2-shRNA and hRGC^Rapamycin groups, respectively (Fig. 2E-H). That the mTOR pathway specifically influenced the generation of hRGCs and E18 RGCs was further corroborated by similar changes in the proportion of cells expressing the RGC regulator BRN3A (GFP⁺BRN3A⁺ cells) as that of GFP⁺β-tubulin⁺ cells in response to a perturbation in the mTOR pathway (Figs S6 and S7). The results obtained by immunocytochemical analysis was consistent with the readout of mTOR pathway activities at the single cell level; there was a significant increase and decrease in the proportion of p-S6⁺GFP⁺ cells in hRGC^TSC2-shRNA and hRGC^Rapamycin groups, respectively, compared with controls (Fig. 2I-L), whereas the proportion of S6⁺GFP⁺ cells remain unchanged across the group (Fig. 2M-P). Similar effects of mTOR pathway perturbations were observed on the generation of RGCs from rat RPCs in response to perturbation of the mTOR pathway (Fig. S2).

It has been observed that the mTOR pathway regulates dendritic morphology and axon formation and projection of rodent RGCs (Morquette et al., 2015; Nie et al., 2010). To test the premise that the mTOR pathway regulates neuritogenesis and axon growth we first determined the complexities of neurites elaborated by RGCs by Sholl analysis (Sholl, 1953). This and all subsequent experiments were done on hRGCs. The analysis demonstrated that both the number of intersections and distance of the processes from the soma were significantly greater in cells belonging to the hRGC^TSC2-shRNA groups, compared with controls (Fig. 3A-D). In contrast, the number of intersections and distance of the processes from the soma were significantly reduced in the hRGC^Rapamycin groups versus controls (Fig. 3A-D). These observations suggest that the mTOR pathway has a significant impact on the generation and complexities of neurites, which might involve the recruitment of factors that regulate the growth and guidance of the neurites (Nie et al., 2010). To test this premise, we first examined the expression of growth regulatory factors – the Kruppel-like family of transcription factors (KLFs) and SRY-related HMG-box transcription factor 11 (SOX11) (Li et al., 2018; Moore and Goldberg, 2011) (Fig. 4). Fifteen of the 17 known KLFs are expressed in the rodent retina and their expression patterns, particularly that of KLF4, KLF6, KLF7 and KLF9, are associated with the developmental decline in axonal growth of RGCs after birth (Moore et al., 2009; Moore and Goldberg, 2011). For example, KLF4 and KLF9, expression of which increases postnatally, and that of KLF6 and KLF7, which decreases postnatally, repress and facilitate axonal growth, respectively, in an experimental model (Moore et al., 2009; Moore and Goldberg, 2011). Sox11 is involved in RGC differentiation and optic nerve development and has been observed to promote regeneration of RGCs in rodents (Li et al., 2018; Norsworthy et al., 2017). Quantification of specific transcripts following perturbation of the mTOR pathway in hRGCs revealed that expression of KLF6 and SOX11 increased and of KLF4 decreased in hRGC^TSC2-shRNA groups compared with controls (Fig. 4A-C). In contrast, the expression of KLF4 was upregulated and that of KLF6 and SOX11 attenuated when the mTOR pathway was inhibited, suggesting a functional correlation between the mTOR pathway and expression of these growth regulatory factors during neuritogenesis (Fig. 4A-C). Second, we examined the expression of genes encoding guidance molecules that regulate the pathfinding of RGC axons, specifically those involved in intra-retinal guidance (de la Torre et al., 1997) (e.g. DCC), guidance at the optic chiasm (Erskine et al., 2000) (e.g. ROBO2), retinotopic connection in the superior colliculus (Hindges et al., 2002) (e.g. EPHB3), and general growth of the processes (Kruger et al., 1998) (e.g. GAP43). We observed significantly higher levels of transcripts corresponding to GAP43, ROBO2, DCC, EPHB3 and EPHA4 in cells belonging to the hRGC^TSC2-shRNA groups than in controls (Fig. 4D-H). Levels of these transcripts, except for DCC, remained unchanged in the hRGC^Rapamycin group, compared with controls. To address the possibility that the lack of influence of rapamycin on transcript levels was due to their global measurement, which might not detect small but significant changes in a subpopulation of cells, we quantified single hRGCs, expressing immunoreactivities corresponding to GAP43 (Fig. S8A-D) and ROBO2 (Fig. S8E-H) in the three groups. A significant increase and decrease in GFP⁺GAP43⁺ and GFP⁺ROBO2⁺ cells were observed in the hRGC^TSC2-shRNA and hRGC^Rapamycin groups, respectively, compared with controls, demonstrating that perturbation of the mTOR pathway affects the expression of guidance receptors/molecules. Lastly, in order to determine whether the mTOR pathway affected guidance molecule-dependent hRGC axon growth, we examined whether or not hRGCs responded to guidance through DCC-netrin interactions, which is essential for intra-retinal guidance and navigation through the optic nerve head (de la Torre et al., 1997). We observed an extensive elaboration of filopodia in growth cones of hRGCs exposed to netrin, compared with controls (Fig. 5A-D). The extended filopodia collapsed when exposed to netrin pre-incubated with DCC antibodies, demonstrating the specificity of netrin's effects on the morphology and motility of the growth cones.

Next, we examined whether activated growth cones resulted in the growth of axons. Time-lapse examination of the hRGCs' axons demonstrated a significant increase in their length when exposed to 50 ng/ml of netrin, compared with controls, over time (Fig. 5E-G). However, the difference between the experimental and control groups disappeared after 25 min. Having ascertained that hRGC growth cones respond to netrin, we examined the effect of the mTOR pathway on the efficiency of the growth of hRGC axons in the presence of netrin. This experiment was carried out in a microfluidic device in which the hRGC axons (characterized by the co-expression of Tau1/SM132) are segregated from the soma in the microgrooves of the device (Fig. S9). We observed a three-fold increase (3.61±0.84, P=0.0151) in the number of axons that reached the axonal chambers of the microfluidic devices in hRGC^TSC2-2shRNA groups compared with controls (Fig. 6). hRGCs exposed to rapamycin failed to extend axons in the microgroove (data not shown). Together, these results demonstrated that hRGCs were growth and guidance competent; they express guidance molecules and display a conserved response to them, facilitated by the mTOR pathway.

Next, we were interested to know whether the mTOR pathway's influence on hRGC differentiation included their functional features and maturity. To address this, we performed whole-cell voltage- and current-clamp recordings to measure voltage-gated currents, passive membrane properties, and spiking behavior. In voltage-clamp, a series of depolarizing voltage steps (−70 to +40 mV in 10 mV increments) revealed fast inward Na⁺ currents (I_Na) and slower outward K⁺ currents (I_K) (Fig. 7A-C). The amplitude of I^Na peaked at −20 mV and was similar in hRGC^Control and hRGC^TSC2-shRNA (4.1±0.5 nA in hRGC^Control, n=14; 4.3±0.6 nA in hRGC^TSC2-shRNA, n=14), but was dramatically reduced in hRGC^Rapamycin (1.7±0.3 nA, n=14, P=0.001). In current-clamp recordings, we monitored resting membrane potential (V_rest) and spontaneous spiking behavior (Fig. 7D-G). Both hRGC^Control and hRGC^Rapamycin were fairly depolarized at rest (hRGC^Control: −31±2 mV, n=7; hRGC^Rapamycin: −37±3 mV, n=7) and were not firing action potentials, presumably due to Na⁺ channel inactivation. hRGC^TSC2-shRNA, in contrast, were less depolarized (V_rest=−45±4 mV, n=9) and fired regular spontaneous action potentials at 0.9±0.2 Hz. We next applied a hyperpolarizing DC current to maintain V_rest at approximately −75 mV and recorded spiking behavior in response to depolarizing current injections (+10 to +70 mV, 500 ms). All groups fired fewer action potentials at very strong depolarizing injections due to Na⁺ channel inactivation and depolarization block (Fig. 7F,G). hRGC^TSC2-shRNA had a higher peak number of evoked spikes at +20 pA (3.9±0.6, n=9) and hRGC^Rapamycin fired 2.8±0.6 (n=9). hRGC^Rapamycin had a significantly elevated input resistance (2.4±0.2 GΩ, n=14) relative to hRGC^Controls (1.0±0.1 GΩ, n=11; P=0.00005) whereas input resistance for hRGC^TSC2-shRNA was not significantly different from controls (hRGC^TSC2-shRNA: R_in=1.3±0.2 GΩ, n=11; P=0.19) The differences in spiking patterns are likely the result of both the lower Na⁺ current amplitude in hRGC^Rapamycin and their relatively high input resistance, which means that depolarizing current injections will more readily move the membrane potential to spike threshold and, at higher stimulus strengths, to depolarization block.

We next used calcium imaging to measure spontaneous Ca²⁺ transients to examine further the effects of mTOR signaling on the spontaneously active neural network in culture (Teotia et al., 2017b) (Fig. 8). Cell fate specification leads to expression of a select combination of ion channels, which generate particular patterns of spontaneous activities, accompanied by transient elevations of intracellular calcium concentrations that sustain a specific neural network (Spitzer et al., 2004). We found that cells in hRGC^TSC2-shRNA and hRGC^Rapamycin groups displayed significantly higher and lower spontaneous Ca²⁺ transients versus cells in the hRGC^Control group, respectively (Fig. 8A-C). Spontaneous spikes were retained in cell bodies and did not propagate to processes in cells in the hRGC^Rapamycin group compared with cells in the hRGC^TSC2-shRNA and control groups (Fig. 8A-C, upper panels). Moreover, the mean amplitude of spontaneous Ca²⁺ transients was significantly larger in cells in the hRGC^TSC2-shRNA group than in controls (Fig. 8D). In contrast, exposure of cells to rapamycin almost erased the mean amplitude of spontaneous Ca²⁺ transients, compared with controls. Next, to determine the mechanism underlying the influence of mTOR signaling on the local neural network, we quantified the expression of the presynaptic protein synaptophysin (Fig. S10A-D) and the postsynaptic protein PSD95 (DLG4) (Fig. S10E-H) in response to a perturbation in the mTOR pathway. We observed that the proportion of cells expressing synaptophysin or PSD95 in the hRGC^TSC2-shRNA group was significantly higher than in the hRGC^Control group. In contrast, the proportion of these cells in the hRGC^Rapamycin group was significantly lower than in the hRGC^Control group (Fig. S10D,H). Together, these observations suggest that mTOR signaling facilitates functional maturation of RGCs, including the formation of spontaneously active local neural networks, presumably through synaptogenesis.

In the rodent retina, activation of the mTOR pathway has been found to promote axon regeneration in the optic nerve crush model (Park et al., 2008). Whether or not this premise holds true for hRGCs was tested in the microfluidic model of hRGC axotomy, established in our lab (Fig. 9A,B). Here, hRGC axons that had reached the axon chamber were axotomized by exposure to a detergent and allowed to regenerate back into the axon chamber, where the numbers of axons from lentivirus-transduced-hRGCs were quantified. hRGC axons degenerated within 10 min of exposure to the detergent. Remnants of degenerated axons in the axon chamber were removed by the wash, and culture medium restored in the chamber. The axons began to regenerate back into the axon chamber within 48 h (Fig. 9B), and their quantification in hRGC^TSC2-shRNA and control groups was conducted 5 days after axotomy, following labeling with CTB-CY3 to account for only those axons that were transduced (GFP⁺) (Fig. 9C-E). Quantification of GFP⁺CY3⁺ axons revealed that the proportion of axons in the axonal chambers was significantly higher in the hRGC^TSC2-shRNA group, compared with controls (75.33±1.76% versus 45.40±2.70%, P=0.0008) (Fig. 9C,D,F). The number of axons in the hRGC^TSC2-shRNA group fell significantly below that of controls when rapamycin was added to the culture medium (Fig. 9C,E,F). A recent study reported that alpha-RGCs (αRGCs) preferentially survive axotomy and selectively regenerate in mice owing to high endogenous levels of mTOR activity (Duan et al., 2015). We have determined the range of subtypes generated from human iPSCs (P.T. and I.A., unpublished observations). αRGCs are present among different RGCs subtypes (Fig. S11). Therefore, we examined whether or not hiPSC-derived αRGCs account for the regeneration in response to the activation of the mTOR pathway. We observed that the majority of regenerating axons (GFP⁺CY3⁺) also expressed immunoreactivities corresponding to neurofilament SMI32, an αRGC marker (Duan et al., 2015; Krieger et al., 2017; Lin et al., 2004) (Fig. 9G), suggesting that hiPSC-derived αRGCs are responsive to mTOR pathway-mediated regeneration. Next, to confirm that other pathways implicated in the regeneration of axons on rodent are functional in hRGCs and may be interactive with the mTOR pathway, we examined whether recruitment of the JAK-STAT3 pathway alone (Fischer, 2017) or in combination with the mTOR pathway (Leibinger et al., 2016; Park et al., 2008; Smith et al., 2009) could facilitate hRGC axon regeneration. It has been recently demonstrated that hyper-interleukin-6 (hIL6), a designer cytokine that directly binds to the signaling receptor GP130 (IL6ST) without the need of additional co-receptors to activate the JAK-STAT3 pathway, promotes RGC neurite growth in vitro and axon regeneration in the optic nerve crush model (Fischer, 2017). hRGCs were transduced with Baculovirus-hIL-6/Baculovirus-GFP control virus and subjected to axotomy and regeneration protocol as described above (Fig. 10). Quantification of GFP⁺CY3⁺ axons revealed that the proportion of axons in the axonal chambers was significantly higher in the hRGC^hIL-6 group, compared with controls (71.00±2.43% versus 49.87±6.23%, P=0.0342) following axotomy (Fig. 10A,B,D). The number of axons in the hRGC^hIL-6 group decreased significantly to the levels observed in controls when AG490, an inhibitor of the JAK-STAT3 pathway (Leibinger et al., 2016), was added to the culture medium, demonstrating a positive influence of the JAK-STAT3 pathway on hRGC axon regeneration (Fig. 10A,C,D). The lack of abrogation of regeneration beyond the levels in control was likely due to the low concentration (10 µM) of AG490, which was used to prevent cytotoxicity (Yadav et al., 2005). Furthermore, we observed that the proportion of axons in the axon chamber following axotomy was significantly higher in hRGCs co-transduced with TSC2-shRNA lentivirus and hIL-6 baculoviruses than those transduced with either the former or latter alone, demonstrating an additive effect of the mTOR and JAK-STAT3 pathways on axon regeneration (Fig. 11A,B,D,E). The number of regenerating axons in double-transduced RGCs decreased significantly in the presence of rapamycin+AG490, compared with controls (Fig. 11A,C,D). Together, these observations suggest that hRGCs axons display regenerative capacity, which could be influenced by the mTOR pathway alone or in concert with the JAK-STAT3 pathway.

The mTOR pathway is a ubiquitous nutrient-sensing intracellular pathway that promotes growth and survival mainly by regulating protein synthesis (Costa-Mattioli and Monteggia, 2013; Sarbassov et al., 2005; Swiech et al., 2008; Saxton and Sabatini, 2017). It is suppressed in response to cellular stress and is associated with abnormalities in the developing nervous system (Costa-Mattioli and Monteggia, 2013; Swiech et al., 2008). That the mTOR pathway is intrinsically involved in CNS axon regeneration was demonstrated when genetic deletion of PTEN in adult RGCs was observed to promote robust regeneration of their axons after optic nerve crush (Park et al., 2008). The abrogation of PTEN deletion-mediated regeneration in the presence of rapamycin and its restoration upon the genetic deletion of TSC2, further determined that RGC axon regeneration is facilitated when the mTOR pathway is activated (Park et al., 2008; Choi et al., 2008). Subsequently, similar approaches to perturb the pathway demonstrated its involvement elsewhere in the CNS, including in the regeneration of axons in the corticospinal tracts (Du et al., 2015; Liu et al., 2010; Zukor et al., 2013). We demonstrate that, much like in rodents, the mTOR pathway plays an important role in the development of hRGCs (Choi et al., 2019) and regeneration of their axons (Park et al., 2008), and is therefore an important molecular target for establishing a disease model of glaucomatous degeneration and therapeutic RGC regeneration. This premise is supported by the following observations. First, the mTOR pathway activity was temporally correlated with RGC differentiation in both rat and human retina, suggesting a conserved relationship between the pathway and RGC development. Second, the efficiency of hRGC differentiation was low, their morphology abnormal in terms of neurites, and function compromised when the mTOR pathway was inhibited during differentiation. These abnormalities, compared with controls, were rescued when the mTOR pathway was experimentally activated, suggesting that, regardless of specific mechanisms for lineage-specific differentiation, a basal level of the mTOR activity within the differentiating RPCs may be essential for the optimal differentiation of hRGCs. This notion is supported by observations that in rodents the mTOR pathway regulates (1) dendritic morphology of embryonic neurons in vitro (Jaworski et al., 2005) and mature RGCs upon injury to the optic nerve (Morquette et al., 2015), and (2) axon RGC growth/projection (Nie et al., 2010). Third, hRGCs expressed a battery of guidance molecules, and, as observed for rodents (de la Torre et al., 1997; Erskine and Herrera, 2007; Marcus et al., 1996; Stuermer and Bastmeyer, 2000), the response of their growth cones to those molecules was influenced by the activity of the mTOR pathway, suggesting that it supports the growth and guidance of hRGC axons. Furthermore, the mTOR pathway, besides involving guidance molecules, may recruit intrinsic factors such as KLF4, KLF6 and SOX11 to regulate axon growth. Lastly, axons of hRGCs, particularly those expressing the αRGC subtype marker SMI32, like those in rodents (Duan et al., 2015; Leibinger et al., 2012; Park et al., 2008), are capable of regeneration, the efficiency of which is regulated by the mTOR pathway, thus corroborating that mTOR activity-mediated axon regeneration is evolutionarily conserved from worms to humans (Byrne et al., 2014; Liu et al., 2010; Nawabi et al., 2012; Park et al., 2008).

Our observations suggest that, besides the mTOR pathway, other intrinsic factors and their interactions underlying axon regeneration are conserved in hRGCs. A variety of approaches that included constitutive expression of the effector of the JAK-STAT3 pathway, STAT3 (Pernet et al., 2013), conditional knockout of the JAK-STAT3 pathway inhibitor SOCS3, and the direct recruitment of the GP130 receptor by hIL6 (Leibinger et al., 2016) in rodent RGCs have demonstrated the essential role of the pathway in RGCs axon regeneration and growth following optic nerve injury. Evidence has emerged that the mTOR and JAK-STAT3 pathways may act in concert to promote RGC axon growth. For example, combined activation of both pathways by co-deletion with PTEN and SOCS3 (Sun et al., 2011) or continuous expression of hIL6 in combination with PTEN deletion causes robust axon regeneration (Fischer, 2017; Leibinger et al., 2016). That the JAK-STAT3 pathway is recruited in hRGCs axon regeneration and may act in concert with the mTOR pathway is suggested by hIL6-mediated axon regeneration following axotomy and the additive effect on the process when the mTOR pathway was co-activated. Though the evidence that cytokines increase the levels of p-S6 and that rapamycin inhibits the JAK-STAT pathway-mediated neurite and axon growth (Leibinger et al., 2016) suggests direct interactions between the two pathways, our observation suggests that the mTOR pathway may involve KLF4, a negative regulator of axon growth (Moore et al., 2009), for the ancillary recruitment of the JAK-STAT3 pathway. For example, we observed a significant decrease in KLF4 expression upon activation of the mTOR pathway. As KLF4 is known to inhibit the transcriptional function of STAT3 by binding directly to it (Qin et al., 2013), a decrease in its expression upon the mTOR activation may disinhibit STAT3, thus facilitating STAT3-mediated axon growth.

The mTOR-KLF4-JAK-STAT3 axis may be involved in the development of the disease phenotype in glaucoma. For example, the axis is dysregulated in hRGCs generated from the SIX6 risk allele-carrying primary open angle glaucoma (POAG) patient-specific induced pluripotent cells (Teotia et al., 2017b). SIX6, one of the eye field genes, regulates the generation of retinal cell types by maintaining retinal progenitors (Heavner and Pevny, 2012; Teotia et al., 2017b; Zuber et al., 2003). Therefore, its absence leads to hypoplastic retina in mouse due to the depletion of retinal progenitors (Li et al., 2002; Teotia et al., 2017b). POAG patients with this SIX6 risk allele variant (rs33912345; C>A; His141Asn) have reduced retinal nerve fiber layer (RNFL) thickness (Carnes et al., 2014; Iglesias et al., 2014). SIX6 risk allele RGCs displayed short and simple neurites and reduced expression of guidance molecules compared with controls. The differentially expressed genes mapping on the mTOR and JAK-STAT3 pathways were repressed in SIX6 risk allele hRGCs, compared with controls, suggesting that both pathways were inhibited in the disease condition. In contrast, expression of KLF4 was significantly higher in SIX6 risk allele RGCs versus controls. Together, these observations suggest that, given the observations that both the mTOR and JAK-STAT3 pathways play important roles in neurite growth and KLF4 inhibits it (Fischer, 2017; Moore et al., 2009; Park et al., 2008), the dysregulated axis as observed in SIX6 risk allele RGCs, if maintained in adulthood, may make them susceptible to degeneration. Clinical targeting of the mTOR pathway to rescue RGC degeneration may involve pharmacological activation of the pathway. Recently, a small molecule, NV-5138, has been shown to activate mTORC1 both in vivo and in vitro (Kato et al., 2019; Sengupta et al., 2019). Alternatively, to address the concern that pharmacological activation of the mTOR pathway may affect the function of other cell types because of its ubiquitous presence, RGCs could be targeted for silencing the TSC2 gene by adeno-associated virus sera 2 (AAV2) transduction, which is RGC specific in the adult retina (Hanlon et al., 2017; Harvey et al., 2009). In summary, with the caveat that hRGCs studied here do not reflect the age of the onset of glaucomatous degeneration, our study demonstrates that the mTOR pathway is an evolutionarily conserved modulator of hRGC development and regenerative ability, and thus emerges as a valid target for understanding glaucomatous degeneration and potential therapeutic regeneration.

The use of animals and experimental protocols were approved by the Institutional Animal Care and Use Committee, at the University of Nebraska Medical Center (UNMC) and conducted in accordance with the Association for Research in Vision and Ophthalmology (ARVO) Statement for the Use of Animals in Ophthalmic and Vision Research. Timed-pregnant Sprague Dawley rats (Charles River Laboratories) were used for harvesting retinal cells from E14 and E18 embryos. Rats were euthanized by CO₂ exposure followed by decapitation with sterile surgical scissors to ensure death.

hiPSCs generated from normal adult human peripheral blood mononuclear cells using a non-integrating approach as previously described (Teotia et al., 2017b) were maintained in mTeSR1 medium (Stem Cell Technologies) on Matrigel-coated 6-well plates. Cells were passaged at 70-80% confluency as small clumps using cell dissociation reagent (Stem Cell Technologies) and maintained in a 37°C/5% CO₂ incubator. Cell passaging was performed at a 1:3 to 1:4 split ratio every 5 days.

Differentiation of hiPSCs along the retinal lineage was performed using a previously published protocol (Lamba et al., 2010). The directed differentiation of both E18 and hiPSCs-derived neural rosettes (NRs) into functional RGCs was performed using chemically defined medium following our established protocol (Teotia et al., 2017a). Briefly, NRs were manually picked and plated onto Matrigel-coated dishes (Corning), and RGC differentiation was initiated by treating cells for 2 days with Shh (250 ng/ml), FGF8 (100 ng/ml) and DAPT (3 μM). RGC differentiation was facilitated by treatment with follistatin (100 ng/ml), Shh (250 ng/ml) and DAPT (3 μM) for 3 days. Finally, RGC maturation and survival were promoted by supplementing the medium with BDNF (100 ng/ml), forskolin (10 μM), NT4 (5 ng/ml), CNTF (10 ng/ml), cAMP (400 μM), Y27632 (10 μM) and DAPT (3 μM) for the next 10 days. The medium was changed every 2-3 days. All reagents were purchased from R&D Systems.

The TSC2-shRNA:GFP and Control-shRNA:GFP constructs (Di Nardo et al., 2009) were generous gifts from Prof. Mustafa Sahin (Boston Children's Hospital, USA). Lentivirus transduction was carried out as previously described (Xia et al., 2018). Recombination lentiviral particles were generated through transient transfection of T293 cells, using the ABM lentivirus packaging system (Applied Biological Materials). Viral particles were concentrated using a BioVision PEG lentivirus precipitation kit (BioVision). Viral particles were collected 48 h and 72 h after transfection and filtered through a 0.45 μm membrane. Virus titers were determined using the ABM lentivirus titration kit. Twelve hours after transduction, the virus-containing medium was replaced by fresh medium. Transduction efficiency was determined 24 h post-transduction by direct observation of GFP⁺ cells. GFP⁺ cells per total DAPI⁺ cells were used for quantitative analyses of transduction efficiency in five visual fields per coverslip.

Recombinant baculoviruses were produced using the ViraPower BacMam Expression System (Thermo Fisher) according to manufacturer's protocols. In brief, recombinant bacmid DNA was purified and transfected into adherent Sf9 cells (Thermo Fisher) using Cellfectin reagent to generate P1 recombinant baculovirus stock. Baculoviruses were amplified by inoculation of 50 ml Sf9 suspension cultures (106 cells/ml) in Sf-900 III SFM Medium supplemented with 12.5 U/ml penicillin/streptomycin (Biochrom) in 125 ml polycarbonate Erlenmeyer flasks with vent cap (Corning) with 1 ml virus stock solution and incubation at 27°C and 130 rpm for 2-3 days. Amplified viruses were purified and concentrated by ultracentrifugation of 27 ml virus-containing supernatant underlaid with 2.7 ml sucrose solution (25% sucrose with 5 mM NaCl and 10 mM EDTA in H₂O) in OptiSeal polypropylene tubes (Beckmann Coulter) at 80,000 g and 4°C for 80 min. Viral pellets were re-suspended in 0.5 ml PBS and passed through 0.22 μm low protein-binding sterile syringe filters (Merck Millipore). Baculovirus preparations were pre-tested on HEK293 cells (seeded at ∼3-5×104 cells per well in 96-well plates) by adding 1 μl virus per well overnight. Transduction efficiencies of ≥90% were regarded appropriate for further use. Otherwise, virus stock was subjected to further amplification cycles.

hRGCs were infected with TSC2-shRNA-GFP/Control-shRNA:GFP lentivirus at 1 day in vitro (DIV) of RGC differentiation in the presence of polybrene at 0.6 μg/ml. Twelve hours after infection the virus-containing medium was replaced by fresh RGC differentiation medium. GL3: hRGCs were exposed to rapamycin (Sigma) at a final concentration of 2.5 μM to inhibit the mTOR activity. After infection, hRGCs were kept in culture for an additional 14 days. The perturbation experiments were carried out three times in triplicate. For JAK-STAT3 pathway perturbation, hRGCs were transduced at 1 DIV with hIL6:GFP-baculoviruses or GFP (empty vector)-baculoviruses as described earlier (Fischer, 2017). hIL6 hRGCs were exposed to AG490 (Calbiochem) at a final concentration of 10 μM to inhibit JAK-STAT3 activity.

Polydimethylsiloxane microfluidic devices with 450 μm microgrooves (SND 450, Xona Microfluidics) were assembled and prepared as per the manufacturer's instruction. Briefly, cleaned, sterilized, and dry devices were reversibly attached to poly-D-lysine (PDL, Sigma)-coated cover glass by applying gentle force to seal them (Corning, 50×24 mm) (one device per cover glass) for axonal outgrowth and regeneration studies. The cover glasses were coated overnight with 1 mg/ml poly-D-lysine (Sigma-Aldrich). Once assembled, a solution of 1% Matrigel (Corning) in DMEM/F-12 medium (Invitrogen) was added to the four reservoirs/well (both side soma and axonal compartment) of the device. Devices were coated with Matrigel for at least 1 h at room temperature before cell seeding. After accutase dissociation at 5 DIV, hRGCs were re-suspended at 100,000 cells/ml in RGC maturation and survival media and loaded into the micro-channel in a 2 μl droplet. Cells were allowed to attach for at least 30 min, after which medium was added into the adjacent wells. The axonal compartment was filled with similar medium to facilitate axonal growth. Volumes in the wells were adapted to ensure flow across compartments to gradually provide medium and to establish a flow from the soma (200 μl per well) to the axonal (150 μl per well) compartment. Media in the devices was changed every 2 days.

Axotomy was performed between 10 and 12 DIV by first removing media from the axonal compartment and adding 50 μl RGC medium with 0.5% saponin (Sigma) for 3 min. To prevent the flow of the detergent into the soma compartment, a hydrostatic pressure was maintained by volume difference between soma (200 μl/well) and axonal (50 μl/well) compartments. At the end of a 3 min period, followed by two PBS washes, the axonal compartment was re-coated with 1% Matrigel for 30 min at 37°C. After Matrigel coating, media was returned immediately to the axonal compartment for the duration of the culture time. For the axotomy and regeneration experiment, hRGCs post-axotomy were cultured for an additional 5 DIV followed by retrograde labeling to identify regenerated axons. hRGC media post-axotomy was supplemented with rapamycin (Sigma; 2.5 μM), AG490 (Calbiochem; 10 μM) or rapamycin+AG490 to check the effect of perturbation of the mTOR, JAK-STAT3, and mTOR+JAK-STAT3 pathways, respectively.

Retrograde labeling was performed using 1% cholera toxin subunit B Alexa Fluor 594 conjugate (Thermo Fisher) dissolved in RGC media and added to the axonal compartment (100 μl per well) of the microfluidic device and incubated overnight at 37°C. After overnight incubation, the axonal compartment media was removed, rinsed, and replaced using fresh RGC culture media before imaging.

Protein samples (10-50 μg), extracted from hRGCs (Control, TSC2 and Rapamycin groups in triplicate) at the end of differentiation using RIPA buffer (Thermo Fisher), were denatured and separated by sodium dodecyl sulfate-polyacrylamide gel (12%) electrophoresis, and transferred onto a 0.45 µm PVDF-Plus Transfer Membrane (GE Water&Process Technologies). Membranes were incubated for 2 h at room temperature in blocking solution [5% non-fat milk powder dissolved in TBST buffer (0.15 M NaCl, 20 mM Tris, 0.05% Tween-20)], at room temperature for 2 h, followed by overnight incubation at 4°C with primary antibodies at recommended dilutions (Table S1). Membranes were washed and incubated with anti-mouse or anti-rabbit horseradish peroxidase-conjugated secondary antibodies (Cell Signaling Technology) for 1 h at room temperature and visualized with an enhanced chemiluminescence reagent (ECL Plus Western Blotting Detection System, Amersham) using Syngene gel documentation system (Fredrick). Immunoreactive bands corresponding to S6, p-S6 and GAPDH (loading control) were scanned using Gene Tools image analysis software (Syngene). All western blots shown are representative of at least three independent experiments.

For immunocytochemical analysis, cells were fixed in 4% paraformaldehyde for 15 min. After two rinses with PBS, fixed cells were exposed to 5% normal goat/donkey serum in PBS for 30 min at room temperature and permeabilized with Triton X-100 (0.4% or 0.2% for nuclear or cytoplasmic staining, respectively), followed by overnight incubation with primary antibody at 4°C. The next day, cells were treated with fluorescence (Cy3/FITC)-tagged secondary antibodies (Life Technologies) diluted in a 5% donkey serum and 0.2%/0.4% Triton X-100 solution for 2 h at room temperature, followed by three washes with PBS. Samples were mounted using VectaShield (Vector Laboratories) and fluorescent images were acquired with Zeiss ApoTome Imager M2 upright microscope (Axiovert 200M) and Axiovision 4.8 software (Carl Zeiss). The percentage of cells expressing specific markers was determined by counting immunopositive cells co-expressing GFP per total GFP-positive cells in five visual fields per coverslip. For each experimental condition, three biological replicates were used to calculate means and standard deviation. A list of antibodies and working dilutions is provided in Table S1. For growth cone analyses, Alexa Fluor 594-phalloidin staining (Cytoskeleton) for F-actin was carried out by incubation at 1:100 dilution in PBS containing 5% normal goat serum for 30 min after fixation and permeabilization, as per manufacturer's instruction. Neurite complexity and length quantification was performed using Fiji software (ImageJ, National Institutes of Health) (Schmidt and Kofuji, 2009). The percentage of GFP⁺CY3⁺ axons was calculated by counting the numbers of both GFP⁺ and CY3⁺ axons out of total numbers of CY3⁺ axons that crossed the microgroove and were present in an axonal compartment at the time of imaging.

hRGCs were plated on glass-bottom Petri dishes/microfluidics device for time-lapse imaging and images were acquired on an inverted confocal laser-scanning microscope (Zeiss 710 confocal laser scanning microscope, combined with Zeiss Zen Blue software) using an EC Plan-NEOFLUAR 60× objective lens with additional 1.5× magnification. Following 30 min of imaging, purified recombinant netrin-1 at 50 ng/ml concentration was bath-applied, and the same growth cones were imaged for an additional 30 min. For DCC function-blocking experiments, either anti-DCC (5 μg/ml, BD Biosciences) or the control antibody (control mouse IgG; 5 μg/ml, Millipore) was added to hRGC medium to achieve a final concentration of 10 μg/ml before and during the experiment. To check the response of Netrin-1 under the influence of the mTOR pathway, 50 ng/ml Netrin-1 was added to the hRGC medium in the axonal compartment in a microfluidics device. After 3 days, cells were stained with Alexa Fluor 594-phalloidin and imaged with a confocal microscope as above.

Quantitative polymerase chain reaction (Q-PCR) analysis was carried out as previously described (Teotia et al., 2017a). Total RNA from cells was extracted using Mini-RNeasy kit (Qiagen) according to the manufacturer's instructions. Five μg of total RNA per sample was used for reverse transcription into cDNA, using Superscript III RT kit, following the manufacturer's instructions. Q-PCR was performed using Quantifast SYBR Green Master Mix (Qiagen) on Rotor Gene 6000 (Corbett Robotics). All qPCR results represent each sample measured in triplicate. No-template blanks were used for negative controls. Amplification curves and gene expressions were normalized to the housekeeping gene GAPDH. The primer sequences used in this study are listed in Table S2.

Electrophysiological analysis was performed as described previously (Teotia et al., 2017b) using an upright fixed-stage microscope (Olympus BX51WI or Scientifica SliceScope Pro 6000) equipped with a 40× (Olympus) or 20× (Scientifica) water immersion objective lens. Cells were superfused with Ames' Medium (US Biologicals) at room temperature. Recording micropipettes (tip resistances 8-12 MΩ) were filled with an internal solution composed of (in mM) 120 potassium gluconate, 5 NaCl, 10 KCl, 2 EGTA, 10 HEPES, 5 ATP-Mg, 0.5 GTP-Na2, 5 phosphocreatine-Na2. The pH of the pipette solution was adjusted to 7.4 with KOH, and the osmolality was measured with a vapor pressure osmometer (Wescor) and adjusted to 270-275 mOsm. Reported voltages were corrected for liquid junction potential (10 mV). Electrophysiological recordings were obtained with Multiclamp 700A and 700B amplifiers, a Digidata 1500B digitizer, and pClamp 10.4 software (Molecular Devices). Na⁺ and K⁺ currents were evoked by voltage steps from −70 to +40 mV in 10 mV increments and action potentials evoked with a series of depolarizing current injections in current clamp (+10 to +70 pA) while passive membrane properties were measured with hyperpolarizing current injections (−10 to −40 pA). Resting potential was measured without any current injection, and a hyperpolarizing DC injection was used to hold the resting membrane potential at approximately −75 mV for measurements of evoked spiking behavior.

Calcium imaging in hRGC^Control, hRGC^TSC2 and hRGC^Rapamycin to investigate spontaneous Ca²⁺ transients was performed as described previously (Teotia et al., 2017b). Briefly, cells at the end of differentiation from all the groups were loaded with 3 μM Fluo-4-AM dye (Life Technologies) in HEPES-buffered physiological saline solution, supplemented with 0.04% Pluronic-F27 (Life Technologies) for 60 min at room temperature, followed by three to five washes with HEPES-buffered physiological saline solution. For each group, three Ca^2+-imaging sessions (each session contains three fields of view) were collected from independent samples. All imaging was carried out using an inverted confocal laser-scanning microscope (Zeiss 710 confocal laser scanning microscope, combined with Zeiss Zen Blue software) using an EC Plan-NEOFLUAR 10×/0.30 M27 objective lens (Carl Zeiss). Images were taken every 3 s, with 300 frames recorded per field. For representation and analysis, a total of three independent samples were tested per group and up to four 600 s videos of spontaneous calcium transients analyzed. Using Zen Blue software, a region of interest (ROI) was manually selected and mean pixel intensity of each ROI followed over time, generating time trace data for each ROI. The mean amplitude of spontaneous calcium transients was presented as relative fluorescence changes of averaged ROI [ΔF/F0=(F-F0)/F0; signal to baseline ratio of fluorescence]. The number of spontaneous calcium transients per minute was manually calculated.

Cells dissociated to a single cell suspension were fixed with 4% paraformaldehyde for 15 min and permeabilized with 0.4% Triton X-100 for additional 15 min at room temperature. Cells were further incubated for 1 h at 4°C with BRN3A antibody followed by two PBS washes. Cy3-tagged secondary antibody was used to allow for data acquisition and analysis. For Live/Dead assay, cells were measured using the LIVE/DEAD Fixable Dead Cell Stain Kit (Thermo Fisher Scientific). This method uses blue fluorescent reactive dye to stain the cells with damaged membranes as per manufacturer's protocol. Briefly, cells were pelleted, washed with 1× PBS, and stained with 1 µl of fluorescent reactive dye from the LIVE/DEAD Stain Kit. Cells were incubated on ice for 30 min in the dark. Data was captured using the LSRII flow cytometer system (Beckman Coulter). At least 10,000 events were recorded for each sample. Data analysis was performed BD FACSDiva software version 6.1.2.

Values are expressed as mean±s.e.m. Data were analyzed and plotted using GraphPad Prism (GraphPad). Statistical significance was determined by either a paired Student's t-test (two-tailed) or by one-way analysis of variance (ANOVA) for multiple group comparison. P-values of <0.05 were considered significant. Statistical analysis was performed when each group had at least three replicate samples from two independent experiments.

Thanks are due to Dr Mustafa Sahin for TSC2 lentivirus constructs and Helen Erickson for technical assistance.