Direct Reprogramming of Human Neurons Identifies MARCKSL1 as a Pathogenic Mediator of Valproic Acid-Induced Teratogenicity

SUMMARY

Human pluripotent stem cells can be rapidly converted into functional neurons by ectopic expression of proneural transcription factors. Here we show that directly reprogrammed neurons, despite their rapid maturation kinetics, can model teratogenic mechanisms that specifically affect early neurodevelopment. We delineated distinct phases of in vitro maturation during reprogramming of human neurons and assessed the cellular phenotypes of Valproic acid (VPA), a teratogenic drug. VPA exposure caused chronic impairments in dendritic morphology and functional properties of developing neurons, but not those of mature neurons. These pathogenic effects were associated with VPA-mediated inhibition of histone deacetylase (HDAC) and glycogen synthase kinase-3 (GSK-3) pathways, which caused transcriptional down-regulation of many genes including MARCKSL1, an actin-stabilizing protein essential for dendritic morphogenesis and synapse maturation during early neurodevelopment. Our findings identify a developmentally restricted pathogenic mechanism of VPA and establish the use of reprogrammed neurons as an effective platform for modeling teratogenic pathways.

Graphical Abstract

eTOC BLURB

Chanda, Südhof, and colleagues used reprogrammed human neurons to model the neurodevelopmental toxicity induced by prenatal exposure to Valproic acid, a potent teratogenic compound. By analyzing different phases of neuronal maturation in vitro, the authors identified Valproate-mediated cellular mechanisms that selectively impair the neuronal properties at early developmental stages.

INTRODUCTION

We have recently demonstrated that embryonic stem (ES) cells can be directly and efficiently converted into human neurons bypassing the intermediate progenitor state (Chanda et al., 2014; Pang et al., 2011; Yang et al., 2017; Zhang et al., 2013). These reprogrammed cells acquire the morphological and functional properties of mature neurons within few weeks of transgene expression. The rapid maturation of ES cell-derived neurons offers advantages for translational research as it facilitates the phenotypic characterization of relevant disease models (Pak et al., 2015; Patzke et al., 2015; Yi et al., 2016). However, it remains unknown if the accelerated maturation of induced neurons might conversely limit their application in studying early neurodevelopment and identifying teratogenic mechanisms that affect developmental processes. To address this potential concern, we sought to inquire whether reprogrammed neurons can be employed to model the pathogenic mechanisms of a potent teratogenic compound, Valproic acid (VPA).

VPA is one of the most effective and commonly prescribed therapies for epilepsy and bipolar disorder. In early 1980s, however, VPA was found to be teratogenic (Brown et al., 1980; Robert and Guibaud, 1982). Children born to mothers who were treated with VPA during pregnancy exhibit an increased incidence of neurological diseases, especially neural tube defects and fetal valproate syndrome, which includes severe cognitive defects and autism spectrum disorder (Christensen et al., 2013; Jentink et al., 2010; Koren et al., 2006; Moore et al., 2000; Rasalam et al., 2005; Williams et al., 2001). In contrast, VPA produces no overall cognitive impairment in adults (Gillham et al., 1990). Despite well-documented teratogenicity, VPA is still being used during pregnancy because it is a highly effective anti-epileptic drug (AED) for which, at least in some patients, has no alternative.

The mechanism-of-action for VPA’s teratogenic effects is not well-understood. In rodents, prenatal exposure to VPA causes an array of behavioral and neurodevelopmental changes of unknown origin (Manent et al., 2007; Schneider and Przewlocki, 2005; Wagner et al., 2006). Several conflicting hypotheses have been suggested to explain how VPA might impair brain development at the cellular level. As example, VPA was shown to increase neural differentiation, promote neurite formation in mature neurons (Cho et al., 2013; Hall et al., 2002; Hsieh et al., 2004; Vukicevic et al., 2015; Williams et al., 2002), and was demonstrated to both inhibit or enhance neural precursor cell (NPC) death (Fujiki et al., 2013; Go et al., 2011). However, the pathogenic mechanisms of VPA in newly-born neurons at early developmental stages, remain unclear.

In this current study, we have performed a systematic analysis of the cellular effects of VPA in human neurons directly reprogrammed from ES cells. This approach allowed us to deconstruct the VPA-associated neuronal phenotypes at different phases of in vitro development, and characterize the pathogenic mechanisms of VPA leading to its teratogenic effects. We demonstrate that chronic VPA-treatment differentially affects early vs. late stages of neurodevelopment. During early development, VPA induces profound changes in gene expression and causes severe impairments in dendritic morphology and functional parameters. These findings validate the use of reprogrammed neurons as a model system for understanding the mechanisms of neurodevelopmental disorders, which could be utilized for preventive therapies.

RESULTS

Directly reprogrammed human neurons mature gradually in vitro.

We induced neuronal fate in human H1-ES cells using a well-established protocol that efficiently trans-differentiates ES cells into cortical glutamatergic neurons by forced expression of a transcription factor, neurogenin-2 (Ngn2) (Zhang et al., 2013). We analyzed the maturation kinetics of reprogrammed cells by monitoring them at different time-points (Fig. 1A). We noticed that these cells underwent a rapid morphological transition, as evidenced by immediate neurite formation, and almost completely suppressed pluripotency marker Nanog, within only 4 days after Ngn2 induction (Fig. 1B, C).

Figure 1: Directly reprogrammed human neurons rapidly acquire neuronal identity, but undergo distinct developmental stages and gradual maturation in vitro.

A-B. Neurogenesis was induced in H1-ES cells by Ngn2 expression. Neurons were co-cultured with glia and analyzed (arrowheads) at different time-points (A). In some experiments, cells were additionally infected with a lentivirus expressing EGFP. Example images of day 4 human neurons derived from an ES-cell colony (B).

C. Sample images (left) of ES cells and day 4 neurons (EGFP expressing) stained with DAPI and antibody for Nanog. Summary graph (right) of average intensity for Nanog signal normalized by DAPI-positive nuclear area.

D. Representative images of ES cells and human neurons at post-induction day 4, day 7, day 14, day 21, day 30, and day 45 (vertical columns), as stained with DAPI and immunolabeled for indicated markers (horizontal rows). White arrowheads = human cells (ES cells or neurons), yellow arrowheads = glia, and cyan arrowheads = dividing ES cells.

E-M. Quantifications of signal-intensities for different markers expressed in ES cells and at different time-points of neuronal maturation in vitro. E, Sox2 normalized to nuclear area; F, Ki-67 normalized to nucleus count; G-I, Nestin (G), Dcx (H) and Tuj1 (I) normalized to total surface area; J, NeuN normalized to nuclear area; K, Map2 normalized to neurite area; L-M, Synapsin normalized to threshold-adjusted MAP2 area (L, summary graph; M, sample images).

N. Example traces (top) of APs produced by a 90 pA step-current injection (protocol is shown on top, V_hold = −60 mV) at day 4, day 7, day 14, day 21, day 30, and day 45; summary graphs (bottom) of the percentage of cells capable of firing APs (left) and the total number of APs (right).

O. Summary graphs of C_m (left), R_m (middle), or V_m (right) at different time-points of neuronal maturation.

P. Sample traces of evoked AMPAR-EPSCs at different time-points (left); summary graphs for the percentage of cells with AMPAR-EPSCs (middle), and peak-amplitude of AMPAR-EPSCs (right).

All summary data are presented as means ± SEM, with total number of cells and dendritic sections analyzed (for immunostaining) or cells patched (for electrophysiology) / number of independent experiments.

To further assess the onset of neurogenesis and maturation of reprogrammed cells, we immunolabeled them for specific markers affiliated with different developmental stages. Again, we found that already by day 4, cells lacked dividing nuclei (monitored by DAPI-stain), and lost expressions of pluripotency and proliferation markers (i.e. Sox2 and Ki-67, respectively; Fig. 1D–F). They transiently upregulated NPC marker Nestin (Fig. 1D, G), and rapidly acquired bona fide neuronal markers that begin to express at immature stages (e.g. Dcx and Tuj1; Fig. 1D, H–I). However, the cells only slowly acquired mature neuronal markers (i.e. NeuN, MAP2, or Synapsin; Fig. 1D, J–M). These results suggest that reprogrammed cells lose ES cell-identity and gain neuronal identity quickly after Ngn2 induction, display hallmarks of early neurodevelopmental stages in terms of relevant marker expression, and then undergo gradual maturation.

We next probed the development of functional properties in reprogrammed human neurons by characterizing their electrical and morphological parameters. We found that most cells started firing action potentials (APs) already at day 7, and the number of APs increased with time (Fig. 1N). This further correlated with a gradual growth in neurite complexity, a continuous increase in membrane capacitance (C_m), and a steady decrease in membrane-resistance (R_m) and resting membrane-potential (V_m) (Fig. 1O). To assess the functional maturation of synapses, we measured AMPAR-mediated excitatory postsynaptic currents (EPSCs) evoked by presynaptic stimulations. We observed synaptic responses in ~50% cells already at day 14 and in nearly 100% cells at day 21, with continuous increase in synaptic strength over time (Fig. 1P). Together, these data suggest that Ngn2-induced cells rapidly acquire neuronal properties, and then gradually develop into mature neurons in vitro.

VPA causes chronic impairments in developing human neurons.

We then asked whether these gradually maturing reprogrammed neurons can be used to identify the pathogenic effects of VPA, which exhibits severe teratogenicity during fetal neurodevelopment. At day 1 after transgene induction, we treated developing human neurons with VPA for 72 hr and analyzed them at different time-points thereafter (Fig. 2A). VPA did not alter neuronal induction efficiency, inhibit reprogramming factor expression, or affect cell survival (Fig. S1A–E). However, VPA triggered severe impairments in neurite development (Fig. 2B). VPA altered neurite morphology in a dose-dependent manner, with major effects observed at concentrations as low as ≈ 0.1 mM (Fig. 2C, D). VPA also caused prominent changes in neuronal cell shape. VPA-treatment dramatically increased the soma size and produced a large number of thin filopodia (Fig. 2E, F). VPA-induced pathogenic effects were similarly observed in human neurons trans-differentiated from H9-ES cells using Ngn2, and with murine neurons trans-differentiated from ES cells using either Ngn2 or Ascl1 (Fig. S1H–J), confirming that the phenotype was not dependent on specific cell line or reprogramming factor used for directed neurogenesis. Finally, recurring VPA exposures of 4 hr/day for 3 days or 3 hr/day for 6 days, or a continuous VPA exposure for only 24 hr was sufficient to significantly impair the early morphological maturation in human neurons (Fig. S2A–I).

Figure 2: VPA impairs dendritic morphogenesis and synaptic maturation of developing neurons.

A. Experimental time-course: the Ngn2-induced human neurons (co-expressing EGFP) were treated with VPA or equal volume of dH₂O (control) at day 1 (blue arrowhead) for the next 72 hr, and subsequently analyzed at different time-points (black arrowheads).

B. Representative images of VPA-treated day 4 neurons immunostained for Tuj1 (left); summary graph of total neurite length normalized by the count of DAPI-stained nucleus (right).

C-D. Sample images (C) of EGFP-expressing neurons that were exposed to VPA of indicated concentrations; summary graphs (D) of neurite length (top) and number of branches (bottom).

E-F. Sample images (E, gray arrowhead = conventional neurites, white arrowhead = filopodia-like extensions); summary graphs (F) of soma size (top left), percentage of cells with filopodia (top right, connected circles are values from individual batches), filopodia number (bottom left) and length (bottom right).

G-H. Kinetics of neurite outgrowth in cells that were examined every fifth day after VPA exposure (G, example images, arrowheads = cell-bodies; H, summary graph of neurite length). Single neuron labeling was achieved by sparse infection with a low titer lentivirus expressing EGFP.

I-J. Representative images (H) of control vs. VPA-treated cells, double-labeled for dendritic MAP2 and synaptic Synapsin proteins (boxes expanded on right); summary graphs (I) of dendritic (top, left to right: dendrite length, branch number, and primary dendrites) and synaptic (bottom, left to right: computed total number of synapses, synapse density and size) parameters.

K. Summary graphs of C_m (left), R_m (middle), or V_m (right), for control vs. VPA-exposed neurons.

L. Example traces (left) of APs produced by step-currents (protocol is shown on top, V_hold = −60 mV); summary plot (right) of AP numbers as a function of injected current-amplitude.

M. Superimposed sample traces (left) of AMPAR-mediated evoked EPSCs; summary graphs (right) of the EPSC amplitude, and CV of EPSCs as an indirect measure of presynaptic release probability.

All data are means ± SEM, with number of frames analyzed (imaging), or cells patched (electrophysiology) / number of batches. Statistical significance was weighed by two-way ANOVA (for L) or two-tailed, unpaired, Student’s t-test (all bar-graphs), with *** P < 0.005; ** P < 0.01; * P < 0.05; ns = not significant, P > 0.05.

We next asked whether human neurons can recover from impairments caused by a developmentally early VPA exposure. Following a 72 hr VPA treatment at day 1, we monitored neurite outgrowth in cells every 5 days from day 4 until day 29 (Fig. 2A). We found that the untreated cells elaborated complex neurites during this culture period, whereas the single VPA treatment at early developmental stage permanently retarded the kinetics of neurite arborization (Fig. 2G, H). Using immunocytochemistry of dendritic marker MAP2, further analysis at day 24 revealed that VPA significantly decreased the dendrite length and branching, but not the number of primary dendrites (Fig. 2I, J). VPA had no effect on synapse density or size, but reduced the total number of synapses owing to the decrease in dendritic arborization (Fig. 2I, J). VPA-exposure also lowered the C_m as measured by patch-clamp recordings, but had no effect on R_m, V_m, or AP firing (Fig. 2K, L). VPA-exposure also decreased synaptic strength as measured by AMPAR-EPSC amplitude, without changing presynaptic release probability as assessed by EPSC coefficient-of-variation (CV), consistent with the VPA-mediated reduction in total number of morphologically defined synapses (Fig. 2M). Together, these results suggest that a transient VPA-exposure at early developmental period can impose significant cellular impairments.

VPA treatment does not affect the properties of mature neurons.

Since Ngn2-induced cells slowly acquire morphological and functional maturation (Fig. 1), we therefore asked if VPA can generate pathogenic effects at later developmental stages. To this end, we exposed neurons to 1 mM VPA at three different time-points (post-induction day 1, day 21, or day 50–56), and analyzed them 72–96 hr after each treatment had started (Fig. 3A).

Figure 3: VPA-mediated pathogenic effects are restricted to early developmental stages, and do not affect mature neurons.

A. Experimental strategy for examining the cellular effects of VPA at different developmental phases. Neurons were treated with 1 mM VPA (blue arrowheads) at day 1, day 21 or day 50–56, and analyzed after 72–96 hr (black arrowheads) for each time-point.

B-D. Threshold-adjusted MAP2 immunostainings (top) of control vs. VPA-treated cells, and summary graphs (bottom, left to right) of dendrite length, number of branches, and primary dendrites, for day 1 (B, arrowheads = cell-bodies), day 21 (C), or day 50–56 (D) protocols.

E. Average C_m, (left), R_m (middle), or V_m (right) for neurons that were pre-treated with VPA at day 21.

F. Representative traces (left) of current-induced APs, and summary plot (right) of AP numbers as a function of injected current, for neurons pre-treated with VPA at day 21.

G-H. Sample images (G) of control and VPA-exposed neurons double-labeled for MAP2 and Synapsin (boxed areas expanded on right); summary graphs (H) of synaptic parameters (left to right: computed total number of synapses per cell, synapse density and size).

I. Superimposed traces (left) of evoked AMPAR-EPSCs, and summary graphs of peak EPSC amplitude and CV (right), for neurons pre-treated with VPA at day 21.

J-N. Same as E-I, except for neurons that were pre-treated with VPA at day 50–56.

All numerical data are presented as means ± SEM, with total number of field-of-views analyzed (for imaging) or cells patched (for electrophysiology) / number of batches. Statistical significance was evaluated by two-way ANOVA (for F and K), or two-tailed, unpaired, Student’s t-test (bar-graphs), with *** P < 0.005; * P < 0.05; ns = not significant, P > 0.05).

Using MAP2 immunostaining, we found that chronic VPA exposure at day 1 strongly reduced neurite outgrowth (Fig. 3B, consistent with Fig. 2). Day 21 VPA treatment also caused significant defects in dendritic arborization, but these effects were less severe than those observed for day 1 neurons (Fig. 3C). Day 50–56 VPA-treatment, however, exhibited no obvious changes in dendrite morphology (Fig. 3D). None of the VPA treatments affected cell survival (Fig. S1F, G). In sum, these data show that VPA impairs dendritic outgrowth only at early developmental stages and does not impair the morphology of mature neurons.

To further explore the effects of VPA on the functional properties of day 21 neurons, we next performed patch-clamp recordings. At day 21, 72–96 hr VPA exposure decreased the C_m without changing R_m or V_m, consistent with the decreased dendritic arborization (Fig. 3E). This treatment also lowered AP numbers, and reduced the total number of synapses without affecting synapse density (Fig. 3F–H). Furthermore, VPA treatment of day 21 neurons decreased the amplitude of AMPAR-EPSCs without changing the EPSC CV, as would be expected from a reduction in total synapse number owing to the impaired dendritic branching (Fig. 3I). We also detected a small increase in synapse size in VPA-exposed cells, the mechanism of which remained unclear (Fig. 3H). The pathogenic effects of VPA in day 21 neurons, in contrast to day 1 neurons, did not emerge after 24 hr and required at least 48 hr exposure, suggesting that only long-term VPA exposure can impair neuronal properties at this later developmental stage (Fig. S2J–O). Importantly, VPA treatment of day 50–56 neurons had no effect on their intrinsic properties or current-induced APs (Fig. 3J, K), and displayed a modest increase in synapse density and enhanced synapse size (Fig. 3L, M). However, electrophysiological recordings revealed that VPA did not augment the synaptic strength at day 50–56, suggesting that VPA-induced synapses in developmentally mature neurons are functionally ‘silent’ (Fig. 3N). In sum, these data indicate that VPA-mediated morphological and functional impairments are limited to early development and do not affect mature neuronal population.

VPA impairs neurodevelopment by inhibiting HDAC and GSK-3 pathways.

We explored the mechanism of VPA-phenotypes. Prior studies implied that VPA depletes inositol levels and attenuates inositol-tris-phosphate (InsP₃) signaling (O’Donnell et al., 2000; Vaden et al., 2001), inhibits glycogen-synthase kinase 3β (GSK-3β; Leng et al., 2008) and protein kinase C (PKC; Manji et al., 1999), activates MAP kinases (MAPK; Yuan et al., 2001), and inhibits class I histone deacetylases (HDACs; Gottlicher et al., 2001). To evaluate if these pathways might be responsible for VPA-phenotypes, we treated day 1 neurons with pathway-specific pharmacological agents and tested whether these drugs replicate or reverse the VPA-induced morphological defects (Fig. S3A). We found that VPA-mediated inhibition of early neuritogenesis could be mimicked by administration of specific inhibitors for GSK-3β (CHIR99021) or class I HDACs (TSA, CI994), but not by inhibitors for class III HDACs (nicotinamide), PKC (GO6983), or Valpromide (VPD, structural analog of VPA). Moreover, activators of InsP₃ pathway (myo-inositol, KYP2047) and blockers of MAPK (PD98059, FR180204) failed to prevent VPA-induced defects in neurite outgrowth (Fig. 4A, B). Consistent with inhibition of class I HDAC and/or GSK-3β pathway as a potential mechanism of VPA action, VPA-treatment elevated both histone acetylation and β-catenin levels in human neurons (Fig. 4C–E). These results indicate that inhibition of class I HDAC and/or of GSK-3β could be responsible for the VPA-mediated impairments at early stages of neurite development.

Figure 4: VPA impairs the morphological and functional maturation of developing neurons by inhibiting HDAC and GSK-3 pathways.

A-B. Neurons were treated with indicated drugs at day 1, and analyzed after 72–96 hr. [Drug concentrations: VPD (1 mM), GO6983 (5 μm), CHIR99021 (10 μm), TSA (20 nM), CI994 (50 nM), Nicotinamide (1 mM), VPA (1 mM), Inositol (1 mM), KYP2047 (100 μm), PD98059 (5 μm), FR180204 (5 μm)]. A, sample images; B, average neurite length for indicated drug-combinations.

C-D. Sample images (C) of control vs. VPA-exposed neurons immunostained for indicated markers; summary graphs (D) of size-normalized histone-acetylation (left) or DAPI signal (middle), and ratio between acetylated histone H3 / DAPI signal intensities (right).

E. Representative images (left) of control vs. VPA-treated neurons, co-expressing EGFP and immunostained for β-catenin; summary graph (right) of β-catenin signal intensity normalized to EGFP area.

F. Neurons were treated with the indicated drugs at day 21 for 72–96 hr. Immunostainings for MAP2 (left), and summary graphs of dendritic length (right).

G-H. Sample trace (left) and summary plot (right) of AP number as a function of injected current-amplitude, for cells treated with CHIR99021 (G) or TSA (H) at day 21, and analyzed after 72–96 hr.

I. Voltage-clamp traces for control vs. TSA (left, experimental protocol is shown on top with V_hold = −70 mV; insets display expanded Na⁺-current traces; green lines are time-periods for calculating K⁺-currents); summary plots (right) of K⁺- (I_K) and Na⁺ (I_Na) -current amplitudes as a function of voltage-pulses.

J. Same as I, except for 72–96 hr VPA treatment at day 21.

K-L. Experiment protocol. K, lentiviruses expressing U6-promoter driven class I HDAC (HDAC1/2/3/8) shRNAs were used with selection cassette (PGK-promoter driven puromycin-resistance (Puro)), and an empty pLKO.1 vector was used as control (knockdown efficiencies reported in Fig. S3). L, experimental time-lines. For day -1 knockdowns, cells were co-infected with a lentivirus expressing EGFP.

M-N. Representative images (M) and average neurite length (N) for day -1 knockdown of HDAC1/2/3/8.

O. Sample traces (left) and summary plots (right) of APs generated by current injections in cells expressing knockdown constructs of HDAC1/2/3/8 from day 20 experiments.

P. Sample traces of voltage-clamp experiments (left) and average amplitudes of I_K and I_Na (right) in cells from control vs. HDAC3 knockdown (shRNAs #1 and #2) conditions for day 20 knockdown experiments.

The values on graphs represent means ± SEM for number of field-of-views analyzed (imaging) or cells patched (electrophysiology) / number of independent batches. Statistical significance was weighed by two-way ANOVA (G-J and O-P) or two-tailed, unpaired, Student’s t-test (all other data), with *** P < 0.005; ** P < 0.01; * P < 0.05; ns = not significant, P > 0.05.

To determine if HDAC and/or GSK-3β pathways contribute to VPA-phenotypes observed in day 21 neurons, we treated them with TSA or CHIR99021 and analyzed after 72–96 hr. We used VPD as a negative control. We found that chronic exposure to both TSA and CHIR99021 substantially reduced the dendritic length (Fig. 4F). Both drug-treatments, similar to VPA, also decreased C_m without affecting R_m (Fig. S3B, C). Thus, inhibitors of HDACs and GSK-3 phenocopied the morphological impairments induced by VPA also in day 21 cells. TSA and CHIR99021, however, had significant but opposite effects on V_m that remained unaffected by VPA (Fig. S3B, C, compared to 3E), Therefore, VPA-effects likely reflect a convergent product of multiple signaling pathways, where no single mechanism can fully reproduce all VPA-phenotypes.

Neither TSA nor CHIR99021 impaired cell survival, but interestingly, only TSA and not CHIR99021 effectively reduced AP firing and altered AP properties at day 21, suggesting that VPA may inhibit cellular excitability by HDAC inhibition (Fig. 4G, H, and Fig. S3D–G). In support to this hypothesis, we found that TSA decreased the amplitude of Na⁺-channel inward currents without changing K⁺-channel outward currents, a mechanism for the TSA-induced reduction in AP firing that was also replicated by VPA (Fig. 4I, J). Furthermore, VPA also strongly elevated histone acetylation levels of day 21 neurons and impaired the AP properties similar to TSA (Fig. S3H–J). Exposure to VPD (VPA-analog, negative control) did not affect either intrinsic or AP properties (Fig. S3K–M). These data suggest that VPA impairs morphological maturation of developing neurons by inhibiting HDAC and/or GSK-3β, but decreases cellular excitability only by HDAC-inhibition-mediated reduction of Na⁺-currents.

To gain further insight into the relative roles of class I HDACs in VPA-phenotypes, we used shRNA-mediated knockdowns of all four members (HDAC1/2/3/8) of class I HDAC family (Fig. 4K, L, and Fig. S3N–P). We found that downregulation of both HDAC1 and HDAC2 impaired neurite morphology (Fig. 4M, N), but downregulation of only HDAC3 phenocopied the VPA-mediated inhibition of AP firing and decreased Na⁺-channel peak-current (Fig. 4O–P). Knockdown of HDAC8 did not alter neuronal properties. These results highlight the convergence of mutually exclusive functional contributions from different HDACs that together produce VPA-phenotypes.

To test whether our conclusions apply to neurons generated by a different method, we obtained neural stem cells (NSCs) from human ES cells using dual SMAD inhibition (Chambers et al., 2009). We differentiated the NSCs into neurons and treated them with VPA (Fig. S4A–C). Similar to directly reprogrammed neurons, VPA also inhibited neurite outgrowth in NSC-derived neurons, elevated histone acetylation and β-catenin level (Fig. S4D–F). In addition, inhibitors of class I HDAC and GSK-3 successfully replicated VPA-mediated morphological defects in these cells (Fig. S4G). VPA also impaired the AP properties and reduced Na⁺-channel peak-currents in NSC-derived neurons, similar to Ngn2-induced neurons (Fig. S4H–M, compare with Figs. 3F, 4J, and S3J). Therefore, VPA had comparable effects in developmentally early human neurons produced by very different methods. We also noticed a small but significant decrease in K⁺-channel peak-current in VPA-treated NSC-derived neurons that was not observed for Ngn2-induced neurons, indicating additional effects of VPA in these cells (Fig. S4M). Finally, to further test the generality of our findings, we injected VPA into pregnant mice and analyzed the VPA-exposed fetal brains (Fig. S4N, Video V1). We found that prenatal VPA exposure strongly enhanced histone acetylation and decreased neurite arborization in the cortex of developing embryos (Fig. S4O–P). Thus, VPA-associated pathogenic mechanisms observed in directly reprogrammed human neurons were highly consistent across different in vitro and in vivo models of neurodevelopment.

VPA dysregulates the expression of many genes, most prominently MARCKSL1.

Developmentally early VPA exposure seems to impair neurite morphogenesis by inhibiting class I HDAC and/or GSK-3β pathways (see Fig. 4). Because both HDAC and GSK-3 are involved in gene regulation (Graff and Tsai, 2013; Patel and Woodgett, 2017; Peixoto and Abel, 2013), VPA-induced pathogenic effects may originate from transcriptional changes. To examine this hypothesis, we performed RNA-sequencing analyses of neurons treated with VPA at day 1, day 21 or day 50–56, for 72–96 hr.

We found that VPA profoundly altered the gene transcription in human neurons by dysregulating progressively higher number of genes at early maturation stages (Fig. 5A, Table T1). To identify key proteins involved with VPA-phenotypes, we focused on day 21 treatment, because at this time-point VPA had composite effects on both neuronal morphology and function (see Fig. 3). Further analysis of day 21 RNA-sequencing data revealed that a majority of VPA-affected genes was highly enriched in brain, and many were linked to mental disorders (Fig. S5A, B, Table T2). Gene Ontology (GO) analyses suggested that VPA affected sets of genes linked with oxidation-reduction, transcriptional regulation, stress response, and development, without affecting neuronal maturation (Fig. S5C–I, Table T2). Among the list of genes that were directly associated with neuronal parts, highly expressed during development, and most significantly or strongly down-regulated by VPA, one gene that immediately caught our attention was MARCKSL1 (Fig. S5J–L, Table T2).

Figure 5: VPA exposure alters gene expression and suppresses MARCKSL1 in developing neurons.

A. Day 1, day 21, and day 50–56 neurons were treated with 1 mM VPA or equal volumes of H₂O (Ctrl) for 72–96 hr, isolated by fluorescence-activated cell sorting, and analyzed by RNA-seq. Heat-maps represent average FPKM values of significantly affected genes.

B. Scatter plot of P-values (−log₁₀ scale) for genes that were significantly down-regulated by VPA at day 1 (Y-axis) or day 21 (X-axis). Dotted lines (black) are P = 0.05. The genes commonly down-regulated at both time-points are highlighted (red), and some highly significant genes are labeled.

C. Summary graph of MARCKSL1 FPKM values (connected circles are independent batches) from RNA-seq with day 1 (left), day 21 ((middle), and day 50–56 (right) VPA-treatments.

D. Same as C, except the MARCKSL1 mRNA levels were estimated by qRT-PCR, and normalized to those of GAPDH (internal control) and control-treatments (dH₂O).

E. Same as D, except from neurons treated with indicated drugs [VPD (1 mM), TSA (20 nM), CI994 (50 nM), CHIR99021 (10 μm), SB216763 (5 μm), SB415286 (20 μm), GO6983 (10 μm), SP600125 (10 μm), Nicotinamide (1 mM)] at day 1 (left) or day 21 (right), and analyzed after 72–96 hr.

F. qRT-PCR for MARCKSL1 mRNA in untreated neurons from same experimental batches at indicated stages of in vitro development.

G. Sample images of an untreated day 4 neuron visualized by EGFP (top), immunolabeled for MAP2 (second top), MARCKSL1 (second bottom), and stained for DAPI (bottom). Arrowheads = MARCKSL1 protein clusters in the growth cone.

H. Untreated day 21 human neurons stained for EGFP (top left), MAP2 (top right), and MARCKSL1 (bottom left); boxed areas are enlarged as a merged view (bottom right).

I. Representative images of dendritic sections from untreated day 21 neurons, co-immunostained for Synapsin and MARCKSL1 (left); summary graph for the co-localization of MARCKSL1 and Synapsin-positive clusters, as measured via Mander’s correlation coefficient (right).

J. Images (left) of neurite terminals with growth-cones from neurons that were treated with VPA (1 mM) at day 1 and analyzed after 72–96 hr; summary graphs (right) of MARCKSL1 cluster density and size.

All data are presented as means ± SEM, with the number of experiments (batch-wise comparison, A-F), or the field-of-views analyzed / number of batches (I-J). Statistical significance was assessed by two-tailed, unpaired (paired for A-F), Student’s t-test, (*** P < 0.005; ** P < 0.01; * P < 0.05; ns = not significant, P > 0.05).

MARCKSL1 was the only neuronal as well as cytoskeletal gene expressed with FPKM >200 in control day 21 neurons, and at least two-fold dysregulated by VPA (Fig. S5M–P). MARCKSL1 is a member of MARCKS family that is predominantly expressed in developing brain and has been linked to actin-dependent cell motility and migration (Bjorkblom et al., 2012; Li and Aderem, 1992). Genetic knockouts of MARCKSL1 in mice impair brain development and cause spina bifida (Chen et al., 1996; Wu et al., 1996), diseases that are commonly linked to prenatal VPA-exposure in human patients. In reprogrammed neurons, MARCKSL1 was the highest expressed gene consistently down-regulated by VPA at both day 1 and day 21, but not at day 50–56, making it a potential candidate for major VPA-phenotypes (Fig. 5B, C). Therefore, we next asked if VPA-induced down-regulation of MARCKSL1 may mediate the cellular impairments caused by developmentally early VPA exposure.

Using quantitative RT-PCR (qRT-PCR), we confirmed that VPA treatments substantially reduced MARCKSL1 expression of day 1 and day 21, but not day 50–56 human neurons (Fig. S6A, B, and Fig. 5D). We tested if the pharmacological agents that target VPA-affected pathways can also suppress MARCKSL1 mRNA level. In day 1 and day 21 neurons, administration of inhibitors for class I HDACs (i.e. TSA and CI994) or GSK-3β pathways (i.e. CHIR99021, SB216763, and SB415286) both down-regulated MARCKSL1 expression (Fig. 5E). Of note, MARCKSL1 expression was also affected by inhibition of PKC pathway (by GO6983), but not JNK pathway (by SP600125), which are known to regulate MARCKSL1 phosphorylation (Bjorkblom et al., 2012), whereas class III HDAC inhibitor (Nicotinamide) and VPA-analog VPD consistently failed to decrease MARCKSL1 expression.

During neuron development, MARCKSL1 mRNA levels increased by 4-fold (Fig. 5F). MARCKSL1 protein was abundantly localized in dendritic clusters that were mostly non-synaptic, and was strongly depleted in VPA-treated developmentally early neurons (Fig. S6C, and Fig. 5G–J). VPA also reduced MARCKSL1 expression in reprogrammed mouse and human neurons independent of induction protocols, and in cortical neurons in vivo for VPA-exposed mouse embryos, but had no effect on MARCKSL1 levels in glia (Fig. S6E–M). These data suggest that MARCKSL1 is a bona fide downstream target of chronic VPA exposure in developing neurons.

MARCKSL1 knockdown partially phenocopies VPA exposure.

To examine the effects of MARCKSL1 down-regulation at early developmental periods (Fig. 6A, B), we next used multiple shRNAs that target endogenous MARCKSL1 (Fig. S6D, N, O). We found that Day-1 knockdown of MARCKSL1 caused significant impairments in neurite outgrowth, although less severe than VPA (Fig. 6C). MARCKSL1 knockdown at day 20 also strongly decreased dendritic arborization and reduced the total number of synapses, similar to those observed for VPA (Fig. 6D, E). In addition, day 20 MARCKSL1 knockdown decreased the C_m without affecting R_m or V_m, and thus phenocopied VPA exposure (Fig. 6F). Interestingly, MARCKSL1 knockdown did not alter neuronal excitability as monitored by AP generation, indicating that MARCKSL1 down-regulation does not contribute to inhibition of AP firing by VPA (Fig. 6G). MARCKSL1 knockdown, however, significantly decreased the synaptic strength of AMPAR-EPSC without changing EPSC CV, a phenotype fully consistent with decreased number of synapses caused by impaired dendritic arborization (Fig. 6H). Viewed together, these data suggest that partial loss-of-MARCKSL1 replicates, to a large extent, the morphological impairments caused by developmentally early VPA exposure.

Figure 6: MARCKSL1 knockdown partially reproduces VPA-induced developmental defects.

A-B. Experiment strategy. A, lentiviral vectors expressing U6-promoter driven MARCKSL1-shRNAs were used with PGK-promoter driven puromycin-resistance cassette (Puro), an empty pLKO.1 vector was used as control (knockdown efficiencies in Fig. S6). B, time-line of knockdown experiments. For day -1, cells were additionally infected with a lentivirus expressing EGFP.

C. Sample images (left) and summary graphs (right) of neurite length and branch number, for MARCKSL1 knockdown initiated at day -1.

D-E. MARCKSL1 knockdown at day 20: sample images (D) of control or MARCKSL1 knockdown cells stained for MAP2 and Synapsin (boxes magnified on right); quantification (E) of dendritic and synaptic parameters (top, left to right: total dendrite length, number of branches, and primary dendrites; bottom, left to right: total synapse number, density and size).

F. Summary graphs of C_m (left), R_m (middle), and V_m (right), for MARCKSL1 knockdown initiated at day 20.

G. Representative traces (left) of APs induced by step-current injections; summary plot (right) of AP numbers as a function of current amplitude.

H. Overlaid traces (left) of evoked AMPAR-EPSCs, and summary graphs (right) of EPSC amplitude and CV.

All quantifications are means ± SEM. Summary data also include the number of cells patched or field-of-views analyzed / batches. Statistical significance was assessed by two-tailed, unpaired, Student’s t-test (bar-graphs), or two-way ANOVA (summary plot, G), with *** P < 0.005; ** P < 0.01; * P < 0.05; ns = not significant, P > 0.05.

MARCKSL1 overexpression partially reverses VPA-induced impairments.

To further test the hypothesis that down-regulation of MARCKSL1 contributes to VPA-mediated pathogenic effects, we next asked whether overexpression of MARCKSL1 could reverse, at least partially, the VPA-phenotypes that presumably reflect its teratogenic activity. To address this question, we used lentiviruses that overexpress MARCKSL1 together with EGFP (to identify MARCKSL1-expressing cells), and analyzed human neurons that were pre-exposed to VPA at day 1 or day 21 time-point (Fig. S6P, Q, and Fig. 7A, B).

Figure 7: MARCKSL1 overexpression partially rescues the VPA-induced cellular impairments.

A-B. Experiment plan. A, Lentiviral construct to express full-length human MARCKSL1 trailed by IRES-EGFP under Tet-on promoter, and EGFP-only vector as infection control. B, Experimental time-line. Doxycycline was used for continuous transgene expression, throughout the experimental period.

C. Sample images (left) of EGFP-expressing developing neurons from indicated conditions; summary graphs (right) of total neurite length and branch numbers.

D-F. Sample images (D) of control and MARCKSL1-overexpressing neurons ± VPA treatment, immunostained for MAP2 and Synapsin (boxes expanded on right). Summary graphs of dendritic and synaptic parameters for the indicated conditions (E, left to right: dendrite length, branch numbers, and primary dendrites; F, left to right: computed total number of synapses, synapse density and size).

G. Summary graphs of mean C_m (left), R_m (middle), and V_m (right), for control and MARCKSL1-overexpressing neurons ± VPA treatment.

H. Representative AP traces (left) induced by step-current injections, and summary plot (right) of AP numbers as a function of current injections, for control vs. MARCKSL1-overexpressing neurons ± VPA.

I. Superimposed traces (left) of evoked AMPAR-EPSCs; summary graphs (right) of the EPSC amplitude and CV, as a function of MARCKSL1 overexpression and VPA exposure.

J. Sample EPSCs (left) induced by AMPA puffs; summary graphs (right) of total charge and charge-density (normalized to cell size) of AMPA-induced EPSCs, plotted as a function of MARCKSL1 overexpression ± VPA.

All quantifications are means ± SEM with indicated number of cells patched or field-of-views analyzed / number of experiments. Statistical significance was assessed by two-tailed, unpaired, Student’s t-test (bar-graphs), or two-way ANOVA (panel H), with *** P < 0.005; ** P < 0.01; * P < 0.05.

In developmentally early control neurons, MARCKSL1 overexpression caused a small but significant increase in neurite outgrowth and complexity (Fig. 7C). However, in VPA-exposed cells that exhibited a nearly complete suppression of neurite outgrowth, MARCKSL1 overexpression enabled a partial, but highly significant rescue of that phenotype (Fig. 7C). MARCKSL1 overexpression enabled an identical rescue of neurite arborization in mouse ES-cell derived neurons treated with VPA (Fig. S6R, S). These findings suggest that MARCKSL1 is a major contributing factor for VPA-induced morphological defects.

MARCKSL1 rescued the VPA-phenotypes even more profoundly at a later developmental stage, i.e. when we overexpressed it in day 19 cells and VPA-treated them at day 21. MARCKSL1 expression had no measurable effect in control cells, but completely reversed the VPA-mediated defects in dendritic morphology and synapse numbers (Fig. 7D–F). MARCKSL1 overexpression did not reverse the VPA-induced increase in synapse size (Fig. 7F). However, MARCKSL1 expression fully rescued VPA-mediated reduction in C_m, without affecting R_m, V_m, or depolarization-induced AP firing (Fig. 7G, H). Owing to VPA-induced reductions in dendritic arborization and synapse numbers, both the surface levels of synaptic receptors (measured by AMPAR responses to puff-applied AMPA) and synaptic strength (measured by AMPAR EPSC amplitude) were reduced. All of these VPA-induced pathogenic phenotypes were completely reversed by MARCKSL1 overexpression (Fig. 7I–J). Viewed together, these results suggest that MARCKSL1 overexpression largely prevents VPA-induced morphological and synaptic impairments. These results are consistent with the notion that VPA’s pathogenicity in developing neurons are caused, in part, by cytoskeletal dysregulation due to VPA-mediated transcriptional suppression of MARCKSL1.

DISCUSSION

VPA is one of the most effective AEDs and being used even by pregnant women despite its well-documented teratogenicity. The cellular mechanisms underlying the teratogenic actions of VPA remain unclear to this day. Here, we used directly reprogrammed human neurons at different phases of in vitro development to analyze the cellular basis for VPA-teratogenicity. Our data suggest that VPA exposure potently impairs the morphology and function of neurons, and these deleterious effects of VPA are restricted to early developmental stages and do not occur in mature neurons. These findings establish the use of reprogrammed neurons to assess various teratogenic pathways that may differentially impair neurodevelopment.

Based on our findings, we propose that reprogrammed human neurons recapitulate some of the key features of neurodevelopmental processes, e.g. defined expression of relevant neuronal markers, gradual increase in morphological complexity, and acquisition of functional properties (Fig. 1). In this in vitro model of neuronal maturation, chronic VPA exposure at early developmental stages severely retarded the dendritic arborization of day 1 neurons in a dose-dependent manner (Fig. 2). This VPA-induced pathogenic effect caused a permanent delay in the morphological maturation of neurons that persisted long after drug removal (Fig. 2). Interestingly, VPA also impaired the dendritic development and/or maintenance of day 21 neurons that represented a later developmental stage, although to a lesser extent (Fig. 3). VPA had no effect on synapse density but decreased the total number of synapses because of reduced dendritic arborization (Figs. 2, 3). Importantly, VPA did not alter the morphology of more mature day 50–56 neurons (Fig. 3). These results indicate that VPA-mediated dendritic dysgenesis is restricted to early/mid developmental stages, and does not affect mature neurons.

Interestingly, previous studies with adult mice and primary neuronal cultures indicated that VPA exposure may accelerate neurite outgrowth and enhance synaptogenesis by astrocyte-dependent mechanisms (Cho et al., 2013; Long et al., 2015). Because gliogenesis follows neurogenesis during development and partially coincides with synaptogenesis (a trait of mature neurons), a glia-dependent phenotype may allude to VPA-mechanisms related to late developmental stages. In fact, VPA treatment of relatively mature day 60 human neurons in our experiments also caused a robust increase in synapse size and number (Fig. 3). However, the cellular effect of VPA in newly-born and developmentally early neurons was found to be different. For example, VPA treatment significantly reduced collateral neurite branching in rat dorsal root ganglia, and VPA-injection in pregnant mice decreased Tuj1 levels in the E16 fetal cortex (Go et al., 2012; Williams et al., 2002), similar to what we found in our in vitro and in vivo experiments.

Previous studies indicated that VPA affects multiple cell-signaling pathways (summarized in Fig. S3A). Using pathway-specific agonists or antagonists, we found that VPA-induced impairments in dendritic arborization can be reproduced by inhibitors of both class I HDACs and of GSK-3β (Fig. 4). However, neither HDAC nor GSK-3 inhibitors completely phenocopied all pathogenic effects of developmentally early VPA exposure. VPA had no effect on membrane potential, whereas inhibitors for HDAC and GSK-3 pathways, respectively, increased and decreased the V_m (Fig. S3). Inhibition of both pathways, although by a different extent, reduced the expression of MARCKSL1, a gene that was commonly down-regulated by VPA (Fig. 5). Thus, inhibition of both HDAC and GSK-3 contributes to VPA-pathogenicity, but neither pathway exclusively accounts for all phenotypes.

VPA treatment of day 1 neurons did not limit them from acquiring AP-firing properties at later maturation stages (Fig. 2). However, long-term VPA exposure of day 21 neurons severely impaired their existing capacity of AP firing by reducing the peak amplitude of voltage-gated Na⁺-current (Figs. 3, S2). This particular VPA-phenotype was mediated via HDAC but not GSK-3 inhibition, highlighting the differences in functional contributions by these two VPA-affected signaling pathways (Fig. 4). Interestingly, VPA treatment of already-mature day 50–56 neurons did not alter AP properties, suggesting that this phenotype is restricted to a neuronal population undergoing functional maturation (Fig. 3). Knockdown of class I HDAC family members also led to interesting revelations about VPA-mechanism, as downregulations of HDAC1 and HDAC2 impaired neurite morphology, whereas downregulation of only HDAC3 inhibited AP firing and decreased Na⁺-current (Fig. 4). Thus, VPA-mediated perturbations of distinct pathways can operate independently to generate different cellular effects.

Although this present study found strong evidence for the contributions of select pathways (HDAC and GSK-3) in VPA-phenotypes, other cellular mechanisms can also add to its teratogenic effects that was not captured or investigated in our experimental model. For example, VPA did not alter the number of neurons generated by Ngn2-induction (Fig. S1). However, previous reports indicate that repeated VPA treatment can enhance neural differentiation from NPCs (Hsieh et al., 2004; Vukicevic et al., 2015). More recently, two seminal studies have also used VPA as a chemical inducer to directly convert human fibroblast into neurons (Hu et al., 2015; Li et al., 2015). In these studies, VPA was included in the neuronal induction media but avoided in the maturation media, and therefore, effects of VPA on neuronal maturation were not directly analyzed. In our current study, VPA was applied to the cells after neurogenesis was already induced by Ngn2 with drug-selection. Thus, our study mainly examines VPA’s cellular effects at a post-neurogenesis and early neurodevelopmental stage.

We found that VPA dramatically altered the gene expression profile, with hundreds of genes affected including many linked with neurological disorders (Fig. S5). The total number of VPA-affected genes was substantially reduced with neuronal maturation, suggesting that early developmental stages are particularly susceptible to VPA (Fig. 5A, Table T1). GO-analysis of day 21 neurons revealed that VPA, despite being an HDAC inhibitor, down-regulated the expression of many genes (Fig. S5). As a potential reason for this effect we noticed that VPA exposure, at least at the mRNA level, upregulated transcriptional repressors TXNIP, ID3 etc. and also chromatin condenser H1F0, which may trigger transcriptional repression (Fig. S5G). VPA-treatment of human neurons also dysregulated many genes that were previously reported by other teratogenic VPA-models, although analyzed at different time-points using different experimental protocols with different VPA doses (Table T1, compare with Krug et al., 2013; Robinson et al., 2016; Zhang et al., 2017; Zhang et al., 2018).

Of all VPA-affected neuronal and cytoskeletal genes combined, the largest effect was seen with MARCKSL1, an abundant protein that was consistently down-regulated by VPA, but only during early development (Fig. 5). VPA-induced decrease in MARCKSL1 expression was reproduced in neurons generated by different protocols, and thus designated a canonical effect of VPA in immature neurons (Fig. S6). MARCKSL1 knockdown partially reproduced the VPA-induced dendritic dysgenesis and synapse loss (Fig. 6), and MARCKSL1 overexpression substantially rescued those phenotypes (Fig. 7). Although at day 1, MARCKSL1 overexpression did not fully rescue the VPA-induced neurite impairment, it increased neurite branching by 6–8 fold in VPA-treated neurons, compared to that in control neurons (1.3 fold) (Figs. 7C, and S6R, S). MARCKSL1 is a member of MARCKS family that plays major roles in dendrite and synapse formation, regulates the dynamics of F-actin bundling (Bjorkblom et al., 2012; Calabrese and Halpain, 2005; Li et al., 2008), and its knockout causes diseases that are related to prenatal VPA-exposure, e.g. abnormal brain development and spina bifida (Chen et al., 1996; Wu et al., 1996). Therefore, down-regulation of MARCKSL1 by VPA may provide a major mechanism for VPA-associated teratogenic effects in fetal neurodevelopment.

In summary, we show that VPA differentially affects the cellular properties of developmentally early vs. mature neurons (Fig. S7A). Chronic VPA exposure at early development causes massive defects in dendritogenesis, synapse formation, and cellular excitability, which are not observed in more mature neurons, and are caused at least in part, by down-regulation of MARCKSL1 expression and reduction in Na⁺-channel current (Fig. S7A, B). We propose that these developmentally early pathogenic effects may account for VPA’s teratogenic actions in fetal brain, and suggest that the analysis platform utilized here, i.e. the use of reprogrammed human neurons, offers a great opportunity for future studies to evaluate the potential teratogenic activity of various therapeutic agents.

STAR*METHODS (Details)

• KEY RESOURCES TABLE

Provided as a separate attachment.

• CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Soham Chanda (soham.chanda@colostate.edu).

• EXPERIMENTAL MODEL AND SUBJECT DETAILS

ES cell-derived human neurons

Human neurons were obtained from H1 (male) or H9 (female) ES cell lines by ectopic expression of proneural transcription factors, e.g. Ngn2 and Ascl1, as described previously (Chanda et al., 2014; Zhang et al., 2013).

Mouse ES cell-derived neurons

Mouse neurons were directly converted from mouse ES cells with stably-integrated and Doxycycline-inducible Ngn2 or ASCL1 transgene, as described previously (Chanda et al., 2014; Wapinski et al., 2013).

Human ES cell-derived NSCs

Differentiation of human H1-ES cells into NSC monolayer was achieved by dual SMAD inhibition, as described previously (Chambers et al., 2009).

Mouse embryos

CD1 mice (Charles River) were used in this study. Embryos (both male and female) were collected from PBS-only or VPA-injected pregnant animals.

• METHOD DETAILS

General experimental design

Experiments were performed in accordance with institutional guidelines and regulations. All experiments, except for the initial findings of VPA-induced cellular impairments (Fig. 2B–E, Fig. 3B–D), and standardizing the protocol for chemical induction of NSC monolayer from ES cells followed by their differentiation into neurons (Fig. S4A, B), were performed in a blinded fashion; i.e., the experimenters were unaware of the sample types being studied.

Viral constructs

The lentiviral constructs used for the neuronal reprogramming of human/mouse ES cells included Ngn2-t2APuromycin^Resistance (driven by Tet-on promoter) and rtTA (driven by ubiquitin promoter), with an optional virus expressing EGFP (driven by Tet-on promoter) for morphological analyses (Zhang et al., 2013). For knockdown experiments of human MARCKSL1 and class I HDACs, shRNA constructs (Mission shRNA, Sigma) targeting MARCKSL1 (TRC# 0000157260, 0000156469, 0000152478), HDAC1 (TRC# 0000195467, 0000195103), HDAC2 (TRC# 0000004819, 0000004823), HDAC3 (TRC# 0000194993, 0000004825), and HDAC8 (TRC# 0000350469, 0000314872) were used with rtTA, Ngn2-t2a-Blasticidin^Resistance (driven by Tet-on promoter), and EGFP. An empty pLKO.1 vector was used for infection control. For MARCKSL1 overexpression experiments, we generated lentivirus encoding human cDNA of wild-type MARCKSL1, cloned under Tet-on promoter and followed by IRES-driven EGFP. This lentivirus was used to co-infect ES cells in addition to rtTA and Ngn2-t2a-Puromycin^Resistance –expressing virus. The control lentivirus contained an IRES-EGFP construct.

Lentivirus production

The expression vectors (20 μg) and three helper plasmids (pRSV–REV, pMDLg/pRRE, and VSV-G, 10 μg each) were co-transfected with Polyethylenimine transfection method into 70–75% confluent human embryonic kidney 293T (HEK) cells plated on polyornithine-coated 10 cm dishes. At 7–8 hr after transfection, the HEK cell culture medium was completely replaced with fresh media and supernatants were collected at 36 hr and 60 hr. The pooled supernatant was first briefly centrifuged for 5 min at 1000 × g to remove the cell debris and further centrifuged at 23,000 rpm for 2 hr to obtain viral pallets. The pallet was re-suspended overnight in 100 μl HBSS, collected in Eppendorf tubes, snap frozen, and stored at −80°C for no more than 6 weeks before use.

Generation of human neurons

Human ES cells (H1 or H9 line) were maintained under feeder-free conditions in mTeSR media (STEMCELL Technologies). Media was changed every day. When cell density reached 70%–80% confluence, colonies were dissociated using PBS-EDTA (0.5 mM) and plated onto Matrigel (BD Biosciences)-coated plates at a 1:6 dilution. During passaging, the media was supplemented with 2 μM thiazovivin overnight. For human neuron generation, dissociated single cells were plated on Matrigel-coated glass-coverslips placed in 24-well plates. Cells were infected with lentiviruses containing expression constructs of rtTA and Ngn2-t2a-Puromycin^Resistance with an optional EGFP-expressing virus, in the presence of polybrene (8 μg/ml). Next day, media was replaced with N3 media (DMEM/F12 [Thermo Fisher], N2 [Thermo Fisher], and B27 [Thermo Fisher], supplemented with 12.5 mg of insulin [Sigma], penicillin/streptomycin [Thermo Fisher]) and doxycycline (2 μg/ml). Puromycin selection started from post-induction day 1 and continued till day 4–5. For long-term culture (> 5 days), neurons were further dissociated into single cells using accutase (Innovative Cell), and seeded on passage 3 mouse glia (derived from C57 pups, postnatal day 3). The day after, media was replaced with Neurobasal-A media (Neurobasal-A [Thermo Fisher], L-glut [Thermo Fisher], B27 [Thermo Fisher], penicillin/streptomycin [Thermo Fisher], doxycycline [2 μg/ml], BDNF [10 ng/ml, PeproTech], GDNF [20 ng/ml, PeproTech], NT3 [10 ng/ml, PeproTech], and fetal bovine serum [5%, GE Healthcare]). Media was half-exchanged every 3–4 days and additionally supplemented with 4 μM Ara-C (Sigma) to inhibit glial growth after reaching 70–80% confluency.

Generation of mouse neurons

Doxycycline-inducible flag-Ngn2 or flag-ASCL1 mouse ES cell lines were established using protocol described previously (Chanda et al., 2014; Wapinski et al., 2013). In brief, a CAGGS-promoter driven rtTA-t2a-Puromycin^Resistance cassette was targeted into the ROSA26 locus of the V6.5 mouse ES cell line. A targeted clone was then infected with TetO-flag-Ngn2 or TetO-flag-Ascl1 lentivirus, and successfully infected colonies were hand-picked to establish Ngn2 or Ascl1-inducible mouse ES cell lines. The cells were maintained and expanded in mouse ES cell media (50 ml KOSR [Thermo Fisher], 12.5 ml CCS [Thermo Fisher], 4.2 ml penicillin/streptomycin [Thermo Fisher], 4.2 ml non-essential amino acids [Thermo Fisher], 4.2 ml Na-Pyruvate [Thermo Fisher], 417 μl LIF [Stemgent], 4 μl β-Mercaptoethanol [Sigma], and 341.5 ml DMEM [Thermo Fisher]). For neuronal induction, cells were plated on matrigel-coated coverslips placed in 24-well plates, and the media was replaced with N3 plus doxycycline the day after seeding. Media was half-exchanged every day.

Generation and neural differentiation of human NSCs

To obtain human NSC monolayers, we used the dual SMAD inhibition protocol (Chambers et al., 2009) with following modifications. Cultures of H1-ES cell at about 20% confluence were treated with two small molecules to inhibit Activin/BMP/TGF-β pathways: LDN193189 (100 nM, inhibits ALK1/2/3/6 receptors, STEMCELL Technologies) and SB431542 (10 μM, inhibits ALK4/5/7 receptors, STEMCELL Technologies), and the culture media was switched from mTeSR to N3 media. The N3 media with LDN193189+ SB431542 was replenished every day, and a highly confluent Nanog⁻/Oct3/4⁻ but Sox1⁺/Pax6⁺ NSC monolayers with many visible neural rosettes were obtained after 6 days. The NSCs were dissociated into single cells with accutase, and replated on Matrigel-coated coverslips at 1:50 ratio. The cells were subsequently cultured in N3 media with BDNF, GDNF and NT3. During this period, NSCs spontaneously differentiated into Tuj1⁺/Dcx⁺ neurons. For prolonged cultures and electrophysiology experiments, cells were infected with lentiviruses expressing rtTA and TetO-EGFP at day 1 after NSC dissociation, doxycycline was added at day 2, and further NSC-proliferation was by inhibited by AraC treatment at day 5. The cells were further dissociated at day 7, and co-cultured with passage 3 mouse primary glia in Neurobasal-A media. Doxycycline was used to visualize the neurons, Ara-C was used to limit glial overgrowth, and the media was changed every 3–4 days.

Drug treatments

For most experiments, VPA (Tocris, Sigma-Aldrich) was used at a concentration of 1 mM, except when mentioned otherwise (VPA dose-dependence, Fig. 2C, D; VPA-injection into pregnant mice, Figs. S4 and S6). Other drugs used in this study included TSA (Tocris), CI994 (Selleckchem), Nicotinamide (Tocris), GO6983 (Tocris), VPD (Sigma-Aldrich), CHIR99021 (Tocris), SB216763 (Tocris), SB415286 (Tocris), myo-inositol (Sigma), KYP2047 (Sigma), PD98059 (Tocris), FR180204 (Tocris), and SP600125 (Tocris). For all experimental procedures, relevant drug concentrations and treatment protocols were described in the figure legends. In control conditions, cells were treated with equal volumes of dH₂O or DMSO.

In vivo mouse experiments

All animal procedures were approved by the administrative panel on laboratory animal care (APLAC, Stanford University). The embryonic day was considered E0 by vaginal plug detection. The pregnant animals received a single intra-peritoneal (IP) injection with VPA (200 mg/kg body-weight) dissolved in PBS, or an equal volume of PBS (control), every day for four consecutive days from E12.5 to E15.5. We observed an immediate reduction in animal activity within 5–10 min of VPA injection that lasted for several hrs (not quantified), consistent with its anti-epileptic effects (Video V1). These animals were sacrificed on E16.5, and embryos were collected on ice. The fetal brains were isolated, immediately placed in 4% paraformaldehyde, fixed overnight at 4°C, and cryo-protected afterwards in 25% sucrose (in PBS) solution. Brains were then cryo-embedded with Tissue-Tek OCT compound (Sakura Finetek), and coronal sections (50 μm) were obtained using a cryostat (CM 3050S, Leica). The sections were washed 4 times with PBS, and processed for antibody staining.

Immunostaining

For immunofluorescence stainings, cultures were washed with PBS and then fixed with 4% paraformaldehyde for 15–20 min at room temperature. Cells were permeabilized and blocked with 0.1% Triton X-100 (Sigma) and 10% cosmic calf serum (CCS) in PBS for 30 min. Primary and secondary antibodies were diluted in blocking buffer. Neurons were incubated with primary antibodies for 1 hr at room temperature, washed three-times with PBS, and then incubated with secondary antibodies for 1 hr. The fetal brain slices from VPA-exposed embryos were immunostained similarly, except incubations with primary antibodies were performed overnight at 4°C. The neurons were then washed three more times with PBS, the coverslips were mounted on glass slides using Fluoromount-G (Southern Biotech), and the mages were acquired with A1Rsi (Nikon) confocal microscope. Sequential acquisition was used to avoid bleed through between channels. The laser power, PMT gain, and offset parameters were adjusted to avoid background signals and pixel saturation, but kept constant for all conditions within same experiment.

Antibodies used were rabbit anti-β-catenin (1:500, Abcam, Cat# ab32572), goat anti-Dcx (1:500, Santa Cruz Biotechnology, Cat# sc-8066 and sc-8067), chicken anti-EGFP (1:1000, Aves Labs, Cat# GFP-1020), rabbit anti-Histone H3 acetyl-K27 (1:1000, Abcam, Cat# ab4729), mouse anti-Ki-67 (1:500, BD Biosciences, Cat# 550609), chicken anti-MAP2 (1:500, Abcam, Cat# ab5392), mouse anti-MAP2 (1:500, Sigma-Aldrich, Cat# M1406), rabbit anti-MARCKSL1 (1:500, Proteintech Group, Cat# 10002–2-AP), rabbit anti-Nanog (1:200, Abcam, Cat# ab21624), mouse anti-Nestin (1:500, R&D Systems, Cat# MAB1259), rabbit anti-NeuN (1:500, Millipore, Cat# ABN78), mouse anti-Oct3/4 (1:200, Santa Cruz Biotechnology, Cat# sc-5279), rabbit anti-Pax6 (1:300, Covance, Cat# PRB-278P), mouse anti-Synapsin (1:500, Synaptic Systems, Cat# 106 001), rabbit anti-Synapsin (1:500, Südhof lab, Cat# E028), goat anti-Sox1 (1:500, R&D Systems, Cat# AF3369), goat anti-Sox2 (1:500, R&D Systems, Cat# AF2018), mouse anti-Tuj1 (1:1000, BioLegend, Cat# 801202), rabbit anti-Tuj1 (1:1000, BioLegend, Cat# 802001), and DAPI (1:50000, Thermo Fisher Scientific, Cat# D1306), and Alexa Fluor 488/555/647-conjugated secondary antibodies (Invitrogen).

Image analysis

All image analyses were performed using the Image J (NIH, USA) software with maximum-intensity projection of optical sections (5–10 sections, 0.4–0.5 μM thickness). Quantifications of neurite (EGFP-labeled) or dendritic (MAP2-immunoreactive) arborization were performed with NeuronJ plug-in and manual curation, or automated with NeuronStudio software (Icahn School of Medicine). For Synapsin puncta and MARCKSL1 cluster analysis, images were thresholded by the intensity to exclude background signals and only the puncta number and/or puncta density were calculated. The co-localization analyses between Synapsin and MARCKSL1 signals were achieved using JACoP plug-in with Image J.

Electrophysiology

Electrophysiology experiments were performed similarly to those described before (Chanda et al., 2013). All recordings were performed after VPA/drug-washout following the treatment periods, and using VPA/drug-free internal and external solutions to avoid any acute effects of VPA/drug-application. In brief, EGFP-positive human neurons were patched using internal solution containing (for current clamp, in mM) 130 KMeSO₃, 10 NaCl, 2 MgCl₂, 0.5 EGTA, 0.16 CaCl₂, 4 Na₂ATP, 0.4 NaGTP, 14 Tris-creatine phosphate, and 10 HEPES-KOH (pH 7.3, 310 mOsm); or (for voltage clamp, in mM) 135 CsCl₂, 1 EGTA, 1 NaGTP, and 2 QX-314, and 10 HEPES-CsOH (pH 7.4, 310 mOsm). The extracellular solution contained (in mM) 140 NaCl, 5 KCl, 2 CaCl₂, 1 MgCl₂, 10 glucose, and 10 HEPES-NaOH (pH 7.4). Current-induced AP recordings were performed at around −60 mV, using a small holding current to adjust the membrane potential accordingly. Voltage-clamp recordings for voltage-gated Na⁺/K⁺ channel currents and AMPAR-mediated EPSCs were performed at a holding potential of −70 mV. Puff-application of 100 μM AMPA (R-S AMPA hydrobromide) was performed for 50 ms using Picospritzer III.

Library preparation and RNA-sequencing

The Smarter Ultra Low Input RNA kit (Clontech, Catalog# 634848) or TruSeq kit (Illumina) was used to generate cDNA from the total RNA extracted from FACS-sorted Ngn2 neurons using the standard protocol. Amplified cDNA was purified using SPRI Ampure Beads (Beckman Coulter) and the quality and quantity was measured using a High Sensitivity DNA chip on the Agilent 2100 Bioanalyzer. The cDNA was sheared to an average length of 300 bp using a Covaris S2. The indexed libraries were pooled, quantitated, and sequenced, producing paired ends 2×75 reads. Raw reads were then mapped to the human reference genome (hg19) using Tophat. RNA sequencing for the two conditions was performed in 3–4 biological replicates. Expression levels for the coding genes were estimated using Cufflinks. Candidates genes were filtered and clustered using GeneCluster 3.0 (Eisen et al., 1998) and visualized as a heat map using TreeView (Saldanha, 2004).

The Tissue Expression, Protein Domains, and GO analysis were performed using DAVID 6.8 with annotations for UP_TISSUE (93.8% coverage), INTERPRO (91.2% coverage), and GOTERM_BP_DIRECT (85.6% coverage), respectively. Top candidates from the KEGG_PATHWAY annotations were reported for some VPA-affected gene-sets. For Disease Ontology (DO) analysis, batch query was performed against Human Disease database using MGI software (Jackson Lab), and the diseases were sorted by highest occurrences in VPA-affected significant gene list. Scatter plots for different GO-terms (source: MGI, Jackson Lab) were generated by plotting the FPKM values vs. corresponding P-values of associated genes with/without VPA treatment.

Quantitative RT-PCR

A total of 200 ng of total RNA was reverse-transcribed into cDNA using the First-Strand cDNA Synthesis kit (Life Technologies) with SuperScript II or III reverse transcriptase. Template cDNA was amplified using SYBR Master Mix, and quantitative RT-PCR was carried out on the AB7900HT (Life Technologies). Relative quantity (RQ) values were calculated by the delta-delta $\text{Ct}(RQ=2^{-\left[\text{Δ}C{t}_{Sample}-\text{Δ}C{t}_{control}\right]})$ methods. GAPDH was used to normalize the expression levels of each sample (ΔCt), and untreated samples were used as calibration control. Primer information for individual assays are provided in Fig. S1, S3, and S6.

Fluorescent western blot

Samples were lysed and harvested using RIPA buffer and boiled with sample buffer. They were separated by SDS-PAGE and transferred to a PDVF membrane. Immunoblots were then blocked using 0.1% Tween-20 and 5% BSA in PBS for 30 minutes and incubated overnight at 4°C with primary antibodies. The blots were wa shed with 0.1% Tween-20 in PBS for three times before the secondary antibody in blocking solution was added. The blots were next incubated for 2 hours, washed twice with 0.1% Tween-20 in PBS, and the fluorescence signals were scanned with Odyssey-system (LiCor), followed by quantification with ImageStudio software (LiCor). The antibodies used were: rabbit anti-MARCKSL1 (1:500, Proteintech Group, Cat# 10002–2-AP), and rabbit anti-HSP90 (1:5000, Cell Signaling Technology, Cat# 4874).

• QUANTIFICATION AND STATISTICAL ANALYSIS

For most experiments, the “n” represents total number of cells patched (for electrophysiology) or field-of-views analyzed (for imaging) / number of independent batches, and indicated with corresponding average values. For batch-wise comparisons, only the number of batches were indicated. The data from in vivo experiments were presented as number of brain-sections analyzed / number of embryos sectioned / number of pregnant animals. All average data are presented as means ± SEM (SD of parameter tested / square root of number of samples). In most cases, statistical comparisons between conditions were made using unpaired, two-tailed, Student’s t-test (*** P < 0.005; ** P < 0.01; * P < 0.05; ns = not significant, P > 0.05), except for batch-wise comparisons, where paired t-test was used (as mentioned in the figure legends). In some cases (group assessments), two-way Anova (*** P < 0.005; ** P < 0.01; * P < 0.05; ns = not significant, P > 0.05) was performed (as mentioned in the figure legends).

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit anti-β-catenin	Abcam	Cat# ab32572, RRID: AB_725966
Goat anti-Dcx	Santa Cruz Biotechnology	Cat# sc-8066, RRID: AB_2088494
Goat anti-Dcx	Santa Cruz Biotechnology	Cat# sc-8067, RRID: AB_2088491
Chicken anti-EGFP	Aves Labs	Cat# GFP-1020, RRID: AB_10000240
Rabbit anti-Histone H3 acetyl-K27	Abcam	Cat# ab4729, RRID: AB_2118291
Rabbit anti-HSP90	Cell Signaling Technology	Cat# 4874, RRID: AB_2121214
Mouse anti-Ki-67	BD Biosciences	Cat# 550609, RRID: AB_393778
Chicken anti-MAP2	Abcam	Cat# ab5392, RRID: AB_2138153
Mouse anti-MAP2	Sigma-Aldrich	Cat# M1406, RRID: AB_477171
Rabbit anti-MARCKSL1	Proteintech Group	Cat# 10002–2-AP, RRID: AB_513892
Rabbit anti-Nanog	Abcam	Cat# ab21624, RRID: AB_446437
Mouse anti-Nestin	R&D Systems	Cat# MAB1259, RRID: AB_2251304
Rabbit anti-NeuN	Millipore	Cat# ABN78, RRID: AB_10807945
Mouse anti-Oct¾	Santa Cruz Biotechnology	Cat# sc-5279, RRID: AB_628051
Rabbit anti-Pax6	Covance	Cat# PRB-278P, RRID: AB_291612
Mouse anti-Synapsin	Synaptic Systems	Cat# 106 001, RRID: AB_887805
Rabbit anti-Synapsin	Südhof lab	Cat# E028, RRID: AB_2315400
Goat anti-Sox1	R&D Systems	Cat# AF3369, RRID: AB_2239879
Goat anti-Sox2	R&D Systems	Cat# AF2018, RRID: AB_355110
Mouse anti-Tuj1	BioLegend	Cat# 801202, RRID: AB_10063408
Rabbit anti-Tuj1	BioLegend	Cat# 802001, RRID: AB_2564645
DAPI	Thermo Fisher Scientific	Cat# D1306, RRID: AB_2629482
Chemicals, Peptides, and Recombinant Proteins
Accutase	Innovative Cell	Cat# AT104
Ara-C	Sigma-Aldrich	Cat# C6645
B-27 supplement	Thermo Fisher	Cat# 12587010
BDNF	PeproTech	Cat# 450–02
Blasticidin	Sigma-Aldrich	Cat# 203350
CHIR99021	Tocris	Cat# 4423
Doxycycline	Sigma-Aldrich	Cat# D9891
DMEM media	Thermo Fisher	Cat# 11965092
DMEM/F-12 media	Thermo Fisher	Cat 11320082
FR180204	Tocris	Cat# 3706
GDNF	PeproTech	Cat# 450–10
GO6983	Tocris	Cat# 2285
HyClone cosmic calf serum (CCS)	GE Healthcare	Cat# SH30087.04
HyClone fetal bovine serum (FBS)	GE Healthcare	Cat# SH30071.03
Insulin	Sigma-Aldrich	Cat# I6634
KYP2047	Sigma-Aldrich	Cat# SML0208
LDN193189	STEMCELL Technologies	Cat# 72147
L-Glutamine	Thermo Fisher	Cat# 25030164
Matrigel	BD Biosciences	Cat# 356230
mTeSR1 media	STEMCELL Technologies	Cat# 85850
Myo-inositol	Sigma-Aldrich	Cat# I5125
N-2 supplement	Thermo Fisher	Cat# 17502048
Neurobasal-A Medium	Thermo Fisher	Cat# 10888022
Nicotinamide	Tocris	Cat# 4106
NT-3	PeproTech	Cat# 450–03
PD98059	Tocris	Cat# 1213
Penicillin-Streptomycin	Thermo Fisher	Cat# 15140163
Polybrene	Sigma-Aldrich	Cat# H9268
Polyethylenimine (PEI)	Polysciences Inc	Cat# 23966
Puromycin	Sigma-Aldrich	Cat# P8833
SB216763	Tocris	Cat# 1616
SB415286	Tocris	Cat# 1617
SB431542	STEMCELL Technologies	Cat# 72232
SP600125	Tocris	Cat# 1496
Tacedinaline (CI994)	Selleckchem	Cat# S2818
Thiazovivin	Axon medchem	Cat# 1535
Trichostatin A (TSA)	Tocris	Cat# 1406
Valproic acid (VPA), sodium salt	Sigma-Aldrich	Cat# P4543
VPA, sodium salt	Tocris	Cat# 2815
Valpromide (VPD)	Sigma-Aldrich	Cat# V3640
Deposited Data
RNA-sequencing data	This study	GEO accession # GSE129241
Experimental Models: Cell Lines
Human H1-ES cells	WiCell Research Institute	Cat# WA01
Human H9-ES cells	WiCell Research Institute	Cat# WA09
Mouse ES cells-Inducible Ngn2	Wernig lab	Wapinski et al., 2013
Mouse ES cells-Inducible Ascl1	Wernig lab	Wapinski et al., 2013
HEK 293T cells	ATCC	Cat# CRL-11268
Experimental Models: Organisms/Strains
C57BL/6J mouse	The Jackson Laboratory	Cat# 000664
CD-1® IGS mouse	Charles River	Cat# Crl:CD1(ICR)
Oligonucleotides
shRNA HDAC½/3/8	Mission shRNA, Sigma	Detailed in Supplementary Fig. S3N
shRNA MARCKSL1	Mission shRNA, Sigma	Detailed in Supplementary Fig. S6D
Recombinant DNA
TetO-EGFP	Südhof lab	Zhang et al., 2013
TetO-MARCKSL1-IRES-EGFP	This study	N/A
TetO-Ngn2-T2A-BlasticidinResistance	This study	N/A
TetO-Ngn2-T2A-PuromycinResistance	Südhof lab	Zhang et al., 2013
Ub-rtTA	Südhof lab	Zhang et al., 2013
Software and Algorithms
ImageJ	NIH	RRID: SCR_003070
Igor Pro	Wavemetrics	RRID: SCR_000325
NeuronStudio	Icahn School of Medicine	RRID: SCR_013798
pCLAMP 10	Molecular Devices	RRID: SCR_011323

HIGHLIGHTS.

1. Reprogrammed human neurons progress through distinct developmental stages.

2. VPA-treatment impairs the morphology and functional maturation of developing neurons.

3. VPA-exposure alters the neuronal properties by inhibiting HDAC and GSK-3 pathways.

4. The cytoskeletal protein MARCKSL1 mediates VPA-induced morphological defects.