Female C57BL/6 mice (7–8 weeks old) were mated with male mice at 8 pm. The vaginal plug was examined at 8 am the following morning. Those females having plugs were singly housed and denoted as embryonic at day 0.5 (E0.5). VPA (500 mg/kg body weight, in sodium salt, Sigma, Ronkonkoma, NY, USA) was prepared in 50 mg/mL sterile saline and injected into the peritoneal cavity of pregnant mice at E10.5. The control group received 0.1 mL sterile saline. Male offspring were used for further assays. The animal protocol was approved by the Jinan University Institutional Animal Care and Use Committee.

To characterize postnatal developmental patterns of mice, we utilized a test battery of developmental milestones as previously described (Zhang et al., 2014). Briefly speaking, juvenile mice from both VPA-treated and saline-treated groups were examined for different neural reflex and body development markers, including cliff avoidance (the avoidance behavior when the mouse’ forepaws were approaching the edge of one platform), grasping reflex (to grasp one small metal bar with the forepaws), negative geotaxis (to recover from a head-down position on an inclined plane), surface righting (turning its body around from an upside-down position), air righting (recovering a normal prone position when releasing from the height in a supine position), bar holding (ability to hold a metal bar with the forepaws for more than 5 s), pinnae detachment and eye opening. The day of onset was recorded, and any positive reflexes were confirmed the following day. We examined two cerebellar associated motor reflexes: (1) negative geotaxis: the mouse was placed on a 30° inclined plane with its head facing downwards. The positive reflex was identified when the mouse could turn its body around to a head-up position within 30 s; and (2) air righting reflex: the mouse was released from a 30 cm height using an inverted supine position (with its belly facing upwards). The positive reflex was identified when the mouse could recover the normal prone position (belly facing downwards against the soft bedding) when landing.

To validate the ASD phenotype of the VPA model, the 3-chamber sociability assay was employed first, as previously described (Peter et al., 2016). In brief, the mouse was first placed into the central zone to freely explore the three chambers (divided from a 60 cm × 45 cm clear box) for 15 min. During the second 15-min stage, one age-matched male mouse (S1) was placed into one chamber within a round wire cage. The time for the test mouse to explore and interact with the S1 mouse was recorded. For the third stage, a second stranger mouse (S2) was introduced to the other side of the chamber and the interaction time with both the S1 and S2 mice, was recorded during the 15-min session.

For the open field assay, a clear box (50 cm × 50 cm) was utilized to measure the locomotor activity of mice in a single 15 min session, during which the total distance of each mouse was analyzed.

The accelerating Rota-rod test was performed as previously described (Zhang et al., 2014). The mouse was trained on the rod in six consecutive daily sessions, with 5 rpm initial velocity and 80 rpm maximal at 5 min.

To further elaborate motor learning function, we employed the LadderScan apparatus (Clever System Inc., Reston, VA, USA; Figure 1A). The design of this learning paradigm was modified from a previous report (Vinueza Veloz et al., 2015). The whole apparatus consists of one runway and two side-chambers, which are equipped with LED lights and fans. The runway includes 37 rungs (metal rods, 2 mm diameter, in two lines) with a 15 mm interval. On one side, rungs with an odd number (1, 3, 5, …, 37) were at a high level, whilst even numbered rungs (2, 4, 6, …, 36) were located at a lower level. There were alternative patterns (odd number at low, even number at high level) on the other side. The vertical distance between high and low rungs was 6 mm in height, and mice typically walked on the high rungs only. During the middle segment (rung 7–31), each lower rung was elevated by 18 mm to create one extra obstacle with a 12 mm height. All rungs were equipped with pressure sensors to time each gait. One single trial was initiated after placing the mouse into the chamber on either side. After a 9–12 s delay, light and air puffs were sequentially applied in the chamber to create negative stimuli, which forced the mouse to leave the chamber. The trial was terminated when the mouse successfully walked across the runway to reach the chamber on the other side. The whole behavioral paradigm started with a 4-day training run (25 daily trials for each mouse) to ensure the mouse ran smoothly on the runway. Starting on day 5 to day 8 (25 trials per day), one extra perturbation was introduced and consisted of one randomly “pop-up” rung ahead of the mouse. The elevation of rung occurred at any lower rung in the middle segment of the runway and was induced before a mouse approached, to avoid any spatial specific memory. The rung elevation was preceded by a tone cue (80 dB, 1,000 Hz, 250 ms duration, 250 ms before perturbation). The perturbation event occurred when the mouse performed at least three consecutive “regular steps,” which were defined as having a step length with two or four rungs. All the gait parameters were recorded for each trial and were analyzed.

EdU assay was employed to describe the migration of the GC precursors. Following previously documented methods (Wang et al., 2017), EdU (50 mg/kg, from Click-iT EdU Alexa Fluor 594 Imaging Kit, Invitrogen, Waltham, MA, USA) was injected into the peritoneal cavity of P7 mice. The whole brain was then harvested at 2 h or 72 h later and was prepared in 8 μm cryo-sections. The staining procedure followed manual instruction of the Click-iT EdU Alexa Fluor 594 Imaging Kit. A fluorescent microscope (Zeiss, Germany) was used to capture images. The number of EdU-positive cells in the EGL and internal granular layer (IGL) was counted by the ImageJ 1.48 (NIH, Bethesda, MD, USA).

We used a terminal deoxynucleotidyl transferase (TdT) dUTP nick-end labeling (TUNEL) assay to measure the apoptosis of the cerebellar neurons and precursor cells. The assay was performed using an in situ Cell Death Detection Kit (Roche, Swiss) following its manual instructions. In brief, cryo-sections (8 μm thickness) were permeabilized using a 0.1 M sodium citrate buffer (with 0.1% Triton X-100), followed by PBS washing. A freshly prepared TUNEL reaction buffer (50 μL per slide) was added at 37°C incubation for 1 h. After PBS rinsing, the DAPI (Roche, Switzerland) was used for nuclei staining. Fluorescent images were captured as previously described.

Golgi staining was performed using the Rapid Golgi Stain Kit (FD NeuroTech Inc., Ellicott City, MD, USA) according to the manual’s instructions. The Purkinje cells at the apical region of the cerebellar lobule were imaged using a bright field microscope (Zeiss, Germany). NeuroLucida software (MBF Bioscience, Williston, VT, USA) was used to plot the soma and dendrites of the Purkinje cells under a manually assisted mode. The Sholl analysis was adopted using previous methods (Wang et al., 2017). Briefly, a series of concentric circles (10 μm interval) were plotted around the soma. The number of intersections of dendrites against each circle, and the total dendrite lengths within each 10 μm segment, were quantified using NeuroLucida software.

Mice were deeply anesthetized by isoflurane and were perfused with 4% paraformaldehyde (PFA). Whole brain tissues were extracted and were dehydrated in a 30% sucrose solution at 4°C overnight. Brain slices (30 μm thickness) were prepared using a cryostat (Leica, Germany). Slices were rinsed in 0.01 M PBS and blocked in 3% FBS (with 0.1% Triton X-100) for 1 h. Primary antibody against calbindin D-28K (Abcam, Cambridge, MA, USA) was added for 48 h incubation at 4°C. Donkey anti-rabbit Alexa 488 secondary antibody (Life Technology, Camarillo, CA, USA) was added at room temperature for 2 h incubation. After PBS rinsing, DAPI was added for nuclear staining. Fluorescent images were taken for cell enumeration by the ImageJ (NIH).

The cerebellar tissues were homogenized by a RIPA buffer. Equal amounts of proteins were separated by an SDS-PAGE and were transferred to a PVDF membrane (GE Healthcare, Chicago, IL, USA). The membrane was blocked by a 5% bovine serum albumin (BSA) for 2 h and was incubated in a primary antibody against BDNF, p-TrkB (p-tyrosine kinase B), t-TrkB and β-actin (Cell Signaling Technology, Danvers, MA, USA) overnight at 4°C. The membrane was then incubated in horseradish peroxidase (HRP) conjugated anti-IgG antibody (Cell Signaling) for 2 h at room temperature. Protein bands were visualized by an ECL chemiluminescence substrate. A protein imaging system (Bio-Rad, Hercules, CA, USA) was used to capture images, and their integrated density was measured by the ImageJ (NIH).

Mice (P18–P21) were anesthetized with isoflurane and decapitated. Sagittal slices (300 μm) were prepared in ice-cold “cutting solution” containing (in mM): 240 sucrose, 2.5 KCl, 10 D-Glucose, 26 NaHCO₃, 1.25 Na₂HPO₄, 2 MgCl₂ and 1 CaCl₂, using a Leica VT1200S Vibratome. The slices were incubated in a submerged chamber containing an equal volume of cutting solution and recording solution, at 32 ± 1°C for 30 min and subsequently at room temperature. The recording solution contained (in mM): 126 NaCl, 26 NaHCO₃, 10 D-glucose, 1.25 NaH₂PO₄, 2.5 KCl, 2 CaCl₂ and 2 MgCl₂. All solutions were bubbled with 95% O₂ and 5% CO₂ and maintained at 7.4 pH. For the whole cell recordings, slices were perfused with the recording solution at room temperature. The Purkinje cells were visualized using an infrared differential interference contrast with a Nikon Eclipse FN-1 microscope with a 40× water-immersion objective. Recordings were performed using the MultiClamp 700B (Molecular Devices, San Jose, CA, USA). The recording electrodes (4–8 MΩ) were filled with an intracellular solution containing (in mM): 126 K-gluconate, 4 KCl, 10 HEPES, 4 Mg-ATP, 0.3 Na-GTP, 10 PO-creatinine (pH 7.25 with an osmolarity of 295 ± 5). Data were low-pass filtered at 3 kHz and digitized with the DigiData 1550B using a pClamp 10 at 10 kHz sampling frequency. Miniature excitatory postsynaptic currents (mEPSCs, in present of 1 μM tetrodotoxin) were recorded as inward currents at a holding potential of −70 mV and miniature inhibitory postsynaptic currents (mIPSCs) were recorded as outward currents at a holding potential of 0 mV. Both mEPSCs and mIPSCs were measured using automated event detection in the Clampfit software (Molecular Devices, San Jose, CA, USA) using 5.5 as the match threshold, followed by a manual inspection to exclude any artifacts. All events were extracted from the records with 2 min durations. For plotting the cumulative distribution curve, 40 events were extracted from the recording series of one neuron. The averaged amplitudes and frequencies were calculated from all events in each neuron.

All data were presented as mean ± standard error of means (SEM), unless otherwise specified. Each group of data were first tested for normality using the KS test. Data that did not fit the Gaussian distribution, were compared by a non-parametric Mann-Whitney U test. For data with a normal distribution, a 2-sample student t-test was used for comparison between the two groups. A one-way or two-way analysis of variance (ANOVA), with repeated measures, was used for comparison among the multiple groups, depending on the number of independent variables. The Tukey or Bonferroni post hoc comparison was used to compare means between two specific groups after one-way or two-way ANOVA, respectively.

The cerebellum is closely related with the acquisition of complex motor skills. We thus employed different motor learning paradigms to study the cerebellar associated motor functions on adult (P56) mice. The classical Rotarod assay showed that compared to the saline-treated controls, the VPA-treated group generally performed worse, especially at later training sessions (Supplementary Figure 1A). This impaired motor skill seems to not be caused by a general motor deficit, as the VPA-treated mice showed a similar movement distance in an open-field assay (Supplementary Figure 1B). However, as no significant difference was found in the learning ability on the Rota-rod (two-way ANOVA with repeated measures, interaction effect, F_(5,120) = 0.9056, P = 0.4799, Supplementary Figure 1A), more sensitive motor learning paradigms were expected. We thus introduced a tone-conditioned complex motor learning paradigm named the LadderScan (Figure 1A, see “Materials and Methods” section for detailed information) which was modified from a previous report (Vinueza Veloz et al., 2015). The whole test paradigm consisted of a 4-day habituation session followed by a second 4-day perturbation session. In general, VPA-treated mice showed a similar time duration completing a single trial compared to the saline-treated control group during the habituation stage (Figure 1B), and both groups showed decreased durations, indicating a normal motor function in the VPA-treated mice. After introducing the perturbation, however, the VPA-treated group showed a sharp increase in duration on the walkway, which was maintained at relatively low levels in the control group (two-way ANOVA, treatment × training day effect: F_(7,1496) = 2.54, P = 0.0134; Bonferroni post hoc comparison between VPA- and saline-treated group: P = 0.0137, 0.0479 and 0.0371 from day 6 to day 8; Figure 1B). Therefore, the VPA-treated mice seem to have deficits in the acquisition of this complex motor task with perturbation. When checking the step regularity index, which was defined as the number of consecutively regular steps in a single trial, the VPA-treated mice did not improve significantly with repeated training as compared to the control group (two-way ANOVA, treatment × training day effect: F_(7,1496) = 2.232, P = 0.0294; Bonferroni post hoc comparison: P = 0.0018, 0.0024, 0.0001, 0.0253, 0.0173 and 0.0005 from day 3 to day 8; Figure 1C). Similar results were obtained when checking the average duration of each regular step, which was continuously decreased in both the VPA-treated mice and the control group, with repeated training, but was re-elevated after perturbation introduction only in the VPA-treated group (two-way ANOVA, treatment × training day effect: F_(7,1496) = 2.524, P = 0.0139; Bonferroni post hoc comparison: P = 0.0423, 0.0374, 0.0109 and 0.0035 from day 5 to day 8; Figure 1D). We also analyzed the response of animals toward perturbation and found that the VPA-treated mice had a lower rate of successfully crossing the perturbation (“jump over”) compared to the control group (4.8%–10.4% vs. 9.6%–16.8%, t₍₆₎ = 3.232, P = 0.0179; Figure 1E). Moreover, the VPA-treated mice presented a higher rate of “step on” on the elevated rung (31.2%–40.8% of all sessions at day 5 to day 8) compared to the saline-treated control group (6.4%–24.0%, t₍₆₎ = 5.024, P = 0.0024; Figure 1E). In summary, the LadderScan paradigm suggested that the VPA-treated mice developed motor incoordination when faced with unexpected perturbation, while they maintained normal motor function during relatively simple motor training. Such impairment in complex motor learning did not improve significantly with repeated training. These results supported the impaired motor learning after prenatal VPA exposure.

Having observed the motor learning deficit in adult VPA-treated mice, we next explored if such impairment occurred at an early age. A series of developmental “milestones” was thus employed to describe both body development and motor reflex patterns during the first three postnatal weeks (Zhang et al., 2014). In examining two markers for the body development: pinnae detachment and eye opening, we found similar days of onset between the VPA- and saline-treated mice (Figure 2A), indicating unaltered general development. We further investigated the onset of several sensorimotor reflexes in juvenile mice including cliff avoidance, grasping, negative geotaxis, surface righting, air righting and bar holding. Among those markers, we found that prenatal exposure of VPA, led to the retardation of the negative geotaxis reflex, which is defined as the ability to turn the body around from a head-down position on an inclined plane. The VPA-treated mice also presented a late onset of the air righting reflex, in which the mouse can adjust its body to a prone position when landing on soft beddings, after releasing from a height in a supine posture. Statistical analysis showed the late onset of both reflexes in the VPA-treated group (Negative geotaxis: 5.9 ± 0.24 vs. 5.1 ± 0.20 days, Mann Whitney test, U = 122.5, P = 0.0275; Air-righting reflex: 12.6 ± 0.21 vs. 11.6 ± 0.32 days, U = 114, P = 0.0139; n = 20 in each group; Figures 2B,C). As previous reports have illustrated the impairment of the negative geotaxis (Holmes et al., 2006) and the air righting reflex (Wolf et al., 1996) in cerebellar developmental disorders, it is thus reasonable to speculate that prenatal VPA exposure may cause early postnatal developmental disorders of the cerebellum.

The abovementioned motor deficits and retardation of motor reflexes in the VPA-treated mice, indicate altered cerebellar development patterns. As the most abundant neuron type, the GC forms the major excitatory input to the Purkinje cells via a parallel fiber, in addition to a climbing fiber originating from the inferior olive nuclei. During early postnatal stage, GC precursors in the EGL undergo active proliferation followed by an inwardly radial migration to the IGL where they differentiate into mature GCs (Butts et al., 2014; Marzban et al., 2015). Using the EdU incorporation assay, we tracked the migration of newly generated GC precursors between P7 and P10 (Figure 3A). The VPA-treated mouse juveniles displayed significantly more EdU+ cells in the IGL at P7 (Mann Whitney test, U = 104, P = 0.0191; Figures 3B,D) but less cells in the EGL at P10 compared to the saline-treated control group (t₍₅₃₎ = 3.416, P = 0.0012; Figures 3C,E). No difference was found in the P7 EGL or P10 IGL (P = 0.3545 and 0.8849). Since the GC precursors migrate from the EGL to the IGL, it indicates that VPA-treated mice presented premature inward migration. Newly formed cells also undergo programmed cell death if not undergoing migration or maturation. We further used a TUNEL assay and found that the VPA-treated mice showed elevated apoptosis at P9 in the EGL, but fewer deaths in the IGL (EGL: t₍₁₆₎ = 2.673, P = 0.0167; IGL: t₍₁₆₎ = 3.75, P = 0.0017; Figures 3F,G). Such accelerated apoptosis persisted until P15 in the EGL but not for the IGL (Supplementary Figure S2). Combining these results, prenatal VPA exposure leads to premature GC precursor migration in the cerebellar cortex during the second postnatal week, in association with higher apoptosis.

Cerebellar development is one well-orchestrated process involving neurogenesis, dendritic arborization and synaptogenesis. After demonstrating premature migration and excessed apoptosis of GC precursors, we next investigated the Purkinje cells. In adult mice, we found a significantly decreased number of Purkinje neurons in most of the lobules (two-way ANOVA, group effect: F_(1,265) = 248.5, P < 0.0001; Bonferroni post hoc comparison: P < 0.001 for all lobules but lobule VII and lobule X, Figures 4A,B). On average, the Purkinje cell number was decreased by 15.3%–52.3% across the lobules in the VPA-treated mice, suggesting an impaired Purkinje cell population. Due to normal cerebellar size and lobular formation of the cerebellum (Figure 4A), VPA exposure decreased the Purkinje cell density. Moreover, we examined the dendritic formation of the Purkinje cells and found that at P9, the VPA-treated mice showed a lower dendritic field radius (Supplementary Figure S3). A Sholl analysis indicated altered dendritic arborization patterns at this stage, as indicated by the higher complexity in the proximal dendritic segment but fewer branches at the distal site (Supplementary Figure S3). At the adult stage (P30), the Purkinje cells in the VPA-treated group presented a similar dendritic field radius compared to the control group (t₍₁₆₎ = 0.7583, P = 0.4593; Figures 4C,D). However, the dendritic arborization was remarkably impaired, especially at the middle-to-distal segment (70–120 μm from the soma) of the dendritic tree, as suggested by fewer intersections or shorter dendrite lengths (intersection number: two-way ANOVA, group effect; F_(1,235) = 56.64, P < 0.0001; Bonferroni post hoc comparison: P < 0.05 from 70 μm to 120 μm segment; Figures 4C,E. Dendrite length: F_(1,235) = 17.85, P < 0.0001; Bonferroni post hoc comparison: P < 0.05 from 70 μm to 100 μm segment; Figures 4C,F). The data therefore suggested that prenatal VPA exposure ablated the normal population and dendritic formation of the Purkinje neurons, in a progressive manner, from the juvenile until the adult stage.

The Purkinje neurons receive excitatory synaptic inputs from the GCs via parallel fibers and from climbing fibers originating from the inferior olive nuclei, as well as inhibitory inputs from the interneurons in the molecular layers. Therefore, the impaired dendritic arborization of the Purkinje cell in the VPA-treated mouse model may lead to an aberrant synaptic transmission. To test this hypothesis, we prepared acute cerebellar slices from both VPA- and saline-treated mice and recorded the excitatory and inhibitory transmission on the Purkinje cells. When examining the mEPSC, the VPA-treated group had unchanged amplitudes (Mann Whitney test, U = 237, P = 0.5759; Figures 5A,B) but remarkably decreased frequencies compared to the control group (0.58 ± 0.12 Hz vs. 1.06 ± 0.15 Hz, U = 136, P = 0.0155; Figure 5C). This evidence strongly supports the impaired excitatory transmission in the cerebellar cortex. We also examined the mIPSC, which is evoked by the interneurons including the basket cells and stellate cells in the molecular layer. VPA-treated mice also showed decreased mIPSC amplitudes (11.88 ± 1.73 pA vs. 19.17 ± 2.28 pA, U = 45, P = 0.0446; Figures 5D,E) and suppressed frequencies (1.74 ± 0.53 Hz vs. 3.57 ± 0.32 Hz, U = 54, P = 0.0155; Figure 5F). In sum, prenatal VPA exposure led to suppressed excitatory and inhibitory transmission, which is consistent with the morphological alternations and motor learning deficits mentioned above.