All mice were group-housed at the University of Fribourg, Switzerland in temperature-controlled animal facilities (24°C, 12:12 h light/dark cycle), fed ad libitum with free access to water. C57Bl/6J mice were mated overnight until a vaginal plug was detected in the morning. The day of sperm plug detection was defined as gestational day 0 (GD0). At GD12, 600 mg/kg valproic acid sodium salt (VPA; P4543 Sigma–Aldrich, Buchs, Switzerland) diluted in 0.9% NaCl was administered by oral gavage. Control animals were administered with 0.9% NaCl. Pups were not weaned; brains were taken after cerebral dislocation at postnatal day (PND) 25 ± 1. Only male mice were used in this study. All experiments were performed with permission of the local animal care committee (Canton of Fribourg, Switzerland) and according to the present Swiss law and the European Communities Council Directive of 24 November 1986 (86/609/EEC).

Mice at PND25 were anesthetized (Esconarkon, Streuli Pharma AG, Uznach, Switzerland) and perfused with 0.9% saline solution followed by 4% PFA. Brains were removed and post-fixed for 24 h in 4% PFA before being processed in 30% sucrose-TBS at 4°C. After cryopreservation, coronal sections were cut rostro-caudally using a freezing microtome (Frigomobil, Reichert-Jung, Vienna, Austria); six series of equidistant sections were collected using stereological systematic random sampling principles (see below). Free-floating sections were initially blocked with TBS 0.1 M, pH 7.3 plus 0.4% Triton X-100 and 10% bovine serum albumin (BSA) for 1 h at room temperature. Next, sections were washed three times with TBS, and incubated with PV antibody (anti-rabbit PV25, Swant, Marly, Switzerland) at a dilution of 1:1000 and with Vicia Villosa Agglutinin (biotinylated-VVA, Reactolab, Servion, Switzerland) at a concentration of 10 μg/ml in TBS containing MgCl₂, MnCl₂, CaCl₂ (final salt concentration: 0.1 mM each) overnight at 4°C. Sections were rinsed once with TBS, then twice with Tris-HCl 0.1 M, pH 8.2; they were then incubated (protected from light) at room temperature with anti-rabbit Cy3-conjugated antibody (1:200 dilution) and Cy2 streptavidin-conjugated antibody (at a dilution of 1:200, from Milan Analytic AG, Switzerland) in Tris-HCl. DAPI staining allowed to visualize nuclei of fixed cells (1:1000 dilution, LuBio Science GmbH, Luzern, Switzerland) in PBS 0.1 M, pH 7.4. After final rinsing, slides were coverslipped with Hydromount (National Diagnostics, Atlanta, GO, USA).

We used the optical fractionator method (West et al., 1991) to estimate the total number of PV-positive (PV⁺) and Vicia Villosa Agglutinin-binding (VVA⁺) neurons in brain regions of interest (ROIs) using the Stereo Investigator system (Version 11, MicroBrightField, Williston, VT, USA). The Stereo Investigator system was connected to a Zeiss Axioplan microscope with a motorized x-y stage (Ludl Electronic Products, LTD, NY, USA) and coupled to a Hamamtsu Orca Camera. ROIs were defined based on stereotactic coordinates provided by the Paxinos and Franklin atlas (Paxinos, 2001). The mPFC was defined at 1.94 to 1.10 mm from bregma comprising the anterior cingulate cortex, prelimbic area and infralimbic area; the corpus callosum, the midline between the hemispheres and fissura longitudinalis cerebri served as borders. The striatum (caudoputamen) was defined at 1.10 to –0.82 mm from bregma; the lateral ventricle, corpus callosum, capsula externa and globus pallidus externum served as borders, whereas the commisura anterior and rhinal fissure served as reference points. The SSC was defined at 1.94 to –1.82 mm from bregma; the corpus callosum and capsula externa served as borders, whereas the lateral ventricle, cingulum, hippocampus and outer brain curvature served as reference points. Cell counting was carried out on images obtained with oil immersion objective lenses (100x; NA = 1.40 and 63x; NA = 1.30). The volume of the analyzed brain structure was determined using the Cavalieri estimator (Gundersen et al., 1988). Five animals were analyzed per group, with pups from at least 3 different litters for each group. All results obtained from stereological quantification are reported in Table 3.

Sampling parameters are reported in Table 1. VVA⁺ and PV⁺ cells were counted independently and according to the following criteria: (1) Well visible DAPI-stained nucleus; (2) well-defined perineuronal net (PNN) with a web-/lattice-like morphology for VVA⁺ cells; examples are shown in Figure 1. (3) PV staining surrounding the DAPI-stained nucleus for PV⁺ neurons. The thickness of individual sections was measured at every fifth sampling location, and the mean of all measurements was used for all computations. At each selected location, the microscope was focused down through the disector sample in order to count any positive cell within that particular counting frame according to disector counting rules. Of importance, the cell number estimate is legitimate, even if the tissue volume changes during processing, because the fractionator method does not necessitate a measurement of tissue volume or any other dimensional quality. The total number of cells (N) in the selected ROIs was estimated as summarized by West et al. (1991, 1996) using the equation:

where Q^- signifies the number of objects (counts), and ssf, asf, and tsf correspond to the section sampling fraction, the area sampling fraction and the thickness sampling fraction, respectively. The accuracy of the estimates N is defined by the coefficient of error CE (Table 1), which represents the sampling error related to the counting noise, systematic uniform random sampling and variances in section thickness (Gundersen et al., 1999). For most biological samples, a CE value of 0.10 is largely accepted. CEs for both m = 1 and m = 0 are provided in Table 1 and are expected to bracket the true CE values of the estimates. In our double-labeling experiments, strictly individual fluorescence images for either PV⁺ or VVA⁺ structures were counted as reported before (Filice et al., 2016) without crosschecking the other channel. This ascertained that a cell (or a small part of a cell resulting from the sectioning) with “weak” staining intensity (i.e., below the predefined threshold) and/or characterized by an “atypical” shape (not easily discernable as cell-like) was considered negative, even if a check of the other channel would have probably identified this cell/cell segment as positive for the second marker.

Mice were euthanized by cervical dislocation and the brain was quickly removed and put in ice-cold 0.9% saline solution. The brain was cut in half along the midline and the cerebellum was removed. The hippocampus and striatum were dissected by carefully removing (pulling) these structures as described in Supplementary Figure S2. The remaining parts of the brain consisting essentially of cortex (approximately 80% of the total volume) and to minor extent (<20%) subcortical structures including thalamus and pallidum were collected as “forebrain” samples. All tissue samples were snap-frozen in liquid nitrogen and stored at –80°C for further use. The left hemisphere was always dissected first and used for qRT-PCR. Total RNA was extracted from mouse brain tissue (striatum and “forebrain” from the left hemisphere) using the peqGold TRIzol reagent (Peqlab, VWR International GmbH, Erlangen, Germany). cDNA was synthesized using ThermoFisher’s Verso cDNA Synthesis Kit (ThermoFisher, Lausanne, Switzerland). qRT-PCR was carried out to examine the expression of mRNA of the 18S rRNA, Ubc, Pvalb, Gad67, Kcnc1, Kcnc2, Kcns3, Hcn1, Hcn2, and Hcn4 genes using the universal 2X KAPA SYBR FAST qPCR Master Mix (Axonlab AG, Mont-sur-Lausanne, Switzerland). Details about the primer sequences are listed in Table 2. Gene expression quantitation was carried out in a DNA thermal cycler (Corbett Rotor gene 6000, QIAGEN Instruments AG, Hombrechtikon, Switzerland), according to the following two-steps protocol: a denaturation step of 95°C for 3 min; 40 cycles of denaturation at 95°C for 3 s and annealing/extension/data acquisition ranging from 52 to 62°C for 20 s. The housekeeping genes 18S ribosomal RNA (18S) or ubiquitin C (Ubc) were used as endogenous controls to normalize the mRNA content for each sample. In the second cohort of animals, Ubc mRNA signals were found to show lower variability (smaller S.D.) compared to 18S, both within individual animals and within the groups (VPA vs. control, data not shown). Thus Ubc was used to normalize levels of Kcnc2, Kcns3, Gad67, Hcn1, Hcn2, and Hcn4 mRNA levels in the striatum of these animals. However, normalization with 18S mRNA levels resulted in essentially similar values. mRNA levels were quantified by the 2^-ΔΔCt method and normalized to (I) housekeeping genes and furthermore (II) to the control mice group as described before (Livak and Schmittgen, 2001).

Frozen brain tissue from the dissected second (right) hemispheres as described above were homogenized and soluble or membrane proteins extracted for Western blotting experiments as described before (Racay et al., 2006; Maetzler et al., 2009). Denatured proteins (30 μg) were separated by SDS-PAGE (10–12.5%). After electrophoresis, the proteins were transferred on nitrocellulose membranes (MS solution, Chemie Brunschwig, Basel, Switzerland). The membranes were then blocked with 5% bovine serum albumin (BSA) in TBS for 60 min at room temperature and incubated with primary antibodies: rabbit anti-PV25 (Swant, Marly, Switzerland), rabbit anti-GAPDH (Sigma–Aldrich, Buchs, Switzerland) diluted 1:10,000, rabbit anti-K_v3.1b (Merck Millipore, Schaffhausen, Switzerland), rabbit anti-HCN1 (Alomone Labs, Jerusalem, Israel) diluted 1:200 in 2% BSA in TBS-T overnight at 4°C. Membranes were washed three times in TBS-T and incubated for 1 h with secondary antibody (goat anti-rabbit IgG HRP-conjugated, Sigma–Aldrich, Buchs, Switzerland) diluted at 1:10,000 in TBS-T. Finally, membranes were repeatedly rinsed in TBS-T and developed using ECL (Merck Millipore, Schaffhausen, Switzerland). Bands visualized by ECL were quantified using Image Studio Light Version 5.0. GAPDH signals were used as loading control.

Stereological data, mRNA and protein levels were compared between groups by the Student’s t-test. Data were analyzed using the GraphPad Prism 6 software (San Diego, USA). As no significant differences were observed when comparing ROIs in the two hemispheres of the same mouse, analyses were carried out with the pooled data of both hemispheres. The morphological data were initially checked for normal distribution by the Kolmogorov–Smirnov test and further analyzed with the Student’s t-test. A p-value < 0.05 was considered statistically significant.

Different protocols for the VPA animal model of ASD have been previously applied. We chose to use a dose of 600 mg/kg administered at GD12, because of the well-described behavioral phenotype observed in these mice (Gottfried et al., 2013; Mabunga et al., 2015). Oral gavage was performed to accurately control the dose and minimize risk to harm the pregnant females or the fetus. We did not detect any physical malformation or conspicuous features in the pups from VPA-treated mothers, except for five pups from two different litters that manifested few hairless spots at PND15. However, occurrence of hair loss is a characteristic of the C57Bl/6J and related mouse strains. Litter size, sex distribution and weight of the pups were not different between VPA and control mice (Supplementary Figure S1).

To evaluate the involvement of Pvalb neurons in the VPA mouse model, we performed stereology-based analysis of VPA mice at PND25 ± 1. The optical fractionator method was used to reliably quantify cell numbers in three ASD-associated brain regions, namely the striatum, medial prefrontal cortex (mPFC) and somatosensory cortex (SSC). Coefficient of error (CE) values ranged from 0.06 to 0.11 (Table 3) indicating a high precision of the estimates for both PV⁺ and VVA⁺ populations (Gundersen et al., 1999). One of the most crucial points when counting PV⁺ cells stained with an anti-PV antibody, is, whether the absence of a signal signifies “Pvalb neuron loss” or rather “PV down-regulation”. Therefore, in addition to counting PV⁺ cells via anti-PV immunostaining, we simultaneously quantified the number of Vicia Villosa Agglutinin VVA⁺ cells, i.e., neurons surrounded by PNNs in the same brain regions to obtain an alternative estimate for the number of Pvalb cells. VVA is a lectin that binds to N-acetylgalactosamine residues of PNNs, which specifically surround Pvalb neurons (Hartig et al., 1992; Haunso et al., 2000), as also shown in our previous study (Filice et al., 2016). VVA⁺ cells were considered to serve as a correlate for Pvalb neurons; a typical example of PV and VVA co-localization is shown in Figure 1. In the selected cortical region, essentially all PV⁺ cells are surrounded by a PNN identified as VVA⁺ cells. The one cell showing strong PNN labeling, but weak to none PV staining was mostly out of focus and cut very tangentially, i.e. containing a minimal part of the somatic region within the section, also evidenced by the low to absent DAPI staining.

PV⁺ and VVA⁺ cells were counted independently without crosschecking the other channel to ensure unbiased cell estimates. In the striatum of control mice, ∼90% of PV⁺ cells were also VVA⁺ (Figure 2A), whereas in the cortex, the selectivity was slightly lower with ∼75% of PV⁺ cells also being positive for VVA (Figures 3A and 4A). Within the pool of VVA⁺ cells, ∼70% and ∼65% of cells in control mice were also identified as PV⁺ in the striatum (Figure 2A) and the cortex (Figures 3A and 4A), respectively. Similar values were obtained in previous studies (Ye and Miao, 2013; Filice et al., 2016). The number of PV⁺ cells in the striatum of VPA mice was reduced by ∼15% (p = 0.0093) compared to controls (Figure 2A). However, there was no difference in the number of VVA⁺ cells between the two groups in the same region (Figure 2A), indicating that the number of Pvalb neurons was not decreased in VPA mice. We thus determined the number of double-labeled cells in the striatum. In saline-treated control mice, 71% of VVA⁺ cells were also PV⁺, whereas in VPA mice, this ratio was significantly decreased by ∼15% (p = 0.0448) (Figure 2A). This hinted that PV protein expression levels might be down-regulated in VPA mice. No difference was seen for PV-positive (PV⁺) cells that also stained positive for VVA in the three investigated regions (PV pool; Figures 2A, 3A, and 4A), suggesting that globally VVA⁺ PNNs were not affected by VPA exposure. Representative immunofluorescence images of the striatum are shown in Figure 2D. Although cell estimates obtained by the fractionator method are absolute, we decided to determine the volume of the analyzed ROIs to exclude major macroscopically discernable developmental abnormalities. The volume of the striatum, as measured by the Cavalieri estimator, was not different between VPA and control mice (Figure 2A). To certify that PV expression was decreased in VPA mice, PV levels were determined by qRT-PCR and Western blotting. In line with the stereological counts, striatal Pvalb mRNA levels were decreased by ∼50% (p = 0.0246) in VPA mice compared to control animals (Figure 2B). Likewise, PV protein expression levels in the striatum were decreased by ∼30% (p = 0.0218) in VPA mice, fully supporting the stereological findings (Figure 2C). GAPDH was used to normalize the PV signals on the Western blots (Figure 2C). Since all Pvalb neurons in the striatum are GABAergic and moreover express the potassium voltage-gated channel subfamily C member 1 (KCNC1/K_v3.1) (Chow et al., 1999) and subfamily S member 3 (KCNS3/K_v9.3) (Georgiev et al., 2012), we quantified transcript levels for glutamic acid decarboxylase 67 (Gad67), the major GABA-synthetizing enzyme in the brain, Kcnc1 and Kcns3 in VPA and control mice. For all 3 Pvalb neuron markers, transcript levels were not different in the striatum between VPA and control mice (Figures 5B1,D,F). Altogether, these results strongly indicated that VPA treatment resulted in decreased striatal PV expression. Reduced PV expression was also shown before in the striatum of PV^+/- mice and moreover, in the same region in Shank3B^-/- knockout mice (Filice et al., 2016).

Pvalb neurons in the rat and mouse VPA model have previously been investigated and animals were reported to exhibit a decreased number of PV⁺ cells, considered as the result of Pvalb neuronal loss in the neocortex and the colliculi superior (superficial and intermediate/deep layers), respectively (Gogolla et al., 2009; Dendrinos et al., 2011). To confirm these findings, but at the same time to investigate the alternate possibility that the decrease in PV⁺ neurons was the result of PV down-regulation, we examined two ASD-associated cortical structures, namely SSC and mPFC. Stereological analysis of both regions did not reveal altered numbers of Pvalb neurons in VPA compared to control mice, since neither the number of PV⁺ cells nor the one for VVA⁺ cells was different from control mice (Figures 3A and 4A). Representative immunofluorescence images of the SSC and mPFC are shown in Figures 3B and 4B, respectively. Of note, the volume of the mPFC was slightly increased by ∼20% (p = 0.0286) in VPA mice compared to control mice (Figure 4A). Indeed, brain overgrowth during infancy is a hallmark of ASD pathophysiology that has been observed in human cases (Courchesne et al., 2011), in VPA mice (Go et al., 2012) and several ASD mouse models including Shank3^-/- mice (Ellegood et al., 2015). Nonetheless, calculating the density of Pvalb cells per unit volume did not result in a significantly decreased number of Pvalb neurons in the mPFC of VPA mice. The volume of the SSC was similar between VPA and control mice (Figure 3A).

Consistent with IHC stereology results from mPFC and SSC (Figures 3 and 4) and global appearance of PV staining in the cortex (data not shown), Pvalb mRNA and protein levels from forebrain samples were similar between groups (Figure 5A1,A2). To avoid cross-contamination, the forebrain tissue mostly comprising neocortical tissue, but also including the subcortical structures thalamus and pallidus was not further dissected; thus, reduced spatial resolution is a limitation in the qRT-PCR and Western blot analyses. Next, we quantified Gad67, Kcnc1 (K_v3.1), and Kcns3 (K_v9.3) mRNA levels in the forebrain samples. Gad67 expression was not decreased in VPA mice; rather slightly increased, supporting the finding that there was no loss of GABAergic interneurons in VPA mice (Figure 5D). Moreover, transcript levels of Kcns3, which is selectively expressed in Pvalb neocortical neurons, were even slightly, but not significantly (p = 0.1095) increased in VPA compared to control mice (Figure 5F), in support of an unaltered (certainly not decreased) number of Pvalb neurons. The precise localization of K_v9.3 giving rise to a Kcns3 transcript signal in the striatum is currently unknown. Irrespective of its origin, Kcns3 was unaltered in the VPA mice (Figure 5F).

In contrast, forebrain mRNA levels of Kcnc1 coding for K_v3.1 were significantly decreased by ∼40% in VPA mice (p = 0.0114) (Figure 5B1). This decrease was also confirmed at the protein level: K_v3.1b protein levels determined by Western blot analysis were decreased to a similar extent, i.e., ∼40% reduction (p = 0.0002) of K_v3.1b in forebrain samples of VPA mice compared to controls (Figure 5B2), a finding in line with previous results from lysates of SSC of PND15-21 VPA mice (Iijima et al., 2016). We estimate that these results reflect decreased expression of K_v3.1 channels rather than loss of neurons expressing K_v3.1, since the overlap between cortical Pvalb neurons (unchanged in mPFC and SSC; Figures 3A and 4A) and K_v3.1b is >95% (Chow et al., 1999). Since results with respect to Pvalb neurons in VPA mice showed a high similarity with what we had observed before in Shank3B^-/- mice, i.e., a significant decrease in striatal PV levels (Filice et al., 2016) and moreover the recent findings that Shank3 mutations resulted in a decrease in hyperpolarization-activated cation (I_h) currents likely being the result of decreased HCN3 and HCN4 levels (Yi et al., 2016), we investigated transcript levels of the various hyperpolarization-activated cyclic nucleotide-gated (HCN) channels in VPA mice including transcript levels of Hcn1, Hcn2, and Hcn4, the predominant forms expressed in the brain. While Hcn2 and Hcn4 transcript levels were unchanged (Figures 5G,H); Hcn1, which does not co-localize with Pvalb neurons in the neocortex and the basal ganglia (Morris et al., 2004) was significantly up-regulated by ∼40% (p = 0.0484) in forebrain samples of VPA mice (Figure 5C1). The increase in HCN1 was also verified at the protein level; Western blotting showed a ∼40% increase (p = 0.0342) in VPA mice compared to controls (Figure 5C2). Interestingly, when we quantitatively determined transcript levels of Hcn1, Hcn2, and Hcn4 in the striatum, no differences were observed between VPA and control mice (Figures 5C1,G,H). Finally, we determined mRNA levels of Kcnc2 (K_v3.2) to see whether K_v3.2 quantitatively compensates for the K_v3.1 deficit. K_v3.2 is a close relative of K_v3.1, but co-localizes to a lesser extend with Pvalb neurons (Chow et al., 1999). There was no difference in forebrain mRNA levels of K_v3.2 between VPA and control mice (Figure 5E). This was also true for the striatum (Figure 5E). In summary, our results indicate that VPA mice do neither exhibit Pvalb neuron loss nor a decrease in PV expression levels in the forebrain; instead, they display reduced levels of K_v3.1b in combination with augmented expression of HCN1. The functional implications of these alterations in forebrain regions including the neocortex and the selective down-regulation of PV in the striatum, possibly linked to homeostatic plasticity, are discussed below.