Pharmacological Characterization of Cultivated Neuronal Networks: Relevance to Synaptogenesis and Synaptic Connectivity

Preparation of Primary Hippocampal Cultures

Hippocampi were dissected (Kaech and Banker 2006) from wild-type E18 FVB mouse embryos in HEPES (7 mM) buffered Hanks balanced salt solution (HBSS), followed by trypsin digestion (0.05 %; 10 min; 37 °C) and mechanical dissociation by trituration through 2 fire-polished glass pipettes with decreasing diameter. After centrifugation (5 min at 200×g), the cell pellet was resuspended in minimal essential medium (MEM) supplemented with 10 % heat-inactivated normal horse serum (Innovative Research, MI, USA) and 30 mM glucose (MEM-horse medium). Cells were plated in poly-d-lysin-coated 96-well plates (Greiner Cell coat, µClear, Wemmel, Belgium), at 30,000 cells/cm², and kept in an humidified CO₂ incubator (37 °C; 5 % CO₂). After 4 h, the medium was replaced with B27 supplemented Neurobasal medium, containing sodium pyruvate (1 mM), glutamax (2 mM), and glucose (30 mM). To suppress proliferation of non-neuronal cells, arabinosylcytosine (AraC; 2 µM; Sigma-Aldrich, Bornem, Belgium) was added at the fourth day after plating. All cell culture supplies were purchased from Life Technologies (Merelbeke, Belgium), unless stated otherwise.

Preparation of Primary Astrocyte Cultures

Cerebral cortex from wild-type E18 FVB mouse embryos was dissected (Booher and Sensenbrenner 1972) and cells were further processed as described for the hippocampal cultures. After plating, the cells were kept in MEM-horse medium for 12 days, until the astrocytes formed a nearly confluent monolayer, with minimal neuronal presence. 2 days before splitting the astrocytes, the medium was replaced by B27 supplemented Neurobasal medium, allowing it to be conditioned for 48 h. Afterward, the conditioned medium was collected, the astrocytes were trypsinized (0.05 %; 10 min; 37 °C) and seeded in 96-well plates at different plating densities. After 24 h of recovery in MEM-horse medium, hippocampal cultures were plated on top of the pre-plated astrocytes. Conditioned medium was diluted 1:2 in fresh B27 supplemented Neurobasal medium, and used after 4 h in MEM-horse medium. Cultures grew in the absence of AraC.

Characterization of Distinct Cell Types in Primary Neuronal Cultures

At different days in vitro (DIV), cultures were fixed [4 % paraformaldehyde in 0.1 M phosphate buffer, 10 min at room temperature (RT)] and immunostained for a neuronal marker, HuC/HuD neuronal protein, and an astrocyte marker, glial fibrillary acidic protein (GFAP), followed by DAPI staining to obtain the total number of nuclei. Briefly, after pre-incubation with blocking buffer containing 1 % Triton X-100 for 10 min, the primary antibodies were applied for 4 h at RT (mouse monoclonal against HuC/HuD neuronal protein, 1/1,000, Life Technologies; rabbit polyclonal against GFAP, 1/400, Dako, Heverlee, Belgium). After washing with PBS, secondary antibodies were added for 2 h (Cy3-conjugated donkey-anti-mouse and FITC-conjugated goat-anti-rabbit F_ab fragments, Jackson Immunoresearch, Suffolk, UK). Finally, DAPI was applied to the cultures for 2 min at a concentration of 2.5 µg/ml, followed by a PBS wash.

Stained cultures were automatically imaged for DAPI, HuC/HuD and GFAP in a BD pathway 435 bioimager (20× magnification NA 0.75, Becton–Dickinson, Erembodegem, Belgium). An overview of the well was created using 4 × 4 tiles with an inter-frame gap of 500 µm. Images were subsequently exported to Cell P software (Olympus, Münster, Germany) and nuclei, neurons and astrocytes were manually counted on 4 × 4 tiles, using the touch-count function. Results were expressed as # cells/cm².

Live Cell Calcium Imaging

Cell cultures were loaded with 2 µM Fluo-4-AM in Neurobasal B27 medium at 37 °C and 5 %. After 30 min, the medium was replaced with recording medium, containing (in mM): CaCl₂ 0.9; MgCl₂ 0.5; KCl 2.67; NaCl 138; KH₂PO₄ 1.47; Na₂HPO₄·7H₂O 8; glucose 10. To study the sensitized NMDA-R, the same recording medium was used with exclusion of MgCl₂ and addition of 10 µM glycine (Sigma-Aldrich). Cells were imaged on an inverted dual spinning disk confocal microscope (UltraVIEW ERS, PerkinElmer, Seer Green, UK), for 5 or 8 min, with a 25× objective lens (NA 0.80). To distinguish neurons from non-neuronal cells, 30 µM glutamate was added during the last minute of calcium recording (Pickering et al. 2008). Immediately after finishing the calcium recording, DAPI (2.5 µg/ml DAPI, 1 % TritonX-100, 4 % paraformaldehyde) was added to obtain an image of the nuclei in the recorded field.

Automated Analysis of Calcium Recordings

The resulting calcium recordings were automatically analyzed with a custom-made analysis algorithm (Cornelissen et al. 2013), that provides multiple parameters that reflect the activity of the neurons. Briefly, the neurons were recognized based on a nucleus image and the response to 30 µM glutamate. For each region of interest, traces of the fluorescence intensity over time were generated and used for subsequent signal analysis. The average bursting frequency and amplitude were measured from the normalized (F/F ₀) traces of individual neurons, as was the average decay time, which is the time needed to go from a peak amplitude to 50 % of the baseline fluorescence. Neurons with at least 1 peak during the 4 min recording were counted and expressed as the percentage of active neurons. Whole fluorescence signals of individual neurons were compared and an average correlation score was calculated. Neurons that showed similar calcium patterns (coincident peaks and valleys) returned a high correlation score (Pearson correlation; between −1 and 1). To measure the synchronous bursting frequency and amplitude, all neuronal traces were combined into 1 average trace depicting only the correlated network bursts, while uncorrelated peaks were averaged out. For acute experiments, calcium recordings were analyzed in 2 or 3 separate stretches, corresponding to different treatments.

Dendritic Spine Density

Cultures were fixed (4 % paraformaldehyde in 0.1 M phosphate buffer, 10 min at RT) and stained with the carbocyanine dye CM-DiI (1 µg/ml; 20 min; Life Technologies). At least 24 h passed between staining and imaging, allowing the hydrophobic dye to spread throughout the plasma membrane. High-resolution confocal images were acquired using a dual spinning disk microscope (UltraVIEW VoX, PerkinElmer, Seer Green, UK) with a 63× oil objective lens (NA 1.4), followed by image deconvolution (resolution ≈ 0.25 µm). Dendritic spine density was calculated by manually counting the dendritic protrusions and correcting for the length of the dendrite under investigation (Volocity image analysis software, PerkinElmer, Seer Green, UK). At least 1,500 µm of dendrite length was considered for each condition, on multiple non-overlapping images.

Neurite Outgrowth and Synapse Density

After fixation, cultures were immunostained for the neurite marker β-III-tubulin (rabbit polyclonal against β-III-tubulin, 1/1,000, Covance, Brussels, Belgium) and the synaptic protein synaptophysin-I (guinea pig polyclonal against synaptophysin-I, 1/1,000, Synaptic Systems, Göttingen, Germany). FITC-conjugated goat-anti-rabbit Fab fragments and Cy3-conjugated donkey-anti-guinea-pig (Jackson immunoresearch) secondary antibodies were used. Images of the stained cultures were automatically acquired using a BD pathway 435 bioimager (4 × 4 tiles with 500 µm inter-frame gap, 20× objective) and exported to Cell P software.

To measure neurite outgrowth, images were first processed with a differential contrast enhancement filter, followed by automatic thresholding and finally the surface of the β-III-tubulin-positive structures was measured. To adequately measure the number of synaptophysin-I-positive puncta, images were first treated with a Laplace filter, followed by automated thresholding and counting of the puncta. Synapse density was expressed as the number of synaptophysin-I puncta per µm² β-III-tubulin surface.

Culture Treatment and Pharmacology

To study acute effects on neuronal network activity, tetrodotoxin (TTX; 2 µM), forskolin (FSK; 10 nM–10 µM), MK-801 (1–100 µM; in-house J&J Pharmacy), ketamine (1–100 µM; in-house J&J Pharmacy), KA (10–150 µM, QA (10–150 µM), acetylcholine (10 µM), muscarine (10 µM), nicotine (1 µM), or serotonin (10 µM) was dissolved in the recording medium and added to distinct wells, 3 min after initiation of baseline calcium recording. Effects were further followed for 4 min before the glutamate stimulus was given. All compounds were purchased from Sigma-Aldrich, unless stated otherwise. For successive experiments, 3 min of baseline recording were followed by 3 min in the presence of ketamine (50 µM), KA (100 µM) or QA (100 µM), in turn followed by another 3 min of ketamine (50 µM), KA (100 µM) or QA (100 µM) without washing the previous treatment away. Acute effects were studied on 6 and 7 DIV cultures.

For NGF deprivation experiments, cultures were incubated from the fourth day in vitro with neutralizing mouse monoclonal antibodies against NGF (3–9 µg/ml; Peprotech, London, UK). A nonsense mouse monoclonal antibody (anti-BAX; 9 µg/ml; Abcam, Cambridge, UK) was included as negative control. Neither NGF nor nonsense antibodies contained preservatives.

RT qPCR for NMDA-R Subunits

Hippocampal cultures were grown under control conditions until the time of cell lysis at 4, 7 or 10 DIV. Cells were lysed in 100 µl/well RLT buffer (Qiagen, Germantown, MD, USA), supplemented with 1 % β-mercaptoethanol. RNA was extracted using a RNeasy mini kit (Qiagen), combined with DNase treatment (RNase-free DNase set, Qiagen) using prescribed procedures. RNA concentration and integrity were assessed with the nanodrop spectrophotometer (Thermo scientific, Wilmington, DE, USA) and the BioAnalyzer (Agilent Technologies, Diegem, Belgium), typically yielding highly intact RNA (RIN ≥ 9) with a concentration of about 60 ng/µl. cDNA was synthetized starting from 150 ng RNA, using a SuperScript III First-Strand Synthesis SuperMix (Life Technologies, Merelbeke, Belgium), yielding 20 µl cDNA of which 5 µl (1:10) was used in the subsequent qPCR reaction. Primers and probes were purchased from Tib molbiol (Berlin, Germany) for 5 target genes: Grin1, Grin2A, Grin2B, Grin2D, and Grin3A (for primer sequences, see Table 1). qPCR was conducted in triplo in a LightCycler 480 instrument, using LightCycler 480 Probes Master and LightCycler 96-well plates (95 °C for 10′; 40 cycles of 95 °C for 10″/60 °C for 15″/72 °C for 1″). Primers were used at 0.5 µM and probes at 0.1 µM. The efficiency (E) of each primer/probe mix was measured by the determination of the slope of a standard curve on a serial dilution of initial cDNA.

In total, 5 reference genes were included in the PCR (PSMD1, eIF3, GAPDH, RPL38, eEF2), of which a combination of PSMD1 (Taqman gene expression assay Mm01295362_m1; Life Technologies), eIF3 (for sequence, see Table 1) and GAPDH (Taqman gene expression assay Mm99999915_g1; Life Technologies) was used to normalize the experimental data, based upon analysis by geNorm software (M value <0.5; qBasePlus, Biogazelle, Zwijnaarde, Belgium). The PCR products were run on a 2 % agarose gel to confirm the right amplicon length.

Cycle number was measured using the baseline-independent second derivative maximum method. Normalized relative gene expression was determined by the $E^{-\mathit{\Delta }\mathit{\Delta }{C}_{\text{T}}}$ method and was rescaled to the expression level at 4 DIV.

Statistics

The number of biological and technical replicates of each experiment is listed in supplemental Table 1, and is mentioned in the respective figure legends. A biological replicate refers to 1 dissection and plating procedure. A technical replicate refers to 1 well of a 96-well plate from within a biological replicate.

For statistical analysis of normally distributed data from calcium imaging (Fig. 1f), a linear mixed effects model was used to model the data of each analysis parameter. The nested structure of the data was accounted for using a mixed effects analysis to cope with the heterogeneity between and within biological replicates: several technical replicates were measured within one biological replicate (mouse) and there were repeated measures of the mice on the different DIV. In order to accommodate the dependency of the measurements within the biological replicate and the repeated measurements, a random intercept for mouse has been included in the model. The DIV entered the model as fixed effect term (as a categorical covariate) to allow for modeling flexible patterns over time (instead of imposing a linear trend). The model also accounted for the unequal variances over time. For synchronous bursting frequency and correlation score, data were log-transformed prior to analysis. For synchronous bursting frequency, a value of 0.25 (corresponding to the detection limit) was assigned in case of no occurrence. Post hoc t tests were conducted from the model to detect significant changes between consecutive days. For visual representation, means and standard deviations from the raw data were plotted.

For morphological parameters (Fig. 2b, d), each analysis parameter was modeled using an ANOVA model containing time of maturation of the cells as covariate. Only two biological replicates were measured per DIV. Therefore, the nested structure of the data was not accounted for. The presence of significant differences over time was tested using an overall F-test for each analysis parameter. Conditional on the overall significance of the time effect, post hoc t tests were conducted from the model to detect significant changes between consecutive days. For visual representation, means and standard deviations were plotted from the original data.

For astrocyte conditioning and NGF deprivation experiments, normally distributed calcium imaging parameters were analyzed using an ANOVA model with DIV and treatment as fixed effects (Figs. 3a, 4a). No significant interactions were found between DIV and treatment for any of the calcium imaging parameters. The nesting was not accounted for due to the limited number of biological replicates (n = 2). First, an overall F test was performed to assess any differences over DIV and treatment. Conditional to the overall F tests, post hoc tests were carried out within DIV; for the astrocyte conditioning experiments, post hoc Dunnett’s t tests were carried out for each analysis parameter (equal variances assumed), while Dunnett’s T3 tests were used in the NGF deprivation experiment (equal variances not assumed). For amplitude and synchronous bursting frequency, a log-normal distribution was applied. For synchronous bursting frequency, a value of 0.25 (corresponding to the detection limit) was assigned in case of no occurrence. Morphological parameters (Figs. 3c, 4b) were analyzed using a one-way ANOVA. Conditional to the overall F test, a post hoc Dunnett’s T3 test was carried out to detect differences in the spine density in the astrocyte conditioning experiment. Again, the nesting was not accounted for due to the limited number of biological replicates (n = 2).

To detect the influence of the recording medium on spontaneous network activity (Fig. 6b, f), an unpaired t test was performed. For dose–response experiments of ketamine, KA and QA (Fig. 6c–e, g–i), data on synchronous bursting frequency were log-transformed prior to analysis to cope with the log-normal nature of the data. For synchronous frequency, a value of 0.25 (corresponding to the detection limit) was used in case of no occurrence. No transformation was performed on the correlation parameter. For each compound, a linear model was estimated where the relevant parameter (synchronous bursting frequency or correlation) is modeled as a function of concentration (categorical covariate), recording medium (categorical covariate) and the interaction of concentration and recording medium. Contrasts were estimated to compare normal and sensitizing medium at each individual concentration and (back-transformed) least-squares means were generated (online document: Lenth 2013) for convenient display of the data (LS mean ± 95 % CI) and were investigated for significant changes from before treatment (the first 3 min of recording in the absence of the compound). The p values associated to the estimated contrasts (differences at each concentration level) are used to designate significance on the plot.

Neuronal Networks Develop Synchronized Activity Within a Well-Defined Time-Window

Since the dissection procedure of hippocampi from mouse embryos is crucial for the reproducibility of the cultures and the density of the neurons in vitro is decisive for their connectivity and spontaneous activity (Ivenshitz and Segal 2010), we assessed the proportion of the most prominent cell types in the cultures over time. Nuclei, neurons, and astrocytes were quantified at specific timepoints from 1 to 21 days in vitro (DIV; Fig. 1a, b). In addition to the neurons (HuC/HuD neuronal protein-positive) and astrocytes [glial fibrillary acidic protein-positive (GFAP+)], the number of cells immunonegative for HuC/HuD and GFAP were classified as “other” and include fibroblasts and a small number of pre-oligodendrocytes (Neural/Glial Antigen 2-positive/2′,3′-cyclic nucleotide 3′-phosphodiesterase-negative/myelin basic protein-negative; data not shown). At these timepoints, no microglia could be identified through immunostaining for CD11b, ionized calcium binding adaptor molecule 1, purinergic receptor P2X₄ or Major Histocompatibility Complex class II, both on cultures and on cytospin preparations of culture medium, leading us to conclude that microglia were absent in these cultures.

The total number of nuclei increased during the first week in vitro, followed by a decrease after 2 weeks. The number of neurons, which were post-mitotic at the time of isolation, showed a slight decrease over 3 weeks in vitro. However, the neuronal population was relatively stable within the experimental window (4–14 DIV). The initial rise in nuclei represented the proliferation of astrocytes and other cells. At 4 DIV, arabinosylcytosine (AraC) was added to the medium inhibiting the proliferation of dividing cells. Other non-neuronal cells showed a higher sensitivity to AraC than astrocytes, the proliferation of which ceased from 7 DIV onwards. However, this AraC concentration decreased astrocyte proliferation markedly between 4 and 7 DIV compared to cultures that were not treated with AraC (data not shown).

Spontaneous calcium peaks could be observed in the neuronal cultures after loading with Fluo-4 (Fig. 1c–e). Both neurons and non-neuronal cells showed calcium fluxes, with different kinetics. Neurons exhibited relatively short (about 5 s) and steep calcium peaks, while the other cells showed slower waves (about 30 s), which often spread to neighboring cells. In order to distinguish neurons from non-neuronal cells, 30 µM of glutamate was added to the well during the last minute of each calcium recording. The applied glutamate concentration was neurotoxic, and resulted in a sustained rise of calcium in the neurons only (Pickering et al. 2008) (Fig. 1e, arrows). Non-neuronal cells responded with a slow and transient calcium wave (Fig. 1e, arrowheads), and were discarded from further analysis. The spontaneous neuronal activity evolved from asynchronous at 5 DIV (Fig. 1d) to synchronized activity at 7 DIV (Fig. 1e), reminiscent of the maturation of a neuronal network.

Calcium recordings were analyzed with a custom-made algorithm (Cornelissen et al. 2013), extracting different parameters, as shown in Fig. 1f. Synchronization of the activity was evident from 2 parameters: the correlation score, which expresses the amount of synchrony between neurons in a given calcium recording, and the frequency of synchronous bursts. At 5 DIV, the first synchronous bursts arose, but the activity was still largely asynchronous. 2 days later, at 7 DIV, the activity became mostly synchronous, as reflected in the correlation score and synchronous bursting frequency. The formation of the network coincided with a significant increase in the fraction of active neurons. At 3 and 4 DIV, before the onset of synchronous bursting, only 45 % of the neurons were active during a 4-min recording, while more than 90 % of the neurons showed activity at 7 DIV, demonstrating that the growing network incorporates and activates new neurons, since the total number of neurons was nearly constant between 4 and 7 DIV (Fig. 1b). The frequency of individual bursts remained rather constant over the different DIV while the proportion of bursts that occurred in synchrony increased. The amplitude of the calcium bursts of individual neurons showed an upward trend, whereas the decay time decreased, i.e., by day 7, more calcium entered the neuron compared to day 4, but at the same time, the neuron was better equipped to buffer the excess intracellular calcium. Though Fluo-4 is a calcium indicator of the non-ratiometric type, we found a linear relationship between fluorescence intensity and intracellular calcium concentration in the dynamic range that was typical for our experiments using the calcium ionophore ionomycin (data not shown).

Network Formation Coincides with the Outgrowth of Neurites and the Formation of Synapses, While Dendritic Spines Appear only After Synchronization of the Activity

Neurite outgrowth started after the first contact of the neuron with the bottom of the 96-well plate. At the fourth day after plating, the neurons already contained some presynaptic synaptophysin-I-immunoreactive puncta (Fig. 2b), which are thought to form functional synapses after contacting a postsynaptic partner. Synaptophysin is present on synaptic vesicle membranes and is believed to be involved in regulating vesicle endocytosis (Gordon et al. 2011).

A striking increase in neurite outgrowth was observed between 4 and 7 DIV (Fig. 2b). This expansion slowed down between 7 and 11 DIV after which the neurite surface remained stable, as the bottom of the plate was largely covered by neurites. The number of synaptophysin-I-positive puncta, on the other hand, kept on increasing until 14 DIV. A significant increase in synapse density, expressed as the number of puncta per surface unit (µm) neurite, was seen from 4 to 14 DIV (Fig. 2a, b). As such, the synchronization of neuronal activity (5–7 DIV, Fig. 1d–f) coincided with an important increase in neurite outgrowth and synapse density. Although synaptogenesis continued afterward, it can be assumed that the critical number of functional synapses to obtain synchronous network activity was reached by 7 DIV.

Dendritic spines were visualized via labeling with the lipophilic dye DiI and subsequent confocal microscopy (Fig. 2c). The number of dendritic spines per µm of dendrite showed a marked increase over time, reflecting the maturation process of the neurons (Fig. 2d). While synchronization of the bursting occurred between 5 and 7 DIV, spine density started to increase only at 7 DIV and was most pronounced between 11 and 21 DIV. These data suggest that robust network activity aids in the formation of dendritic spines, which offer electrical and biochemical compartmentalization of intensively used synapses.

Chronic and Acute Modulation of Neuronal Network Activity

The presented in vitro model holds great potential to facilitate fundamental and drug discovery-related research on 2 levels. The first one is the neurodevelopmental level, which is likely to be disturbed in mental diseases such as schizophrenia, autism spectrum disorders, intellectual disability, etc. To validate the proposed setup as a tool for studying neurodevelopment, it is essential to show that the formation of the neuronal network is responsive to chronic modulatory treatments. The second level is synaptic communication in the mature network, which is compromised in diseases like Alzheimer’s disease, schizophrenia, mood disorders, etc. Therefore, we explored if mature neuronal network activity can acutely be influenced by known neuromodulatory compounds.