Screening for modulators of neural network activity in 3D human iPSC-derived cortical spheroids

Human induced Pluripotent Stem Cells (iPSCs) are a powerful tool to dissect the biology of complex human cell types such as those of the central nervous system (CNS). However, robust, high-throughput platforms for reliably measuring activity in human iPSC-derived neuronal cultures are lacking. Here, we assessed 3D cultures of cortical neurons and astrocytes displaying spontaneous, rhythmic, and highly synchronized neural activity that can be visualized as calcium oscillations on standard high-throughput fluorescent readers as a platform for CNS-based discovery efforts. Spontaneous activity and spheroid structure were highly consistent from well-to-well, reference compounds such as TTX, 4-AP, AP5, and NBQX, had expected effects on neural spontaneous activity, demonstrating the presence of functionally integrated neuronal circuitry. Neurospheroid biology was challenged by screening the LOPAC®1280 library, a collection of 1280 pharmacologically active small molecules. The primary screen identified 111 compounds (8.7%) that modulated neural network activity across a wide range of neural and cellular processes and 16 of 17 compounds chosen for follow-up confirmed the primary screen results. Together, these data demonstrate the suitability and utility of human iPSC-derived neurospheroids as a screening platform for CNS-based drug discovery.


Introduction
Modern conventional high-throughput drug screening typically uses recombinant cell lines that overexpress a drug target of interest. The advent of human induced pluripotent stem cells (iPSCs) promises, among other things, the development of relevant cellular disease models for use in high-throughput screening. Human iPSCs offer many advantages over recombinant cell lines or primary rodent cells for use in drug screening. Because these cells are derived from human donors, human genetic diseases can be more accurately modeled, especially when used in combination with modern genome editing techniques. Furthermore, the ability of iPSCs to be differentiated into a variety of cell types further expands their utility, as screens can be without any known deleterious mutations and a normal healthy phenotype. As described previously [16], the microBrain 3D platform was generated, differentiated, and matured for 8 weeks prior to shipment. Over the time of differentiation and maturity, the spheroids spontaneously develop synchronized calcium activity. Upon receipt, the 3D co-differentiated population of cortical neurons and astrocytes [16] was maintained in and replenished every two days for one week prior to assay with BrainPhys (StemCell Technologies, 05790) media supplemented with 1x SM1 (StemCell Technologies, 05711), 20ng/mL of BDNF (StemCell Technologies, 78005), 20ng/mL of GDNF (StemCell Technologies, 78058), and 1x pen/strep (Corning, 30-002-CI).

Immunofluorescence
3D neurospheres were fixed in 4% paraformaldehyde solution in PBS for 10 minutes at room temperature. Next, neurospheres were washed 5x with half volume changes of PBS. Neurospheres were permeabilized in 0.4% Triton-X in Odyssey Blocking Buffer (LiCor) for 15 minutes at room temperature followed by addition of blocking solution (Odyssey Blocking Buffer + 0.1% Triton-X) and incubated for 4 hours at room temperature. After blocking, primary antibodies (MAP2 from Millipore, MAB3418; GFAP from Abcam, ab134436) were added at 1:250 dilution in blocking solution and incubated overnight at 4˚C with gentle agitation. The next day, neurospheres were washed 8x with half washes of PBS + 0.1% Tween-20 with 2 minutes for each wash. Secondary antibodies were then added in blocking solution and incubated for 1 hour at room temperature, protected from light. After the secondary antibody incubation, neurospheres were washed 4x with half washes of PBS + 0.1% Tween-20. The nuclei were stained using DAPI for 8 minutes at room temperature followed by 8 half washes with PBS. Neurospheres were imaged in PBS using an ImageXpress Confocal Microscope (Molecular Devices) at 20x magnification.

Screening assay
Three half-washes were performed on 3D neurosphere plates with phenol red-free BrainPhys Medium (StemCell Technologies, 05791) supplemented with 1x SM1, 20ng/mL of BDNF, 20ng/mL of GDNF, and 1x pen/strep. Next, Calcium 6 Dye (Molecular Devices, R8190) was added to a concentration of 1x and incubated for 2 hours at 37˚C in 5% CO 2 in preparation for the baseline reading. The baseline signal was read for 10 minutes on a FLIPR Tetra 1 (Molecular Devices) with a capture rate of 2 Hz. Compounds were diluted in the above medium. After the baseline reading, compounds were added to the neurospheres and activity was measured after 30 minutes, 2 hours, and 4 hours for 10 minutes with a capture rate of 2 Hz.

Compound library
The LOPAC 11280 library was purchased from Millipore-Sigma, and all compounds were solubilized in DMSO. In the primary screen, each compound was tested in duplicate (on separate plates) at 1 μM concentration. For concentration-response experiments, fresh powders were ordered and tested in ten-point half-log concentration intervals. For concentration response curves, each dilution was tested in triplicate.

Data analysis
FLIPR Tetra 1 data were analyzed using ScreenWorks PeakPro Software (Molecular Devices). Spatial uniformity correction was enabled, and peak detection was utilized with the following parameters: smooth width = 5, fit width = 10, and dynamic amplitude threshold = 20. Each well was normalized to its baseline reading before compound addition. Data were graphed using GraphPad Prism version 7. In the primary screen, compounds were counted as hits only if both replicates were greater or less than 2 times the standard deviation of the DMSO treated wells. Significant differences between vehicle and compound treated wells were determined by one-way ANOVA followed by Dunnett's test. GraphPad Prism version 7 was used to perform all statistical analyses.

3D neurospheres display synchronized calcium oscillations
Two-dimensional (2D) cultures and 3D spheroids of human iPSC-derived cortical neurons and astrocytes were obtained from StemoniX, Inc. The presence of astrocytes and neurons were confirmed by positive staining with MAP2, a protein highly expressed in neurons, and GFAP, a protein expressed by astrocytes. We confirmed the distribution of neural cells on the surface ( Fig 1A) and interior of the neurospheres ( Fig 1B). As reported before for cultures of primary rodent neurons [12,13], we found that these neurospheres display spontaneous activity that can be visualized with a calcium dye and that activity increased with culture age over time in culture (S1 Fig). Initial experiments utilized fluorescence microscopy for imaging calcium activity (Fig 1B), however the platform was quickly switched to a 384-well fluorometric imaging plate reader (FLIPR Tetra 1 ) as this platform is more amenable to measuring compound-induced effects on the calcium waveform properties (such as oscillation frequency, amplitude, peak rise time, peak decay time) in a high-throughput fashion ( Fig 1C). One benefit of the neurospheres used in this study is their mixed composition of cortical neurons and astrocytes, making them more physiologically relevant than a pure neuronal culture. As functional activity could arise from either or both cell populations, we sought to confirm the neuronal contribution by limiting the calcium indicator presence to only neurons. To this end, we transduced neurospheres with a calcium indicator driven by the neuronal-specific synapsin promoter (S2 Fig). Spontaneous activity was detected by the synapsin-driven Ca 2+ indicator, and the frequency pattern was similar to that of the ubuiquitous Ca 2+ indicator used in the FLIPR recordings, thus confirming that neuronal activity contributes to at least a portion of the whole spheroid signal. The ability to isolate activity to a single cell population is important in that it could also be used in follow-up screens to elucidate cellular and molecular mechanisms of action triggered by unknown compounds. We next compared the spontaneous activity generated by cells grown in 2D cultures to that of the 3D cultures. Activity heat maps for peak count in 384-well plates of 2D and 3D cultures are shown in Fig 1C. In the 2D plate, 149 out of 384 wells (39%) did not show any detectable calcium oscillations (dark green squares) while the 3D plate, in comparison, showed detectable activity in 100% of the wells.
Well to well consistency in oscillatory activity was also more reproducible in the 3D plate, with peak count ranging from 18 to 49 peaks over ten minutes (34.5 ± 5.5, mean ± s.d.) in the 3D plates versus 0 to 69 peaks over ten minutes (18.5 ± 18.7, mean ± s.d.) in the 2D plates ( Fig  1D). Similarly, the consistency and magnitude of peak amplitude measured in relative fluorescence units (RFU) was greater in the 3D cultures (332.1 ± 59.0, mean ± s.d.) compared to 2D cultures (116.4 ± 74.7, mean ± s.d.; Fig 1E). Thus, the spontaneous activity generated by 3D cultures was more homogenous and robust from well-to-well than that of the 2D cultures, making the 3D platform more suitable for screening.

3D neurospheres respond to excitatory and inhibitory tool compounds
We next tested whether the spontaneous activity from 3D neurospheres could be modulated by reference compounds that activate or inhibit various ion channels and neurotransmitter receptors. Treating neurospheres with 1 μM tetrodotoxin (TTX), a potent voltage-gated sodium channel inhibitor, completely blocked spontaneous activity (Fig 2B, 2H and 2I). Treatment with 1 μM AP5, a selective NMDA receptor antagonist, decreased both the peak frequency and amplitude (Fig 2H and 2I) but did not completely block activity ( Fig 2C). Treatment with 1 μM 4-aminopyridine (4AP), a K + channel inhibitor, significantly increased peak frequency (Fig 2D and 2H) and decreased peak amplitude ( Fig 2I). Treatment with 1 μM cyclothiazide, a positive allosteric modulator of AMPA receptors, induced seizure-like activity ( Fig 2E) characterized by bursts of rapid and irregular activity with decreased peak amplitude and minimal changes in overall peak frequency (Fig 2H and 2I). Treatment with 1 μM NBQX, an AMPA receptor antagonist, significantly reduced both peak frequency and peak amplitude (Fig 2H and 2I) but did not completely block activity ( Fig 2F). Finally, treatment with 1 μM muscimol, a GABA A receptor agonist, completely blocked spontaneous activity (Fig 2G, 2H and 2I). Taken together, these findings demonstrate that excitatory activity is mediated by both NMDA and AMPA receptors, as antagonists of either class of receptors modulated but did not completely block spontaneous activity. Since the GABA A receptor agonist muscimol blocked all neuronal activity, this suggests that GABA receptors are present and can mediate inhibitory activity.

Screen with 1280 pharmacologically active compounds
To determine the potential utility of this platform for screening purposes, we performed a small screen with 1280 active compounds (LOPAC 11280 ) tested in duplicate at 1 μM concentration. Neurospheroid activity was measured via calcium fluctuations at 10-minute intervals before compound addition and at 30 minutes, 2 hours, and 4 hours following compound addition. We aimed to identify compounds that could increase or decrease activity and thus selected hits based on both replicates being greater than or less than two-times the standard deviation of the DMSO treated wells. We analyzed two parameters, peak frequency and peak amplitude, across vehicle controls and LOPAC 11280 compounds. Low variability for both parameters was observed in vehicle control (DMSO) wells within and across all 384-well plates for all timepoints (Table 1).
We defined a compound as hit when it changed neural activity (either peak frequency or amplitude) by twice the respective standard deviation of the negative control values; corresponding to 30% (peak frequency) and 40% (peak amplitude) changes. Scatter plots for each time point and peak frequency/amplitude are shown in Fig 3. In total, 111 compounds significantly affected peak frequency or amplitude on at least one timepoint with 17 of the 111 hits (15%) affecting both peak frequency and amplitude at all timepoints measured (Table 2) suggesting that many compounds presented a transient neural effect, while others have a longer lasting modulatory effect. In addition, 40 compounds (36%) only changed frequency or amplitude at 1 timepoint. Most of the hits, 100 out of 111 (90%), decreased peak frequency. Of the 11 compounds that increased peak frequency, only two (amitriptyline hydrochloride and forskolin) increased peak frequency at all three time-points (Table 2). Overall the hits spanned many different cellular processes (Fig 4). Most of the hits (58%) impacted neurotransmissionrelated pathways, while the rest affected diverse processes including (but not limited to) ion channels, cell biology/biochemistry, kinase signaling, as well as immune and hormone modulators (Fig 4). The diversity of hits, as well as the reproducibility of the vehicle controls, in this small-scale screen clearly validates the potential of this platform for identifying modulators of neuronal activity.

Hit confirmation and concentration response with select compounds
We selected 17 compounds for further confirmation based on their ability to alter peak frequency and amplitude at the majority of tested timepoints. Of the 17 compounds tested in concentration response, three are adenosine receptor agonists, four are antagonists of NMDA or AMPA receptors, and three are positive allosteric modulators of GABA receptors. The remaining seven compounds act on various ion channels and other cellular processes (Table 3). Concentration response curves for peak frequency and amplitude for each     compound at the 30-minute timepoint are displayed in Figs 5 and 6. EC50s for each compound at each timepoint are listed in Table 3. Of the 17 compounds tested, 16 were confirmed active in concentration response assays. Notably, the adenosine receptor agonists had effects in the low nanomolar range. The only compound that was not confirmed was amitriptyline hydrochloride, which is a tricyclic antidepressant. This compound increased peak frequencỹ 1.5 fold and decreased peak amplitude~0.5 fold in the primary screen at 1 μM (Table 2), but when tested in a concentration-response, it did not increase frequency and blocked all spontaneous activity at the highest concentrations (Fig 5E). Forskolin was the only compound confirmed to increase peak frequency with an EC 50 of 45 nM at 30 minutes (Table 3). Forskolin also decreases peak amplitude at 30 minutes with an EC 50 of 45 nM at 30 minutes (Table 3).

Discussion
Development of a high-throughput, functional readout for neuronal network activity has been a challenge over the years, and a few methods of assessing neuronal function in a highthroughput format have recently been described. The MANTRA™ system developed by the Galenea Corporation uses a pH-sensitive synaptic reporter to monitor synaptic vesicle cycling as a readout for neuronal activity [17]. Virdee et al. reported on a similar technology using a Values are fold-change from DMSO treated wells and indicate that frequency and/or amplitude was greater or less than two-times the standard deviation from DMSO.
No value corresponds to no change from DMSO treated wells. https://doi.org/10.1371/journal.pone.0240991.t002 calcium sensitive dye to measure responses in networks of cultured rat neurons in vitro [18]. Both systems measure responses evoked by electrical field stimulation in 2 dimensional primary rodent neurons. In this study we extend these investigations to assess human-based neural activity using iPSC-derived 3D neurospheroids with the goal of developing a higherthroughput functional screening platform. Similar to previous work we show that functional  activity can be monitored via a calcium sensitive dye [18]. Further, we provide several lines of evidence supporting a neuronal contribution to the whole spheroid Ca 2+ signal including; 1) using a synapsin promotor-driven Ca 2+ fluorophore to demonstrate temporal correlation between neuronal specific and whole spheroid Ca 2+ signals, 2) demonstrating neuronal connectivity by showing synchronization of the neuronal-specific Ca 2+ signals across discrete areas of the neurospheroid, and lastly 3) modulating the whole neurospheroid Ca 2+ signal with neuronal specific ion channel and neurotransmitter agonists and antagonists. While not ruling out other sources, e.g. astrocytes, these data and that of others [19] provides strong evidence for a neuronal contribution to the whole neurospheroid Ca 2+ signal. Synchronous bursting of neurons in culture is believed to be a result of spontaneous miniature synaptic conductances in combination with random depolarizations that exert an intrinsic timing of bursting in the network. Though the exact mechanism of spontaneous network bursting is not known, changes in the frequency and amplitude of calcium transients associated with these network bursts likely represent many aspects of synaptic transmission and the activity of many neurotransmitter systems and are thus suitable surrogate phenotypic marker for neural activity. This may be an advantage in a compound or genetic screening since hits have the potential to affect a broad number of pathways and mechanisms of neurotransmission and phenotypic-based screening is gaining traction as a successful approach in drug-discovery [16]. The potential utility of simultaneously interrogating multiple pathways with a phenotypic approach was demonstrated here with the results showing that the diverse compound collection of the LOPAC 11280 library impacted diverse cellular processes of the human neural spheroids. Traditionally, incorporating phenotypic and/or native functional readouts from 2-dimensional cultures has been difficult in screening paradigms has been difficult. Similarly, we observed greater well-to-well variation across the 2D plates used in these experiments. One potential reason for the greater well-to-well variation in the 2D cultures is that the amplitude of the oscillations could be below the detection limit of our instrument. However, this still highlights the utility of the consistent activity of the 3D cultures that was detected in every well. The observed very low well-to-well variability on 3D neurospheres illustrates how they are able to bring native function into high-throughput screening applications. By analyzing the frequency and amplitude of calcium oscillations in these 3D cultures of human neurons and astrocytes, we were able to screen for modulators of network activity. Features of the calcium oscillations such as frequency, amplitude, and synchronization can be used to describe network activity and quantified to rank the effects of various compounds on the system. We demonstrated that spontaneous activity can be modulated with known excitatory and inhibitory compounds. Subsequently, a library of 1280 compounds with known activity was screened at 1 μM, and 111 (8.6%) compounds were identified that modulated neuronal network activity through a wide variety of biological mechanisms. Of these, 17 compounds were tested in concentration response experiments, and 16 of the 17 compounds were confirmed as active, including three classes of compounds that affected peak frequency and amplitude at all time points: adenosine receptor agonists, GABA receptor modulators, and glutamate receptor modulators. Additionally, seven other compounds acting through various mechanisms inhibited spontaneous activity. It is possible that some compounds identified in our screen are toxic and thus decrease activity in a non-specific manner. While we did not perform a viability assay due to the acute nature of the treatment, this could be done either simultaneously or as a follow-up exercise to address cell viability with novel hits in subsequent screens similar to previous work using longer incubation periods [19]. There were two compounds in the primary screen, forskolin and amitriptyline hydrochloride, that increased peak frequency at all timepoints. Forskolin activates adenylate cyclase and thus likely increases spontaneous activity in a non-specific manner. Amitriptyline hydrochloride is a tricyclic antidepressant and was the only compound that was not confirmed in the secondary concentration-response experiment. In the primary screen, amitriptyline hydrochloride increased peak frequency, but when fresh powder was obtained and the experiment repeated for concentration response, this compound failed to repeat and abolished spontaneous activity at the highest concentration.
One advantage of screening neurospheres system used on this study is the balanced composition of neurons and astrocytes, better mimicking the in vivo condition. The 3D system is more physiologically relevant as the two cell types exist in a more natural state. This may be a benefit in screening for novel compound modulators of network activity. While the compounds we tested here are well characterized and have known effects on neurons and neurotransmitter systems, any unknown compounds discovered that impact neurosphere activity could be specific to astrocytes or neurons. We presented one method for uncovering cell-specific effects through the use of a neuronal specific Ca 2+ sensor. Moreover, the use of calcium reporters driven by targeted promoters could help to visualize the activity modulation on different cell types when under the influence of compounds.
Modeling disease using human iPSCs has revealed consistent synaptic deficits in several disorders including schizophrenia and autism [1,4,20]. The assay described herein may have the potential for future disease-modeling studies of psychiatric disorders and phenotypic drug screening. New therapies for neurological disorders are a vital area of unmet need partially due to the unknown complexity of the central nervous system. This 3D neurosphere platform may provide a phenotypic approach to interrogating neurological diseases and ultimately lead to therapies. The microBrain 3D neurospheres have also been used to profile neurotoxic compounds, which highlights the broad applicability of this platform [19].
In summary, we have described the utility of 3D human neural cultures in drug screening for modulators of neural network activity. We generated a data set that can serve as a reference for compounds that alter spontaneous network activity. Moreover, we confirmed that compounds that increase or decrease network bursting can be identified in a reproducible fashion using this assay, making it ideal for screening campaigns. This platform is robust and can be used for future drug screening and disease modeling efforts.