Transferable neuronal mini-cultures to accelerate screening in primary and induced pluripotent stem cell-derived neurons

The effort and cost of obtaining neurons for large-scale screens has limited drug discovery in neuroscience. To overcome these obstacles, we fabricated arrays of releasable polystyrene micro-rafts to generate thousands of uniform, mobile neuron mini-cultures. These mini-cultures sustain synaptically-active neurons which can be easily transferred, thus increasing screening throughput by >30-fold. Compared to conventional methods, micro-raft cultures exhibited significantly improved neuronal viability and sample-to-sample consistency. We validated the screening utility of these mini-cultures for both mouse neurons and human induced pluripotent stem cell-derived neurons by successfully detecting disease-related defects in synaptic transmission and identifying candidate small molecule therapeutics. This affordable high-throughput approach has the potential to transform drug discovery in neuroscience.

raft), and after one to two days the rafts are released, en masse, into cell culture media reservoirs where they are cultured together for days or weeks to attain proper maturity with negligible evaporative losses ( Fig. 1C-E). Rafts are then sorted as needed into high content well plates for compound screening via pipette and used for treatment or staining. Alternatively, rafts can be transferred via a magnetic wand 11 . A magnetic plate placed under the well plate quickly directs the magnetic raft to the bottom of the well. Importantly, once rafts are located to the floor of the well, they stay in place during normal transport or automated stage movements. The localization of these rafts at the bottom of the well also allows for imaging via high resolution microscopy and makes the rafts compatible with autofocus functions. To facilitate fluid exchanges and treatment proto-cols, rafts are held in place on the well floor with a plate magnet. The slight concavity of the raft (Fig. 1D) and the surface tension of the media surrounding the raft protect neurons during release and transfer, without compromising the high-content imaging required for most screening strategies. Unlike contact spotting, which has been used to generate small-sized cultures for non-neuronal cell types [12][13][14] , the mobile raft mini-cultures can be independently screened in parallel. This method increases screening potential .30-fold over traditional well-plate cultures, while remaining compatible with wellplate-based high-throughput drug screening equipment (Fig. 1F).
Mobile mini-cultures of human stem cell-neurons. Advances in stem cell technology now allow generation of unprecedented models of human neurological disease states in vitro; however, the time and resources for such stem cell work presents a notable bottleneck. We reasoned that our micro-raft approach might allow screens using neurons derived from stem cells. We first generated neurons from H9 embryonic stem cells (ESC-neurons; Fig. 2A-E) 15,16 . We then cultured ESC-neurons onto raft arrays and released rafts within 2 days (see Fig. 1D). Neuronal marker expression and function ( Fig. 2F-H) were confirmed by 5 days on released rafts, further validating that the raft micro-cultures remain functional after they are released from arrays.
Mini-cultures versus well-plate cultures. To evaluate the utility of raft micro-cultures for drug screening, we performed a viability assay using human ESC-neurons on released rafts compared with 384-well plates (Fig. 3). We assessed the neuronal viability of our ESC-neurons by determining the percentage of the dead cell marker SYTOX green 17 -positive cells and using nuclear markers (DAPI or Hoechst) to label all cells (Fig. 3A). ESC-neurons were plated into a 384-well plate at 20,000 neurons/well (initial plating density: 1,800 neurons/mm 2 ), as performed previously 18 . Within the wells of the well-plate, neurons are unevenly distributed, with the highest density around the periphery. We chose to compare neuron viabilities on day 5, the earliest time point we could detect neuronal function (Fig. 2H) from 300 mm x 300 mm regions around the center of each raft or well to simulate future automated imaging. By this day, there was a significantly lower fraction of dead cells on rafts than in wells (Fig. 3B). To ensure that the difference in viability was not due to the difference in neuron density at this time [612.5632.5 live neurons/mm 2 on raft vs. 291.4621.9 live neurons/mm 2 in wells; p,0.001], we compared the percentage of SYTOX1 neurons within a subset of rafts and wells containing equivalent densities of live neurons (see methods; Fig. 3B). Within these subsets, the difference in neuronal viability was retained (p , 0.05).
Next, we quantified sample-to-sample variability by plotting the percentages of SYTOX1 neurons of either rafts or wells on day 5 (Fig. 3C). Our analysis suggests that the percentage of dead cells is more consistent between rafts than between wells. Indeed, the cumulative distributions of percentages of SYTOX1 neurons were significantly different, with the distribution for the rafts significantly steeper, indicating less variability between rafts compared with wells (p,0.05). In addition, the between-sample consistency was maintained using rafts up to day 14 ( Fig. 3C-D), whereas neurons in wells required daily media changes to survive (data not shown). This lack of maintenance requirements for raft arrays is a significant advantage over high-density well plates. These differences could be due to the higher amount of neuronal material required for well plating more quickly depleted the media of nutrients. Using rafts would drastically decrease the number of neurons per screen. Alternatively, over this time media composition in small volume well plates might be significantly altered by evaporation. Released rafts can be maintained in larger volume reservoirs of media less sensitive to evaporative effects and later moved to high content plates for imaging. Taken together, the increases in both neuronal viability and sample-to-sample repro- ducibility represent distinct advantages of raft arrays over conventional well-plate cultures.

Measurement of synaptic phenotypes using human iPSC-neurons.
While sample-to-sample consistency between screens is vital to drug discovery, it is also critical that the model system recapitulates human disease phenotypes 9 . Often this is accomplished by harvesting neurons from readily-available transgenic rodent lines, but human iPSCs are increasingly used to generate neurons from neurological patients. To confirm that raft mini-cultures could successfully be used as a platform to model human disease phenotypes, we used iPSC-neurons from a patient with Fragile X (FX iPSC-neurons), the most prevalent inherited form of intellectual disability 19 , and we functionally compared these neurons to iPSCneurons from a neurotypical individual (CNTL iPSC-neurons) ( Fig. 4A-H). We concentrated on FX for these proof-of-principle experiments because: 1) the genes underlying FX are well-known, 2) the disease state can be confirmed by the lack of Fragile X mental retardation protein (FMRP) expression, and 3) there is a wellcharacterized, transgenic rodent model (the FMR1 KO mouse) for comparison. Importantly, iPSC-neurons exhibited neuronal markers (Fig. 4C), and neither the reprogramming nor differentiation altered the level of FMRP expression (Fig. 4D).
We plated FX and CNTL iPSC-neurons onto raft arrays, released them after 1 day, and then used lipophilic FM dyes to evaluate differences in presynaptic function at 14 days post-plating 20,21 . We found a significant decrease in the volume of FM puncta following activitydependent loading in FX iPSC-neurons, compared to CNTL-iPSC neurons (Fig. 4E), suggesting a smaller recycling pool or a reduction in the rate of endocytosis. In addition, we found significantly enhanced unloading kinetics in FX iPSC-neurons, compared to CNTL iPSC-neurons ( Fig. 4F-H). Importantly, these differences were also found between primary neurons from FMR1 KO mice and WT mice ( Fig. 4E-H), and are consistent with the previous findings that FX neurons have decreased numbers of synaptic vesicles 22 and a higher probability of neurotransmitter release 23 . These results demonstrate that, despite the reduced number of neurons in microrafts compared to culture-wells, there remains sufficient statistical power to screen and identify reproducible functional phenotypes.
Drug screening using released micro-rafts. Finally, we performed a proof-of-concept study to demonstrate the use of raft arrays in a drug screening assay, by reproducing previous results from a much lower throughput screen 18 . We used embryonic neurons from a mouse model containing a Ube3a-YFP transgene within the normally silent paternal allele previously used to screen potential therapeutic compounds for Angelman syndrome 18 . Just as in the previous screen, the topoisomerase inhibitor, topotecan, successfully increased expression of YFP in raft mini-cultures (Fig. 5), indicating that topotecan activates the normally dormant paternal allele of Ube3a. Together, these data show that drug treatment in raft mini-cultures reproduces findings reported using traditional well-plate cultures.

Discussion
Micro-raft arrays allow significantly enhanced sample sizes and a reduction in the number of neurons needed from existing supplylimited sources. An extraordinary amount of time and resources are dedicated to generating neuron cultures. The lack of efficient differentiation protocols for stem cells and the length of time required to differentiate and maintain these neurons (.30 days), present bottlenecks for their use in high-throughput screening 9 . For neurons obtained directly from embryonic mice, which yield the largest number of live cortical neurons, the generation of animals presents a major bottleneck, compounded by poor reproductive success (,50%), small litter size, long gestation times, and short fertile windows. For example, cortical neurons harvested from a single mouse embryo (,3,000,000 neurons) could generate 4,800 rafts; the same number of samples using 384-well-plates would require 32 embryos and multiple litters.
Micro-raft arrays also overcome key technical challenges in highthroughput screening using neurons. First, they overcome difficulties with evaporation that plague traditional long-term micro-cultures in high-density well plates 9 . After micro-rafts are released, they can be cultured for extended lengths of time in large media volumes that are not affected significantly by evaporation until they are ready to be transferred for screening and imaging. Second, the use of arrays reduces the plating time, supplies, and reagents, because dissociated neurons are plated en masse onto each disposable array (,3 arrays per embryonic mouse cortex) (movie S1).
Finally, additional experimental possibilities result from advantages in handling capabilities. These arrays are produced rapidly and reproducibly 11 , do not require large investment in expensive equip- Human iPSCs were generated from purchased human fibroblast lines. The Fragile X positive fibroblast line, GM05131A was purchased from Coriell Institute for Medical Research at passage 11 and non-Fragile X line CCD-1079sk was obtained from ATCC and used for experiments at passage 7. Both lines were maintained and cultured according to the vendor's instructions. To induce pluripotency, both fibroblast lines were plated into 6-well plate (200,000 cells/well) and maintained overnight in DMEM (Gibco) supplemented with 10% FBS (Gibco). Cells were then infected with lentiviral particles carrying the following reprogramming genes: Oct4,  Sox2, Klf4, cMyc, Lin28 and Nanog. Individual hiPS clones were selected after 14 to 21 days, expanded and characterized using the germ layer characterization kit (Millipore). To verify their pluripotency hiPS clones were plated onto 10 cm tissue culture dishes and grown to 80% confluency. At this point cells were dissociated using EDTA and grown for three weeks as embryoid bodies (EBs) in suspension on Petri dishes in IMDM (Lonza) supplemented with 10% FBS (HyClone). In addition, Rock inhibitor (Selleck Chemicals) was maintained in the medium for the first 24 hours. After three weeks embryoid bodies were seeded on Matrigel-coated 6-well plates, fixed and processed for expression of germ layer biomarkers (Fig. 2).
At day 24, the neuroepithelial cells in the center of colonies that had formed neural tube-like rosettes and were attached loosely were manually lifted, dissociated either manually or with Accutase (Life Technologies) and then seeded on poly-L ornithine coated micro-raft arrays or well plates in N2B27 medium overnight. The following day, micro-rafts were released from the array and the medium was changed to N2B27 supplemented with BDNF (10 ng/ml) for all cells. For ease of understanding, days were renumbered such that day 1 corresponded to the first day on raft arrays ( Fig. 2A) and day numbers in throughout the text use this numbering system unless otherwise stated. Stem cell neurons on both released rafts and in wells began extending neurites by day 2-3. Stem cell neurons were then maintained for up to 14 days post plating as described.
Stem cell-neuron viability assay. To assess viability of H9-derived neurons, cells were plated on either micro-raft arrays (1 million neurons/array) or in wells of a 384well plate (20,000 neurons/well) in N2B27 containing BDNF. Rafts were released and transferred on day 2 as described. On either day 5 or day 14, cells were treated with SYTOX green (1 mM; Life Technologies) and either NucBlue (2-3 drops/mL; Molecular Probes) or DAPI (300 nM; Life Technologies) for 5-10 minutes to visualize dead/dying cells and all cells, respectively. Neurons were then rinsed with PBS and mounted with Fluoromount-G (Southern Biotech) for imaging and assessment. Live neuron densities were extrapolated from the number of SYTOXnegative neurons within a 300 mM x 300 mM image from the center of the raft or well. To standardize neuronal density when calculating neuronal viability, percentages of SYTOX-positive neurons were also compared from a subset of rafts or wells containing between 100 and 300 live neurons per 300 mM x 300 mM image. , and Anti-anti [100x; Gibco, 1%]). Next, hippocampi were manually dissociated in fresh NBM, centrifuged (67 xg) for 7 minutes at 4uC, and resuspended (12 million neurons/mL) in NBM. Neurons were then plated on poly-D-lysine-coated raft arrays (1 million neurons/array). Primary cortical neuron cultures from Ube3a-YFP knockin mice were prepared as previously described 18 . Rafts were then released the following day (DIV 2) and maintained in NBM until DIV14 as described.
FM dye assays. To assess puncta-unloading rates, released rafts with either stem cell neurons or mouse neurons were transferred via pipette to a devised micro-chamber (8 mm diameter polydimethylsiloxane; PDMS; reservoir on a glass coverslip). Cell culture media was replaced with HEPES-buffered solution (HBS; 119 mM NaCl, 5 mM KCl, 2 mM CaCl 2 , 2 mM MgCl 2 , 30 mM glucose, 10 mM HEPES) and allowed to recover for at least 30 min at 37uC. HBS was replaced with the FM dye loading solution containing 10 mM N-(3-trimethylammoniumpropyl)-4-(6-(4-(diethylamino)phenyl)hexatrienyl)pyridinium dibromide (FM 5-95; Invitrogen), 20 mM CNQX (Tocris Bioscience), 50 mM d-AP5 (Tocris Bioscience), and high KCl buffered saline (90 mM KCl, 34 mM NaCl, 2 mM CaCl 2 , 2 mM MgCl 2 , 30 mM glucose, 10 mM HEPES) for 1 min to allow for endocytosis of the dye. Loading solution was then removed and neurons were rinsed with 10 mM FM 5-95 in HBS for 1 min. Next, neurons were rinsed three times with a high-Mg 21 , low-Ca 21 solution (106 mM NaCl, 5 mM KCl, 0.5 mM CaCl 2 , 10 mM MgCl 2 , 30 mM glucose, 10 mM HEPES) containing 1 mM Advasep-7 (Biotium) for 1 min. Finally neurons were rinsed three times with HBS containing 20 mM CNQX and 50 mM d-AP5 for 1 min. Unloading was performed using electrical field stimulation. Positive and negative electrodes were placed in on either side of the raft. Stimulation was generated with a two-channel stimulus generator (STG4002; AD Instruments) in current mode with an asymmetric waveform (2480 mA for 1 ms and 1 1600 mA for 0.3 ms) at 20 Hz for 600 pulses. Z-stacks (.30 slices) were captured every 15 s using auto-focus for each time-point to prevent drifting. At least five baseline images were acquired before starting electrical stimulation.
To assess puncta volumes, neurons were loaded as described above with the fixable version of FM 4-64 (Invitrogen) replacing FM 5-95. Following rinses, neurons were fixed with 4% paraformaldehyde in phosphate buffered saline (PBS) containing 40 mg/ml sucrose, 1 mm MgCl 2 , and 0.1 mm CaCl 2 for 15 minutes, rinsed 3 times with PBS, then mounted in Fluoromount-G for imaging.
Immunocytochemistry. Neurons on rafts or in well plates were fixed with 4% paraformaldehyde in PBS containing 40 mg/ml sucrose, 1 mm MgCl 2 , and 0.1 mm CaCl 2 for 20-45 min. Neurons were permeabilized in 0.25% Triton X-100 for 15 min then blocked in PBS containing 10% goat serum for 15 min. Primary antibodies to btubulin (152000; chicken; Aves Labs), MAP2 (151000; rabbit; Millipore), VGLUT1 (15200; mouse; NeuroMab), and FMRP (1510; mouse; 2F5-1 [developed by Tartakoff, A.M./Fallon, J.R. and obtained from the Developmental Studies Hybridoma Bank developed under the auspices of the NICHD and maintained by The University of Iowa, Department of Biology, Iowa City, IA 52242]) were diluted in PBS with 1% goat serum and incubated for 1 h at room temperature. AlexaFluor goat anti-mouse, anti-rabbit, and anti-chicken antibodies conjugated to fluorophores with 488 nm, 568 nm, or 633 nm excitation wavelengths (151000; Invitrogen) were diluted in PBS and incubated for 1 h at room temperature. Immunostaining for neurons from Ube3a-YFP knock-in mice were performed as previously described 18 .
Image processing and analysis. Confocal slices were sum projected in ImageJ and converted to 8-bit depth. For the FM image analysis, we thresholded the first frame of the z-stack to a minimum pixel value of 30. We analyzed puncta .2 pixels 2 and measured the intensity of each of these puncta throughout the registered stack. The intensity of each puncta was normalized to the frame preceding stimulation and slope normalized to baseline. Puncta that unloaded .5% after 1 min were classified as ''unloading'' and were included in the analysis. For puncta volume measurements, zstacks were analyzed using ImageJ plug-in ''Foci Picker3D'' 24 using the following settings: uniform background 5, tolerance setting 5, and minimum pixels in focus 20.
Statistics. Statistical analyses were performed using GraphPad Prism. When comparing distributions we used the non-parametric t-test (Kolmogorov-Smirnov test). Two-tailed unpaired Student's t test was used when comparing means for two conditions. Error bars represented the standard error of the mean (SEM) and the threshold for statistical significance was set at p,0.05 throughout the study.