The data that support the findings of this study, including the original single-cell RNA Sequencing data, are publicly available at this doi: www.doi.org/10.17881/lcsb.20190326.01.
Furthermore, a previous version of this manuscript is available as pre-print under: https://doi.org/10.1101/589598.
Pluripotent stem cell culture
hiPSC lines were provided by Bill Skarnes, Wellcome Trust Sanger Institute (iPSC Bill), Alstem (iPS15, derived from human peripheral blood mononuclear cells, episomal reprogrammed) or previously described in Reinhard et alia (Reinhardt et al. 2013b). The cells were cultured on Matrigel-coated (Corning, hESC-qualified matrix) plates, maintained in Essential 8 medium (Thermo Fisher Scientific) and cultured with and split 1:6 to 1:8 every 4 to 5 days using Accutase (Sigma). Ten μM ROCK inhibitor (Y-27632, Abcam) was added to the media for 24 h following splitting.
Derivation of midbrain floorplate neural progenitor cells
The derivation and maintenance of midbrain floorplate neural progenitor cells (mfNPCs) has been described previously (Smits et al. 2019).
In brief, embryoid bodies (EBs) were formed with 2000 iPSCs each, using AggreWell 400 (Stemcell Technologies). The cells were cultured in Knockout DMEM (Invitrogen) with 20% Knockout Serum Replacement (Invitrogen), 100-μM beta-mercaptoethanol (Gibco), 1% nonessential amino acids (NEAA, Invitrogen), 1% penicillin/streptomycin/glutamine (Invitrogen), freshly supplemented with 10-μM SB-431542 (SB, Ascent Scientific), 250-nM LDN-193189 (LDN, Sigma), 3-μM CHIR99021 (CHIR, Axon Medchem), 0.5-μM SAG (Merck) and 5-μM ROCK inhibitor (Sigma). After 24 h, EBs were transferred to a non-treated tissue culture plate (Corning). On day two, medium was replaced with N2B27 medium consists of DMEM-F12 (Invitrogen)/Neurobasal (Invitrogen) 50:50 with 1:200 N2 supplement (Invitrogen), 1:100 B27 supplement lacking vitamin A (Invitrogen) with 1% penicillin/streptomycin/glutamine, supplemented with 10-μM SB, 250-nM LDN, 3-μM CHIR and 0.5-μM SAG. On day four and six, medium was exchanged with the same but including 200-μM ascorbic acid (AA, Sigma). On day eight, EBs with neuroepithelial outgrowth were triturated into smaller pieces and diluted in a 1:10 ratio. For following passages, 1× TrypLE Select Enzyme (Gibco)/0.5-mM EDTA (Invitrogen) in 1× PBS was used and 10,000 to 20,000 cells per 96-well ultra-low attachment plate (round bottom, Corning) were seeded. The cells were always kept under 3D culture conditions and from passage 1 on cultured in N2B27 medium freshly supplemented with 2.5-μM SB, 100-nM LDN, 3-μM CHIR, 200-μM AA and 0.5-μM SAG. After every cell split, the ultra-low attachment plate was centrifuged for 3 min at 200×g to assure the aggregation of single cells at the bottom of the well. Additionally, a 5-μM ROCK inhibitor was added. The cells were split every 7 to 14 days and the medium was changed every third day. After four to five passages, mfNPCs were used as a starting population for hMOs.
Generation of midbrain-specific organoids
To start the generation of hMOs, 3000 cells per well were seeded to an ultra-low attachment 96-well round bottom plate, centrifuged for 3 min at 200×g and kept under maintenance conditions for 7 days. LDN and SB were withdrawn of mfNPC expansion medium and after three additional days, the concentration of CHIR was reduced to 0.7 μM. On day nine of differentiation, medium was changed to neuronal maturation N2B27 medium including 10-ng/ml BDNF (Peprotech), 10-ng/ml GDNF (Peprotech), 200-μM AA (Sigma), 500-μM dbcAMP (Sigma), 1-ng/ml TGF-β3 (Peprotech), 2.5-ng/ml ActivinA (Life Technologies) and 10-μM DAPT (Cayman). The organoids were kept under static culture conditions with media changes every third day for 35 or 70 days. Detailed information about the generation of hMOs has been published recently (Smits et al. 2019).
hMOs were fixed with 4% PFA overnight at 4 °C and washed 3× with PBS for 15 min. After treatment, they were embedded in 3–4% low melting point agarose in PBS. The solid agarose block was sectioned with a vibratome (Leica VT1000s) into 50 or 70-μm sections. The sections were blocked on a shaker with 0.5% Triton X-100, 0.1% sodium azide, 0.1% sodium citrate, 2% BSA and 5% normal goat or donkey serum in PBS for 90 min at RT. Primary antibodies were diluted in the same solution but with only 0.1% Triton X-100 and were applied for 48 h at 4 °C.
After incubation with the primary antibodies (Supplementary Table 2), sections were washed 3× with PBS and subsequently blocked for 30 min at RT on a shaker. Then sections were incubated with the secondary antibodies in 0.05% Tween-20 in PBS for 2 h at RT and washed with 0.05% Tween-20 in PBS and Milli-Q water before they were mounted in Fluoromount-G mounting medium (Southern Biotech).
STAINperfect Immunostaining Kit (ImmuSmol) was used according to manufacturer’s protocol to detect dopamine, serotonin, GABA and L-glutamine. Nuclei were counterstained with Hoechst 33342 (Invitrogen).
For qualitative analysis, three randomly selected fields per organoid section were acquired with a confocal laser scanning microscope (Zeiss LSM 710) and images were further processed with OMERO Software. Three-dimensional surface reconstructions of confocal z-stacks were created using Imaris software (Bitplane).
Quantitative image analysis
Immunofluorescence 3D images of hMOs were analysed in Matlab (Version 2017b, Mathworks). The in-house developed image analysis algorithms automate the segmentation of nuclei, astrocytes and neurons with structure-specific feature extraction. The image preprocessing for the segmentation of nuclei was computed by convolving the raw Hoechst channel with a Gaussian filter. By selecting a pixel threshold to identify apoptotic cells, a pyknotic nuclei mask was identified and subtracted from the nuclei mask.
For the segmentation of neurons, a median filter was applied to the raw TUJ1 channels. The expression levels were expressed in two ways as follows: (i) positive pixel of the marker, normalized by the pixel count of Hoechst; (ii) cells positive for a marker expressed as a percentage of the total number of cells. In this latter case, the nuclei were segmented and a watershed function was applied. Considering the high cell density of the specimens, steps to ensure high quality in the segmentation process were implemented and structures with a size higher than 10,000 pixels were removed (this indicated incorrected segmentation, e.g. clumps). In the nuclei successfully segmented as a single element, a perinuclear zone was identified. In case the marker of interest was positive in at least 1% of the perinuclear area, the corresponding cell was considered as positive.
Single-cell RNA sequencing using droplet-sequencing (Drop-Seq)
Single-cell RNA sequencing (scRNA-seq) data were generated using the Droplet-Sequencing (Drop-Seq) technique (Macosko et al. 2015) as described previously (Walter 2019). In this work, we performed scRNA-seq of hMOs derived from hiPSC line H4 (see Supplementary Table 1). For each time point, 35 days and 70 days after dopaminergic differentiation, we pooled and analysed 30 hMOs each.
Pre-processing of the digital expression matrices from scRNA-seq
The result of the Drop-Seq scRNA-seq pipeline and subsequent bioinformatics processing is a digital expression matrix (DEM) representing the number of mRNA molecules captured per gene per droplet. Here, we obtained two DEMs, one corresponding to 35-day hMOs and the other to 70-day hMOs. After quality cut based on knee plots, we retained for each sample 500 cells with the highest number of total transcripts measured and performed normalization of the DEM separately. Finally, the two DEMs were merged for the comparison analysis of the two time points based on 24,976 expressed genes in 1000 cells. The data was analysed by our customized Python analysis pipeline (Python version 3.6.0, with anaconda version 4.3.1) including dimensionality reduction by t-distributed stochastic neighbourhood embedding (t-SNE) (van der Maarten and Hinton 2008) and differential gene expression analysis.
Analysis of differentially expressed genes from scRNA-seq data
To determine which and how many genes were differentially expressed between 35-day and 70-day hMOs, we applied a one-way ANOVA test, a one-way ANOVA test on ranks (Kruskal-Wallis test), and a Mutual Information based test. The minimum p value obtained for each gene across these three tests was retained and statistical significance was set to p < 0.01 after Bonferroni correction for multiple hypothesis testing of differentially expressed genes (DEGs).
Cumulative gene expressions from scRNA-seq data
From literature, we extracted cell type–specific gene lists (Supplementary Table 3) for stem cells, neurons and neuronal subtypes (dopaminergic, glutamatergic, GABAergic and serotonergic neurons) (Reinhardt et al. 2013a; La Manno et al. 2016; Cho et al. 2017). Note that not all genes listed therein have been measured in our dataset; these were highlighted in Supplementary Table 3.
For each list, we defined a score, which we refer to as cumulative gene expression, computed as the sum of the expression of the corresponding genes from normalized DEM for each cell. Since the expression levels were measured at single cell level, we can consider the cells’ distributions across the cumulative genes expression scores (Fig. 2a). These histograms exhibit the cumulative gene expression scores normalized to their maxima on the horizontal axis. Thus, on the horizontal axis, a value of 1 corresponds to the maximal cumulative gene expression for one list of genes, while 0 corresponds to no expression of any genes from that list. The vertical axis exhibits the number of cells falling into the corresponding bin of the histogram. In each subpanel, the distributions for day 35 and for day 70 are shown. Population differences were assessed by Z-test of the means with Bonferroni correction.
Gene-gene correlations from scRNA-seq data
From the scRNA-seq data, we also computed gene-gene Pearson correlation coefficients for stemness-specific and neuron-specific genes. Analysis was performed independently for the two samples (35-day DA dif and 70-day DA dif) resulting in two correlation matrices (Fig. 1c).
In the lower triangular matrix, all correlation values are shown, whereas the upper triangular matrix contains only statistically significant correlations (p value < 0.05 after Bonferroni correction). For visual clarity, diagonal elements and undetected genes were excluded.
Fold changes of gene expression from scRNA-seq data
For individual genes, we considered the normalized gene expression across cell populations. For each selected gene, we compared its expression within the 35-day cells with the one within the 70-day cells by computing the logarithmic fold change (log2FC). We performed this analysis for the genes specific of neuronal subtypes including glutamatergic neurons, GABAergic neurons and dopaminergic neurons (Fig. 2c–d), where negative values indicate that a gene is less expressed at day 35 than at day 70 and positive numbers the opposite. p values are based on Z-test with Bonferroni correction and significance levels correspond to * = p value <0.05, ** = p value <0.01, *** = p value < 0.001, and **** = p value < 0.0001. Error bars represent SEM based on the individual sample average and error propagation.
scRNA-seq data analysis for UMAP plot, dot plot and violin plot
ScRNA-seq data were generated using the Droplet-Sequencing (Drop-Seq) technique (1). After bioinformatics processing, we obtained two digital expression matrices (DEM), corresponding to day 35 and day 70 after differentiation into human midbrain organoids (hMOs).
In an alternative analysis approach, which is independent and complementary to the scRNA-Seq analysis described above, further data processing was performed using the Seurat v.3.0.0 R package (Satija et al. 2015). Cells with more than 4000 or less than 500 detected genes, as well as those with mitochondrial transcripts proportion higher than 7.5% were excluded. We collected a total of 1295 cells (505 cells at day 35 and 790 cells at day 70). The datasets were log normalized and scaled to 10,000 transcripts per cells. The top 2000 highly variable genes for day 35 and day 70 were determined using the variance-stabilizing transformation method. The datasets from day 35 and day 70 were integrated using canonical correlation analysis (CCA) in the Seurat package (Stuart et al. 2019). The datasets were integrated based on the top 30 dimensions from CCA using the Seurat method by identifying anchors and integrating the datasets. The resulting integrated data were scaled and principal component analysis (PCA) was performed. Clustering was performed based on the top 30 principal components (PCs), using the shared nearest neighbour (SNN) modularity optimization with a resolution of 0.8. Cluster identities were assigned based on cluster gene markers as determined by the “FindAllMarkers” function in Seurat and gene expression of known marker genes.
Sixty-three-day-old hMO specimens were immersion-fixed in a solution of 2% PFA and 2.5% glutaraldehyde in 0.1-M sodium cacodylate buffer (pH 7.4, Electron Microscopy Sciences, Hatfield, PA) for 3 h, rinsed several times in cacodylate buffer and further post-fixed in 2% glutaraldehyde in 0.1-M sodium cacodylate buffer for 2 h at room temperature on a gentle rotator; fixative was allowed to infiltrate an additional 48 h at 4 °C. Specimens were rinsed several times in cacodylate buffer, post-fixed in 1.0% osmium tetroxide for 1 h at room temperature and rinsed several times in cacodylate buffer. Samples were then dehydrated through a graded series of ethanols to 100% and dehydrated briefly in 100% propylene oxide. Tissue was then allowed to pre-infiltrate 2 h in a 2:1 mix of propylene oxide and Eponate resin (Ted Pella, Redding, CA), then transferred into a 1:1 mix of propylene oxide and Eponate resin and allowed to infiltrate overnight on a gentle rotator. The following day, specimens were transferred into a 2:1 mix of Eponate resin and propylene oxide for a minimum of 2 h, allowed to infiltrate in fresh 100% Eponate resin for several hours, and embedded in fresh 100% Eponate in flat moulds; polymerization occurred within 24–48 h at 60 °C. Thin (70 nm) sections were cut using a Leica EM UC7 ultramicrotome, collected onto formvar-coated grids, stained with uranyl acetate and Reynold’s lead citrate and examined in a JEOL JEM 1011 transmission electron microscope at 80 kV. Images were collected using an AMT digital imaging system with proprietary image capture software (Advanced Microscopy Techniques, Danvers, MA).
The Maestro microelectrode array (MEA, Axion BioSystems) platform was used to record spontaneous activity of the hMOs. A 48-well MEA plate containing a 16-electrode array per well was precoated with 0.1-mg/ml poly-D-lysine hydrobromide (Sigma-Aldrich). Sixty to seventy days old organoids of two different passages were briefly treated for 5 min with 1× TrypLE Select Enzyme, resuspend in 10 μg/ml laminin (Sigma-Aldrich) and placed as a droplet onto the array. After 1 h incubation, neuronal maturation media was added and cells were cultured for 1–2 weeks. Spontaneous activity was recorded at a sampling rate of 12.5 kHz for 5 min at 37 °C over several days. Axion Integrated Studio (AxIS 2.1) was used to assay creation and analysis. A Butterworth band pass filter with 200–3000 Hz cutoff frequency and a threshold of 6× SD were set to minimize both false positives and missed detections. The spike raster plots were analysed using the Neural Metric Tool (Axion BioSystems). Electrodes with an average of ≥ 5 spikes/min were defined as active, for the pharmacological treatment 24 electrodes were analysed. The organoids were consecutively treated with Gabazine, D-AP-5, NBQX (Cayman Chemical, end concentration: 50 mM each), and Quinpirole (Sigma Aldrich, end concentration: 5 μM). To block all neuronal activity and thus verify spontaneous spiking activity of the cells, tetrodotoxin (TTX, Cayman Chemical, 1 μM) was applied at the end. The spike count files generated from the recordings were used to calculate the number of spikes/active electrode/min. Further details regarding the MEA system were previously described (Bardy et al. 2015).
If not stated otherwise, experiments were performed with three independently generated organoid cultures from three different cell lines (n = 9). Gaussian distribution was evaluated by performing D’Agostino and Pearson omnibus normality test. In case the data were normally distributed, Grubbs’ test was performed to detect significant outliers. Unpaired t test with Welch’s correction or nonparametric Kolmogorov-Smirnov test was performed to evaluate statistical significance. Data are presented as mean ± SEM. The statistical analyses of scRNA-seq data are described in the corresponding sections.