Journal Pre-proofs Reproducible generation of human midbrain organoids for in vitro modeling of Parkinson’s disease

The study of human midbrain development and midbrain related diseases, like Parkinson`s disease (PD), is limited by deficiencies in the currently available and validated laboratory models. Three dimensional midbrain organoids represent an innovative strategy to recapitulate some aspects of the complexity and physiology of the human midbrain. Nevertheless, also these novel organoid models exhibit some inherent weaknesses, including the presence of a necrotic core and batch-to-batch variability. Here we describe an optimized approach for the standardized generation of midbrain organoids that addresses these limitations, while maintaining key features of midbrain development like dopaminergic neuron and astrocyte differentiation. Moreover, we have established a novel time-efficient, fit for purpose analysis pipeline and provided proof of concept for its usage by investigating toxin induced PD.


INTRODUCTION
Understanding the development of the midbrain dopamine system is the basis for elucidating the underlying mechanisms of diseases such as drug addiction, schizophrenia, autism, and neurodegeneration 1 . The most common age-associated midbrain disorder is Parkinson's disease (PD).
PD is characterized by the irreversible loss of dopaminergic neurons in the midbrain substantia nigra, leading to typical motor symptoms such as bradykinesia and tremor. Despite an aging population and numerous studies in different experimental models, the progress in discovering disease-modifying treatments is slow. The lack of drugs reaching phase two clinical studies is mainly due to a deficiency in representative human models that recapitulate complex multifactorial neurological diseases [2][3][4] . A novel, state of the art, model system that holds great potential for studying the dopaminergic system are midbrain organoids [5][6][7] . These three-dimensional (3D) self-organising cellular structures mimic the actual human midbrain by showing a defined spatial organization that allows differentiation into different cell types and favours their communication 8,9 . As such, compared to 2D cell culture or animal models, midbrain organoids are able to model aspects of human brain histology and physiology.
Nevertheless, they still exhibit -similar to other organoid systems -certain limitations. A technical drawback is the absence of vascularization leading to an insufficient nutrient or oxygen supply in the centre of the organoid that results in a necrotic area, the so called dead core 5,7,10 . Another issue related to organoid studies is the heterogeneity in size and cellular composition. This variability affects the reproducibility of results obtained with organoids 5 . Last, the analysis techniques for 3D cell cultures are still limited, and every existing technique needs to be optimized for 3D organoids. Thus, there is a need for optimized and standardized organoid model systems. Here, we focus on midbrain organoid generation to address these limitations. We intended to generate more reproducible and viable organoids to further reduce the gap between the actual human midbrain and the currently available in vitro models.

Protocol optimization leads to smaller and earlier differentiated midbrain organoids.
In the first step of organoid standardization, we speculate that starting from a more committed cell type reduces variability between organoids. Therefore, two different guided midbrain organoid generation strategies were combined in this study 7,11 . Organoids were derived from human floor plate neuronal progenitor cells (mfNPC) similar to Smits et al. 11 . The mfNPCs were derived from three independent induced pluripotent stem cells lines (Figure S1A and S1B) and the quality of the derivation was assessed by staining with antibodies against a set of cell identity markers for stemness and midbrain ( Figure S1C) 11,12 . Moreover, iPSC quality had been extensively assessed before for all three cell lines in Nickels et al., 2019 38 , Supplementary Figure 1. The actual midbrain organoid generation protocol was optimized based on Monzel et al. 7 . In the following, we compared three different conditions for midbrain organoid generation. Condition A represents the current standard conditions combining the two previously described protocols 7,11 (Figure 1A). In this condition, mfNPCs were cultivated for 8 days under maintenance conditions (Ascobic Acid (AA), LDN193189 (LDN), SB431542 (SB), Smoothened Agonist (SAG) and CHIR99021 (CHIR)) in 96 well ultralow adhesion plates in order to form 3D colonies. At day 8 they were embedded into 30 µl geltrex droplets as described by Monzel et al., favouring regionalisation 5,8 . Next at day 8, maturation was initiated in 24 well ultralow adhesion plates, under shaking conditions 7 . Maturation consists in a pre-patterning step for 3 days (Patterning I media with no LDN and SB), a patterning step for 6 days (Patterning II media reducing the concentration of CHIR) followed by a differentiation step for 21 days (AA, dbc Adenosine monophosphate (AMPc), Brain derived neurotrophic factor (BDNF), Glial derived nerurotrophic factor (GDNF), Activin A (ACT-A), Transforming growth factor beta (TGFß), (2S)-N-[(3,5-Difluorophenyl)acetyl]-L-alanyl-2-phenyl]glycine 1,1-dimethylethyl ester (DAPT)) to reach 30 days in total 11 (details described in the methods section).
In order to optimize this standard approach (Condition A), we assumed that smaller and less dense organoids, having a reduced proliferation phase, will lead to better nutrient and oxygen supply in the core and hence reduce cell death and variability. The first optimization step consisted in modifying the timing of the maturation protocol and embedding in order to reduce the maintenance phase from 8 days to 2 or 5 days respectively (Condition B and C) ( Figure 1A). In condition B, we additionally accelerated the patterning and pre-patterning process from a total of 9 to 6 days ( Figure 1A). A significantly accelerated differentiation could be seen under conditions B and C, independent of the cell lines ( Figure 1B, 1C, S2A-C). This effect was more pronounced in condition B than in condition C ( Figure 1C, S2A and S2C). Differentiation was assessed by measuring the length of the neurite outgrowth from the organoid into the surrounding matrix, the so called differentiation cloud at a specific early timepoint, here 2 days after embedding (day 10 of the process). At later stages the differentiation cloud represents the neurite mass surrounding the cell bodies, however, then it is no longer representative of the neurite length, as the border of the geltrex droplet is a limiting factor forcing the neurites to bend and to intermingle.
In order to further optimize the process we investigated the effect of starting with different cell numbers for all three conditions. A total of 1000, 2000, 3000, 6000, and 9000 cells were used as starting material for colony seeding and embedding into 10 µl (1000, 2000) or 30 µl (3000, 6000, 9000) of geltrex. Colony/organoid core diameter was measured 1 day before embedding (Figure 1D-F, S2D and S2E) and 2 days after embedding (Figure 1G-I , S2F and S2G). Moreover, the size of the organoid was assed at day 30 of differentiation by measuring the Hoechst positive area representing the cell bodies, as the size of the organoid itself is limited by the geltrex droplet ( Figure 1J-L, and S2H-J). Before embedding (day 7), the size of the colonies is strictly defined by the number of starting cells ( Figure   1D-F, S2G and S2E), however once embedded the accelerated differentiation of condition B is also significantly reducing the organoid diameter in comparison to the other conditions ( Figure 1G-I, S2F and S2G). At 30 days of differentiation, when measuring the Hoechst positive area, the size of the organoid is even more dependent on the different conditions and less on the starting number of cells ( Figure 1J-L, S2H, S2J and S2K). These observations indicate that changing the differentiation protocol is sufficient for generating smaller organoids (Figure 1I and 1L). Repeating the size assessment with three different cell lines (cell line names: K7, T12 and COR), having different genetic backgrounds, ages and sex ( Figure S1B) for the following conditions A6000, B9000, and C3000, confirmed that condition B leads to significantly smaller 1 mm (B) vs. 2 mm (A and C) organoid centre size (Figure S2I and S2J).
Interestingly, this is despite the fact that condition B9000 starts with more cells at seeding than the other two conditions. A6000 (2 mm) was chosen as a control condition because it showed reduced size compared to A9000 at 30 days of differentiation ( Figure 1K). B9000 and C3000 were chosen as further optimized conditions as they were of identical size and half the size of A6000 at 30 days (1 mm) ( Figure 1K). Additionally, the organoid production efficiency was assessed by quantifying the operators handling ability of different sized colonies during embedding. For the smallest colonies (1000 cells), half of the colonies were lost during the process (Figure S2L and S2M).
Protocol optimization reduces organoid density and enables whole mount stainings.
In a next step, the density of the dopaminergic neuron network (TH positive cells) was assessed.
Therefore, the outer layer (first 205 µm) of 30 day old immunostained organoids of K7 was 3D reconstructed with the Imaris software package and the TH positive volume per organoid volume (cylinder defined by the area of the organoid and the height of the stack) was quantified (Figure 2A Figure 2I). Antibody penetration in condition B for up to 800 µm is far enough to enable whole mount stainings (Figure 2D and 3A and S4A-C). As whole mount stains drastically improve productivity and reduce labour intensity compared to staining and imaging of individual sections, we optimized this step further (Figure 3). Using the optimized protocol, we were able to perform whole mount stains in all of the three conditions A6000, B9000 and C3000 at day 30 of differentiation for all three cell lines (Representative images are shown) (Figure 3). Moreover, mounting the organoids allowed us to take 20x images throughout the whole organoid (28 stacks a 10µm) (Figure 3 and S4B). Last, the whole mount stainings enabled, 3D reconstruction in Imaris of the whole organoid using either the 5x ( Figure   S4A) or 20x objective ( Figure S4C).
Optimized midbrain organoids have increased viability and no dead core.
One of the most severe limitations of many organoid systems is the presence of dead cells in the centre, reaching a size of up to 4 mm² (the so-called dead core). We next aimed at analysing the dependency of the dead core size on the starting cell number and differentiation conditions. Accordingly, we measured the size of the necrotic area in the centre of the organoid at 30 days of differentiation. Due to instability of this necrotic area, it typically falls out of organoid sections and hence appears as an empty centre in the images of the sectioned organoids ( Figure 4A). Sections of organoids generation in condition A showed a significant dead core, which was slightly smaller when a lower number of starting cells was used (e.g. A9000 vs A3000) ( Figure 4B). However, noticeable differences were observed when comparing conditions B and C to A. Strikingly, organoids generated according to condition B showed no dead core. While those generated according to condition C showed a small dead core in organoids starting with more than 3000 cells ( Figure 4A, 4B and 3). This observation is in agreement with the antibody penetration distance -and though probably nutrient supply -highlighted above ( Figure 2D-F). Based on these results, we decided to use for the further validation, for both conditions B and C, the biggest organoids with no dead core, these correspond to condition B with a starting cell amount of 9000 cells (B9000) and condition C with a starting cell amount of 3000 cells (C3000). Importantly, B9000 and C3000 show a reduced size compared to A6000, which we chose as control condition in the further analysis ( Figure 1K).
In order to investigate how reproducible these results are in different cell lines, the dead core measurements were repeated ( Figure 4C-E), the pyknotic nuclei were counted ( Figure 4F-G), live cells were counted ( Figure 4H) and an ATP viability assay was performed (Figure 4I and 4J) using three independent cell lines and the above-determined conditions. Condition B significantly reduced the dead core formation to nonexistence ( Figure 4C and 4E), the amount of pyknotic cells (measured in whole mount stainings) from 45 % to 30 % (± 2.1% standard error of difference between means) ( Figure 4F) and more than doubled the number of live cells as well as organoid viability (Figure 4H-I).
No significant difference between the cell lines were observable, which indicates a high degree of reproducibility for this approach (Figure 4D, 4G, and 4J).

Midbrain organoids reveal a distinct spatial regionalisation and the presence of glial cells.
After determining condition B as most favourable according to the so far analysed criteria, we investigated if the midbrain organoids derived under this condition are able to generate different cellular identities and a spatial organization ( Figure 5). Therefore, we performed stainings for stem cell-, neuronal -and glial markers at day 30 of differentiation ( Figure 5A-O). Both, whole mount stainings using the optimized immunohistochemistry protocol, followed by Zeiss confocal image quantification, as well as section stainings using high content automated microscopy were performed ( Figure 5A and 5B). Whole mout stainings show slightly squeezed organoids from the top, including all cross sections ( Figure 5C). Sections show the radial and tangential sections of the organoid, which is disc shaped ( Figure 5C). For whole mount stainings the volume of the regions stained with different markers, Hoechst, TH, TUJ1 and FOXA2 was quantified by 3D reconstruction in the Imaris software ( Figure 5A). For sections, the mask of the markers TUJ1, GFAP, S100ß, SOX2, Hoechst, and Ki67 was generated by an in house implemented Matlab script ( Figure 5B). Neuronal quantification in whole mount stainings by TUJ1 showed a significant increase in the amount of neurons in condition B when normalized to the volume of nuclei, independent of cell lines (Figure 4D and 4E). This increase was also observed for the dopaminergic neuron marker TH (Figure 4F and S5B). The increase can be explained by an equal volume of TH (Figure 2G (Figure 4G II). Interestingly, the vast majority of the TUJ1 positive cells expressed TH ( Figure S5C). Moreover, the quantification of TUJ1 over Hoechst, using the whole mount staining approach reflects the observations made in sections, validating this novel method (Figure 5D and S5E). Glial cell analysis in sections demonstrated that the total amount of S100ß and GFAP positive cells was 10.51 % (± 2.4 SEM) and 7.13 % (± 1.2 SEM) respectively, for condition B, which is higher than for the other conditions ( Figure 5H and 5I). In addition, neural stem cells and progenitor cells were quantified using the stem cell marker SOX2, the proliferation marker to form a neuronal layer, followed by a second wave of astrocyte differentiation 18 (Figure S5I).

The optimized protocol reduces batch to batch and cell line variability.
Next, we addressed variations stemming from batch to batch differences or cell line variability ( Figure   5P and 5Q). Therefore, the coefficient of variation (CV) for 9 different conducted experiments was calculated. As parameters for this we chose three measurements for size (the diameter before and after embedding as well as the area of Hoechst at 30 days (Figure 1)); three measurements for differentiation (size of the differentiation cloud (Figure 1), the volume TUJ1/Hoechst and the volume S100ß/Hoechst from the sections (Figure 5)) as well as three measurements related to cell death (the ATP variability assay, the size of the dead core and amount of pyknotic nuclei (Figure 3)). Using these 9 experimental data sets the CV was calculated for each of the 27 data points (3 cell lines, 9 experiments) across all three batches. The same was done for the CV across cell lines (27 data points from 3 batches and 9 experiments). The average CV for condition B in comparison to the control condition A, shows that the optimized protocol is significantly reducing the variability between batches from 46.62 % to 25.77 % (±8.8 standart error of differences between the means) and of differences and between cell lines from 40.33 % to 27.83 % (±5.9 standart error of differences between the means) ( Figure 5P and 5Q). Condition C on the other hand is not reducing the variability significantly ( Figure 5J and S5K).

Optimized midbrain organoids harbour different neuronal subtypes.
In order to further reveal cellular type identities in the midbrain organoids, we assessed the presence of neuronal subtypes by immunocytochemistry at day 30 (Figure 6). Besides dopaminergic signalling, abundantly characterized hereafter, glutamatergic and serotonergic signalling play an import roles within the midbrain. A subset of dopaminergic neurons for instance express vesicular glutamate transporter 2 (GLUT), which when abolished leads to neurodegeneration 14 . Moreover, serotoninexpressing neurons (SERO), are thought to have besides their physiological roles, important functions in midbrain related diseases such as Parkinson's disease 15 . Here, we were able to detect both GLUT positive and SERO positive neurons that either co-expressed TH or not (Figure 6A and 6B). The midbrain substantia nigra is further subdivided in different regions, and mainly dopaminergic neurons of the region A9 are affected in PD. These neurons are positive for the G protein-coupled inwardlyrectifying potassium channel 2 (GIRK2). Indeed, the here described optimized midbrain organoids consist of TH/GIRK2 double positive cells, indicating A9 dopaminergic neuron identity ( Figure 6C).
Moreover, we detected dopamine transporter (DAT) and dopamine (DOPA) itself, both bearing witness of dopamine synthesis and transport (Figure 6D and 6E), as shown in more details by Smits et al. 2019 11 . Lastly, we could show that pre-and post-synaptic proteins such as synaptophysin (SYP) and post-synaptic density protein 95 (PSD95) are expressed, which are necessary for synaptic transmission ( Figure 6F).

Optimized midbrain organoids are suitable to investigating toxin induced PD.
In order to validate the fitness for purpose of the optimized human midbrain organoids for neurotoxicology studies and in vitro modeling of PD, we used the 6-OHDa model, which is well established in rodents 16,17 . Midbrain organoids of all three iPSC cell lines produced according to condition B were treated with 50 µM, 100 µM, 150 µM and 175 µM of 6-OHDA (Figure 7). Treatments were performed at day 30 of differentiation for 2 consecutive days and analysis was done after 6 additional days. The number of GFAP positive cells remained unaffected by the treatment (Figure 7A and 7B). The volume of TH positive dopaminergic neurons, however, decreased in a dose dependent manner (Figure 7C and S5I). The specificity of the loss of dopaminergic neurons was assessed by quantifying the number of TUJ1 positive cells, which slightly but insignificantly decreased ( Figure 7D).
Moreover, when normalizing TH to TUJ1 the decrease upon 6OH-Da treatment remains significant ( Figure 7E), which validates the fitness for purpose of this optimized midbrain organoid system as potential model for PD.

DISCUSSION
By modulating the organoid maturation process, we have generated a novel optimized and standardized human midbrain-like organoid model that exhibits many features reminiscent of the human midbrain dopaminergic system. The optimized generation strategy led to a spatially organized organoid model, including different cell types of neuroectodermal origin. The different cell types encompassed stem cells, neurons and glial cells and emerged in a sequence reminiscent of neuronal development, forming defined structural layers and ventricular-like structures 18 . On one hand, the optimized organoid model has a relative amount of TH-expressing dopaminergic neurons, which is not achieved in any other model 7,11 . On the other hand, the occurrence of different neuronal subtypes underlines the cellular complexity of the neuronal network. Importantly, both characteristics are necessary to model the heterogeneity of midbrain development and pathology. As an example, in PD, selectively A9 dopaminergic neurons expressing vGLUT2 have an increased vulnerability 19,20 .
In this study, we successfully addressed several of the main drawbacks related to organoid production, usage, and analysis, including issues with organoid variability in properties of interest such as between lines as well as between batches, amenability to whole mount imaging, and massive cell death in the organoid centre 21 . Previous studies showed that healthy neurons are limited to the outer perimeter of the organoid because of the necrotic core 22,23 . Using a further committed cell type as starting population and an accelerated differentiation protocol, we were able to eliminate the dead core completely and to reduce cell death by 15 % in the remaining tissue. Moreover, upon the size and density reduction, a viable, FOXA2+ inner cell mass could be observed and ATP production was increased by 40% compared to control condition. In contrast to the most recent strategies that have attempted to reduce the necrotic core and enhance diffusion by specific bioreactors and the reduction of the starting cell numbers, we could show that the modulation of the organoid maturation process itself represents the critical factor 10,24,25 . Reducing tissue density seemingly resulted in a better oxygen and nutrient supply 26 . The optimized neuronal induction is the key towards reduced cell death 24 and represents a big step in the usage of organoids for therapeutic approaches.
In addition, the described approach enabled whole mount stainings that permit 3D reconstruction and quantificative analysis of the whole organoid. This concept represents an efficient method to overcome the challenging and labour-intensive step of sectioning, and the mounting strategy enabled the necessary depth of stain penetration when using confocal microscopy 21 . Moreover, this approach offers a holistic perspective on the spatial distribution, localization, and relative amounts of different cell types. The more guided maturation process did not interfere with regionalization and cellular diversity. Further, compared to previous 3D reconstructions using the clarity approach, we were able to reconstruct and quantify the detailed architecture of the 3D neuronal network, enabling future automated phenotyping strategies 9 . This optimization is a breakthrough in cost and time efficient organoid analysis.
Lastly, the accelerated and more unidirectional maturation process, helped to overcome the variability resulting from spontaneous cellular commitment 27 . Using a more defined patterning strategy in order to optimize reproducibility has been previously observed in dorsally patterned forebrain organoids derived from iPSCs 28,29 . These authors have also shown that exogenous patterning, using a guided strategy, has no effect on terminal cell identities 28,30 . By starting from a more committed cell type and adapting a defined patterning strategy, we were able to reduce the variability between cell lines and between different organoid batches by half in the here described approach. Significantly, the optimization of the timing during the maturation process was revealed to be a critical factor for increased reproducibility and to overcome the batch to batch variability, also called batch syndrome 31 . The increased reproducibility, independent of the batch of the starting material, genetic background, and sex of the original donor, confirms the robustness of the method, which is a crucial characteristic for its use in clinical applications and commercialisation. Nevertheless, it should be noted that although the variability has been reduced there remain some differences.
These variations, were expected as we speak of a living biological system and should be accounted for. In order to draw valid conclusions results should ideally be observed in independent batches and cell lines.
Taken together, we demonstrate that we can generate standardized midbrain organoids in a reproducible manner, while maintaining cellular complexity of the model. The optimization strategy could potentially be extended to other organoid models facing similar limitations. As proof of concept, we validated and demonstrated the fitness for purpose of this novel organoid system, by recapitulating toxin induced dopaminergic neuronal cell death in the midbrain, modeling a form of PD 32 . This achievement, does not only demonstrates the models usefulness for analysing real-life problems, as toxin induced PD represents in some areas of the world the major cause for PD 36,37 , but can also replace the well-established, highly-used and ethically-compromised 6OH-Da rodent models 16,17 . As such, the model has great potential for drug screening and will complement other model systems, as well as clinical studies designed to mechanistically investigate neurodegeneration or other midbrain diseases, in order to discover novel therapeutic approaches.

Online METHODS mfNPC cultivation and derivation from iPSCs
mfNPCs were maintained for up to 15 passages (after their generation from iPSCs) under maintenance medium, splitted at a confluency of 80% using Accutase and seeded 600.000 cells per well on a geltrex  Figure 1A for condition A, B and C. For colony formation they were kept in maintenance medium for 8d for condition A, for 2 days for condition B and 5 days for condition C. Patterning I medium consists in maintenance medium without SB and LDN and was added for condition A and C for 3 days and for condition B for 2days. For patterning II medium the concentration of CHIR was additionally reduced to 0.7 μM. Cells were kept for 6 days for condition A and C, and 4 days for condition B, in patterning II media. Differentiation medium consist in N2B27 medium with 10 ng/ml BDNF (Peprotech), 10 ng/ml GDNF (Peprotech), 200 µM AA, 500 µM dibutyryl cAMP (Sigma), 1 ng/ml TGF-β3 (Peprotech), 2.5 ng/ml ActivinA (Peprotech) and 10 µM DAPT (Cayman). Differentiation medium was added for condition A and C for 21 days, and for condition B for 24 days. At day 8, independent of the patterning strategy, the colonies were embedded into geltrex droplets as described by Monzel et al, and cultivated under shaking conditions in a 24 well ultralow adhesion plates. 7

Immunocytochemistry
Immunofluorsecent stainings using mfNPCs were performed as described in Smits et al., Antibodies are found in Supplementary Table 1. For staining of sections midbrain organoids were fixed at day 30 of differentiation with 4% PFA overnight at 4 °C, washed three times with PBS for 15 min and embedded in 3% low-melting point agarose. The solid agarose block was sectioned with a vibratome (Leica VT1000s) into 80 µm sections.
The sections were blocked on a shaker with 0.5% Triton X-100, 2% BSA, and 5% goat or donkey serum in PBS for 90 min at RT. Primary antibodies were diluted in the same solution with 0.1% Triton X-100 and were applied for 72h at 4 °C. After incubation with the primary antibodies (see Supplementary   Table 1), sections were washed three times with PBS for 10min at RT on a shaker. Then, the sections were incubated with the secondary antibodies for 2 h at RT (same solution as primary) and washed three times with PBS and once with Milli-Q water before they were mounted in Fluoromount-G mounting medium (Southern Biotech).
Whole mount stainings were performed according to a modified protocol described by Koehler et al. 34 Organoids were fixed at day 30 of differentiation with 4% PFA overnight at 4 °C, washed three times with PBS for 5 min. They were incubated for 8min in ice cold Acetone, to optimize antibody penetration (previously described for whole mount immunohistochemistry in Zebrafish eggs by Abcam). Blocking was done with 0.1% Triton X-100, 10% goat or donkey serum in PBS o/n at RT.
Primary antibodies were diluted in 0.1% Triton X-100, 3% goat or donkey serum in PBS and incubated 3 days at 4 °C on a shaker. Organoids were then washed three times with PBS for 60 min at RT and secondary Antibodies were added (same solution as primary) for 2 days at 4 °C on a shaker. Finally, three washes for 5min at RT were performed using 0.05% Tween-20 in PBS and one with water. Then the organoids were mounted in Fluoromount-G mounting medium.
STAINperfect Immunostaining Kit (ImmuSmol) was used according to the manufacturer's protocol to detect dopamine in whole mount stains.

Imaging Analysis and quantification
Confocal images of entire whole-mount organoid stainings were taken using the 5x or 20x objective of the Zeiss PALM/Axiovert fluorescence microscope. Imaris 3D reconstruction software was used in order to quantify the volume of the marker (TH, TUJ1, Hoechst) expression, using the surface function.
Thresholds were set manually according to visual alignment. FOXA2 and nuclei's were counted with the spot function (threshold for size per count set to 10 µm for detecting viable cells only). Pyknotic nuclei were analysed using their smaller size as discriminating factor (7 µm -10 µm = pyknotic nucleis only) compared to the total amount of cells (threshold at 7 µm for viable and non-viable) within the spot functions.
For qualitative and representative images the 60x objective was used.
The organoid sections of 80 µm thickness were acquired with an Operetta High-Content Imaging System (Perkin-Elmer) and analysed with the following image analysis algorithm. Immunofluorescence 3D images of midbrain organoids were analysed in Matlab (Version 2017b, Mathworks). The in-house developed image analysis algorithm for automated marker quantification based on voxels was described before 11,35 and the script can be found in the online repository https://github.com/LCSB-DVB/Nickels_2019. Brightfield images were taken using the Stereomicroscope Nikon SMZ25 or the AxioVert.A1 from Zeiss.

ATP assay
The ATP essay was perfomed using the CellTiter-Glo 3D cell viability assay (Promega) according to the manufactors protocol. Luminescence was measured using the Microplate Cytation5M Cell imaging Multi Mode Reader (Biotek). ATP luminescent levels were normalized to the organoid diameter.

Size measurements
Size was determined using the bright field or fluorescent images and by measuring the diameter or area with the Zeiss software.

Variability
The coefficient of variation was calculated for 9 experimental data sets. Three quality attributes for size: the diameter before and after embedding as well as the area of Hoechst staining at 30 days. Three quality attributes for of differentiation: the size of the differentiation cloud, the volume ratio TUJ1/Hoechst and the volume ratio S100ß/Hoechst. Three quality assessments related to cell death: the ATP production, the dead core size and the pyknotic nuclei count.

6-OHDa and rotenone treatment
The 6-OHDa treatment was performed at day 30 of differentiation at two consecutive days for 24h. 6-OHDa was dissolved at concentration 50 μM, 100 μM, 150 μM and 175 μM in N2 media without growth factors (500 μl/organoid) kept on ice in the dark. The organoids then rested for additional 6 days in differentiation media before analysis.     (Figure 3). (F and G) Pyknotic nucleis were counted in whole mount stainings using Imaris for A6000, B9000, and C3000 and for three cell lines K7, T12 and COR and three batches. Pyknotic nucleis were normalized to total cell count. Mean ±SEM n=9 either grouped by condition F or cell line G. (H) Live cells were counted in whole mount stainings using Imaris for A6000, B9000, and C3000 and for three cell lines K7, T12 and COR and three batches. (I and J) ATP luminescence was normalized to the diameter of the organoid for A6000, B9000, and C3000 and for