Imaging three-dimensional brain organoid architecture from meso- to nanoscale across development

In recent years, advances in stem cell technologies have enabled rapid progress in the field of pluripotent stem cell-based 3D cultures such as brain organoids. These culture formats represent self-organizing structures that recapitulate certain aspects of in vivo brain development. They display complex structures that recapitulate several aspects of early neurogenesis, including the formation of an apical and basal surface, polarized neuroepithelium, neurogenic ventricular and outer radial glia (oRG), the formation of layered, cortex-like architectures and maturation to the level of synapse formation (Paşca et al., 2015; Lancaster et al., 2017). At the same time, proper visualization of these diverse processes in 3D has remained challenging, and most analyses of their cyto- and histoarchitecture are still based on conventional 2D histology of organoid cryosections. However, recently more sophisticated approaches for whole-organoid clearing and imaging that allow addressing the structural features in all three dimensions and preserving the 3D organization by avoiding cryosectioning were introduced and have shown rapid development (Adhya et al., 2021; Albanese et al., 2020; Edwards et al., 2020). Although several studies that employ whole-organoid clearing have been published in recent years, they mainly use small organoids, e.g. to assess early neural differentiation (Benito-Kwiecinski et al., 2021). A number of parameters remain rather challenging in the context of whole-mount analysis of larger organoids. This applies particularly to structures appearing at late stages of organoid differentiation, such as imaging of dendritic spines and synapses (Masselink et al., 2019; Dekkers et al., 2019; Albanese et al., 2020).

Parallel to the development of organoids, novel and fast large-volume imaging methods were introduced that are able to depict fine cellular and subcellular structural features within geometrically extended tissue samples in 3D. The ability to image large neuronal tissue fragments came into reach with the advent of light-sheet fluorescence microscopy (LSFM). This technique allows us to observe, with one or two microscope objective lenses, a fluorescently labeled specimen, the focal plane of which is illuminated perpendicular to the detection axis by a thin sheet of light (Huisken et al., 2004; Dodt et al., 2007). Thus, LSFM offers intrinsic optical sectioning, which can further be amended by confocal line detection (Silvestri et al., 2012; Baumgart and Kubitscheck, 2012) to yield optimal contrast in a scattering specimen. Substantial progress has been made in the past few years by using illumination beam shaping to achieve very thin light sheets to enhance optical resolution over large fields of view. Lattice light sheets (Chen et al., 2014; Ellefsen and Parker, 2018; Stockhausen et al., 2020), Airy beams (Vettenburg et al., 2014) or modifying the Gaussian light-sheet waist across the field of view (Dean et al., 2015; Fu et al., 2016; Neyra et al., 2020) have resulted in extra thin light-sheets with large extensions. Moving the specimen through the illuminating sheet within the detection plane of the imaging objective provides 3D image stacks. When using sensitive and fast CMOS cameras, image rates of hundreds of frames per second can be achieved, significantly decreasing the imaging time for large specimens compared with confocal laser scanning microscopy.

Optimal use of light-sheet microscopy requires the elimination of refractive index inhomogeneities in the probe by using an immersion medium with a refractive index matched to the cellular components of the probe. This can be accomplished by tissue clearing (for a review on different clearing procedures, see Ueda et al., 2020). Combining tissue clearing with LSFM allows effective optical resolutions in the range of 0.3 µm laterally and 1.0 µm axially when using long-distance objective lenses for imaging with a numerical aperture (NA) of 1.0 or greater. Thus, LSFM is especially well suited for the fast analysis of complex arrangements of large cleared cell clusters and tissue fragments, where it enables, for example, fast light microscopic assessment of the complex 3D architecture of organoids (Albanese et al., 2020; Benito-Kwiecinski et al., 2021).

However, LSFM cannot reveal the very fine details of neuronal networks as these structures are well below the optical diffraction limit. To visualize synapses with spatial information conserved in a 3D human cerebral organoid, super-resolution imaging is necessary. Classical point scanning light microscopy in super-resolution is challenging and restricted to small regions (single synapses), making it virtually impossible to scan entire organoids or even distinct structures within organoids in 3D.

A solution to this issue came into reach with the development of light-sheet fluorescence expansion microscopy (LSFEM), which has enabled the analysis of extended neural circuits in super-resolution (Bürgers et al., 2019; Gao et al., 2019; Schwarz and Kubitscheck, 2021). As in standard expansion microscopy (ExM; Chen et al., 2015; Ku et al., 2016; Tillberg et al., 2016; Chozinski et al., 2016), LSFEM uses water-absorbent polymers to physically expand enzymatically treated tissue samples. Before synthesizing the expandable polymer within the fixed-tissue sample, proteins of interest are labeled with fluorescent antibodies that bind the antigen of interest. The tissue is then partially digested to allow subsequent expansion of the polymer matrix containing the fluorescent labels. As a result of expansion, fluorescent moieties spaced closer than the optical diffraction limit (∼250 nm) can be optically resolved, resulting in effective super-resolution images of organoids. Using LSFEM, we were able to rapidly image extended neuronal circuits in effective super-resolution (Bürgers et al., 2019).

Here, we present a novel brain organoid analysis pipeline, which employs LSFM and LSFEM to image entire brain organoids in 3D during different developmental stages. The methods allow us to zoom in and out on an entire single organoid from meso- to nanoscale optical resolution in order to obtain a comprehensive view of both the brain organoid architecture and subcellular aspects. Careful sample preparation allowed the conservation of fluorescent proteins, which are frequently used to label e.g. neuronal subpopulations. Using effective super-resolution imaging, the finest details of neuronal network parameters within a larger context can be depicted succeeding, for the first time, in identifying clusters of synaptically connected neurons in the context of an entire cleared mature brain organoid.

Fixed brain organoids represent opaque structures. We developed specific sample preparation techniques and imaging approaches to exploit the inherent information optimally. To assess the applicability of our approach to different organoid protocols, we have used two different protocols in our study: a classic ‘organoid’ protocol adapted from Lancaster et al. (2013), and a protocol adapted from Pasça et al. (2015), who term their structures ‘spheroids’. However, the distinction into ‘organoids’ and ‘spheroids’ is somewhat arbitrary, because both are based on self-organized architecture formation. For that reason, we uniformly refer to these structures as ‘organoids’. We have applied our LSFEM approach to both protocols, which we refer to as protocol I and II (for details, see Materials and Methods), to ensure broad applicability.

Following fixation, we permeabilized the organoid tissue using CHAPS instead of Triton X, following a procedure suggested by Zhao and coworkers (Zhao et al., 2020). This enabled especially the preservation of the activity of autofluorescent proteins and staining of complete large organoids using commercial antibodies with high efficiency (see Fig. S1). We usually employ DNA staining to mark all cell nuclei and also include markers for specific cell types or cellular structures highlighting specific types of neurons or subcellular neuronal structures. To allow a light microscopic analysis of these samples, optical clearing is mandatory (Fig. 1A). Therefore, the first steps of ExM include the addition of bifunctional linker coupling protein residues, thereby creating a polyacrylamide gel within the organoid, which keeps the fluorescently labeled structures in place. Digestion by proteinase K renders the sample transparent. Keeping the sample in a buffered aqueous immersion medium such as PBS induces a 1.5-fold expansion (Fig. 1B), whereas the exchange of the medium to bi-distilled water results in an approximately 4-fold physical expansion (Fig. 1C). In either state, the sample can be analyzed using LSFM (Fig. 1D). We performed extensive controls demonstrating that the overall organoid structure was not distorted during the clearing process (Fig. S2). Labeling the tight junctions (ZO1) and the progenitor cells (SOX2) using respective antibodies yielded identical structures. In this comparison it was evident that the clearing process greatly improved contrast and overall image quality. The opaque nature of uncleared organoids prevented imaging of structures located inside the organoids.

Mesoscale imaging of cleared brain organoids allowed us to follow organoid development from generation until maturation. The key steps of the pipeline for organoid sample preparation and imaging are shown in Fig. 1E, and the various approaches to imaging are summarized in Table 1. Cerebral organoids can reach up to several millimeters in diameter. To cover such dimensions, light-sheet imaging was performed using a low magnification objective lens (10×) with a relatively low NA (0.3) in order to achieve a large field of view for covering the complete organoid with a limited number of mosaic tiles (Fig. 1F,G).

We employed mixed organoids containing 10% EGFP-expressing cells [90% induced pluripotent stem cells (iPSCs) mixed with 10% doxycycline-inducible EGFP-labeled iPSCs from the same genetic background prior to seeding them] and with all cell nuclei labeled by Hoechst. Use of our optimized sample preparation protocol (Bürgers et al., 2019; Stockhausen et al., 2020) allowed us to maintain the fluorescence of autofluorescent proteins, e.g. EGFP, by first optimizing the permeabilization of the sample and then the digestion conditions and the expansion buffer, which avoided the necessity of antibody staining in this case. One month after the generation of the organoid, the distribution of the EGFP cells was found to be not uniform throughout the volume, as could be concluded from 2D images (Fig. 2A-D). Rather, the labeled cells tended to form large clusters. Interestingly, such clusters of EGFP-labeled cells still existed after 14 months. We suspect that the non-uniform distribution of EGFP-positive cells was due to local proliferation of subpopulations of these cells (Fig. 2E-G).

The functional architecture of brain organoids extends over lengths ranging from more than a centimeter to nanometers, and we became interested in devising an approach that enables recording across these ranges. Cleared and 1.5-fold-expanded complete brain organoids were imaged using a 10× objective lens (Figs 2 and 3A,B). The use of 4-fold expansion, a 1.1 NA objective lens for imaging, an axial step size of 0.3 µm and subsequent deconvolution allowed visualization of selected sample regions at the 100 nm scale (Fig. 3C). Numerous cell somata and neurites with extensions up to hundreds of micrometers were visible and traceable. Close examination of magnified sample regions revealed numerous spine-like structures, suggesting advanced differentiation and formation of neuronal connections (Fig. 3D-F).

Our approach enables detailed insights into the cellular architecture of brain organoids, because numerous structural parameters may be evaluated. This also allows quantitative analyses, as we demonstrate here using neuroepithelial rosettes as example. Structural analyses can be improved even more when exploiting the fact that antibodies generally penetrate expanded tissue particularly well (Edwards et al., 2020).

These structures typically appear in cerebral organoids, forming ventricular zone (VZ)-like areas, the apical surface of which can be labeled with antibodies to the tight junction protein zonula occludens protein 1 (ZO1, also known as TJP1). ZO1 immunofluorescence thus enables the evaluation of the topology of neuroepithelial rosettes and the ventricle-like space they could enclose (Fig. 4A,B). Fig. 4C shows the cropped apical surface across the whole organoid, demonstrating that these structures exhibit a large variation in size and shape. Some were closed structures, i.e. enclosing a ventricle-like lumen; others appeared to be relatively flat with complex geometry and a sheet-like topology. Therefore, the measurement of the apical surface of the neuroepithelium appeared to be an appropriate parameter for their characterization. The surface size distribution of the segmented structures of the organoid shown in Fig. 4A is displayed in Fig. 4C. Data like that shown in Fig. 4C allowed the quantification of numerous parameters (Table 1), e.g. characterizing the apical surface topology (Fig. 4D). Mesoscale parameters were quantified by imaging three different 2-month-old organoids.

The combination of antibody staining and LSFEM allows straightforward detection of neural subpopulations. As an example, we used co-labeling with antibodies to SOX2 and TBR2 (EOMES), which can, for example, delineate oRG cells, a distinct population of neuronal progenitors in the developing human brain, which are located in the outer subventricular zone (oSVZ). These cells are essential for neurogenesis and expansion of the human cortex (Bershteyn et al., 2017). ORGs are characterized by expressing SOX2, but not TBR2, in the outer region of the VZ (see green arrows in Fig. 5A-C).

However, only identifying SOX2-positive cells and their position is not sufficient for an unequivocal delineation and quantification of oRGs (Pollen et al., 2019). Recent studies have revealed that the vast majority of cells expressing HOPX also expressed the radial glia marker SOX2, but not the intermediate progenitor marker TBR2 (Bhaduri et al., 2020; Pollen et al., 2019). Thus, HOPX is considered a useful marker for oRGs, and we applied it in 3-month-old organoids. In Fig. 6, we show several examples for the identification of oRGs in different samples. Co-staining with an antibody to N-cadherin enabled the delineation of the apical surface of the neuroepithelium and thus a spatial relationship of the HOPX-positive cells to the histoarchitecture (Fig. 6C,D).

A key parameter in neurogenesis during human brain development is the orientation of mitotic cleavage planes of neuronal progenitors with regard to the apical surface of the neuroepithelium (Fig. 7A). The orientation of the mitotic spindle modulates the orientation of the cleavage plane and, therefore, the position of the two daughter cells. The correct spindle orientation during the early stages of human corticogenesis is vital for accomplishing the right ratio between symmetric and asymmetric cell divisions. Most actively dividing neuronal progenitors during early cortical development exhibit a horizontal orientation (i.e. an angle of 0 to 30°) in relation to the ventricular surface cleavage plane, which leads to expansion of the cortical progenitor pool via symmetric cell division. Vertical (60 to 90°) and oblique (30 to 60°) mitotic cleavage planes start to become more predominant immediately before neurogenesis (LaMonica et al., 2013; Yingling et al., 2008). This asymmetric mode of cell division results in the generation of two different daughter cells and leads to an increase in neuronal differentiation. Several studies have shown that a disrupted orientation of the mitotic cleavage planes led to abnormal corticogenesis, reflected in various developmental phenotypes. Lancaster and colleagues performed one of the first landmark studies showing a crucial role of the shift in mitotic cleavage plane orientation and its effect on the development of microcephaly (Lancaster et al., 2013). In two more-recent studies using Miller-Dieker syndrome patient-derived organoids in comparison with controls, they have successfully shown that, under disease conditions, there is a clear shift from the vertical to the horizontal plane of cell division of radial glia without a significant increase in oblique planes, causing early neurogenesis and smaller size of patient-derived 3D cortical cultures (Bershteyn et al., 2017; Iefremova et al., 2017).

We found the combination of DNA staining by Hoechst and immunofluorescence staining for ZO1 and the neural progenitor marker SOX2 suitable for delineating the orientation of cleavage planes with respect to the surface of the neuroepithelium (Fig. 7B,C). Importantly, our image analyses revealed that it is not always possible to assess the orientation of the cleavage plane with regard to the apical surface using 2D projections alone. As shown in Fig. 7D, the xy-section suggested a horizontal orientation of the cleavage plane of the mitotic cell with regard to the apical surface, whereas the xz-section of the very same cell nucleus indicated a vertical orientation (Fig. 7E). The evaluation of the true orientation requires the full 3D view (Fig. 7F). This was performed for three different 2-month-old brain organoids. The results are shown in Fig. 7G. From the performed analysis, it can be concluded that the majority (53%) of diving cells along the ventricle-like structure exhibit a vertical cleavage plane orientation, which is in line with previously reported data from early stages of cortical organoid cultures (Bershteyn et al., 2017; Iefremova et al., 2017).

The existence of functional neuronal connections in cerebral organoids has been reported previously (Paşca et al., 2015; Quadrato et al., 2017; Giandomenico et al., 2019). However, there are no reported studies using whole cleared organoids showing the co-existence of both pre- and postsynaptic proteins at synapses in a spatial manner (for a review of current organoid imaging, see Brémond Martin et al., 2021). We used again chimeric organoids generated with 10% EGFP-containing cells for such structures. Fig. 8 shows a 14-month-old organoid with a diameter of ∼1.5 cm (Fig. 8A). Regions near the surface of the organoid contained numerous cells with neuronal morphology and long neurites (Fig. 8B). LSFEM with antibodies to the presynaptic protein synapsin1 (SYN1) and the postsynaptic protein HOMER1 revealed diffraction-limited spots within distances of 150±61 nm (mean±s.d.; n=26), with a median value of 130 nm of each other along with neural projections (Fig. 8C-F). In this very same organoid, we determined the number of spines on three different dendrite segments. For that purpose, the expanded organoid was labeled by antibodies against MAP2. We determined spine densities of 0.06, 0.036 and 0.045 spines per µm² on dendritic segments with lengths of 269, 187 and 175 µm lengths, respectively. This means that we could find one dendritic spine for every 212 µm² on average. Considering the above demonstration of the existence of pre- and postsynaptic structures, these values show that the 14-month-old organoid exhibited well-developed and abundant neuronal connections. Thus, using LSFEM spine structures (Fig. 3D-F) and synaptic structures (Fig. 8D,E) can be detected in complete 3D organoids without the need for physical cutting of the sample.

Cerebral organoids are opaque 3D structures with a size in the range of a few millimeters. Opacity and volume make light microscopic analysis difficult because the opacity impedes classical imaging with sufficient contrast and the organoid size prohibits the use of high-resolution optical microscopy. The latter requires objectives with a high NA, which generally have very short working distances. Classical imaging approaches using organoid slices miss the natural 3D features of such complex samples.

Here, we demonstrate the potential of LSFM combined with a clearing and expansion of organoids by a factor of 1.5- to 4-fold, which has also recently been used for the examination of mouse brain sections (LSFEM; Bürgers et al., 2019; Gao et al., 2019; Stockhausen et al., 2020). We demonstrate that this approach allows detailed analysis of organoids up to 15 mm diameter.

The preservation of the fluorescence of autofluorescent proteins during the expansion procedure enabled us to follow the location and fate of selected cell types during the course of development over a time period of up to 14 months. The use of a fraction of cells expressing fluorescent proteins when generating the organoids, in addition to nuclear staining, allowed a deeper analysis of the location and distribution of cells groups within the volume. Thereby, we confirmed the well-known observation that the development of organoids varied largely due to the batch-to-batch variability (Quadrato et al., 2016, 2017; Qian et al., 2019; Velasco et al., 2019).

The crucial details of neuronal connectivity occur on length ranges of ∼100 nm. Such small structures can optically only be resolved using super-resolution light microscopy. We have already demonstrated that LSFEM can yield effective super-resolution laterally down to less than 100 nm and axially down to 300 nm. Thereby, individual synaptic connections can be identified (Bürgers et al., 2019). Achieving this requires clearing followed by an expansion of the sample by a factor of four and subsequent high-resolution imaging. Further improvement of image resolution may be achieved by deconvolution techniques. This approach allowed us to detect single spine-like structures in a 5-month-old brain organoid. Usually, imaging at such a high resolution is feasible but not applicable for large sections of organoids due to the immense amount of data produced. An organoid of 1 mm³ original size would yield an object of 64 mm³ size after expansion. Imaging that structure at a resolution of 100 nm laterally and 300 nm axially at 16 bit, considering the Nyquist theorem, would yield a dataset of 340 TB. Current data processing workstations are at or beyond their computational limit when handling such amounts of data. Therefore, imaging at a mesoscopic scale is used to locate specific regions, for which super-resolution data can be obtained. Notably, this allows examining a single specimen on length ranges from 1 cm to 100 nm, corresponding to five orders of magnitude.

A hallmark of developing cerebral organoids is the generation of rosette-forming neuroepithelial structures with an apical-basal polarity. The space enclosed by these structures has been shown to form a ventricle-like system, which can span across large volumes of the organoid (Di Lullo and Kriegstein, 2017). As these areas correspond to a pendant of the neurogenic ventricular zone in vivo, their qualitative and quantitative assessment is of great importance. The study of these structures was straightforward using LSFEM because they are immediately visible in the 3D data we acquired. The lumina of VZs could well be visualized and analyzed using additional staining against ZO1 or N-cadherin.

A further determinant of brain organoid structure are oRGs. These cells can be stained using antibodies against TBR2, SOX2 and HOPX (Pollen et al., 2019). The identification of oRGs is generally challenging because of the low abundance of this cell type in brain organoids. Therefore, it is even more problematic if only 2D sections are employed because the mapped volume is quite small. The use of the complete 3D image stack clearly improved the chance of detecting this important cell type. For the success of these experiments and to achieve labeling with high contrast it was especially important to employ a new permeabilization strategy before labeling, namely to use CHAPS instead of Triton X (Zhao et al., 2020). This approach yielded a much better penetration of antibodies to their target sites inside the mature, rather large, organoids.

The orientation of cleavage planes of dividing cells with regard to the VZ surface is important for the growth properties of a brain organoid. We noted that evaluating the 3D data revealed the correct orientation of cleavage planes with regard to the lumen surface – use of 3D data avoided possible misinterpretations compared with using only 2D optical sections. Furthermore, we noticed that the use of SOX2 is sufficient to visualize the orientation of cleavage planes with our method, without the need for specialized antibodies such as phospho-vimentin.

Clearly, LSFEM cannot compete with electron microscopy in terms of resolution and detection of fine structural details of specimen. However, in contrast to electron microscopy, LSFEM is compatible with multicolor fluorescence imaging, thus enabling molecular contrast for diverse neuronal populations and nanoscale resolution within a single large-tissue preparation. We performed triple color staining of a brain organoid using EGFP and pre- and postsynaptic markers in order to identify synapses unambiguously. Using LSFEM, we succeeded for the first time to detect both presynaptic synapsin 1 and postsynaptic HOMER1 within distances of 150 nm of each other along with neural projections, clearly proving the existence of synapses in a 14-month-old organoid by light microscopic means. The capability of imaging across many length ranges allows, for example, to count and analyze the spatial distribution of synapses in certain brain areas. In principle, our technique provides a sufficient optical resolution to allow the identification of different kinds of spines, e.g. mushroom spines or stubby spines, in 3D. In the future, for such experiments, it would be helpful to label dendritic structures using MAP2.

Clearly, LSFEM has specific limitations. An obvious one is the anisotropic optical resolution. Generally, the axial resolution is about 3-fold lower than the lateral resolution. Approaches to reduce that difference exist, but they do not function at the ultimate resolution limit. Furthermore, it is not simple to examine a specimen of arbitrary size with very high resolution. One reason is that the penetration time of antibodies into a spatially extended specimen increases nonlinearly with sample thickness. Thus, it is not trivial to achieve homogeneous staining of large samples. Also, very thick specimens present a problem, because the working distance of high resolution imaging objective lenses is limited.

In summary, we demonstrated that the combination of LSFM and ExM – LFSEM – allows imaging of mature brain organoids in toto down to synaptic resolution in a single imaging session, when they are combined with careful specimen preparation that preserves autofluorescent proteins and optimization of imaging results using deconvolution. The situation may be further improved when exploiting the fact that antibodies penetrate expanded tissue particularly well (Edwards et al., 2020). Thus, LSFEM is optimally suited for the analysis of brain organoid development.

Human IPSCs used in these experiments are cell line iLB-C-133bm-s4 (hPSCreg name UKBi013-A) and cell line iLB-C-133bm-s4 AAVS1-GFP (hPSCreg name UKBi013-A-1). Cells were maintained in six-well tissue culture plates (Nunc) coated with 1% Geltrex membrane matrix (Thermo Fisher Scientific) in StemMACS iPS-Brew (Miltenyi Biotec) with regular passaging using EDTA/PBS. Cultures were tested for mycoplasma contamination and were maintained mycoplasma free.

Organoids were generated along previously established protocol with slight modifications (Iefremova et al., 2017). In brief: on day 0 of organoid culture, iPSCs were dissociated into single-cell suspension using TrypLE Express (Gibco), followed by plating 18,000 cells in each well of an ultra-low-attachment round-bottom 96-well plate in StemFlex medium (Gibco) with 50 µM of ROCK inhibitor Y-27632 (10 µM, Hiss Diagnostics). In order to visualize the EGFP-positive cells, iPSCs from the same genetic background with and without doxycycline-inducible EGFP construct were mixed in the ratio 10/90, respectively. Organoids were fed every other day for up to 5 days with StemFlex media and then transferred to low-adhesion 6 cm plates in the neural induction medium containing 50% Neurobasal, 50% DMEM/F12 supplemented with 1× N-2 supplement, B-27 supplement (all Gibco) and glucose (0.4 mg/ml, Carl Roth). Neural induction medium was supplemented with 1% Minimum Essential Medium, non-essential amino acids solution (MEM-NEAA, Gibco), 1% GlutaMax, LDN-193189 (180 nM, Axon Medchem), A8301 (500 nM, Miltenyi Biotec) and XAV939 (10 µg/ml, Enzo Life Sciences) before medium change. After 5-6 days, the medium was changed to the neural differentiation medium containing 50% Neurobasal, 50% DMEM/F12 supplemented with 1× N-2 supplement, B-27 supplement, glucose (0.4 mg/ml), cAMP (0.15 μg/ml, Sigma-Aldrich), 1% MEM-NEAA, 1% GlutaMax. Over the next 5 days, organoids were embedded in Matrigel (Corning Life Sciences) and further cultured on a cell culture shaker with a medium change every 2-4 days until the day the cultures were fixed for further analysis. From day 35 onwards the medium was changed to 50% Neurobasal, 50% DMEM/F12 supplemented with 1× N-2 supplement, 1× B-27 supplement, glucose (0.4 mg/ml), cAMP (0.15 μg/ml), insulin (Sigma-Aldrich), 1% Matrigel, 20 ng/ml brain-derived neurotrophic factor (BDNF; CellGS) and 10 ng/ml glial cell-derived neurotrophic factor (GDNF).

Here, we used a modified protocol from Paşca et al. (2015). In brief, iPSCs were dissociated into single cell suspension with StemPro Accutase (Gibco) and organoid formation was performed by transferring 1.5×10⁶ iPSCs (5000 cells/microwell) into AggreWell 800 plates (Stemcell Technologies) in medium [50% DMEM-F12 GlutaMax, 50% Neurobasal, 1:100 B-27, 1:200 N-2, 1:200 MEM-NEAA, 1 mM L-Glutamine, 1:1000 β-mercaptoethanol (all Gibco), 10 µg/ml insulin (Sigma-Aldrich)] supplemented with the two SMAD pathway inhibitors dorsomorphin (1 µM, Sigma-Aldrich) and SB-431542 (10 µM, AxonMedChem), as well as with the ROCK inhibitor Y-27632 (10 µM, Hiss Diagnostics). As described above, iPSCs from the same genetic background with and without doxycycline-inducible eGFP construct were mixed in the ratio 10/90, respectively. For the first 5 days, the medium without ROCK inhibitor was changed daily. Afterwards, the organoids were transferred into a CERO tube (OLS OMNI Life Science) and cultivated in a rotating CERO table-top bioreactor (CERO 3D bioreactor, OLS OMNI Life Science). From day 5 to day 12, organoids were fed every other day. On day 12, medium containing bFGF (10 ng/ml, Biotechne) instead of SMAD inhibitors was used for 4 days. From day 16 on, organoids were maintained in unsupplemented medium with medium changes every other day.

In order to generate mixed 3D cortical organoids containing EGFP cells according to protocol I or II, we used iPSCs carrying a doxycycline-inducible EGFP cassette, which was knocked into the AAVS1 locus as previously described (Qian et al., 2014; Peitz et al., 2020). IPSCs were dissociated to a single cell suspension with StemPro Accutase (Gibco). Then, 90% of unlabeled iPSCs were gently and thoroughly mixed with 10% EGFP-labeled iPSCs from the same genetic background and seeded into AggreWell plates as described above. Doxycycline (1 µg/ml, Sigma-Aldrich) was added continuously with every medium change from day 0 (day of mixing) onwards. Except for the doxycycline treatment, the 3D cultures were maintained under the same conditions as iPSC-derived cultures containing 100% of unlabeled cells.

Here, we give an overview of the methods for specimen preparation, expansion and microscopy. Detailed experimental procedures for both short and extended protocols, and the solutions required, can be found in the supplementary Materials and Methods.

Immunohistochemistry was optimized from standard protocols. All used chemicals are summarized in Table S1. Details of the procedure are given in the supplementary Materials and Methods. In brief, the fixed 3D cultures were first permeabilized using CHAPS in the permeabilization buffer (1× PBS, 0.5% CHAPS) on a shaker at 37°C. The time varied depending on the size of the sample, e.g. 1 h for a 1-month-old sample. After permeabilization, samples were washed three times with 1× PBS at room temperature (RT). To prevent unspecific binding of the primary antibody, the samples were incubated with blocking buffer (1× PBS, 5% normal goat serum, 0.3% Triton X-100, 0.02% sodium azide) on a shaker overnight (ON) at RT. After blocking, the organoids were incubated ON in blocking buffer with the primary antibody (see Table S2) on a shaker at 4°C. The following day, slices were washed at RT in blocking buffer three times for 30 min and incubated ON in secondary antibody (see Table S3) on a shaker at 4°C. For nuclear staining, all samples were stained using Hoechst 33342 (H3570, Invitrogen).

The expansion microscopy protocol was adopted from Chozinski et al. (2016). Details of the procedure are given in the supplementary Materials and Methods. In short, the immunostained organoids were incubated with 2 mM methylacrylic acid-NHS linker for 24 h on a shaker at RT. After washing three times in PBS, the organoids were incubated for 16 h in the monomer solution (8.6% sodium acrylate, 2.5% acrylamide, 0.15% N,N′- methylenebisacrylamide and 11.7% NaCl in 1× PBS) on a shaker at 4°C.

The gelling solution was prepared by adding 4-hydroxy-TEMPO (0.01%), TEMED (0.2%) and ammonium persulfate (0.2%) to fresh monomer solution. During gelling, the organoids were placed in a 24-well plate on ice to avoid early polymerization. After applying the gelling solution, samples were put on a shaker at 4°C for 5 min and then transferred to the gelling chamber, followed by 3 h incubation at 37°C. After the gel formation, the samples were incubated at 37°C in the digestion buffer (50 mM Tris, 1 mM EDTA, 0.5% Triton X-100, 0.8 M guanidine HCl, and 16 U∕ml of proteinase K; pH 8.0), exchanging the buffer every 24 h. In general, a 2-month-old organoid takes about two complete days to be completely digested. After digestion, the buffer was removed and the samples were washed three times with PBS.

For light-sheet microscopy we used a custom-built setup. In short, for fluorescence excitation, four fiber-coupled lasers emitting at 405, 488, 561 and 638 nm (Hübner Photonics) were employed. The horizontally scanned light sheet was generated by a galvanometer system with silver-coated mirrors. The adjustment of the beam waist position within the sample chamber was realized by relay optics mounted on a linear precision stage. The beam waist in the object plane was adjusted to a 1/e² diameter of 6.5±0.02 µm for the 405 nm, 7.3±0.02 µm for the 488 nm, 7.0±0.02 µm for the 561 nm and 8.3±0.02 µm for the 638 nm laser lines. For illumination we used a Mitutoyo 10× NA 0.28 air objective. Our custom-designed sample chamber featured an illumination window formed by a conventional 24×24 mm coverslip with a thickness of 0.17 mm. The sample was observed from the top using different objective lenses (Table 2). The sample was mounted on a coverslip, which could be moved in three spatial directions by motorized micro-translation stages. In some experiments an optional 1.5× magnification (Nikon) was used. We used a sCMOS camera (2048×2048 pixels, pixel size 6.5 μm, Orca Flash 4.0 V2, Hamamatsu Photonics K.K.) for data acquisition in global shutter mode. All electronic components were controlled by a custom-written LabView program.

Mesoscopic imaging and analysis yield information on the topology of complete organoids, which is helpful, for example, to analyze batch-to-batch differences.

For imaging, the digested specimen was fixed on a coverslip with poly-L-lysine to avoid movements during the measurement. Then, the coverslip was inserted into the sample holder and placed into the sample chamber filled with PBS solution, which resulted in an expansion by a factor of 1.5.

Before image acquisition, a visual inspection of the sample was performed to verify a successful sample preparation. Then, the samples were analyzed using LSFM employing a 10× water immersion (WI) objective lens with an NA of 0.3 and an effective field of view of 998 μm². The achieved real and effective optical resolutions are given in Table 2. Owing to the large size of the sample, imaging in a mosaic fashion was needed in order to image the whole organoid.

For imaging at the microscopic scale, the organoids were prepared as described above, yielding a transparent 1.5-fold expanded specimen. Now, however, they were examined with a high-resolution long-distance objective with an NA of 1.1 enabling an optical resolution of about 0.3 µm laterally and 1.1 µm axially. This resulted in effective resolutions of 0.2 and 0.7 µm, respectively, when considering the sample expansion (Table 2). Thus, the lateral resolution increased by a factor of ∼4 and the axial resolution by a factor of ∼15 compared with mesoscopic imaging. This made structural characterization possible at the cellular length range. Imaging of complete organoids at this resolution would produce about 500 GB data per 1 mm³ and per channel, which would require high-end image processing workstations for analysis. Therefore, imaging of complete organoids at this resolution is often not advisable, although it is principally possible. Rather, certain regions of interest (ROIs) should be selected in the mesoscale data for subsequent analysis at the microscopic scale.

For imaging at the nanoscopic scale, organoids were prepared as described above, yielding a transparent 1.5-fold expanded specimen. Then, the buffer solution into which the sample was placed was replaced by bi-distilled water. This led to an ∼4-fold expansion compared with the original sample size. Such samples were examined with a high-resolution long-distance objective (NA 1.1) enabling an optical resolution of about 0.3 µm laterally and 1.1 µm axially. This yielded effective super-resolution of 0.1 and 0.3 µm, respectively, when considering the sample expansion (Table 2), which made structural characterization possible at the subcellular length range. The resolution can further be improved by 3D image deconvolution. When imaging, the axial step size must be adjusted such that deconvolution can optimally be performed.

We processed 3D stacks of raw 16-bit images using custom-written MATLAB scripts, which allowed parallel data processing (Gonzalez et al., 2009). In a first step, the intensity histograms were adjusted to normalize brightness and contrast throughout the complete dataset.

Complete 3D representations of the samples were possible after several 3D datasets were stitched together using Fiji (Schindelin et al., 2012) and the stitching plugin of Preibisch et al. (2009). In order to optimize the stitching process, especially when datasets exceeded the available RAM of the workstation, the process was performed in two steps. First, substacks of the 3D datasets were created using a Fiji script. Each substack contained about 15% of the information located in the center of the full stack. Secondly, each substack was stitched to its respective neighboring substack yielding the best overlap in terms of the cross-correlation measure. Based on the localization information of each substack after stitching, the full 3D stacks were stitched.

A final step to improve the contrast throughout the 3D data was performed after stitching, to compensate for possible intensity variations of the sample in the axial direction. To this end, a histogram equalization was performed in every image plane of the stitched dataset. For calculation of z-projections, the maximum intensity projection algorithm of Fiji was used.

As outlined in the Results section, selected image stacks were spatially deconvolved using Huygens (Professional version 21.10, Scientific Volume Imaging). Deconvolution was performed using theoretical point spread functions (PSF), based on microscopic parameters, or a measured PSF determined by analysis of fluorescent microbeads embedded in 1% agarose gel. The classical maximum likelihood estimation algorithm was used, and a signal-to-noise ratio value between 12 and 20 for a maximum number of iterations between 60 and 100 were selected.

The 3D representation of the data was achieved using the Surpass view in Imaris (Version 9.7.2, Bitplane). Data processing was performed on a workstation equipped with two Intel Xeon Platinum 8160 CPU (2.1 GHz, 24 cores), 512 GB memory, and an Nvidia Quadro P5000 GPU (16 GB GDDR5X) running under Windows 10 Pro.

We gratefully acknowledge expert support by the micromechanical workshop of the Institute of Physical and Theoretical Chemistry of the University of Bonn under the guidance of Daniel Poetes.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.200439.