Validation of non-destructive morphology-based selection of cerebral cortical organoids by paired morphological and single-cell RNA-seq analyses

Summary

Organoids, self-organized cell aggregates, contribute significantly to developing disease models and cell-based therapies. Organoid-to-organoid variations, however, are inevitable despite the use of the latest differentiation protocols. Here, we focused on the morphology of organoids formed in a cerebral organoid differentiation culture and assessed their cellular compositions by single-cell RNA sequencing analysis. The data revealed that organoids primarily composed of non-neuronal cells, such as those from the neural crest and choroid plexus, showed unique morphological features. Moreover, we demonstrate that non-destructive morphological analysis can accurately distinguish organoids composed of cerebral cortical tissues from other cerebral tissues, thus enhancing experimental accuracy and reliability to ensure the safety of cell-based therapies.

Graphical abstract

Highlights

• •Organoids in a cerebral differentiation culture could be classified by morphology
• •Cell types in organoids could be identified by morphology and scRNA-seq
• •Organizing signals were locally activated at the early stages of differentiation
• •Non-destructive morphological selection enabled the collection of desired organoids

Ikeda et al. paired scRNA-seq and morphological analyses to identify the cellular composition of a heterogeneous collection of morphologically distinct organoids generated in cerebral organoid differentiation. This work illustrates the potential for the non-destructive morphological selection of target organoids to reduce tissue heterogeneity for research and improve the safety of cell-based therapies.

Introduction

Injury to the corticospinal tract (CST) by stroke or brain trauma leads to various disabilities, including motor dysfunction and dysphagia. Current treatments for these disorders are limited to the protection of surviving neural circuits by medical treatment and rehabilitation but often are accompanied by aftereffects such as motor paralysis (Mansour et al., 2018). Therefore, cell replacement therapy to reconstruct neural circuits represents a much-needed alternative.

Previous studies have shown fetal mouse brain tissues to extend axons along the CST in adult mice (Gaillard et al., 2007; Péron et al., 2017; Samata et al., 2020; Sano et al., 2017), thus suggesting fetal brain tissue could repair damaged neuronal circuits after injury. However, limited resources and ethical issues have been hindering the progress of the clinical use of fetal cortical tissues. Recent advances in organoid technology have enabled the in vitro production of various tissues, including the cerebral cortex, by mimicking developmental processes (Kadoshima et al., 2013; Lancaster and Knoblich, 2014; Pasca et al., 2015). Cerebral organoids display stepwise self-organized laminar formation and similar cellular composition to those in the developing brain. Moreover, they extend axons and exhibit electrophysiological activity after transplantation (Kitahara et al., 2020; Mansour et al., 2018). Recent reports demonstrated cerebral organoids integrated into the sensory cortex to form functional synaptic connections with host neurons in the mouse brain and drive behavioral changes (Jgamadze et al., 2023; Revah et al., 2022). These findings indicate that cerebral organoids have vast potential to be a viable source for cell replacement therapy to reconstruct damaged neuronal circuits.

A critical problem for the clinical application of cerebral organoids is the variability among individual organoids (Chiaradia and Lancaster, 2020). Even with sophisticated induction protocols, the emergence of non-target cells is inevitable. Morphological variations among human induced pluripotent stem cell (hiPSC)-derived organoids generated during cerebral organoid differentiation are readily detectable and may thus serve as a non-destructive visual indicator of cellular composition or organoid properties. Single-cell RNA sequencing (scRNA-seq) has enabled the analysis of cellular characteristics at single-cell resolution and revealed variations in cellular components among individual organoids (Fleck et al., 2021; Pollen et al., 2019; Quadrato et al., 2017). However, the relationships between morphological characteristics of organoids and cellular composition remain elusive. To investigate such relationships, we morphologically classified hiPSC-derived cerebral organoids and analyzed the cellular makeup of each morphological type by scRNA-seq. Through this morphology-based approach, we revealed the types of non-target cells and successfully selected cerebral cortical organoids to enhance experimental reproducibility and the safety of cell therapy.

Results

Morphological classification of cerebral organoids

We induced cerebral organoid differentiation from hiPSCs (S17), as reported previously (Figure 1A; Kitahara et al., 2020), and observed morphological variations between differentiation batches after a five- to six-week induction period, in which deep-layer neurons represent the predominant cell type generated according to our previous report (Figure 1B; Kitahara et al., 2020). Such variations were also observed in other pluripotent stem cell lines (Figure S2A). We morphologically classified the induced organoids into seven categories based on structural characteristics (Figure 1C) according to the following criteria: variant 1, organoids with rosette-like concentric layered structures throughout; variant 2, those with low transparency and no clear internal structures; variant 3, those with balloon-like cystic structures; variant 4, organoids with fibrous epithelial-like structures; variant 5, organoids with pigmentation; variant 6, transparent organoids and cyst-like internal structures; and variant 7, organoids with a transparent periphery and no clear internal structures. We grouped each organoid into one or a combination of these categories and observed substantial differences between differentiation batches and cell lines (Figures 1D and S2B), indicating variations in differentiation tendencies.

Figure 1.

Classification of organoids by morphological features

(A) Schematic illustration of the conditions used to induce cerebral organoids from iPSCs. KSR, KnockOut serum replacement; CDLC, chemically defined lipid concentrate.

(B) Bright-field images of organoids at 5–6 weeks of differentiation from 12 independent differentiation batches using the S17 iPSC line. Scale bar, 1 mm.

(C) Representative image of organoids at 5–6 weeks of differentiation displaying each defined morphological feature. Scale bar, 1 mm.

(D) Percentage of organoids with single or combined morphological features from 12 differentiation batches. For example, “variant 1 + 2” represents organoids possessing structures of variant 1 and 2. The number of organoids subjected to analysis ranged from 23 to 78 organoids per batch.

Cell type identification in cerebral organoids with each morphological feature

To identify the cellular composition of organoids designated for each category, we performed scRNA-seq analysis of nine organoid samples from different or combined categories (Figure 2A). As expected, based on the nearest-neighbor algorithm, each organoid showed a specific clustering pattern with some overlaps (Figures 2B and 2C). Next, we annotated these clusters into cell types based on canonical characteristic tissue marker genes (Figure 3A, Table S4, and Figure S3). Figure 3B shows marker gene expression profiles of each cell type. Generally, neuronal cells expressed TUBB3 and MAP2, while fibroblasts and epithelial-like cells expressed KRT8 and KRT19 (Figures 3B, S3A, and S3C). Cortical neurons and their progenitors, characterized by the expression of EMX1, a dorsal forebrain marker, were further grouped into radial glia (expressing SOX2, PAX6, and MKI67), glutamatergic neurons (expressing BCL11B, NEUROD6, SLC17A7, etc.), intermediate progenitors (EOMES), and Cajal-Retzius cells (expressing RELN). As radial glial cells are highly proliferative, they contained cells in S and G2/M phases and expressed proliferative markers (MKI67 and TOP2A; Figures S3B and S3C). Outer radial glia (co-expressing HOPX, TNC, and EMX1) were not detected, consistent with outer radial glia appearing at a later stage of development (upper layer producing phase, 10 weeks of organoid culture in our protocol). GABAergic neurons expressing glutamate decarboxylase (GAD2) consisted of progenitor cells (early and late GABAergic neuron progenitors expressing ASCL1, DLX1, and DLX2) and mature GABAergic neurons (expressing DLX5 and DLX6). In the GABAergic neuron cluster, SST expression was observed, thus suggesting that some cells have already differentiated into this subtype of GABAergic neurons (Figures 3B and S3A). HOX genes, including HOXA2 and HOXB2, characterized caudal neurons. Non-neuronal cells, such as cells from the choroid plexus, fibroblasts, neural crest cells, and endothelial cells, were also identified. Choroid plexus cells expressed TTR and an ion channel TRPM3. CNS fibroblasts expressing COL1A1 were divided into meningeal cells (MGP) and choroid plexus (AQP1, IGF2, and IGFBP7). Neural crest cells consist of cells in different developmental stages: neural plate border-like stage (early neural crest, expressing ZIC1 and MSX1), epithelial-mesenchymal-transition (EMT) stage (middle neural crest, expressing IDs and EMT-related genes including TWIST1, PRRX1, SHOX2, GPC3, and SNAI1), and migration stage (late neural crest, expressing SOX10) (Simões-Costa and Bronner, 2015). Vascular endothelial cells were characterized by PECAM1, KDR (VEGFR-2), and CDH5 (VE-Cadherin).

Figure 2.

Single-cell RNA sequencing analysis of organoids with different morphological features

(A) Bright-field images of organoid samples (6 weeks of differentiation) for scRNA-seq analysis. Image headers show the morphological features observed in each organoid based on the classification in Figure 1C. Scale bar, 1 mm.

(B) UMAP plot of scRNA-seq data showing transcriptomic clustering of single cells derived from 9 organoid samples described in Figure 2A. Each color represents an individual organoid sample.

(C) UMAP plot of scRNA-seq data displayed separately for each sample. The data are the same as in Figure 2B.

Figure 3.

Identification of cell types induced by cerebral organoid differentiation system

(A) UMAP plot of scRNA-seq data after clustering by the shared nearest neighbor algorithm. The data are the same as in Figure 2B, with each color representing a different cluster.

(B) Dot plot of marker genes in each cluster. The color scale represents the gene expression level in each cluster. Normalized UMI counts of each gene were averaged, centered, and scaled (transformed to Z score). The circle size represents the percentage of cells expressing each marker gene in each cluster. See also Figure S3.

(C) Ratio of cell types contained in each organoid sample as determined by clustering in scRNA-seq. “cortical neuron” represents three clusters: radial glia, glutamatergic neuron/intermediate progenitor, and Cajal-Retzius cells. “GABAergic neuron” represents three clusters: GABAergic neuron early progenitor, GABAergic neuron late progenitor, and GABAergic neuron. “neural crest” represents three clusters: neural crest (early, middle, and late).

As a result of these analyses, the cellular components of each organoid were revealed (Figure 3C). For example, organoids with variant 1 morphology were primarily composed of cortical neurons, whereas organoids with variant 2 morphology predominantly comprised GABAergic neurons. This result indicates that there are some relationships between morphology and cell identity.

Morphological characterization of each cell type

To examine the relationship between organoid morphology and cellular composition, tissues with specific morphology were isolated and subjected to gene expression analyses (Figure S4A). First, we performed Pearson correlation analysis after microarray analysis to examine the similarity in global gene expression of each category of organoids (Figure 4A). We found a clear correlation between morphological features and global gene expression, suggesting that cellular composition, based on organoid morphology, could be deduced. Moreover, marker genes identified by the scRNA-seq analysis were explicitly expressed in the structures of each morphological variant (Figure 4B). These marker genes are summarized in Table 1.

Figure 4.

Investigation of correlation in organoid morphology and gene expression

(A) Microarray analysis of tissues with each morphological feature. Tissues of each morphological feature were dissected and analyzed by microarray (n = 3). The color represents pairwise-correlation coefficients (Pearson).

(B) RT-qPCR analysis of organoids with each morphological feature for marker genes of each tissue type (n = 3). Statistical significance for each morphological category was calculated using one-way analysis of variance (ANOVA) with Dunnett’s multiple-comparison test; ^∗∗p < 0.01, ^∗∗∗p < 0.001, ^∗∗∗∗p < 0.0001; ns, not significant. Three organoids for each morphological group were analyzed.

(C) Immunostaining of organoids at 5–6 weeks of differentiation with each morphological feature for marker proteins of each tissue type. Red represents signals of immunostaining for each marker protein. White represents signals for DAPI staining. Scale bar, 200 μm. See also Figure S4A for high-magnification images.

Table 1.

Summary of tissue/cell types and marker genes of each morphological variants

Morphological variants	Tissues/cell types	Marker genes
variant 1	cortical tissue/glutamatergic neuron	SLC17A7, EMX1, NEUROD6
variant 2	GABAergic neuron	GAD2, DLX1, DLX2, DLX5, DLX6
variant 3/4	CNS fibroblast	COL1A1
variant 5	melanocyte	TYR
variant 7	choroid plexus	TTR
variant 8	caudal neuron	HOXA2, HOXB2, HOXA5, HOXB5

Organoid morphology, tissue, cell types, and representative marker genes are listed.

Next, we confirmed that these marker genes could be used to identify specific structures cytologically (Figures 4C and S4B). By immunofluorescence study, we found that variant 1 organoids were composed of EMX1-expressing cortical neurons. In addition, variant 2 was composed of GAD-expressing GABAergic neurons, variant 3 and 4 were composed of COL1A1-expressing fibroblasts, variant 5 was composed of TYR-expressing melanocytes, and variant 6 was composed of TTR-expressing choroid plexus. Similar expression was observed with other cell lines, suggesting that correlations between organoid morphology and marker expression are reproducible across multiple cell lines (Figure S4C).

Organizing signals are locally activated during the early stages of differentiation

Since each tissue is induced by the function of its organizer, we analyzed when and how fate determination occurs in these organoids. During development, organizers form during the early stages and induce differentiation into various tissues by secreting signaling molecules at specific time points. Thus, to evaluate whether organizing signals are locally activated, we performed immunostaining against downstream targets of organizing signals for each tissue in the organoid differentiation system (Figure S5). In a differentiation batch where differentiation into the choroid plexus (variant 6) was prominently observed, activation of the BMP-Smad1/5 pathway, a crucial signal for choroid plexus differentiation, was locally observed on day 7. Additionally, expression of Lmx1a, a roof plate region marker, was observed in the protrusion area on day 14 (Figure S5A). Subsequently, in a differentiation batch where differentiation into the basal ganglionic eminence (variant 2) was prominent, Nkx2.1 expression, a marker of the medial ganglionic eminence, was observed on days 7 and 14 (Figure S5B), earlier than the expression of GAD65, a marker for GABAergic neurons (day 35). In addition, in a differentiation batch with significant differentiation into neural crest-derived tissue (variant 3, 4, 5), Pax3/7, a marker for the neural plate border, was observed on day 3, and Sox10, a neural crest marker, was detected near Pax3/7 on day 12 (Figure S5C). In some organoids, organizer markers for multiple lineages were observed in an organoid (Figure S5D). These results suggest that the emergence of organizer-like cells during the early stages of organoid differentiation determines the fate of various organoids.

Selection of cerebral cortical organoids by morphological selection

The aforementioned results indicated that organoids containing only cortical neurons could be distinguishable by morphology. To test this, we selected organoids with variant 1 structure from three differentiation batches and analyzed their gene expression profiles by scRNA-seq (Figures 5A and 5B). Selected organoids showed similar gene expression profiles, suggesting that morphological selection reproducibly identifies organoids with identical cellular components. Furthermore, we found the entire cell population to express regional markers for cortical neurons (e.g., FOXG1 and EMX1) and subsequently differentiate into two subpopulations: radial glial cells expressing PAX6 and HES1 and cortical glutamatergic neurons expressing SLC17A7. Most glutamatergic neurons were subcortical projection neurons expressing BCL11B (CTIP2), and some were callosal projection neurons expressing SATB2. Notably, the expression of non-target markers, including GAD2 (GABAergic neuron), TTR (choroid plexus), COL1A1 (fibroblast), TYR (melanocyte), and PECAM1 (endothelial cell), was not detected (Figure 5C). Immunofluorescence study also revealed that the selected organoids were composed of rosette-forming progenitor cells and CTIP2-expressing cortical neurons without any non-target cells. (Figure 5D). These results showed that we could identify organoids with only cortical neurons in a non-destructive manner via morphological selection.

Figure 5.

Purification of cortical tissues by morphological selection

(A) Bright-field images of organoids at 6 weeks of differentiation classified as “variant 1” analyzed by scRNA-seq. Scale bar, 1 mm.

(B) UMAP plot of scRNA-seq data of organoids in Figure 5A, colored by each differentiation lot.

(C) UMAP plot of scRNA-seq data of organoids in Figure 5A, colored by marker genes. The color intensity represents the expression level. TYR and PECAM1 were excluded at the data filtering process due to their low expression levels.

(D) Immunostaining of organoids at 5 weeks of differentiation classified to the “Rosettes” category for marker proteins of each tissue type. Red represents signals of immunostaining for each marker protein. White represents signals for DAPI staining. Scale bar, 200 μm.

Discussion

Since organoids are formed by self-organization under stochastic control, there is inevitable heterogeneity between each organoid. This heterogeneity, however, is a difficult challenge for both basic research and clinical applications of organoids (Hofer and Lutolf, 2021). When used to study developmental processes, disease mechanisms, and drug responses, this heterogeneity creates noisy data, thus preventing accurate data interpretation. When using organoids for cell-based therapies, intra- and inter-batch variations in their cellular composition threaten their safety and efficacy. In this study, we revealed specific relationships between organoid morphology and cellular composition. In a non-destructive manner, we successfully identified organoids containing only cortical neurons based solely on morphology. This finding aligns with the recent work of Chiaradia et al. (2023), which established a relationship between morphological features of organoids and their cellular composition and fate. They found that organoids with complex cytoarchitecture are correlated with cortical layer formation, consistent with our observation that rosette-like structures are indicative of cortical neurons. Since morphology is a crucial indicator of tissue characteristics, it is a valuable non-invasive marker for selecting organoids.

During neurodevelopment, soluble morphogens establish the anterior-posterior and dorsal-ventral axes. BMP, sonic hedgehog, and retinoic acid signals induce dorsal, ventral, and caudal regions, respectively (Chiaradia and Lancaster, 2020; Eiraku and Sasai, 2012; Gaspard and Vanderhaeghen, 2010). Neural crests emerge during early neurogenesis from the regions adjacent to the neural plate and are defined by WNT5A expression (Ideno et al., 2022). In our experiment, one week after differentiation onset, we observed local BMP-pSMAD1/5 signaling activation, Nkx2.1 expression, and Pax3/7 expression in differentiation batches that tend to turn into choroid plexus, GABAergic neurons, and neural crest-derived tissues, respectively (Figures S5A–S5C). Previous studies also suggest morphogen-secreting organizer-like tissues generated early on during organoid formation may direct the trajectory of organoid development (Kadoshima et al., 2013; Renner et al., 2017). Therefore, the spontaneous formation of these organizer regions may trigger the formation of various tissues in organoid differentiation. Targeting these signaling pathways at early time points of differentiation may thus improve the differentiation efficacy of cerebral organoids.

In addition to signaling pathways, recent studies have highlighted the critical role of extrinsic and intrinsic extracellular matrices as triggers for activating differentiation and morphogenesis pathways (Chiaradia et al., 2023; Jain et al., 2023; Martins-Costa et al., 2023). Since the precise mechanisms underlying the self-formation of organizers in organoids remain unclear, further investigation into these mechanisms could be instrumental in improving differentiation efficiency.

Using our differentiation method, no cardiac, liver, or kidney cells were produced, thus suggesting that our protocol was specific for cerebral induction. Induced cells were cortical neurons, GABAergic interneurons, caudal neurons, choroid plexus, and neural crest-derived cells (melanocytes and fibroblasts), consistent with previous reports (Fleck et al., 2021; Pollen et al., 2019; Quadrato et al., 2017). Notably, among the induced intracranial cells, our methodology had sufficient sensitivity to detect TTR, COL1A1, and TYR as markers representative of non-target cell populations to evaluate the purity of cortical neurons to ensure the quality, efficacy, and safety of organoids. The markers identified by our study are vital for quality control of the final products for cell-based therapies.

Using single-cell analysis to correlate gene expression with organoid morphology, we examined the cellular compositions of different types of organoids and distinguished morphologically cerebral cortical organoids from ones containing non-cortical cells. This non-destructive morphology-based selection does not require special instruments and is more efficient than antibody-based methods, making it an ideal method to select candidate organoids for transplantation studies or therapies.

Experimental procedures

Maintenance culture of human pluripotent stem cells

All experiments using a human embryonic cell (hESC) line and hiPSC lines were approved by the Ethics Committee of the Center of iPS Cell Research and Application, Kyoto University. One hESC line (KhES-1) and 4 hiPSC lines (S17, 201B7, Ff-I14s03, and Ff-WJs524) were used for experiments. KhES-1 was used for experiments between passages 33 and 35. S17 is derived from peripheral blood mononuclear cells, established using Sendai virus vectors with Oct3/4, Sox2, Klf4, and LMyc (between passages 12 and 15). 201B7 is derived from fibroblast, established by retrovirus vectors with Oct3/4, Sox2, Klf4, and c-Myc (between passages 60 and 63). Ff-I14s03 and Ff-WJs524 are derived from peripheral blood mononuclear and cord blood cells, respectively, and are established using episomal vectors, as reported previously (Yoshida et al., 2023). The passage number for Ff-I14s03 was between 17 and 20 and between 10 and 13 for Ff-WJs524. All hESC and hiPSC lines were maintained on iMatrix (Nippi) in StemFit medium (Ajinomoto). For passaging (every 7 days), induced pluripotent stem cells (iPSCs) were dissociated into single cells by treatment with 0.5× TrypLE Select with 250 μM EDTA for 8 min and re-plated onto 6-well plates at a density of 1–1.5 × 10⁴ cells per well.

Differentiation of cerebral organoids

To generate cerebral organoids, we employed the differentiation method previously described by Kitahara et al. (2020) with some modifications. One day before starting differentiation, iPSCs were treated with StemFit medium without solution C supplemented and with 5 μM SB431542 (transforming growth factor β inhibitor; Tocris). 0.5× TrypLE Select with 250 μM EDTA was used to optimize cell dissociation for feeder-free culture. hiPSCs were dissociated into single cells by treatment with 0.5× TrypLE Select with 250 μM EDTA for 10 min. Cells were plated onto low-cell-adhesion-coated V-bottomed 96-well plates (PrimeSurface MS-9096V; Sumitomo Bakelite) in a differentiation medium supplemented with 50 μM Y-27632 (FUJIFILM Wako Pure Chemicals) at a density of 9,000 cells per well. The differentiation medium was DMEM/F-12 GlutaMAX (Thermo Fisher Scientific) supplemented with 20% (v/v) KnockOut serum replacement (KSR, Thermo Fisher Scientific), 5 μM SB431542, and 3 μM IWR1e (Wnt inhibitor; Calbiochem). With day 0 defined as the day that the floating culture of cell aggregates was started, half the medium was changed once every three days until day 15. On day 18, aggregated cells were transferred to 90-mm non-coated dishes (MS-1390R; Sumitomo Bakelite) and further cultured in DMEM/F-12 GlutaMAX supplemented with 1% (v/v) N-2 supplement (Thermo Fisher Scientific), 1% (v/v) chemically defined lipid concentrate (CDLC; Thermo Fisher Scientific), 0.25 μg/mL amphotericin B (Thermo Fisher Scientific), 100 U/ml penicillin, and 100 μg/mL streptomycin (Thermo Fisher Scientific) until 5–6 weeks. Organoids were cultured on an orbital shaker and maintained at 50 rpm in 20% O₂ and 5% CO₂ to eliminate the need for cell culture in 40% O₂. Complete medium changes were performed once every 3 or 4 days from day 18 to day 35. The S17 hiPSC line was used for all experiments except those described in Figures S2, S4B, and S5. Specific cell lines are listed in the “Maintenance culture of human pluripotent stem cells” section.

scRNA-seq

Organoids were dissociated into single cells with neuron dissociation solutions (FUJIFILM Wako Pure Chemical Corporation). Dissociated cells were resuspended with Hank’s balanced salt solution supplemented with 10% (v/v) KSR and 10 μM Y-27632 at a density of 1,000 cells/μL. The cell suspension was loaded onto a Chromium Next GEM Chip G (2000177 10× Genomics), targeting 3,000 cells for capture per well, and processed in the Chromium controller to obtain gel beads-in-emulsion. Libraries were generated with Chromium Next GEM Single Cell 3′ Reagent Kits v3.1 (1000121; 10× Genomics) according to the manufacturer’s protocol (CG000204 Rev C). Sequences were analyzed by NovaSeq 6000 (Illumina).

scRNA-seq data analysis

RNA-seq reads were aligned to the GRCh38 human genome reference sequence using Cell Ranger (v6.6.1) with default parameters. Unique molecular identifier (UMI) counts were analyzed using the Seurat R package (v4.1.0). Barcodes with low UMI counts (nCountRNA), low gene number (nFeature), high mitochondrial gene proportion, or abnormally large counts were filtered. Cutoff values were individualized for each library by reference to each distribution. (Figure S1 and Table S1). UMI counts were normalized for each cell by the “LogNormalize” method: divide by the total count, multiply by 10⁵, and transform to log1p. For correcting batch effects among differentiation induction samples, datasets were integrated using the harmony algorithm (v1.0) to acquire the first 50 principal components after principal-component analysis based on the 2,000 genes with the greatest cell-to-cell variation. The first 30 harmony dimensions were converted into two-dimensional data using uniform manifold approximation and projection (UMAP) (Becht et al., 2018). Cells were classified into clusters based on a shared nearest-neighbor graph with a resolution of 0.01. Differentially expressed genes were defined as those with more than 1.5-fold differential expression compared to other clusters. Cell types constituting each cluster were identified based on sets of characteristic genes according to marker genes reported in the literature.

Gene expression analysis by quantitative reverse-transcription PCR

Total RNA was obtained by RNeasy micro kit (QIAGEN) and reverse transcribed using the SuperScript III first-strand synthesis system (Thermo Fisher Scientific) according to the manufacturer’s protocols. Quantitative PCR was performed with the TaqMan gene expression master mix (Thermo Fisher Scientific) and QuantStudio (Thermo Fisher Scientific) according to the manufacturer’s instructions. Gene expression levels were normalized to GAPDH using the ΔΔCt method. Probes and primers are listed in Table S2.

Immunostaining analysis

Organoids were fixed with 4% paraformaldehyde for 30 min, dehydrated by 30% (w/v) sucrose in phosphate-buffered saline, and embedded in O.C.T. compound (Sakura Finetek). Frozen sections (16-μm thick) were produced using a cryostat (CM1850, Leica Biosystems). Double- and triple-labeled staining was performed after permeabilization with 0.3% or 2% (v/v) Triton X-100, antigen retrieval (if necessary), and blocking with Block ACE (KAC). Primary and secondary antibodies used in this study are listed in Table S3.

Microscopy and image analysis

Bright-field images of organoids in culture were taken with a digital microscope (Leica Biosystems, DMS1000). Confocal images were taken using an LSM800 confocal microscope (Zeiss) and analyzed with ZEN Blue image processing software (Zeiss).

Organoid classification

Bright-field images were captured from 17 to 78 organoids obtained from 5 cell lines and 22 production batches in total. Organoids were classified manually into 7 categories, based on their morphology according to criteria described in the Results section. Organoids were classified into 14 groups based on one or combinations of these categories. A digital microscope DMS 1000 (Leica Biosystems) was used for morphological observations.

Microarray analysis

Organoids with distinct morphological features (variant 1 to 7) were isolated by dissection (n = 3). Tissues were lysed with Buffer RLT (QIAGEN) containing 1% (v/v) 2-mercaptoethanol, and total RNA was extracted using an RNeasy micro kit (QIAGEN) according to the manufacturer’s protocol. Microarray analysis using a GeneChip array was performed by Kurabo Industries, Ltd. In brief, RNA quality was measured using a bioanalyzer (Agilent Technologies), and total RNA was synthesized by reverse transcription and amplified by in vitro transcription of second-strand cDNA templates using T7 RNA polymerase. Labeled cRNA was purified and fragmented, loaded onto a GeneChip human genome U133 plus 2.0 array (Applied Biosystems). Statistical analysis of gene expression was performed using GeneSpring GX software (Agilent Technologies). Correlation analysis was performed to assess the similarity of gene expression between samples. Pairwise-correlation coefficients (Pearson) were computed between each sample pair and represented as a heatmap by GeneSpring.

Statistics

Correlation analysis of microarray data was performed by GeneSpring. Pairwise-correlation coefficients (Pearson) were calculated between each sample pair and represented as a heatmap. Statistical analysis for quantitative reverse-transcription PCR (RT-qPCR) data was performed by Prism 9 (GraphPad Software). The significances of differences between each morphological category were determined by one-way ANOVA with Dunnett’s multiple-comparisons test for seven groups. Differences were considered statistically significant when probability values were <0.05.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Jun Takahashi (jbtaka@cira.kyoto-u.ac.jp).

Materials availability

This study did not generate new unique reagents.

Data and code availability

Raw data and images are available upon request to the corresponding author. Microarray data are available through NCBI GEO: GSE235892. RNA-seq data are available through the Japanese Genotype-phenotype Archive (JGA) for FASTQ files (JGA: hum0472) and Genomic Expression Archive (GEA) for analyzed data files (GEA: E-GEAD-865).

Acknowledgments

We thank Dr. Kelvin Hui for the critical reading of this manuscript, Ms. Haruka Asano and Ms. Hiroko Koreishi for sample preparation, and all members of the Takahashi Lab for discussions. This research was supported by a grant from the Network Program for Realization of Regenerative Medicine from the Japan Agency for Medical Research and Development (AMED) (21bm0204004h0009 to J.T.).

Author contributions

M.I. designed the study, collected and assembled the data, performed the data analysis and interpretation, and wrote the manuscript. D.D., H.E., Y.O., and M.F. collected and analyzed the data. T.K. and K.Y. performed the data analysis and interpretation. J.T. conceived and designed the study, assembled the data, carried out the data analysis and interpretation, wrote the manuscript, and approved the final manuscript.

Declaration of interests

J.T. receives a grant for collaborative research from Sumitomo Pharma Co., Ltd. A patent application partially related to this manuscript has been published as WO 2022/265086.

Published: October 10, 2024