Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • Article
  • Published:

Automated high-speed 3D imaging of organoid cultures with multi-scale phenotypic quantification

Nature Methods volume 19, pages 881–892 (2022)Cite this article

Subjects

Abstract

Current imaging approaches limit the ability to perform multi-scale characterization of three-dimensional (3D) organotypic cultures (organoids) in large numbers. Here, we present an automated multi-scale 3D imaging platform synergizing high-density organoid cultures with rapid and live 3D single-objective light-sheet imaging. It is composed of disposable microfabricated organoid culture chips, termed JeWells, with embedded optical components and a laser beam-steering unit coupled to a commercial inverted microscope. It permits streamlining organoid culture and high-content 3D imaging on a single user-friendly instrument with minimal manipulations and a throughput of 300 organoids per hour. We demonstrate that the large number of 3D stacks that can be collected via our platform allows training deep learning-based algorithms to quantify morphogenetic organizations of organoids at multi-scales, ranging from the subcellular scale to the whole organoid level. We validated the versatility and robustness of our approach on intestine, hepatic, neuroectoderm organoids and oncospheres.

This is a preview of subscription content, access via your institution

Access options

Buy this article

  • Purchase on Springer Link
  • Instant access to full article PDF
Buy now

Prices may be subject to local taxes which are calculated during checkout

Fig. 1: Working principles.
Fig. 2: Matching JeWell chips to different types of organoid.
Fig. 3: Subcellular quantification of cell proliferation in oncospheres.
Fig. 4: Multi-scale analysis of neuroectoderm organoids.
Fig. 5: Efficient rare events detection by multi-magnification imaging pipeline.
Fig. 6: Correlative imaging of live versus transcription factors expression.

Similar content being viewed by others

Data availability

Due to the size of the dataset, the raw images of the organoids acquired in this study as well as the training datasets for the CNNs will be available upon request. We will also provide free access to JeWell chips for testing by a Material Transfer Agreement. Source data are provided with this paper.

Code availability

All the Neural networks used in this paper are publicly available and have been used without any modifications.

References

  1. Kim, J., Koo, B. K. & Knoblich, J. A. Human organoids: model systems for human biology and medicine. Nat. Rev. Mol. Cell Biol. https://doi.org/10.1038/s41580-020-0259-3 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  2. Takebe, T. & Wells, J. M. Organoids by design. Science 364, 956–959 (2019).

    Article  CAS  Google Scholar 

  3. Kratochvil, M. J. et al. Engineered materials for organoid systems. Nat. Rev. Mater. 4, 606–622 (2019).

    Article  CAS  Google Scholar 

  4. Rossi, G., Manfrin, A. & Lutolf, M. P. Progress and potential in organoid research. Nat. Rev. Genet. 19, 671–687 (2018).

    Article  CAS  Google Scholar 

  5. O’Connell, L. & Winter, D. C. Organoids: past learning and future directions. Stem. Cells Dev. 29, 281–289 (2020).

    Article  Google Scholar 

  6. Vives, J. & Batlle-Morera, L. The challenge of developing human 3D organoids into medicines. Stem Cell Res. Ther. https://doi.org/10.1186/s13287-020-1586-1 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  7. Busslinger, G. A. et al. The potential and challenges of patient-derived organoids in guiding the multimodality treatment of upper gastrointestinal malignancies. Open Biol. https://doi.org/10.1098/rsob.190274 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  8. Lukonin, I. et al. Phenotypic landscape of intestinal organoid regeneration. Nature https://doi.org/10.1038/s41586-020-2776-9 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  9. Renner, H. et al. A fully automated high-throughput workflow for 3D-based chemical screening in human midbrain organoids. eLife https://doi.org/10.7554/eLife.52904 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  10. Bock, C. et al. The Organoid Cell Atlas. Nat. Biotechnol. https://doi.org/10.1038/s41587-020-00762-x (2021).

    Article  PubMed  Google Scholar 

  11. Rios, A. C. & Clevers, H. Imaging organoids: a bright future ahead. Nat. Methods https://doi.org/10.1038/nmeth.4537 (2018).

    Article  PubMed  Google Scholar 

  12. Dekkers, J. F. et al. High-resolution 3D imaging of fixed and cleared organoids. Nat. Protoc. 14, 1756–1771 (2019).

    Article  CAS  Google Scholar 

  13. Lukonin, I., Zinner, M. & Liberali, P. Organoids in image-based phenotypic chemical screens. Exp. Mol. Med. https://doi.org/10.1038/s12276-021-00641-8 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

  14. Wan, Y., McDole, K. & Keller, P. J. Light-sheet microscopy and its potential for understanding developmental processes. Annu. Rev. Cell Dev. Biol. https://doi.org/10.1146/annurev-cellbio-100818-125311 (2019).

    Article  PubMed  Google Scholar 

  15. Yang, Q. et al. Cell fate coordinates mechano-osmotic forces in intestinal crypt formation. Nat. Cell Biol. https://doi.org/10.1038/s41556-021-00700-2 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

  16. Serra, D. et al. Self-organization and symmetry breaking in intestinal organoid development. Nature https://doi.org/10.1038/s41586-019-1146-y (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  17. Eismann, B. et al. Automated 3D light-sheet screening with high spatiotemporal resolution reveals mitotic phenotypes. J. Cell Sci. https://doi.org/10.1242/jcs.245043 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  18. Voleti, V. et al. Real-time volumetric microscopy of in vivo dynamics and large-scale samples with SCAPE 2.0. Nat. Methods https://doi.org/10.1038/s41592-019-0579-4 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  19. Yang, B. et al. Epi-illumination SPIM for volumetric imaging with high spatial-temporal resolution. Nat. Methods https://doi.org/10.1038/s41592-019-0401-3 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  20. Maioli, V. et al. Time-lapse 3-D measurements of a glucose biosensor in multicellular spheroids by light sheet fluorescence microscopy in commercial 96-well plates. Sci. Rep. https://doi.org/10.1038/srep37777 (2016).

    Article  PubMed  PubMed Central  Google Scholar 

  21. Galland, R. et al. 3D high-and super-resolution imaging using single-objective SPIM. Nat. Methods https://doi.org/10.1038/nmeth.3402 (2015).

    Article  PubMed  Google Scholar 

  22. Sorrentino, G. et al. Mechano-modulatory synthetic niches for liver organoid derivation. Nat. Commun. https://doi.org/10.1038/s41467-020-17161-0 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  23. Fedorova, V. et al. Differentiation of neural rosettes from human pluripotent stem cells in vitro is sequentially regulated on a molecular level and accomplished by the mechanism reminiscent of secondary neurulation. Stem Cell Res. https://doi.org/10.1016/j.scr.2019.101563 (2019).

    Article  PubMed  Google Scholar 

  24. Yoshida, S., Miwa, H., Kawachi, T., Kume, S. & Takahashi, K. Generation of intestinal organoids derived from human pluripotent stem cells for drug testing. Sci. Rep. https://doi.org/10.1038/s41598-020-63151-z (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  25. Weigert, M., Schmidt, U., Haase, R., Sugawara, K. & Myers, G. Star-convex polyhedra for 3D object detection and segmentation in microscopy. In Proc. 2020 IEEE Winter Conference on Applications of Computer Vision, WACV 2020 3655–3662 (IEEE, 2019); https://doi.org/10.1109/WACV45572.2020.9093435

  26. Sun, X. & Kaufman, P. D. Ki-67: more than a proliferation marker. Chromosoma https://doi.org/10.1007/s00412-018-0659-8 (2018).

    Article  PubMed  PubMed Central  Google Scholar 

  27. Lancaster, M. A. et al. Cerebral organoids model human brain development and microcephaly. Nature https://doi.org/10.1038/nature12517 (2013).

    Article  PubMed  PubMed Central  Google Scholar 

  28. Hříbková, H., Grabiec, M., Klemová, D., Slaninová, I. & Sun, Y. M. Calcium signaling mediates five types of cell morphological changes to form neural rosettes. J. Cell Sci. 131, jcs206896 (2018).

  29. Meinhardt, A. et al. 3D reconstitution of the patterned neural tube from embryonic stem cells. Stem Cell Rep. https://doi.org/10.1016/j.stemcr.2014.09.020 (2014).

    Article  Google Scholar 

  30. Chandrasekaran, A. et al. Comparison of 2D and 3D neural induction methods for the generation of neural progenitor cells from human induced pluripotent stem cells. Stem Cell Res. https://doi.org/10.1016/j.scr.2017.10.010 (2017).

    Article  PubMed  Google Scholar 

  31. Ronneberger, O., Fischer, P., Brox, T. U-Net: Convolutional Networks for Biomedical Image Segmentation. In: Navab, N., Hornegger, J., Wells, W., Frangi, A. (eds) Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015. MICCAI 2015. Lecture Notes in Computer Science(), vol 9351. Springer, Cham. https://doi.org/10.1007/978-3-319-24574-4_28

  32. Huang, G., Liu, Z., Van Der Maaten, L. & Weinberger, K. Q. Densely connected convolutional networks. In Proc. 30th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, 21–26 July, Honolulu, HI, USA (IEEE, 2017); https://doi.org/10.1109/CVPR.2017.243

  33. Karzbrun, E., Kshirsagar, A., Cohen, S. R., Hanna, J. H. & Reiner, O. Human brain organoids on a chip reveal the physics of folding. Nat. Phys. https://doi.org/10.1038/s41567-018-0046-7 (2018).

    Article  PubMed  PubMed Central  Google Scholar 

  34. Driehuis, E. et al. Pancreatic cancer organoids recapitulate disease and allow personalized drug screening. Proc. Natl Acad. Sci. USA https://doi.org/10.1073/pnas.1911273116 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  35. Beccari, L. et al. Multi-axial self-organization properties of mouse embryonic stem cells into gastruloids. Nature https://doi.org/10.1038/s41586-018-0578-0 (2018).

    Article  PubMed  Google Scholar 

  36. Kumar, S. V. et al. Kidney micro-organoids in suspension culture as a scalable source of human pluripotent stem cell-derived kidney cells. Dev. https://doi.org/10.1242/dev.172361 (2019).

    Article  Google Scholar 

  37. Perche, F. & Torchilin, V. P. Cancer cell spheroids as a model to evaluate chemotherapy protocols. Cancer Biol. Ther. https://doi.org/10.4161/cbt.21353 (2012).

    Article  PubMed  PubMed Central  Google Scholar 

  38. Seglen, P. O. Preparation of isolated rat liver cells. Methods Cell. Biol. https://doi.org/10.1016/S0091-679X(08)61797-5 (1976).

    Article  PubMed  Google Scholar 

  39. Quelennec, E. et al. Generation of two induced pluripotent stem cell lines IMAGINi004-A and IMAGINi005-A from healthy donors. Stem Cell Res. 48, 101959 (2020).

    Article  CAS  Google Scholar 

  40. Alessandri, K. et al. A 3D printed microfluidic device for production of functionalized hydrogel microcapsules for culture and differentiation of human neuronal stem cells (hNSC). Lab Chip 16, 1593–1604 (2016).

    Article  CAS  Google Scholar 

  41. Sofroniew, N. et al. napari/napari: 0.4.9rc1. Zenodo https://doi.org/10.5281/ZENODO.4905201 (2021).

  42. Legland, D., Arganda-Carreras, I. & Andrey, P. MorphoLibJ: integrated library and plugins for mathematical morphology with ImageJ. Bioinformatics https://doi.org/10.1093/bioinformatics/btw413 (2016).

    Article  PubMed  Google Scholar 

  43. Kingma, D. P. & Ba, J. L. Adam: a method for stochastic optimization. In Proc. 3rd International Conference on Learning Representations, ICLR 2015 - Conference Track Proceedings (2015).

  44. Fedorov, A. et al. 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magn. Reson. Imaging https://doi.org/10.1016/j.mri.2012.05.001 (2012).

    Article  PubMed  PubMed Central  Google Scholar 

Download references

Acknowledgements

The research is supported by the CALIPSO project supported by the National Research Foundation (NRF), Prime Minister’s Office, Singapore under its Campus for Research Excellence and Technological Enterprise program. V.V. acknowledges the support of NRF investigator grant nos. NRF-NRFI2018-07, MOE Tier 3 MOE2016-T3-1-005, MechanoBiology Institute (MBI) seed funding and ANR ADGastrulo. J.B.S. and R.G. are supported by the Labex BRAIN, the ‘program investissement d’avenir’ (grant no. The grant for the Agence Nationale de la Recherche (ANR) ANR-10-IDEX-03-02), the ANR soLIVE (grant no. ANR16-CE11-0015) and France BioImaging infrastructure (grant no. ANR-10-INSB-04). T.D. holds a CNRS PhD fellowship. We thank F. Saltel (BaRiTon laboratory) for providing the HEP3B-H2B-GFP stable cell line. We thank P. Cohen, K. Alessandri and A. Leonard (TreeFrog Therapeutics) for providing the encapsulated stem cell cysts. We acknowledge the kind gift of Lifeact/H2D Esc from O. Reiner (Weizmann) and C. Butler for their discussions. A.B. and G.G. acknowledge support from MBI core funding. We all thank A. Wong and D. Pitta de Araujo for help editing the manuscript.

Author information

Authors and Affiliations

  1. Mechanobiology Institute, National University of Singapore, Singapore, Singapore

    Anne Beghin, Gianluca Grenci, Geetika Sahni, Su Guo, Harini Rajendiran, Saburnisha Binte Mohamad Raffi, Richard de Mets, Hui Ting Ong, Anais Monet, Vidhyalakshmi Acharya & Virgile Viasnoff

  2. Immunology Translational Research Programme, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore

    Anne Beghin

  3. Department of Microbiology and Immunology, National University of Singapore, Singapore, Singapore

    Anne Beghin

  4. Biomedical Engineering Department, National University of Singapore, Singapore, Singapore

    Gianluca Grenci

  5. University of Bordeaux, CNRS, Interdisciplinary Institute for Neuroscience, IINS, UMR 5297, Bordeaux, France

    Tom Delaire, Xareni Galindo, Florian Levet, Remi Galland & Jean-Baptiste Sibarita

  6. QuantaCell, Pessac, France

    Damien Blanc & Victor Racine

  7. University Bordeaux, CNRS, INSERM, Bordeaux Imaging Center, BIC, UAR 3420, US 4, Bordeaux, France

    Florian Levet

  8. Department of Biological Sciences, National University of Singapore, Singapore, Singapore

    Virgile Viasnoff

  9. IRL 3639 CNRS, Singapore, Singapore

    Virgile Viasnoff

Authors
  1. Anne Beghin
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Gianluca Grenci
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Geetika Sahni
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Su Guo
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Harini Rajendiran
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Tom Delaire
    View author publications

    You can also search for this author in PubMed Google Scholar

  7. Saburnisha Binte Mohamad Raffi
    View author publications

    You can also search for this author in PubMed Google Scholar

  8. Damien Blanc
    View author publications

    You can also search for this author in PubMed Google Scholar

  9. Richard de Mets
    View author publications

    You can also search for this author in PubMed Google Scholar

  10. Hui Ting Ong
    View author publications

    You can also search for this author in PubMed Google Scholar

  11. Xareni Galindo
    View author publications

    You can also search for this author in PubMed Google Scholar

  12. Anais Monet
    View author publications

    You can also search for this author in PubMed Google Scholar

  13. Vidhyalakshmi Acharya
    View author publications

    You can also search for this author in PubMed Google Scholar

  14. Victor Racine
    View author publications

    You can also search for this author in PubMed Google Scholar

  15. Florian Levet
    View author publications

    You can also search for this author in PubMed Google Scholar

  16. Remi Galland
    View author publications

    You can also search for this author in PubMed Google Scholar

  17. Jean-Baptiste Sibarita
    View author publications

    You can also search for this author in PubMed Google Scholar

  18. Virgile Viasnoff
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

A.B. performed and conceived the experiments and wrote the paper. G.G. conceived and fabricated the microchips. G.S. performed the experiments with stem cells. S.G. performed the experiments with cancer cells. H.R. performed RT–qPCR and some quantifications. T.D. developed the automated acquisition for high-content imaging. S.B.M.R. fabricated the chips. D.B., R.M., X.G., A.M., V.R. and F.L. performed the artificial intelligence image analysis. H.T.O. performed the image analysis. V.A. performed the experiments on hepatocytes. R.G. conceived the imaging system and some experiments and participated to the paper writing. J.-B.S. conceived the technology, wrote the paper, supervised and funded the work. V.V. conceived the experiments, wrote the paper, and supervised and funded the work.

Corresponding authors

Correspondence to Anne Beghin, Jean-Baptiste Sibarita or Virgile Viasnoff.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Methods thanks Bon-Kyoung Koo, Prisca Liberali and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Madhura Mukhopadhyay was the primary editor on this article and managed its editorial process and peer review in collaboration with the rest of the editorial team.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended Data Fig. 1 JeWell chip fabrication and multi-well design.

a. Schematic of the fabrication steps of the JeWell chips. 1- The PDMS working mold displaying positive truncated shaped pyramids is flipped and placed on top of a flat cut of PDMS. 2- UV curable resin (NOA73 in this case) is poured next to one edge and by capillarity action fills the cavity (that is the space in between the truncated pyramids) (the red arrow shows the direction of flow). 3- After complete filling of the cavity, NOA73 is cured by UV flood exposure. 4- The PDMS working mold is peeled-off, leaving the cured NOA73 on the flat PDMS cut, which serves as a handling substrate. 5- Excess of cured NOA73 from the 4 edges is trimmed away. 6- Au thin film is deposited onto the cured NOA73 (by sputter-coating). 7- A standard coverslip is coated with thin layer of NOA73, which is then partially cured to become solid while retaining adhesive properties; the Au-coated film is flipped and pressed on the pre-cured NOA73 and finally exposed to UV flood-exposure (from the coverslip side) to fully cured the thin adhesive NOA73 layer. 8- The flat PDMS cut is Peeled-off, which leaves the Au coated JeWell structured film adherent to the coverslip and with the top of the pyramidal micro-wells open. The final JeWell chip is composed of a reflective structured NOA73 film on a glass coverslip with open access at the top. b. Examples of mounting vessels for the JeWell chips from single Petri dish to 6-well plate and 96-well plate.

Extended Data Fig. 2 JeWells dimensions and designs available.

a. Shapes and dimensions of different JeWell chips that are currently produced. b. Examples of chips with JeWells of different sizes and shapes. c. Si wet etching produces cavities in the bulk of the wafer accordingly to the geometry of the opening in the masking/protecting layer and constrained by the crystal planes orientation. The latter constrain is an effect of the dependence of etching velocity from the directions in the Si crystal structure. 45° mirrors can emerge only along edges aligned to {110} crystal family planes when a Si wafer with its surface aligned to a (100) plane is used. For edges aligned in different directions, planes with higher miller indexes are exposed and etched with faster rates than those in the principal direction, that is (100), (110) or (111). The planes that result from edges not aligned with the latter have inclinations comprised between 45° and 54.7° (this last is the angle between (111) and (100) planes). As a consequence, squared or rectangular etching windows with their edges aligned to the (110) directions will be etched down along the {110} planes and preserve their original geometries, other than for the inclined translation, which results in a restriction of the initial dimension.

Extended Data Fig. 3 Initial number of cells and growth organoids rate estimation in JeWells.

a. Image stitching of the chips using MetaMorph Scan Slide at 20x magnification. The stitched image was subdivided in 8 quadrants and each individual wells cropped according to the method described in material and method, independently of their shapes. b. The U-Net convolutional neural network (CNN) was trained to segment individual cells and cell clusters with an accuracy of detection of 93%. c. Left: Image of a whole JeWell chip (~1cm2) with color coding corresponding to the number of cells in each JeWell (N=775) (white square corresponds to 96 JeWells). Middle: Distribution of the number of cells per JeWell (221± 67) (mean±STD). The color codes of the bar match the heat map of the JeWells. The radial gradient in cell numbers reflects the localization of the addition of cells with the pipette. Since the initial density of cells can be monitored in each well this experiment shows that the influence of the initial number of cells can be performed in 1 chip. Right: The initial number of cells depends on the cell seeding density and sedimentation time (N=100 JeWells). The histogram shows that the chip contains at least 96 wells for any average cell number between 150 and 300 cells per well with a maximum spread of ± 15 cells. More homogeneous distribution in the chip could be achieved with stirring during the seeding phase. The evolution of cells in JeWells with time was performed on N=15 JeWells for each time point. d. Cells are left to aggregate for 6 hours and the size of each aggregate is assessed by its projected area in bright field trans-illumination mode (20x). We repeatedly obtained a reproducible distribution of aggregate sizes within 6 hours using the same cell seeding density (0.5M/ml) (N=3 batches, 96 organoids measured for each). Bars show the average value and whiskers the min max of the distribution. e. The Neuroectoderm organoids grew inside the JeWells over days (here 10 days) and their growth rate was monitored. The JeWell occupancy was typically around 90% (N=221 JeWells). Reference marks were made by onsite laser engraving, using the soSPIM excitation beam-steering device. It permits the repositioning of the plate on different acquisition systems or after manipulations of the plate outside the platform (medium exchange, ECM addition…). A specific ID is attributed to each organoid for correlative microscopy. All scale bars represent 50 µm (N=100 JeWells for the panels c and d).

Source data

Extended Data Fig. 4 soSPIM microscope.

a. Schematic representation of the soSPIM principle showing a sample holder comprising 45° micro-mirrored cavities (JeWells) deposited on a glass coverslip that can be mounted in 35 mm petri dishes or along multi-wells (6 to 96) plate format (blue), an excitation beam-steering unit mounted on a conventional inverted microscope (green), and the soSPIM software to control and synchronize the optical elements to perform 3D imaging (red box). b. The light-sheet position is controlled by the laser beam steering1 using two galvanometric mirrors (GM) and a focal tunable lens (TL), all conjugated by relay lenses (RL) to the objective back focal plane. A motorized iris (MI) allows adjusting the laser beam diameter to control the light-sheet dimensions (thickness and length). Upper right: cartoon illustrating the 3D acquisition principle of the soSPIM system. The light-sheet position onto the 45° mirror is synchronized with the objective axial position and the tunable lens to maintain the excitation and detection plane superposed, and the light-sheet thinnest part at the sample position for 3D imaging. c. Light-sheet thickness (FWHM) as function of its length (FWHM) obtained for different iris opening within the beam steering unit and two different objectives (red dots for 60x WI 1.27 NA, black triangles for 20x Air 0.75NA). The light-sheet parameters were computed from the images of the light-sheet after reflection onto a 45° mirror into a JeWell filled with a fluorescent solution. The light-sheet thickness corresponds to the minimum width of the beam perpendicular to its propagation direction as fitted by a Gaussian function. The light-sheet length corresponds to the width of the beam along its propagation direction as fitted by a Gaussian beam. The black line represents the theoretical relationship between the thickness and the length of a focused Gaussian beam at 488 nm.

Extended Data Fig. 5 High content imaging acquisition workflow.

a. Schematic representation of the relation between the XY mirror position and the light-sheet illumination depth Z created by a laser beam reflected onto the 45° mirrors of a JeWell. In absence of correction, a spatial displacement of the device perpendicular to the 45° mirror axis (X-axis on the schema) leads to an equivalent displacement of the light-sheet along the Z-axis, whereas the imaged plane remains at the same position, corrupting the image quality. Therefore, precise repositioning of the JeWells as well as lateral drifts need to be corrected to ensure optimal 3D imaging during long-term and HCS-like multi-position acquisitions. b. Principle of the cross-correlation-based repositioning system. Left: Comparison between a reference position and a translated position to correct, and the overlay before (Magenta) and after (Cyan) correction. Middle: Cross-correlation matrix between the two brightfield images represented on the left panel before correction. The computation of the cross-correlation maximum position using a center of mass approach with a sub-pixel accuracy returns the drift vector to be corrected to ensure a proper JeWell positioning, and therefore 3D soSPIM imaging, according to the reference position. Right: Repositioning accuracy obtained from the computation of the drift between consecutive 3D acquisitions, before and after a random displacement of the stage. The acquired images were binned to evaluate the effect of the image size reduction in the repositioning error and on the computational time. It resulted in a repositioning error of 45 nm and a computation time of 188 ± 13 ms for a binning of 4, compared to 30.2 nm and 1760 ± 40 ms respectively with a binning 1x. c. FlowChart representation of the high-content 3D imaging acquisition process: A brightfield preview of the full chip containing well-arrayed JeWells is performed using the Scan Slide module of MetaMorph software (Step 1). The automatic detection of all the JeWells based on Fourier transform filtering and automatic thresholding (Step 2) and their interactive selections to select the JeWells to image (Step 3) are handled through a home-made plugin integrated into MetaMorph with the soSPIM beam-steering control (yellow box). The acquisition process is then steered using the Multi-Dimensional Acquisition (MDA) module of MetaMorph (Step 4) in which journals for home-made routines are executed enabling JeWells accurate positioning by auto-correlation at a user-defined frequency and 3D soSPIM imaging (yellow rectangles). The acquisition process ends once all the positions have been acquired at every time-point. d. Left: Scan Slide bright-field image of a JeWell chip containing 299 JeWells with their automatic detection (ROIs) and user-assisted selection (green=selected, red=not selected). The overall Scan Slide acquisition, followed by the 299 JeWells detection and selection processes took 25 min.

Extended Data Fig. 6 Imaging performances of soSPIM.

a. We used the same microscope body (Nikon Eclipse), camera (Hamamatsu Orca Flash 4), laser launch (Oxxius laser diodes 405, 488, 630 nm) and objective (60x WI 1.2NA) to compare the imaging quality of the soSPIM and the spinning-disc confocal techniques. Out of focus signal rejection was superior for spinning-disc between 0 and 25 µm above the coverslip. Deeper than 25 µm, the light-sheet illumination and collection schemes surpassed the ones of the spinning-disc more impacted by light scattering, leading to an increase of collected signal. This is an expected characteristic when comparing confocal approaches to light-sheet microscopy. b. Comparison of the total laser power received by an organoid during the acquisition of 70 Z-planes stacks. For equivalent signal to background ratio (S/B), the sample is exposed to light up to 700 time less for soSPIM than for spinning-disc. The laser power was measured at the objective’s output. c. Left: depth color-coded maximum intensity projection of selected time-points of a 18h time-lapse soSPIM acquisition performed on 4 different HEP3B spheroids stably expressing H2B-GFP. 3D stacks of 70 planes (1 µm plane distance) were acquired every 90 seconds for 721 time-points (total duration of 1,080 min). Right: normalized mean intensity of the spheroids signal over the whole acquisition duration (dots) and mean curves (black line) fitted with a single exponential decay (gray line) illustrating the low to no bleaching of the signal along the time-course of the acquisition (half-life = 11078 ± 900 min). The difference of the mean intensity on a volume encompassing the spheroid and a volume outside of the spheroid were computed for each time-points and positions. The signal was then normalized according to its mean value for all time-points. All these effects illustrate that prolonged live 3D imaging is possible using the soSPIM acquisition platform.

Extended Data Fig. 7 Deep Learning-based image analysis.

a. Schematic description of the deep learning-based analysis pipeline for mitosis and apoptosis detection using YOLOv5 CNN published by ZeroCostDL4Mic with default settings, using a NVIDIA Quadro RTX6000 24GB. The ground truth consisted of a training dataset of mitosis and apoptosis manually pinpointing by drawing bounding boxes around the central plane of the apoptosis (Blue) and of the mitosis (Red). A total of 344 mitosis and 1,123 apoptosis were manually picked for the training, augmented to a final training set of 4,128 mitosis and 13,476 apoptosis (See Methods). We used 90% of this dataset for the training and 10% for validation of the network. The AI detection results (Prediction) showed an accuracy of 89% as compared to the ground truth. The prediction was obtained by processing each z image plane independently, and post-processed to obtain the 3D assignment of all events (See Methods). b. Illustration of the CNN architecture used for the whole organoid shape classification, corresponding to a 3D adaptation of the Densenet121 model2. It resulted in a 99% accuracy as compared to the ground truth for the classification of organoids in ‘B-shape’ and ‘O-shape’ (representative images). c. Illustration of Features Pyramid Network (FPN) architecture with successive Resblocks in the backbone (C0-C5) of the bottom-up pathway (blue). The remaining features maps (P0-P5) were generated by a combination of convolution operation applied to the feature maps of the bottom up pathway and up-sampling transformation applied on each level of the pyramid (green). The probability score and segmentation output (streaks outline) was given by applying softmax layer on P0.

Extended Data Fig. 8 Reproducibility and absence of analysis spatial bias.

a:left: representative image of a neuroectoderm organoid after 8 days of differentiation in JeWells with the deep learning 3D detection of dead/mitosis cells (left, YoloV5) and of nuclei (right, Stardist). N= 3 batches of 40 organoids each. Right : at D8, the total number of cells per organoids are relatively similar for Batch 1 and 2 seeded at the same cell density (1.105 cells/ml) and reduced by half in Batch3 seeded with 0.5.105 cells/ml. The bar represents the average value, The box the std and the whiskers the min-max.Organoids (Batch1 and 2) filling the JeWells or smaller organoids (Batch3) hence display similar growth dynamics. b : left: representative single plane raw image of a HCT116 tumoroid grown in JeWells (yellow stripes: direction of the light sheet illumination). Right: Integrative z-projection of the binarized segmentation the nuclei in the organoid using deep learning 3D segmentation (Stardist). Each pixel value hence represent the value of the integrated nuclear volume over the z axis for each x,y coordinate. Right: Representative lateral profiles from 3 different organoids of nuclei volume along the x axis (yellow line in top panel). We measured a homogeneous distribution of nuclei along the light sheet direction despite the gradient in optical contrast. The homogeneous detection results from the training of Stardist with nuclei with different contrasts covering the whole gradient range in raw images. N=20 organoids c. Comparison of the same ‘B shape’ Neuroectoderm organoid imaged from the left and right side of the JeWell. Both single sided illuminations revealed the same left right asymmetry of the distribution of sox2 positive cells within the organoids. It demonstrates that a single side illumination sufficed to perform quantitative measurement of transcription factor gradients within the organoids. Multi-side (up to 4) illumination is possible with the soSPIM but was not implemented in an automatically manner. d. 3D segmentation comparison between left-side and right-side illumination images. Left: Single plane intensity overlay between left-side (cyan) and right-side (magenta) illumination. Middle: Cross-sections overlays of 3D nuclei segmentation computed from left-side (cyan) and right-side (magenta) illumination data. Objects are the maximum intensity projections of the nuclei intersecting the cross-sections in (x,y), (x,z) and (y,z) directions. Right: Quantification of the difference between the absolute number of detected nucleii in the two illumination directions (green, N = 32), and of the number of nuclei that overlaps (orange, N = 64). Box plots indicate the mean (bar) and Std. The butterfly represents the distribution over all the organoids in each condition.

Source data

Extended Data Fig. 9 Scatter plot of the rosette streak volumes vs the average Sox2.

Expression levels per nucleus were compared for each ‘O-shape’ organoid. Each individual point represents one organoid (N=96 organoids, N=415 streaks). The populations were classified in 4 groups using gating based on the Sox2 intensity threshold established in Fig. 4b and Methods. Images display examples of organoids in each of the 4 groups. White lines represent the Deep learning-based segmentation of the streak contours. Box plots represent the distributions of the number of nuclei and mitotic indexes for the −/− and +/+ gated populations. ** : p-value = 0.006; * : p-value =0.05 were obtained by unpaired Student T-test. Grey values (687 and 2,0) represent the mean values for the entire ‘O-shape’ population of organoids. The graph displays the mean (red line), the Standard deviation (box) and the min-max (whiskers).

Extended Data Fig. 10 3D live imaging using extracellular dyes.

a. Introducing calcein as an extracellular dye, revealed the intercellular spaces and luminal cavities (marked by *) in the 3D culture (inverted LUT). The illumination comes from one side (left here). We then introduced a second no-permeant dye (Alexa 647) and used high speed imaging (12 stacks/min) to monitor the dynamics of infiltration of the dye in the 3D culture. b: RTqPCR on neuroectoderm markers for organoids grown in the JeWells (left) and in U-Bottom dish. RT-qPCR performed on 11 genes, demonstrating the expected reduction in pluripotent reporters (green) and the increase of the neuroectoderm markers (blue) (N=3–6 replicates, ~ 100-200 organoids/replicate). Statistical tests: P values were obtained from Multiple unpaired t-test to compare the relevance of the fold change of gene expression to the profile of stem cells at day=0 on the same batch. P values: ****<0.0001, ***<0.001, **<0.01, *<0.05, ns: non-significant. Exact p values can be found in Supplementary Table 2. Mean, Stdv and whole distributions are shown for each replicate of each batch. c: Representative images of rosettes for organoids grown in Jewells and in U-Bottom dish. We did not notice any significant changes in lumen organization between both cases. N-=400 organoids.

Source data

Supplementary information

Supplementary Information

Supplementary Tables 1–4.

Reporting summary

Supplementary Video 1

96 primary rat hepatocytes spheroids and z stack. Primary rat hepatocytes were seeded in the JeWell chip and left for aggregation over 2 days. They were then fixed and stained for Actin (Phalloidin, gray) and MRP2 (gold). MRP2 is an apical transporter that is recruited at the apical poles of the hepatocytes. It is a reporter for the bile canaliculi. The medial z plane is displayed for each of the 96 spheroids on the left, and a z stack of one of the 96 spheroids on the right.

Supplementary Video 2

3D segmentation of nuclei based on DAPI with StarDist. Example of the quality of the 3D segmentation of a HCT116 oncosphere using StarDist neural network. Nuclei are stained with DAPI. The StarDist predictions are the white contours. Right, zoom on the region inside the yellow square.

Supplementary Video 3

3D classification of proliferation stages using YoloV5. Example of the quality of the 3D detection of Early G1, late G1-S-G2 and mitosis cells of the same HCT116 oncosphere as Video 2 using YoloV5 neural network. The proliferating nuclei are stained with Ki67. The 3D bounding boxes are color coded: blue, Early G1 phase; yellow; late G1-S-G2 and red, mitosis.

Supplementary Video 4

Segmentation of streaks in an organoid. Organoid made from hESC in differentiation process to neuroectoderm (fixation and staining at day 8 in JeWells). 3D image stacks were acquired in soSPIM mode for Actin (Phalloidin, gold) and nuclei (DAPI, blue). Left, probabilities map for each pixel to belong to a streak, based on the Actin staining. Right, segmentation over raw images.

Supplementary Video 5

Deep Learning-based segmentation of a streak in 3D. The streak is the region encompassed within the white volume that results from the thresholding of the artificial intelligence scores shown in Video 7. Note that the detected region corresponds to the core of the rosette and does not encompass the surrounding cells. The segmentation was performed on the Actin staining images, but it corresponds to the N-cadherin positive region of the rosettes.

Supplementary Video 6

Live imaging of neuroectoderm. Live imaging of neuroectoderm differentiation stained with Alexa647 extracellular marker from days 4 to 7 (two planes out of 50 are displayed). One image every 15 min.

Supplementary Video 7

Parallelized live soSPIM acquisition of differentiating organoids. Organoids made from hESC-lifeact GFP-H2BmCherry in differentiation process to neuroectoderm in JeWells. Green, lifeact and purple, histone-H2B. 15 different organoids were acquired every 15 min. Top line for high z, middle line for median z and bottom line for low z.

Supplementary Video 8

Eight days time-lapse recording of neuroectoderm differentiation. Organoid made from hESC-lifeact GFP in differentiation process to neuroectoderm (from day 2 to day 8 in JeWells) (inverted gray LUT) (median plane of the organoid). The yellow line corresponds to a rosette-shape organization around an internal streak.

Source data

Source Data Fig. 3

Cell counts in wells.

Source Data Fig. 4

Cell counts in organoids.

Source Data Extended Data Fig. 3

Organoid sizes.

Source Data Extended Data Fig. 8

Segmentation counts.

Source Data Extended Data Fig. 10

qPCR results.

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Beghin, A., Grenci, G., Sahni, G. et al. Automated high-speed 3D imaging of organoid cultures with multi-scale phenotypic quantification. Nat Methods 19, 881–892 (2022). https://doi.org/10.1038/s41592-022-01508-0

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/s41592-022-01508-0

This article is cited by

Search

Advanced search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing