Imaging 3D cell cultures with optical microscopy

Abstract

Three-dimensional (3D) cell cultures have gained popularity in recent years due to their ability to represent complex tissues or organs more-faithfully than conventional two-dimensional (2D) cell-culture monolayers. Advantages include more realistic cell-to-cell and cell-to-matrix interactions along with natural molecular and physical phenotypes that are desired for basic and clinical investigations. There is typically a need to characterize such 3D cell culture models with high-resolution imaging. This article reviews the application of both 2D and 3D microscopy approaches for 3D cell cultures. We first summarize the most popular optical microscopy methods that have been utilized with 3D cell cultures. We then discuss the general advantages and disadvantages of various microscopy techniques for several broad categories of investigation involving 3D cell cultures. Finally, we provide perspectives on key areas of technical need in which there are clear opportunities for innovation. Our goal is to guide microscope engineers and biomedical end users towards the most optimal imaging methods for specific investigational scenarios involving 3D cell cultures, and to identify use cases in which additional innovations in high-resolution imaging could be helpful.

Introduction

Cell culture techniques have evolved well beyond early simplistic two-dimensional (2D) monolayers to mimic nearly all features of a complex in vivo tissue environment. These new advanced models seek to replicate fully human tissues with complex architectures, multiple cell types and accurate cell-cell interactions. Furthermore, these systems support exquisite control of the cellular microenvironment with implementation of in vivo biochemical and biophysical features such as chemical gradients and peristaltic forces. The ability of these models to accurately replicate human physiology in health and disease is driving their adoption for investigational, translational, and clinical applications where these cell culture models are increasingly able to replace animal models. Despite the many advantages offered by these in vitro human model systems, the added complexity, tissue layers and size/thickness present imaging challenges often akin to that of living organisms. Just as imaging methods have revolutionized clinical diagnostics, 2D and 3D microscopy methods adapted for these more in vivo-like culture systems will enable monitoring of both cell-level behavior as well as organ-level physiology. In this article, we review the state-of-the-art instrumentation and application of 2D and 3D microscopy techniques as well as ongoing challenges in imaging advanced cell-culture systems.

We focus on four classes of 3D cell culture models that all have associated imaging challenges: advanced 2D cultures, spheroids/organoids, organ-on-a-chip, and slice cultures (Fig. 1). All 3D cell culture systems have unique strengths and weaknesses, making them optimal for different applications. Conventional 2D cultures or monolayers, in which cells grow on a flat surface, are simple, low-cost and easily imaged, but fail to recapitulate many key features of living organs ¹. Advanced 2D cultures employ cell patterning or layering to replicate a greater set of in vivo tissue behaviors, for example, cell segregation or complex stimuli responses. Examples include surface-confined gastruloids ², self-patterned cultures on biological or artificial scaffolds ^{3, 4}, and cells on micropatterned or curved substrates ⁵. These cultures can exhibit one or more imaging challenges beyond what is faced with simple monolayers, such as the existence of multiple cell layers, or cells growing on topologically complex and thick surfaces. Building on these innovative 2D cultures are cultures that mimic complex cell–cell and cell-matrix interactions in a fully three-dimensional (3D) environment ⁶. Spheroids and organoids are the most common form of 3D culture ⁷. Spheroids are typically defined as 3D cell aggregates derived from immortalized cell lines while organoids are 3D cell cultures originating from stem cells or primary tumor cells. Both spheroids and organoids can be generated using a wide range of strategies to facilitate cell-cell interactions and are often encapsulated within natural or synthetic hydrogels or supported within other complex environments e.g. hanging droplets in microcontainers ^8–11. Spheroid and organoid cultures have gained widespread acceptance as biologically relevant model systems for basic biomedical research, pharmaceutical development, and disease models. Examples include tumoroids (primary tumor organoids) ¹², embryoid bodies (pluripotent stem cell aggregates) ¹³, and organoids formed from primary cells such as the brain ¹⁴, lung ¹⁵, heart ¹⁶, liver ¹⁷, intestines ¹⁸, kidney ^{19, 20} and others. However, the surrounding light-scattering matrix, heterogeneous size and shape of the structures, and the desire to image with sub-cellular resolution at multiple focal planes create unique imaging challenges and opportunities for organoids/spheroids relative to those of simple monolayers and advanced 2D cultures.

Fig. 1. Illustration of 3D cell culture models with 2D and 3D microscopy.

In this article, four types of 3D cell culture are discussed: advanced 2D cultures, spheroids and organoids, organ-on-chip systems, and slice cultures. Two-dimensional microscopy provides a projection view (path-averaged), while 3D microscopy offers “optically sectioned” images, enabling accurate characterization of volumetric morphologies, including intricate internal structures. Example images of colon culture models are shown at the bottom: advanced 2D crypts with central stem cells slightly protruding above the surface (a projection of the 3D images captured by confocal fluorescence microscopy) ⁴, colonic organoids (imaged by epifluorescence microscopy) ³, colon-on-a-chip (detached crypts settled horizontally on a glass slide and imaged with epifluorescence microscopy) ⁴⁹, and colon slice cultures (10-μm thin section imaged with epifluorescence microscopy) ⁴⁹. (green: stem/proliferative cells; red: differentiated cells; blue: nuclei) (inspired by BioRender.com)

Microphysiological systems or organ-on-chip systems strive to culture cells under realistic physiological conditions beyond what is typically employed for static organoid/spheroid cultures ²¹. Organ-on-chip systems typically employ microfabrication technologies to build micro-scale scaffolds and channels replicating the architecture, biophysical features, and chemical microenvironment of in vivo tissues ²². These systems can incorporate complex chemical gradients, fluid flows, mechanical forces and electrical stimuli, with an ultimate goal of replicating the complex physiology of an organ subunit such as a lung alveolus, liver lobule, kidney nephron or intestinal crypt. However, this comes at an obvious cost in terms of complexity, size, and throughput. Although there are many designs for organ-on-chips, by far the most common incorporate topographically accurate extracellular matrix (ECM) scaffolds and/or perfusion chambers for nutrient supply ²³. Nearly every organ has been simulated on these systems including brain ²⁴, lung ²³, heart ²⁵, intestine ²⁶, kidney ²⁷, liver ²⁸, prostate ²⁹, blood vessels ³⁰, skin ³¹, bone ³², cartilage ³³ and more. The myriad chip designs and formats of organ-on-chips bring additional imaging demands relative to spheroids/organoids. Such constructs often require sophisticated housing elements with multiple device layers, large-sized tissues with adjacent fluidic compartments, and complex natural or synthetic scaffolding with significant opacity or scattering behavior that can be amongst the most significant challenges for high-resolution imaging. Organ-on-chip systems will continue to require innovative and flexible microscopy designs, staining methods, and analytical pipelines to enable these 3D culture systems to achieve their fullest potential for biological investigations and translational/clinical research.

Finally, slice cultures, also known as organotypic cultures, are derived from excised primary tissues, often in the form of 100- to 400-μm thick tissue sections. Assuming that fresh primary tissues are available, which can be a challenge, slice cultures can be rapidly prepared in contrast to long organ-on-chip maturation times. Slice cultures preserve the anatomical features of a living organ and incorporate all native cell types, both aspects that remain challenging for organ-on-chip systems ^34–36. Imaging of tissue slices can be challenging due to their short lifespan, as well as their unpredictable arrangement of cells and matrix that are often highly scattering and aberrating for light propagation. Accordingly, innovations in microscopy will continue to transform and broaden the utility of slice cultures.

This review focuses on the above 3D cell culture models that hold great promise for diverse applications such as developmental biology, infection biology, pharmacology, and cancer biology. Other 3D culture systems that are primarily envisioned for therapeutic applications (tissue repair and organ replacement) or bioreactor systems are not covered. As mentioned, there are unique challenges in performing optical microscopy in thick 3D cultures, some of which are summarized in the side box (Fig. 2). Although imaging live samples is preferred in many scenarios, it should be noted that for non-living (i.e. fixed) samples, a variety of optical clearing approaches have been developed to allow virtually any tissue type to be rendered highly homogeneous in terms of refractive index and therefore non-aberrating/scattering (i.e. transparent) ^37–40. Specialized clearing and imaging protocols have even been developed for 3D cultures (organoids) ⁴¹. Finally, we intentionally avoid discussing challenges related to image analysis, as this is an important but lengthy topic that deserves its own review.

Fig. 2. Major challenges for microscopy of 3D cell cultures.

To illustrate some of the basic processes that make optical microscopy challenging in 3D cell cultures, a beam of light is shown being focused to a localized spot beneath a sample surface (i.e. on the order of 200-microns deep). Living cells and tissues consist of diverse components and interfaces, such as lipid membranes, organelles, cytoplasm, cell filaments, and fluids that all exhibit differences in refractive index. These heterogeneous refractive-index distributions lead to light scattering and aberrations (i.e., changes in photon propagation paths) as light transits through the tissue. These challenges are greater in 3D cell cultures compared to 2D cultures, as 3D cultures are typically larger and have more structural complexity. Light scattering is a pseudo-random process in which light is dispersed in various directions after interacting with small refractive objects. The behavior of any single photon is difficult to predict due the stochastic nature of light scattering. The accumulation of multiple scattering events over millions of photons leads to a reduction in signal at the focus and an increase in a “blur” of undesired background light that reduces image contrast (here defined as signal to background ratio, or SBR). For relatively thin and/or transparent specimens in which this scattering can be limited to an acceptable degree, refractive-index heterogeneities can still result in significant wavefront aberrations. Wavefront aberrations, arising from the refraction of light as it passes through irregular interfaces with different refractive indices, cause the shape of the focus to be distorted and enlarged, which degrades the image quality. In the absence of such aberrations, the size and shape of the beam focus should be close to the ideal limit (i.e., the “diffraction limit”) as predicted by optical diffraction theory. As light propagates within tissue, light is absorbed exponentially as a function of depth. However, at microscopic length scales (< 1 mm deep), the effects of scattering and aberrations typically dominate over the effects of absorption. Other challenges with imaging 3D cultures include the need for uniform staining of the sample, avoiding aberrations and scattering from the sample holder/substrate (if the optical setup requires the light path to transmit through the substrate), minimizing phototoxicity and photobleaching, maintaining ideal environmental conditions for live-culture growth and imaging, and the need for a long working distance to image deeply within larger specimens, especially when using high numerical aperture (NA) objectives that typically have shorter working distances. Note that NA is related to the total range of angles at which light is focused or collected, where a higher NA enables higher spatial resolution. Higher objective magnification is often used as a proxy for higher resolution; however, it is important to note that NA and magnification are distinct parameters and that NA is ultimately the driver of spatial resolution.

Previous reviews have summarized the application of optical microscopy techniques for various biological fields ^42–48, including imaging and single-cell genomics tools for embryology ⁴², evaluation of host-pathogen interactions in infectious biology ⁴³, drug discovery in pharmacology ⁴⁴, and live-cell imaging methods for assessing tumor heterogeneity and drug mechanisms in cancer biology ⁴⁵. These reviews have mainly focused on 2D cell cultures, tissues, and animal models, and/or on various considerations such as biosensor/reporter design and data post-processing rather than on microscopy instrumentation ⁴⁵. Given the unique characteristics and challenges of high-resolution imaging of 3D cell cultures, this review first provides an overview of the most-common optical microscopy techniques employed for analyzing 3D cell cultures. We then analyze the benefits and drawbacks of these microscopy techniques for selected research topics. Finally, we provide perspectives on technical challenges and emerging methods that could have a significant impact on the field in upcoming years.

The following section outlines the major categories of 2D and 3D optical microscopy techniques that have found prevalence for the study of advanced cell cultures. We intentionally list the most common (i.e. commercially available) microscopy modalities (with micron-level resolution) below to allow for a generalized discussion of the major advantages and disadvantages facing all optical microscopy methods for diverse applications (next section). Similar tradeoffs exist for other forms of microscopy that are not discussed at length in this review, but that may be relevant and optimal for certain applications. For example, we do not discuss optical-coherence tomography (OCT) or photoacoustic microscopy (PAM) techniques, which have also been effectively used in some 3D cell culture applications ^{50, 51}, such as monitoring the morphology of large heart organoids via OCT ⁵². These techniques typically do not provide the level of sub-cellular resolution and/or multiplexed imaging capabilities that are often desired in the research scenarios covered in this article. We also do not cover ultra-high-resolution (i.e. super-resolution) optical microscopy techniques as they remain less popular for studies involving 3D cultures but can have great value for certain applications. A brief survey of some emerging microscopy approaches, which may be ideal for niche applications, will be provided in the technical perspectives section in the second half of this article.

1. Transmission light microscopy

Transmission light microscopy (TLM), such as phase-contrast and differential interference contrast microscopy, are the most utilized microscopy techniques in biological labs for 2D imaging. With TLM, light traveling through samples is partially absorbed, refracted, or subjected to phase perturbations, resulting in image contrast without the need for exogenous labels or stains ^{53, 54}. Thin and translucent specimens (e.g., monolayer cells) are readily visualized using TLM, while structures within thicker samples (greater than several hundred micrometers) may appear partially or wholly opaque due to the lack of optical sectioning (i.e., rejection or suppression of out-of-focus background signal). TLM lacks the ability to image specific molecules labeled with exogenous stains but is a cheap and simple imaging approach with many use cases. For 3D cultures, a limitation of 2D microscopy approaches like TLM and epifluorescence microscopy is that there is limited depth resolution to assess the height of structures or to create a side view of the sample.

2. Epifluorescence microscopy

In epifluorescence microscopy, a microscope objective is used to illuminate a tissue sample at a specific excitation wavelength whereby the fluorescence signal generated by that excitation light is collected through the same objective lens. In thick tissues, fluorescence is excited along the entire excitation beam path (before and after the focal plane), which can lead to significant out-of-focus background. In other words, a 2D projection view of the sample is achieved without optical sectioning ⁵⁵. However, computational deconvolution techniques can provide some degree of 3D information ⁵⁶. In comparison to TLM, an advantage of epifluorescence microscopy is that specific tissue targets can be fluorescently labeled to generate signal in relation to a non-fluorescent (unlabeled) background, which allows for high-contrast imaging of those labeled objects. This is especially true in thin tissue sections where out-of-focus background is also minimized. In comparison to 3D fluorescence imaging modalities, epifluorescence microscopy is a wide-field camera-based technique (no laser scanning or pulsed lasers required) that is relatively inexpensive and easy to use.

3. Confocal fluorescence microscopy

With thick specimens, image contrast (signal to background ratio) and imaging depth are fundamentally limited by the presence of out-of-focus light (i.e., light emitted from below and above the focal plane) as well as multiply scattered light. In order to collect only in-focus signals, confocal microscopes focus light to a small point (or to a set of points) within the tissue and image the signal (typically fluorescence) generated at that focus back through a pinhole (or pinhole array) placed at an intermediate image plane within the collection path. This pinhole allows the in-focus light to pass through, where it is recorded by a photodetector element, but rejects a large amount of the out-of-focus and multiply scattered light. Scanning the light focus enables the generation of 2D images, while stacking 2D images at various focal planes yields volumetric imaging data ⁵⁷. Confocal fluorescence microscopes are the most popular commercial imaging platforms for high-resolution 3D fluorescence imaging of cells and tissues. However, they are typically expensive and often have a limited imaging speed due to their reliance on scanning a tight focus of light (or multiple focal points simultaneously) to generate an image over time. In addition, there is a large amount of wasted light, rejected by the pinhole, which is not used for signal generation. This excess background light traveling through the tissue specimen can result in photodamage and photobleaching. Note that while commercial systems are expensive, low-cost confocal microscope systems have also been developed ^58–60.

4. Multiphoton microscopy

For standard fluorescence imaging, photons are linearly absorbed by fluorescent molecules and converted to fluorescence signal at a shifted wavelength. In multiphoton microscopy, the simultaneous absorption of multiple photons is needed to generate the fluorescence signal of interest, which can only occur when photon flux is very high, such as at the tight focus of a beam. Therefore, the generation of out-of-focus light is suppressed, and high-resolution 3D images can be generated by scanning the tight focus within the sample ⁶¹. One notable advantage of multiphoton microscopy is the use of longer wavelength excitation, which reduces scattering in biological samples and generally allows for deeper imaging. As with confocal microscopy, disadvantages include slow imaging rates and high costs. Like all 3D microscopy approaches, including confocal microscopy and light-sheet microscopy, multiphoton microscopy provides depth-resolved information, including the ability to provide a “side view” of the sample. However, as is true for nearly all microscopy approaches, the resolution in the depth direction is typically worse than the in-plane (lateral) resolution.

5. Light-sheet fluorescence microscopy

Light-sheet fluorescence microscopy (LSFM) systems illuminate samples with a thin sheet of light to selectively excite fluorophores within that sheet, which are then imaged over a relatively large 2D plane (defined by the light sheet) onto a sensitive detector array (camera) ^62–68. Similar to confocal fluorescence and multiphoton microscopy, 3D images can be produced by stacking 2D images, but the use of a light sheet (rather than a small focal point) allows for faster image acquisition by collecting light from many pixels simultaneously on a camera (improved scalability to large samples). In addition, since light-sheet microscopy typically uses different optical components and paths for illumination and collection, there is more design flexibility, such as to tailor the axial and lateral resolutions compared with other 3D microscopy approaches. Finally, light sheet microscopy is optically efficient, with little wasted light, resulting in low photodamage and bleaching of the sample ⁶⁹. However, since LSFM does not efficiently reject or suppress multiply scattered background light (compared to confocal fluorescence or multiphoton microscopy), samples must be transparent, either naturally or through optical-clearing methods. The high cost of LSFM systems is another disadvantage, but simple systems have been published and commercialized ^{70, 71}.

2D vs. 3D microscopy techniques for different research scenarios

Imaging challenges and trade-offs between various microscopy approaches are highly application dependent. Surveying all applications of optical microscopy for 3D cultures is not possible. However, there is value in exploring a subset of examples to allow for a meaningful discussion of technical considerations. Therefore, this section surveys four broad example areas where optical microscopy of 3D cell cultures has been applied: developmental biology, infection biology, pharmacology, and cancer biology. For technologists, this section provides insights into the specific questions being asked by biologist end-users and highlights where imaging methods could add value, as well as identifying technical gaps that could inspire future innovation. For biologists, this section provides tangible examples to describe the strengths and weaknesses of various microscopy approaches for a particular application, aiding in the selection of appropriate methods. This section also highlights the remaining needs/challenges that will be further discussed in the technical perspectives section.

As an introductory note, we should acknowledge that the most common use of light microscopy for any cell culture is for the quality-control process prior to any experimental investigation, wherein periodic inspections are necessary to ensure that cultures are growing healthily and with desired physical phenotypes. Quality-control inspections can often use low-throughput imaging methods if sampling a small number of specimens (on the order of tens) ^75–77 is sufficient to gauge the conditions of a larger set of cultures. However, higher-throughput imaging methods could be of value if evaluation of many cultured samples is needed ⁷⁸. Two-dimensional imaging is sufficient in many cases, such as characterizing basic morphological features using TLM without staining ^{75, 76, 79–81} or to assessing cell viability with live-dead staining and epifluorescence microscopy ^{76, 82–84}. For more complex quality control scenarios, such as evaluating cell morphology in the context of the local 3D microenvironment, volumetric imaging (optical-sectioning) methods may be valuable ^{76, 85–90}.

1. Development biology

Imaging has played a major role in investigating the growth and developmental processes of multicellular organisms. Organoids, which follow developmental pathways comparable to in vivo organogenesis or embryo development (gastruloid), are a valuable platform for studying developmental processes ^91–94. In addition to models following normal developmental processes, those undergoing disrupted development are used to study various disorders or diseases ^95–97. Advanced 2D cultures and organoids serve as an alternative to traditional model organisms that are challenging due to availability and ethical concerns ^{2, 98}. Moreover, complex organ-on-a-chip systems have been developed ^99–101, as they can provide a more realistic and detailed microenvironment for certain investigations. Such organ-on-chip cultures often utilize gradients of growth factors, and matrix or mechanical cues to guide cell growth and differentiation to enable studies of developmental processes ¹⁰². With these intricate engineered structures, assays requiring cell dissociation are typically less desirable than in situ and in vitro imaging-based analyses.

1.1. Morphogenesis

Morphogenesis is one key aspect within developmental biology, focusing on the processes of cell growth, maturation, and migration to form organs or organisms. Recent advances in gene-editing tools and computational biology have led to several non-imaging approaches, such as CRISPR-based barcodes, to track the fates of individual cells and their progeny ^103–105. Nevertheless, imaging-based methods have the advantage of providing spatial-temporal information encompassing not only cell division and differentiation but also cell movement ¹⁰⁶, death, and important sub-cellular structures such as cilia ⁹⁶.

Time-lapse volumetric imaging is often desirable for studying morphogenesis in 3D cell cultures ^{99, 103, 107, 108}. Since morphogenesis is usually a slow process, capturing image sequences at modest (e.g. 60 min - 24 hr) time intervals can be sufficient to capture relevant dynamics. However, given that the timescale of morphogenesis can span several days or even weeks, photobleaching and phototoxicity are two potential challenges. While the high-throughput imaging capabilities of LSFM may not be needed for morphogenesis studies, the ability to minimize bleaching and phototoxicity makes LSFM attractive. For example, LSFM with nuclei-labeling techniques was applied to image developing cerebral organoids for several days and to track cell lineages relevant to various parts of the brain. This provided vital insights into how cells divide and migrate to form different brain regions ¹⁰³ (Fig. 4(a)).

Fig. 4. Imaging 3D cell cultures used in diverse research domains.

(a) Developmental biology. Spatial lineage analysis is performed in cerebral organoids using time-lapse imaging with LSFM over several days. Nuclei positions and division events are monitored to elucidate the morphogenesis process. (Scale bar: 100 μm) ¹⁰³ (b) Infection Biology. The infection behavior of bacteria in intestinal organoids is observed using LSFM, helping to uncover the mechanism of bacterial translocation. ¹⁴³ (c) Pharmacology. The efficacy of a drug to mitigate radiation-therapy damage is assessed by measuring villus heights in a gut-on-a-chip model. Confocal microscopy results are shown with and without pre-treatment. (Scale bar: 100 μm) ¹⁷³ (d) Cancer Biology. The killing effects of engineered T cells is tracked in patient-derived tumor organoids using confocal microscopy. (Scale bar: 30 μm; time: hr: min) ²²⁴ (inspired by BioRender.com)

Imaging a large sample such as an entire organoid or one functional unit within an organ-on-a-chip (e.g., colon crypt, liver lobule) can facilitate a comprehensive understanding of tissue development and organization. Many microscopy devices are limited in their ability to accommodate large 3D specimens and/or to clearly visualize the deeper regions of cultures due to wavefront aberrations and light scattering. In the previous example of cerebral organoids, nuclei in the inner regions were less distinguishable for these reasons (Fig. 4 (a)). Moreover, in complex organ-on-a-chip models, the use of scaffolds and matrix-rich substrates can exacerbate optical challenges if they add refractive heterogeneities to the tissue or the imaging path. Potential techniques to overcome such issues will be discussed in the technical perspectives section.

1.2. Function of genes

There is significant interest in studying gene functions during developmental processes. Microscopy assays are typically used to provide spatial-temporal information from small numbers of gene targets, while non-imaging methods, such as CRISPR-based high-throughput knockout screening, can help to identify important genes from a large pool of candidates ^{109, 110}. However, recent advances in 2D spatial genomics/transcriptomics have allowed microscopy to be used both for high-throughput screening of large numbers of targets (often for hypothesis generation) in addition to targeted imaging of modest numbers of genes (often for hypothesis testing) ^{111, 112}.

For investigating small numbers of molecular targets, both live and end-point imaging have been utilized ^{99, 101, 113–115}. 3D microscopy is needed to resolve and quantify gene expression in individual cells. While both confocal microscopy and LSFM have been used for this purpose, the need for time-lapse imaging makes LSFM an attractive choice due to its reduced photobleaching and phototoxicity (similar to morphogenesis research). Additionally, the higher speed of LSFM can be useful for studying transient gene expressions when endogenous markers are available. In one study examining the development of intestinal organoids from single stem cells, LSFM was used to observe the spatiotemporal expressional dynamics of a stem-cell marker over several days to elucidate its key role in organ development ¹¹⁵.

While imaging entire cultures is not always needed, identifying regions of interest across large 3D cell cultures for subsequent localized interrogation is often required. This is challenging, especially for rare events in a large sample, highlighting the need for tailored multi-scale imaging workflows, potentially automated through computational pipelines, as discussed in the technical perspectives section.

2. Infection biology

Imaging is invaluable for investigating host-pathogen interactions in infection biology. Most types of 3D cell culture models discussed in this article have been employed for such purposes ^116–119. Spheroids and organoids can be useful for mimicking multiple-cell-type or organ-like environments during infections ^{120, 121}, complementing traditional 2D cell cultures that lack complexity ¹¹⁶. In addition, researchers have designed more complex organ-on-a-chip models to study organ systems ^{122, 123}, including the response of the host immune system upon infection ¹²⁴. Slice cultures preserve the primary structure and function of real organs, allowing even sub-organ regions to be studied ^125–127.

2.1. Viral infection

The investigation of viral infection in 3D cell culture models has included the analysis of viral copies ¹²⁸, targeted cell types or regions ^{126, 129–132}, as well as structural ^133–135 and gene expression ^{125, 129} changes in hosts. In certain studies, non-imaging methods have been used to evaluate viral load and changes in host-gene expression dynamics post-infection ^{125, 128, 129}. Alternatively, imaging assays have been used to identify cell types and regions susceptible to viral invasion ^{126, 129–132} and to observe architectural changes in the host following infection ^133–135.

Complex and heterogeneous relationships exist between viruses and host cells in 3D cell culture models. Confocal microscopy has often been used to quantify such relationships ^{126, 129, 130, 132–135}. Similar to multiphoton microscopy, confocal microscopy provides a high level of spatial resolution that facilitates the analysis of co-localized cells and viruses, the latter of which can range in size from several tens to several hundreds of nanometers ¹³⁶. In one study exploring viral infection in human colon organoids, confocal microscopy was used to observe the distribution of the virus and the architecture of fixed colon organoids post-infection. This approach led to the identification of a novel anti-infection mechanism involving the extrusion of infected cells ¹³³. While issues of aberration or scattering are less significant in this example, where the organoid diameter is roughly 200 μm or less, they can be more pronounced in larger 3D cultures. Tissue clearing methods could be applied to mitigate aberrations and scattering for the analysis of fixed 3D cell cultures. It should be noted that there are some cases where time-lapse imaging of live samples is essential, such as to study the behavior of viruses hijacking the cytoskeleton for transport ^{137, 138}.

2.2. Infection by bacteria and parasites

Similar to research on viral infection, studies on bacterial or parasitic load, and changes in host-gene expression, can be monitored using non-imaging approaches ^139–141. However, in the cases of bacteria and parasites, time-lapse analyses can be useful since most bacteria and protists are dynamically active when infecting their hosts ^142–145. LSFM is an attractive approach since it minimizes photobleaching/phototoxicity and its resolution is typically sufficient for both bacteria and protozoa, which are several micrometers or larger ^{146, 147}. Additionally, high imaging speeds can be valuable for capturing the movements of certain bacteria or parasites that invade rapidly, sometimes on the order of minutes ¹⁴⁸. In one study investigating the infection mechanisms of a lethal bacterium, researchers used time-lapse LSFM to observe the location of bacteria within intestinal organoids, revealing previously unknown infection pathways and offering insights into future treatment strategies ¹⁴³ (Fig. 4 (b)).

Most infection studies focus on surface-exposed layers ^{143, 144, 149} where issues such as tissue scattering and wavefront aberrations are less problematic, especially for smaller spheroids/organoids. However, for organ-on-a-chip or slice-culture models with topologically complex structures that introduce refractive heterogeneities, addressing these issues will likely become necessary ¹⁴⁴. Another challenge is spatially locating host-pathogen interactions of interest, particularly due to the large number of pathogens introduced and the substantial size of 3D cell cultures. Automated multi-scale workflows may be of value in this context, as described in the technical perspectives section. Finally, while low- to medium-throughput imaging of tens of samples has typically been performed in prior studies ^{140, 143, 149}, high-throughput imaging methods offer the potential to investigate personalized treatments or to screen for drug candidates to combat infections ^{116, 150}.

3. Pharmacology

Pharmacology studies the effects of drugs/chemicals on biological systems. Advanced 2D, spheroid and organoid cultures have been popular for drug-screening experiments ^151–153, where non-imaging population-level analyses are often sufficient, such as with microplate readers ^154–157. Another area of research aims to delve deeper into the effects of a smaller number of drug candidates, including their efficacy, toxicity, and mechanisms of action. Organ-on-a-chip models have emerged as attractive tools for such investigations. However, given their larger size and spatial complexity, optical challenges abound with ample opportunities for innovation ^158–161. Slice cultures are less commonly used due to the difficulties and costs of obtaining primary tissues ¹⁶², and the challenges in obtaining precision-cut slices with minimal variations between slices ¹⁶³. However, recent techniques, such as microdissected “cuboid” models coupled with microfluidic handling, offer some promise for studying drug effects on large numbers of primary tissue cultures ¹⁶⁴.

3.1. Efficacy / Toxicity

Drug efficacy is mainly investigated using disease models (e.g., disease-on-a-chip)¹⁶⁵, while drug toxicity is often studied using liver and kidney models, as these organs are responsible for detoxification and are often susceptible to damage ^{161, 166}. Both 2D and 3D fluorescence microscopy have been utilized in previous studies ^{156, 159–161, 167–171}. For drug effects that can be adequately represented by cell population metrics, such as viability ¹⁷², wide-field 2D epifluorescence imaging can be ideal since it is fast and simple. Examples include using a fluorescent tracer to measure the permeability of endothelial barriers in a retinal microvasculature-on-a-chip model to quantify drug efficacy ¹⁶⁰. 3D microscopy can be valuable for evaluating drug-induced morphological or structural changes ^{168, 173}, effects on specific cell types (single-cell analysis) ^{170, 174–176} or pathological features such as protein aggregations ¹⁶⁹. For example, confocal microscopy has been applied to assess villus heights in gut-on-a-chip models as a measure of a drug’s ability to suppress radiation injury ¹⁷³ (Fig. 4(c)).

3.2. Mechanisms of action

Methods that are typically non-imaging, such as mass spectrometry and RNA sequencing, have often been used to assess metabolic or gene expression changes and to identify underlying mechanistic pathways ^177–179. For microscopy-based investigations that provide valuable spatial context, end-point analyses of treated and untreated groups are common. With less concern for imaging speed, photobleaching, and phototoxicity, confocal fluorescence or multiphoton microscopy can be an optimal choice due to their high levels of resolution and optical-sectioning abilities in thick tissues. For example, certain key molecules in signaling pathways, such as reactive oxygen species (ROS), can only be analyzed in situ with living cultures as they are small and will quickly diffuse across cell membranes in excised tissues ¹⁸⁰. In one study, confocal microscopy was employed to detect a decrease in ROS levels in radiation-injured gut-on-a-chip models after drug treatment, suggesting a possible response pathway ¹⁷³. It should be noted that tissue clearing is not applicable for this purpose. Therefore, issues of aberration and scattering need to be resolved through other methods (see technical perspectives section).

4. Cancer biology

Cancer biologists are often trying to elucidate the molecular and biomechanical mechanisms of cancer progression at multiple spatial scales from the sub-cellular to the organism level, all of which is vital for improving the ability to prevent and treat such diseases. Microscopy techniques have been commonly used to assess cancer cell motility, especially in the context of invasion and metastasis ^{181, 182}. Imaging methods also facilitate the examination of cellular interactions within tumor tissues, which commonly include not just cancer cells but surrounding structures such as blood vessels, lymphatics, nerves, fibroblasts, ECM, and immune cells ^183–185. Various 3D cell culture models have been developed to provide realistic and complex microenvironments for such studies. Due to the complexities of cancer, the majority of 3D cell culture models have only attempted to mimic certain features of the cancer of interest, and the analytical methods employed have varied greatly based on the specific goals of the investigations.

4.1. Cancer invasion and metastasis

Tumor spheroids/organoids have been derived to study their invasion into a surrounding matrix ^186–189. Another strategy has been to co-culture normal and tumor spheroids/organoids to model invasion properties ^{190, 191}. Organ-on-a-chip models, which can replicate more complex microenvironments, have been used to study the migration of cancer cells toward surrounding stroma or vasculature ^{176, 192–194}, as well as extravasation of cancer cells from blood vessels ^{174, 175}. Additionally, normal tissue-slice cultures co-cultured with tumor spheroids have been employed to study invasion within real tissues ^{195, 196}.

Time-lapse imaging of cell dynamics is critical for many studies. In many cases, time-lapse 2D microscopy can be sufficient to capture processes such as cell/tumor protrusions or migration patterns ^{176, 186–188, 191–193}. For example, TLM has been applied to quantify the invasion distance of breast cancer cells toward stromal regions in organ-on-a-chip models ¹⁷⁶. 3D time-lapse imaging can be useful for monitoring 3D morphological changes during invasion. Confocal microscopy has been used to assess the deformation of the ECM surrounding whole breast-cancer spheroids, hence providing insights into invasion mechanisms ¹⁸⁹. In this study, the small spheroid size (about 100 μm) likely reduced the impacts of aberration and scattering but such challenges may need to be addressed in larger 3D cell cultures.

High-resolution 3D imaging of fixed samples at the end of an invasion assay is also common for studying structures and molecular biomarkers within tumor cells ^{176, 187, 190–192, 195, 196}. Confocal microscopy is a popular choice, as it offers high resolution and optical sectioning capabilities. Here, issues of wavefront aberrations and scattering can be avoided through optical-clearing methods applied to fixed specimens. For example, confocal microscopy has been used to image cleared brain organoids invaded by glioma stem cells (including spheroid-like stem-cell clusters), where a specific goal was to observe cell-membrane structures ¹⁹⁰. Similarly, cleared brain slices have also been imaged to quantify glioma spheroid invasion ¹⁹⁵.

4.2. Angiogenesis and vascular changes

Organ-on-a-chip platforms can exhibit semi-realistic vascular networks and are often used for 3D cell culture research on angiogenesis ¹⁹⁷. Engineered structures such as microfluidic channels can partially mimic the spatial organization of blood vessels and tumors, allowing researchers to study the process of angiogenic sprouting ^198–203. Standard 2D epifluorescence microscopy can be suitable for many studies in which basic vascular metrics such as area, density, and length are desired. For instance, in a study that established a tumor spheroid-on-a-chip with endothelial cells seeded on the sides, researchers utilized epifluorescence microscopy to assess the total area of the vasculature after days ¹⁹⁸. For nuanced examination of subtle changes in vessel morphologies, 3D microscopy may be needed. For example, the intricate network of the tumor vasculature has been analyzed using confocal microscopy to evaluate the effects of anti-angiogenic RNA nanoparticles ²⁰². For large scale analysis of angiogenesis or the co-opting of vessels by tumor cells, high-throughput LSFM can be of value, especially if performed in a multi-scale workflow that is time- and data-efficient (see technical perspectives section).

4.3. Fibroblasts and extracellular matrix (ECM)

Fibroblasts and ECM provide an essential structural scaffold and chemical support to tumors ^204,205. The four types of 3D cell culture discussed in this article have all been utilized to study the roles of fibroblasts and ECM in tumor progression through co-culture techniques ^206–212. Non-imaging approaches, such as quantitative real-time reverse-transcription polymerase chain reaction (qRT-PCR), are often used to examine expressional profiles of fibroblasts or tumor cells ²⁰⁹. Imaging techniques, on the other hand, are mainly used to characterize ECM or fibroblast structures ^206–212.

3D microscopy has clear advantages for observing the spatial organization of tumor cells, fibroblasts, and polymers in the ECM ^206–212. Confocal microscopy and multiphoton microscopy are often preferred for various reasons. For example, since fibroblasts and ECM do not actively migrate inside the TME, end-point analyses are typically sufficient and are compatible with slower imaging modalities like confocal/multiphoton microscopy ^{206, 207, 209, 211, 212}. In addition, the hierarchical structure of collagen, one of the main components in ECM, produces strong second harmonic generation (SHG) signal, which can be visualized using label-free multiphoton microscopy ^{207, 208, 210}. Such nonlinear microscopy methods can provide not only structural information, but also certain molecular details such as crosslinking ²¹³. In a study examining the effects of heat therapy on pancreatic tumor spheroids, researchers employed multiphoton microscopy to quantify changes in collagen fibers, providing insights into the underlying mechanisms of heat treatments ²⁰⁷.

4.4. Cancer-immune interactions

Immune cells modulate cancer behavior through direct interactions or the generation of inhibitory molecules ^214–216. Immunotherapy, such as T-cell therapies and checkpoint inhibitors, harnesses the immune system to fight cancer and have gained much interest ^217–219. The co-culturing of tumor cells and immune cells in 3D allows researchers to study complex and spatially heterogeneous cancer-immune interactions with much greater realism than in 2D ^220–222.

3D microscopy has been valuable for providing comprehensive spatial insights into tumor-immune interactions ^{163, 223–227}. For example, in research focused on identifying genes that boost the efficacy of engineered T cells against patient-derived tumor organoids, confocal microscopy was applied to quantify killing behaviors of T cells (against organoids) with different genetic modifications at the single-cell level (Fig. 4 (d)). This time-lapse analysis provided essential information for selecting genes that enhance killing effects, paving the way for further optimization of T cell therapies ²²⁴. In this case, where rapid time-lapse imaging is needed to capture cell-killing processes with minimal photobleaching and phototoxicity, LSFM could be an ideal imaging modality.

Major goals in this field include molecular profiling and behavioral analysis of the interactions between immune and cancer cells. Integrating these two aspects is particularly valuable. Many studies have predominantly employed microscopy techniques, especially live imaging, for behavioral analysis ^{163, 223–225, 228, 229}. In contrast, non-imaging methods have often been used to ascertain RNA or protein profiles (e.g., immune cell markers) ^{224, 226, 230–232}. A high-impact goal would be to integrate 3D microscopy and molecular profiling at the single-cell level in the form of spatial omics. In the previously mentioned research, single-cell sequencing was conducted, but the absence of a method to correlate single-cell molecular profiles with 3D microscopy-observed behaviors was a limitation ²²⁴. Rather than relying upon sequencing methods that lack spatial context, in situ 3D examination of RNA/protein expression would be transformative. Current multiplexed imaging techniques have been largely confined to 2D sections, but some promising 3D techniques are beginning to emerge ^233–237 (see technical perspectives section).

Technical perspective

In the previous section, we surveyed different use cases for microscopy of 3D cultures and provided examples of how certain microscopy techniques could be optimal for different investigational topics. We also identified remaining technical challenges that should be overcome: (1) imaging the deeper or inner regions of 3D cell cultures, such as for studying morphogenesis in developmental biology; (2) developing optimized workflows to identify and image regions of interest in large samples, such as for monitoring host-pathogen interactions during bacterial or parasitic infections; and (3), imaging a holistic set of gene expression and structural targets with sub-cellular resolution within a large volume, as is particularly important for assessing tumor-immune interactions.

This section explores three technical areas for addressing the challenges listed above: aberration correction and scattering mitigation, multi-scale imaging, and multiplexed imaging. We introduce some promising methods and discuss current technical gaps, highlighting needs and opportunities for future development. As an introductory note, we anticipate one key consideration for any novel imaging system for 3D cell cultures is the need for integrated environmental control (i.e., temperature and humidity), especially since large 3D cell cultures will often require longer imaging durations than 2D cultures and may be more sensitive to environmental conditions.

Finally, while we discuss the concept of “computationally enhanced microscopy” to improve the image-generation process (section 4 below), we do not discuss the lengthy topic of image analysis in this article. Nor do we cover important topics related to large imaging datasets, such as the development and standardization of image formats, analysis pipelines, quality-control processes, and data-sharing solutions. These topics will need to be addressed in the future to maximize the impact of advanced optical microscopy techniques ^238–240.

1. Aberration correction and scattering mitigation

Adaptive optics (AO) technologies have been developed to address aberrations caused by refractive heterogeneities in complex 3D tissues. In general, wavefront aberrations are measured in AO systems as feedback to inform a wavefront-correction step ^{241, 242} (Fig. 5. (a)). The measurement methods can be broadly divided into wavefront-sensor-based and sensorless methods. The former use optical wavefront sensors to precisely quantify aberration patterns ^243–245, while the latter rely on analyzing image metrics like contrast or sharpness as a proxy to estimate the degree of aberrations ^246–249. Aberration correction is achieved either with specialized “adaptive elements” like deformable mirrors and spatial light modulators, which create inverse wavefront patterns to counteract the aberrations ^{243–245, 247, 248}, or by directly adjusting the positions/angles of optical components to optimize the alignment of the imaging system ^{246, 249}. Wavefront sensor-based methods enable rapid operation but add system complexity. Sensorless AO methods typically are slower, as multiple images are typically needed to characterize the aberrations or to iteratively mitigate their effects ^{241, 242}. AO-based microscopy methods have been applied for in vivo imaging of animal models with high structural complexity ^250–252. These innovations demonstrate the potential for addressing aberrations in 3D cell cultures. While the application of AO in 3D cell cultures is still emerging, pioneering studies have yielded promising results. For instance, one study successfully developed an AO-based LSFM system for imaging organoids (Fig. 5. (a)), capturing the dynamics of cellular ingestion over several minutes ^{243, 253}.

Fig. 5. Principles of aberration correction and scattering mitigation techniques for 3D cell cultures.

(a) Generic illustration of AO microscopy, demonstrating aberration detection using either a wavefront sensor or sensorless methods, and correction via a deformable mirror. The example images are of organoids captured by AO-based LSFM before and after aberration correction ²⁴³ (green: dynamin; magenta: clathrin). (b) The principle of two-sided illumination/collection in an LSFM system that uses four different combinations of objectives to image different parts of the sample and to avoid the effects of scattering/aberrations at deeper tissue regions. The final image can be generated through a computational image fusion algorithm. (inspired by BioRender.com)

Light scattering is a challenge to mitigate due to its stochastic nature. A common strategy to overcome this is to use longer wavelengths in the near-infrared region, which exhibit reduced tissue scattering, thereby enabling deeper light penetration in biological samples. Recent advances, such as the development of multiphoton light-sheet fluorescence microscopy, which uses a near-infrared light sheet to enhance imaging depths while maintaining low photobleaching and phototoxicity ^254–256, show promise for imaging living 3D cell cultures. Additionally, the integration of confocal gating in LSFM has been used to improve the rejection of scattering background ^{257, 258}. In this method, a laterally scanned “pencil beam” generates a light sheet over time and the position of the pencil beam is synchronized with the rolling shutter of a sCMOS camera to reject out-of-focus blur from scattered light ²⁵⁸. Another strategy for dealing with light scattering is to image specimens from multiple directions, such as with dual-sided light-sheet illumination/collection ²⁵⁹. For example, some studies of embryos employed LSFM with two-sided illumination/collection to address tissue scattering along with adaptive optics to address aberrations and sample-induced misalignments ^{246, 260} (Fig 5. (b)). Similar techniques have also been applied for organoid imaging with customized sample holders ^{115, 261}. However, these multi-sided illumination/collection strategies are limited to specimens with certain geometries and sizes. Recently, open-top dual-view LSFM has been developed, further enhancing the capability to accommodate larger specimens while addressing the challenge of imaging deeply in scattering tissues ²⁶². Finally, similar to AO, novel techniques have been explored in recent decades to undo the effects of light scattering in tissues, including the use of a fluorescent or ultrasound “guide star” ^263–265. For example, by leveraging the reduced scattering properties of ultrasound in biological tissues, focused ultrasound waves can be used to tag light (in this case, shifting the frequency of specific photons) at the location of a desired focus. Through holographic detection and play-back methods, the tortuous path of the scattered light from the focus can then be reversed, enabling light to be spatially confined at the location of the original ultrasound-tagged focus. While these novel techniques remain slow and complex, future advances may allow them to be applied for imaging 3D cultures. Most challenging is the fact that such methods require constant and rapid updating of the illumination patterns that are custom generated to counteract the effects of scattering. This is because the small scatterers in living samples are constantly moving and randomizing at timescales on the order of milliseconds.

In addition to microscopy hardware innovations, issues of aberration or scattering can also be mitigated through the appropriate co-design of 3D cultures and imaging systems. For example, modifications could be made to minimize the thickness of sample substrates and to position tissue regions of interest closer to the microscope objectives. Instead of spatially arranging the cells as multilayers along the z-axis, some new designs of organ-on-chip systems arrange the functional subunits (e.g. crypt-villus in intestines ⁸⁰, sinusoid in liver ²⁶⁶) in the x-y plane to enable optimal light penetration and imaging of structures/processes of interest (e.g. cell migration along the crypt axis) over time. In addition, microengineered scaffolds or structures (for example, hydrogel microcavity arrays) can be applied to 3D cultures to guide the formation of organoids at predefined locations along the same focal plane, significantly facilitating high-throughput and high-content imaging ¹⁶⁸.

2. Multi-scale imaging

To interrogate areas of interest within large sample volumes, multi-scale imaging is often necessary, ideally with a single microscope that offers multiple magnifications. This is especially crucial for 3D imaging, where imaging times and dataset sizes scale with resolution to the 3^rd power. In short, low-resolution imaging enables quick surveys of large sample areas /volumes to identify areas of interest with minimal phototoxicity and time. Subsequent high-resolution imaging enables detailed and quantitative analysis of those localized regions (Fig. 6 (a)). A recent study has introduced a novel multi-well plate equipped with micro-mirrors, enabling both transmitted light and light-sheet fluorescence imaging with a conventional microscope. This innovation along with an automatic multi-scale imaging pipeline was used to identify rare cell clustering events during the development of intestinal organoids ²⁶⁷ (Fig. 6 (b)).

Fig. 6. Principles and applications of multi-scale imaging in 3D cell cultures.

(a) An example workflow using multi-scale imaging in which low-resolution imaging is first used to rapidly screen an array of 3D cultures to identify regions of interest, followed by high-resolution imaging of specific structures/regions for detailed quantitative analysis. (b) In one example, a custom-designed multi-well plate, which incorporates micro-mirrors, is used with a conventional microscope equipped with switchable objectives. Initially, simple widefield TLM (2D imaging) is used to record the position of each organoid, followed by the application of moderate-resolution LSFM to detect rare cell-cluster events. High-resolution LSFM is then used to carefully characterize those rare structures (scale bar: 30 μm) ²⁶⁷ (c) In another example, a multi-scale “hybrid” OTLS microscope enables whole cleared mouse brains to be rapidly screened within several hours using a low-resolution imaging path to identify the locations of brain metastases, followed by detailed high-resolution interrogation of small regions of interest containing those metastases for quantitative analysis. (scale bar: 1 mm for low-resolution imaging and 100 μm for high-resolution imaging) ²⁷⁸. (inspired by BioRender.com)

Furthermore, future integration of super-resolution techniques into multi-scale imaging could expand the range of observable scales in 3D cell cultures while maintaining efficient identification of regions of interest. While super-resolution microscopy generally offers a limited field of view, it remains valuable for studying the morphology and/or spatial correlations of small (< 200 nm) cellular components ^268–271. For example, stimulated emission depletion (STED) microscopy has been used to observe the contact points between microglia processes and synapses in brain organoids, providing insights into microglia-neuron interactions ²⁷².

One approach for multi-scale imaging is to utilize a single objective (and single imaging path) to achieve both a high NA (high resolution) and a large field of view. A number of recent systems have utilized such specialized optical components, which can be expensive and bulky ^273–275. However, there are some cost-effective options that have been mass produced for metrology applications in manufacturing industries ²⁷⁶.

Multi-scale imaging with a turret of objectives is routinely used in traditional microscope modalities like epifluorescence and confocal microscopy. However, the benefits of multiscale imaging are only beginning to be deployed in LSFM systems ^277–279. This is especially true for LSFM microscopy systems that require specialized water- or oil-immersion objectives or customized sample-mounting strategies, where it can be difficult to rapidly switch objectives without disturbing the sample – this is also true for other microscopy modalities that utilize immersion objectives. A previously developed open-top light-sheet (OTLS) system addressed this challenge by incorporating a solid immersion meniscus lens (SIMlens), a specialized “wavefront-matching” element that allows air objectives to be used as immersion objectives, thereby enabling easier switching between objectives ²⁸⁰. A multi-scale “hybrid” OTLS microscopy system was also recently developed, featuring integrated low- and high-resolution imaging arms within one system, eliminating the need for mechanical switching of objectives ²⁷⁸ (Fig. 6 (c)). In another example, two different imaging modalities were used to provide multi-scale imaging information – in this case, a combination of confocal microscopy and optical coherence tomography (reflectance microscopy) – to enable observation of both whole-spheroid growth and intricate internal structures ²⁸¹. To complement such hardware improvements, automated workflows will be essential to overcome the manual and time-consuming process of determining when and where to switch between imaging scales, especially for large 3D cell cultures and/or multiple 3D cultures in high-throughput screening applications. Additional examples and details will be explored in a section below, “Computationally enhanced microscopy.”

Beyond the common microscopies mentioned above, an emerging technique, light-field microscopy (LFM), also shows high potential for multi-scale 3D imaging of cell cultures. LFM utilizes a novel optical configuration that enables the computational reconstruction of 3D images from a single camera exposure at extremely high volumetric frame rates ^282–287. For example, Fourier LFM has been successfully applied to observe the response of a whole colonic organoid subjected to external physical stimuli at a volumetric imaging rate of 10 Hz ²⁸⁸. While one major tradeoff of LFM is the need to sacrifice resolution to obtain depth information, recent developments in high-resolution LFM (with a smaller field of view) have addressed some of these limitations. One such LFM system was used to characterize mitochondrial dynamics in a single living cell, demonstrating the scalability of LFM ²⁸⁹. Future innovations in multi-scale LFM systems could greatly benefit 3D cell cultures research, particularly for high-throughput screening of living cultures. However, LFM would likely struggle with thick specimens that are densely labeled since the limited dynamic range of the detector and the cumulative shot noise due to myriad signals collected from a large volume would lead to image deterioration in comparison to 3D microscopy methods that restrict the generation (i.e. LSFM, multiphoton) or collection (i.e. confocal) of out-of-focus signals.

3. Multiplexed imaging

As 3D cell cultures become more complex, there is a growing need to study numerous targets within a single specimen. Emerging methods for multiplexed imaging present new opportunities in modern studies of 3D cultures but they also introduce inherent tradeoffs. The choice of multiplexing strategy can differ significantly based on whether 3D cultures need to be examined in their living or fixed states. If the study of dynamics and live interactions is not required and imaging speed is not a primary concern, a larger number of molecular species can be probed and a broader array of tools is available. Among these, immunofluorescence imaging stands out due to its high molecular specificity and ability to concurrently visualize multiple molecular distributions. Immunofluorescence imaging is generally limited to the visible spectrum (400-700 nm) and can accommodate, in most cases, the excitation and emission spectra of up to ~5 fluorophores. To image more probes, highly multiplexed immunofluorescence strategies (e.g., MxIF ²⁹⁰, t-CyCIF ²⁹¹, IBEX ²⁹², SWITCH ²³⁶) use sequential cycles of staining and imaging of 3-5 probes per cycle where fluorescence is inactivated between each cycle to avoid spectral overlap (Fig 7. (a)). This method has been applied to thin slices of fixed organoids ^{293, 294} (Fig. 7. (a)). Alternatively, a single round of staining with barcoded antibodies, each containing unique oligonucleotide sequences, has been combined with a sequential readout of barcodes using complementary fluorescent oligonucleotides (e.g., CODEX ²⁹⁵, Immuno-SABER ²⁹⁶). Multiplexing has also been combined with spatial transcriptomics to map mRNA distributions across tissues (e.g., MERFISH ²⁹⁷, seqFISH ²⁹⁸, and EASI-FISH ²³³). In these cases, tissues are stained with barcoded probes for gene panels in a single round, followed by sequential rounds of reagent delivery and imaging to spatially localize the transcripts. All these techniques allow for probing tens of molecular targets within the same sample. However, the iterative nature of staining and/or imaging in these approaches can be prohibitively time-consuming for thicker 3D cell cultures ^233–236. It should be noted that we only summarized a few popular methods for spatial omics here, and that numerous reviews provide much more detail for this rapidly evolving field ^{299, 300}.

Fig. 7. Principles of highly multiplexed imaging and their application in 3D cell cultures.

(a) An example workflow for cyclic staining and imaging of fixed samples typically involves staining 3 - 4 protein targets with antibodies, followed by photobleaching and re-staining after each imaging round to enable the detection of multiple targets across cycles. Bottom part is an example multiplexed image of a thin retinal organoid section with 32 “tissue units” in different colors ²⁹³. (b) For live samples, hardware techniques such as hyperspectral imaging can be applied to capture multiple labels. The mixed signals are computationally unmixed to create a final multiplexed image. (c) For virtual staining, AI models are used to predict the staining of multiplexed targets from label-free (or low-plex) images

For studies that aim to capture the physiological dynamics of 3D cell cultures, time-lapse imaging of live specimens is necessary and can impose constraints on the level of multiplexing achievable (i.e., detection times typically scale with the number of imaging channels). Maintaining a viable living sample is incompatible with the harsh chemical treatments used in the iterative immunofluorescence-based techniques mentioned above. One recently published strategy has achieved biocompatible cyclic staining and rapid quenching of multiple fluorescent antibodies to label the surfaces of living cells in tissues ³⁰¹ and organoids ²⁹⁴. In most cases, multiplexed imaging of live 3D cultures relies on a handful of biocompatible fluorescent molecules and a small but growing palette of genetically expressed fluorescent proteins ³⁰². Several promising approaches use advanced optical hardware and software to discern the subtle differences between overlapping spectra of chemical species in the sample. For example, hardware-based strategies have aimed to accommodate more fluorophores at each time point by rapidly changing the excitation wavelength of light using a tunable filter ³⁰³, or using a hyperspectral imaging detector ³⁰⁴. Computational algorithms such as phasor-based methods are required to unmix the signals from multiple fluorophores with overlapping excitation or emission spectra ^{305, 306} (Fig. 7. (b)). While these strategies are currently limited to identifying 5-10 chemical species, their capabilities are growing.

Studies of 3D cultures would benefit from the ability to stain and image more targets simultaneously, allowing for faster profiling than cyclic approaches. A recently developed method demonstrated single-round 21-channel imaging using novel fluorescent semiconducting polymer dot (PDot) probes that have a highly tunable Stokes shift and produce a palette of spectrally distinguishable probes based on a combination of excitation and emission wavelengths ³⁰⁷. In the future, software advances may also relax current practical tradeoffs, as artificial intelligence (AI) models have been used to predict molecular expression patterns and phenotypes (i.e. virtual staining) ^308–311 (Fig. 7. (c)), thereby reducing the degree of multiplexing necessary or improving image quality/speed in multiplexing experiments ³¹² (see next section on “Computationally enhanced microscopy”).

4. Computationally enhanced microscopy

Besides hardware innovations, future advances in software, especially artificial intelligence (AI), will undoubtedly play a significant role in facilitating the imaging of 3D cell cultures. Computationally enhanced microscopy refers to the integration of advanced software with traditional microscopy techniques to overcome certain hardware limitations and practical imaging challenges, thereby improving image acquisition, analysis, and interpretation. One well-established technique is deconvolution microscopy, which has been used in both 2D and 3D microscopy to improve resolution and contrast by computationally removing the effects of aberrations and out-of-focus light ^{56, 313}. While deconvolution routines have traditionally been iterative and slow, AI techniques are now being explored to accelerate and improve deconvolution performance ^{314, 315}. The computational removal of out-of-focus light and scattering-induced background can also be achieved using structured illumination techniques in which an engineered pattern of light appears with high contrast within a focal plane of interest but is blurred out in the background ^316–318. This difference between the in-focus and background signal allows for computational removal of the background after one or more images is acquired, and in some cases allows for resolution enhancement as well. As a final example, others have acquired images of tissue targets co-stained with visible and near-infrared probes to train deep-learning models to restore the quality of the visible-fluorescence images at depths that are typically only accessible with near-infrared imaging. This has the advantage of allowing biologists to continue to rely on popular visible fluorescent agents and fluorescent proteins, which tend to be brighter than near-infrared fluorophores and readily available in diverse forms and protocols ³¹⁹.

As an example of an AI-assisted multi-scale workflow, a recent study integrated epifluorescence and LSFM microscopy to detect rare cell-division events ³²⁰. In this example, rapidly acquired epifluorescence microscopy images were input into a deep-learning classifier, which identified the coordinates and states of each cell to automatically guide the 3D imaging of dividing cells with LSFM. In another example, AI has been used to facilitate multi-scale imaging in the temporal domain ³²¹. In this innovative event-driven acquisition workflow, lowspeed imaging was used for broad long-term monitoring of cells with minimal photobleaching, phototoxicity and computational demands. Upon detecting cellular events of interest through the deep-learning model, the system automatically switched to a high-speed imaging mode for detailed analysis of these transient processes of interest. Although these multi-scale imaging strategies have so far been applied only to cell monolayers, they hold promise for optimization in 3D cell cultures.

The flourishing field of computer-based virtual staining, now widely reported for its applications in single-cell and 2D histological analysis ^{309–311, 322}, promises to streamline staining and imaging workflows in 3D cell cultures. For instance, one study developed deep-learning models to infer six fluorescence-labeled images from label-free images of cell monolayers ³¹¹. These approaches could add even more value for 3D microscopy applications in which large volumetric samples could be extremely difficult or time-consuming to stain, and where each additional imaging channel adds significant challenges in terms of imaging times and cost ³²³. In addition, AI-based image analysis has been effectively applied for segmenting, tracking, and characterizing organoids from standard TLM ^324–329 and confocal microscopy images ^{330, 331}.

A major challenge in integrating AI in 3D cell culture imaging pipelines lies in the requirement for substantial training data. This limitation is being addressed with newer foundation models ^332–335, which are pre-trained on diverse datasets and are less limited to specific biological domains. Foundation models have recently been used in organoid image analysis to predict drug responses with less training data and computational effort ³²⁹, demonstrating their potential to streamline the implementation of AI in 3D cell culture imaging. Nonetheless, effective applications of computationally enhanced microscopy will require concerted efforts in both hardware and software development, including hardware-software co-designs.

Summary

The emergence of 3D cell culture models serves to bridge the gap between traditional 2D cell cultures and animal models, facilitating an extensive range of biological research directions with increased realism and insights. Optical microscopy has been an indispensable and ubiquitous tool for biological investigations involving cell cultures and animal models in the past, and therefore has an obvious and important role in extracting maximal value from modern 2D and 3D cell cultures. However, continued advancements in microscopy are needed to accommodate the increase size and complexity of such tissue constructs. To facilitate high-impact innovations in this field, this article first surveyed existing optical microscopy approaches within the context of a diverse set of biological applications. By discussing the technical needs of biologist end-users, along with the strengths of weaknesses of existing microscopy approaches for specific use cases, we identified areas in which future technical developments could have a large impact. These include the need to combat the effects of tissue scattering and aberrations to image deeply within 3D specimens; the need for multi-resolution workflows to efficiently interrogate large volumes; the need for improved multiplexed imaging strategies to analyze the relationships between various cell types and tissue structures; and finally the need to incorporate computational methods, including AI, to facilitate all of these advances plus other workflows (e.g. image analysis, which we did not cover in this article). Since our goal was to identify areas of greatest impact to a broad set of users, this article covered only a subset of 3D cell culture types, research disciplines, and imaging methods. Nonetheless, we hope the discussions can serve as a guide for both end-users and engineers using and advancing the development of novel microscopy techniques for 3D tissue cultures.

Fig. 3. Optical diagram, general characteristics and examples of common 2D and 3D microscopy methods used to image 3D cell culture models.

The performance of six classes of microscopy techniques is roughly generalized and compared for six parameters: optical-sectioning ability (volumetric ability), phototoxicity and photodamage, speed, cost, ease-of-use, and spatial resolution. Example images: transmission light microscopy: intestinal organoid ⁷²; epifluorescence microscopy: ileum organoid ⁷³; confocal fluorescence and multiphoton microscopy: colonic organoids ⁴¹; light-sheet fluorescence microscopy: brain organoid ⁷⁴