Multiphoton intravital microscopy of rodents

Abstract

Tissues are heterogeneous with respect to cellular and non-cellular components and in the dynamic interactions between these elements. To study the behaviour and fate of individual cells in these complex tissues, intravital microscopy (IVM) techniques such as multiphoton microscopy have been developed to visualize intact and live tissues at cellular and subcellular resolution. IVM experiments have revealed unique insights into the dynamic interplay between different cell types and their local environment, and how this drives morphogenesis and homeostasis of tissues, inflammation and immune responses, and the development of various diseases. This Primer introduces researchers to IVM technologies, with a focus on multiphoton microscopy of rodents, and discusses challenges, solutions and practical tips on how to perform IVM. To illustrate the unique potential of IVM, several examples of results are highlighted. Finally, we discuss data reproducibility and how to handle big imaging data sets.

Intravital microscopy (IVM) provides an unrivalled ability to appreciate the principles regulating tissue and organ function at cellular and subcellular levels by capturing cell behaviours in situ and in real time. IVM is, broadly, any kind of microscopy performed on any kind of living organism. In this Primer, we focus on the high-resolution imaging of individual cells in living rodents with techniques such as multiphoton microscopy. In the last two decades, these high-resolution microscopy technologies in rodents have rapidly become popular in the fields of cell biology and medicine, as they shed new light on physiological and pathological processes. Cellular behaviour is driven by the tissue (micro) environment, which is often incomplete and oversimplified in the in vitro context. Indeed, fundamental differences in cell behaviour have been observed between in vivo and in vitro conditions. For example, IVM in tumour-bearing mice showed that the mode of action of taxanes, an important class of chemotherapeutics, is different between in vitro and in vivo conditions¹. Although cancer cells undergo mitotic cell death in vitro upon treatment with taxanes, in vivo the majority of the same cancer cells die without any signs of mitotic defects, underscoring the importance of in vivo validation of chemotherapeutic agents. IVM studies not only can confirm or dispute hypotheses generated through in vitro or histological studies but have also enabled the discovery of biological principles that were otherwise unexpected.

In this Primer, we do not describe the principles of microscopy techniques in detail, which have been extensively described elsewhere^2–5. We aim to generate a guide that helps researchers to set up or optimize IVM in rodents in their own laboratory. For this, we discuss microscopy set-ups, optimizing an IVM experiment and the type of analyses that can be performed. We present some examples of the use of advanced microscopy techniques — such as photobleaching, photoactivation, photo-conversion, Förster resonance energy transfer (FRET) and fluorescence lifetime imaging microscopy (FLIM) — where the in situ and real-time nature of IVM expands the degree of information that can be achieved with these approaches. IVM researchers face challenges based on the nature of the specimens and the need to maintain physiology in anaesthetized animals. We outline how to overcome these challenges, including procedures on how to correct for movement aberrations due to breathing and heartbeat and guidelines on how to implement techniques to excite and detect fluorophores deep within tissues and organs that are most often difficult to access. Finally, we provide some considerations on the complex relationship between IVM and the principles of the 3Rs of animal work: although IVM procedures are often complex and more invasive for the animals, they often enable reduction of the overall number of animals used and are irreplaceable in terms of providing unique information. Indeed, IVM in rodents generates rich data sets that can be analysed deeply and uncover unexpected and otherwise untestable biological mechanisms in healthy and diseased tissues. For example, IVM has revealed the cooperation of multiple cell types to enable the intravasation of immune cells during inflammation⁶ or metastatic cells leading to tumour spread⁷, and the unexpected migratory behaviour of multiple stem cells^8–11 and chemoresistant tumour cells^12,13.

Experimentation

Microscopy modalities

Modern IVM platforms consist of a laser, a scan head and detectors, which enable either point, line or multifocal scanning, or light sheet detection of intact tissue in living animals^14,15 (TABLE 1). Wide-field and confocal microscopes detect fluorophores by single-photon excitation with light in the visible wavelength range (<700 nm). Single-photon excitation achieves high resolution (down to 200 nm) and imaging speed (multiple frames per second) within around 100 μm from the tissue surface, including precise detection of cell organelles, membrane trafficking and cytoskeletal dynamics^16,17. Excitation by visible light suffers from strong scattering by tissue structures, leading to loss of signal and resolution when focusing deep inside tissues¹⁸ (FIG. 1a). For fast imaging of superficial tissue layers, wide-field microscopy enables frame rates up to 1,000 s⁻¹, which is sufficient for monitoring fast calcium currents and arterial blood flow.

Table 1 |.

Microscopy techniques used for intravital microscopy

Microscope	Strength for IVM	Weakness for IVM
Wide-field	Very sensitive Fast imaging Commonly available microscope Cost-efficient	Imaging depth is limited to tens of micrometres in non-cleared tissues Poor z-resolution
Light sheet	Very fast imaging of large areas in three dimensions Minimum photodamage and fluorophore bleaching Optimally suited for embryos, brain and fixed tissue after clearing	Very large data sets Imaging depth is limited to tens of micrometres in non-cleared tissues Cannot be applied to rodents Requires specialized microscopes and expertise Susceptible to shadowing artefacts
Single-point scanning confocal	Thin optical section Commonly available microscope	Imaging depth is limited to 100 μm in non-cleared tissues Slow imaging
Spinning disc confocal	Thin optical section Fast imaging of single planes	Imaging depth is limited to tens of micrometres in non-cleared tissues Requires bright samples due to large light losses
Multiphoton	Improved imaging depth with hundreds of micrometres for three-photon and four-photon excitation in non-cleared tissues Thin optical sections Enables higher harmonics of tissue structures Up to seven channels simultaneously detected	Requires specialized microscopes and expertise Costly

Fig. 1 |. Intravital microscopy by multiphoton excitation.

a | Multiphoton microscopy platform. A pulsed pump laser is introduced into a tunable oscillator or amplifier to generate high-power laser light in the range of 1,000–1,700 nm. Multiphoton-excited signal in the sample is detected without the need of a pinhole. b | Jablonski diagrams of one-photon, two-photon, three-photon and four-photon-excited fluorescence and higher harmonic generation. Arrows indicate individual photons causing excitation (upward arrows) and emission (downward arrows). c | Single-photon excitation excites all fluorophores potentially causing photodamage in and out of the imaging focus; two-photon and multiphoton excitation occurs only in the imaging focus and, thereby, minimizes photodamage in out-of-focus planes. d | Example of a multicolour intravital microscopy (IVM) image including second harmonic generation (SHG), third harmonic generation and multiple colour channels. Displayed tissue volume in xyz represents 210 × 210 × 110 μm³. Part d adapted from REF.²¹, CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).

Multiphoton imaging.

To excite deeper layers and allow 3D imaging in complex in vivo environments, multi-photon imaging is the method of choice. Near-infrared (700–1,000 nm) or infrared (1,000–1,700 nm) laser light is introduced into the sample by pulsed lasers, causing high photon density in the focus and the excitation of fluorophores by multiple photons^18,19 (FIG. 1a). In contrast to single-photon excitation, two-photon imaging fluorophores are excited by simultaneous absorbance of two photons with half the excitation energy^18,19 (FIG. 1b). More recently, infrared pulsed light sources can achieve three-photon and four-photon excitation of tissue-intrinsic fluorescent molecules and exogenous fluorescent reporters, as well as generate high harmonic light scattering^20,21 (FIG. 1b). Multiphoton excitation allows penetration up to 1 mm in some tissues and causes less photodamage due to a lack of excitation in out-of-focus regions, rendering this approach suitable for deep tissue microscopy in small animals²² (FIG. 1c). As a recent development, increasing the pulse energy of the laser light but reducing the pulse repetition rate (from 80 MHz down to 1–2 MHz) allows a dramatic increase in photon density within the laser focus, without increasing photodamage to the sample^21,23,24. This high photon density enables three-photon excitation at an excitation wavelength of 1,300 nm, and at 1,650–1,700 nm a combination of three-photon excitation of red fluorophores together with four-photon excitation of green and blue fluorophores (FIG. 1b). Up to twofold increased tissue penetration is achieved in the brain and peripheral tissues when using three-photon and four-photon microscopy compared with conventional two-photon microscopy, due to increased pulse energies, due to improved signal-to-noise ratio and because 1,300 nm and 1,650–1,700 nm both reside in a region of low water and biomolecule absorption, known as the near-infrared window^21,25,26. Using an integrated platform design, it is now possible to combine multicolour confocal microscopy with multiphoton imaging in the near-infrared and infrared excitation spectrum^27–29 (FIG. 1d).

Commercial microscope vendors have made great strides in microscope hardware, incorporating features such as FLIM electronics³⁰, optical parametric oscillators^27,29, group velocity dispersion compensators³¹, multicolour detection³² or video-rate acquisition³³, all of which were once only research endeavours and of limited accessibility. To allow for further customization of multiphoton microscopy systems, companies offer kits and publish open-source designs³⁴ to reduce the complexity of system design and remove technical barriers. Novel systems are still pushing the envelope of system capabilities, incorporating new features such as multispectral/hyperspectral imaging^35,36, unique light sources^21,37 or mesoscale imaging³⁸. Software for system control and data acquisition is intimately tied to the hardware chosen. Commercial systems, or those that build upon them, must use the software provided by the company. Kits and home-built systems provide the option either to write software from scratch or to employ one of several open-source initiatives^39–43. Image analysis can be performed in various different commercial or open-source platforms.

Upright and inverted configurations.

The two major microscope configurations used for IVM are upright and inverted microscopes. For inverted IVM, the sample is immobilized on the stage above the objective. Conversely, for upright IVM the objective is positioned above the sample and can be dipped into surgically exposed samples or positioned above a coverslip, which helps to immobilize and flatten the sample. Although the majority of tissues can be imaged using both configurations (FIG. 2a), intravital brain microscopy is commonly performed on upright microscopes (FIG. 2a), which allows immobilization/stabilization of the animal head using stereotactic head holders^44–46.

Fig. 2 |. Intravital microscopy of tissues through optical imaging windows.

Murine models can be imaged with upright or inverted microscopes (part a). Examples of chamber types (cartoons) and representative example multicolour intravital microscopy (IVM) images taken through the imaging windows of the dorsal skinfold chamber (part b), cranial window (part c), lung window (part d), mammary imaging window (part e) and abdominal imaging window (part f). Scale bars represent 20 μm. eGFP, enhanced green fluorescent protein; SHG, second harmonic generation. Part b (IVM image) reprinted with permission from REF.³⁸⁴, Rockefeller University Press.

Gaining optical access

Several vertebrate animal models, including zebrafish, are naturally transparent during early developmental phases, and therefore well suited to perform IVM. In small rodents, including commonly used mouse models, a few organs present optically non-turbid surfaces — for example, skin, eyes and the oral cavity — that allow direct access for in vivo microscopy. Gaining optical access at a cellular or subcellular resolution at much deeper sites is essential for many experiments.

Microscope access to most tissues and organs requires surgical intervention, either by ad hoc surgery to expose the tissue for direct coupling to the optics or by implantation of an optical window (FIG. 2b–f). Ad hoc approaches restrict imaging to limited periods of time (a few hours) due to the risk of side effects caused by extended use of anaesthetics and by limitations in maintaining the physiological conditions of exposed organs. Ad hoc approaches are therefore mostly used to visualize short-term cellular dynamics such as cell migration or invasion^47–49. Optical windows protect the tissue, reduce motion artefacts and enable longitudinal imaging over days to months^50–54 (FIG. 2b–f). A multitude of imaging windows have been developed over the past years providing optical access to nearly all organs in the mammalian body, including brain, breast and abdominal organs⁵⁰. Windows were initially fabricated using passivated stainless steel, titanium or selected bio-compatible polymers. Recent advances include the development of flexible silicon windows to image subcutaneous organs such as muscles and mammary tissue^55,56; the imaging window with a replaceable lid, which provides physical access to the tissue of interest over the entire duration of the imaging experiment⁵⁷; as well as the development of an imaging window to follow embryonic development in utero at a high resolution⁵⁸. Other windows that are not frequently used but can improve imaging depth beyond the limits imposed by the scattering property of the tissue include gradient-index lenses and prisms^59–61. Gradient-index lenses are based on either a single lens or a series of thin lenses (diameter 1–2 mm) inserted into soft tissues (primarily the brain) to reach the desired depth. Prisms are inserted with the hypotenuses coated with a reflective material, enabling the translation of the rastering pattern of the excitation beam from an x–y plane to an x–z plane.

Animal maintenance

In order to limit sample movement, rodents typically need to be anaesthetized for IVM. Gaseous isoflurane is commonly used for induction and maintenance of anaesthesia (5% or 0.75–3% anaesthesia in oxygen or air, respectively, at an oxygen flow rate of 1–2 l min⁻¹). Imaging through chronic imaging windows is not invasive and therefore requires a minimal level of anaesthesia, allowing rapid recovery of the animal, even after extended periods of time (>8 h)^50,62–64. For more invasive procedures where the tissue of interest is surgically exposed for IVM (such as skin flap surgery or acute imaging windows)^64–66, analgesia is required for pain relief in combination with anaesthetic agents. For example, a combination of zoletil or ketamine and xylazine has been used for non-recovery IVM of subcutaneous tumours exposed via skin flap surgery, which involves euthanizing the animal after imaging^65–67. Other combinations involving the use of midazolam, fluanisone and fentanyl have also been described for acute non-recovery IVM^68,69. In order to avoid a too light or too deep plane of anaesthesia, the animal can be monitored by checking reflexes (such as in response to a toe pinch), by observing respiration or by monitoring the heartbeat, respiration and blood oxygenation^57,65,67,70.

Although anaesthetics and analgesics can help to limit animal movement and reduce pain sensation, they may also interfere with physiological processes. For example, it has been shown that different anaesthetic and analgesic agents can distinctly affect cellular activity, behaviour and function^71–73 compared with the awake state. Care should therefore be taken to choose an anaesthetic and analgesic regimen which does not affect the biological processes to be observed by IVM and to design complementary (non-imaging) experiments to further support the experimental hypothesis.

Keeping the sample happy should be the hallmark of in vivo imaging⁷⁴ (BOX 1). During IVM in rodents, the body temperature needs to be maintained near 37 °C to ensure physiological cell and tissue function and vascular flow (which regulates tissue oxygenation, nutrient supply/metabolism as well as the behaviour of circulating cells such as cancer cells)⁷⁵. This is essential as anaesthetized animals lose the ability to maintain body temperature independently. Temperature fluctuations have been shown to affect IVM readouts, such as fluorescence lifetime measurements⁷⁶, highlighting the importance of a physiological environment throughout the IVM session. In anaesthetized rodents, heated environmental enclosures, heating blankets and/or a heated stage are commonly used to maintain body temperature, which can be monitored using a rectal probe. The use of heated immersion objectives or dry objectives may help to maintain temperature in the tissue of interest, counteracting heat transfer away from the sample via a cooler immersion medium⁷⁷. Where possible, temperature maintenance near the tissue of interest can be monitored using a small probe close to the tissue and coverslip. IVM using a microscope set-up that is fully integrated into an enclosed incubating chamber can further help to maintain animal/sample temperature while also reducing temperature fluctuations of the IVM equipment.

Box 1 | Responsible use of intravital microscopy: beyond the 3Rs.

Intravital microscopy (IVM) touches upon the public and ethical debate on animal testing. This requires careful reflection on the application of IVM.

There are legal frameworks that provide guidance in the context of IVM, such as EU Directive 2010/63. Although the technique is used on a global level, legal frameworks often function only for a specific region or country. A mere reference to a legal framework will not suffice for the responsible use of IVM in general. There exist diverse views on the value of animals in research, and there are differences between regions and between animal species. For instance, the use of dogs in research often raises more public discussion than the use of mice, and replacing mice with zebrafish in IVM is sometimes considered a preferred alternative, even though these three animal species have similar capacity to experience pain and can be harmed by the testing procedures. Finally, the 3R principles (replacement, reduction and refinement)³⁸⁵ have been presented as an essential step to more responsible practice in animal research, with the aim to decrease animal use and reduce the impact of the procedures for the involved animals. Although these principles are important for responsible animal research, a simple reference to these principles is not sufficient, and the principles can be in conflict with each other. For instance, the tension between using a small number of animals for an invasive procedure where the tissue of interest is surgically exposed for IVM against a design that is less invasive but requires ten times more animals. The 3R principles provide no direct guidance in this case and often overlook the underlying reasons for or against the use of IVM. These principles function as guidance to improve the system of animal testing, but as such do not provide the ingredients for the societal and ethical justification^385,386.

The applications of IVM should be in line with recent views on responsible animal research. At the experimental design stage, this entails a testing strategy that reduces in vivo experiments. A comparison of non-invasive IVM methods with in vivo results is an important step to find ways to apply IVM techniques that do not harm animals or that no longer require the use of animals. Exchange between research groups and consorted action are required to adhere to the international landscape standards. This exchange may also contribute to harmonization of guidelines on an international level. The harms and benefits of IVM should be carefully assessed, including the application of the 3R principles. Establishment and maintenance of a culture of care³⁸⁷ is also needed for all who are involved in the use of animals for IVM, including an attitude of care for the animals, careful compliance with protocols and guidelines, and professional development.

Ethical reflection should be an integrated part of IVM to provide unique insights into the in situ behaviour of cells, and to solve fundamental biological and medical problems.

For long imaging sessions (>3 h), the animal should also be supplemented intraperitoneally, subcutaneously or via intravenous catheters with fluids, such as saline, in order to prevent animal dehydration. Dehydration of the tissue of interest itself is another challenge in IVM, which can lead to tissue shrinkage as well as motion artefacts during imaging. Whereas optical imaging windows above the tissue of interest can prevent tissue dehydration, surgically exposed tissues require a moist environment, such as constant flushing with warm Tyrode’s salt solution or saline^78,79 or applications of selected gels⁸⁰, to protect from dehydration. Incubating chambers enclosing the microscope set-up can help to circulate humidified air around the sample to maintain humidity/hydration.

Fluorescent labelling

In order to visualize molecular and cellular behaviour in vivo, many IVM approaches rely on the use of fluorescent protein tags to provide spatial and temporal information on a protein of interest. In particular, red-shifted fluorescent proteins are well suited for IVM. Red-shifted emission light has long wavelengths leading to less scattering in tissues, thereby permitting increased imaging depth in tissues and reducing the contribution of endogenous tissue fluorescence to the overall signal. However, red and far-red fluorescent proteins often suffer from poor quantum yield and low photostability, which limits the available photon budget per sample, reduces the signal-to-noise ratio and impedes long-term imaging. Furthermore, many red-shifted fluorescent proteins present as oligomers, which can impose steric hindrance if tagged to a protein of interest and lead to biological artefacts. The recent generation of monomeric red and far-red fluorescent proteins with improved brightness and photostability^81,82, or the genetically engineered red-shifting of existing green fluorescent proteins⁸³, has helped to overcome these limitations for IVM studies.

Labelling strategies for IVM may include the implantation of fluorescent cells into animals, for example in tumour xenograft settings^84–86, studies of immune cell behaviour^87,88 or to visualize exchange of molecules between cells carried by extracellular vesicles^89–91. Where cells and tissues can readily be accessed, in vivo electroporation may be used for delivery of fluorescent protein-encoding genes as demonstrated for IVM of embryos⁵⁸, postnatal tissues, such as the brain⁹², retina⁹³ and skin⁹⁴, or tumours⁹⁵. Viral gene delivery can also be achieved in vivo, for example via viral inhalation to label the lung⁹⁶ or systemic administration to label abdominal organs such as the liver⁹⁷. Other labelling agents may include fluorescent dextrans or quantum dots for IVM of the vasculature^7,69, fluorescent antibodies for affinity labelling of specific cell-surface proteins (intravital immunofluorescence)^98,99, probes for intracellular pH changes¹⁰⁰, reactive oxygen species generation^100,101, protease activity¹⁰² or dyes to label fatty acids¹⁰³, lipid droplets¹⁰⁴ or vesicles¹⁰⁵.

For specific and permanent labelling of cell types and tissues, genetically engineered mouse models may be used, where fluorescently labelled proteins are expressed from ubiquitous gene loci, such as Rosa26 or Hprt, or from endogenous gene loci to recapitulate native expression patterns. Knock-in of fluorescent proteins in the endogenous locus can be used to visualize cell state changes in real time using IVM. For example, endogenous labelling of CDH1 with a fluorescent protein was used to monitor real-time CDH1 localization and expression levels as a readout for epithelial-to-mesenchymal transition during cancer progression^106,107. Similarly, using a mouse model in which enhanced green fluorescent protein (eGFP) is knocked-in behind the Lgr5 gene, in vivo stem cell plasticity was visualized over time in homeostatic and tumorigenic conditions^108–110. To follow cell cycle progression using IVM, transgenic fluorescence ubiquitination-based cell cycle indicator (FUCCI) reporter mice were generated wherein a cyclic change in colour from green over yellow to red can provide information on the cell cycle phase in which an individual cell resides¹¹¹. This has, for example, been used to demonstrate that myogenic cells migrate faster during S/G2/M phase during muscle regeneration¹¹², whereas invadopodia formation predominantly occurred during G1 phase in a mouse model of breast cancer¹¹³. FUCCI IVM has also been used to optimize response to standard-of-care chemotherapies in preclinical cancer models⁸⁴. Furthermore, transgenic mice have been generated using fusion proteins of fluorescent timers, where a change in spectral properties occurs following synthesis over the course of fluorescent protein maturation. For example, a fluorescent timer fusion to histone H2B knocked into the Hprt locus was used to determine/map individual cell cycle length in vivo¹¹⁴. Similarly, fusion of the fluorescent timer to the TCR downstream gene Nr4a3 or the transcription factor Foxp3 was used to characterize the temporal dynamics of TCR signalling and transcription factors^115,116.

Cre-inducible multicolour reporter mouse models can be used to study long-term cellular dynamics and lineage tracing. For example, transgenic Brainbow mice express a Cre-inducible brainbow cassette consisting of up to four fluorescent proteins under the control of the neuron-specific Thy1 promoter. Here, Cre recombination allows for stochastic fluorescent protein expression to individually label Thy1-expressing cells with an arbitrary combination of fluorescent proteins in order to map neuronal circuits in the brain^117,118. This concept was expanded upon by inserting brainbow into the Rosa26 locus (Confetti mice)¹¹⁹ or the β-actin locus (Rainbow mice) to induce fluorescent protein expression in any tissue or cell type of interest using Cre driver lines, which allowed live lineage tracing in developing embryos¹²⁰ and the mammary gland^66,121 or the in vivo tracking of stem cell fate in intestinal crypts^108,119, the mammary gland^121,122 and tumours.

Label-free techniques

The inherent fluorescent properties of endogenous molecules can also be used to provide a label-free context to IVM data. Label-free strategies have the advantage that they provide environmental information without the need for genetic modification or in vivo labelling of tissues. Label-free strategies exploit the excitation of select endogenous molecules by either linear or non-linear methods. For example, NADH, flavonoids and other small molecules are extensively used to visualize cellular architecture and metabolism by using either confocal or multiphoton microscopy¹²³. Less commonly used techniques that are technically challenging, such as coherent anti-Stokes Raman scattering and stimulated Raman scattering, and label-free ultrasound and opto-acoustic imaging, have been used to image, respectively, lipids and myelin, or DNA^88,124,125 and the vasculature or tissue oxygenation^126,127.

A commonly used label-free imaging method in IVM is higher harmonic generation, a light scattering process that arises when light interacts with electrons bound to molecules via non-linear restorative forces, causing the emission of photons with frequencies that are integer multiples of the illumination light¹²⁸. Substantial higher harmonic generation is only possible either with high-illumination intensities (such as occurring with high peak power ultrashort femtosecond pulses that simultaneously reduce the risk of photodamage and phototoxicity), highly non-linear restorative forces or in polymeric molecules where the monomers are spaced such that emitted photons from each monomer can efficiently transfer their momentum along the macromolecule to generate large aggregate signals¹²⁹. Second harmonic generation (SHG) results in the frequency doubling of photons. Biological structures efficient in generating SHG include fibrillar collagen, actomyosin and biomaterials^15,130. Third harmonic generation results in the frequency tripling of light and can originate from cell and tissue interfaces with a mismatch of refractive index, such as tissues that contain water–lipid and protein–lipid interfaces^131,132 (such as cell membranes or lipid droplets) or endogenous compounds such as metabolites^132,133.

Lastly, the visualization of endogenous fluorophores, such as the intracellular metabolites NAD(P)H and FAD, can be used as a label-free imaging technique to map cells and their metabolic state in live tissues¹³³. Metabolite IVM has, for example, previously shown promise for the detection or delineation of diseased and healthy tissues due to the differential expression and abundance of endogenous fluorophores¹³⁴. This ability for label-free FLIM imaging may also open up opportunities for applications in human volunteers and patients to diagnose disease tissue or the disease stage using IVM and endomicroscopy, demonstrating the bench to bedside capacity of in vivo imaging^135,136. Closely related to label-free imaging of endogenous molecules is the use of autofluorescence generated by molecules introduced into a biological system for other reasons. For example, some anti-cancer therapeutics are themselves fluorescent and this property can be used to track drug distribution and uptake^137,138.

Advanced techniques

Intravital photobleaching, photoactivation and photoconversion applications.

Whereas fluorescent labels can provide spatial information on a cell or protein of interest, tracking this signal over time can provide insights into molecular and cellular dynamics and behaviour in vivo^139,140.

Photobleaching techniques, such as fluorescence recovery after photobleaching (FRAP) (FIG. 3A) or fluorescence loss in photobleaching (FIG. 3B), can be used to assess the mobility of fluorescently labelled proteins or agents. FRAP involves the laser-mediated bleaching of the fluorophore in a defined region of interest followed by tracking of fluorescence recovery over time (FIG. 3A). This approach provides information on the mobile and immobile fractions of a molecule of interest, its half-time of recovery and the mode of movement — for example, lateral diffusion versus cytoplasmic exchange for membrane proteins. Fluorescence loss in photobleaching serves as a complementary approach where the loss of fluorescence adjacent to a bleached region is monitored over time (FIG. 3B). Intravital FRAP and fluorescence loss in photobleaching imaging have been used to assess cell–cell junction dynamics during cancer progression^68,141, subcellular protein dynamics in actomyosin cytoskeleton networks of the salivary gland^16,140,142, granule trafficking in dendrites of epidermal T cells¹⁴³, glutamate receptor dynamics in dendritic spines¹⁴⁴ or the dynamics of protein aggregation in the brain, which can often underlie neurodegenerative diseases¹⁴⁵. Similarly, intravital FRAP using fluorescent dyes can also be used to characterize changes in vessel leakiness upon wound healing¹⁴⁶, lymphatic viscosity and flow in a mouse model of arthritis¹⁴⁷, cell–cell communication via gap junctions^148–150 or molecule diffusion and interstitial connectivity in extracellular spaces^151,152.

Fig. 3 |. Advanced imaging techniques.

A | Fluorescence recovery after photobleaching (FRAP) imaging involves the laser-mediated bleaching of a fluorophore in a region of interest (yellow dashed circle) followed by monitoring of fluorescence recovery into the same region of interest (black dashed circle). B | During fluorescence loss in photobleaching imaging, the fluorophore in a region of interest is continuously bleached (yellow dashed circle) while observing loss of fluorescence in adjacent/connected regions (black dashed circle). C | Photoactivation studies use caged (non-fluorescent) fluorophores, which can be uncaged and activated by a laser pulse to emit a fluorescent signal (yellow dashed circle) followed by monitoring the loss of fluorescence in the same region of interest (black dashed circle). D | During photoconversion the spectral properties of a fluorophore are permanently changed, which can, for example, allow/facilitate the tracking of individual cells in large populations. E | Examples of Förster resonance energy transfer (FRET) biosensors: intermolecular FRET from the donor fluorophore (cyan) to an unlinked receptor fluorophore (yellow) (part Ea); cleavage-dependent FRET biosensors whereby an irreversible cleavage of the substrate peptide leads to irreversible loss of/decrease in FRET (part Eb); distance-dependent FRET biosensors whereby a reversible change in the linker region leads to a reversible modification in the distance between donor and acceptor fluorophores, thus affecting FRET efficiency (part Ec); and orientation-dependent FRET biosensors whereby a reversible modification in the linker region leads to a reversible change in the alignment between donor and acceptor fluorophores, thus affecting FRET efficiency (part Ed). F | Fluorescence lifetime imaging microscopy (FLIM) is used to quantify the fluorescence lifetime, which is the time a fluorophore spends in the excited state prior to emission of a photon, an inherent feature of each fluorophore, which also depends on environmental conditions. Parts A–C and E adapted from REF.¹³⁹, CC BY 3.0 (https://creativecommons.org/licenses/by/3.0/).

Photobleaching imaging can be complemented by photoactivation studies, wherein an initially caged (non-fluorescent) agent is activated via a laser pulse to emit a fluorescent signal (FIG. 3C). The expansion of this localized pool of activated fluorescent proteins (or loss of fluorescence in the activated region of interest) can be tracked over time (FIG. 3C) to provide information on protein mobility, movement and kinetics, such as tracking protein dynamics in live tumours¹⁴¹, the bile flux mechanisms in the live liver¹⁵³ or transcription factor dynamics in living embryos¹⁵⁴. For further details on the set-up of photobleaching imaging studies, we refer the reader to two publications^155,156.

Photoconvertible agents are another excellent tool for IVM studies that involve tracking of dynamic events, such as cell fate mapping^51,157, the movement of cancer cells^70,158–163 or immune cells^164–166, or the dynamics of subcellular compartments¹⁶⁷. Upon excitation at a specific wavelength, photoconvertible proteins, such as the green to red converting proteins Kikume, Kaede or Dendra2, for example, undergo an irreversible switch in their spectral properties due to a light-induced peptide cleavage (FIG. 3D). This can be explored to precisely label and track individual cells (FIG. 3D) with the aforementioned advantage over photoactivation that a fluorescent signal is already obtained prior to photoconversion.

FRET imaging and biosensors.

FRET involves the transfer of energy from a donor fluorophore to an acceptor fluorophore provided that the donor and acceptor display sufficient spectral overlap, proximity and alignment to each other (FIG. 3E). FRET can be quantified using ratiometric microscopy or FLIM.

Intermolecular FRET involves the energy transfer between unlinked donor and acceptor fluorophores, which can be used to assess heterotypic or homotypic protein–protein interactions¹⁶⁸ (FIG. 3Ea). Intramolecular FRET in FRET biosensors occurs between donor and acceptor fluorophores that are physically connected to each other via a linker substrate, and can be used to study molecular activity, such as proteolytic cleavage or phosphorylation by protein kinases. For example, cleavage-dependent FRET biosensors contain a linker sequence that can serve as a protease substrate, where protease-mediated cleavage of the linker irreversibly separates donor and acceptor fluorophores from each other, decreasing FRET efficiency (FIG. 3Eb). Examples of cleavage-dependent FRET biosensors include biosensors to assess caspase activity (which coordinates cell apoptosis)¹ or matrix metalloproteinase activity (which regulates remodelling and turnover of extracellular matrix)¹⁶⁹.

Proximity-dependent and orientation-dependent FRET sensors contain a linker that serves as a substrate for post-translational modification, thereby affecting the proximity and alignment of donor and acceptor and, subsequently, FRET efficacy (FIG. 3Ec,Ed). These reversible FRET biosensors include reporters for kinase activity such as ERK^170–174, AKT/PKB^175,176, SRC^69,177,178, FAK^84,179,180 or ROCK^181,182 and can provide information on drug target inactivation dynamics in vivo to see whether, for how long and to what extent a given drug target can be inhibited. This information can be used to fine-tune drug treatment regimens in vivo for maximized drug target engagement while minimizing overtreatment, which is often associated with side effects and toxicities^{69,84,135,159,177,183,184}.

Furthermore, proteins themselves can serve as a linker where changes in protein domain binding or protein folding and configuration determine FRET efficiency. Examples include the Raichu FRET biosensor series of small GTPases where the small GTPase binds to or dissociates from the binding domain dependent on GDP (inactive) or GTP (active) loading, thereby affecting FRET efficacy to quantify rapid changes in these notoriously difficult to assess signalling events in live tissue^159,184,185. Similarly, IVM of a glucose FRET biosensor, which contains the bacterial MglB protein as a substrate for reversible glucose binding, was used to reveal single-cell heterogeneity of breast cancer cell metabolism in vivo with higher levels of intracellular glucose detected at the tumour borders¹⁸⁶. Lastly, FRET biosensors of microenvironmental conditions have also been described, where changes in oxygenation or pH can lead to altered fluorescent protein folding of biosensor configuration, which consequently affects FRET efficacy^187,188.

Exciting new developments allow multiplexed imaging of several FRET biosensors to simultaneously assess several biological or molecular processes at the same time. Especially, the generation of new red-shifted and far-red-shifted FRET pairs facilitated the simultaneous imaging of several FRET biosensors^{81,83,189,190} with application for deep in vivo imaging as well as multiplexing with optogenetic tools to switch on distinct signalling pathways^190,191. Recent live cell imaging studies have further advanced the field and highlight what may be possible in IVM in future by specifically interrogating protein activity at a distinct/endogenous subcellular location using split fluorescent proteins and split FRET biosensors¹⁹², mapping of signalling molecule gradients using FRET-based nanorulers¹⁹³ or the tracking of entire signalling networks using biosensor barcoding in combination with deep learning-based analysis approaches¹⁹⁴. For a detailed overview on FRET biosensors, the reader is referred to the following review articles^180,195,196.

Fluorescence lifetime imaging microscopy.

FLIM can be used to quantify the fluorescence lifetime, which is the time a fluorescent molecule spends in the excited state before emitting a photon and returning to the ground state (FIG. 3F). The fluorescence lifetime is an inherent property of each molecule which varies depending upon the molecule’s environment. The fluorescence lifetime can therefore be used to identify individual fluorescent species (FIG. 3F). FLIM imaging can help to resolve or separate signals from overlapping spectra or be used to quantify changes in FRET efficiency, for example in FRET biosensors. This has been used to assess many key aspects of in vivo biology such as calcium dynamics in B cells in vivo¹⁹⁷, real-time signalling dynamics during cancer cell extravasation into the live liver¹⁷⁸, to map endogenous fluorophores such as the intracellular metabolites NAD(P)H and FAD, in live tissues^133,198 or for drug target engagement in a model of ovarian cancer metastasis using intravital endomicroscopy¹⁹⁹.

Image acquisition

Microscope acquisition parameters such as the frame rate, resolution and magnification are dictated by the biological processes of interest, but must also consider limitations of the microscope, the signal source and the tissue properties as well. In principle, aiming for the Nyquist condition — where acquisitions are performed at more than twice the highest spatial or temporal frequency component in each of these parameters — is ideal. However, the interplay between the frame rate, magnification and light exposure means that this ideal is not always attainable. For example, higher magnification often means a laser beam scans over a tissue at a slower rate, increasing the amount of light exposure, phototoxicity and photodamage. Attempting to capture rapid dynamics requires high frame rates, but high frame rates increase the overall exposure of the tissue to light and may lead to photodamage. Optimal conditions attempt to maximize tissue exposure while minimizing photodamage and phototoxicity. Although this is highly dependent on the microscope and experiment, there are some guidelines that can be provided. Resolution is limited by either the optical resolution (laser spot size determined by the numerical aperture of the objective lens) or the pixel size (determined by the scan angle and magnification of the objective). Scan angles and line scan speeds determine the pixel dwell time (how long a laser spends illuminating a pixel). Longer pixel dwell times increase exposure, but acquisition electronics typically do not capture this increased signal²⁹. In addition, continuous excitation may induce photodamage more rapidly than allowing molecules a period of rest. Thus, it is better to increase the signal-to-noise ratio by averaging multiple scans, rather than increasing the illumination intensity. When averaging, line averages can accommodate motion artefacts better than frame averages. Phototoxicity and photodamage can be assessed by recording a large number of images in succession and plotting the signal intensity over time, and by observing the tissue behaviour in real time (for example, with cell stress reporter activity) or cell death and inflammation after imaging. The highest laser power that does not cause cell stress and/or photodamage at a given magnification and scan angle should be used.

Overcoming physiological motion

Physiological motion, such as from respiration, heartbeat, peristalsis or muscle and vascular tone, can affect tissue geometry over longitudinal IVM sessions and lead to blurred images²⁰⁰, and is one of the major challenges of IVM regardless of the type of IVM imaging platform. Motion correction and compensation strategies include gating approaches to synchronize image acquisition to the cardiac and/or respiratory cycles in order to obtain IVM data when physiological motion is minimal. However, it should be noted that this approach can result in a trade-off for temporal resolution as IVM only occurs at dedicated time points. Many studies have focused on suppressing sample motion during IVM. This often poses a trade-off or compromise between stabilizing the sample without introducing artefacts that might affect the biology to be observed. In order to reduce motion of surgically exposed tissues or readily accessible tissues, such as the tongue, during IVM, custom-made holders can be used with care to not interfere with the blood circulation between the exposed tissue and the animal^16,201,202. Optical imaging windows can also be surgically inserted above the tissue of interest with adhesives between the tissue and the window frame providing stability^{50,52,53,70,203}. Additional stability for IVM through optical imaging windows can be achieved if the window can be fitted into a custom-made window stage, which can be readily added to an inverted IVM microscope set-up. Similarly, stereotactic head holders have been glued to surgically exposed skull bone to stably assess the interactions between osteoblasts and osteoclasts during bone turnover^204,205. Extremely slow, non-periodic movements such as tissue growth can also occur which do not necessarily impede IVM data acquisition itself but might have to be accommodated for during longitudinal IVM studies. Here, the suture-less insertion of silicone-based imaging windows above the tissue of interest for up to 35 days can prevent growth restriction^55,56.

Results

Correcting for physiological motion

In addition to physical restraint, real-time or post-IVM correction approaches have also been used. For example, an off-path light beam was used to measure motion displacement followed by adjustment of the objective position in order to correct for axial motion^206,207. In another example, a feedback system was used that is based on anatomical 3D landmarks, which can be obtained from the SHG signal or blood vessels, in combination with high-speed computing to correct for sample drift²⁰⁸. Furthermore, the ImageJ macro Intravital_Microscopy_Toolbox can be used to automatically remove severely distorted images if they exceed a dissimilarity score to a reference frame²⁰⁹. The software tool IMART achieves image stabilization using alignment and similarity to a reference frame by rigid registration of images to correct for overall movement followed by non-rigid registration to correct for small-scale aberrations/movements²¹⁰. StackReg (recently modified as HyperstackReg to work on 4D hyperstack data) is a sub-pixel registration algorithm for Fiji that minimizes the mean square difference of intensities between images^211–213. Similarly, Galene was developed to correct for physiological motion in FLIM data, for instance when performing FRET measurements in intestinal crypts, which are subject to peristaltic movement¹⁷⁸. FIGURE 4 shows the Galene workflow for correction of FLIM–FRET data correction, which can also be followed to correct motion in other IVM data sets.

Fig. 4 |. Overcoming tissue motion.

General workflow for motion correction of intravital fluorescence lifetime imaging microscopy (FLIM)–Förster resonance energy transfer (FRET) data using the software Galene. Physiological motion, such as respiration, heartbeat, peristalsis or muscle and vascular tone, can render intravital microscopy (IVM) data unintelligible as shown here for intravital FLIM–FRET in the intestine as an example. After extracting the photon arrival times, individual frames are reconstructed and aligned against a reference frame based on fluorescence intensity information. Following alignment, the individual photon arrival positions are reassigned for each frame to allow quantification of fluorescence lifetimes per pixel. Similarly, Galene can also be used to correct for motion in other IVM data including fluorescence time series or endogenous fluorescence data. ECFP, enhanced cyan fluorescent protein; SEYFP, super enhanced yellow fluorescent protein. Adapted from REF.¹⁷⁸, CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).

Segmentation and quantification

IVM can provide highly complex information and many routinely encountered challenges, including moving image position, low signal and inconsistent signal strength between channels, can be mitigated by using image processing, such as broadly used open-access software (ImageJ/FIJI, Cell Profiler). Segmentation can help to select those data for subsequent quantification which are of relevance to the experiment, such as specific fluorescence channels and spectra, individual cells or cell populations, or subcellular structures. This can, for example, be achieved by manual selection, fluorescence intensity thresholding as well as supervised, semi-supervised or automatic data segmentation. Tracking tools, such as the TrackMate^214–216 plug-in for ImageJ/FIJI or the Icy plug-in Spot Tracker²¹⁷, can be used to segment cells in an image and track their dynamics during an acquired time series. This information can then be used for track visualization and quantification of dynamic cell behaviours, for example, cell migratory parameters, such as cell displacement, migration speed or directionality. Similarly, surface segmentation — provided in the commercial software Imarisor Amira — can be used to delineate the boundaries of cells or other structures in volumetric data for subsequent quantification of parameters such as cell volume, shape or position. To improve the speed, objectivity and reproducibility of IVM data analysis, deep learning methods have been developed. These approaches, which are based on the training of computer algorithms or artificial neural networks, are not mainstream yet but are progressively integrated into segmentation of IVM data. For machine learning assisted data segmentation, the user can initially train a model by manually labelling areas containing a signal compared with areas containing no signal, from which the neural networks can then extract thresholds to develop a model. If the model was sufficiently trained, it can segment all remaining data. This concept was used in an IVM database of leukocyte movement, a highly heterogeneous and plastic cell type. Manually annotated tracks for different leukocyte populations upon exposure to a range of stimuli served as a training and validation data set for machine learning approaches to automate cell migration and tracking analysis²¹⁸.

Furthermore, a cell segmentation approach based on a trainable clustering algorithm was developed, which separates highly heterogeneous immune cell populations into individual cells²¹⁹. Similarly, a semi-supervised learning approach assisted by pixel classification was used to segment and track immune cell dynamics in the lymph node in vivo²²⁰ as well as fluorescently labelled synapses in the cortical volume of mice upon whisker stimulation¹⁴⁴. In addition to cell segmentation algorithms for specific data sets, U-Net, a plug-in for ImageJ/Fiji, although pre-trained for cell segmentation and detection, allows the assignment of new tasks²²¹, and has recently been used to track cells in the live retina from phase contrast data²²² or to segment cells for subsequent ratiometric FRET analysis of small GTPase activity²²³. In order to improve accessibility to deep learning tools, several platforms have been released, providing a range of pre-trained machine learning-based models for image processing and analysis without requiring extensive coding skills, including the ImageJ/Fiji plug-in DeepImageJ²²⁴ which is compatible with BioImage Model Zoo, an open repository for deep learning models developed by and for the scientific community, and ZeroCostDL4Mic, a platform which allows users to train and run deep learning models on free to access cloud spaces, thus reducing the amount of computing hardware required for deep learning assisted analysis²²⁵. For a detailed discussion on best practices in the use of deep learning models for image analysis, we refer the reader to the following excellent article²²⁶.

Quantification

Depending on the biological question, quantification of IVM data following segmentation can include quantification of fluorescence intensity, to analyse changes in gene expression (if the fluorophore is coupled to a protein); quantification of size or shape, to assess changes in cellular or subcellular morphology; and quantification of x–y–z coordinates/location over time, to track subcellular dynamics, migration of individual cells or cell populations, as well as cell–cell or cell–environment interactions. Furthermore, quantification of microenvironmental parameters, such as the extracellular matrix, can reveal dynamic changes in matrix deposition or remodelling, which can lead to fibrosis under pathological conditions and accompanies diseases, such as cancer or cardiovascular disease^227,228. Similarly, quantification of vascular features can be performed to monitor vascular structure, flow, perfusion and patency. Analysis of fluorescently labelled drugs can be used to assess drug uptake, distribution or metabolism to provide valuable information about drug pharmacokinetics and pharmacodynamics in vivo in a preclinical context, which may inform future clinical applications or mechanisms of action^135,136,183.

Available software

Paralleling the vast range of potential IVM readouts, there is an abundant source of software tools available to the scientific community to facilitate data analysis. Many software packages have been developed, which are bundle solutions to image data correction (motion/drift correction post IVM) and segmentation/quantification/visualization (cell tracking, quantification of cellular/subcellular/molecular features). Proprietary software packages include IMARIS, Amira, MetaMorph, ImagePro Plus and Arivis Vision 4D, whereas open-source applications include FIJI, QuPath, Ilastik and Cell Profiler. Supplementary Table 1 presents a non-exhaustive list of image analysis tools that were made available to the scientific community over recent years. In an effort to incorporate user-driven need and collaborative effort, ImageJ is an open-source image analysis software that readily allows addition of new features or analysis workflows, which are distributed and added to ImageJ via plug-ins and macros^229,230. In order to expand upon its applicability, ImageJ2 was released, which for example allows the import, handling and analysis of multidimensional data, data types and size²³¹. Fiji serves as a bundled release of ImageJ2 with pre-selected plug-ins, thus presenting a more complete installation package²³². A comprehensive overview of the diverse applications of ImageJ and its derivatives is presented elsewhere^233,234. Similarly, the platform TAPAS (Towards an Automated Processing and Analysis System) has recently been published with an enhanced focus on image data organization and the establishment and sharing of workflows/protocols for automated image processing in order to improve data standardization and reproducibility²³⁵.

Applications

IVM has provided seminal insights into the dynamics of biology across many different research fields, ranging from neurobiology and stem cell biology to cancer biology. As it is impossible to highlight all applications of IVM, we provide a few examples in the field of expertise of the authors.

Imaging immunity

The ability of IVM to visualize the immune response in an ever-broadening array of tissues and physiological situations has dramatically expanded — and fundamentally altered — our understanding of inflammation and immunity. IVM has allowed the mapping of the multistep leukocyte recruitment cascade, defining specific, sequential and regulated stages of leukocyte tethering, rolling, activation, adhesion and transmigration, identifying the role of selectins, integrins and chemokines as well as interaction with tumour cells followed by cytotoxic effector function^236–239. Subsequent application of fluorescently labelled antibodies and reporters, and of confocal and multiphoton microscopy, further refined this understanding, allowing differential labelling of leukocyte subsets and improved resolution over conventional light microscopy^240–242 (FIG. 5). Although these early studies were limited, they were ground-breaking in the adoption of fluorescent reporters²⁴¹, the adaptation of laser-scanning confocal microscopy²⁴⁰ and the use of charge-coupled device (CCD) cameras to achieve high imaging frame rates²⁴². Recently, IVM recordings of immune cells were used to introduce a new framework whereby a large number of motility and shape parameters can be quantified and analysed in an automated way, using computational tools typically employed for transcriptional and proteomic profiling. These behavioural profiles can then be used as a metric to determine the impact of individual genes and proteins on the behavioural state of individual cells²⁴³. Moreover, adaptation of intravital approaches to additional vascular beds, including the application of epifluorescence to the mouse brain²⁴⁴, spinning disc imaging to the liver²⁴⁵ and two-photon imaging to the breathing lung²⁴⁶, has identified tissue-specific exceptions to this central leukocyte recruitment cascade^244,246,247, further refining our understanding of inflammation and immunity.

Fig. 5 |. Bone marrow imaging in situ in the leukaemia mouse model.

Representative intravital two-photon skull images. Bone tissue was detected by second harmonic generation (SHG), which captures collagen molecules without staining (represented in blue). Scale bar: 50 μm. a,b | Images of mouse LIN⁻SCA1⁺KIT⁻ (LSK) cells within the bone marrow niche: mouse LSK-tdTomato cells transplanted into Cxcl12-GFP recipient mouse (green, CXCL12-abundant reticular (CAR) cells; red, LSK cells) (part a); and mouse LSK-tdTomato cells transplanted into Col1a1 2.3-CFP mouse (cyan, osteoblasts; red, LSK cells) (part b). c | Intravital image of a murine chronic myeloid leukaemia model made by transplantation of leukaemia cells (BCR-ABL-ires-GFP transfected LIN⁻SCA1⁺KIT⁺ cells). Green, leukaemia cells; red, blood vessels. d | Intravital image of a murine acute myeloid leukaemia (AML) model made by transplantation of leukaemia cell line (C1498-eGFP). Green, leukaemia cells; red, blood vessels.

IVM has also been transformative in our understanding of cell–cell and cell–pathogen interactions. Using models of bacterial infection and spinning disc confocal microscopy, the function of tissues such as the liver in pathogen clearance has been elucidated, identifying roles for both antibody-mediated and independent bacterial capture^248,249. Other elegant studies have employed multiphoton imaging approaches in animals with cells expressing fluorescent protein reporter molecules and labelled antigens to define mechanisms involved in immune-complex drainage through lymphatics and antigen trafficking within lymph nodes^250–252, lymphocyte entry, trafficking and retention within secondary lymphoid tissues²⁵³, the initiation of neutrophil swarming within sites of infection and/or damage²⁵⁴ and graft rejection following transplant²⁵⁵. Wide-field microscopy has helped to define the role of high endothelial vessels in regulating cellular entry to lymph nodes and to tumours^256–258, whereas other work using CCD-based epifluorescence has been critical in defining shear-dependent alterations to neutrophil behaviour²⁵⁹. Additionally, the balance achieved between image resolution and the acquisition rates possible with spinning disc confocal imaging has been essential to our evolving understanding of numerous immune effector mechanisms including the formation and function of neutrophil extracellular traps^260,261, and understanding how viral infection reprogrammes the host immune response²⁶². Within the context of viral infection, the higher resolution achieved with resonant scanning confocal IVM has allowed for the direct visualization of viral delivery to tissues^263,264, and has allowed for the mapping of the innate immune response to acute viral infection²⁶⁵. Over the past decade, IVM has also brought the role of the platelet in the host immune response to the fore. Imaging of direct platelet–neutrophil interactions, with a range of platforms from CCD-camera based epifluorescence to two-photon microscopy, has identified roles for platelets in neutrophil recruitment and activation^266–268, helping to facilitate lymphocyte adhesion and antigen recognition within the virally infected liver²⁶⁹ and as key contributors to immunothrombosis^102,270. More recently, the imaging depth possible with two-photon microscopy has allowed us to observe the function of the immune system within previously inaccessible tissues such as the bone marrow, providing single-cell resolution analysis of engineered T cell therapies in murine models of cancer^271,272.

Stem cell behaviour in solid tissues

Adult stem cells self-renew and give rise to all specialized cells in the body, and are therefore the key driver cells for tissue homeostasis, regeneration and diseases such as cancer. Our knowledge on stem cells was predominantly derived from static genetic and molecular data that defined stem cells as a homeostatic pool of quiescent cells that only occasionally give rise to transient amplifying cells, which in their turn give rise to a burst of specialized offspring. However, pioneering IVM studies in hair follicle, intestine and pubertal mammary epithelia drew a completely different picture, and showed that adult stem cells can be highly migratory and proliferative and that their behaviour and fate are dynamic, heterogeneous and compartmentalized^{108,121,273,274}. Stem cells are identified in IVM experiments directly by a fluorophore that is expressed by a stem cell promoter, or indirectly by their self-renewal behaviour and potential using fluorescent lineage-tracing approaches. The latter requires multi-day imaging of the same cells and is therefore often combined with optical windows. One particular aspect which has been elucidated using multi-day IVM is the high degree of plasticity in their behaviour and fate. In the skin, IVM experiments revealed that stem cell fate decisions (that is, symmetric or asymmetric division) are directed by the behaviour of the neighbouring cells²⁷⁵ and the location of the stem cells within the niche²⁷³. In the hair follicle, IVM showed that stem cells move along the niche during the hair follicle growth cycle, and, depending on their position, these stem cells generate distinct differentiated cell types of the hair²⁷⁴. Multi-day imaging of the pubertal mammary gland showed that stem cell movement enables stem cells to enter various positions and niches so that they all have equal behaviour and fate^121,122. IVM also showed that cell movement is important for the fate decisions of intestinal tissues^276,277. Differentiated cells have been shown to move away from the stem cell niche²⁷⁷ and migration in the reverse direction has also been shown to enable cells to enter or stay in the stem cell niche and, therefore, function as stem cells regardless of their molecular profile²⁷⁶. Lastly, multi-day multiphoton microscopy directly visualized the competition between wild-type intestinal stem cells and intestinal stem cells that acquired a (oncogenic) mutation²⁷⁸. Interestingly, imaging every few weeks the same area of surgically exposed intestine with multiphoton microscopy showed that caloric restriction increases the number of functional stem cells per stem cell niche, leading to slower but stronger competition and a more frequent loss of mutations^279,280 (FIG. 6). Together, these IVM studies highlight that stem cell fate and differentiation potential are highly plastic and dependent on environmental cues.

Fig. 6 |. Intravital microscopy of calorie restriction reveals slower and stronger stem cell competition.

a | By repeated intravital microscopy (IVM) of the same crypts 48 h and 12 weeks after fluorescent labelling of the stem cells, stem cell competition can be visualized. Stem cell clones are either lost or remain within the crypts, represented by the presence of red cells. b,c | IVM images of the same intestinal areas in control conditions (part b) and upon caloric restriction (part c). Dotted lines highlight the crypts in which labelled stem cell clones were present at 48 h and 12 weeks after stochastic fluorescent labelling. Zoomed-in images (bottom panels) show examples of clones that were lost (1 and 3) or remained (2 and 4) in both conditions. Scale bars represent 250 μm (top panels) or 50 μm (bottom panels). Adapted with permission from REF.²⁷⁹, Elsevier.

Imaging haematopoietic and leukaemic stem cell behaviour and dynamics.

IVM has contributed to our understanding of the behaviour and dynamics of haematopoieticstem cells (HSCs) and leukaemic stem cells (LSCs) that reside within the bone marrow through minimally invasive surgery to expose the skull calvarium or tibia and imaging of the bone marrow cavities using time-lapse multiphoton microscopy (FIG. 5). It has been important to reveal spatio-temporal interactions of HSCs or LSCs with their bone marrow niche, which supports HSC maintenance and haematopoiesis or which is used by LSCs to escape from immune cell attacks.

Transplantation of stained HSCs into mouse calvaria bone marrow revealed that steady-state HSCs engrafted near the endosteum and resided within the endothelial niche^11,281–283. In addition, differentiating progeny of HSCs reside within their specialized bone marrow niches, which can regulate the dynamics, localization and proliferation of stem and progenitor cells²⁸⁴. To precisely analyse the behaviour of HSCs in vivo, transgenic mouse strains with fluorescently labelled HSCs are key tools, such as the α-catulin-GFP and Fgd5-mCherry strains, in which HSCs and vascular endothelial cells are labelled²⁸⁵, and the Hoxb5-tri-mCherry (REF.²⁸⁶), Mds1^GFP/+Flt3^Cre (REF.¹⁰) and Pdzk1ip1 (REF.⁹) strains, in which only some of the long-term HSCs are labelled. Even though these strains are useful for HSC imaging, there is still no ideal HCS-specific reporter mouse strain that labels all phenotypically identifiable HSCs with bright fluorescence, and it is essential to develop new reporter mouse strains for further studies. Imaging analyses of leukaemic cells in mouse bone marrow have also advanced in recent years using time-lapse multiphoton microscopy. IVM studies showed that B cell acute lymphoid leukaemia cells resided within an endothelial bone marrow niche²⁸³, and interact with the vascular endothelium²⁸⁷. Moreover, T cell acute lymphoid leukaemia cells were shown to be highly migratory, especially following chemotherapy treatment¹³. By contrast, several studies reported localization of acute myeloid leukaemia (AML) cells around the endosteum. Bone marrow imaging of MLL-AF9-driven murine AML showed that remodelling of the endosteal region by local AML cells resulted in a reduced supporting capacity of the endosteal niche for HSCs²⁸⁸, and that a subset of AML cells are more migratory than the rest²⁸⁹, potentially taking a role akin to that of invasive tumour cells. In addition, IVM showed that the chemotherapy drug cytarabine suppressed cell motility of the transplanted mouse AML cell line via the RHO–MRTF–SRF pathway, suggesting a relationship between residual AML cell dynamics and chemo-resistance²⁹⁰. Recently, it has become possible to visualize the function of cells with their dynamics in living bone marrow using fluorescent probes. For example, by applying a metabolic sensor to human AML cells, it was shown that high-glycolysis LSCs prefer to reside within the endosteal niche and proliferate to maintain their leukemogenic activities²⁹¹. Also, dynamic measurements of a real-time indicator of mTORC1 activity in other mouse AML models showed that the mTORC1 activity of AML cells varied according to their distance to the bone and AML cells had high mTORC1 activity under cytarabine chemotherapy²⁹².

Progression of solid tumours

IVM has been important to reveal dynamic processes that drive several distinct progression steps of solid tumours of epithelial origin (such as breast and pancreas tumours). At the very early stage of tumour development, solid tumours are confined within a basement membrane and are identifiable by the presence of tumour cells filling up and expanding individual ducts, acini or lobules. The process of cellular expansion within these spatially well-defined units has been visualized through mammary imaging windows by time-lapse multiphoton microscopy¹²¹. It was found that during the in situ stage, the observed cellular movement is predominantly passive and occurs as a consequence of increased proliferation of cells located near the basement membrane, which push luminal cells further towards the centre of the unit (FIG. 7). It is widely thought that dissemination to distant sites requires cancer cells to actively degrade this basement membrane. IVM through organ-specific imaging windows and time-lapse multiphoton microscopy has demonstrated that interstitial tissue invasion is mediated by single-cell dissemination as well as collective invasion, which can be visualized using genetically encoded reporters for cytoplasm or membrane markers and nucleus^293,294. By co-registration of tumour cells with SHG and third harmonic generation, the guiding tissue structures and tracks of least resistance can be detected, including collagen bundles, perfused blood vessels, myofibres, nerves and adipocytes¹³¹.

Fig. 7 |. Intravital microscopy of the ductal carcinoma in situ in the PyMT mouse model.

a | Large-volume high-resolution intravital microscopy (IVM) of the ducts provides an overview of the tissue and allows identification of the tissue stage and of incompletely filled ducts (white box). A 5 × 5 mosaic of the mammary gland imaged through a mammary imaging window implanted in a transgenic mouse model that expresses the PyMT oncogene and the fluorescent proteins Dendra2 in the mammary epithelium (green) and CFP in macrophages (blue). b | Stills from a continuous time-lapse movie of the duct showing migration of the cells over 7 h imaging duration. Red outline indicates position of cells at time t = 0 min. After only a brief imaging period (left), displacement of the cells is not obvious. However, after 7 h of imaging (right), cells have moved appreciably. Dashed line indicates the distance the cells have moved and can be used to calculate the speed of cell movement during imaging. c | Ki67 staining (brown) of fixed tissues taken from another mouse shows ~60% of cells are actively proliferating in similar structures. d | High-resolution stills from the IVM movie captures the division of a single cell (white dashed outline drawn by hand indicates the boundary of the dividing cell and the daughter cells), as evidenced by chromosomal separation over ~30 min.

IVM has further led to a paradigm-shifting observation demonstrating that cancer cells can intravasate during the in situ stage²⁹⁵. This observation explained the finding of disseminated tumour cells (DTCs) in the bone marrow of patients with in situ breast cancer²⁹⁶. Using IVM, the patterns of invasive cancer cell migration, single cell^7,297 and collective^{294,298–300}, as well as the effect of cancer cell interaction with non-cancerous cells within the tumour microenvironment on cancer cell phenotype^{109,212,260,261,301} have been described in detail. For example, IVM studies indicate that whereas intravasation of cancer cells into lymphatics occurs by collective cell migration²⁹⁸, intravasation of cancer cells into blood vessels occurs by single cancer cells^{7,106,212,302}. IVM showed that in breast cancer, invasive cancer cells intravasate through tumour microenvironment of metastasis doorways, stable structures on blood vessels composed of a pro-angiogenic macrophage, a cancer cell overexpressing the actin regulator Mena and an endothelial cell. Tumour microenvironment of metastasis doorways are sites where intravasation of single cancer cells has been observed in real time^{7,212,302,303}.

The latest stages of solid tumour progression, such as extravasation of DTCs and their outgrowth at the secondary sites, has been studied by multiphoton microscopy through specific imaging windows in several models. IVM studies revealed the fate of DTCs immediately upon extravasation into the brain³⁰⁴, liver^53,305,306, bone²⁸³ and, most recently, lungs³⁰¹. For example, IVM demonstrated that the primary and metastatic tumour microenvironment induces changes in the cancer cell states (including stemness¹⁰⁹, epithelial-to-mesenchymal transition¹⁰⁶ or a dormant phenotype^260,301), which in turn drives cancer cell dissemination by providing these cells with a survival advantage within the metastatic site, and the potential to grow into metastatic nodules when conditions become suitable. Furthermore, IVM determined that cancer cells can re-disseminate haematogenously from already formed metastatic nodules in the lymph nodes^307,308 and lungs^52,161, potentially increasing the tumour burden.

Imaging haemostasis and thrombosis

Thrombosis, leading to the occlusion of vessels and, subsequently, to strokes and heart attacks, is a major cause of morbidity and mortality worldwide. To study this lethal process, several thrombosis models for IVM have been developed³⁰⁹, leading to the discovery of unexpected mechanisms which fostered drug development to target these major human diseases. For example, a ferric chloride-injured thrombosis model in mesenteric arterioles has been developed³¹⁰ and, later, used to visualize the formation of platelet plugs, thereby revealing a role for fibronectin in this process^311–313. An alternative laser-induced endothelial injury model was also developed, and confocal and wide-field microscopes were used to image in real time the formation of a thrombus. The authors combined multiple fluorescent channels with a bright-field channel, and visualized the deposition of platelets, accumulation of tissue factor and generation of fibrin³¹⁴. Furthermore, IVM showed that fully activated platelets were overlaid with unstable shell and less-activated platelets³¹⁵. Additional studies have made use of activatable protein substrates to reveal protease activity in living animals, defining where and when thrombin, the terminal enzyme in the clotting cascade, is active in the blood vessels^102,316,317. These studies have been critical in knowledge linking infection, inflammation and coagulation^{265,318–320}. Moreover, these types of IVM experiments have helped to find new potential ways to prevent thrombosis. For example, an IVM study has found a potential link between apoA-IV and platelet activity, suggesting that raising this plasma protein, either upon infusion of this protein or upon increased synthesis on consumption of unsaturated fat, may act as a strategy to prevent thrombosis and strokes³²¹. Other studies have attempted to uncouple infection and coagulation, allowing us to prevent pathological immunothrombosis while preserving the critical functions of haemostasis^322,323.

Imaging subcellular processes

High-resolution IVM depends upon procedures minimizing motion artefacts caused by the heartbeat and respiration, improved high numerical aperture optics, high scan speed⁶⁴ and rapid engineering of mouse models expressing suitable bright reporters. These advances allow the study of in vivo cell biology at a level of subcellular detail previously reserved to in vitro microscopy, thus providing the unique opportunity of unravelling how intracellular events are controlled by the tissue environment and how they contribute to tissue pathophysiology. Although confocal and spinning disc microscopy are an ideal choice for subcellular imaging, their use is restricted to a maximum of 50–75 μm from the tissue surface. Multiphoton microscopy is the obligatory choice for deeper penetration in tissues. For example, seminal studies of transcytosis of albumin and folate receptor carried out by two-photon microscopy in the kidney^324,325 led to the discovery that a significant fraction of proteins is recaptured in the kidney proximal tubules. This finding revealed an additional mechanism aside from the well-established glomerular filtration, with significant consequences for the understanding of kidney diseases and the development of novel therapies^326,327. Work in various secretory organs, in immune cells and cancer models, has revealed how unique combinations of extracellular cues are transduced into cascades of intracellular signals that trigger the assembly and activation of force-generating structures to remodel intracellular membranes^{16,79,90,105,201,328}. For example, stimulation of β-adrenergic signalling in exocrine glands elicits the assembly of a novel actomyosin structure composed of geodesic cages, revealed by spinning disc microscopy, which drive the regulated exocytosis of secretory vesicles¹⁶, a process that is coordinated with the activation of mitochondrial dynamics and bioenergetics³²⁹. Another study performed by two-photon microscopy revealed that during neutrophil extravasation the autocrine/paracrine release of chemokine signalling activates myosin II to control integrin trafficking to drive cell adhesion on the endothelium and to generate the forces required for trans-endothelial migration³³⁰. Notably, these unique structures and processes have not been described in cell or organ cultures, further highlighting the power of this approach in bridging cell biology with organ pathophysiology^{16,17,105,328}.

Reproducibility and data deposition

Regardless of the chosen imaging modality, IVM data are typically voluminous compared with conventional digital imaging data, as they contain multidimensional spatial information, temporal information by time-lapse imaging and multichannel information for multiple fluorescent labels. The raw data can be of multiple terabytes in size, and often researchers can only save processed data after analysis, which requires systematic and efficient management of processing, communication and information security to prevent mistakes. Building a system to deposit and back up IVM data in laboratories is therefore necessary for proper data management, and managing the data on the computer controlling the microscope is not recommended. Connecting the computer to the network poses a risk of data leakage and storing high volumes of raw data on microscope-controlling computers can interfere with the control of the microscope. Raw data should therefore be transferred to and stored on external devices and network-attached storage connected to a closed network. When performing analyses on other platforms or computers, the required data can be requested from the network-attached storage, and after processing the processed data are also stored and managed on the network-attached storage. Backing up the data is even more important because it requires much effort and time to acquire a single image in IVM experiments. Even if the requested data are lost, they can be recovered if the raw data remain on the external server. Sufficient data storage capacities and net bandwidth for exchanging data of multiple terabytes in size are essential.

Biological image management systems that manage imaging data together with metadata related to microscope information and imaging conditions are available online. OMERO (Open Microscopy Environment Remote Object)³³¹ and Bisque (BioImage Semantic Query User Environment)³³² are frequently used and good examples of web-based open-source repositories. OMERO can be used to further improve/streamline the workflow from image processing to data visualization, which also allows deposition of image data in a secure repository and can help to improve the organization, analysis and sharing of big IVM data^331,333. OMERO supports more than 150 image file formats and converts them to an OME-based model that any software can use, and the platform and related toolkits are frequently updated^334,335. Bisque supports more than 100 image file formats, and all related data have a unique Uniform Resource Locator (URL) to access them via a web browser. The repository also provides a 3D multi-fluorophore image viewer. The RIKEN Center for Advanced Photonics in Japan also released a cloud-based system for sharing, processing and visualizing 3D biomedical images³³⁶. These repositories are useful when applying previous analyses to other projects or data. However, sharing and comparing data remain difficult due to the complexity of imaging technology. Details on the experimental animal (including genotype, sex and age) and imaging details (including excitation and emission wavelengths, fluorophores and used objectives) should be included as metadata. Full recommendations for the standardization of life science imaging data formats and the improvement of data sharing were announced in 2021 to advance open science by creating an environment in which large amounts of imaging data can be shared appropriately³³⁷. In the future, data management in a standardized format may become a global standard.

Reproducibility in live imaging results

The generation of each image is extremely complex, and therefore all imaging settings need to be well annotated. A high number of samples is required for IVM because of large individual differences among animal models, and even if the imaging conditions are the same, the appearance may change depending on the conditions of the microscope system. Furthermore, due to the limited observation range, it takes an appreciable amount of time and effort to generate one image. Imaging data are not only used for qualitative analyses but also for image processing and quantitative biological measurements to demonstrate that the data are reproducible. Although many tools have been developed for quantitative analysis, detailed methods and parameters must be chosen subjectively. These details are often not reported in papers given their word restrictions and that no unified method can be described. However, even relatively small differences in unreported processing methods can lead to conclusions with inconsistent results and many differences. Therefore, future establishment of standardized reporting guidelines to accurately and completely describe all image processing methods can help to improve the reproducibility of IVM studies.

Limitations and optimizations

The main limitations of IVM are linked to resolution, throughput and the use of animals both ethically and as a model for human disease (BOX 1). Although important differences exist between mice and humans, when appropriately designed, murine models can be strikingly similar in physiology and pathology with respect to disease initiation and progression including the driving genetic alterations and factors such as inflammation. Yet the applicability of murine models to human disease needs to be evaluated for each research question. Mice remain an important animal model to obtain fundamental knowledge on stimuli and processes that drive normal physiology and alterations that lead to disease³³⁸. Although a wealth of studies have provided striking data sets of tissue dynamics in murine models at cellular resolution, subcellular resolution is harder to achieve, especially when cells deep within tissues are observed. In mature tissue, high cell density, fibrillar collagen, fat tissue, blood vessels and mineralized structures scatter the incident light and decrease the signal intensity with scanning depth²¹.

The development of imaging protocols that can gradually increase laser power and detector gain in deeper sections or 3D stacks, and the choice of multiphoton over confocal microscopy, improve resolution at depth. However, in most cases, we are limited to imaging the 100–200 μm of tissues that are closest to the objective lens in locations that are surgically exposed, optically accessible through imaging windows or directly physically accessible by the objective. Still, subcellular resolution has been achieved⁶⁴ and workflows including FLIM and FRET can be used to identify interactions between specific molecules^133,178,196.

Biological tissues can display a very complex, multilayered structure with variations in the refractive index of the sample introducing aberrations and blurring, which are increased further with increasing imaging depth, obscuring image quality and limiting resolution and readouts. Mismatches between the refractive index of the sample and the immersion medium can also contribute to aberrations. Once light enters a sample, these aberrations can distort the wavefront, leading to aberrations and reducing image quality. Although still not commercialized, adaptive optics has been developed to artificially introduce a compensatory aberration to the light wavefront of equal but opposite magnitude to the sample aberration, correcting the wavefront and improving image quality and resolution at depth. For an in-depth discussion on adaptive optics, we refer the reader to recently published, comprehensive review articles^339,340. Initially developed to fix light aberrations in telescopes used to observe remote stars, adaptive optics promise to expand the next frontier of deep tissue microscopy.

An alternative imaging approach is provided by the use of endoscopy-based approaches. Multimode optical fibres have been used to reach deep regions of the brain with minimal tissue damage³⁴¹. In general, these approaches are still very much at the prototype stage, but it is hoped that further collaborations between physicists and engineers will lead to widely applicable solutions.

Currently, the most difficult limitation to overcome is the low throughput of IVM. Because real-time biological processes that occur over several hours or days can only be tracked in only one or a very few animals at a time, IVM is typically time-intensive and labour-intensive. Sometimes very rare cells are observed (for example, HSCs)^9,11, although even when the cells and structures observed are more abundant (such as in the case of intestinal crypts or hair follicles) it is challenging to monitor several fields of view at a time, particularly if 3D data sets are generated. The speed of acquisition is a still an unbroken limiting step in determining the throughput of acquisition. When working with confocal and two-photon microscopes, alternative scanners that allow faster acquisition rates are still under-used. At the low end of temporal scales, scanning imaging systems, such as confocal and multiphoton microscopes, have employed various means for rapid scanning including spinning polygons³⁴², acousto-optic deflectors³⁴³ and resonant galvanometers³⁴⁴, although these advances have yet to become commercialized and, thus, widely available. In the cases where samples are limited in the number of available fluorescent molecules, fast-scanning techniques may be of limited utility. However, although it is commonly thought that short dwell times do not allow sufficient time for an appreciable amount of fluorescence to be generated, this is actually not a limitation to fast scanning. As acquisition electronics only sample a very small portion of light emitted during a pixel’s dwell time, the signal collected by most systems is rather independent of scan speed²⁹. Indeed, many video-rate confocal and multiphoton systems have been successfully developed and used. For example, video-rate acquisition of 2D images using an opto-acoustic scanner has been demonstrated²⁸², and an acousto-optic deflector-based multiphoton systems was developed to visualize neurotransmitter release in the intact brain using a pH-sensitive GFP variant fused to a synaptic vesicle membrane protein³⁴³. Ideally, microscope arrays that would allow monitoring cells in two to four animals simultaneously would increase the number of cells observed in each session; however, at the present time the cost of such set-ups, even if multiple microscopes could share single and multiphoton lasers, remains prohibitive. Attempts to widen the spatial scale of acquisitions have taken the form of mosaicked imaging (large-volume high-resolution intravital imaging)^64,202, or optical systems specifically designed for large field of view, mesoscopic imaging³⁴⁵, which have been used to solve problems where rare and stochastic events are captured and followed over time to determine their fate. These include spontaneous tumour initiation²⁰² and DTC arrest in the lungs³⁰¹.

More recently, new techniques have focused on reducing scan times by the selective illumination of subregions of interest³⁴⁶. These techniques are particularly useful for imaging immobile but functionally active structures such as neurons. Because neural networks comprise multiple neurons connected to the same functional circuit, but potentially separated by millimetres of distance, labelling the cells with calcium indicators can allow them to be probed together. Innovative new techniques that are focused either on unique scan patterns³⁴⁷ or on combining multiphoton with light sheet microscopy³⁴⁸ are still being developed.

A further limitation is posed by the duration that an animal can be kept anaesthetized while maintaining its physiology, and therefore the length of each imaging session. Although implanted windows have not been found to cause inflammation or a major alteration in the physiology^50,53, the accumulated effects of multiple procedures (window, anaesthesia, imaging and so on) on physiology over extended periods of time can and should never be excluded. Thus, it is crucial that proper control experiments be designed to ensure that an observed effect is due to the phenomenon being studied, and not a response to the technical approach. Although there are examples of tissues being constantly monitored for up to 72 h (REFs.^{8,51,108,349,350}), the challenge of maintaining animal physiology increases as the time of anaesthesia is extended beyond a few hours. Electrophysiology and imaging set-ups where animals are awake while recordings are taking place are not common but have been des cribed^{24,72,347,351}, although sessions remain relatively short; time limits in these cases are linked to the amount of time an animal can be kept restrained without affecting its physiology. Miniaturized microscope devices would revolutionize our ability to observe cells in situ over long periods of time^352,353. Recently, a lightweight two-photon miniscope was developed to perform calcium imaging in awake non-restrained mice over time spans of weeks³⁵⁴, but significant technological developments will be required for this to happen in other organs.

Finally, image analysis is a major bottleneck in the generation of data from IVM sessions. Traditional segmentation pipelines do not deal well with depth-linked changes in signal and customized approaches are needed³⁵⁵. Artificial intelligence approaches hold the promise to increase our ability to recognize, label and measure cells and structures efficiently, and will undoubtedly enable a wide implementation of multispectral imaging approaches, so that we will eventually be able to follow several — if not all — different cell types found in a certain tissue area and truly grasp the cellular dynamics underpinning health and disease^225,226.

Outlook

Future perspectives on the 3Rs

Although recent advances in cell culture techniques allow for live imaging of cell dynamics and behaviour of many different cell types, these in vitro systems can still not fully replace in vivo biology. Therefore, the observed behaviour can significantly differ for cells in the in vitro and in vivo settings. In the absence of in vitro systems to study cell behaviour and dynamics in their near-native environment, IVM will remain an important technology to provide unique insights into the in situ behaviour of cells. In this light, it will be important to consider and reflect on the ethics regarding animal use for experimentation and further develop the 3Rs (replacement, reduction and refinement) for IVM (BOX 1).

Replacement.

Before designing an IVM study, it should always be considered whether animal models are necessary to address the research questions. For example, human samples can be analysed with histopathology or sequencing approaches to identify cell types and gene alterations, and many organoid models (mini-organs) have been developed to study cell behaviour in a 3D environment, and to enable the study of dynamics of cell–cell interactions^356,357. However, histopathology and sequencing only provide a snapshot of highly dynamic processes and tissues, and organoids are a simplification of the full in situ environment. Therefore, it is still inevitable that animal models will be used to address key biological and medical questions.

Reduction.

The past decade has seen a wide variety of implantable imaging windows for various organs^358,359. The introduction of windows has helped to reduce the number of required animals because it allows measurements over time in the same animal rather than at different time points in multiple animals. In addition to the increase of data obtained from a single animal, the number of mice required for a statistically significant result is reduced as paired statistics can be performed. Another important development to reduce the use of animals is new approaches to cure data after, for example, movement distortions. These technologies reduce the number of experiments that fail, and therefore the number of required animals.

Refinement.

In addition to reducing the number of required mice, imaging windows are important refinement tools as they dramatically reduce the invasiveness and expand the accessibility of tissues for in vivo investigation even to the point that imaging can be performed for some tissues in freely moving, awake mice³⁶⁰. Combined with development in anaesthetics and analgesics, IVM experiments can be optimally refined.

Rodent models are typically considered the gold standard of biological research due to the high level of similarity in genetic, physiological and pathological features to humans³⁶¹. However, non-mammalian models can, potentially, be a refinement, such as nematodes³⁶², Drosophila³⁶³ and zebrafish³⁶⁴. Zebrafish are likely the most common model after mice and have been utilized for a diverse range of investigations that span cancer cell metastasis³⁶⁵, mitophagy³⁶⁶ and sensory hair-cell development and regeneration³⁶⁷, to name just a few. Their small size and high level of transparency make them suitable for imaging with a wide range of optical microscopy techniques³⁶⁴.

Priorities for the next decade

IVM could be advanced to decode cause and effect relationships of paracrine and juxtacrine communications between cells during specific physiological and pathological processes, as well as the effect of pharmacological interventions on tissue microenvironments. These capabilities would enhance our understanding of the natural progression of various diseases and the ways current therapies could be modified to achieve maximum benefit. In pursuit of this goal, IVM has seen several waves of innovation that have built upon each other. Advances in microscope technology enabled better visualization of cells and non-cellular components within tissues and drove the development of better methods of visualizing cellular dynamics. Advances in fluorescent proteins, and techniques for their insertion into somatic and germ-line cells in a cell type-specific manner, drove the need for developments in surgical engineering. Some of the newest advances have been the use of fluorescent reporters and biosensors to go beyond the identification of cell type and location, venturing into the cell state, lineage tracing³⁶⁸ and fate mapping^70,110. Reporters for stemness^109,212, phase of the cell cycle³⁶⁹ and exposure to hypoxia³⁷⁰ are just a few examples.

The next frontier that is ripe to be addressed is that of functional imaging, where the natural progression of diseases and the effect of pharmacological interventions can be tested within the context of intact tissues. This can take several approaches: microfluidic-containing implantable devices for the controlled release of biological or chemical agents, where the first proof-of-principle forays have already begun^371,372 but only a very few applications have been realized^373,374; or optogenetics³⁷⁵, a discipline where genes encoding light-sensitive proteins control protein expression^376,377, activity^378,379 or function³⁸⁰. To date, the predominant use of optogenetic tools has been in the discipline of neuroscience³⁸¹, with limited applications in cardiovascular³⁸² and developmental research³⁸³. IVM is envisioned to provide advanced understanding of biological processes within intact healthy and diseased tissues which would ultimately impact patient care.

Conclusion

Taken together, IVM is a powerful technology that has the unique potential to reveal cell and tissue dynamics in the unperturbed tissue context. IVM has significantly contributed to our understanding of in vivo cell behaviour and plasticity. We foresee that in the next years, technical advances will even further push the boundaries of in vivo imaging depth, resolution, throughput and complexity to explore the unseen worlds of tissue biology.

Fluorescence lifetime imaging microscopy (FLIM).

An imaging technique capable of assessing the chemical environment of a fluorophore by detecting the characteristic time from excitation to emission for a fluorophore rather than the number of photons emitted.

Intravasation

The entry of a cell or material from tissues into the lumen of a blood or lymph vessel.

Pulse repetition rate

The frequency at which high-energy laser light is emitted, typically 1–2 MHz for low-repetition rate and 80 MHz for high-repetition rate lasers.

Optical parametric oscillators

Femtosecond pulsed light sources capable of emitting light in the range of 1,100–2,000 nm, increasing the wavelength of a pulsed femtosecond laser beyond 1,000 nm.

Group velocity dispersion compensators

Optical devices capable of maintaining the duration of femtosecond light pulses in the presence of dispersive optical elements such as objective lenses.

Optical window

A surgically implantable device that creates an optically clear aperture through which deeper tissues (such as internal organs) may be viewed.

Gradient-index lenses

Very small diameter lenses (~1–2 mm) made of a material of non-uniform index of refraction that can be positioned at physical sites that cannot be accessed by standard lenses.

Quantum yield

The ratio of photons emitted to photons observed, informing on the efficiency of fluorescence emission.

Photon budget

The number of emitted photons per molecule before becoming non-fluorescent, which can depend on the quantum yield and photostability of a fluorophore.

Invadopodia

Plasma membrane protrusions with proteolytic activity to break down extracellular matrix proteins, which can be used by cancer cells to locally invade into adjacent tissues.

Calvarium

The top part of the skull. In the mouse, this includes the frontal bones, with the central suture joining them.

Endosteum

The interface between bone and bone marrow, typically lined by osteoblasts and osteoclasts.

MLL-AF9-driven murine AML

A murine model of poor-prognosis acute myeloid leukaemia (AML). The model can be based on the doxycycline-induced expression of the oncogene MLL-AF9 or on retroviral transduction of MLL-AF9 into haematopoietic progenitor cells. The different variants have some differences, for example, in the expression levels of the oncogene and latency of disease, but all develop into AML.

Acini

A cluster of cells that form a lobed structure that looks similar to a cluster of grapes.

apoA-IV

A plasma protein that shares structural features with other apolipoproteins.

Juxtacrine

signalling that requires cell–cell or cell–matrix contact.

Surgical engineering

The combination of novel surgical protocols and engineering designs to expose these tissues for short-term and longitudinal microscopic analysis.

RELATED LINKS

Fiji: https://fiji.sc/

ImageJ: https://imagej.nih.gov/ij/

Intravital_Microscopy_Toolbox: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0053942

Principles of the 3Rs: https://nc3rs.org.uk/who-we-are/3rs

TAPAS: https://imagej.net/plugins/tapas