Intravital imaging to study cancer progression and metastasis

Abstract

Navigation through the bulk tumour, entry into the blood vasculature, survival in the circulation, exit at distant sites and resumption of proliferation are all steps necessary for tumour cells to successfully metastasize. The ability of tumour cells to complete these steps is highly dependent on the timing and sequence of the interactions that these cells have with the tumour microenvironment (TME), including stromal cells, the extracellular matrix and soluble factors. The TME thus plays a major role in determining the overall metastatic phenotype of tumours. The complexity and cause-and-effect dynamics of the TME cannot currently be recapitulated in vitro or inferred from studies of fixed tissue, and are best studied in vivo, in real time and at single-cell resolution. Intravital imaging (IVI) offers these capabilities, and recent years have been a time of immense growth and innovation in the field. Here we review some of the recent advances in IVI of mammalian models of cancer and describe how IVI is being used to understand cancer progression and metastasis, and to develop novel treatments and therapies. We describe new techniques that allow access to a range of tissue and cancer types, novel fluorescent reporters and biosensors that allow fate mapping and the probing of functional and phenotypic states, and the clinical applications that have arisen from applying these techniques, reporters and biosensors to study cancer. We finish by presenting some of the challenges that remain in the field, how to address them and future perspectives.

Introduction

Although most cancer deaths are due to metastasis¹, the mechanisms of primary and metastatic tumour progression continue to elude definitive resolution, despite many studies². This makes the opening sentence of Warren and Gates’s 1936 article as relevant today as when it was first published: “In spite of a large volume of information from clinical and experimental observations on tumour metastasis, the subject is still largely a matter of speculation”³. The ultimate result of this uncertainty is that many promising new drugs that target metastatic cancer fail during clinical trials owing to lack of therapeutic efficacy^4,5. Furthermore, more than 10 years of investigation into the tumour microenvironment (TME) has revealed that the TME is a major driver of metastatic phenotypes^6,7. A full understanding of the TME, in both primary and secondary sites, is therefore crucial to reveal commonalities and differences between these sites that could lead to successful treatments.

Since the TME encompasses host and tumour cells, the most ‘high-fidelity’ experimental method to study this biology would be in vivo studies in mammals that measure cells of the TME in real time, longitudinally (at multiple time points) and with single-cell resolution. Intravital imaging (IVI) is beginning to fill this role as it provides insights into the TMEs and cell phenotypes that drive the timing, location and aggressiveness of tumour cell growth, invasion, dissemination to distant sites and therapeutic response, all within intact live tissues.

Here we review some of the recent advances in IVI that make investigations into cancer progression and metastasis, and their treatment, possible. We begin by making the case for the necessity of investigating tumour cell phenotypes in vivo, and then discuss some of the advances that have enhanced the ease of use (via improved methods and protocols) and utility of IVI (by increasing the amount and type of information the technique can record).

We confine this Review to the imaging of mouse models of cancer, as these are currently the best models available for the investigation of human cancer. In addition, this Review is focused on IVI using multiphoton microscopy because multiphoton microscopy offers significant advantages over clinical techniques such as magnetic resonance imaging and positron emission tomography (for example, single-cell resolution and functional labelling^8,9), as well as other types of optical microscopy, such as confocal microscopy (for example, reduced photobleaching, increased penetration depth, more robust optical alignment and increased sensitivity to weak fluorophores¹⁰). Furthermore, we describe the clinical impact that IVI-discovered phenotypes and pathways are having on the discovery of new biomarkers and therapeutics.

Advances in multiphoton IVI

Introduction to multiphoton IVI

The fundamentals of confocal and multiphoton microscopy^10,11, and of IVI and its history, have been described in many publications^12-18 that provide excellent explanations for how these imaging methods generate images of single slices of intact tissues. They describe how confocal microscopes achieve this by blocking out-of-focus light with pinholes before the detector, and how multiphoton microscopes limit the generated signal to the focal plane by using ultrafast pulsed infrared laser sources that allow absorption to occur only at regions of the sample with the highest photon density (that is, the plane of focus). Confocal and multiphoton microscopes can non-destructively produce images of optical sections that are comparable to the mechanically sliced tissue sections of standard histopathology. Several excellent reviews have additionally been published covering applications of IVI, such as the use of fluorescent proteins for visualizing single^19-22 and multiple^23,24 cell types; the utility of implantable imaging windows^25,26; the use of IVI for measuring metastasis^9,27, cell signalling²⁸, molecular dynamics²⁹ and stem cell plasticity³⁰ in vivo; the use of IVI in anticancer drug research³¹; and the use of IVI directly in the clinic³².

Enhancing the accessibility of multiphoton IVI

Although IVI is typically considered a technically difficult process³³ requiring specialized equipment ^27,34 and significant skills, the past decade has seen substantial growth of IVI use, and barriers to its adoption have been dramatically reduced. Advances (Box 1) include the advent of a plethora of new microscope technologies^35-38, fluorescent biosensors for identifying cell types and biochemical pathways^39-41, and protocols for accessing various tissues throughout the body, both short-term and long-term^25,26,32. Of particular importance are the numerous surgical protocols for accessing and stabilizing tissues that have enabled extended time-lapse intravital multiphoton imaging in a variety of murine organs, including the brain⁴², liver^43,44, cremaster muscle⁴⁵, spinal cord^46,47, mammary glands⁴⁸ and tumours^27,49, ovaries⁵⁰, kidneys⁵¹, lymphatic vessels^52,53, lymph nodes^54,55, salivary glands⁵⁶, lungs^57,58, and even developing embryos⁵⁹. The images presented in Fig. 1 are examples of the range and variety of tissues that can be imaged, and visually demonstrate how IVI has improved beyond the days of blurry black-and-white images to a level that is now comparable to that of in vitro or ex vivo studies.

Box 1. New tools for intravital imaging.

Encoded fluorescent proteins (CFP, GFP and YFP) and markers (fluorescent dextran and quantum dots)

Most commonly used markers for intravital imaging²⁵⁸.

Near-infrared fluorescent probes

Allow increased multichannel imaging²⁵⁹.

Fluorescence Lifetime Imaging

Used to image intrinsic fluorescence in unlabelled tissues^38,260.

Encoded photoconvertible fluorescent proteins (for example, Dendra2)

Used for fate mapping cells in living tissues at single-cell resolution²⁶¹.

Biosensors

Used for imaging of hypoxia³⁹, NF-κB²⁶², cancer stem cells^41,121, dormancy⁴⁰, Rho GTPases²⁶³ and the cell cycle²⁶⁴.

Second-harmonic generation

Used to image intrinsic nonlinear signal generation by extracellular matrix fibres²⁶⁵.

Third-harmonic generation

Allows imaging of intrinsic nonlinear signal generation that relies on the light-scattering properties and mismatches of refractive indices of water–lipid and water–protein interfaces²⁶⁶.

Fig. 1 ∣. Example images obtained by intravital imaging taken from a variety of tissues.

a, Kidney. Red, vasculature labelled with an Intravenous (IV) injection of tetramethylrhodamine (TMR)-tagged 155-kDa dextran; bright green, GFP-labelled macrophages; dim green, autofluorescence from collecting ducts and tubules. Scale bar, 70 μm. b, Liver. Red, sinusoids labelled with TMR-tagged 155-kDa dextran; cyan, CFP-labelled Kupffer cells. Scale bar, 70 μm. c, Lung. Red, vasculature labelled with an IV injection of TMR-tagged 155-kDa dextran; green, GFP-labelled disseminated tumour cells; blue, second-harmonic generation signal from type I collagen fibres. Scale bar, 70 μm. d, Bone marrow: a 3 × 3 mosaic of bone marrow imaged through the calvarium. Orange, vasculature labelled with an IV injection of TMR-tagged 155-kDa dextran; blue and green, autofluorescence from bone. Scale bar, 170 μm. e, Lymph node: a 5 × 5 mosaic of the entire lymph node. Red, blood vasculature labelled with an IV injection of TMR-tagged 155-kDa dextran and lymphatic endothelia (PROX1⁺ cells) labelled with tdTomato; green, GFP-labelled macrophages; yellow, GFP-labelled macrophages that have taken up the TMR-tagged dextran; blue, second-harmonic generation signal from type I collagen fibres. Scale bar, 225 μm. The inset shows a zoomed view of the region in the white box. f, Ductal carcinoma in situ (DCIS) in the mammary gland. Cyan, CFP-labelled mammary epithelium at the DCIS stage; red, vasculature labelled with an IV injection of TMR-tagged 155-kDa dextran. Scale bar, 50 μm. g, Muscle. Green, autofluorescence from striated muscle fibres. Scale bar, 100 μm. h, Pancreas. Green, Dendra2-labelled pancreatic acinar cells; red, vasculature labelled with an IV injection of TMR-tagged 155-kDa dextran. Scale bar, 70 μm.

Crucial to the success of these advances is the ability to prepare tissue for IVI. To achieve this, researchers are turning to surgical engineering, a field combining mechanical engineering, materials science and surgery that has traditionally focused on the unidirectional flow of engineering technology into the clinic (developing novel materials, devices and surgical protocols for use on patients)^60,61. However, for IVI, this flow is reversed, and the skills of and the instruments and procedures used by surgeons are brought into the imaging laboratory. The influence of this process is particularly notable when it comes to the development of chronic, or implantable, imaging windows (Box 2). Material biocompatibility, mechanical design and surgical implantation protocols must all converge to create a portal into the underlying tissue that does not induce local or systemic inflammation, negatively impact the normal physiology or behaviour of the animal, or alter the biology being investigated (Box 3). Validation studies have shown that this goal is attainable^58,62,63. Implantable imaging windows allow access to internal organs (even vital organs), while maintaining survival of the animal over days to weeks (or longer), enabling longitudinal IVI of tumour progression^58,62,63. It is important to note that use of implantable imaging windows (Box 2) is extremely flexible and does not limit the application of any of the new microscope technologies described in Box 1.

Box 2. Imaging windows for use with intravital imaging.

Mammary

The mammary imaging window was developed to eliminate the limitations of the terminal skin flap and dorsal skin fold chamber techniques and allow serial visualization of the mammary gland and orthotopic mammary tumours at high resolution, and over a period spanning days to weeks^261,267.

Lymph node

The chronic inguinal lymph node window provides serial access to this important immunological tissue. This procedure is well tolerated without any significant physiological changes in the mouse 14 days after window implantation²⁶⁸.

Abdominal

An abdominal window provides serial access to visceral organs, such as the spleen, kidneys, small intestine, pancreas and liver⁶².

Lung

A permanent window for high-resolution imaging of the lung was developed to enable longitudinal imaging during cancer metastasis⁵⁸.

Cranial

Two different methods enable imaging access to the cortex of the mouse brain: the cranial imaging window and the thinned-skull cranial imaging window²⁶⁹.

Cerebellum

The chronic cranial imaging window has been modified to image the mouse cerebellum, a region of the brain critical to coordinated motion²⁷⁰.

Long bone

Longitudinal intravital imaging of the bone marrow allows long-term multiphoton imaging in the long bone using a gradient refractive index micro-endoscopic lens⁶³.

Ovarian

The ovarian imaging window is a long-term imaging window developed to study the maturation of the ovarian follicle in response to gonadotropin analogues, as well as tumour invasion into the ovary⁵⁰.

Spinal cord

Long-term imaging of the spinal cord can be accomplished for periods up to months^271,272.

Box 3. Imaging window designs and limitations.

A list of imaging windows (and the tissues they give access to) in current use is given in Box 2. In designing new windows, attention must be paid to a fair number of parameters. Chief among these are shape and ergonomics to avoid irritation during ambulation, which leads to premature failure. Materials must be chosen with consideration given to biocompatibility, ease of manufacture and sterilizability. Biocompatibility must include evaluation of the impact of the material on the biology, as well as the biological impact on the material, as even materials normally considered inert (for example, ultra-high-molecular-weight polyethylene) can degrade over long periods of contact with tissues. Degraded materials can cause inflammation, allergic foreign-body responses or even chemical-induced carcinogenesis²⁷³. The importance of biocompatibility increases with the duration of implantation, but many chronic windows have been demonstrated to have exceptional performance over long periods^58,62,63.

For those not wishing to design their own custom windows, many publications include (and many laboratories are willing to share) manufacturing drawings that can be brought to in-house machine shops. If these services are not available, investigators can perform an Internet search for some of the nascent online custom manufacturing marketplaces where drawings can be uploaded to a website and machine shops with available capacity bid for the work.

Finally, optical performance must be evaluated, ensuring that the tissue is within the working distance of the objective lens, the point spread function of the illumination light is maintained and optical clarity is preserved over time. Use of standard no. 1.5 coverslip glass is usually sufficient to preserve point spread function, and some have reported that light-scattering proteinaceous deposits can be reduced or eliminated by coating coverslips with a polyethylene glycol film⁴⁴.

With this enhanced accessibility, biologists can now use IVI to collect real-time in vivo information about the dynamic processes that cells and tissues undergo within their native TME in the live animal^58,64,65. This ability even extends to measurements of the pharmacodynamics and pharmacokinetics of therapeutic treatments^66-68 and their influence on metastasis⁶⁹ (discussed further in the coming sections).

Expanding the spatial and temporal scale of imaging

TMEs have heterogeneous vascularization, extracellular matrix and cell density that may reflect distinct steps in cancer progression and need to be evaluated across different spatial scales (from submicron to the entire tumour). To record events on these vastly different scales, it is necessary to capture images of large volumes of the tumour tissue (~1–2 mm³) at subcellular resolution (~0.25 μm per pixel)⁴⁸. Unfortunately, capturing low-magnification images with high resolution is not possible with standard multiphoton microscopy, which relies on high-magnification, high numerical aperture objective lenses for efficient and bright signal generation¹⁰. This limits the size of the field of view of the acquired images and results in a dramatic undersampling of the tissues. The impact of this is illustrated in Fig. 2a, where just a few small regions from a picture of a famous person are presented. From just these small regions, it is impossible to discern the identity of the subject.

Fig. 2 ∣. Large-volume, high-resolution intravital imaging.

a, When presented with just a few high-resolution, high-magnification images, it is impossible to determine the subject of the photograph. b, Acquiring many images in a mosaicked pattern (orange arrows) maintains the spatial relationship between the individual acquired images. c, This allows complete coverage of the subject, and enables stitching of the individual images into a low-magnification, high-resolution image revealing the overall large-scale image. d, Large-volume, high-resolution intravital imaging (LVHR-IVI) applied to a mammary tumour using the method described previously⁴⁸ in a mouse mammary tumour virus–polyomavirus middle T antigen (MMTV-PyMT)–Dendra2 MacBlue mouse provides the context required to identify neoangiogenic vessels (NAVs) labelled with fluorescently tagged dextran (red) injected intravenously, distinguish between areas of ductal carcinoma in situ (DCIS) and late-stage carcinoma (LC), and determine their positioning relative to the stroma (S). A10 × 10 mosaic (340 μm × 340 μm tiles; the dashed box indicates the size of one tile) was acquired with a low-magnification, high-numerical aperture (NA) objective lens (25 × 1.05 NA). Fields of view were overlapped to compensate for non-uniform flatness of field illumination. Stitched images show the overall structure of the tissue. The formation of the ductal tree by the epithelial cells (green) is clearly visible, as are the blood vessels. e–g, The individual high-resolution tiles within the low-magnification LVHR-IVI image demonstrate the level of cellular and subcellular detail captured by this technique. e, Nuclei (N) and vacuoles (V). f, Cancer cell streaming migration along collagen fibres (arrow). g, Single-cell migration (arrows). Blue, second-harmonic generation signal from collagen I fibres; cyan, CFP-labelled macrophages; green, Dendra2-labelled tumour cells; red, tetramethylrhodamine-labelled 155-kDa dextran. The photograph in part c (and variations of it in parts a,b) courtesy of Keystone/Stringer via Getty Images.

One solution is to acquire many high-magnification, high-resolution images of the subject in a contiguous mosaicked pattern (Fig. 2b), and then stitch them together to produce a low-magnification image that can reveal the underlying identity of the subject (Fig. 2c), a process called ‘large-volume, high-resolution’ (LVHR) imaging^70,71. The application of LVHR to IVI (LVHR-IVI)⁴⁸ generates a more comprehensive view of the tumour than is possible with traditional IVI. Figure 2d demonstrates the type of images that can be obtained with LVHR-IVI techniques⁴⁸. LVHR-IVI thus allows researchers to use IVI images in a manner similar to how pathologists use sections from formalin-fixed, paraffin-embedded tissue, where the extent of tumour invasion and regions of interest are first identified at low magnification and are then analysed in more detail at higher magnification. With use of tissue morphology, areas of stroma (regions devoid of endothelial or epithelial cells) and neo-angiogenic vasculature⁷² can be identified. LVHR-IVI thus allows critical biological and pathological processes to be distinguished (for example, discerning mammary ductal carcinoma in situ (DCIS) from late carcinoma or branching ducts) following standard histopathology criteria⁷³. More detailed analyses can then be performed at higher magnification, identifying growth patterns associated with tumour progression and dissemination (Fig. 2e-g).

Combining LVHR-IVI with imaging windows (Box 2) enables multiple imaging sessions spanning days to weeks, or even months. However, longitudinal imaging introduces the additional challenge of locating the same imaging field repeatedly from session to session. This challenge has been addressed by the relocalization technique in vivo microcartography, which can relocate structures over multiple days in a variety of tissues, including the lung⁵⁸ and melanoma tumours⁷⁴.

These techniques are crucial for investigating early stages of tumorigenesis or metastatic seeding, which begin as just a few rare cells stochastically dispersed within an entire tissue. With use of these techniques, the transformation of normal oral tissues into premalignant and then malignant lesions was visualized by repeated imaging of the entire tongue⁷⁵. In the mammary gland, where epithelial cells are a small and dispersed fraction of the entire tissue, LVHR-IVI captured early dissemination by directly observing single tumour cells escaping the confines of mammary DCIS lesions via intravasation⁷⁶. Finally, in the lung, LVHR-IVI enabled visualization of the arrival and subsequent fate of disseminated tumour cells (DTCs)⁷⁷.

Imaging metastatic colonization

One of the earliest uses of IVI to investigate the process of metastatic colonization was that of Wood, who, in 1958, used a transparent chamber implanted into the rabbit ear⁷⁸ (developed by Sandison more than 30 years earlier⁷⁹). Wood used this chamber to visualize the fate of tumour cells injected directly into the vasculature.

Since then, metastatic colonization of secondary sites has been investigated in a number of different cancer types and tissues. These studies give a detailed account of the process of metastatic colonization, transforming it from a simple descriptive catch-all phrase into a process that encompasses several individual steps, including cancer cell arrival, extravasation, survival and growth into secondary nodules. In the following subsections, we briefly describe what has been learned to date about the metastatic process using in vivo imaging of various anatomical sites and tissues. Unless otherwise noted, each of the studies mentioned used IVI in one form or another.

Muscle

Early studies of metastasis focused on visualizing the arrival of tumour cells in easily imaged vascularized muscles such as the rabbit ear⁷⁸ and the cremaster muscle⁸⁰. However, metastasis of tumours to muscle tissues is very rarely encountered in patients⁸¹. Skeletal muscle, spleen, thyroid and adipose bone marrow (yellow bone marrow) constitute antimetastatic niches⁸², which may be interesting to study to decipher why these sites do not lead to overt metastases. Indeed, recent work studying skeletal muscle (using ex vivo techniques) has determined that, while tumour cells do migrate to muscle, oxidative stress within the tissue creates a microenvironment that prevents outgrowth of these cells⁸³. Thus, we may expect further development of imaging windows to study antimetastatic niches in the future.

Lymphatics and lymph nodes

IVI of locoregional dissemination to lymphatics and lymph nodes⁸⁴ determined that mechanical pressure placed on tumours can increase tumour cell trafficking to the draining nodes. However, Das et al. showed that, in unperturbed tissues, entry of tumour cells into the lymphatic sinuses occurs through active tumour cell migration and can be prevented by blocking the cytokine receptor C-C motif chemokine receptor 8 (CCR8)⁵². Although development of metastatic foci in lymph nodes (positive lymph nodes) has been associated with worse outcome in patients, it was only recently discovered (using fixed tissue analyses) that cancer cells in positive lymph nodes can migrate haematogenously to distant sites^85-87. IVI of positive lymph nodes could provide a better understanding of the mechanisms of cancer cell redissemination to distant sites and lead to novel targeted therapies.

Lung

One of the more challenging tissues to image is the lung. Its delicate, yet complex nature has made it a target of imaging since the 1600s⁸⁸, with in vivo visualization reported as early as 1925 (ref.⁸⁹). Modifying a vacuum-stabilized imaging window for cats first published in 1939 (ref.⁹⁰), Funakoshi et al. became among the first to image metastatic nodules in the lungs of mice⁹¹. The group found the vasculature in metastatic nodules to be irregularly shaped compared with normal vessels and to have reduced blood flow⁹². More recently, Headley et al. used this type of imaging window to show that upon arrival in the lung, DTCs fragment and shed tumour microparticles that accumulate in the lung interstitium and generate immune responses⁹³. Finally, our group developed an implantable imaging window that allows the mouse to survive, giving multiple views of the lung vasculature for the first time^58,94. This window can be used to directly visualize the arrival and extravasation of DTCs and follow them as they grow into micrometastases⁷⁷. This work demonstrated that the TME of the primary tumour induces an invasive, stem-like and dormant phenotype within the migrating cancer cells, giving these cells a survival advantage within the metastatic site and the potential to grow into metastatic nodules when the conditions become suitable.

Brain

A similar longitudinal IVI study followed DTCs in the brain over time and observed that DTCs in the brain stop at vascular branch points, extravasate soon after arrival and maintain close contact with the microvasculature⁶⁵. After extravasation, growth of the cancer cells occurred in cancer type-specific patterns, with melanoma metastases using vessel co-option (perivascular growth) and lung cancer-derived metastases forming nodules that induced angiogenesis. Another recent study using serial IVI of the brain showed that escape from dormancy is a rate-limiting step in the development of brain metastases, with astrocytes driving DTC dormancy⁹⁵.

In addition to its use for studying patterns of metastatic growth in the brain, IVI could be used to learn more about the mechanisms by which cancer cells break the blood–brain barrier and enter the brain parenchyma. IVI could also potentially help design measures to prevent brain metastases, which are inevitably associated with terminal disease⁹⁶.

Liver

One of the first IVI investigations of metastasis in the liver visualized the arrival of fibrosarcoma DTCs, and found cells mechanically trapped in narrow liver sinusoids, without apparent influence from platelets⁹⁷. This was confirmed by Scherbarth and Orr⁹⁸, who found that B16 melanoma cells are mechanically trapped in hepatic sinusoids. However, they also found that pretreatment of mice with the pro-inflammatory cytokine interleukin-1α (IL-1α) caused tumour cell adhesion to endothelial surfaces of vessels that were more than twice the size of the tumour cells. Schluter et al.⁹⁹ then found that colon carcinoma cell lines had high adhesion rates, independent of their metastatic potential, and that DTCs were arrested in microvessels with diameters much larger than the adherent tumour cells. These studies demonstrated that in addition to vasculature size, host-derived soluble and cancer cell-intrinsic factors affect metastatic seeding of the liver.

Most recently, Ritsma et al. developed an implantable imaging window for abdominal organs (that is, small intestine, liver, spleen, kidneys and pancreas) and used it to study the steps of colorectal cancer metastasis formation in the liver over the course of 14 days^44,62. They found that DTCs proliferated to form motile, yet confined, ‘pre-micrometastases’, which then condensed into micrometastases in which cell migration was suppressed and proliferation persisted.

Bone

IVI of metastatic homing to the bone marrow of the calvarium showed that, when injected into the vasculature, most tumour cells (leukaemic, multiple myeloma and prostate) homed to spatially restricted vascular domains within the marrow¹⁰⁰. Further analysis showed that these domains express stromal cell-derived factor 1 (SDF1; also known as CXCL12), and that benign circulating haematopoietic stem/progenitor cells, mature T lymphocytes, leukaemic cells and tumour cells all extravasate at these locations. Thus, tumour cells use these particular areas of vascular endothelium in a manner that mimics the multistep tissue-homing mechanisms of benign leukocytes. It would be of interest to investigate whether any therapies can target the extravasation of cancer cells at these specific sites without affecting normal physiology.

Summary

Taken together, the studies discussed in this section typify the breadth of applications of IVI, the landmark discoveries that have been made in cancer and metastasis research using IVI, and a vision for the future application of IVI.

Imaging drug action

The development of therapeutics that can prevent or treat metastasis is a major unmet need in cancer research. As IVI can analyse intact tissues that are connected to the systemic circulation, it can be used to track systemically administered drugs, measuring tissue clearance (pharmacokinetics) and the drug’s effect on targeted tissues (pharmacodynamics)¹⁰¹. As some therapeutics are administered as prodrugs, which must be metabolized in the body before the intended drug action is achieved¹⁰², they can be studied only in vivo. Standard pharmacokinetics assays involve low-resolution techniques such as blood sampling or tissue homogenization¹⁰³. Pharmacodynamics are usually assessed by histopathology and by low-resolution clinical imaging techniques such as magnetic resonance imaging and positron emission tomography¹⁰⁴. This leaves the mechanism of action of some therapeutics unknown¹⁰⁵.

A number of IVI investigations have tracked drug distribution and uptake of chemotherapeutics^68,106, nanoparticle drugs¹⁰⁷ and prodrug conjugates¹⁰⁸, targeted therapies^66,109, immunotherapies⁶⁷ and radiation sensitization to therapeutics¹¹⁰. These studies show that IVI allows assessment of drug distribution and tissue penetrance, uptake by multiple cell types and evaluation of the resultant cell death, all in real time^31,105,111.

Biosensors

One of the most common uses of IVI has been to visualize and quantify cellular motility. Parameters such as migratory path, velocity, turning frequency and chemotactic index can be extracted from time-lapse videos^112,113. With multicolour^114,115 and multispectral^116,117 imaging, multiple cell types can also be imaged simultaneously and cell–cell interactions can be visualized and quantified (for example, using interaction times¹¹⁸). However, cells, particularly those in tissues, carry a wealth of information that goes beyond motility parameters. Expression of genes, alone or as part of a signature, marks cells as having specific functions or phenotypic states. Use of these genes to drive the expression of fluorescent proteins allows their direct visualization in vivo¹¹⁹. This can be particularly important for understanding metastasis and tumour progression.

Imaging of phenotypic states of cells

It has been postulated that a stem cell-like phenotype is needed for metastatic colonization. However, given the plasticity of DTCs, the exact location where stemness induction occurs along the metastatic cascade has been difficult to ascertain. With use of an inducible fluorescent colorectal carcinoma model designed to intravitally visualize cancer stem cells (CSCs; cells that express leucine-rich repeat-containing G protein-coupled receptor 5 (LGR5)), it was recently demonstrated that disseminated colorectal cancer cells retain a non-stem-like, highly motile phenotype (and are LGR5⁻) while in the circulation and during colonization of distant organs (liver)¹²⁰. However, these cells possess intrinsic plasticity and can switch to become LGR5⁺ cells in response to microenvironmental factors produced at secondary sites, such as hepatocyte growth factor (HGF) and fibroblast growth factor 2 (FGF2). Furthermore, selective ablation of LGR5⁺ cells prevented metastatic outgrowth, confirming that a stem-like phenotype is required for the initiation of metastatic outgrowth in this model.

Contrary to the reduction of stemness during dissemination found in colorectal cancer, recent IVI analysis of breast cancer using a biosensor for stemness¹²¹ found that a stem-like phenotype is enriched within specific microanatomic locations of the primary tumour that contain intravasation portals, called ‘tumour microenvironment of metastasis (TMEM) doorways’⁴¹. That study found that the density of CSCs is further enriched during metastatic dissemination, with CSCs making up more than 60% of circulating cancer cells. This stemness phenotype persists during cancer cell arrival at a distant organ (the lungs) and early metastasis formation, and then decreases as metastatic nodules increase in size. Tracking the fate of DTCs upon arrival in the lungs with IVI through implantable lung imaging windows shows that these spontaneously disseminating cancer cells also exhibit invasive and dormant phenotypes, which provide extravasation and survival advantages⁷⁷.

Thus, it seems that the need for activation of a programme of stemness during the metastatic cascade is cancer type specific and target organ specific. However, in both studies, non-CSCs displayed higher migration velocities than CSCs, indicating that some properties of CSCs are universal across cancer types. Further investigation using IVI should allow delineation of therapeutic targets to block stemness induction and exit from dormancy.

Imaging of oxygen tension and hypoxic tumour cells within the TME

Oxygen tension and pH can affect both cancer progression and response to treatment. Although one would expect that these parameters correlate with each other (for example, hypoxic areas exhibit low pH), a study by Jain and colleagues using high-resolution imaging combined with fluorescence ratio imaging microscopy and phosphorescence quenching microscopy showed that no such correlation exists, demonstrating an extraordinarily complex connection between perfusion and metabolism within the TME¹²². Although these studies were performed intravitally using human colon carcinoma cells, the microenvironment was potentially not reflective of that in human tumours as the xenografts were grown ectopically in a dorsal skin fold chamber, which may have a microenvironment substantially different from the one these cells encounter physiologically.

Despite this limitation, discrepancies between areas of high oxygen tension and expected high intracellular oxygenation were also observed at the single-cell level by real-time IVI of orthotopically implanted breast cancer cells. This was accomplished using tumour cells expressing a novel hypoxia biosensor that is sensitive to the expression of hypoxia-inducible factor 1 α (HIF1α)³⁹. The study showed that hypoxic tumour cells can paradoxically be found located adjacent to well-perfused blood vessels. Furthermore, it showed that metabolically hypoxic cells display an invadopodium-rich, slow migratory phenotype with enhanced matrix degradation activity that is characteristic of migrating cancer cells¹²³.

Imaging of apoptotic cell death

Observing cancer cell death within the native TME can be useful to determine the effect of drugs on apoptosis. BODIPY-labelled peptides that bind to apoptotic cells in a Ca²⁺-independent manner and fluoresce in green (Apo-15 (ref.¹²⁴)) or red (Apotracker Red¹²⁵) may be useful to study the effects of experimental therapeutics on cancer cells in their native microenvironment. These studies would be of higher fidelity than studies performed in vitro, and could help understand which cells within the TME become therapy resistant and which drugs can alleviate this resistance.

Imaging of intrinsic signals

In addition to biosensors, signals that are native to the tissue, such as second-harmonic generation signal¹²⁶ or autofluorescence, can provide useful information. Fluorescence lifetime imaging measures the time of arrival of individual photons after pulsed excitation of auto-fluorescent molecules such as NADH and FAD¹²⁷. While the arrival time of each photon is stochastic, the distribution of arrival times follows an exponential decay, with the characteristic decay time dependent on the chemical microenvironment surrounding the fluorescent molecule³⁸. Intravital fluorescence lifetime imaging has been used to investigate cell type-specific metabolic signatures^128-130, and therapeutic responses¹³¹. Use of intrinsic signals greatly simplifies experiments and makes possible the extension of IVI tools to clinical applications (where exogenous labelling of tissues is not possible)^132,133.

Imaging metastatic dissemination

Understanding how tumour cells intravasate is critical for discovering approaches to decrease overall metastatic burden. Blocking intravasation could be used for patients with local and regional disease who receive systemic preoperative (neoadjuvant) therapies, which, in certain instances, may increase haematogenous dissemination^69,134. In addition, blocking dissemination may still be important in patients who had their primary tumour resected, because evidence is accumulating that haematogenous dissemination also occurs from metastatic lesions^135,136. Indeed, the relative number of circulating tumour cells is an important prognostic and predictive factor in metastatic breast¹³⁷, colon^138,139 and pancreatic¹⁴⁰ cancers.

Despite the importance of understanding how tumour cells gain access to the blood vasculature, there is considerable debate regarding the answer to this question. Two (not mutually exclusive) theories exist: that of collective migration followed by collective vascular invasion, and that of single-cell and streaming (coordinated single-cell) migration, followed by single-cell intravasation.

Evidence for collective migration

The concept of collective migration¹⁴¹ dates to early observations of the motility of tumour cells grown in 3D matrices in vitro. These studies noted that small clusters of cells can move together in vitro¹⁴² and grow in ‘ribbons’, reminiscent of tumour cell aggregates observed embedded in connective tissue in histological sections of patient tumours¹⁴³. However, only three studies have directly observed collective migration in vivo in orthotopically injected tumour cells^144-146. Of these three studies, Giampieri et al.¹⁴⁴ found that collective migration was not associated with haematogenous dissemination but was instead observed during cancer cell invasion into lymphatic vessels.

These studies^144-146 used tumour cell lines that fail to recapitulate the histology and progression of cancers observed clinically. IVI studies in transgenic mouse models of cancer, which do follow the clinical disease progression and histology^147,148, have not observed collective migration, nor have they observed that migrating tumour cells need to invade the stroma for them to intravasate. Indeed, by study of whole-mount excised tumour tissues to evaluate the quantity and location of intravasation sites across the entire tissue (tumour and adjacent stroma), it was determined that more than 98% of intravasated cells are found at the tumour interior, not in the stroma¹⁴⁹. This result is consistent with the findings of studies using IVI⁶⁴ and immunohistopathology^85,150-152.

While patterns of tumour cells extending into the stromal tissue that are consistent with those observed in 3D in vitro assays have been observed histologically in fixed tissues from mice¹⁵³ and humans¹⁵⁴, it is impossible to determine from fixed tissues whether these patterns arose from the collective migration of tumour cells or from asymmetrical tumour cell division and growth. Still, both tumour strands¹⁵⁴ and tumour buds (single or small tumour cell clusters (four cells or fewer))^155,156 found at the invasive tumour front correlate with clinical outcome in patients. It is possible that these histological patterns are related to an advanced stage of tumour progression, but not necessarily collective migration.

Finally, collective migration can be a mode of tumour cell dissemination only if, as speculated, it leads to the collective vascular invasion of tumour cells. Evidence supporting collective vascular invasion stems only from analysis of fixed haematoxylin and eosin-stained mouse and human tissue sections, which is a method incapable of distinguishing between lymphatic vasculature and blood vasculature¹⁵⁷. Most importantly, it has since been determined by immunostaining that in more than 97% of cases of intravascular tumour clusters, the vessels are lymphatic vessels, not blood vessels¹⁵⁸.

The observation of tumour cell clusters^159-161 in the circulation of mice and human patients has also been used to support the hypothesis of collective migration and collective vascular invasion. However, those studies used non-physiological conditions to promote cell cluster generation and may reflect tissue damage rather than dissemination ability. Thus, although recent work seems to indicate that tumour cell clusters in the lung are more potent at generating metastases than single tumour cells¹⁶², direct evidence for this conclusion and the origin of cell clusters is still lacking. Invasion of cohesive clusters of tumour cells directly into the blood vasculature, as inferred from cell clusters detected in peripheral blood, has, to our knowledge, never been seen by IVI.

The paradox that single tumour cells enter the blood, but that tumour cell clusters seed distant metastases, may be resolved by the finding that tumour cell cluster formation is associated with CD44-mediated intercellular adhesion¹⁶³. This additionally increases DTC survival and improves DTC seeding as CD44 is associated with a CSC phenotype¹⁶⁴. Consistent with this finding are the observed stem-like gene expression patterns of tumour cell clusters found in the blood circulation in live mice¹⁶⁵. Together, these results support a pattern of dissemination where single tumour cells intravasate and subsequently form clusters (supported by gene expression patterns associated with stemness) in the circulating blood.

Evidence for single-cell/streaming migration during dissemination

Multiphoton IVI has played a major role in defining the mechanism of single tumour cell dissemination. IVI by our group^{64,123,166-169}, and by others^144,148,170, has established that (1) single-cell and streaming migration are seen frequently in invasive carcinoma in a variety of mouse models, including transgenic, xenograft and tissue transplant models^{64,123,144,148,166-170}; (2) macrophages are essential to this process^64,168; and (3) these single-cell migration patterns of tumour cells and macrophages are associated with haematogenous dissemination^64,167,169. Even at secondary sites such as the lung, our IVI studies^57,58, and ex vivo imaging studies by others^171,172, have demonstrated the involvement of macrophages in the extravasation of tumour cells.

IVI of mammary tumours has been used to observe that, during streaming migration, solitary tumour cells and macrophages communicate with each other via a paracrine interaction¹⁷³ and move together (as unattached cells at speeds of 10–100 times the speeds commonly seen in similar cell types in vitro^123,166,168) along collagen fibres towards blood vessels¹⁶⁸. Macrophage–tumour cell interactions in the migrating stream alter gene expression in the tumour cells in a pattern called the ‘invasion signature’¹⁷⁴. One of the most highly altered proteins in this signature¹⁷⁵ is the actin-regulatory protein MENA (also known as ENAH), which has several splice variants¹⁷⁶: MENA11a (associated with an epithelial phenotype) is downregulated during streaming migration, and MENA^INV (an isoform that leads to higher levels of motility and increased sensitivity to growth factors such as epidermal growth factor (EGF)¹⁶⁶, HGF¹⁷⁷ and insulin-like growth factor (IGF)¹⁷⁸) is overexpressed. Another change in the MENA-specific expression pattern, MENA^Calc, reflects the relative amount of MENA11a compared with total MENA levels and is prognostic for risk of metastatic recurrence in patients with breast cancer^179,180.

Evidence for cancer cell intravasation portals

IVI of mammary carcinomas has revealed that once the streams of macrophages and tumour cells reach the vasculature, they slow down and associate with TMEM doorways. TMEM doorways are composed of one macrophage expressing high levels of the receptor tyrosine kinase TIE2 (also known as TEK) and vascular endothelial growth factor (VEGF), one MENA-expressing tumour cell and one endothelial cell, all in direct and prolonged stable contact^{64,151,168,181}. Long-duration time-lapse multiphoton IVI has shown that the TMEM doorways themselves are stable over extended periods, during which they regulate transient, localized blood vessel opening and cancer cell entry into the bloodstream^64,169. IVI has been used to directly visualize single-cell intravasation through these doorways in several transgenic and xenograft mouse models of breast cancer^{64,76,114,168}.

Active TMEM doorways are found in pre-invasive^76,167 and invasive ductal⁶⁴ breast cancer as well as in metastatic foci in lungs⁵⁸ and positive lymph nodes¹⁵², indicating that TMEM doorway-mediated cancer cell dissemination occurs not only at the primary tumour site but also at metastatic sites, which could perpetuate metastatic dissemination even after removal of the primary tumour. The observation of TMEM doorways in lymph nodes supports indirect studies that have concluded that tumour cell dissemination from lymph nodes occurs via a haematogenous route^85,87,182.

Recent work using IVI with CSC biosensors and immunofluorescence of fixed tissues has shown that TMEM doorways are microenvironments enriched in CSCs and MENA^INV-expressing cancer cells owing to increased cancer cell–macrophage contact that occurs around TMEM doorways⁴¹. This work provides an explanation as to why tumour cells crossing into the blood circulation are CSCs, and why tumour cells that make contact with macrophages around TMEM doorways exhibit greatly enhanced transendothelial migration activity (a consequence of MENA^INV expression increasing their ability to enter the blood circulation¹⁸³). Thus, this observation provides the mechanism through which single breast tumour cells entering the blood may lead to tumour cell clusters seeding distant metastases; macrophages are responsible for increased intercellular adhesion (CD44 expression leading to clustering in the blood), and the induction of a stem-like phenotype, leading to enhanced tumour-initiating capacity.

The observation that tumour cells migrate from metastatic nodules^85-87,135, and that metastasis-to-metastasis seeding is a common event^184,185, indicates that it would not be too late to inhibit TMEM doorway function after removal of the primary tumour. This is because, in some patients, cancer cells might have already migrated from the primary site and formed clinically undetectable micrometastases, which could be a source of further metastatic tumour cell dissemination and could increase overall metastatic burden.

Indeed, IVI showed that a TIE2-specific inhibitor, rebastinib, can inhibit TMEM doorway-associated vascular opening and intravasation^69,186. Rebastinib substantially decreases the numbers of circulating tumour cells in mice and humans with breast cancer^69,186,187. Furthermore, mice with metastases from a mammary tumour survive after tumour resection and treatment with rebastinib plus cytotoxic chemotherapy, whereas those treated with resection and chemotherapy alone do not¹⁸⁶. As discussed further later, rebastinib is currently in phase I clinical trials for use in breast cancer and other cancers^188,189.

Clinical translation of IVI findings

Biomarkers for metastasis

Some of the discoveries pertaining to the metastatic cascade made by IVI have been successfully translated to the clinic in the form of biomarkers capable of prognosticating metastatic recurrence in patients^{150,151,179-181,190}. In particular, the density of TMEM doorways is significantly associated with distant recurrence, specifically early recurrence (5 years after diagnosis)¹⁸¹, and MENA^Calc has been positively associated with the risk of death in three breast cancer cohorts^179,180, independent of other traditionally measured clinical parameters.

These retrospective studies^{150,151,179-181}, which comprise thousands of patients, demonstrate how insights gained from IVI can yield prognostic tests specific for metastatic dissemination. Further, the fact that tests such as the TMEM doorway score have a very poor correlation with assays that measure proliferation (for example, Oncotype DX Breast Recurrence Score) confirms that a distinct biological process is being measured by the TMEM doorway-related prognostics¹⁸¹ (Box 4). These new prognostics that specifically measure the process of dissemination should enable the evaluation of a new class of desperately needed antidissemination drugs^{69,186,187,191}.

Box 4. Complementary use of the TMEM doorway score and the Oncotype DX Breast Recurrence Score in treatment decisions.

The density of tumour microenvironment of metastasis (TMEM) doorways (TMEM doorway score) is an independent prognosticator of metastasis in breast cancer^150,151,181. The TMEM doorway score offers the opportunity to fine-tune breast cancer prognosis obtained by the Oncotype DX Breast Recurrence Score (RS). RS is a 21-gene prognostic assay of recurrence based on genes predominantly associated with cancer cell proliferation²⁷⁴. In breast cancers from patients, there is a lack of correlation between these scores¹⁸¹, indicating that they measure different aspects of cancer biology: the TMEM doorway score is a measure of cancer cell dissemination, while RS is a measure of cancer cell proliferation¹⁷⁴. Indeed, patients with a low RS but a high TMEM doorway score have significantly increased risk of distant recurrence^174,181. Thus, precision medicine should strive for the optimal assessment of prognosis by taking into consideration both scores when patient prognosis is being assessed and treatment is being planned.

Effect of therapy on the TME as seen by IVI

IVI has been used to demonstrate the effects of systemic therapies (for example, chemotherapy, radiotherapy, targeted therapy and anti-angiogenic therapy) on the TME (Fig. 3), a phenomenon known as the host response¹⁹². Changes in the TME of the primary tumour and metastatic sites (for example, influx of bone marrow-derived progenitors or modulation of immune response within the primary tumour) can influence drug delivery, cancer progression, cancer dissemination and drug resistance.

Fig. 3 ∣. The contribution of high-resolution intravital imaging to our understanding of the effects of various therapies on the tumour microenvironment and cancer phenotype.

a, Chemotherapy-and radiotherapy-induced tissue damage and hypoxia lead to an influx of myeloid cells (1) into the tumour microenvironment (TME)⁶⁸ and a subsequent increase in macrophage-mediated localized transient vascular openings that may affect drug delivery (2)¹¹⁰ and may increase the density and activity of TME of metastasis (TMEM) doorways (triangle) (3), leading to enhanced localized TMEM doorway-mediated transient vascular opening (4), TMEM doorway-mediated single cancer cell intravasation and an increase in the number of circulating tumour cells (CTCs)⁶⁹. b, Targeted therapy (for example, anti-BRAF) can lead to extracellular matrix remodelling via cells of the TME such as cancer-associated fibroblasts. The newly formed extracellular matrix may, by an alternative route, activate growth signalling pathways, thereby bypassing the inhibited pathway and rendering the targeted therapy ineffective²⁰¹. c, Targeted inhibition of C-X-C motif chemokine receptor 4 (CXCR4) prevents interaction of acute lymphoblastic leukaemia (ALL) cells with the bone marrow niche, allowing them to escape the bone marrow and become sensitive to chemotherapy (1). By contrast, other factors within the niche keep acute myeloid leukaemia (AML) cells within the marrow (2), rendering them insensitive to chemotherapy^202,203. d, Anti-angiogenic therapy (1) can induce expression of endothelial adhesion molecules (for example, intercellular adhesion molecule 1 (ICAM1) and vascular cell adhesion molecule 1 (VCAM1)), which enhances leukocyte adhesion (2), leukocyte extravasation (3) and a subsequent antitumour immune response (4)²⁰⁴.

Systemic therapy depends on drug delivery into the TME (a dynamic process best studied in vivo). Numerous important conclusions regarding drug delivery have been made with the help of IVI. IVI has shown that tumour vasculature is permeable to subcellularsized nanoparticles^193,194, leading to the concept of increased vascular ‘leakiness’ in tumour beds, and associating this leakiness with poor therapeutic delivery¹⁹⁵. However, increased vascular permeability may be beneficial for the selective delivery of chemotherapeutics into, and their retention in, tumour interstitial tissues, an effect termed ‘enhanced permeability and retention’ (EPR)^196,197. While EPR is primarily associated with solid tumours, there is evidence that nanocarriers can enhance the interaction of therapeutics with lymphoma cells¹⁹⁸. This demonstrated that drug delivery into tumours can be improved via vascular disruption of intratumoural blood vessels. This is accomplished using the recently developed technique of acoustic vaporization and the application of microparticle- and nanoparticle-carrying drugs¹⁹⁹.

IVI has also revealed that, in breast cancer, radiotherapy and chemotherapy can induce transient vascular disruption (depending on perivascular macrophage density)¹¹⁰. In particular, perivascular macrophages elicit dynamic bursts of serum extravasation, enhancing delivery of nanoparticles and subsequently increasing drug uptake in neighbouring tumour cells (Fig. 3a). Moreover, combining radiotherapy with the chemotherapy cyclophosphamide increased macrophage influx and delivery of nanoparticles further, suggesting that conventional anticancer therapies may be used to prime the TME for drug delivery¹¹⁰. Similarly, other groups demonstrated that chemotherapy increases the density of perivascular TIE2^hi macrophages and TMEM doorways, and increases TMEM doorway opening^69,110,134 (Fig. 3a). These changes, along with a chemotherapy-induced increase in MENA^INV expression in primary breast tumours, contribute to increased vascular opening that is linked to cancer cell dissemination and metastatic seeding at secondary sites^69,134. This chemotherapy-promoted TMEM doorway assembly and function⁶⁹ is suppressed by rebastinib, which targets TMEM TIE2^hi macrophages^69,186.

The aforementioned observations may be a part of essentially the same biological phenomenon: a TME response resulting from a chemotherapy-mediated increase in the density of perivascular macrophages. It would be interesting to investigate whether drug delivery can be improved in the clinic by exploiting macrophage-induced vascular opening in the TME upon radiotherapy and cytotoxic chemotherapy, as suggested by Miller et al.¹¹⁰, while at the same time suppressing the TMEM doorway-associated increase in tumour cell dissemination.

IVI has shown that, in breast cancer, the tumour’s response to chemotherapy, as well as the delivery of chemotherapeutics, depends on tumour size⁶⁸, demonstrating that chemotherapy increases macrophage density in the TME, which subsequently hampers a beneficial response of the tumour to chemotherapy (Fig. 3a). Another IVI study showed the effect of chemotherapy on the induction of dormancy and the protective effect of osteopontin on dormant acute lymphoblastic leukaemia (ALL) cells in mice²⁰⁰. Using a cranial imaging window, the researchers demonstrated that multiple rounds of chemotherapy result in minimal residual disease, which is refractory to treatment, and that treatment response can be increased by injection of osteopontin-targeting antibodies.

IVI was also used to understand the response of certain solid tumours to targeted therapy. For example, IVI of an ERK/MAPK biosensor was used to study mechanisms underlying the resistance of melanoma cells carrying a Braf mutation to anti-BRAF therapies²⁰¹ (Fig. 3b). Specifically, tolerance to the BRAF inhibitor PLX4720 developed rapidly in a TME with high stromal density. Mechanistically, PLX4720 activates melanoma-associated fibroblasts, leading to increased matrix production and elevated integrin β1–focal adhesion kinase (FAK)–SRC signalling to ERK in melanoma cells. Co-inhibition of BRAF and FAK decreased tumour size by abolishing PLX4720-induced ERK activation and resensitizing melanoma cells to BRAF inhibition.

Furthermore, IVI has been used to study the effect of targeted therapy on the most common types of leukaemia, acute myeloid leukaemia (AML) and T cell ALL (Fig. 3c) and helped explain the disappointing results from clinical studies of patients with AML treated with C-X-C motif chemokine receptor 4 (CXCR4) inhibitors. Visualization of the effect of the CXCR4 inhibitor AMD3100 on the motility of AML cells and T cell ALL cells indicated that AMD3100 inhibited retention of ALL cells within the bone marrow niche (which is supportive of a chemo-resistant phenotype) but AML cells were still retained within the niche, most likely owing to compensatory signals from other chemokines, such as C-C motif chemokine ligand 2 (CCL2)²⁰². Thus, IVI contributed mechanistic information critical for guiding clinical trials in patients with leukaemia. The same group previously used a similar IVI approach to demonstrate that the dynamic interactions of T cell ALL with the bone marrow microenvironment are responsible for refractory disease²⁰³. This helped redirect the clinical therapeutic strategy of combating chemo-resistant T cell ALL from targeting the bone marrow stroma to targeting cell migration and the interaction of cells with the bone marrow niche.

IVI was also instrumental in evaluating the role of anti-angiogenic therapy in melanoma; this therapy facilitated tumour infiltration of lymphocytes, which led to an adaptive immune response and a reduction in tumour growth in mice with subcutaneously grown B16 melanoma²⁰⁴ (Fig. 3d). Mechanistically, this work exploited the observations that the expression of endothelial adhesion molecules (for example, vascular cell adhesion molecule 1 (VCAM1)) is crucial for leukocyte extravasation, and that the expression of endothelial adhesion molecules can be suppressed by proangiogenic factors such as VEGF. Indeed, using IVI, the researchers visualized the promoting effect of anti-angiogenic therapy on vascular rolling, adherence and transmigration of leukocytes into the melanoma microenvironment, which increased the effectiveness of adoptive immunotherapy.

In addition to evaluating the effect of various therapies on the TME of primary tumours, IVI was used to assess the response of cancer cells to chemotherapy in metastatic sites. IVI through an abdominal imaging window indicated that, regardless of the type of chemotherapeutic used, liver-metastatic colon carcinoma cell lines show similar changes such as cell fragmentation, condensation, swelling and intracellular vacuoles²⁰⁵. This indicates that cytotoxic chemotherapy induces nonspecific tissue damage and explains why a similar host response can be observed, regardless of chemotherapy^206-209. Likewise, IVI was used to follow the fate of colon cancer cells migrating from spleen to liver. That study followed, at the single-cell level, the effect of 5-fluorouracil on cancer cell adhesion to endothelial cells in liver sinusoids, as well as the cytotoxic effect of 5-fluorouracil on early metastases (4–10 days after intrasplenic tumour injection) and late metastases (2–8 weeks after intrasplenic injection). 5-Fluorouracil did not affect cancer cell extravasation, but showed cytotoxic activity at early and late stages of metastasis²¹⁰.

In summary, IVI has substantially contributed to our understanding of the tumour–host interaction during various anticancer treatment modalities. These insights should enhance our ability to successfully treat patients with advanced cancer.

Future perspectives

Limitations of IVI

Despite advances that have made IVI more accessible and versatile, several limitations still hamper widespread adoption of the technique. These include limitations on imaging depth; garnering information on cellular identity and functional state; the ability to control experimental conditions; and image analysis. In this section we present some of the initial attempts to address these shortcomings and lay out the need for further development.

Increasing depth of penetration

Although still extremely shallow compared with clinical imaging techniques, multiphoton imaging provides some of the greatest penetration depths compared with other optical microscopy techniques, with an approximate twofold to fivefold increase over confocal microscopy^211,212. The ultimate depth of penetration of multiphoton microscopy varies greatly with the scattering properties of the imaged tissue. Relatively uniform tissues such as brain offer the greatest depth of penetration (~500 μm), while those consisting of multiple layers of lipid–water–air interfaces (such as the lung) rapidly destroy the illumination’s point spread function (a crucial parameter for multiphoton signal generation) and limit imaging depth (~30 μm).

Attempts to increase this limit have mainly focused on reducing tissue scattering by increasing the wavelength of illumination and/or emitted light. Thus, the development of far-red fluorophores has been an active area of research^213-216, and has motivated the push to extend the wavelength range of multiphoton microscopes by using pulsed light sources that can efficiently excite them (for example, using optical parametric oscillators^114,217, fibre lasers²¹⁸ or three-photon excitation²¹⁹). More recently, another tack has been to try to compensate for the point spread distortions using adaptive optics with two-photon^220-222, three-photon²²³ and even four-photon²²⁴ excitation. These techniques have extended the depth of penetration for some tissues (such as brain) to more than 1 mm (ref.²²⁵).

Correlative IVI

Multiphoton microscopy is excellent at revealing the behaviour and dynamics of cells and tissues. However, structures that lack fluorescence- or harmonic-generating molecules do not produce any optical signal, and thus do not appear in acquired images. Histochemistry and immunostaining, on the other hand, are extremely versatile in their ability to label cells and structures, but, as they require fixation and mechanical sectioning of tissues, they can give only single time point snapshots and cannot capture cellular or tissue dynamics.

To overcome these limitations, two groups have developed methods for mechanically sectioning the fixed tissue after IVI to capture the cells contained within the IVI plane of focus for further analysis by fixed–frozen histochemistry and immunostaining²²⁶ or by electron microscopy^227,228. Although limited by their labour-intensiveness, and by the small number of cells (~5–10) that can be correlated, these techniques hold the promise to link behavioural phenotypes with cellular, molecular and ultrastructural identity and composition. Improved methods that can increase throughput, as well as the number of aligned cells by orders of magnitude, are still needed.

Combination with other preclinical research modalities

A wealth of information can be obtained by the rapidly advancing methods for accessing and analysing the DNA, RNA, protein, lipid and metabolite content of cells. Powerful techniques capable of extracting information on limited numbers of targeted analytes (for example, limited sets of genes, transcripts, proteins and so on) are now being expanded into ‘omics technologies for unbiased discovery-based research^229,230. As originally developed, many of these approaches required the isolation and destruction of cells and tissues during the analysis, resulting in loss of their original spatial relation to other cells in the tissue. However, recent advances have made it possible to perform these analyses while preserving the architecture of the tissues. Mass spectrometry imaging²³¹, multiplexed fluorescence in situ hybridization²³², highly multiplexed immunohistochemistry²³³ and immunofluorescence²³⁴, and spatial transcriptomics and proteomics²³² could all be readily combined with IVI, and we expect such combinations to lead to new insights into mechanisms of metastasis.

Functional imaging

Even with the advantages that correlative microscopy provides, the aforementioned techniques still deliver information on the state of the cells at single time points. While advances have been made in the design and use of fluorescent reporters and biosensors, without more sophisticated methods for manipulating cells in vivo, IVI remains largely an observational and correlative technique.

Conventional methods for studying molecular mechanisms in vivo use approaches where gain-of-function and loss-of-function experiments are affected either via the systemic application of drugs or via the genetic manipulation (constitutively or chemically induced) of germline cells or somatic cells introduced into naive host animals. While powerful, these approaches lack the spatial and temporal control to precisely target cells without unwanted on-target and off-target side effects.

To truly realize the goal of performing in vivo studies that can define and test molecular mechanisms at single-cell resolution, in real time and longitudinally, researchers need to experimentally manipulate cellular processes locally and transiently, all in a native tissue context that is connected to the lymphatic and cardiovascular circulatory systems (which could, for example, provide immune infiltrates). The field of optogenetics (the combination of genetics and optics to control protein function with light²³⁵) holds the promise of fulfilling this need. The past two decades have seen an enormous growth in the number and variety of optogenetic tools that can regulate cellular processes^236-238.

Optogenetic tools now exist to control a variety of cellular functions, including membrane potentials (via light-controlled ion channels and pumps)²³⁹, enzymes²⁴⁰, signalling receptors²⁴¹, protein activity (by controlling protein localization²⁴² or binding²⁴³), and gene expression²⁴⁴ and repression²⁴⁵. Despite these advances, the use of optogenetic tools in vivo has mainly been in neuroscience applications²³⁷, although use is increasing in cardiovascular research²⁴⁶ and developmental biology²⁴⁷. Since new CRISPR technologies are greatly simplifying the incorporation of optogenetics into mouse models^248,249, the main stumbling block to widespread adoption of optogenetics is the slow and often complex illumination schemes that are required to activate processes such as transcription. Miniaturized wearable light-emitting electronic devices are now being developed, although almost exclusively for neuroscience applications^250-253.

Artificial intelligence and deep learning

Finally, one of the major bottlenecks for IVI experiments is image analysis, which has been limited to manual (or at best semi-automated) techniques. This is because, unlike images of in vitro cultures, where cells can be sparse and easily separated, cells in tissues are closely packed, and often of similar intensities. The lack of fully automated analyses is particularly problematic when time-lapse IVI or LVHR-IVI is used, as these techniques generate voluminous amounts of data, making better machine vision algorithms sorely needed. Very few investigations have used artificial intelligence on IVI data to date^254,255, although efforts are under way to establish databases of IVI images²⁵⁶, a crucial prerequisite for the development of machine learning algorithms.

Furthermore, IVI images present a wealth of data that may not be easily interpretable without a better understanding of the behaviour of cells within tissues. Thus, application of artificial intelligence or deep learning techniques may enable the discovery of hidden phenotypes^123,257 that are predictive of cellular behaviours under specific stimuli, or within specific microenvironmental niches.

Conclusions

There is a consensus that the TME determines tumour phenotype and response to treatment in ways that transcend the genetic mutations that drive tumour growth. Understanding the mechanisms behind the role of the TME requires real-time and longitudinal observations in living animals at single-cell resolution. Multiphoton IVI has now progressed to the stage where it can be relied on as a major tool in the effort to study the TME and its role in the host response.

New technologies can be used to define clinically actionable mechanisms responsible for tumour cell dissemination and metastatic recurrence, and can overcome previous obstacles to visualizing tumour cell motility phenotypes in vivo. These technologies have allowed the visualization of the tumour cell migration phenotypes associated with tumour cell dissemination and metastasis, including the intravasation of single tumour cells, de novo tumour cell cluster formation inside blood vessels and extravasation at metastatic sites. They also hold the promise of resolving remaining paradoxes in the literature regarding the relative contributions of single-cell migration and intravasation compared with collective migration and collective vascular invasion. IVI has led to the discovery of TMEM doorways involved in dissemination of tumour cells at both primary tumours and metastatic sites, and new prognostics and treatment options for use in combating distant metastatic recurrence in patients with breast cancer. IVI has also given direct insights, at single-cell resolution, into the effects of systemic therapies on, and the contribution of the microenvironment to, the host response.

Finally, IVI has played a central role in elucidating the mechanisms underlying clinically observed therapy failures, in particular those associated with changes in the TME, and has additionally pointed to optimal drug combinations that can overcome this failure. As interactions between cancer cells and the TME are often dynamic, IVI will continue to be the method of choice to understand how cancer cells behave in their natural environment, longitudinally during cancer progression and in response to treatment.

Glossary

Acoustic vaporization

Vascular disruption induced by vaporization of microscale or nanoscale droplets using ultrasound sonication.

Chemotactic index

A metric capable of measuring the directed movement of an organism or entity in response to a chemical stimulus.

Collective migration

The coordinated movement of a large cluster of adherent tumour cells that retain homotypic cell–cell junctions out of the tumour mass and towards the stromal blood vasculature.

Confocal microscopy

A form of high-resolution optical microscopy that captures optical sections with single-cell resolution by collecting light only from a single slice of an otherwise intact tissue.

Fluorescence ratio imaging microscopy

An optical microscopy technique that measures the ratio of two different wavelengths to quantify shifts in the fluorescence spectra of probes without the influence of system-specific parameters (for example, detector quantum efficiency, excitation intensity and optical path length).

Liver sinusoids

Highly specialized endothelial cells forming fenestrated blood vessels of the hepatic microcirculation.

Machine vision

Imaging-based automatic inspection and analysis.

Magnetic resonance imaging

A relatively low resolution but deeply penetrating imaging technique, commonly used in the clinic, which measures the effect of strong magnetic fields on the nuclei of water molecules to form an image of the structure, and sometimes function, of tissues and organs.

Mosaicked pattern

A combination or merger of multiple high-resolution, high-magnification images of a sample into a single image, producing a low-magnification, high-resolution image.

Multiphoton microscopy

A form of optical microscopy that captures optical sections with single-cell resolution by using femtosecond pulsed lasers to limit signal generation to a single slice of an otherwise intact tissue.

Optical sections

Images of a single thin plane from within a thick sample captured by removal of out-of-focus light. The name originates from the similarity these images have to those obtained from mechanically sectioned and stained tissues.

Phosphorescence quenching microscopy

In the biological sciences, an optical microscopy technique for the sensitive evaluation of oxygen consumption using optical oxygen sensors whose phosphorescence changes with oxygen concentration.

Point spread function

The 3D diffraction pattern of light formed at the focus of a lens.

Positron emission tomography

A low-resolution but deeply penetrating imaging technique, commonly used in the clinic, which forms images of the physiological function (blood flow, metabolism, neurotransmitters and drug accumulation) of organs using the emissions of an intravenously injected radioactive drug (called a ‘tracer’).

Second-harmonic generation signal

Light originating from second-order non-linear optical scattering by non-centrosymmetric molecules such as collagen or microtubules.

Vessel co-option

The ability of tumour cells to incorporate pre-existing vessels from surrounding tissues, rather than using angiogenesis.