Molecular mobility and activity in an intravital imaging setting – implications for cancer progression and targeting

Intravital or in vivo microscopy (IVM) has emerged as a powerful technique for the anatomical and functional mapping of cell biology in live mice from the subcellular level to whole-body approaches (reviewed in Conway et al., 2014; Ellenbroek and van Rheenen, 2014; Vennin et al., 2016). The discovery of fluorescent proteins, such as green fluorescent protein (GFP) and related variants (Chalfie et al., 1994; Giepmans et al., 2006; Miyawaki et al., 1997; Prasher et al., 1992), and their introduction into mice by genetic engineering has opened exciting opportunities to track live cell biology and molecular dynamics in vivo. Combined IVM of fluorescent proteins with other fluorescently tagged probes, such as antibodies, drugs or nanoparticles, while also exploring the inherent or autofluorescent properties of endogenous compounds, such as extracellular matrix (ECM) or metabolites (see Table 1 for further detail) can provide a comprehensive biological insight into live animals with an improved fidelity compared with cell and tissue-culture models. Furthermore, the use of multiphoton microscopy with pulsed infrared lasers provides key advantages for IVM as this can significantly extend the imaging depths owing to the reduced absorption and scattering at longer excitation wavelengths, while providing strong optical sectioning. This is essential for imaging thick samples, without the need for a confocal pinhole, thereby reducing the effect of sample aberration on the image quality relative to confocal microscopy. The relative advantages of different microscope platforms are discussed in detail elsewhere (Follain et al., 2017; Pantazis and Supatto, 2014; Timpson et al., 2011).

In recent years, the promise of IVM has been vividly realised, moving from proof-of-concept studies to providing critical quantitative insights into biological processes, which would be impossible using in vitro studies. In this Review, we discuss recent advances in IVM in mice, highlighting intravital applications to study development and tissue homeostasis, along with cancer aetiology and treatment. For IVM applications in other areas of biology, such as neuroscience or immunology, the reader is referred to excellent recent reviews (Osswald and Winkler, 2013; Secklehner et al., 2017). We also provide a summary of the emerging use of more advanced, subcellular imaging technologies in IVM and an overview of imaging applications in humans, while concluding with a discussion on the challenges and bottlenecks faced in IVM.

IVM in living mice poses several challenges that need to be considered; the first is tissue accessibility. To overcome this challenge for short-term and medium-term imaging, the tissue of interest can be surgically exposed, while preserving its structure, connectivity and perfusion in the anaesthetised mouse. For example, IVM of lipid droplets in epithelial cells of a surgically exposed lactating mammary gland highlighted the fate of lipid droplets from their synthesis in the endoplasmic reticulum to secretion at the apical plasma membrane in a live mouse (Fig. 1A; Masedunskas et al., 2017). Furthermore, IVM in surgically exposed tissues was used to characterise the role of the actomyosin cytoskeleton during exocytosis in the salivary gland (Milberg et al., 2017), to probe secretion and paracellular transport in the liver (Porat-Shliom et al., 2016) and to visualise T- and B-cell interactions in the germinal centre of intact lymph nodes (Xu et al., 2013a).

Some IVM approaches do not require surgical exposure of the tissue of interest. For example, live cell imaging in the retina was recently performed to observe dynamic radial infiltration patterns of immune cells in a mouse model of autoimmune retinal inflammation (Bremer et al., 2016). The auricular skin of the ear provides another easily accessible site to extract valuable in vivo cell biology without the need for surgical intervention. Here, Li et al. (2016a) characterised how neutrophils take up fluorescent immune complexes and transmigrate with their cargo across the endothelium into the interstitium (Fig. 1B). Similarly, intravital photoconversion of the fluorescent protein Kikume, whose fluorescence irreversibly changes from green to red upon excitation with ultraviolet or violet light, was used to delineate pathways of Kikume-expressing neutrophil migration and response to pathogen introduction (Hampton et al., 2015). Such novel insights into neutrophil dynamics are important, considering that neutrophil recruitment has been shown to have a role in injury and diseases, such as infection or cancer progression, as well as their treatment (Hind et al., 2016; Powell et al., 2017; Steele et al., 2016).

Surgical insertion of optical imaging windows adjacent to an organ of interest allows the transient stabilisation of surgically exposed tissues and repeated long-term imaging of live tissue in the anaesthetised mouse (Bochner et al., 2015; Kedrin et al., 2008; Nobis et al., 2017; Ritsma et al., 2014, 2012, 2013b; Rodriguez-Tirado et al., 2016; Tabuchi et al., 2008). For instance, IVM of megakaryocytes and platelets in lung, liver, spleen and bone marrow through windows in fluorescent, genetically engineered mouse models (GEMMs) revealed that the lungs are a major site of platelet production and contain a pool of hematopoietic progenitors (Fig. 1C; Lefrançais et al., 2017). Similarly, intravital lung imaging of fluorescent antibody-labelled immune cell populations was used to characterise the in vivo recruitment dynamics of invariant natural killer T-cells, neutrophils and dendritic cells (Thanabalasuriar et al., 2016).

Cranial windows have been used to describe a detailed pathway that cells use to arrive and transmigrate into bone marrow sinusoids during homing (Bixel et al., 2017) and to track the fate and behaviour of transplanted bone marrow cells in the native bone microenvironment (Le et al., 2017). Other window approaches are specifically suited for surgical insertion at the cervical, thoracic or lumbar level along the spinal cord for IVM approaches in the live spinal cord microenvironment during tissue homeostasis or inflammation (Bartholomäus et al., 2009; Davalos et al., 2008; Haghayegh Jahromi et al., 2017; Kim et al., 2010; Odoardi et al., 2007; Yang et al., 2017; Zenaro et al., 2013).

Longitudinal tracking through imaging windows has also revealed novel insights into the positioning, maintenance and plasticity of mammary and intestinal stem cells (Fig. 1D; Ritsma et al., 2014; Scheele et al., 2017). Similarly, IVM through imaging windows demonstrated a previously unknown mechanism of intestinal crypt fusion alongside the previously described crypt fission, which may underlie the regulation of the crypt number in the adult mouse (Fig. 1E; Bruens et al., 2017).

This wide range of IVM tools and applications has not only shed new light on normal biology and tissue homeostasis but also had an impact on our current understanding of cancer progression and treatment, which will be discussed in detail below.

Several approaches have been taken to label the tumour and its microenvironment using GEMMs, bioluminescence, or untargeted or targeted labelling of cells or their surrounding environment (see Table 1), with the aim of tracking cancer progression, metastasis and treatment (Ilhan-Mutlu et al., 2016; Jung et al., 2017; Osswald et al., 2015; Park et al., 2016b). In this context, IVM can be used to monitor cancer and stromal cells, thereby helping to characterise spatiotemporally dynamic events, such as local cancer invasion and distant metastases, as well as evaluating targets of drug treatment or pharmacodynamics and pharmacokinetics (Conway et al., 2014; Dubach et al., 2017; Miller and Weissleder, 2017).

For example, IVM tracking of the interaction of malignant B-cells with their microenvironment in the calvaria was used to delineate the upstream regulators of α4β1 integrin activity in cancer cell adhesion in vivo, which could in future be targeted to disrupt interactions between cancer cells and their environment in this hard-to-treat site (Martínez-Moreno et al., 2016). Another study described the bone marrow seeding of leukaemia cells in vivo and their transient and promiscuous interactions with their microenvironment (Fig. 2A; Hawkins et al., 2016). Here, IVM revealed a progressive remodelling of the bone marrow microenvironment during disease progression that was confirmed in biopsies from human patients, which emphasises the value of IVM in pre-clinical models to gain insights into disease aetiology (Hawkins et al., 2016).

Similarly, IVM was used to demonstrate heterogeneity in cell motility and the mode of cell migration in melanoma. These intravital findings were combined with transcriptomic analysis to identify gene networks that are differentially expressed between motile and non-motile cells and these may represent potential predictors of cell mobility and invasiveness in patient samples (Manning et al., 2015). IVM tracking also showed that changes in breast tumour cell velocity, in response to the LIM domain kinase (LIMK) inhibitor Pyr1, depended on the morphology of the tumour cell in vivo (Prunier et al., 2016), where, upon drug treatment, elongated cells slowed down, whereas round cells sped up. This demonstrated a transient response to drug treatment in individual cells depending on cellular morphology and treatment timing, which cannot be observed in fixed tissue samples, but is important to guide future treatment regimens. Similarly, single-cell analysis in pancreatic cancer has also revealed heterogeneity in response to treatment (Erami et al., 2016; Nobis et al., 2017, 2013).

Furthermore, the pharmacodynamics and pharmacokinetics of chemotherapy can often be elucidated using IVM. For example, IVM demonstrated that tumour cells display delayed onset of cell death following treatment with paclitaxel in vivo compared with cancer cells examined in vitro, highlighting that the tumour microenvironment influences chemotherapeutic efficacy (Orth et al., 2011). Moreover, for detailed characterisation of in vivo pharmacokinetics, a fluorescent analogue of the chemotherapeutic drug eribulin was administered and imaged through a window chamber in a mouse model of spatially heterogeneous expression of taxane resistance (Laughney et al., 2014). Here, resistant cells, visualised by the high expression of a fluorescent multi-drug resistance 1 (MDR1) fusion protein, did not accumulate the eribulin analogue, whereas MDR1 inhibition improved its intracellular accumulation. Importantly, such single-cell readouts of drug uptake and efficacy can help to provide real-time insight into the mechanisms of drug efficacy and drug resistance in vivo and to establish optimised treatment regimens to overcome the intratumoural heterogeneity in drug response and resistance.

IVM can also provide novel insights into dynamic aspects of the metastatic cascade in vivo, such as epithelial-to-mesenchymal transition (EMT), which can often precede and permit cancer metastasis, and mesenchymal-to-epithelial transition (MET), which might allow metastatic cancer cells to colonise secondary sites. Zhao et al. (2016) simultaneously observed in vivo EMT alongside changes in cell migration and morphology in the MMTV-PyMT model of invasive and metastatic breast cancer, in which expression of the oncoprotein polyoma middle T antigen (PyMT) under the control of the mouse mammary tumour virus long terminal repeat (MMTV-LTR) drives cancer progression (Lin et al., 2003). Here, IVM can provide insight into single-cell EMT events in their native environment, as well as distinct real-time readouts of drugs targeting the EMT process or cells that have undergone EMT (Zhao et al., 2016). De-regulated expression of the cell–cell junction protein E-cadherin is also often associated with EMT and cell invasiveness (Gregory et al., 2008; Lehmann et al., 2016b). Using a GEMM with fluorescent protein expression from the endogenous E-cadherin locus to monitor changes in E-cadherin expression intravitally, Beerling et al. (2016) demonstrated epithelial–mesenchymal plasticity in breast cancer. They show that mesenchymal cells with low E-cadherin expression have a similar potential as epithelial cells with high E-cadherin expression to give rise to epithelial metastases. This indicates that cancer cells can rapidly switch between epithelial and mesenchymal states in vivo, which may support them in adapting to new environments and promoting secondary growth (Fig. 2Bi; Beerling et al., 2016). Therefore, IVM can help to determine the epithelial–mesenchymal status of cancer cells and the optimal timing of drug treatment to most effectively counteract EMT and MET switching during the transit of cancer cells to secondary sites.

Cancer cells can also invade and metastasise in the absence of EMT, retaining and positively utilising E-cadherin expression for locomotion. Indeed, it was recently demonstrated that cancer-associated fibroblasts (CAFs) can support collective cancer cell invasion (Gaggioli et al., 2007; Labernadie et al., 2017; Sanz-Moreno et al., 2011). For example, heterophilic interaction between CAF N-cadherin and cancer cell E-cadherin results in the transduction of mechanical forces that pull cancer cells away from the primary tumour and re-polarise CAFs, resulting in a net collective movement (Fig. 2Bii; Labernadie et al., 2017). Similar heterotypic interactions have been observed in border cell migration in Drosophila and other collective movements in development (Cai et al., 2014; Kardash et al., 2010; Wang et al., 2010) and thus may be a process co-opted in cancer to facilitate collective migration (Friedl and Gilmour, 2009; Ilina and Friedl, 2009).

Photoconversion and the use of photoswitching proteins offer another excellent opportunity to follow cell invasion and metastasis. For example, tracking of photoswitched mammary tumour cells showed that, compared with primary tumour cells distant from blood vessels, cells in close proximity to blood vessels migrate extensively within and outside of the primary tumour, and later on, are also detectable at secondary sites (Kedrin et al., 2008). Similarly, genetically engineered murine and patient-derived colon organoids that express known mutational drivers of colorectal cancer progression and metastasis were longitudinally tracked following photoswitching to define the cumulative mutational load required for tumour cell migration and subsequent metastasis (Fumagalli et al., 2017). This study demonstrated that a combined mutational load affecting several signalling pathways allows colorectal cancer cells to grow independently of signals from their microenvironment and so promote metastasis.

IVM therefore serves as a versatile tool to obtain detailed spatiotemporal information of disease progression and to pinpoint those stages of cancer progression that are most vulnerable to drug action in vivo for improved pre-clinical characterisation of drug treatment regimens.

In vivo imaging has been used to characterise the dynamic interactions between cancer cells and the tumour-associated immune system, which are known to have a key role in cancer progression, metastasis and treatment. For instance, labelling of tumour-infiltrating immune cells by in vivo photoconversion showed that some immune cell types are retained in the primary tumour, whereas others, such as effector T-cells, which play a role in anti-tumour immunity, can readily migrate to draining lymph nodes and even to secondary tumour sites (Torcellan et al., 2017). These findings could be further explored to promote T-cell-mediated anti-tumour immunity at secondary sites. IVM has also been used to monitor the pro-metastatic support that immune cells can confer to cancer cells. Intravital lung imaging through windows showed the formation of neutrophil extracellular traps (NETs), which are an extracellular network of DNA and cytotoxic enzymes released by neutrophils to bind and kill pathogens (Park et al., 2016a). In the absence of infection, breast cancer cells could induce NET formation, thereby stimulating cancer invasiveness and metastasis, which could be inhibited upon administration of DNaseI-containing nanoparticles (Park et al., 2016a). Similarly, IVM of the lung immune response upon metastatic seeding showed that many metastatic pioneer cells, which are released from the primary tumour into the circulation, fragmented upon shear stress within blood vessels (Headley et al., 2016). The resulting microparticles were taken up by immune cells that arrived at the metastatic niche in distinct waves. Using GEMMs, which lack distinct subsets of immune cells, the authors showed that microparticle-containing myeloid cells might promote metastatic seeding, whereas a rare dendritic cell subtype, which also ingested microparticles, was found to have anti-metastatic effects, highlighting the dual nature of the immune response to metastasis (Headley et al., 2016). These observations exploited the transfer of the fluorescent label from cancer cells into the tumour-ingesting immune cells by phagocytosis; many IVM approaches make use of such inherent cellular processes. For example, IVM demonstrated that fluorescently labelled bisphosphonates, which have been shown to possess anti-tumour activity, were taken up by tumour-associated macrophages (TAMs), potentially corrupting their tumour-promoting functions (Junankar et al., 2015). Here IVM identified the cell types targeted by bisphosphonates in cancer and provided novel insight into the previously uncharacterised anti-cancer activity of these drugs.

IVM has also been used to assess the performance of immunotherapy or blockade of immune checkpoints. For example, treatment of colorectal cancer, lung cancer and melanoma xenograft models with fluorescently labelled anti-PD1 (PD1 is also known as PDCD1) showed that TAMs take up anti-PD1 from the surface of cytotoxic T-cells (CTLs), which is mediated by FCγ receptors (Fig. 2C; Arlauckas et al., 2017). Based on these observations, combination treatments consisting of FCγR inhibition and anti-PD1 antibodies were tested in vivo, which improved immunotherapy performance by extending the time anti-PD1 localised to CTLs (Arlauckas et al., 2017). Similarly, longitudinal IVM of interactions between cancer and immune cells revealed the formation of an immunosuppressive ‘ring’ of T-regulatory cells around the tumour that can inhibit CTL efficacy (Qi et al., 2016). Indeed, breaking this ring of T-regulatory cells by using cyclophosphamide accelerated CTL infiltration into the tumour.

Moreover, IVM of melanoma immunotherapy contributed to a better therapeutic outcome in mice as it demonstrated that treatment with an anti-CD137 agonist antibody improved the anti-tumour effector functions of both endogenous and transplanted CTLs (Weigelin et al., 2015). Here, anti-CD137 agonist antibody treatment resulted in increased intratumoural CTL viability, reduced CTL migration speed, as well as prolonged interaction between CTLs and tumour cells and thus increased killing efficacy upon contact with a tumour cell (Fig. 2D; Weigelin et al., 2015). IVM has also been used to test the efficacy of new anti-cancer immunotherapy agents. For example, treatment with a new T-cell bispecific antibody that directly links T-cells with cancer cells resulted in faster and more persistent migration of freely moving T-cells, an increase in tumour-associated T-cells, and prolonged interactions between T-cells and cancer cells, all of which accelerate tumour cell death (Lehmann et al., 2016a). Taken together, these studies demonstrate the usefulness of IVM in fine-tuning or enhancing anti-cancer therapy.

Blood vessels serve as carriers for oxygen, nutrients and drugs, but can also allow tumour cells to spread to distant sites in the body. IVM offers an opportunity to establish functional landmarks of tissue and cancer biology in addition to the anatomical landmarks obtained from immunohistochemistry. Indeed, a distinct tumour microenvironment of metastasis (TMEM) has been described that can be detected in both GEMMs of breast cancer and human breast cancer patients (Harney et al., 2015). IVM in MMTV-PyMT mice demonstrated a local loss of vascular junctions at TMEMs followed by cancer cell intravasation, thereby providing novel insights into the mechanisms of cancer metastasis (Harney et al., 2015). Similarly, IVM of pancreatic tumour xenografts demonstrated different mechanisms of vascular permeability with transient openings of the vessel wall, as well as vascular bursts visualised by the leakage of fluorescently labelled nanoparticles into the adjacent tissue; both mechanisms may support drug delivery into the extravascular tissue (Matsumoto et al., 2016). Future investigation may show whether these observations align with previous research on the TMEM in breast cancer (Harney et al., 2015).

Localised vascular leakage has recently gained attention with regard to potential cancer treatment. IVM of tumour-associated blood vessels, TAMs and fluorescently labelled or therapeutic nanoparticles revealed that radiotherapy increased TAM density and resulted in enlarged and permeabilised blood vessels with an increase in vascular bursts followed by nanoparticle flux into the tumour tissue (Fig. 2E; Miller et al., 2017). Interestingly, TAM accumulation correlated with and was required for vascular bursts following irradiation, and the combination of radiotherapy and therapeutic nanoparticles reduced tumour growth in a TAM-dependent manner (Miller et al., 2017).

However, the induction of localised vascular bursts and extravascular leakage at breast cancer TMEMs has also been observed upon neoadjuvant chemotherapy (Karagiannis et al., 2017). Chemotherapy using paclitaxel enhanced TMEM assembly through an increased infiltration of macrophages and thereby promoted metastasis despite a reduced growth of the primary tumour, whereas inhibition of TMEM activity reduced these pro-metastatic effects. Another approach to exploit vascular permeability for improved nanoparticle delivery is the localised depletion of tumour-associated platelets, which have been shown to preserve the integrity of tumour-associated blood vessels (Li et al., 2017).

Future detailed intravital monitoring of the tumour microenvironment in the presence or absence of therapy is likely to provide additional insights into the role of localised vessel permeability with the aim to balance localised vascular permeability to improve drug delivery without opening the floodgate to metastasis.

IVM can be used to better understand cancer cell–cell communication at a distance. Extracellular vesicles (EVs) or exosomes can carry nucleic acids, proteins or lipids to transfer information between cells (reviewed in detail in Becker et al., 2016; Raposo and Stoorvogel, 2013). For example, invasive and metastatic cancer cells can secrete EVs or exosomes that contain pro-invasive and pro-metastatic ‘signals’ (Singh et al., 2016; Yokoi et al., 2017; Zomer et al., 2015). Recently, Zomer et al. (2015) used Cre-recombinase-carrying exosomes that induce a switch in fluorescent protein expression upon uptake to identify EV-recipient cells in vivo. This showed that EVs derived from malignant cancer cells can enhance the metastatic potential of less-malignant cells, revealing the global challenge we face when treating primary and secondary tumour burden. Similarly, IVM in a glioma mouse model provided novel insight into cancer–stroma communication by demonstrating the uptake of glioma-derived EVs by the stromal microglia and other immune cells in vivo (van der Vos et al., 2016). Exosomes have also been shown to promote fast and directional tumour cell migration. Autocrine exosome communication can support and stabilise cell–substrate adhesion and increase the persistence of migration directionality (Sung et al., 2015). Furthermore, exosomes have been shown to be transferred to distant sites, such as liver (Costa-Silva et al., 2015) or lung (Liu et al., 2016), where they manipulate resident host cells, preparing a pre-metastatic niche for efficient metastasis (Hoshino et al., 2015; Peinado et al., 2017).

A number of recent studies investigated the potential of exosomes to deliver anti-cancer cargo into the tumour. For example, application of tumour-targeting exosomes engineered to carry shRNA or siRNA against Kras^G12D, a common driver mutation in pancreatic cancer, as well as in human pancreatic cancer patients, resulted in improved survival and reduced metastasis (Kamerkar et al., 2017). With refined labelling techniques for EVs and exosomes, further studies may inform us how best to treat primary versus metastatic disease (Steeg, 2016) and prevent communication with other cancer cells and distant sites (Peinado et al., 2017). Future IVM studies may thus provide further insight into the multifaceted role of EVs in disease progression.

Subcellular IVM approaches (Box 1) can be used to probe dynamic subcellular processes, such as protein dynamics, protein–protein interactions or protein activity, making these techniques attractive to assess the molecular basis of cancer progression and treatment at a level beyond the single-cell level discussed above. For example, photobleaching techniques, such as fluorescence recovery after photobleaching (FRAP; Fig. 3A) and fluorescence loss in photobleaching (FLIP, Fig. 3C), can be used to characterise the mobility of fluorescently tagged proteins or fluorescent agents (see Fig. 3 and Box 1 for further details).

The use of photobleaching IVM has provided many insights into the molecular dynamics of key biological processes in intact tissues. For example, Machado and Mitchell (2011) used a dorsal skinfold chamber model to track blood vessel ingrowth into a wound site upon heat injury. Using intravital FRAP imaging of FITC-dextran during re-vascularisation they were able to distinguish between vessels of different maturity based on their fluorescence recovery profiles and half-times of recovery. Elegant studies in human tumour xenografts have used FRAP to show how ECM abundance and cellular geometry affects interstitial mobility and diffusion of macromolecules and propose the reduction of ECM content as a potential measure to overcome chemoresistance (Netti et al., 2000; Pluen et al., 2001).

Other IVM studies suggest that increased mobilisation of E-cadherin in cell–cell junctions may be a feature of migrating cells (Canel et al., 2010; Serrels et al., 2009). Recently, an E-cadherin–GFP GEMM was combined with subcellular FRAP and FLIP imaging to provide further insight into in situ E-cadherin dynamics in a tissue-specific microenvironmental context (Fig. 3B,D; Erami et al., 2016). FRAP and FLIP analyses revealed that E-cadherin is mobilised in invasive pancreatic cancers that are driven by mutations in both Kras and Tp53 compared with a non-invasive pancreatic cancer model driven by Kras mutation and loss of Tp53. Because KRAS and TP53 are commonly mutated in the human disease (Biankin et al., 2012), live FRAP and FLIP imaging could be used to visualise the subtle but important dissolution of tumour cell–cell junctions during early stages of pancreatic tumour mobilisation in vivo. Such an approach was subsequently used to assess anti-invasive drug targeting in vivo to target early invasion or pre-EMT events.

Genetically expressed Förster resonance energy transfer (FRET) biosensors have been used to visualise the subcellular activation status of selected proteins or signalling pathways in an intravital context (see Fig. 4 and Box 1 for further details). For instance, intravital FRET imaging of cell cycle arrest using a CDK1-FRET biosensor (Fig. 4B) was used to optimise chemotherapy performance in murine and patient-derived models of pancreatic cancer by whole-body priming with the ROCK inhibitor Fasudil (Fig. 2F; Vennin et al., 2017). By imaging the biosensor alongside the ECM architecture and vessel leakiness, a transient manipulation of the stromal ECM, rather than chronic treatment, was shown to be effective in improving chemotherapy efficacy. Similarly, Hirata et al. (2015) used intravital FRET imaging of an ERK-FRET biosensor (Fig. 4B) to demonstrate that the BRAF inhibitor PLX4720 activates melanoma-associated fibroblasts, which drive ECM production, remodelling and stiffening to protect melanoma cells against BRAF inhibition. Here, the combination of PLX4720 with FAK (also known as PTK2) inhibition, which targets the integrin-β1–FAK–Src signalling axis restored the vulnerability of melanoma cells to BRAF inhibition. Simultaneous fluorescence and FRET imaging was used in a different study to assess both cell cycle progression through photoswitching of H2B-Dendra and the onset of apoptosis with a Caspase-3 FRET biosensor (Fig. 4B) in the same cell upon docetaxel treatment (Janssen et al., 2013). Here, differences in drug response between in vitro and in vivo settings were observed: in vitro, docetaxel induces cell death via mitotic perturbation, whereas in vivo, docetaxel induces apoptosis independently of mitotic defects, emphasising the need to assess drug targeting in vivo. Validated FRET biosensors have been discussed in detail previously (Conway et al., 2014, 2017; Newman et al., 2011).

Mice can also be genetically engineered to express FRET biosensors to assess protein activity in situ. For instance, novel FRET biosensor mice for Caspase-1 and Caspase-3 activity (Fig. 4B) were used to image single-cell apoptotic events (Liu et al., 2014; Yamaguchi et al., 2011). Similarly, Kamioka and colleagues recently reported the generation of Eisuke-FRET and PKAchu-FRET mice to quantify the dynamics of two major signal transducers, ERK1/2 (collectively known as ERK and also known as MAPK3 and MAPK1) and PKA, respectively (Fig. 4B; Kamioka et al., 2012). Intracellular signalling via ERK and PKA during tissue homeostasis, wound healing, inflammation and upon drug targeting could be correlated with the live cellular phenotype observed (Kamioka et al., 2012; Mizuno et al., 2014, 2016; Sano et al., 2016). Furthermore, these FRET biosensor mice allowed the spatiotemporal assessment of cell signalling and pharmacodynamics in live tissue, such as the temporal propagation of ERK signalling in the epidermis (Fig. 1F; Hiratsuka et al., 2015) or the spatial heterogeneity in ERK signalling in breast tumours (Kumagai et al., 2015).

FRET biosensor mice represent a valuable tool to study the activity of difficult-to-assess signalling molecules such as rapidly switching small GTPases, which are active in the GTP-loaded state and inactive in the GDP-loaded state. This rapid change and the lability of active GTPases ex vivo make it difficult to quantify their activation via traditional biochemical approaches (Samuel et al., 2011). Recently, the generation of FRET biosensor mice for two prototypical members of the Rho family of small GTPases, Rac1 and RhoA, has been reported (Fig. 4B; Johnsson et al., 2014; Nobis et al., 2017). These FRET biosensor mice were used to showcase the dynamic regulation of small GTPase activity in vivo; live-imaging demonstrated oscillations in activity during directional neutrophil migration (Fig. 1G), regulation of activity during cyclic mammary gestation, co-option of small GTPase activity during breast and pancreatic cancer progression or valuable pharmaco-dynamic readouts during drug intervention.

Simultaneous imaging of several FRET biosensors for multiplexed readout of several signalling pathways has also been achieved through the excitation of several donor fluorophores using the same wavelength (Demeautis et al., 2017; Laviv et al., 2016). For example, Demeautis et al. (2017) reported dual FRET quantification of ERK and PKA FRET biosensors in the same cell to dissect the hierarchies, crosstalk and combinatorial inputs of intracellular signalling pathways. Similarly, a pair of new red fluorescent proteins was imaged in combination with an EGFP-Venus FRET biosensor in vivo, where CyRFP could be excited simultaneously with EGFP without an overlap of their emission spectra (Laviv et al., 2016), demonstrating that the simultaneous assessment of multiple protein activities is feasible.

As discussed above, intravital FRAP and FRET imaging can provide pivotal insights into the subcellular regulation of protein dynamics and activity. However, these approaches are limited by the optical restrictions of confocal laser scanning and multiphoton microscopy, including magnification, resolution or imaging depth. New multimodal imaging approaches whereby IVM is combined with electron microscopy, such as intravital correlative light and electron microscopy (CLEM) can help to reveal ultrastructural details of cells and ECM. The coupling of IVM in live tissue and electron microscopy in fixed tissue imposes several challenges on the user, including tissue fixation or retrieving the IVM region of interest (ROI) in the EM sample (discussed in more detail in de Boer et al., 2015; Follain et al., 2017; Karreman et al., 2016a). For example, near-infrared branding (NIRB) was used to permanently label intravital ROIs and to precisely localise them in the EM sample. This technology was used to reveal ultrastructural details of the microenvironment of rare cell biological events, such as cancer cell extravasation into the brain, which without prior IVM localisation may be challenging to find in a fixed sample (Karreman et al., 2016b).

Another approach to retrieve IVM ROIs ex vivo is cryosection labelling and intravital microscopy (CLIM), where intravitally imaged tissue is locally and permanently labelled by phototattooing of collagen fibres prior to fixation and embedding. This allows the user to identify and further analyse IVM ROIs ex vivo, as exemplified by the immunohistochemical characterisation of the microenvironments of migratory and non-migratory breast cancer cells (Ritsma et al., 2013a). Recent developments in intravital focusing of cells (Galanzha et al., 2016) or in matrix-assisted laser desorption/ionisation (MALDI) imaging (Grey et al., 2009) may open further avenues for combining IVM data with either in vivo flow cytometry applications or spatial proteomics.

Previously inaccessible tissues may be imaged using microendoscopic devices for IVM in awake animals; this recently allowed the simultaneous characterisation of behaviour and the quantification of Ca²⁺ dynamics (Ghosh et al., 2011) or of ERK and PKA signalling activity (Goto et al., 2015). Furthermore, new optical clearing techniques for large mammalian systems, such as Clarity, Scale, CUBIC (clear, unobstructed brain/body imaging cocktails and computational analysis), 3DISCO (3-dimensional imaging of solvent-cleared organs), SeeDB (see deep brain), ISDoT (in situ decellularization of tissues) or PACT (photoacoustic computed tomography), have improved the imaging resolution at depth by removing the cellular components of tissues that cause its opacity (Chung and Deisseroth, 2013; Fischer et al., 2017; Mayorca-Guiliani et al., 2017; Richardson and Lichtman, 2015; Tainaka et al., 2014; Tomer et al., 2014; Yu et al., 2017). Future combination of intravital studies with post-intravital optical clearing and subsequent high-resolution imaging, similar to CLEM, may provide new insight into disease aetiology and treatment.

IVM techniques are also used in targeted nanotherapy and phototherapy. For example, Chen et al. (2017) synthetised a near-infrared fluorescent probe that is cleaved by cathepsin B, which is highly expressed in tumour cells, and can subsequently be excited with near-infrared light to generate a cytotoxic agent. The authors tested their concept in vivo in subcutaneous tumours and showed that their photodynamic therapeutics approach prevented tumour growth without affecting normal tissues. Similar applications have recently been described to specifically induce the generation of cytotoxic singlet oxygen in cancer cells (Huang et al., 2016b), for the targeted labelling of tumour cells (Yan et al., 2016) or for the improved delivery and uptake of drug-carrying nanovesicles (Huang et al., 2016a), including the use of immune cells for the delivery of nanoparticles into the tumour (Chu et al., 2017).

Intravital multiphoton imaging can be used to detect endogenous autofluorescent compounds, such as ECM or metabolites, without introducing any exogenous labels or genetic modifications (Table 1). Here, label-free fluorescence lifetime imaging microscopy (FLIM, see Box 1 for further details) readily lends itself to applications in humans. For instance, FLIM has been demonstrated in human lung and skin samples to delineate the border of cancer lesions (Galletly et al., 2008; Wang et al., 2017), as well as in capillaries of the fingernail bed and of the forearm in vivo (Shirshin et al., 2017) or in the tongue of human volunteers (Cheng et al., 2014), which could also be adapted for the monitoring of oral cancers in humans. In a recent clinical trial, intravital tomography combined with FLIM of metabolites (see Table 1 and Box 1) demonstrated differences in mitochondrial density between healthy and inflamed skin and that an increase of free NADH in mitochondria correlated with the severity of inflammation (Huck et al., 2016). Thus, this approach could potentially be applied for the early identification of low-inflammation events and thus earlier intervention.

IVM has already been employed in human melanoma patients to image intravenously infused fluorescein to assess tumour vasculature and vascular patency (Fisher et al., 2016). By imaging tumour vessels through an incision into the skin, blood vessels were found to have an enlarged diameter in the in vivo setting compared with previous findings from fixed tumour tissues, further emphasising the importance of imaging cells and tissues in vivo in their native environment. Future developments in IVM setups such as these are likely to provide opportunities for IVM in the clinical setting.

IVM imposes a unique set of challenges from a technical perspective, in that users must deal with sample motion, tissue autofluorescence and a complex microenvironment. Another common challenge faced is the large volume of data, as is the case with genomics and proteomics. Here, we briefly discuss recent developments that may help overcome some of these challenges.

Currently, there are few commercial packages available that reliably correct for motion artefacts, hence most researchers rely on custom-written software packages. For stable IVM, the animal can be immobilised by using physical restraints or applying negative pressure to reduce tissue motion (Cao et al., 2012; Edelstein et al., 2014; Lee et al., 2012). Residual sample motion due to respiratory or cardiovascular movements can be overcome by synchronising image acquisition with physiological measurements, for example heartbeat (Aguirre et al., 2014; Vinegoni et al., 2015) or respiration (Lee et al., 2012). Image-based stabilisation can also be used to correct for lateral sample motion in post-processing (Fiole et al., 2014; Lee et al., 2014; Soulet et al., 2013; Vercauteren et al., 2006; Vinegoni et al., 2014). Motion of cell–cell junctions during intravital FRAP and FLIP has also been successfully compensated for using optical flow-based techniques (Erami et al., 2016). Axial sample motion in and out of the focal plane cannot be corrected post-imaging. In this case, it may be possible to use real-time correction approaches, where an optical coherence tomography (OCT)-based tracking system was used to adjust the position of the objective in real-time (Sherlock et al., 2015, 2017).

In IVM, penetration depth is often limited by aberration caused by the sample itself, which may be compensated for by using adaptive optics, for instance using a correction collar (Muriello and Dunn, 2008), a deformable membrane mirror (Caroline Müllenbroich et al., 2014) or segmented pupil illumination (Ji et al., 2012; Wang et al., 2014), which can correct for sample aberrations by increasing signal at depth. Alternatively, image-based metrics can be used to estimate the sample aberrations (Burke et al., 2015; Débarre et al., 2009; Song et al., 2010).

IVM imaging of dynamic and fast events, such as blood flow or mobile immune cells, can be accomplished using resonant scanners, which allow imaging speeds up to 30 frames per second (Kirkpatrick et al., 2012; Matsumoto et al., 2010; Nguyen et al., 2001) or multibeam imaging, which can also reduce photodamage by spreading the excitation power over a number of focal spots (Niesner et al., 2007; Rinnenthal et al., 2013; Shimozawa et al., 2013).

Many IVM setups have been customised to adapt to the numerous applications and imaging modalities of IVM, which can easily be adapted and expanded on. Alternatively, they can be assembled from individual software components that control distinct devices, such as stage or scanner, to enable precise control over image acquisition parameters, including HelioScanner (Langer et al., 2013), Micro-Manager (Edelstein et al., 2014) or ScanImage (Pologruto et al., 2003).

The diverse range of imaging environments for IVM means that custom analysis workflows tend to be developed for each project and few widely used general purpose tools exist. Workflows built with open source packages include ImageJ/Fiji (Schindelin et al., 2012), Icy (de Chaumont et al., 2012), Cellprofiler (Carpenter et al., 2006) or CellCognition (Held et al., 2010), and can complement the use of commercial tools, e.g. Imaris (Bitplane), Amira (FEI), MetaMorph (Molecular Devices), or Volocity (PerkinElmer). FRAP and FLIP data require quantitative analysis of small subcellular regions over long time periods and, as such, are highly sensitive to sample motion. The image stabilisation software discussed above can be applied directly to photo-bleaching data before analysis with conventional analysis tools (Kraft et al., 2014; Rapsomaniki et al., 2012; Schindelin et al., 2012). Intravital FLIM-FRET with a significant contribution from tissue autofluorescence can be analysed with FLIMfit (Conway et al., 2017; Warren et al., 2013) or phasor-based approaches can be used to separate the contributions from the FRET biosensor and tissue fluorescence (Digman et al., 2008).

A wide variety of tools to analyse and model cell migration in vitro and in vivo have been developed for analysis of cell migration (Alexander et al., 2008; Masuzzo et al., 2016), clonal cell development (Coffey et al., 2013) or for the automated 3D tracking of cells or nuclei (Keller et al., 2008; Li et al., 2007). For example, a study used a multiparametric classification approach to automatically identify tumour cell phenotype and to relate the features they identified to their microenvironmental context (Gligorijevic et al., 2014). Machine learning approaches have also been used to automate longitudinal cell cycle profiling using the fluorescence probe FUCCI (Chittajallu et al., 2015) or to identify haematopoietic stem cells in a heterogenous environment (Khorshed et al., 2015). Other software tools have been developed for the characterisation of cell movement in relation to complex microenvironmental cues, such as collagen orientation (Mayorca-Guiliani et al., 2017; Rezakhaniha et al., 2012), grey-level co-occurrence matrix analysis (GLCM; Cicchi et al., 2010) or gap analysis (Acton et al., 2014).

Here, guidelines for standardised image acquisition, annotation, storage, analysis and publication may allow for a better comparison between published works and lead to the generation of comprehensive atlases of single-cell and subcellular events. Several open source projects aim to provide a platform for storing, annotating and managing large datasets, such as OMERO (Allan et al., 2012), Bisque (Kvilekval et al., 2010) and Cytomine (Marée et al., 2016). Other projects such as the Image Data Resource (IDR) aim to facilitate the reuse and reanalysis of previously published data by integrating the analysis of multiple, disparate image studies (Williams et al., 2017). This common use and publication of IVM datasets may also maximise the overall scientific output by facilitating the analysis and extraction of information not assessed by the primary study, similarly to that of protein and gene lists derived from large ’omics datasets.

In conclusion, we have come a long way in this field from improving fluorophores, imaging components and key tools. The future of IVM as a stand-alone technique or in combination with other research areas, including molecular or cellular ’omics approaches, will help to encourage the use of IVM as part of a standard approach to study disease progression. Importantly, using IVM, will allow us to observe disease in a high-fidelity, functional and more physiological setting which is of the utmost importance if we are to fully understand disease progression and how best to target it.

The authors would like to thank Dr Aurélie Cazet, Dr Claire Vennin, James Conway and Daniel Reed for critical reading of the manuscript and apologise to our colleagues for not being able to cite all individual references due to space limitations.