Cellular Plasticity during Metastasis: New Insights Provided by Intravital Microscopy

Abstract

Metastasis is a highly dynamic process during which cancer and microenvironmental cells undergo a cascade of events required for efficient dissemination throughout the body. During the metastatic cascade, tumor cells can change their state and behavior, a phenomenon commonly defined as cellular plasticity. To monitor cellular plasticity during metastasis, high-resolution intravital microscopy (IVM) techniques have been developed and allow us to visualize individual cells by repeated imaging in animal models. In this review, we summarize the latest technological advancements in the field of IVM and how they have been applied to monitor metastatic events. In particular, we highlight how longitudinal imaging in native tissues can provide new insights into the plastic physiological and developmental processes that are hijacked by cancer cells during metastasis.

Metastasis, the dispersion of tumor cells throughout the body, represents the largest cause of cancer-related death. Various multifaceted events occur during the escape of metastatic cells from the primary tumor and dissemination to secondary sites (Chaffer et al. 2016). A detailed understanding of these events is required for the development of efficient antimetastatic strategies (Steeg 2016). Throughout the different steps of the metastatic cascade, tumor cells can change their state, a phenomenon commonly defined as “cellular plasticity,” and this plasticity complicates the identification of therapeutic targets. Investigating these plastic events is rather limited when using conventional techniques, such as biochemical and histology assays, that draw static pictures. In contrast, development of microscopy techniques for in vivo imaging, often referred to as intravital microscopy (IVM), has provided unprecedented insights into the dynamics of cancer progression and metastatic spread (Beerling et al. 2011; Entenberg et al. 2013; Conway et al. 2014; Nobis et al. 2018). Imaging tissues in living animals was first reported in the 19th century, however due to the limited properties of the optics and the lack of contrast, IVM was mainly used to visualize tissue vasculature (Wagner 1839). In 1958, metastatic cells in the rabbit ear were imaged for the first time using an imaging chamber (Wood 1958). Since the 21st century, the development of genetic rodent models that express fluorescent proteins, and technical improvements in the field of microscopy have enabled researchers to visualize biological events deeply into live tissues, with enhanced resolution and timescale. In this review, we describe the latest technical developments for high-resolution fluorescence IVM as well as infrequently used high-end techniques that could be further used to image metastatic spread. Next, we discuss the latest biological findings provided by IVM regarding the plastic behaviors of disseminating cancer cells and stromal compartments during metastasis.

RECENTLY DEVELOPED IVM TECHNIQUES

Here, we provide an update on the most recently developed IVM technologies, for more detailed reviews on the historical development of IVM techniques and the tools now commonly used for IVM, we refer to (Beerling et al. 2011; Pittet and Weissleder 2011; Conway et al. 2014; Ellenbroek and van Rheenen 2014; Nobis et al. 2018).

Novel Technologies to Identify Cancer Cells and Their Microenvironment

The ability to simultaneously acquire signals from several channels using IVM enables the combination of multiple fluorescent probes in a single IVM experiment. While mice genetically engineered to endogenously express fluorescent labels are commonly used to study specific biological processes (Abe and Fujimori 2013; Conway et al. 2014; Ellenbroek and van Rheenen 2014; Suijkerbuijk and van Rheenen 2017; Nobis et al. 2018), they can be combined with different strategies to visualize multiple cancer compartments. For instance, infrared light is often used to excite fluorophores deep inside tissues, but this light can also produce label-free signals from structures within the tumor microenvironment (TME) (Weissleder 2001). For example, noncentrosymmetric structures such as coiled-coil and polymeric proteins can be visualized using second harmonic generation (SHG) imaging (Cicchi and Pavone 2017). Similarly, third-harmonic generation (THG) imaging is a label-free strategy that relies on the light scattering properties and mismatches of refractive indexes of water–lipid and water–protein interfaces (Yildirim et al. 2015). During SHG and THG, two or three photons are combined into one photon with a wavelength of half or a third of the excitation wavelength. SHG has enabled the visualization of extracellular matrix (ECM) remodeling during cancer progression and response to therapy (Fig. 1A; Weigelin et al. 2012; Nobis et al. 2013a; Burke and Brown 2014; Lawson et al. 2015; Vennin et al. 2017) and THG can be used to visualize elastin, erythrocytes, adipocytes, endothelial cells, and leukocytes (Fig. 1B; You et al. 2018; van Huizen et al. 2019). SHG and THG imaging were recently combined to localize implanted biomaterials in mice and to study immunological responses upon surgery (Dondossola et al. 2016). Similarly, two- and three-photon imaging can be used to monitor metabolism via for instance label-free detection of FAD and NADH (Blacker and Duchen 2016). Recently, You et al. (2018) developed a multiharmonic technique relying on a single-excitation source nonlinear imaging platform for concurrent imaging of FAD/NADH along with SHG and THG signals, to map the microenvironment of breast tumors. A recent labeling strategy relies on quantum dots (QD), which are nanoparticles with excellent optical properties for in vivo imaging (Matea et al. 2017). QD are frequently used for real-time imaging of blood vessels (Ornes 2016) and have also been used to monitor cell extravasation, angiogenesis, tumor vessel leakiness and to label rare cell populations such as hematopoietic stem cells (HSCs) (Fig. 1C; Nobis et al. 2013a; Gallego-Ortega et al. 2015; Bruns et al. 2017; Vennin et al. 2017; Wang et al. 2018). In addition, QD can be conjugated with antibodies to visualize specific cell subtypes and to quantify the amount of targeted cell population using in vivo cytometry (Lidke et al. 2007). Similarly, injecting fluorescently labeled dextran or antibodies can mark tumor-associated vasculature or specific cell populations such as immune cells (Fig. 1D; Progatzky et al. 2013; Freise and Wu 2015; Thanabalasuriar et al. 2016). Last, near-infrared (NIR) fluorescent probes have increased the possibilities for multichannel imaging (Shcherbakova et al. 2016). Because of the low absorption of NIR emission light by tissues and due to the low autofluorescence of NIR wavelengths, deeper structures can be imaged. However, the low sensitivity of most detectors at infrared light often lowers the signal-to-background ratio (Luo et al. 2011). NIR probes were recently used for IVM imaging of senescent cells, protein–protein interactions, cell cycle status, RNA detection, signaling cascades, and lipoprotein turnover (Shcherbakova et al. 2016; Bruns et al. 2017). Interestingly, NIR probes such as indocyanine green are used in patients to visualize blood vessels (Carr et al. 2018). The combination of this large range of probes for IVM represents powerful avenues to simultaneously study multiple metastatic events.

Figure 1.

Novel developments to label cancer cells and their microenvironment. (A) Live SHG imaging (purple) of the ECM in a pancreatic cancer patient-derived xenograft with high ECM content. [Panel A adapted from Vennin et al. (2017) with permission from The American Association for the Advancement of Science © 2017.] (B) THG imaging of adipocytes (blue) and SHG imaging of the ECM (magenta) in the healthy mammary fat pad. (C) Quantum dot labeling of tumor-associated vasculature in breast tumors. [Panel C reprinted from Gallego-Ortega et al. (2015) under the terms of the Creative Attribution License.] (D) Visualization of natural killer cells residing in the lungs of a healthy mouse using conjugated-antibodies and injected in vivo. [Panel D based on data in Thanabalasuriar et al. (2016)].

IVM of Fluorescent Properties to Reveal Molecular Events

Fluorescent biosensors developed for IVM can report subcellular events, such as transient protein–protein interactions or protein activity (Nobis et al. 2013b, 2018; Conway et al. 2014). Genetically encoded FRET (Förster resonance energy transfer) biosensors rely on two fluorescent proteins (a donor and an acceptor), where excited donors transfer their energy to acceptors, triggering a loss of donor emission and a gain in acceptor emission. FRET depends on the proximity between the donor and the acceptor, their spectral overlap and their alignment (Rajoria et al. 2014). Intravital measurements of FRET are achieved by either ratiometric measurements of donor and acceptor signals, by measuring the sensitized emission of the acceptor, or by measuring the donor's fluorescence lifetime (FLIM) (Piston and Kremers 2007; Jalink 2013; Conway et al. 2014; Martin et al. 2018). Cell motility can be studied using for example FRET-reporters for RhoA (Nobis et al. 2017), Rac (Johnsson et al. 2014; Yukinaga et al. 2014), Src (Nobis et al. 2013a; Vennin et al. 2017), SWAP70 (Kriplani et al. 2019), Arf6 (Hall et al. 2008), or FAK (Seong et al. 2013) activity. Similarly, FRET biosensors have shed light on the dynamics of tyrosine kinase signaling transduction (Akt, Erk reporters) (Aoki et al. 2017; Conway et al. 2018; Muta et al. 2018), membrane receptor activity (PDGFR, GPCR biosensors) (Seong et al. 2017; Halls 2019), cell response to tissue tension (MMP-9, Src, FAK, or Eph4 biosensors) (Wang et al. 2005; Ouyang et al. 2010; Seong et al. 2011; Stawarski et al. 2014; Pan et al. 2019), immune cell activity (Fyn, ZAP-70 biosensors) (Li et al. 2016; Ouyang et al. 2019), and cancer cell metabolism (ATP, cAMP biosensors) (Imamura et al. 2009; Klarenbeek et al. 2015; Annamdevula et al. 2018; Husada et al. 2018). Last, monitoring cell cycle progression and response to treatments can be assessed using CDK1 (Gavet and Pines 2010a,b; Vennin et al. 2017), PLK1 (Gheghiani et al. 2017) or caspase-3 FRET reporters (van der Wal et al. 2001; Janssen et al. 2013). A caveat of FRET sensors is their rather low dynamic range, with a typical 5%–10% change in FRET upon modification of molecule activity or interactions in vitro (Lam et al. 2001). This limitation is further increased in vivo, where the signal-to-noise level is low. In addition, emission at shorter wavelengths suffers from more scattering when imaging in deep tissues. This leads to differences in detection efficiency of donor and acceptor signals and impairs FRET measurements that depend on ratiometric and sensitized emission (Jares-Erijman and Jovin 2003). In addition, FLIM-FRET imaging requires long acquisition times of the signal, and physiological motions in vivo can impair the measurements. This can be resolved by software developed for the correction of sample motion blurring, which offer new opportunities for FLIM-FRET quantification of in vivo processes (Lowe 2004; Kaifosh et al. 2014; Warren et al. 2018).

Fluorescence recovery after photobleaching (FRAP) also allows to study molecular processes. Here, a fluorescently labeled molecule is photobleached in a targeted subcellular region and fluorescence recovery in this region is recorded by time-lapse microscopy. The speed and extent of fluorescence recovery is used as a readout of the mobility and turnover of the bleached molecule (Fritzsche and Charras 2015). FRAP has increased our understanding of processes driving the mobility of membrane proteins, protein–protein interactions, cell–cell junction turnover, cell motility and invasion, signal transduction, DNA damage, force transmission, and ECM features (Dunham 2004; Goehring et al. 2010; Erami et al. 2016; Khait et al. 2016; Kumar et al. 2016; Steurer et al. 2019). FRAP can be complemented with fluorescence loss in photobleaching (FLIP), where loss of fluorescence in areas adjacent to the bleached region is measured, and thereby strengthens the observed changes in molecule mobility obtained by FRAP (Erami et al. 2016). Moreover, optogenetic probes represent powerful tools for manipulating and monitoring cell behavior with enhanced spatiotemporal resolution. Optogenetic probes are light-responsive proteins that can be used to selectively manipulate molecular processes (Shemesh et al. 2017). Such sensors were initially engineered to probe neural activity, however recently optogenetic tools were used to dissect molecular events occurring during cell motility, epithelial-to-mesenchymal transition (EMT), transcription, mechanotransduction, and immune modulation (Tan et al. 2017; Valon et al. 2017; Fielmich et al. 2018; O'Neill et al. 2018; Qin et al. 2018; Zago et al. 2018; Zhou et al. 2018).

Considering the intricate interdependence of signaling pathways, combined imaging and analysis of multiple fluorescent biosensors in the same cell could in the future increase our understanding of the complexity of biological processes driving metastasis.

Recently Developed Optical Windows

Most sites where solid tumors metastasize to, such as the lungs and the liver, have mainly been imaged upon surgical exposure, which limits the visualization of the metastatic process to up to 72 h (Ewald et al. 2011; Entenberg et al. 2015). To overcome this, optical imaging windows allow the longitudinal monitoring of tumor progression in native tissues (Alieva et al. 2014; Conway et al. 2014; Ellenbroek and van Rheenen 2014). Historically, the first windows were engineered to image tumors transplanted in the brain or skin (Brown et al. 2003; Kienast et al. 2010). More recently, the development of lung and abdominal windows and technologies for IVM in the bone and bone marrow allow us to study metastatic spread in secondary sites. For instance, we previously developed an abdominal imaging window (AIW) (Ritsma et al. 2012), consisting of a titanium ring with a coverslip on top, which can be tightly implanted to the skin and abdominal wall using a purse-string suture. The AIW can be maintained in mice for up to 9 wk without significant alterations of animal behavior. We used the AIW to track metastatic cells and revealed that in the early stages of liver colonization, migration is required for efficient establishment of metastasis (Supplemental Movie S1; Ritsma et al. 2012). Similarly, Entenberg et al. developed a new window for lung imaging, a metastatic site for which the success of treatments remains moderate. The authors engineered a steel, passivated ring which can be inserted into mice between the ribs (Entenberg et al. 2018). While previous approaches for lung IVM required mechanical ventilation of the animal and were therefore terminal procedures (Looney et al. 2011; Entenberg et al. 2015), this novel lung window was used daily for 2 wk and was combined with microcartography to study multiple steps of lung metastasis (Entenberg et al. 2018). Other common metastatic sites are the bone and bone marrow, which are optically dense and highly scattering tissues, and therefore also present difficulties for IVM. While bone metastasis is a critical burden for patients with advanced cancer (Croucher et al. 2016), our knowledge of this disease is derived mostly from static approaches. Repeated surgical exposure of bones can be used to longitudinally study metastasis. For instance, monitoring dormant cancer cells in the endosteal niche was achieved via surgical exposure of the same bone area and using fiducial marks to localize single cells over time (Lawson et al. 2015). Implantation of miniaturized tissue-engineered bone constructs under a skin-fold window was also used to image bone-like tissues (Dondossola et al. 2018; Khosravi et al. 2018; Stiers et al. 2018). For instance, Dondossola et al. (2018), engineered an ossicle that recapitulates important mechanical and biological features of the native bone, while also displaying high optical accessibility. Combining a skin window with this implanted engineered bone constructs allowed visualization of tumor-stroma crosstalk during metastatic spread, as well as the remodeling of the bone niche during therapy (Dondossola et al. 2018). Similarly, endothelial cells in bone marrow vessels can be specifically visualized in GFP-Flk1 mice, and this enabled the direct in vivo monitoring of HSCs homing in the bone marrow, as well as the measurement of blood cell flow dynamics (Bixel et al. 2017). Last, intracranial windows have been optimized to image the dynamics of brain colonization (Kienast et al. 2010; Goldey et al. 2014; Alieva et al. 2017, 2019). Following craniotomy of the parietal bone and removal of the meninges to expose the brain cortex, a glass window is glued along with an adaptor ring for fixation onto the microscope stage. Such strategies have enabled long-term monitoring of the different steps of metastatic spread, cell invasion and changes of the TME in the brain (Kienast et al. 2010; Erapaneedi et al. 2016; Alieva et al. 2017, 2019).

Combining IVM with Multiple Techniques to Enhance Our Understanding of Metastasis

While each microscopy technique comes with its benefits and caveats, combining IVM with complimentary approaches can enhance our understanding of the complexity of metastasis. For instance, we previously developed cryosection labeling and intravital imaging (CLIM) to integrate IVM with histochemistry analyses (Ritsma et al. 2013). Here, an area imaged by IVM can be “tattooed” using light to subsequently relocate this specific area in histological sections (Ritsma et al. 2013). We used CLIM to interrogate how microenvironmental events influence tumor cell invasion, and we found that interactions with T cells lead to enhance cancer cell motility (Ritsma et al. 2013). Similarly, IVM can be combined with superresolution imaging. For instance, correlative light and electron microscopy (CLEM) combines IVM data with electron microscopy (EM) to obtain a complete view at the nanoscale level of tissues (Karreman et al. 2016a; Hyenne et al. 2019). The main bottleneck of CLEM is the difficulty to precisely locate by EM the region imaged previously by IVM, however this can be facilitated using fiducial marks such as blood vessels, nuclei, or myelinated axons (Karreman et al. 2016a; Goetz 2018; Luckner et al. 2018). CLEM has provided new high-resolution information into the subcellular events driving tumor cell invasion, extravasation of circulating tumor cells (CTCs) and uptake of cancer cell-derived extracellular vesicles (EVs) (Karreman et al. 2014, 2016b; Follain et al. 2018; Hyenne et al. 2019). Lastly, single-cell sequencing has also been used to characterize cells identified using IVM (Beerling et al. 2016; Scheele et al. 2017). This enables researchers to link dynamic behaviors visualized via IVM with molecular signatures obtained using single-cell sequencing.

Imaging Plasticity of Cell Lineages

Progression through the metastatic cascade can occur over periods of weeks, months, or even years. During this period, tumor cells can change their state and fate (Chaffer et al. 2016). Fluorescent-based reporters are used together with IVM to track the same cells and their progeny over time and enable linking cell status to cell fate. For instance, photoconvertible fluorophores such as Dendra (Kedrin et al. 2008; Fumagalli et al. 2017), PSmOrange (Subach et al. 2011), EosFP (McKinney et al. 2009), or Kaede (Ando et al. 2002; Torcellan et al. 2017) can change color upon laser exposure and are used to track cells in time and space and during the lifetime of the photoconverted fluorophore.

While photoconvertible probes allow transient marking of targeted cells, lineage tracing tools label cells with an inheritable mark. A common lineage tracing strategy relies on the expression of unbiased or lineage-specific Cre-recombinase to permanently activate a fluorescent reporter (Livet et al. 2007; Snippert et al. 2010; Scheele et al. 2017). Lineage tracing probes are used to study plasticity of tumor cell identity, hierarchy, and clonality. For example, static and IVM-based lineage tracing recently showed that the cellular hierarchy of nontransformed tissues is maintained in a number of tumors (Driessens et al. 2012; Schepers et al. 2012; Zomer et al. 2013). Using lineage tracing, these studies identified a small population of cancer stem cells (CSCs) in skin, colorectal, and breast tumors that have self-renewal capacity and give rise to more differentiated cancer cells (Driessens et al. 2012; Schepers et al. 2012; Zomer et al. 2013). Moreover, IVM of lineage tracing tools revealed that cancer stemness is a plastic state that can be lost or gained over time, rather than a fixed cell property (Zomer et al. 2013). In addition to differences in stem cell properties, cancer cells can acquire different growth potentials, and this could be due to accumulation of mutations or alterations of the microenvironment. For instance, static lineage tracing revealed that a strong selection pressure can occur during progression from benign lesions to carcinoma stages, and this results in the dominance of one single clone (Reeves et al. 2018). Similarly, IVM revealed that precursor lesions in pancreatic cancer are polyclonal, however, this diversity decreases during cancer progression while a significant fraction of metastatic lesions are polyclonal (Maddipati and Stanger 2015). Similarly in breast cancer, metastatic cells homing in distant sites were found to be polyclonal (Cheung et al. 2016). Last, IVM showed that stem cells harboring oncogenic activation of the Wnt/β-catenin pathway in mouse skin epithelium can form tumor lesions (Brown et al. 2017). This study further showed that wild-type cells can actively eliminate transformed stem cells. This leads to tumor regression and even complete disappearance of the tumor, thereby reestablishing normal homeostasis (Brown et al. 2017). Lineage tracing can also be used to label cells that take up material released by other cells. For example, packaging of Cre-recombinase into EVs released by donor cells enables the specific labeling of recipient cells expressing a Cre-dependent reporter (Zomer et al. 2015; Steenbeek et al. 2018). This method of cell labeling, combined with IVM, revealed for instance that aggressive breast tumor cells can transmit their invasive behavior to recipient cells through the exchange of biomolecules carried by EVs (Supplemental Movie S2; Ridder et al. 2014, 2015; Zomer et al. 2015; Steenbeek et al. 2018). Recently, new recombinase-based systems were engineered to label cells in a more controllable manner. For instance, a photoactivatable Cre-recombinase (PA-Cre) was developed to target cells with enhanced spatiotemporal precision (Schindler et al. 2015). The PA-Cre can be activated in native tissues using fiber optic illumination, enabling the modification of gene expression while also permanently labeling native cells (Schindler et al. 2015). In addition, dual-recombinase-activated lineage tracing (DeaLT) was recently developed to trace multiple lineages while also reducing off-target labeling that can occur using the Cre-mediated lineage tracing (He et al. 2018). DeaLT integrates the Cre-LoxP system with a second system, the Dre-rox system, for DNA targeting with enhanced specificity while also enabling the tracing of multiple lineages (He et al. 2018). DeaLT was recently used to specifically label and trace different lineages of cardiac stem cells expressing c-Kit (He et al. 2018). Similarly, a split Cre system, which relies on concurrent expression of two markers for activation of functional Cre was recently developed (Hirrlinger et al. 2009; Buczacki et al. 2013). This system was used to identify quiescent cells in the intestine that are precursors of mature and differentiated secretory cells (Buczacki et al. 2013). In the future, combining these novel Cre-based reporters with IVM may further enhance our understanding of plastic switches undergone by cancer cells during metastasis.

LATEST BIOLOGICAL INSIGHTS INTO CELL PLASTICITY DURING METASTASIS UNCOVERED BY IVM

In the next section, we discuss the latest biological insights into plastic metastatic events as assessed by IVM.

Monitoring Cancer Plasticity during Invasion

In order to be able to leave the primary tumor, cancer cells are thought to adopt distinct modes of invasion (Friedl and Alexander 2011), which have been studied extensively using static approaches. Recently, IVM has facilitated the study of these heterogeneous migratory behaviors. As such, IVM identified single cells escaping from the primary tumor (Supplemental Movie S3; Beerling et al. 2016), as well as multicellular streaming entities, and cells switching from cohesive cell migration to single-cell migration (Giampieri et al. 2009). Interestingly, IVM demonstrated that the switch from collective to single-cell invasion can be regulated by TGFβ1 signaling (Giampieri et al. 2009). Moreover, in vivo measurements of FRET- and FRAP-reporters of cell invasion in primary pancreatic tumors and during the early steps of liver colonization (Johnsson et al. 2014; Erami et al. 2016; Warren et al. 2018) have provided insights into how invasive cancer cells adapt their invasive behavior upon genetic modification or during treatment with anti-invasive drugs (Johnsson et al. 2014; Erami et al. 2016; Warren et al. 2018).

It has been extensively debated whether disseminating epithelial cancer cells acquire migratory and stem cells properties by hijacking the developmental process EMT (Del Pozo Martin et al. 2015; Fischer et al. 2015; Zheng et al. 2015; Zhao et al. 2016; Brabletz et al. 2018). For example, the lack of evidence for EMT in histological sections of human primary tumors that have metastasized, as well as in metastases, is often raised to question the very existence of EMT during metastasis (Bukholm et al. 2000; Kowalski et al. 2003; Zheng et al. 2015). Recently, IVM-studies have shed light on the plastic nature of epithelial and mesenchymal states by demonstrating the ability of cancer cells to switch between the two states. For instance, we used IVM to monitor cancer cells that change color upon EMT. We showed that migratory and disseminating cells undergo EMT but revert to an epithelial state at the metastatic site, a process referred to as mesenchymal-to-epithelial transition (MET) (Fig. 2A; Beerling et al. 2016). Importantly, MET is believed to be necessary for efficient metastatic colonization (Kalluri and Weinberg 2009). Similarly, IVM identified a population of cancer cells in the mammary gland that undergo EMT, are highly motile, commonly reside close to blood vessels (Zhao et al. 2016) and which can revert back to an epithelial state when colonizing secondary sites (Beerling et al. 2016; Harper et al. 2016). Moreover, IVM in melanoma revealed that invasive cells that intravasate into the tumor vasculature at the primary site are poorly differentiated (Pinner et al. 2009). However, at secondary sites, this undifferentiated, highly motile phenotype is not maintained, suggesting that metastatic cells can switch from one state to another (Pinner et al. 2009).

Figure 2.

Imaging plasticity during the metastatic cascade. (A) IVM of migrating tumor cells in primary and secondary sites. Time-lapse microscopy of migrating breast tumor cells (yellow and blue for cancer cells expressing E-cadherin) leaving the primary tumor. [Panel A is reprinted from Beerling et al. (2016) under the terms of the Creative Commons Attribution-NonCommercial-No Derivatives License.] Scale bar, 50 µm. (B) IVM of tumor cells homing to and colonizing secondary sites. (i) IVM through a cranial window of extravasation of cancer cells and formation of micrometastases (green) in close proximity to blood vessels (red). [Panel B(i) is reprinted from Follain et al. (2018) with permission from Elsevier © 2018.] Scale bar, 100 µm. (ii) Longitudinal IVM tracking through a cranial window of metastatic outgrowth of lung cancer cells (red) in the brain adjacent to blood vessels (green). [Panel B(ii) is reprinted from Kienast et al. (2010) with permission from Springer Nature © 2009.] Scale bar, left 100 µm, right 20 µm. (C) IVM microscopy of dormant leukemic cells in the bone marrow and attached to osteopontin (red), labeled with DiR (green), which is retained in noncycling cells. [Panel C reprinted from Boyerinas et al. (2013) with permission from the American Society of Hematology © 2013.] Estimated scale bar, 20 µm. (D) IVM monitoring of the plastic interactions between cancer cells and the immune system. (i) IVM of patrolling monocytes (green) taking up material derived from lung CTCs (red) in the lungs. [Panel D(i) reprinted from Hanna et al. (2015) with permission from The American Association for the Advancement of Science © 2015.] Scale bar, 100 µm. (ii) Tissue resident T cells (green) in primary tumors interacting with melanoma cells (red). [Panel D(ii) reprinted from Park et al. (2019) with permission from Springer Nature © 2019.] Estimated scale bar, 10 µm.

Tumor cell invasion is also dependent on microenviromental cues, such as cell density, collagen fiber crosslinking, proximity to blood vessels and interactions with immune cells (Gligorijevic et al. 2014). For instance, IVM tracking of the early steps of liver colonization by metastatic pancreatic tumor cells revealed that the activity of Src, a key regulator of cancer cell invasion, was reduced upon TME editing with the Rho-kinase inhibitor Fasudil (Warren et al. 2018). Moreover, photoswitchable proteins were used to label invasive mammary tumor cells and to dissect how the TME influences cell invasion (Kedrin et al. 2008; Gligorijevic et al. 2014). Using this approach, cancer cells in close proximity to blood vessels were found to be more migratory than tumor cells located in areas devoid of vessels (Kedrin et al. 2008; Gligorijevic et al. 2014). Following intravasation, CTCs are subjected to hemodynamic forces, and recently CLEM was used to characterize how these forces influence the arrest and extravasation of disseminated tumor cells in the brain (Fig. 2B(i); Follain et al. 2018). In addition, IVM through a cranial window enabled the monitoring of metastatic growth in proximity to blood vessels following extravasation (Fig. 2B(ii); Kienast et al. 2010). IVM also revealed that tumor-associated macrophages (TAMs) can promote tumor cell invasion and metastasis. For instance, TAMs were found to migrate together with breast cancer cells upon stimulation with either epidermal growth factor or colony-stimulating factor 1 (Wyckoff et al. 2007). Similarly, paracrine signaling between TAMs and cancer cells can lead to enhanced expression of the actin-binding protein Mena^INV in the cancer cell compartment, and this can increase invadopodia formation, cancer cell motility, and intravasation (Philippar et al. 2008; Roussos et al. 2011; Gligorijevic et al. 2014). Last, IVM monitoring of breast cancer cells targeted with chemotherapy demonstrated that macrophages with high levels of Tie2 expression (Tie2^Hi) accumulate at tumor-microenvironment of metastasis (TMEM) and can facilitate local transient permeability of the endothelium and tumor cell intravasation (Harney et al. 2015; Karagiannis et al. 2017). This suggests that inhibiting Tie2^Hi macrophages in combination with chemotherapy may be beneficial in the neoadjuvant setting (Harney et al. 2015; Karagiannis et al. 2017).

Plasticity of Dormancy and Senescence

Dormant and senescent (D/S) cancer cells are quiescent cancer cells that survive within primary and secondary tumors until cell intrinsic or microenvironmental cues reactivate their proliferative state. In addition, because they do not proliferate, dormant cells are thought to be resistant to chemotherapy and may therefore drive tumor recurrence (Aguirre-Ghiso 2006; Goss and Chambers 2010). While senescence has long been considered to be irreversible, senescent cells were recently suggested to maintain some plasticity and to regain the ability to proliferate (Milanovic et al. 2018). Fluorescent probes have been engineered to monitor the plastic behavior of D/S cells in native tissues. For instance, lipophilic fluorescent dyes are gradually diluted upon cell proliferation but are maintained by nonproliferating cells (Boyerinas et al. 2013; Lawson et al. 2015). In addition, the fluorescence ubiquitination cell cycle (FUCCI) and Ki67-reporters can be used to identify nonproliferating cells in live tissue (Newman and Zhang 2008; Sugiyama et al. 2009; Basak et al. 2014; Yano et al. 2014; Chou et al. 2018). Similarly, a p16^Ink4a trimodality reporter (p16-3MR), consisting of a fusion protein of synthetic Renilla luciferase, mRFP and Herpes simplex virus thymidine kinase allows visualization and specific killing of cells expressing p16^Ink4a, an important driver of dormancy and senescence (Demaria et al. 2014, 2017). Moreover, labeling β-galactosidase⁺ senescent cells can be achieved using beads loaded with a fluorophore and coated with galacto-liposaccharides (Muñoz-Espín et al. 2018).

IVM of D/S cancer cells has led to the identification of microenvironmental mechanisms driving the plastic switch toward proliferation. For instance, in acute lymphoblastic leukemia (ALL), osteopontin (OPN) can induce dormancy by promoting anchorage of leukemic cells in the endosteal niche (Boyerinas et al. 2013). IVM of label-retaining cells in the calvarial bone marrow demonstrated that blocking OPN triggers dormant cells to switch back to proliferation and renders them more sensitive to chemotherapy (Fig. 2C; Boyerinas et al. 2013). Similarly, repeated imaging of multiple myeloma cells in the endosteal niche identified cancer cells entering dormancy for long time-periods before resuming proliferation (Lawson et al. 2015). Moreover, distinct components of the endosteal niche were found to regulate the dormancy-to-proliferation switch, where interactions with osteoblasts maintained dormancy, while the osteoclast-driven remodeling of the niche reactivated cancer cell proliferation (Lawson et al. 2015). Last, longitudinal tracking of metastatic melanoma and lung cancer cells through an intracranial window revealed that dormant cells reside in close proximity to the perivascular niche and that blocking angiogenesis can induce dormancy (Kienast et al. 2010). Surprisingly, dormant metastatic melanoma cells were also found to be highly migratory in the brain, while metastatic lung cancer cells were mostly static at this secondary site, suggesting that this phenotype is not driven by the brain microenvironment (Kienast et al. 2010).

Recently, non-IVM based studies have demonstrated that cancer cell dormancy can also be regulated by local inflammation and autophagy (Albrengues et al. 2018; Vera-Ramirez et al. 2018), and that tumor cells that exit senescence are reprogramed into a stem-like state to fuel cancer growth (Milanovic et al. 2018). In addition, dormant cells that are mesenchymal can undergo MET during the switch toward proliferation and metastatic growth (Yang et al. 2018). Future IVM assessments of dormant cell plasticity may enhance our understanding of these newly discovered phenomena and can help identify new targets to tackle therapy resistance and tumor recurrence.

Plasticity of the Immune Landscape

While the identity and fate of cancer cells continuously evolve during metastatic spread, the TME is also plastic and can be coopted by cancer cells to support their dissemination (Vennin et al. 2016). For instance, IVM of immune cells during cancer progression and metastatic spread has provided novel insights into the plasticity of both antitumor and protumor behavior of immune cells in response to cancer-related stimuli (Vennin et al. 2016; Suijkerbuijk and van Rheenen 2017; Torcellan et al. 2017; Nobis et al. 2018; Wellenstein and de Visser 2018). For instance, in melanoma, IVM recently identified a subset of dendritic cells (DCs) that present antigens to CD4⁺ T cells but fail to promote their differentiation toward an antitumor phenotype (Binnewies et al. 2019). Interestingly, blocking regulatory T cells (Tregs) removed the brake on DCs and reestablished CD4⁺ T cell antitumor functions (Binnewies et al. 2019). Similarly, DCs were found to cluster with natural killer cells, and this correlated with enhanced responses to immunotherapy (Barry et al. 2018). In addition, monocytes are classically considered to promote metastatic spread, however recently patrolling monocytes were shown to reduce metastatic colonization of the lungs (Hanna et al. 2015). Here, IVM of Nra1-GFP mice to specifically monitor patrolling monocytes revealed that upon intravenous injection of lung, melanoma, or breast cancer cells, patrolling monocytes are recruited to the lungs and prevent tumor cell invasion (Fig. 2D(i); Hanna et al. 2015). Similarly, live imaging in primary breast tumors demonstrated that monocytes that enter tumor tissues differentiate into highly migratory TAMs, which guide cancer cells toward blood vessels (Arwert et al. 2018). These migratory TAMs can undergo a unidirectional switch toward a metastatic-assisting perivascular phenotype (Arwert et al. 2018). In addition, a population of tumor-experienced T cells were recently identified using photolabeling strategies. These T cells can readily egress from the primary tumor and migrate to metastatic sites, where they may mediate tumor-immunity (Torcellan et al. 2017). The interactions between cancer and immune cells can vary over time and space. For instance, in clinically dormant melanoma lesions, resident memory T cells accumulate in stable and persistent melanoma lesions, and can either retain prolonged contact with cancer cells or display a migratory phenotype (Fig. 2D(ii); Park et al. 2019). Similarly, interactions between CTCs and immune cells residing in secondary sites were shown to influence the fate of disseminated cancer cells (Headley et al. 2016). Here, using fluorescent lineage reporters of myeloid cell populations, Headley et al. (2016) visualized waves of myeloid cell populations that take up microparticles released by CTCs and that subsequently accumulate in the lungs to promote metastatic growth. Successful establishment of metastasis can also be driven by interactions of cancer cells with neutrophils. For instance, CTCs that bind to neutrophil extracellular traps (NET) or that extravasated in close proximity to neutrophils in the liver have a higher ability to seed in this tissue (Spicer et al. 2012; Cools-Lartigue et al. 2013).

Plasticity of the ECM and the Vasculature

Remodeling of the ECM provides cancer cells with biochemical and mechanical cues that regulate their growth, invasion, and response to therapies (Pickup et al. 2014). For instance, increased ECM deposition is thought to impair drug delivery (Pickup et al. 2014; Vennin et al. 2018). IVM FRAP combined with histology and transmission EM demonstrated that the diffusion coefficient of macromolecules in patient-derived cranial and skin tumors is reduced in areas with a dense ECM (Netti et al. 2000). In line with this, IVM of mouse tumors and patient-derived xenografts revealed that reducing ECM crosslinking can enhance chemotherapy efficacy (Vennin et al. 2017). Cancer cell invasion and liver metastases can also be reduced upon ECM manipulation, suggesting that reestablishing normal ECM homeostasis could be used as an antimetastatic strategy (Vennin et al. 2017). Cancer-associated fibroblasts (CAFs) are thought to be the main drivers of ECM remodeling, and IVM demonstrated that CAFs can also influence cancer cell metastasis. For instance, CAFs and cancer cells can establish physical interactions that are mechanically active and promote collective migration of cancer cells (Labernadie et al. 2017). Similarly, CAFs can travel along with cancer cells from a primary tumor to secondary sites, and depletion of CAFs was shown to reduce metastatic burden and extended mouse survival (Duda et al. 2010).

Tumor cells can also coopt the vasculature. IVM has shown that vasculature remodeling occurs during the early stages of malignant transformation and that angiogenesis appears when tumors progress from hyperplasia to carcinoma (Wyckoff et al. 2007; Kedrin et al. 2008; Vennin et al. 2016). This results in a leaky and dysfunctional blood vessel network (Hagendoorn et al. 2006; Vennin et al. 2017), which provides a route for metastatic cells toward secondary sites (Kedrin et al. 2008). In addition, IVM of breast tumors revealed that cancer cells migrate toward macrophages in the perivascular niche (PVN), and this correlated with the intravasation of tumor cells (Wyckoff et al. 2007). In line with this, IVM through a cranial window demonstrated that dormant metastatic cells reside in close proximity to the PVN (Kienast et al. 2010) and that the switch from dormancy to growth can either be driven by angiogenesis or by vessel cooption (Kienast et al. 2010). Because of the dysfunctional vasculature, oxygen is not evenly distributed within the tumor and this can lead to the establishment of hypoxia (Wilson and Hay 2011). IVM mapping of oxygen levels in pancreatic tumors recently showed that hypoxic areas move within tumor tissues over time (Conway et al. 2018). In addition, an Akt FRET-reporter was used to reveal that hypoxia induces resistance to the mTORC1/2 inhibitor AZD2014. This resistance was overcome by combining AZD2014 with cytotoxic-prodrugs specifically activated in hypoxia (Conway et al. 2018). Last, an intraoperative epifluorescent platform was recently used to image the tumor vasculature of melanoma patients during surgical removal of the tumor. This suggests that in the future, IVM in a human setting could help monitoring drug delivery and could facilitate the development of new treatments (Fisher et al. 2016).

Plastic Responses to Therapies

IVM has shed new light on cell intrinsic and microenvironmental mechanisms of resistance to therapy. For instance, in glioma, where chemotherapy efficacy is commonly low, IVM identified tumor microtubes, which are long-lived, ultralong and thick membrane extensions that interconnect tumor cells via gap junctions (Osswald et al. 2015). The density of these tumor microtubes increased upon chemotherapy and prevent cancer cell apoptosis (Fig. 3A; Osswald et al. 2015; Weil et al. 2017). Moreover, in T cell acute lymphoblastic leukemia (T-ALL), cells that survive treatments adopt a more motile phenotype and colonize secondary sites (Fig. 3B), in line with recent findings suggesting that cytotoxic treatments may induce metastatic spread (Hawkins et al. 2016; Karagiannis et al. 2017).

Figure 3.

Plastic responses to therapies. Cancer cells display plastic responses to therapies, which can be cell intrinsic or a result of microenvironmental protection. Here, we depict some examples of both types of mechanisms of resistance: (A) A network of tumor microtubes protects glioma cells from dying. After chemotherapy, the number of connections increases and this correlates with reduced response to treatment. White arrows point to microtubes. [Panel A is reprinted from Osswald et al. (2015) with permission from Springer Nature © 2015.] (B) T-ALL cells become more migratory and colonize the bone marrow after chemotherapy. [Panel B reprinted from Hawkins et al. (2016) et al. with permission from Springer Nature © 2016.] (C) CAF activation upon treatments with BRAF inhibitor leads to ECM remodeling and reduced tumor cell response to treatment. [Panel C reprinted from Hirata et al. (2015) under the terms of the Creative Commons Attribution-NonCommercial-No Derivatives License (CC BY NC ND).] (D) Peritumoral Tregs prevent infiltration of adoptive cytotoxic T cells (CTLs) into melanoma lesions, reducing immunotherapy efficacy. [Panel D reprinted from Qi et al. (2016) under the terms of the Creative Commons Attribution License (CC-BY).] (E) Biopsy induces inflammation and tumor cell migration, demonstrated by the change in the red area postbiopsy. [Panel E reprinted from Alieva et al. (2017) under the terms of the Creative Commons Attribution License (CC BY).]

Cancer cells can also modify their niche upon treatment and alterations of the TME can influence response to therapy (Binnewies et al. 2018). For instance, in T-ALL, treatment can increase the ability of tumor cells to trigger apoptosis of osteoblasts, thereby misbalancing normal hematopoiesis and conferring an advantage for T-ALL cells to repopulate the bone marrow (Hawkins et al. 2016). Similarly, in melanoma, tumor cells subjected to MEK and BRAF inhibitors can activate neighboring CAFs, leading to ECM remodeling and reduced cancer cell response to treatment (Fig. 3C; Hirata et al. 2015). While therapy resistance can be driven by ECM stiffness (Hirata et al. 2015), long-term treatment with MEK and BRAF inhibitors can also push cancer cells toward an epithelial phenotype, and this correlated with increased resistance to treatment (Brighton et al. 2018). In addition, IVM was used to show that chemotherapy in acute myeloid leukemia (AML) induces apoptosis of endothelial cells, which subsequently leads to the establishment of hypoxia and reduces drug delivery to the tumor cells (Passaro et al. 2017). Similarly, IVM revealed that AML cells can disrupt the bone marrow vasculature after treatment with cytarabine and doxorubicin, leading to transendothelial migration of HSCs, loss of osteoblasts and increased survival of leukemic cells (Passaro et al. 2017). Restoring normal tumor vasculature can improve treatment efficacy (Vennin et al. 2018). For instance, encaging chemotherapies into transferrin-functionalized nanoparticles that specifically target tumor-associated blood vessels can overcome treatment resistance in glioma (Lam et al. 2018). Immune cells can also undergo dynamic changes upon treatments, and this in turn influences cancer cell responses to therapies. For example, IVM demonstrated that anti-VEGF therapy switched the TME toward an immunosuppressive state, with enhanced extravasation of neutrophils into colorectal tumors and increased the production of immunosuppressive IL-10 (Jung et al. 2017). In addition, in melanoma, peritumoral Tregs can prevent tumor infiltration of adoptive CTLs and thereby reduce immunotherapy efficacy (Fig. 3D; Qi et al. 2016). This was reverted by administering cyclophosphamide, which eliminated Tregs and increased migration of adoptive CTLs into tumors (Qi et al. 2016). In addition, IVM revealed that CCR2⁺ macrophages are recruited to breast tumors upon chemotherapy, and this correlates with tumor relapse (Lam et al. 2018). Similarly, we recently used IVM to demonstrate that biopsies, which are routinely performed in the clinic, can trigger the recruitment of macrophages to the biopsy site and subsequently induce cancer cell invasion and proliferation (Fig. 3E; Supplemental Movie S4; Alieva et al. 2017). Importantly, blocking the recruitment of macrophages prevents this plastic switch and reduces cancer cell invasion (Alieva et al. 2017). IVM can also facilitate the development of novel immune-based therapies. For instance, fluorescently labeled bisphosphonates, which are thought to solely target bone cells, were shown to accumulate in TAMs infiltrating breast tumors, suggesting that bisphosphonates may be used in treatment of breast cancer patients (Junankar et al. 2015). Similarly, Yu et al. (2019) developed α-melittin nanoparticles, which are specifically taken up by liver sinusoidal endothelial cells, triggering a switch in the cytokine milieu to recruit innate and adaptive immune cells to the liver, thereby reducing metastatic. Lastly, interactions between a subset of DCs and T cells were recently shown to be required for efficient anti-PD1 treatment in colorectal tumors (Garris et al. 2018). Here, IVM of cytokine production demonstrated that DCs sense IFNγ produced by T cells during immunotherapy, and in turn release IL-2 to further stimulate antitumor immunity (Garris et al. 2018).

CONCLUDING REMARKS

IVM comprises a range of powerful tools, which over the last decade have shed light on the dynamics of metastasis. In this review, we provide a glance of how IVM enables the monitoring of plastic events followed by cancer cells during metastasis. Discoveries uncovered using IVM may in the future facilitate the development of novel antimetastatic therapeutics. In addition, in order to expand our knowledge of metastasis, IVM should be combined with high-throughput approaches for profiling of cancer cells as well as with new cancer models such as patient-derived models and humanized mouse models.