Advances in fluorescence microscopy have enabled the study of membrane diffusion, cell adhesion and signal transduction at the molecular level in living cells grown in culture. By contrast, imaging in living organisms has primarily been restricted to the localization and dynamics of cells in tissues. Now, imaging of molecular dynamics is on the cusp of progressing from cell culture to living tissue. This transition has been driven by the understanding that the microenvironment critically determines many developmental and pathological processes. Here, we review recent progress in fluorescent protein imaging in vivo by drawing primarily on cancer-related studies in mice. We emphasize the need for techniques that can be easily combined with genetic models and complement fluorescent protein imaging by providing contextual information about the cellular environment. In this Commentary we will consider differences between in vitro and in vivo experimental design and argue for an approach to in vivo imaging that is built upon the use of intermediate systems, such as 3-D and explant culture models, which offer flexibility and control that is not always available in vivo. Collectively, these methods present a paradigm shift towards the molecular-level investigation of disease and therapy in animal models of disease.
Advances in optical microscopy have produced a variety of methods suitable for the study of protein dynamics in cultured cells. This has been accelerated by the development of fluorescent proteins (FPs), which enable not only selective labeling and localization of virtually any protein, but also allow engineering of functional probes that report on the activity of signaling molecules such as kinases (Wang, Y. et al., 2005) and GTPases (Yoshizaki et al., 2003). FPs are increasingly incorporated into sophisticated animal models of disease, in which tissue-specific driver mutations can be linked to the expression of FP probes. As a result, new animal models that are appropriate for the investigation of molecular dynamics in vivo, such as genetically engineered mice (GEM), have been developed.
The need for in vivo imaging has emerged in conjunction with the understanding that many cellular responses are determined by signals within a tissue niche (Coussens and Werb, 2002; Sneddon and Werb, 2007). For example, local mediators of cellular communication, such as tumor necrosis factor α (TNFα), inflammatory cytokines, and chemokines can crucially influence the progression and metastasis of cancer (reviewed in Wilson and Balkwill, 2002). The ability of stem cells to retain their special properties also appears to be regulated by features of the stem cell niche (Zhang et al., 2003). Furthermore, imaging of protein dynamics in response to therapeutic intervention has the potential to dramatically increase the use of mouse cancer models and our understanding of the molecular mechanisms of drug action (Kamb, 2005). By combining advanced imaging techniques with sophisticated new animal models, we are now able to investigate both disease and treatment at the molecular level.
This Commentary will review approaches for imaging molecular dynamics in vivo, including both imaging techniques and experimental systems, with examples drawn primarily from the use of mouse cancer models. We begin the discussion with general approaches to FP imaging and their use to study cellular dynamics in vivo, and progress to more specialized methods for molecular level investigations. Then we discuss methods that complement the use of FPs by providing contextual information that is important for the interpretation of results. Finally, we highlight the factors that constrain in vivo experimental design and the use of intermediate experimental systems, to bridge the gap between glass coverslips and native tissue.
Imaging cellular dynamics with fluorescent proteins
Use of FPs has expanded into developmental models, such as zebrafish and Drosophila melanogaster, following the FP revolution of cell biology in the late 1990s. The application of FPs in vivo has been limited both by the development of appropriate imaging approaches and the understanding of what kinds of questions can be answered using imaging approaches. Therefore, in this section, we begin with an introduction to general approaches for imaging the dynamics of cells in tissues (summarized in Table 1).
Whole-body fluorescence imaging
Whole-body imaging is useful for determining the location and size of tumors in living mice (Fig. 1). This approach is generally non-invasive and can be repeated many times on the same animal, which facilitates longitudinal studies of tumor growth or senescence, metastatic progression and response to therapy (Ahmad et al., 2011; Morton et al., 2010). Basic systems provide qualitative information about the size and location of fluorescent tumors on the basis of epi-fluorescent illumination, a macroscopic lens and sensitive detection using cooled CCD cameras. Some of these systems have sufficient resolution to image single cells, and have been used to image events such as cancer cell extravasation (Yamauchi et al., 2006). Alternatively, advanced tomographic systems generate quantitative fluorescence images by using transmitted laser light to correct for sample absorption and scattering (reviewed in Ntziachristos et al., 2005).
Confocal laser scanning microscopy
Confocal laser scanning microscopy (CLSM) is the technique of choice for many in vitro cell-imaging applications [for a thorough discussion refer to the Handbook of Confocal Microscopy (Pawley, 2006)]. Optical sectioning (Box 1) in CLSM is based on the selective detection of fluorescence from a single point within the specimen by using a spatial filter (pinhole) placed in front of a detector. The main advantages of CLSM compared with multiphoton microscopy are its relative ease of use and its widespread availability, and its low cost and high safety (Table 1). One disadvantage of CLSM is that the illumination method generally leads to photobleaching and phototoxicity throughout the focal path of the laser, including regions above and below the image plane. Such photobleaching can be minimized by reducing the intensity of the laser beam when particularly bright or dim regions of the sample are encountered (Hoebe et al., 2007). A second disadvantage is that confocal detection suffers greatly if light emitted from the focal point in the sample is scattered on its way to the detector. Light scattering is one of the main limiting factors in tissue imaging (Helmchen and Denk, 2005). Scattering primarily occurs as a result of differences in refractive index associated with compartmental boundaries within cells and tissues, such as nuclei, blood vessels and extracellular matrix (ECM) fibers (Fig. 1B). As a photon transits such a boundary, its direction is altered so that it cannot be properly refocused through the detection pinhole and its signal will be lost. Measurements have shown that, on average, a photon will be scattered once for every 47 μm traveled through adult rat brain tissue (Oheim et al., 2001), which suggests an effective imaging depth of <50 μm for CLSM.
CLSM has been used to study the shedding of cells from the surface of the intestinal epithelium (Watson et al., 2005) and also to quantify the frequency and type of movement in tumors that are formed by cells deficient for pyruvate dehydrogenase kinase isoenzyme 1 (Pinner and Sahai, 2008). However, CLSM is more commonly used for thinner translucent specimens, including intermediate systems, such as cell derived matrices (Caswell et al., 2007), tissue explants (Mort et al., 2010) and organoids (Pearson and Hunter, 2007) (refer to Fig. 1 and Box 2 for more detail). For example, Gaggioli et al. recently used CSLM to demonstrate the collective invasion of carcinoma cells into an organotypic matrix (Gaggioli et al., 2007). Contextual information was provided by imaging collagen fibers using confocal reflection microscopy (a detection mode in which reflected excitation light rather than fluorescence is imaged by the detector). Collagen is a main constituent of most tissues, and a common, biologically relevant (compared with glass coverslips) substrate for cell migration (Fig. 1B). Confocal reflection microscopy has also been used to visualize reorganization of collagen matrices, such as local contraction or hydrolysis, during invasion (Wolf et al., 2003).
Intermediate systems such as cell-derived matrices and organotypic gels are amenable to CLSM because of their superior optical properties (less absorption and scattering of light) compared with whole tissues. Kubow and Horwitz used translucent collagen gels and low-light confocal microscopy to visualize formation of faint green fluorescent protein (GFP)-labelled paxillin adhesions (imaged in fluorescence) juxtaposed with collagen fibers (imaged in reflectance) in U2OS cells and, thereby, confirmed the existence of focal-adhesion-like structures in 3D culture (Kubow and Horwitz, 2011).
Spinning disk confocal microscopy (SDCM) has emerged in recent years as a powerful technique for live imaging of ‘semi-thick’ specimens such as Caenorhabditis elegans (Mayer et al., 2010), zebrafish (Wang et al., 2011) and Drosophila (Aldaz et al., 2010). In SDCM an array of pinholes rather than a single pinhole is scanned over the specimen and light from the pinhole array is imaged onto the surface of a CCD camera [for a detailed description see Pawley (Pawley, 2006)]. SDCM has been less widely adopted for in vivo imaging in mice because of its susceptibility to crosstalk between pinholes, which limits image contrast when imaging deep within highly scattering tissue (Egner et al., 2002; Wang, E. et al., 2005). Nevertheless, SDCM has been used to examine the differential effects of hypoxia on T-lymphocyte and myeloid cell migration in MMTV-PyMT mice – a mammary carcinoma model (Egeblad et al., 2008) – and in conjunction with laser ablation to study the signals that guide neutrophil migration into superficial sites of inflammation in the liver (McDonald et al., 2010).
Box 1. Glossary
In vitro and in vivo
Originally, these terms referred to experiments performed outside of and within the body, respectively. However use of in vivo has been stretched to cover experiments performed in cells, which leads to confusion and complicates online searches. Therefore, we suggest in muro (from the latin murinus) to distinguish work performed in living mice.
Laser scanning microscopy (LSM)
It is possible to raster-scan the point of a focused laser beam across a specimen by using pivoting mirrors and a telecentric lens system to achieve single or multiphoton excitation of fluorescence. Images can be generated by confocal or wide-field detection of fluorescence emission.
Fluorescence involves absorption of a photon by a fluorophore, which raises the energy state of an electron within the molecule to an excited state, followed by relaxation of the electron down to ground state and loss of energy in the form of an emitted photon after a brief period of time. The average time between excitation and emission is known as the fluorescence lifetime, which lasts between 2 and 5 nseconds for many common fluorophores.
Processes such as multiphoton fluorescence and second harmonic generation (SHG) are non-linear because they require the interaction of two or more photons and, therefore, only occur at very high photon densities. Doubling the illumination intensity results in the four (22) or eight (23) times increase of SHG or third harmonic generation (THG), respectively.
Optical sectioning refers to the ability to image a single plane with high contrast in the middle of a thick specimen.
An optical parametric oscillator (OPO) laser converts pulsed light from a Ti:sapphire laser into tuneable output (~1100–1300 nm) that has the same repetition rate and pulse-length as the pump laser (Zhang et al., 2010).
PALM and STORM
Photoactivation localization microscopy (PALM) and stochastic optical reconstruction microscopy (STORM) are imaging methods that surpass the classical resolution limit of light microscopy through sequential localization of single fluorophores.
A non-fluorescent process by which a photon can lose or gain energy by interacting with a molecule. The magnitude of the energy change is a function of the illumination wavelength and properties of the molecule.
Stimulated emission depletion (STED) is an imaging method that surpasses the classic resolution limit of light microscopy by using two laser beams that are focussed in tandem to suppress fluorescence in a doughnut-shaped region around a central emission point.
Time correlated single photon counting (TCSPC) is a laser-scanning fluorescence lifetime imaging method that requires pulsed excitation. The laser intensity is attenuated until each pulse excites no more than a single fluorophore within the sample. For each pixel, the time of photon detection is correlated against the time at which the pulse occurred, and many tens of thousands of events are averaged to compute the fluorescence lifetime.
The titanium:sapphire (Ti:sapphire) laser is a sapphire crystal laser that emits pulsed light (~80 MHz repetition rate) over a broad infrared range (~780–1080 nm). Owing to the extremely short pulse duration (<200 fseconds), moderate average power levels are delivered in pulses of mW/cm2 to gW/cm2 intensity.
Multiphoton laser scanning microscopy
Multiphoton laser scanning microscopy (MPLSM) has become the technique of choice for imaging cellular dynamics in tissues. Unlike confocal microscopy, which generates optical sections on the basis of selective detection of light, MPLSM generates optical sections through the selective excitation of fluorescence within the sample. In MPLSM the energy required to excite an electron within a fluorophore (Box 1) is provided by simultaneous absorption of two photons (Denk et al., 1990). The probability of such an event is extremely small and only achieved in the focal point of the objective, where light from a pulsed laser is focussed to the smallest possible volume. This provides one of the main benefits of multiphoton over confocal microscopy, because only the molecules within the focal point are excited and photo-damage above and below the image plane is eliminated. The use of a pulsed laser has additional benefits, including its suitability for second harmonic generation (SHG) and fluorescence lifetime imaging (FLIM). A less obvious but equally important benefit is the flexibility in detection that is inherent to MPLSM. There is no requirement to focus the emitted fluorescence back through the laser-scanning system onto a pinhole aperture (Box 1 and Fig. 2). Instead, non-descanned detection can be used, whereby the detector is placed as close to the objective as possible (Williams et al., 1994). This approach is more effective at detecting the scattered photons that result from fluorescence excitation deep within a tissue sample and, therefore, leads to a higher sensitivity and a better signal–to–noise ratio (Centonze and White, 1998; Wang et al., 2002). However, disadvantages of MPLSM include the need for expensive, potentially hazardous, lasers and substantially increased photobleaching when the illumination intensity of the sample is too high (Table 1) (Patterson and Piston, 2000).
MPLSM has revolutionized the study of immune cell dynamics (reviewed in Garside and Brewer, 2008), Ca2+ dynamics (reviewed in Gobel and Helmchen, 2007) and the mechanisms underlying tumor cell invasion of host tissue (reviewed in Provenzano et al., 2009). In a seminal study Ahmed et al. used multi-photon microscopy to characterise tissue-specific expression of GFP in the mammary glands of transgenic mice (Ahmed et al., 2002). Using time-lapse imaging, these authors could also demonstrate the migration of individual tumor cells towards blood vessels in a manner suggestive of metastatic intravasation. Much subsequent work focused on characterizing cell migration in tumors within living organisms on the basis of, for example, EGFR expression (Xue et al., 2006) and the interaction of cancer cells with tumor macrophages (Wyckoff et al., 2007). In the latter study, the authors observed that, in MMTV-PyMT mice, intravasation only occurred within approximately one cell diameter (~20 μm) of perivascular macrophages; and these observations confirmed the importance of macrophages in promoting metastasis.
Generally, cytoplasmic expression of FPs has been used for tracking cell dynamics. However, some studies have made use of GFP to investigate subcellular protein localization. For example, Wyckoff et al. used GFP to assess the localization of the myosin light chain in metastatic MTLn3E cells within xenograft mouse tumors (Wyckoff et al., 2006). Similarly, Giampieri et al. used GFP to visualize the location of the transcription factor and transforming growth factor β (TGFB) effector SMAD2 to assess the influence of TGFB signaling on the motility of MTLn3E cells in vivo (Giampieri et al., 2009). Nuclear localization of GFP–Smad2 indicated TGFB signaling being active in individually migrating cells.
The redder the better in vivo
Light scattering, autofluorescence and absorption by tissue (which can cause phototoxicity) are attenuated at longer wavelengths, which has driven the development of red FPs for in vivo use (Shaner et al., 2008; Shcherbo et al., 2007). A typical red FP, such as mKate (Shcherbo et al., 2007), has a single-photon excitation peak of 588 nm, suggesting a two-photon excitation peak of ~1180 nm, which lies beyond the tuning range of a Ti:sapphire laser (Box 1). Such far-infrared wavelengths can be generated by an OPO laser (Box 1) (Andresen et al., 2009); however, the need for an OPO to excite red FPs adds complication and expense to an already sufficiently complex and expensive microscope system. Piatkevich and colleagues took a mutagenesis approach to resolve this problem by systematically changing the sequence of mKate to produce LSS-mKate1, which can be excited in the tuning range of a Ti:sapphire laser but retains emission at ~605 nm (Piatkevich et al., 2010). This enabled the authors to simultaneously excite cyan fluorescent protein (CFP; blue), fluorescein isothiocyanate (FITC; green), and LSS-mKate1 (red) at 870 nm in MTLn3 cells grown into mouse xenograft tumors, and to monitor cell polarization near and far away from the blood vasculature in response to a chemotactic gradient. Finally, the infrared fluorescence protein (IFP), which is based on conjugation of a bacterial phytochrome with biliverdin (Shu et al., 2009), emits photons at even longer wavelengths (>700 nm) and should, therefore, enable even deeper tissue imaging.
Imaging molecular dynamics using FPs
A variety of fluorescence techniques have been developed to report on molecular dynamics in cultured cells (Table 1). Although such
Box 2. Intermediate experimental systems
Cell-derived matrix (CDM) is a flexible, fibronectin- and collagen-based matrix that is derived from fibroblasts. It is ~5–10 μm thick and provides a source of growth factors and receptors for integrin engagement (Discher et al., 2005; Ginsberg et al., 2005). CDMs have been routinely used to assess integrin trafficking and the bi-directional, physical and mechanical interaction of cells with an elastic fibrillar network (Discher et al., 2005; Caswell et al., 2007). Organotypic 3D collagen I matrices consist of acid extracted native, crosslinked collagen that is polymerized in the presence of fibroblasts (Sabeh et al., 2004; Edward et al., 2005; Raub et al., 2007). Organotypic cultures are a malleable platform for the investigation of the important roles of stromal cells – either directly or through secretion of cytokines, growth factors or ECM – in the initiation and progression of tumorigenesis. Organotypic cultures can also involve the co-culture of multiple cell types on top of the matrix prior to interaction with stromal fibroblasts during invasion. This way, a great deal of information on tumor growth, survival and differentiation has been established (Edward et al., 2005; Edward et al., 2010; Meier et al., 2000). Ex vivo culture of organs or tissues offers the opportunity to image molecular or cellular dynamics within the complex architecture of authentic tissue with increased sample access and stability. Ex vivo live imaging has been used to examine melanoblast kinetics in embryo mouse skin, intestinal crypt activity within crypt cultures and immune infiltration with regards to leucocytes and dendritic migration in tissue explants (Lammermann et al., 2008; Lammermann et al., 2009). Transplantation models. Tumor growth can be modeled in vivo using transplanted cells, including stably transfected cell lines. Transplantation models mimic native tissue features including a 3D setting, multiple ECM interactions, formation of vasculature networks and interactions with stromal cells. Tumors can have both hypoxic and necrotic regions, and some models result in local invasion of the surrounding tissue or bone (Canel et al., 2010a; Condeelis and Segall, 2003; Serrels et al., 2009). However, such models frequently involve transplantation of cells into a tissue that is different from their origin and use of immunocompromised animals. Genetically engineered mouse (GEM) models of disease have been engineered to recapitulate the progression and histopathology of human disease. Increasingly, fluorescent protein expression has been incorporated to enable the study of cellular dynamics within these tumor models. Murine models of human cancers that use the Cre–Lox technology to target expression of known oncogenes and/or tumor suppressors in parallel with fluorescent proteins to specific tissues have recently facilitated the study of pancreatic, bladder, renal and colon cancer (Ahmad et al., 2011; Cole et al., 2010; Doyle et al., 2010; Morton et al., 2010). The use of intermediate systems described above combined with in vivo work can substantially improve our insight into cell behavior in the given disease state.
techniques have been developed for ideal experimental conditions that use cells grown on glass coverslips, the techniques covered here have also been successfully applied to study molecular dynamics in tissue environments.
Fluorescence recovery after photobleaching (FRAP) (Axelrod et al., 1976; Lippincott-Schwartz et al., 2003) involves rapidly photobleaching the fluorescence within a small region of the sample through intense excitation and then analysing the recovery of fluorescence intensity in the bleached region over time. A FRAP recovery curve can be used to derive two primary parameters: the immobile fraction and the half time of recovery. The immobile fraction is the fraction of bleached molecules that remains trapped in the analysis region and, therefore, limits full recovery of the fluorescence intensity. The half time of recovery is a measurement of the rate at which freely moving molecules diffuse in and out of the bleached region. Molecular diffusion can also be interrupted by transient binding or ‘corralling’ events that reduce the apparent diffusion rate. FRAP acquisition tools are standard features of most commercial CLSM or MPLSM systems and many approaches for data analysis have been described (Sprague and McNally, 2005).
FRAP has been extensively used to assess the dynamics of GFP-tagged proteins in vitro; however, quantitative investigations of molecular dynamics at the subcellular level in vivo remain limited. In Drosophila, FRAP has been used to address the crosstalk between actin dynamics and E-cadherin. These studies identified two sub-populations of actin that regulate either clustering or lateral movement of E-cadherin (Cavey et al., 2008). Similarly, we have used FRAP to compare E-cadherin dynamics within A431 cells grown in vitro and in subcutaneous tumors in mice (Serrels et al., 2009; Timpson et al., 2009). Interestingly, the efficacy of the clinically approved Src inhibitor dasatinib, which impairs E-cadherin turnover, was substantially greater in vivo than in vitro despite the same cells being used (Serrels et al., 2009), which highlights an important caveat in the use of cultured disease models as a replacement for living tissue.
Photoactivation and photoconversion of fluorescence
Photoactivation of fluorescence is a process during which a non-fluorescent molecule is converted into its fluorescent form, usually through a chemical reaction that is driven by ultraviolet light. For instance, activation of photoactivatable GFP (PA-GFP) by intense illumination at 405 nm drives decarboxylation of Glu222, which converts the tripeptide fluorophore at the heart of this GFP mutant from a neutral (dark) to an anionic (fluorescent) state (Patterson and Lippincott-Schwartz, 2002). Photoactivation is, essentially, the inverse of photobleaching; however, it can offer improved sensitivity of detection by creating a positive signal against a dark background and, therefore, allowing the distribution of subcellular pools of PA-GFP-tagged protein to be tracked over time. The main weakness of the photoactivation approach is that labeled proteins are nonfluorescent and, therefore, difficult to target with the 405 nm laser prior to activation. A solution to this problem is to label proteins of interest with a fluorophore, such as Kaede (Ando et al., 2002) or Dendra (Gurskaya et al., 2006), which can be photoconverted from one color to another through targeted illumination at 405 nm. A drawback of photoconversion compared with photoactivation is that it requires the dedication of two detection channels (e.g. green and red) to image the labeled protein before and after conversion.
As with FRAP, photoactivation and photoconversion have found widespread use in vitro, but their application in vivo is more limited. At the cellular level, photoconversion of Dendra2 has been used to track the migration and intravasation of orthotopically injected breast cancer cells in conjunction with a surgically implanted imaging window (Gligorijevic et al., 2009; Kedrin et al., 2008). The authors were able to repeatedly image a pre-selected pool of cells for up to 21 days in the context of the surrounding environment, which included the vasculature and the ECM (see also Gligorijevic and Condelis, 2009). Recently, this technology has been used to quantitatively and repeatedly measure tumor cell motility and proliferation in response to clinically relevant anti-metastatic agents that target the integrin–Fak–Src signaling axis over long time courses (Canel et al., 2010). A weakness of photoactivation and/or photoconversion for long-term tracking of cells is that the cytoplasmic signal becomes diluted with successive cell divisions. Photoactivation of gene expression (see following section) is, therefore, a better approach in this case. Photoactivation in vivo was used for the first time to track plasma membrane dynamics, by using the membrane-targeting sequence of H-Ras linked to photoactivatable GFP (Serrels et al., 2009; Timpson et al., 2009). Proteins attached to the plasma membrane were found to be substantially less mobile in tumor cells that had been grown in living animals compared with the same cells cultured in vitro (Serrels et al., 2009), an observation that is of crucial importance for the initiation of signal transduction at the cell surface (Radhakrishnan et al., 2010).
Photoactivation of protein activity
Advances in protein photochemistry have recently progressed beyond the uncaging of fluorescence and now enable the uncaging of protein activity by using targeted bursts of light. Recently, a constitutively active mutant of Rac1 was fused to the photoreactive domain of phototropin and the resulting steric hindrance impaired the interaction of Rac1 with its effectors (Wu et al., 2009). Upon illumination with light at 458 nm, the photoreactive domain of phototropin unwinds and uncages the fused protein, thereby leading to Rac1 activation. In this way, the precise location and timing of protein activation can be achieved at subcellular resolution. This method has recently been used in vitro to control the location of protrusions and direction of fibroblast migration (Levskaya et al., 2009; Wu et al., 2009), and in vivo to selectively activate Rac in cells that migrate through tissue in Drosophila (Wang et al., 2010) and zebrafish (Yoo et al., 2010).
An alternative approach that is based on phytochrome 2 and CRY2-interacting bHLH from Arabidopsis thaliana has been used to activate protein expression and Cre-recombinase-mediated DNA recombination in cultured cells (Kennedy et al., 2010). In the near future, these techniques will hopefully enable new experimental approaches, such as photoactivation of gene expression, which would allow the activation of oncogenes and fluorescent reporters in specific locations within tissues – such as the intestine or liver – followed by repeated imaging to study the progression of disease and response to drug treatment at the molecular level.
FRET and FLIM
Förster resonance energy transfer (FRET) involves the direct transfer of energy from a donor fluorophore to an acceptor fluorophore. Because this can only occur when the two fluorophores lie within close proximity (~5 nm), FRET indicates a direct molecular interaction between two fluorescently labeled proteins. This has enabled analysis of when and where proteins interact within cells. There are two primary means of detecting FRET, through fluorescence intensity or fluorescence lifetime.
FRET results in the loss of donor fluorescence and an increase in acceptor fluorescence, which can be quantified by imaging both channels and calculating their ratio. By using this approach, Kardash et al. demonstrated the subcellular activation of Rac1 and RhoA at the front of migrating germ cells during zebrafish embryonic development (Kardash et al., 2010). Ratiometric detection of FRET is easy to implement on virtually any live cell imaging system; however, corrections are required to compensate for issues such as excitation and emission bleed-through, and differences in protein expression levels (Berney and Danuser, 2003). A wide variety of FRET probes now exist for the investigation of many aspects of cell migration including adhesion, signaling and membrane dynamics (Sabouri-Ghomi et al., 2008).
FRET can also be detected as a change in the fluorescence lifetime of the donor fluorophore by fluorescence lifetime imaging (FLIM). Fluorescence lifetime is most commonly measured by time-correlated single-photon counting (TCSPC, Box 1), although many alternative approaches, which include frequency domain detection (Lakowicz et al., 1992a) and time-gated wide-field detection (Elson et al., 2004), now exist. TCSPC requires the use of a pulsed laser source and is, therefore, amenable to combination with two-photon excitation and non-descanned detection. A key benefit of FLIM detection of FRET (FLIM–FRET) is that this technique is insensitive to the expression levels of donor and acceptor fluorophores.
Using the FLIM–FRET technique a wide variety of biochemical interactions have been analyzed in vitro. More recently, progress has also been made in the application of this technique to study living mammalian tumor tissue. In an initial study FLIM–FRET was used to image the interaction between GFP-tagged chemokine (C-X-C motif) receptor 4 (CXCR4-GFP) and red fluorescent protein (RFP)-tagged protein kinase C α (PKCA-RFP) in mammary carcinoma xenografts (Kelleher et al., 2009). In another study, FLIM was used to image a caspase-3-based FRET probe to examine the development of drug-resistance in a syngenic mouse tumor model (Keese et al., 2010). In the first case, the authors used fluorescence lifetime to demonstrate differences in receptor binding when comparing deep and shallow tumor regions. In the second study changes in fluorescence lifetime were analyzed to investigate the tumor response following drug treatment. However, in both cases FLIM–FRET was analyzed on the basis of the response of entire fields of cells, rather than single cells or subcellular regions. More recently, our laboratory has used FLIM–FRET to examine the subcellular activation of RhoA in a metastatic model of pancreatic cancer driven by mutations in KRas and p53R172H. In vivo FLIM-FRET demonstrated that Rho is activated specifically at the poles of invasive cells, and that treatment with dasatinib at a concentration that impairs metastasis of primary tumors to the liver, specifically abolishes polar Rho activation without affecting the basal activation level within the cell body (Timpson et al., 2011).
Fluorescence correlation spectroscopy
Fluorescence correlation spectroscopy (FCS) is not a conventional imaging method. However, this powerful technique uses the same basic hardware as CLSM or MPLSM, and several commercial laser scanning microscopes (LSMs) can be used for both correlation spectroscopy and imaging. FCS delivers information about the concentration, mobility and binding affinities of fluorescently labeled molecules in solution, in cells and, more recently, in developing tissues (reviewed in Haustein and Schwille, 2007). Like in CLSM, a laser beam is focused into the sample but, unlike CLSM, the focal point remains fixed in one position and the signal is derived from the diffusion of fluorescently labeled molecules in and out of the focal volume. Comparison of fluctuations in the fluorescence intensity on a time-scale of milliseconds to seconds generates an autocorrelation function that can be analyzed in many different ways, including cross-correlation of fluorescent signals from two different molecules [fluorescence cross-correlation spectroscopy (FCCS), reviewed in Bacia et al. (Bacia et al., 2006)].
In a ‘tour de force’ study that involved many technical refinements, Ries et al. used FCCS to determine the concentrations and binding affinities of fibroblast growth factor (FGF) receptors 1 and 4 for FGF8 in living zebrafish embryos undergoing gastrulation (Ries et al., 2009). This work showed that the affinities derived from the measurement of purified proteins were substantially different from the apparent affinities measured in a complex reaction in vivo. Similar to FRET, FCCS provides information about the association of fluorescently labeled proteins. However, detection of molecular complexes by FCCS is not limited by the distance and orientation between the two fluorophores. This limitation can lead to false-negative FRET results when two proteins interact but their associated fluorophores are too far apart or have the wrong orientation for efficient FRET.
Complementary imaging approaches
Although FPs are the corner-stone of in vivo imaging, their use in animal models is limited by time-consuming and expensive production and breeding strategies, particularly when complex genetic backgrounds are required. However, a variety of techniques can be used to complement FPs. The best complementary techniques are easy to use in conjunction with any genetic model and provide contextual information for the interpretation of results derived from experiments with FP probes.
Use of exogenous fluorescent dyes
Perhaps the easiest method of generating fluorescence signals in living organisms is through the injection of fluorescent dyes. One simple approach is the non-specific use of inert fluorescence markers, such as rhodamine–dextran (Wang et al., 2002), or the use of quantum dots (reviewed in Zrazhevskiy et al., 2010) to visualize, for example, the vascular (Larson et al., 2003; Stroh et al., 2005) and lymphatic networks (Kim et al., 2004). An interesting functional aspect of using quantum dots is that, as a result of their larger size, they tend to only extravasate under pathological conditions (Kim et al., 2009; Stroh et al., 2005). Therefore, the detection of a fluorescence signal within the surrounding tissue provides physiologically relevant information about vascular leakage. One of the oldest specific labels to be used for fluorescence imaging in tissues are Hoechst dyes. They label DNA through intercalation, have a high LD50 in mice, and were originally used to label the nuclei of vascular endothelial cells (Chaplin et al., 1985; Olive et al., 1985). They can also be used effectively to label cell nuclei in the skin and intestinal crypt.
Some drugs contain delocalized electron structures that convey fluorescence. These include anthracycline-intercalating agents, such as doxorubicin (Olive et al., 2009) and epirubicin (Featherstone et al., 2009); photosensitizing agents, such as porphyrins (Pathak et al., 1995); acriflavine (Kiesslich et al., 2007), and psychoactive compounds including LSD (Fisher et al., 2003). This drug autofluorescence offers the ability to directly visualize the penetration of therapeutic agents into their target tissue, which is important for many tumor types in which poor vascularization may limit drug efficacy (Olive et al., 2009).
Injection of fluorescently labeled cells has also proven to be an effective labeling approach. Seminal multiphoton imaging of T- and B-cell interactions was performed by purifying and labeling each population from donor mice, injecting cells back into recipient mice and allowing them to migrate into lymph nodes prior to harvesting the nodes for ex vivo observation (Bousso et al., 2002; Miller et al., 2002; Stoll et al., 2002). Similarly, the uptake of quantum dots into the cytoplasm has been used to label cells in culture that were subsequently injected into mice and imaged on the basis of nano-particle fluorescence (Stroh et al., 2005; Voura et al., 2004).
Use of functional dyes
Over the past 10 years there has been an active development of functional probes, in which fluorescence is activated in response to environmental or enzymatic properties of the sample. Weissleder and colleagues have developed a class of activatable probes that fluoresce in the infrared range only after specific enzyme cleavage (Bremer et al., 2001). One such probe is MMPSense™680, which fluoresces following cleavage by metalloproteinases – including matrix metallopeptidases 2, 3, 9 and 13 (Bremer et al., 2001) (Fig. 1). A related probe, ProSense™680, is activated by proteases such as cathepsin B, L or S and plasmin (Nahrendorf et al., 2007). Instead of measuring enzymatic activity, Blum et al. have designed a fluorescent probe that labels and inhibits cysteine cathepsins of the papain family in mouse tumors (Blum et al., 2007). Kuchimaru and colleagues have taken an alternative approach by using Halotag chemistry to conjugate an infrared fluorophore to the oxygen-dependent degradation and protein-transduction domains (ODD–PTD) of hypoxia-inducible factor α (HIF1A) to generate a probe that is actively taken up by cells in hypoxic environments (Kuchimaru et al., 2010; Palmer et al., 2010). These functional dyes can provide valuable information about the local tumor environment and are easy to combine with any genetic model.
Imaging of auto-fluorescence
Cells and tissues contain a number of endogenous chromophores that generate autofluorescence. These include melanin, keratin, lipofuscin, flavin adenine dinucleotide (FAD) and reduced nicotinamide adenine dinucleotide (NADH). Autofluorescence is typically encountered as a problem that hinders detection of fluorescent probes through the generation of high background fluorescence, although such background can be effectively removed using spectral unmixing (reviewed in Levenson and Mansfield, 2006).
Alternatively, autofluorescence can be used as a primary signal to visualise biologically relevant processes (Zipfel et al., 2003). Of particular interest is the use of NADH fluorescence to read out the oxidative state in cells and as a marker of hypoxic regions in tumors (reviewed in Provenzano et al., 2009). It has been established that free NADH has a short fluorescence lifetime of around 400 pseconds, whereas the lifetime increases to between 1000 and 3700 pseconds when it is enzyme bound (Lakowicz et al., 1992b; Niesner et al., 2008). This difference enables the proportions of free and enzyme-bound NADH to be determined by FLIM. This approach has been used at the tissue level to demonstrate that the gradient in the redox potential, which is associated with healthy epithelial tissue, is lost in pre-cancerous tissue in the DMBA-treated hamster cheek pouch model of oral cancer (Skala et al., 2007). Lifetime imaging of autofluorescence has also been used to image a variety of tumor types (McGinty et al., 2010) and has been applied to image the differences between benign and malignant breast (Tadrous et al., 2003) and colorectal (Zavadil et al., 2005) human tissue biopsies, as well as between basal cell carcinomas and surrounding healthy tissue (Galletly et al., 2008).
Second harmonic generation
Some crystalline materials are able to promote the combination of two photons of one wavelength into a single photon that possesses twice the energy and half the wavelength. This non-linear process, known as second harmonic generation (SHG; Box 1) depends on the electric dipole properties of the crystal and only occurs under extremely intense coherent illumination. SHG is essentially non-toxic because no energy is absorbed by the sample in the process, and it is easy to combine with multiphoton excitation of fluorescence because both require the high illumination intensities derived from a pulsed infrared laser (Fig. 2) (Brown et al., 2003).
Because of their chiral, semi-crystalline structure, collagen fibrils are efficient second harmonic generators (Verbiest et al., 1998). Collagen SHG is routinely used to complement MPLSM of FP-labelled tumor cells by providing contextual information about local tissue structure (Wolf et al., 2009; Wyckoff et al., 2006; Zipfel et al., 2003). Collagen SHG can also be used to identify specific features of a tissue, such as the dense irregular connective tissue that marks the base of the intestinal crypt (Fig. 2D). Images of collagen SHG have been analysed in order to derive a range of parameters including collagen age (Williams, 2005), tissue stiffness (Raub et al., 2007; Raub et al., 2008) and remodelling of the ECM (Perentes et al., 2009). We recently used this approach to assess collagen density and stiffness in mouse skin and, thereby, demonstrated an unexpected feedback between cancer cells and surrounding ECM components (Samuel et al., 2011) (Fig. 2E). Through a similar non-linear optical process, third harmonic generation (THG; the conversion of three photons into one photon that has three times the energy and, therefore, one third the original wavelength) can be generated at interfaces such as the plasma membrane (Müller, 1998). SHG and THG have been successfully combined to study cell division in early zebrafish development (Olivier et al., 2010). In this study, the authors used SHG from mitotic spindles to identify the centre of embryonic cells and THG from the plasma membrane to identify cell-cell boundaries, in order to generate a lineage tree of cell divisions over the course of several hours.
Imaging experiments in vivo differ from in vitro experiments in many ways. Not only do cells look and behave differently but also the questions asked are often subtly different. Furthermore, imaging cells within tissues presents many technical challenges that constrain in vivo experimental design and influence the choice of imaging approach. In this section, we consider some of the issues that relate to in vivo experimental design and offer suggestions how to create an experimental pipeline that promotes successful in vivo imaging.
Use of intermediate systems
In vivo experimental design presents many challenges that are not encountered in vitro, including surgical preparation of animals, anaesthesia, suppression of motion artefacts arising from breathing, heartbeat and muscle twitching, a limited timescale of observation, a reduced number of experimental observations, difficulty synchronizing experiments, and reduced optical sensitivity and resolution. Three dimensional tissue culture systems (Fig. 1; Box 2) offer useful intermediates for the development of in vivo imaging approaches and, compared with 2D cell culture, allow different questions to be addressed. For instance, organotypic cultures (Box 2) are better approximations of the in vivo cellular environment than glass coverslips, and offer superior optical properties and greater ease of use than live animals. Crucially, they are more amenable to experimental manipulation and, therefore, provide a useful system in which to characterize probes, perform controls and undertake initial experiments prior to in vivo work. The development of sophisticated intermediate systems, including recently published models for the intestinal crypt (Sato et al., 2009) (Fig. 2B) and optical cup (Eiraku et al., 2011), suggest that many tissue environments will soon be amenable to in vitro study. Such models will aid the development of in vivo imaging experiments by promoting an experimental pipeline (Fig. 1A; Table 2) in which similar questions are asked, using a progressive series of experimental systems that are characterized by increasing biological fidelity and decreasing experimental tractability.
Choice of experimental set-up
In this Commentary we have presented a variety of imaging techniques and intermediate model systems, which are characterized by many different strengths and weaknesses (Tables 1 and 2). The art of successful experimental design lies in matching an appropriate imaging approach with the required experimental system to obtain clear and meaningful results. Fig. 3 presents a flowchart to assist in the selection of experimental systems and imaging techniques. One guideline to use in conjunction with this is to ‘start as you intend to finish’. Although wide-field CCD imaging is generally superior for imaging cells in culture, we have found it important to use the same probes and imaging techniques throughout an experimental pipeline. This helps to ensure that results are comparable across experimental platforms. Furthermore, in this way in vitro imaging provides a ‘best-case example’ for the sensitivity and resolution that can be achieved in vivo.
Design of cell migration experiments
The study of cell migration serves to illustrate differences in the constraints of in vitro and in vivo experimental design. Migration has long been studied using cells grown on glass coverslips, which has led to the analysis of features such as cell area, polarization of shape, and the rate and persistence of migration on the basis of statistics derived from large numbers of cells. Recent progress in in vivo imaging, however, has highlighted the difficulty of observing specific events in vivo, which results in small numbers of observations and difficulties in planning and synchronizing experiments. Imaging cell migration through tissue has also led to greater interest in the mechansims by which cells move through 3D environments and the environmental signals that promote migration. Such questions have been effectively addressed using reconstituted collagen meshworks to study the requirement for MMP activity to hydrolyze – rather than simply squeeze through – the local environment (Wolf et al., 2007). Likewise, organotypic cultures have been useful for investigating the role of stromal cells in promoting tumor cell invasion (Gaggioli et al., 2007). In both of these examples, results that were first obtained using intermediate model systems were subsequently confirmed in vivo.
A further difference between in vitro and in vivo design is the need to collect information about the cellular environment in vivo. An example of this is the extensive use of confocal reflectance and SHG to image collagen in cell migration studies. More such approaches are needed to image other important environmental parameters, such as hypoxia and local concentrations of growth factors and cytokines. Because the time and place of many important disease-related events is unpredictable, in vivo studies also require the development of steady-state molecular read-outs as surrogates for direct observation of rare events; for example, monitoring GTPase activation or cell adhesion dynamics as surrogates for cell migration. Intermediate culture systems can help to overcome some of these problems through the characterization of surrogates under controlled conditions.
There are many developments on the horizon that have the potential to impact in vivo imaging. New imaging approaches, such as optical frequency domain imaging (Vakoc et al., 2009) and stimulated Raman scattering [SRS (Saar et al., 2010)], have the potential to complement FP imaging by visualizing tissue structure in the absence of exogenous labeling. As a Raman technique (Box 1), SRS is particularly attractive because it offers the possibility of identifying bio-molecules label-free on the basis of their chemical structure. Great progress has recently been made in the development of super-resolution techniques including PALM, STORM and STED (Box 1) (reviewed in Schermelleh et al., 2010); however, challenges remain in applying these methods to image dynamic structures in living tissue. As a laser scanning technique, STED seems to have the greatest potential for dynamic, in vivo measurements. Single-plane illumination microscopy [SPIM (Huisken et al., 2004; Keller et al., 2008)] has produced impressive results in developmental systems, such as zebrafish and Drosophila; however, this approach requires embedding the entire sample in a medium such as agarose and, therefore, may be of limited use for mouse cancer models.
The development of automated Ti:sapphire lasers has brought MPLSM to a wider range of users, and automated OPO lasers – for excitation of red Fps – are likely to continue this trend (Box 1). However, there is an urgent need for better red FPs that overcome photostability and toxicity problems, and can be used for applications requiring multiple labels – especially red FRET reporters. The success of Raichu probes for monitoring the activity of small family GTPases suggests that there are opportunities to develop sensors for other forms of protein activity, especially kinases involved in signal transduction. Other MPLSM developments with the potential to increase the resolution and sensitivity of MPLSM at greater tissue depth are adaptive optics (Girkin et al., 2009), which increases the efficiency of excitation, and ‘total emission detection’ (Combs et al., 2010), which increases the sensitivity of light detection. Finally, new approaches to accessing living tissue in vivo are required to extend the list of (murine) tissues accessible to live imaging. This includes increased refinement of observation windows (Gligorijevic et al., 2009), exteriorizing (Coppieters et al., 2010) and stabilizing (Toiyama et al., 2010) internal organs, and stick lens (Alencar et al., 2005) and endoscopic (Hsiung et al., 2008) approaches, which are compatible with a variety of wide-field and LSM imaging modalities.
The combination of advanced fluorescence imaging techniques with genetically engineered mouse models presents a new paradigm for research on the molecular level in vivo. Previously, cell culture has served as a proxy because most experiments were not technically feasible in vivo. However, it is increasingly clear that some approximations made in vitro fundamentally limit the validity of results, and that basic molecular responses might be different within the same cells cultured in vitro compared with those in vivo (Serrels et al., 2009; Timpson et al., 2011). To fully realize the potential at hand a few things have to happen. First, we must continue to adapt existing methods of imaging molecular interactions for use in living animals. Second, we must develop intermediate model systems that more closely recapitulate conditions in vivo. Last, we have to develop new experimental concepts that are based on the synthesis of imaging technology, intermediate systems, and new mouse models.
We thank Haley Bennett for critical reading of the manuscript, and Bojona Gligorievic for critical reading of the manuscript and help assembling the reference library. We also thank Pat Caswell, Jennifer Morton, David Huels and Yafeng Ma for images provided.
P.T. was supported by a fellowship from AstraZeneca, E.J.McG. and K.I.A. were supported by a Cancer Research UK core grant. The authors have no conflict of interest.