1.

Biological Significance of Tumor Microcirculation

It is a well-established concept that the continued growth and progression of a tumor is dependent on the development of a vascular network to supply oxygen and nutrients.¹ Tumors develop a vascular network through co-option of the surrounding normal vasculature and/or angiogenesis—the development of neovasculature from preexisting vessels.^{2, 3} The blood vessels formed by these processes are characterized by multiple structural and functional abnormalities, including tortuous vascular networks, neovascular shunts, blind ends, plasma channels, and temporally as well as spatially heterogeneous blood flow.^{4, 5, 6, 7, 8, 9} These abnormalities result in tumor regions with poor delivery of oxygen and nutrients, resulting in hypoxia and low pH, as well as high vascular permeability and microvessel resistance, which contribute to elevated tumor interstitial fluid pressure (IFP).^{10, 11} Tumor hypoxia, acidity, and IFP have been related to disease progression and adverse therapeutic outcome.^{12, 13} In addition to the clinical impact of these indirect features of tumor vascular abnormalities, tumor vessel density has been directly related to a poor prognosis in multiple studies.^{14, 15, 16, 17} The key requirement for tumor angiogenesis in tumor progression, coupled with its clinical significance, highlight the tumor vasculature as a potential biomarker of disease status and exploitable therapeutic target. Thus, to take full advantage of the tumor vasculature as a target, it is necessary to understand the processes involved in tumor vascularization, the structural and functional abnormalities that characterize tumor vessels, and their response to therapy. A large part of what is known to date from in vivo studies has been uncovered using optical imaging, more specifically intravital microscopy (IVM). We discuss some of the different IVM approaches—what has been shown to date, and current and future applications of optical imaging in tumor angiogenesis.

2.

Optical Imaging and Intravital Microscopy

Optical imaging describes the use of the visible spectrum extended to include ultraviolet and near infrared $(\sim 300\text{to}1300\mathrm{nm})$ light imaging. Macroscopic bioluminescence and fluorescence techniques are becoming common-place in cancer research for in vivo studies of the tumor microenvironment, particularly induction of and changes in hypoxia and tumor cell metastasis/invasion.^{18, 19, 20} However, microscopic optical imaging of tumors in vivo (intravital microscopy or IVM) enables high resolution studies at a cellular/subcellular level in real time, with applications in clinical as well as experimental animal research, via technological advances such as microendoscopy.^{21, 22} Recent advances in fluorescent/bioluminescent probes, microscopes and associated filters, image acquisition, and analysis technology have ensured continued development of IVM as a valuable research tool in the field of tumor angiogenesis and microcirculation, permitting the study of tumor cell interactions with stroma/blood vessels,^{23, 24} gene expression and vascular function^{25, 26} – blood supply^{27, 28} and vascular permeability,^{29, 30} among others.

3.

Fluorescent Probes for Bioimaging

Many different types of probes are available for optical imaging and IVM^{31, 32} from fluorescent dyes^{33, 34, 35} to photoactivatable and photoswitchable fluorescent proteins^{32, 36, 37} (see Fig. 1 ).

Fig. 1

Fluorescent probes. (a) Multiphoton image of FITC-dextran labeled blood vessels in a murine fibrosarcoma growing in the dorsal skin fold window chamber (DSFC). (b) A neonatal Kaede transgenic mouse (reprinted with kind permission from Tomura ¹⁴¹) before (upper) and after (lower) photoconversion on exposure to ultraviolet light for 30, 60, or $90\sec$ . (c) Fluorescent image of human prostate carcinoma (PC3) tumor constitutively expressing mCherry growing in the DSFC of a transgenic Tie2-GFP mouse. Tumor cells are red, and GFP expression is visible in the blood vessel walls. (d) Multiphoton image depicting the use of quantum dots (reprinted with kind permission from Stroh ⁴²) in vascular imaging of a MCaIV mammary tumor growing in the DSFC of a VEGF-GFP transgenic mouse. The figure shows concurrent imaging of quantum dot labeled blood vessels (QD470, blue) and VEGF-driven GFP expression in perivascular cells. Concurrent imaging at $850\mathrm{nm}$ was possible due to the broad excitation spectrum of the quantum dots (scale bar, $50\mu \mathrm{m}$ ).

Fluorescein isothyocyanate (FITC) is an organic fluorescent dye that continues to be widely used after several decades.^{33, 38} It has excitation and emission maxima at 488 and $520\mathrm{nm}$ , respectively, and demonstrates high fluorescence intensity, but like many fluorophores, particularly organic ones, is prone to photobleaching. This has led to the development of synthetic dyes such as the Alexa-fluor dyes (Alexa-fluor 350, 488, 555, etc.) that demonstrate improved photostability alongside high intensity. FITC can be conjugated to different molecules. For instance, dextrans are polysaccharides composed of glucose chains of varying lengths; consequently, multiple molecular weight versions of FITC-dextran exist, and can be used to provide valuable information on vascular permeability through analysis of vascular leak.³⁸

In addition, there are chemically sensitive derivatives of FITC, such as the pH sensitive fluorochrome $2^{\prime }$ , $7^{\prime }$ -bis-(2-carboxyethyl)-5,6-carboxyfluorescein (BCECF). BCECF excitation occurs in a pH-dependent manner at wavelengths between 440 and $500\mathrm{nm}$ , and has been used to generate spatial and qualitative data on tumor pH.³⁹ The fluorescence spectra of porphyrins are also pH sensitive, and phosphorescence lifetime is oxygen sensitive, enabling elegant studies of the tumor microenvironment.⁴⁰

Although useful, organic dyes have several limitations. In addition to photobleaching, they generally exhibit broad emission spectra, which limit their capacity as dual markers of different events/structures because of cross talk between the dyes. Quantum dots represent a class of fluorescent nanocrystals synthesized to have broad excitation peaks and sharp emission peaks to overcome these limitations.^{35, 41} They exist in a range of colors, and generate bright fluorescence with improved photostability, making them less prone to photobleaching, an important and desirable property for live imaging studies that might involve prolonged and repeated exposure times. Stroh ⁴² exploited the broad excitation peak of quantum dots to image MCaIV murine mammary tumor vasculature (following IV injection of quantum dots) and perivascular cells simultaneously [VEGF-GFP transgenic mice expressing VEGF-driven green fluorescent protein (GFP)] using the same wavelength [see Fig. 1].

In recent years, fluorescent probes have been developed to exploit a process known as Förster resonance energy transfer (FRET). If two different fluorophores are located within the Förster radius (typically $10\mathrm{nm}$ ) of each other and one has a peak emission wavelength that is broadly the same as the peak excitation wavelength of the other (i.e., significant emission-excitation spectral overlap), a nonradiating energy transfer can occur (FRET). The emission energy will transfer from the first fluorophore to the second non-radiatively. For example, on exposure to indigo/blue light, a cyan fluorescent protein (CFP) molecule (the donor) in close proximity to a yellow fluorescent protein (YFP) molecule (the acceptor) will cause the YFP to emit yellow light without having been exposed to the green light normally required to excite its fluorescence. This technique is used to demonstrate molecular interaction and colocalization, and has been used primarily for studies of cells in vitro. However, Zhou ⁴³ recently described in vivo use of a FRET probe molecule consisting of eCFP linked to Venus (a yellow fluorescent protein variant, see discussion later on), with a sequence containing a caspase-3 cleavage site. They used this probe to monitor the dynamics of caspase-3 activation induced by the action of the drug cisplatin and photodynamic therapy in tumors using FRET IVM. On caspase-3 activation, the molecular link between the eCFP and Venus is broken and the FRET signal is lost. Similar FRET probes have been used, in mouse tumor models in vivo, to investigate the activity and therapeutic targeting of matrix metalloproteinases, enzymes intimately involved in angiogenesis.⁴⁴

The discovery of green fluorescent protein (GFP) in 1962,^{45, 46} and the plethora of fluorescent proteins that have been uncovered, refined, and synthesized since, has revolutionized in vivo optical imaging. The literature available on the use of GFP and other fluorescent proteins (BFP, CFP, YFP, RFP)³² is vast. The main use of these proteins is to fuse them to genes of interest that in turn regulate their expression, so that fluorescence is used as a reporter of gene expression in vivo, at multiple time points. IVM studies of tumors using GFP and similar proteins have primarily focused on tumor cells transfected with the corresponding gene, under the control of a constitutively expressed viral promoter. This allows microscopic tracking of tumor cell proliferation and migration patterns.⁴⁷ Fluorescence of GFP does not require specific cofactors but does depend on oxygen concentration,⁴⁸ such that data need to be interpreted with caution. The potential for such large proteins to modify cell behavior is also a concern, although several studies have demonstrated minimal effects of GFP in tumor cells on such properties as tumor progression and metastasis.⁴⁹ Several fruit fluorescent proteins have been developed more recently, which are mutants of a monomeric mutant form of dsRed (mRFP1). These proteins (e.g., mCherry and TdTomato) are potentially less disruptive than GFP, and also provide a range of colors from yellow-orange to red-orange, with improved quantum efficiency and stability.⁵⁰ To date these proteins have not been used in IVM studies of tumors; however, their suitability for in vitro live cell imaging studies has recently been explored,⁵¹ and it is likely that newer gene expression vectors will be synthesized to exploit their advantages.

Proteins demonstrating reversible and irreversible changes to their spectral properties on exposure to specific optical wavelengths have now been discovered and developed through modification and refinement of existing fluorescent proteins. These are referred to as photoactivatable (photoactivated shift from low to high fluorescent intensity) or photoswitchable (photoactivated shift in fluorescent color) fluorescent proteins (PAFP).^{37, 52}

The first natural PAFP to be discovered was Kaede in 2002,⁵³ which initially fluoresces green, but on exposure to UV light undergoes an irreversible conversion to a red-fluorescent form. In the same year, a photoactivatable mutant version of GFP was synthesized, PA-GFP, exhibiting a wavelength-specific conversion to an enhanced fluorescent form.^{52, 54} Dronpa and KFP1 exemplify reversible PAFPs.^{55, 56} Dronpa, which fluoresces green, has the capacity for reversible wavelength-specific bleaching,⁵⁵ whereas KFP1 exhibits both reversible and irreversible changes in fluorescent expression—exposure to specific wavelengths can be used to induce relatively short-lived bright red fluorescence that can be quenched, or the exposure times or intensity can be increased to stimulate irreversible “kindling” of red fluorescence that cannot be quenched and does not fade.⁵⁶ Although the potential of these proteins has yet to be fully explored, they represent novel and highly useful tools. For example, the idea of inducing an irreversible photoconversion in a cell population of interest and then tracking that population, relative to new incoming cells that have not undergone photoconversion, is an attractive one. Care needs to be taken because new cells produced on cell division will express the fluorescent protein in its original form (e.g., Kaede), thus diluting the population of interest. Nonetheless, with careful planning there is a wide scope for use of these proteins.

4.

Dorsal Skin Fold Window Chamber

The dorsal skin fold window chamber model (DSFC) (see Fig. 2 ) is an invaluable tool for IVM of tumor angiogenesis.⁵⁷ It was originally developed in the rabbit ear in 1924⁵⁸ and adapted for use in mice in 1943.⁵⁹ The DSFC model in mouse involves the surgical removal of a skin flap and associated connective tissue and fat from the dorsum to provide optical clarity. A metal chamber (usually aluminium or titanium) holds the skin away from the mouse body, exposing the vascularized layer of muscle and skin on the opposing side of the flap or “window.” A tumor piece or tumor cells (often as a spheroid or cell conglomerate) are placed onto, or injected into, the exposed fascial plane and covered with a glass coverslip. This process is described in detail by Koehl, Gaumann, and Geissler.⁵⁷ It should be noted that for larger animals (e.g., rat) that have thicker skin, the skin layers from both opposing sides of the flap are removed, thereby facilitating an improved optical path.⁶⁰

Fig. 2

Dorsal skin fold window chamber (DSFC). (a) A photograph is shown of a DSFC preparation on the rear dorsum of a SCID mouse. The exposed blood vessels on the fascial surface can be clearly seen. (b) A full view of a DSFC window $1\text{-day}$ postimplantation with a tumor cell aggregate (circled; murine fibrosarcoma). (c) The same tumor shown $6\text{days}$ later, nearly filling the window.

The DSFC model enables repeated acquisition of spatial and temporal data from a growing tumor over periods of days to weeks. Although there are also several disadvantages (tumors are essentially subcutaneous and there are size restrictions due to the width and diameter of the chamber window, which exert high tissue pressures), the model is widely applicable and has been used extensively in a range of different rodent models, enabling the use of syngeneic rodent tumor lines in immunocompetent strains, as well as human xenograft lines using immunocompromised strains such as nude and severe combined immunodeficiency (SCID) mice.

It should be noted that although the DSFC model is the most extensively used, there are alternative preparations suitable for IVM, the most similar of which is the cranial window, used to expose the brain³⁴ and mammary window.⁶¹ Other models often require externalization of an organ and thus sacrifice of the animal immediately on completion of image acquisition. Examples include the lung,^{62, 63} pancreas,⁶⁴ mammary fat pad,⁶⁵ and mesentery.⁶⁶

5.

Intravital Microscope

There are several technical problems to overcome and welfare issues to address when using live animals for microscopy, i.e., positioning and restraint of the mouse, physiological status of the mouse, objective lens properties (sufficient magnification, working distance, and numerical aperture), detector specification, analysis software capabilities, and so on. A standard microscope stage cannot be used, but needs to be specially modified to enable positioning of the mouse such that the window is locked in place. This generally involves the use of a specialized jig that restrains the mouse, with the window in the correct position to be bolted securely to the stage. Further, depending on the time frame of image acquisition, physiological parameters such as body temperature and blood pressure need to be monitored and maintained, necessitating the use of a heating blanket and/or heat lamp and temperature/blood pressure probes and recording equipment for anesthetized animals. Both general anesthesia and physical restraint can confound data interpretation due to alterations in blood pressure and vasoactivity. Isofluorane is commonly considered the most appropriate general anesthetic for small rodents,⁶⁷ although constant infusion of a short-lived IV injectable anesthetic such as Propofol can also maintain a suitable anesthetic depth for restraint.

Most IVM studies have to work around movement in the specimen during data acquisition due to pulse, respiration, peristalsis, muscle contraction etc. Without doubt the simplest and most widely used solution to this problem is real-time (video rate) data acquisition with a digital video (DV) camera and images recorded onto tape or other storage media. The major drawback with this approach is that DV cameras do not normally have sufficient sensitivity for use with many fluorescence applications although they are well-suited to transmitted light microscopy. Wide-field fluorescence and Nipkow disk-based confocal microscopes are normally used with cooled charge-coupled device (CCD) cameras, which are more sensitive. Recent further development of these has seen the introduction of back-illuminated (also called back-thinned) CCD cameras that have much higher quantum efficiency and wider spectral response. When coupled to electron multiplication (EM) technology, these detectors (EMCCD) can achieve $\sim 90\%$ quantum efficiency coupled with short-noise-limited at video rate $(25\text{to}30\text{frames}∕\mathrm{s})$ .⁶⁸ Point scanning laser confocal and multiphoton microscopes usually have photomultiplier tube (PMT) detectors. Although these microscopes are generally slow scanning and the detectors have lower quantum efficiencies than cooled CCDs, nonetheless these systems produce higher resolution images with better contrast than wide-field systems provided that the specimen can be kept still. Multiphoton microscopes do not require pinholes, so can benefit from the increased sensitivity of gallium arsenide phosphide (GAsP) detectors in a nondescanned (NDD) configuration. There are several recent reviews on CCD sensors,^{69, 70, 71, 72} and the books by Sluder and Wolf^{73, 74} present a good introduction to cameras for microscopy.

6.

Conventional Wide-Field Light Microscopy

6.1.

Background

Early IVM studies were confined to conventional wide-field light microscopy using transillumination (light transmitted from the condenser through the tissue sample to the objective) or epi-illumination (light reflected/emitted by the tissue sample through the objective). Basic fluorescent microscopy uses mercury or xenon arc lamps in conjunction with specific excitation and emission filters and a dichroic mirror, which splits the emission and excitation beams. Following exposure to light of an appropriate wavelength, a single photon absorbed by a fluorophore, either intrinsic or an extrinsic fluorescent dye/probe, enters an excited state. While excited, some energy is dissipated so when, after a few nanoseconds, the fluorescent molecule returns to its ground state, a photon of lower energy (and therefore longer wavelength) is emitted. The difference between the excitation and emission wavelength maxima is called the Stokes shift.

Conventional microscopy has been the mainstay of IVM, providing morphological information on the tumor vasculature [see Fig. 3 ] such as vessel length, diameter, and branching patterns, and qualitative information on the leakiness of blood vessels (hemorrhage), as well as video recordings of flowing blood (contrast provided by red cells under transillumination) and of vascular leak of exogenous fluorescent markers (epi-illumination). Some of the earlier observations made using IVM to study tumors growing in chamber models demonstrated notable architectural and blood flow aberrations that are now widely accepted to characterize tumor vasculature,^{75, 76} e.g., inconsistency in microvessel size, commonly with larger vessels located at the tumor periphery and smaller, more chaotic vessels toward the center. Dilated capillary diameters varying along the length of abnormally long capillaries were recorded, as well as intermittent or regurgitant flow and temporary or permanent microvessel occlusion by red cells.^{75, 76} These observations have been strengthened by subsequent refinements and application of modern image analysis techniques.^{27, 38, 77, 78, 79, 80, 81, 82}

Fig. 3

Images acquired using different microscopy techniques. (a) Transillumination images of murine fibrosarcomas expressing single VEGF isoforms (left is VEGF 120, and right is VEGF 188) growing in the DSFC of SCID mice. Obvious differences in the vascular architecture are apparent. (b) Multiphoton image of murine fibrosarcomas expressing single VEGF isoforms (left is VEGF 120, and right is VEGF 188) following IV administration of FITC-dextran. Second harmonic generation imaging shows collagen fibers (blue) between the vessels (green) in the VEGF 120 tumors. (c) Multiphoton 3-D images of a P22 rat sarcoma growing in the DSFC of BDIX rats following IV administration of FITC-dextran. Four sequential images are shown. The time postinjection is indicated next to each figure, demonstrating vascular leak over time. (d) Transillumination vascular image (left), with adjacent oxyhemaglobin saturation overlay derived from hyperspectral imaging (right) of vasculature in a nontumor-bearing DSFC of a SCID mouse (scale bar, $1\mathrm{mm}$ ). The right-hand axis depicts oxygen saturation from 0 to 100%.

7.

Confocal Microscopy

Despite the continued utility of conventional wide-field transmitted light and fluorescence microscopy in IVM, there are several problems contingent on these methods; even in a DSFC preparation, tumor growth results in an increasing depth of tissue through which light has to penetrate (several $100\mu \mathrm{m}$ ). Out-of-focus light scattered by the tissue interferes with and degrades the image at the focal point, severely limiting the depth of tissue that can be examined and confounding intensity measurements and interpretation of the acquired images.^{30, 103} Furthermore, it is only possible to examine the vasculature in 2-D using these approaches, which introduces errors in quantitative measures of vascular morphology and function. Traditionally, the only method of examining the tumor vascular network in 3-D was through reconstruction of a 3-D image following 2-D analysis of serial immunohistochemical sections, or through vascular casting.¹⁰⁴

To visualize a single optical section within the tissue, confocal laser scanning microscopy uses mirrors to scan a beam of laser light of the required fluorescence excitation wavelength point by point across the specimen. The emitted fluorescence signal is then “descanned” through a pinhole, preventing light from outside the focal plane degrading the image, thereby improving image resolution, signal-to-noise ratio, and allowing increased depth discrimination. As the microscope is focused through the specimen, a 3-D composite image is constructed.^{105, 106} However, the high light intensity of the beam, coupled with the need to repeatedly scan the specimen (either during the acquisition of a z-stack or time-lapse sequence), results in increased sample photobleaching and other light-induced tissue damage, which limits the usefulness of confocal fluorescence microscopy for IVM applications. Its use has been restricted primarily to postexcision analyses of fixed samples.^{107, 108} The ability of tumors to induce expression of VEGF in host stromal cells with fibroblast-like morphology was demonstrated using IVM confocal microscopy of murine mammary adenocarcinoma tumors (MCaIV) growing in VEGF-EGFP transgenic mice.¹⁰⁹ However, further analyses to determine spatial relationships was carried out using multiphoton microscopy techniques (see Sec. 8). One recent study utilized confocal reflected light microscopy in conjunction with multiphoton microscopy to develop a method of repeatedly and reproducibly locating the same region within a DSFC window.¹¹⁰ Reflected light confocal microscopy can be used to image microparticles, in this case latex polystyrene microspheres positioned in the DSFC preparation.

Basic confocal fluorescence microscopy can take considerable time, which makes it unsuitable for studies that require high spatial and temporal resolution, for example with a moving specimen. Spinning disk confocal microscopy uses a rotating Nipkow disk (with multiple pinholes each containing a microlens) to scan multiple beams of light across the specimen at the same time. The Nipkow disk can rotate as many as 1800 to 5000 times per minute, and a cooled CCD camera is normally used that allows the acquisition of up to approximately 1000 whole optical sections (frames or fields) per second,¹¹¹ thereby making it possible to view rapidly occurring events in real time.⁶⁸ The application of multiple lower intensity beams reduces photobleaching relative to confocal laser scanning microscopy; however, the Nipkow disks have pinholes of a fixed size and are therefore only optimized for use with one objective lens. The sensitivity and cost of the camera are other limiting factors. Nonetheless, this technique could prove invaluable in studies examining the movement of different cells and molecules during the dynamic process of tumor angiogenesis.

Another approach to real-time confocal IVM is video rate confocal laser scanning microscopy (VRCLSM), which has been applied to in vivo studies of kidney function.^{112, 113} VRCLSMs usually employ an acousto-optical deflector to scan the laser beam at high speed across the specimen. Early applications of this technique usually produced video signals which could be recorded to tape, although more recent VRCLSM systems tend to be modules that are added to conventional CLSM.

8.

Multiphoton Fluorescence Microscopy

Since 1990 when its potential as a biological imaging tool was clearly established,¹¹⁴ multiphoton microscopy has been growing in strength and has played a key role in many biological studies examining tumor angiogenesis.^{29, 38, 115} Multiphoton microscopy uses short pulses (approximately $100\text{femtoseconds}$ ) of high intensity infrared (IR) or near-infrared (NIR) light to induce multiple photon excitation.^{105, 114, 116} For fluorescence to occur, a fluorophore must first absorb a photon of light that has energy appropriate to induce an excited state. This usually happens with the arrival of a single photon, but can also occur if two photons with half the energy (or three with one third…etc.) are absorbed almost simultaneously. Since it is extremely unlikely that a significant number of lower energy (longer wavelength) photons will arrive almost simultaneously by chance, excitation of the fluorophore will only ever occur in the presence of an extremely high density of photons, found only when the beam of a powerful laser is at its most concentrated, right at the point of focus. The beam is scanned over the specimen as before but, unlike a confocal laser scanning microscope, no pinhole is required to exclude out-of-focus light, since there is no excitation of the fluorophore and therefore no photobleaching above and below the plane of focus.^{30, 105}

Multiphoton microscopy offers several advantages over confocal microscopy for live-tissue imaging. The low linear absorption of IR and NIR light by most tissue samples increases the depth penetration possible from the approximately $40\text{to}100\mu \mathrm{m}$ achievable using confocal microscopy to several hundred micrometers (absolute depth is dependent on the tissue properties, wavelengths used, and working distance of the objective lens). Further, as fluorophore excitation occurs exclusively within the focal plane, extensive photobleaching and tissue phototoxicity are avoided. ^{30, 105, 114, 117, 118}

Multiphoton fluorescence microscopy allows the excitation and visualization of fluorescent molecules, therefore unlabeled tissues are only visible when they exhibit autofluorescence (which is generally less in evidence than with single photon imaging techniques). To see weak fluorescence, a low autofluorescence background is desirable, but it can then be difficult to localize and understand the relationship between the fluorescently labeled and unlabeled tissues. It is often not possible or worthwhile to use transmitted light because of the depth of penetration in tissue or opacity of the specimen under examination, so reflection microscopy must be employed. In addition to conventional reflected light imaging, a process called second harmonic generation [SHG; see Fig. 3] can be used to view noncentrosymmetric structures such as collagen.^{118, 119} SHG is the result of multiple photons interacting with structures such as collagen, and, as there is no loss, the energy of two photons is combined to emit twice the energy and thus half the wavelength.¹¹⁸ Put simply, two photons are reflected as one photon with half the wavelength. Although not widely applied to IVM studies for examining tumor angiogenesis, SHG has been used in combination with multiphoton imaging of fluorescently labeled molecules to provide 3-D data on the spatial distribution of vessels and collagen in human melanoma tumors (Mu89),¹¹⁹ and there is scope for its broader use.

Multiphoton microscopy is widely applicable to tumor angiogenesis studies using IVM and has been used to build on and improve existing knowledge. Multiphoton microscopy analyses of murine mammary and human colon adenocarcinoma tumors (MCaIV and LS174T, respectively) treated with a neutralizing antibody against VEGFR2 (DC101) clearly demonstrated therapy-induced normalization of the tumor vasculature.¹⁰¹ This phenomenon was first documented using conventional fluorescence microscopy in hormone-dependent tumors.¹²⁰ Brown ¹⁰⁹ used multiphoton fluorescence microscopy to expand on a previous study demonstrating tumor cell driven expression of VEGF in host stromal cells,¹²¹ and showed that these cells were located both at the tumor-host stroma boundary and also deep in the tumor tissue $\left(200\mu \mathrm{m}\right)$ , where they were often associated with tumor blood vessels. Accurate details on individual cell expression and spatial distribution relative to other cells and structures have only been made possible through the development of multiphoton microscopy techniques. The same study went on to demonstrate specific gene expression mediated tumor cell homing ( $\mathrm{Hif}{1}^{-∕-}$ mutant cells homed to hypoxic regions) to track delivery of a therapeutic agent, characterize tumor vascular architecture in 3-D, and examine functional features such as blood flow velocity and permeability.¹⁰⁹

Multiphoton microscopy is also useful for techniques requiring quantification of fluorescence intensity, since it overcomes some limitations of conventional fluorescence microscopy such as fluorescence scatter from outside the focal plane, for example in estimation of tumor vascular permeability [see Fig. 3]. Using multiphoton microscopy, the VDA, CA-4-P was found to rapidly increase vascular permeability to a 40-kD dextran in the P22 rat sarcoma tumor model,²⁹ and a high molecular weight isoform of VEGF (VEGF188) was found to be associated with development of tumor blood vessels with improved barrier function compared with lower molecular weight isoforms (VEGF164 and 120).³⁸ Further data on the possible molecular mechanisms behind tumor vascular maturity, and subsequent impact on treatment, was provided in a study using a wild-type human glioblastoma model (U87MG) and a variant stably transfected to overexpress PDGF-C.¹¹⁵ Multiphoton microscopy was used to examine vascular architecture and permeability (as a measure of function). PDGF-C was found to play an important role in glioblastoma vascular maturation, with the PDGF-C overexpressing tumors demonstrating a more normalized vascular morphology concurrent with decreased vascular leakiness. Consistent with improved vascular maturity, the PDGF-C overexpressing tumors also demonstrated resistance to treatment with DC101.¹¹⁵

9.

Hyperspectral Imaging

Hyperspectral imaging is a technique based on the absorbance of light across a spectrum of wavelengths by different tissues.¹²² The technique may be used with either transmitted light if the specimen is thin enough or reflected light. The primary application of this technique for in vivo imaging of tumor vasculature is to examine metabolic data, or more specifically vascular oxygen saturation. Oxy- and deoxyhemoglobin demonstrate differences in light absorption (both visual and NIR light can be used) in proportion with the level of oxygen saturation.^{122, 123, 124} Consequently, oxygen saturation within tumor vessels can be measured [see Fig. 3] in real time through the use of IVM in conjunction with an appropriate optical filtering system, such as an acousto-optic or liquid crystal tunable filter, which is fitted in front of the camera.^{123, 125} The potential of this technique to study oxygenation in tumor blood vessels has been clearly demonstrated in a study using the 4T1 mammary carcinoma tumor model. The tumor model was engineered to express GFP under the control of a HRE, as well as to constitutively express RFP. Thus, tumor growth and the onset of hypoxia could be monitored using fluorescent IVM. Through the simultaneous application of hyperspectral imaging, filtering the absorption bandwidth within the appropriate hyperspectral range, oxygen saturation “maps” showing the spatial distribution of oxyhemoglobin were produced.¹²⁵ Using this method, the same group has recently related fluctuations in tumor microvascular oxygenation to the location of arteriovenous shunts, a common anomaly in tumor vascular networks.⁸

10.

Optical Coherence Tomography

Optical coherence tomography (OCT) is a noninvasive optical imaging technique that can be used to generate high resolution 3-D images, at depth, alongside additional information on blood flow.^{126, 127, 128, 129} It shares several similarities to ultrasound imaging but is based on measuring the intensity and time delay of light, rather than sound, back-scattered by tissue. The use of a weakly focused optical beam is used to generate sectioned images over an axial range of several millimeters. Generation of the image is based on a fiber optic Michelson interferometer and low-coherence wavelength light source.¹³⁰ Briefly, light from a broad-band, single-mode source (e.g., a superluminescent diode) is sent through a beamsplitter; one beam, a reference path, is reflected from a mirror or optical delay, while the other passes down a fiber optic probe and is focused on the target specimen. Light reflected back down the reference and specimen paths is recombined, and when the distance traveled by both paths is within the coherence length of the light, interference fringes are produced. These fringes can then be analyzed to reveal changes in the intensity and back-scattered (echo) time delay produced by structures within the tissue, and a map of the tissue structure can be generated.¹³⁰ The use of light scatter to generate image contrast as opposed to fluorescence emission (e.g., confocal or multiphoton microscopy) or molecular absorption (e.g., hyperspectral microscopy) permits the use of much longer wavelengths. OCT typically operates between 800 and $1300\mathrm{nm}$ , where tissue scattering and absorption are minimized; hence, depth penetration can exceed $1\mathrm{mm}$ .

OCT is often used in conjunction with Doppler (DOCT) to examine microvascular blood flow, thereby generating high resolution images of the vascular network overlain with blood flow information.¹³⁰ OCT can provide useful information on the microstructure of the tumor vasculature without the need for fluorescent probes. Further, the application of DOCT and the generation of image analysis techniques such as speckle variance analysis¹²⁷ permits the acquisition of blood flow data that may be spatially attributed to the vascular map.

So far, OCT/DOCT-based technology has been applied primarily to normal tissues.¹³⁰ For instance, it has been used to observe photodynamic therapy (PDT)-induced vascular shut-down of normal blood vessels using IVM in the DSFC.¹²⁷ However, tumor studies are beginning to emerge. In 2008, Standish ¹³¹ used interstitial DOCT, where a DOCT probe was inserted into rat prostate tumors (Mat-Ly-Lu) growing subcutaneously to quantify tumor vascular response to PDT. In 2009, Vakoc ¹³² used a modified DOCT technique, optical frequency domain imaging (OFDI)^{133, 134} for IVM-based high-resolution, deep tissue imaging studies of tumor vascular networks and their response to targeted therapeutics. This extensive study examined mophological features of multiple tumor blood and lymphatic vascular networks at depths beyond $1\mathrm{mm}$ (as opposed to the approximately $400\text{-}\mu \mathrm{m}$ depth possible using multiphoton microscopy), extracting quantitative data pertaining to the structure and extent of the vascular networks through the postimaging application of an in-house automated 3-D vascular tracing and analysis algorithm. In addition to generating anatomical data on the vascular networks, which was found to correlate well with previously extracted multiphoton data, the technique was applied to tumor-bearing mice (murine mammary adenocarcinoma, MCaIV) treated with the VEGFR2 neutralizing antibody DC101, providing data on blood and lymphatic vascular changes in response to therapy. The data obtained were, again, complementary to previously acquired data using multiphoton microscopy; however, additional data were provided through the ability to image frequently ( $2\text{-}\mathrm{h}$ time points over a $48\text{-}\mathrm{h}$ period) without the use of exogenous markers. Thus, the authors were able to acquire vascular networks at depths over a wide tumor volume at multiple time points without any confounding factors such as leak or photobleaching of fluorescent dyes. The ability to generate temporal and spatial data of the lymphatic vascular network, simultaneous to acquisition of blood vessel networks, was particularly novel.¹³²

Thus, there is the potential for future tumor-based IVM studies using DOCT-based technologies to examine the vascular effects of various therapeutic approaches. Further, as has already been done in normal blood vessels, it could be used in combination with other optical imaging approaches such as hyperspectral imaging to examine multiple tumor vascular parameters.¹³⁵

11.

Image Analysis

Quantifying IVM data is an on-going challenge that needs to be undertaken to fully interpret the vast amounts of imaging data that can now be acquired. In addition to expensive commercial software packages, freeware such as ImageJ are available to researchers for this purpose. Alternatively, researchers often employ programming tools such as Matlab^™, Mathematica^™, and the similar freeware options Scilab and Octave, to develop their own algorithms for specialized analytical purposes. A common necessity for studies of tumor microcirculation is accurate identification of the vascular networks in software, so that morphological parameters can be measured, for example, or other features of the tumor microenvironment can be spatially related to the microvasculature. Barber ⁷⁸ proposed a semiautomatic tracing algorithm for tumor blood vessels based on a compact Hough transform and radial maps, where plasma was fluorescently labeled with intravenously administered FITC-dextran to highlight vessels under two-photon excitation. The software requires a high degree of manual tracing, but finds vessel centers and calculates vessel diameters automatically. An adaptive irregular cylinder was used to model vessels in the automated methodology proposed by Abdul-Karim, ⁸¹ again using fluorescent dextran to provide contrast in SCID mice bearing a murine mammary adenocarcinoma. An automatic seeding sequence was used to localize the cylinder model, which found the center as the maximum intensity and fitted the diameter with a median calculation. Another automated methodology to trace vessels identified by fluorescent dextran is presented in Tyrrell, ⁸² where the spatial drop in intensity of the vessels is modeled as super-Gaussian functions. The modeling was similar to Abdul-Karim, ⁸¹ but instead of a cylinder, the shape model was generalized to a super-Gaussian function of the intensity, which benefits from stronger fitting in the presence of noise. Another approach, developed for computed tomography but with clear application to IVM, is based on the intensity spatial variation found through the derivatives in the directions of rows and columns (the eigenvalues of the Hessian matrix) for the discrimination of structures that resemble vessels; the diameters are then calculated from the maximum direction of change of intensity near the vessels.¹³⁶

Data from the different optical imaging acquisition technologies described are acquired in the same format; an intensity value $i$ (which is dependent on the number of photons detected, which in turn depends on the presence of a certain structure or concentration of a certain substance on the sample) is assigned to each element of a 5-D matrix or a subset of it. The matrix $\mathbf{i}(x,y,z,\lambda ,t)$ describes the intensity as a function of the $x,y,z$ spatial coordinates, $\lambda$ is the wavelength or frequency and time is $t$ . The spectral decomposition can be obtained with a multispectral camera or simply with a color camera with filters for the red, green, and blue channels.

IVM is being developed in parallel with a number of other sophisticated imaging technologies such as magnetic resonance imaging (MRI) and positron emission tomography (PET). Therefore, an immediate challenge is to correlate data that are derived from an individual subject with different acquisition modalities [generating separate matrices ${\mathbf{i}}^{*}(x,y,z,t,\dots )$ ] to provide a detailed and spatially resolved profile of vascular parameters, drug distributions, etc. DSFCs have been modified to be used with a MRI scanner¹³⁷ to compare parameters such as tumor area, vascular area fraction, and length of large vessels with initial extraction fraction and perfusion per unit tumor volume of the MRI contrast agent gadopentetate dimeglumine (Gd-DTPA). In Sefcik, ¹³⁸ IVM, microcomputed tomography, and histology were used to assess microvascular remodeling and associated bone healing in vivo in response to the release of the bioactive phospholipid sphingosine 1-phosphate. The potential that may be achieved by combining IVM with genome-wide “-omic” approaches is discussed in Megason and Fraser.¹³⁹ The matrix $\mathbf{i}(x,y,z,t,g)$ is presented as the description of hypothetical subcellular localization patterns for all proteins in the genome, at all developmental times and places, in 3-D within an organism.

While images from IVM can be acquired with high resolutions in a short time, other techniques may require longer times for lower resolutions; therefore it is important to consider the spatiotemporal differences of the datasets while correlating and interpreting them. Another important issue arises from the actual size of the datasets. 5-D sets can easily become too large and complex to be stored, transported, and processed with the current computational resources. A final challenge will be to feed the data extracted from IVM into the mathematical models of tumor angiogenesis that have been developed recently.¹⁴⁰ The assumptions and parameter settings that make up these models need to be tested with experimental data before the models can be used to test hypotheses for therapeutically targeting the tumor microcirculation.

12.

Summary

It is clear that IVM approaches have played, and continue to play, a vital role in studies designed to enhance our understanding of tumor vasculature—the molecular mechanisms involved in its development, the multiple structural and functional anomalies that define, it and the changes wrought by various therapeutic approaches. However, there is still much to learn, particularly with regard to molecular control of angiogenesis and acquisition of fully quantifiable data on vascular function, with sufficient spatial and numerical accuracy to allow detection of subtle changes with therapy. To address these gaps in our knowledge, it is necessary to both fully exploit and continue to build on the multiple and fast-moving technological advances in the field of optical imaging, from image and video capture to data analysis. For instance, the full potential of the more recently developed molecular probes has not been realized in IVM studies of tumor vasculature. The expense of modern microscope systems, such as multiphoton fluorescence, has also limited their application in this field. Nevertheless, the availability of these novel tools, combined with the availability of powerful computers for data analysis and storage, guarantees a place for IVM in future investigations of tumor microcirculation. These studies will not be confined to the preclinical arena. Deep-seated human tumors will be accessible to IVM via microendoscopy, and simultaneous application of IVM with other imaging modalities in clinical use, such as MRI and PET, provides an opportunity to maximize the advantages and overcome the disadvantages of each modality. In conclusion, we expect that IVM will continue to play a valuable role in research into tumor microcirculation and make a valuable contribution to the development of cancer treatments designed to target tumor vasculature.