|Home | About | Journals | Submit | Contact Us | Français|
The use of multiphoton microscopy for imaging mouse brain in vivo offers several advantages and poses several challenges. This tutorial begins by briefly comparing multiphoton microscopy with other imaging modalities used to visualize the brain and its activity. Next, an overview of the techniques for introducing fluorescence into whole animals to generate contrast for in vivo microscopy using two-photon excitation is presented. Two different schemes of surgically preparing mice for brain imaging with multiphoton microscopy are reviewed. Then, several issues and problems with in vivo microscopy - including motion artifact, respiratory and cardiac rhythms, maintenance of animal health, anesthesia, and the use of fiducial markers – are discussed. Finally, examples of how these techniques have been applied to visualize the cerebral vasculature and its response to hypercapnic stimulation are provided.
The ability to visualize the nervous system and its activity in vivo has – by nature – a significant advantage over studies using post-mortem and cultured tissue. In Vivo imaging allows the observation and perturbation of a functioning, intact system directly, rather than in less ethological model systems (Balaban and Hampshire 2001). Furthermore, an in vivo approach is necessary to measure blood flow and its regulation, as both blood and blood vessels are removed during cultured cell preparations, and as blood vessels are severed during slice preparations, rendering them as low resistance pathways subject to artificial flow parameters.
A variety of optical and nonoptical imaging modalities have been used for functional neuroimaging (Villringer and Dirnagl 1997). Examples of optical techniques include near infrared (IR) spectroscopy, intrinsic signals, laser Doppler flowmetry, confocal laser scanning microscopy, and multiphoton excitation laser scanning microscopy. Examples of nonoptical techniques include magnetic resonance imaging, positron emission tomography, and ultrasound. In general, optical techniques can offer the highest resolution but are restricted by the high absorption and scattering of light by the brain tissue. Among the optical techniques, multiphoton microscopy is best suited for in vivo studies due to the (a) quadratic dependence of optical absorption, such that under most circumstances optical sectioning is performed by the incident light, (b) use of IR light which scatters less than visible light and hence increases the depth of focal penetration, (c) pulsing of the IR light to optimize its penetration into the brain yet moderate its average power, and (d) use of a point-scanning system to optimize spatial resolution (Hecht 1992; Denk and Svoboda 1997).
There are five reasons for using mice. First, the cortical thickness in mice is thin as compared to in rats (~1.6 mm vs. 2.2 mm), and thus more of the cortical anatomy may be imaged in mice. Second, since the skull and dura mater are substantially thinner in mice as compared to rats, it is possible to use multiphoton microscopy to image directly through intact, thinned skull in mice. Third, the thin dura mater of the mouse does not impede an intracranial injection via a beveled glass pipette, which is used to introduce viral vectors containing genetically-encoded fluorescent contrast agents. Fourth, the bulk of mammalian transgenic animals are mice. Lastly, the thin dura mater of mice may also allow easier access of drugs into a brain via a cannula over a cortical window than would the thicker dura mater of rats.
Imaging was performed at two different locations, as summarized below. Fluorescein Isothiocyanate Dextran (FITC, MW 77K), Rhodamine B Isothiocyanate Dextran (RITC, MW 73K), Texas Red Dextran (Molecular Probes), yellow fluorescent protein (YFP), cyan fluorescent protein (CFP), and YG fluorescent microspheres (Polysciences) were all excited with the laser mode-locked at 800 +/-10 nm. Laser power was calibrated at 800 nm per objective lens per experiment. While imaging, input power was initially set near zero at the surface of the brain, and then increased in a conservative manner to obtain signal for mapping out the brain surface. The precision of depth estimation is limited by the curvature of the brain, an effect observed to be ~30 μ for NIH Swiss mice. Power was increased with depth of focal penetration (Oheim, Beaurepaire et al. 2001). Data analysis was performed using a variety of software packages, including Btrack, Kaleidageph, IDL, Confocal Assistant, ImageJ (Java-based NIH Image), and Adobe Photoshop. Labview was used for recording non-imaging data, including respiration signals and scan times.
The first set-up consisted of a commercial Bio-Rad MRC1024 MP. This system was equipped for both confocal and multi-photon (MP) imaging. Data acquisition was under software control (Lasersharp vs. 3.1 with time course). Light for one-photon excitation was provided by a Krypton-Argon gas laser, and this was used for viewing slides of brain sections that had been immunohistochemically labeled, as well as for initial comparisons with the MP in vivo imaging. Light for two-photon excitation was provided by a Spectra-Physics Tsunami Titanium-Sapphire laser (model 3955) pumped by 5W Millenia Argon laser (model P). The repetition rate at 800 nm was 82 MHz when pumped at 5W. Bio-Rad filter blocks TS1/T2A were used for MP imaging and T1/T2A were used for confocal imaging. MP signal collection was via the “green/red” filter set (for FITC, RITC, and YG) into external (non-descanned) photomultiplier tubes. It was important to include a customized adjustable beam collimator in the light path, so that the beam was (a) confined within the scan mirror dimensions and (b) slightly underfilling the back aperture for each objective lens. The water immersion objective lenses typically used were from Olympus (20X/NA0.5 UMPlanFL, 40X/NA0.8 LUMPlanFL/IR and 60X/NA0.9 LUMPlanFL).
The second set-up began as a Bio-Rad MRC 600 confocal equipped with an Argon gas laser and was converted into a custom two-photon system, as described elsewhere (Tsai et al. 2002). The set-up was used for imaging experiments during the various stages of this conversion, and all image acquisition was under software control (COMOS vs. 7). Multiphoton excitation was accomplished with the use of a Coherent Mira Titanium-Sapphire laser pumped by a 10 W Verdi Argon laser. The repetition rate at 800 nm was 76 MHz when pumped at 10 W. The excitation light path included a 600 DCLP dichroic. Emitted light was recorded by external photomultiplier tubes after passing through a laser blocking filter (BG39, which removed 95% of light >700 nm) and an appropriate emission filter (485 df22 for CFP and 535 df10 for YFP and FITC). For YFP and CFP, light < 600 nm collected from the specimen was split into two channels by a 505 DRLP dichroic. When recording respiratory signals, a BG40 blocking filter was added to the detection path in order to block light from a diode used to measure respiration. The water immersion objective lenses typically used were from Olympus (20X/NA0.5 UMPlanFL, 40X/NA0.8 LUMPlanFL) and Zeiss (40X/NA0.8 Achroplan, 63X/NA0.9 Achroplan).
Perhaps the easiest method of introducing fluorescent contrast into the brain is to use cerebral angiography, via intravenous injection of a fluorophore (Figure 1). The fluorophore should be conjugated to dextran (minimum MW 70,000) in order to prevent leakage of dye from the vasculature. Points to consider with this method include (a) the “blood dilution factor” due to the injection and (b) the quantity of fluorescent material that is injected. The “blood dilution factor” may be estimated, as the injected volume is a known quantity and the total blood volume (blood cells + plasma) in mice is ~8.5% of body weight; blood plasma alone is ~4% (Brafield and Llewellyn 1982). The total fluorescence may be estimated by verifying what dye-to-glucose substitution was utilized in the preparation of the dextran-conjugated dye (i.e. the moles of dye per mole of glucose used to conjugate the dye to the hydroxyl groups of dextran, or simply how many “dyes per dextran”) and the concentration of the dye solution that was injected. In general, a final concentration of 15-30 ×10-6 M of dye solution in blood is sufficient to yield signal that is well within the detection range without saturating, and presumably not interferring with the detection the physiological changes. Figures 2, ,3,3, ,44 show examples of the contrast obtained when angiography is combined with multiphoton microscopy. While angiography labels blood plasma, it is also possible to use fluorescent contrast to track blood plasma. This may be accomplished via an intravenous (IV) injection of small fluorescent microspheres that dilute within the blood stream and may – with fast acquisition rates – be observed as they streak through vessels (Figure 5). When calculating the amount of fluorescence for these injections, it is important to consider that typically only the outer 10% of each microsphere is fluorescent.
Whereas loading of fluorescent contrast into the brain’s vasculature is fairly simple, the loading of contrast into the brain cells themselves is comparatively challenging (Allport and Weissleder 2001). While preparations of cell cultures and tissue slices have benefited by the use of acetoxymethyl esters (to facilitate the introduction of fluorescent dyes) and biolistic loading with a “gene gun” (to facilitate the introduction of fluorescent protein genes), neither of these vehicles is practical for getting cells to fluoresce in intact, adult animals. Nevertheless, there are at least four possible approaches for labeling cells in vivo. The first of these, extracellular injection of dye, is the most parsimonious but the least effective. While some dye is endocytosed by cells, the labeling pattern is that a few saturated cells are present in a noisy background at the injection site, and autofluorescence is seen elsewhere. Another approach is intracellular injection of dye (Svoboda, Denk et al. 1997). This approach has been successful, although the number of cells that may injected during a single recording session is practically constrained. A third approach is to generate transgenic animals expressing fluorescent proteins (Zhuo, Sun et al. 1997; Nolte, Matyash et al. 2001). While this approach has been successful, it may suffer from unwanted biological effects due to constitutive expression of the transgene throughout development. A fourth approach, demonstrated here, involves the injection of viral vectors that genetically encode for fluorescent proteins (Griesbeck, Yoder et al. 2000). Mice were anesthetized with ketamine-xylazine (50 mg/kg and 15 mg/kg, I.P.) and placed into a stereotactic apparatus. A hole (bur size FG 1/2) was drilled in the skull over sensory cortex. An adenoviral vector for yellow cameleon 2.1 (i.e. genetically encoding for both cyan and yellow fluorescent proteins, under the control of the cytomagaloviral promoter) was loaded into beveled glass electrodes, and pressure injected (20 psi for 100 msec or 1320 nl virus with titer of 20,000 PFU/nl) intracranially at depths of 0.25 mm, 0.5 mm, and 1 mm. Following a minimum in vivo incubation and recovery time of two days, mice were urethane-anesthetized and cortical windows were prepared over sensory cortex. As seen in Figure 6, limited fluorescence is present at the injection site itself. Figures 7 and and88 show the cellular expression pattern of fluorescent proteins as observed in vivo through cortical windows. This “fluorescent loading” method results in the expression of genetically-encoded fluorescence in morphologically distinct cell types in the brain, as seen in Figure 9.
At the end of each imaging session, the mice were transcardially perfused with fixative and their brains were extracted for immunohistochemical processing. Since infected cells were of various morphologies, the injected tissue was immunolabeled in an effort to identify if neurons or astrocytes expressed the genetically encoded fluorescence. Astrocytes were labeled using an antibody specific to either glial fibrillary acidic protein (GFAP) or S100β, and neurons were labeled using an antibody that recognizes neurofilament protein (NF).
YFP was detected in the corpus callosum, ventricles, astrocytes near the cortical surface (glial limitans), cortex, and striatum (Figure 10). Expression in these areas was not always restricted to the local vicinity of the injected area; for example, YFP+ cells near the brain surface were observed as far as 1.8 mm anterior to the injection site; in the corpus callosum, YFP+ cells were observed as far as 1.25 mm posterior to the injection site. Expression of genetically-encoded fluorescence in the striatum and the cerebral cortex was typically restricted to within 225 μ of the injection tract. The majority (73.5%, n=34) of the YFP+ cells near the cortical surface labeled positively with an astrocyte marker, S100β. Examples of such cells are shown in Figure 11. While the remaining fluorescent cells (YFP+/S100β-) in this region were not identified histochemically, some may be leptomeningeal cells. YFP+/S100β+ cells were also observed in the cortex (52%, n= 23), striatum (31%, n=29) and corpus callosum (50%, n=16). Most of the YFP+/S100β- cells in the cortex and striatum were identified as neurons (NF+). The YFP+/S100β- cells in the corpus callosum were not identified histochemically; they appeared to be oligodendrocytes. The fluorescent protein in ventricles did not appear to be positive for S100β and was likely being expressed by ependymal cells. Different viral vectors and promoters may infect different cellular populations (e.g. Chen, Lendvai et al. 2000).
Brains were fixed with 4% paraformaldehyde and sliced coronally into 25 μ sections. Nonspecific aldehyde sites were blocked with PBS/glycine and sections were incubated in a general blocking buffer (PBS/glycine 3% normal goat serum, 1% BSA). This was diluted 1:3 to obtain a working buffer used throughout the remaining procedure. Overnight incubation in primary antibodies plus Triton X-100 (1%) at 4°C was followed by incubation in goat anti-rabbit Cy5 IgG (H + L, Jackson, 1:75) for 1 hr at room temperature. The following primary antibodies were used: Dako rabbit polyclonal to S100β (1:50), Sigma rabbit polyclonal to NF-200kd (1:100), and Sigma rabbit polyclonal to GFAP (1:100). Sections were washed thoroughly, mounted in antifade media, sealed, and viewed with a confocal microscope. Tissue from cameleon–expressing transgenic mice provided a positive control. Negative controls included uninjected but processed tissue, tissue processed without primary antibody, and unprocessed tissue (to determine autofluorescence at the relevant optical settings).
CD-1 NIH Swiss Mice (Charles River) were anesthetized with urethane (1-2 mg/g body weight, administered interperitoneally and as a split dose separated by 10 min) and placed into a small animal stereotaxic apparatus with a mouse adaptor (Kopf 926) and custom-built head supports. Their body temperature was maintained at 36.5°C by a homeothermic blanket system with flexible probe (Harvard Apparatus) that was adapted to fit a murine rectum. Puralube Vet Ointment was applied to the eyes to prevent dryness. Animals breathed oxygen gas saturated with water vapor. Hourly supplements of 0.1 ml physiological ringer (including 10mM glucose) were administered subcutaneously. The skin over the skull was reflected and the imaging region was marked. The skull was dried and the imaging region was prepared by one of the two methods described below.
This method, used previously with rats, was adapted for mice. A 2-3 μ2 window of skull was removed over the imaging region with a dental air drill and FG 1/4 bur (Midwest). Bone wax was applied as necessary to control bleeding from the skull’s vasculature. The dura mater – which, at least in this mouse strain, is of negligible thickness – was left intact and did not impede imaging. A stainless steel headframe (Figure 12) – used to stabilize the animal’s head to a holder attached to the optical bench – was cemented to the surrounding skull with dental acrylic (Lang). Moisture of exposed brain regions was maintained by a surgical gelatin sponge (Upjohn) that was saturated with artificial cerebral spinal fluid (ACSF). This sponge was removed just prior to sealing the exposed brain surface to a glass coverslip via an agarose plug (2% in ACSF). The glass coverslip was secured to the headframe. Imaging was performed primarily with water immersion objective lenses that sat (in water) at the surface of the coverslip. This method allows for the maximum focal penetration into neocortex that may presently be attained with optical techniques. It does not suffer from high light scattering by the skull, or unmatched refractive indices between the skull and aqueous solutions (e.g. ACSF, agarose, or ddH2O). Figure 13 shows a low magnification view of the cerebral vasculature as viewed through a cortical window. A caveat of this method is that the removal of skull renders the brain susceptible to pressure and temperature changes, which hastens the decline of the animals and thus reduces recording time.
In this method, the skull is thinned, but left intact and imaged through (Yoder and Kleinfeld 2002). A 2-3 μ2 area of skull over the imaging region was thinned to 200-250 μ with a dental air drill and FG 1/2 bur (Midwest). Superglue (Loctite 493 Instant Adhesive Superbonder 49350) – which scatters light significantly less than bone wax - was used to control bleeding from the skull. No bone wax was applied. A stainless steel headframe (Figure 12) was cemented to the unthinned skull with dental acrylic. This headframe was specifically designed to fit mouse skull; it has the optimal ratio of skull-to-metal needed to secure a firm attachment without having to build up “fake skull” using dental acrylic. It was secured to the imaging apparatus and used to stabilize the animal’s head position during imaging experiments. ACSF was applied over the imaging region and images were obtained using a water immersion objective lens that was placed directly in the ACSF over the skull. While this method adds precision to the image resolution observed during intrinsic optical imaging techniques (Masino, Kwon et al. 1993), it does possess two caveats. First, the focal depth attainable in cortex is reduced by at least the depth of the thinned skull. Second, the light scattering by (Figure 14), and different refractive index of, the skull somewhat compromises the image resolution. An asset of this method is that it avoids susceptibility to pressure and temperature changes, and enables longer recording periods and multiple imaging sessions (Christie, Bacskai et al. 2001).
One of the most common problems encountered during in vivo imaging is motion artifact. Motion effects may be grouped into two categories - external motion, where the brain is moved as a whole, and internal motion, where motion occurs within the brain itself. An obvious strategy for dealing with sources of motion external to the brain is to improve the stability of the interface between the head and the imaging apparatus, thereby removing or reducing the motion artifact during data acquisition. Failing this, corrections for motion during data analysis may be possible. If a fiducial marker was present (be it from autofluorescence or experimentally-introduced contrast), it might provide a means to track a region of interest (ROI) and reregister images. Generally, this requires that the range of motion for the ROI was fully captured within the dimensions (planar and 3D) recorded by the images. Motion internal to the brain is not readily alleviated, as it is often either part of a genuine physiological response or related to the “vasomotion” indicative of a living animal. Nevertheless, the impact of “physiological noise” may be minimized by collecting images at rates distinct from biological rhythms. Futhermore, signal fidelity may be improved by filtering the frequencies associated with respiratory and cardiac cycles.
Physiological monitoring of animals serves many purposes. It tracks the animal’s viability, to verify when the animal is healthy and to communicate in real time when it is not. Physiological monitoring can provide control data for systemic responses to experimental stimuli (such as the changes in breathing during hypercapnic stimulation). To the extent that respiration, heart rate, or blood pressure generate “physiological noise”, the imaging SNR may be improved by gating image acquisition to a physiological signal or by using the physiological data to filter the imaging signals during post-processing. Further, as anesthetized animals do not properly thermoregulate, this function may be provided by feedback from a temperature monitoring system.
The use of anesthetized animals for in vivo imaging reduces motion artifact and allows for invasive procedures. However, its use introduces some confounding effects that are not fully understood (Lahti, Ferris et al. 1998). Some neuronal activity is resilient during anesthesia, in that electrical responses to stimuli are observed. However, glial activity may be directly antagonized by some anesthetics (Finkbeiner 1992) and anesthetics exert wide-ranging effects on cerebral blood flow (Lindauer, Villringer et al. 1993). These matters should be considered when interpretting data collected on anesthetized animals.
Brain activation is evident in cerebral neurons, glia, and vasculature – although the glial responses have only been recorded in vitro and in situ (Verkhratsky, Orkand et al. 1998). The temporal parameters of neuronal, glial, and cerebrovascular responses are not fully characterized in terms of their durations and latencies, but the peak response times suggest a sequence in which neurons respond first (msec), succeeded by astrocyte responses (1-2 s), and then the hemodynamic response (6 s). Even less is known about the spatial parameters of neuronal, glial, and cerebrovascular responses, or the degree of their overlap. In some cases, neuronal receptive fields have been mapped electrophysiologically and identified as networks of synaptically connected cells. While activity within the astrocyte syncytium has been demonstrated, the glial receptive fields have yet to be identified (Giaume and McCarthy 1996). It has been suggested (Villringer and Dirnagl 1995) that a cerebrovascular receptive field may be defined as the portion of a vascular bed fed by a single penetrating arteriole (Figures 15 and and16),16), although in practice the dimensions of a cerebrovascular response may depend upon what attribute of the vascular response is recorded (Buxton 2002).
When attempting to determine the parameters of a cerebrovascular response, the question of resolution arises. Nonoptical neuroimaging techniques tend toward a “top down” approach, by measuring macrovascular-dominated signals over large areas of brain. Optical techniques have the spatial resolution to afford a “bottom up” approach. The first step is to consider if individual capillaries, the smallest vessels of the microvasculature, are reliably responsive. The cerebral capillary bed is a tortuous structure, marked by bifurcations and loops, as evidenced in its “spaghetti” appearance (Figures 15, ,16,16, ,17).17). This geometry suggests that the net direction of flow within a capillary segment need not be determined, as it is along arterial and venous routes. In order to directly examine flow in single capillary segments, Villringer and coworkers (1994) developed a technique that combined angiography with confocal laser scanning microscopy to watch the motion of blood cells within a capillary. Angiography brightly labels the blood plasma but not the blood cells. Thus, when a vessel’s diameter approaches the size of a blood cell – as is the case with capillaries – the blood cells themselves may be visualized as nonfluorescent “dark regions” and their motion may be recorded (Villringer, Them et al. 1994). This technique was adapted for multiphoton microscopy (Kleinfeld, Mitra et al. 1998). Indeed, the basal motion of blood cells through capillaries exhibits significant variability. Examples of this “biological noise” within the capillary bed include stalls of the blood cells and reversals of their flow direction (Figures 18 and and19)19) and the frequent of occurrence of microstrokes (Figures 20 and and21).21). Interestingly, while the basal flow is characterized by fluctuations in healthy tissue, it is possible that the variability may be reduced under pathological conditions (Figure 22).
Single-trial responses of capillaries following sensory stimulation are either not detected or are modest, perhaps due to the high baseline variability of the flow (Kleinfeld, Mitra et al. 1998). In an effort to step back and establish the scale for vascular changes, the vasculature was stimulated directly by inducing a hypercapnic state in the animal. This state was induced by replacing the O2 gas normally inhaled with CO2 (10% as regulated by a Matheson flowmeter). Identification of surface arteries and veins was confirmed with the use of dichroics: 470DF15 was used to view veins and 560DF10 was used to view arteries, in accordance with the published absorption spectra for hemoglobin and deoxyhemoglobin (Lemberg and Legge 1949). Respiration signals were amplified and filtered before being acquired by Labview at 200 Hz; these signals verified the delivery of CO2. Vessels were viewed through thinned skull using angiography. While venous changes were not readily apparent, consistent and significant changes in arteries (Figure 23) and capillaries (Figure 24 and and25)25) were observed.
This work was made possible by funding from NIH (MH 12420, NS 41096) and NSF (DBI 9604768). Ed Ballinger and I-Teh Hsieh provided assistance with the configuration and calibration of optics and optical recording devices. Dr. Jeff Squier and Dr. Andrew Millard provided valuable technical advice regarding the Spectra-Physics and Coherent systems, respectively. Dr. Suri Venkatachalam, Dr. James Prechtl, and Chris Joseph helped to establish the surgical facility, and Sean O’Connor assisted in the assembly of the gas delivery system. Dr. Oliver Griesbeck constructed and provided the adenoviral vectors. “Btrack” image analysis software was developed at the National Center for Microscopy and Imaging Research with the support of NIH (RR 04050).