In Vivo Multiphoton Microscopy of Deep Brain Tissue

Michael J. Levene, Daniel A. Dombeck, Karl A. Kasischke, Raymond P. Molloy, Watt W. Webb


Although fluorescence microscopy has proven to be one of the most powerful tools in biology, its application to the intact animal has been limited to imaging several hundred micrometers below the surface. The rest of the animal has eluded investigation at the microscopic level without excising tissue or performing extensive surgery. However, the ability to image with subcellular resolution in the intact animal enables a contextual setting that may be critical for understanding proper function. Clinical applications such as disease diagnosis and optical biopsy may benefit from minimally invasive in vivo approaches. Gradient index (GRIN) lenses with needle-like dimensions can transfer high-quality images many centimeters from the object plane. Here, we show that multiphoton microscopy through GRIN lenses enables minimally invasive, subcellular resolution several millimeters in the anesthetized, intact animal, and we present in vivo images of cortical layer V and hippocampus in the anesthetized Thy1-YFP line H mouse. Microangiographies from deep capillaries and blood vessels containing fluorescein-dextran and quantum dot-labeled serum in wild-type mouse brain are also demonstrated.


Noninvasive methods for imaging of deep structures in intact animals include MRI and PET (Jacobs and Cherry 2001). However, the resolution and speed of MRI is more than an order of magnitude lower than that of fluorescence microscopy and offers only a few functional contrast agents compared with the tremendous library of available fluorescent indicators. Although PET accesses a variety of molecular probes, its resolution is on the scale of millimeters and it suffers from slow image acquisition. Therefore the ability to perform minimally invasive deep in vivo fluorescence microscopy represents a breakthrough in intact-animal studies for biology and medicine.

Although routine use of multiphoton microscopy achieves imaging depths in brain of typically <500 μm, the ultimate imaging depth is dependent on many parameters, including the age of the animal and the optical geometry (Oheim et al. 2001). Oheim et al. (2001) have extended the reach of multiphoton microscopy by approximately 100 μm by appropriate use of a low-magnification, high numerical aperture objective. Theer et al. (2003) recently used regenerative amplification of 200-kHz pulses to achieve high pulse peak powers while maintaining reasonable average powers, enabling imaging depths of close to 1 mm, although image quality suffered significantly at greater depths. Gradient index (GRIN) lenses offer the opportunity to reach even deeper lying structures.

GRIN lenses use a negative gradient in the refractive index of glass from the center of the lens to the outside edge to bend and focus light. GRIN lenses are characterized by a length, or pitch, and a numerical aperture (NA). The pitch of a GRIN lens determines how many internal images are formed within the lens. A 0.25-pitch lens focuses a parallel beam incident on the front surface of the lens to a point on the back surface. A lens of pitch 1 forms an upright image on the back surface, with an internal, inverted image plane located at one-half the length of the lens. GRIN lenses are commercially available in lengths of up to several tens of centimeters, long enough, in principle, to access deep brain structures in large animals and humans.

GRIN lenses have been used for fiber bundle–coupled confocal microscopy (Knittel et al. 2001) and in vivo epi-fluorescence microendoscopy (Jung and Schnitzer 2002). The use of large (>1 mm diam) GRIN lenses for in vivo multiphoton microscopy (Levene et al. 2002) and the characterization of thin composite GRIN lenses for multiphoton microscopy (Jung and Schnitzer 2003) have recently been reported. Multiphoton microscopy (Denk et al. 1990) has several important advantages for in vivo imaging applications with GRIN lenses. It is capable of resolving optical sections relatively far from the surface of the lens, while tissue in close proximity to the lens may suffer from mechanical damage or immune response. The superior sectioning ability of multiphoton excitation is critical for observing detailed structure in tissues in which fluorescence sources are distributed throughout the region of interest. Multiphoton microscopy is able to excite intrinsic tissue fluorescence and UV absorbing dyes without the use of UV excitation, which has poor tissue penetration and produces a strong fluorescent background both from tissue regions outside the focal plane and from the high numerical aperture GRIN lenses themselves.


Microscope apparatus

The multiphoton microscope used here is the same as described elsewhere (Kloppenburg et al. 2000) except that, for yellow fluorescent protein (YFP) imaging, the bialkali photomultiplier tube (PMT) is replaced by a GaAsP PMT (H7422P, Hamamatsu, Bridgewater, NJ), which exhibits a fivefold greater sensitivity to yellow wavelengths. The Ti:Sapphire laser source produced ∼100-fs pulses at 80 MHz. The position of the GRIN lens relative to the objective lens is controlled via hydraulic micropositioners (Narishige Scientific Instruments Lab, Tokyo, Japan) fixed to the nosepiece of the microscope. The focus knob of the microscope moves the objective and GRIN lens together as a fixed unit to the appropriate depth within the specimen. A piezoelectric focus controller (Physik Instrumente, Waldbronn, Germany) adjusts the focus of the objective relative to the GRIN lens, allowing z-series of the sample while holding the position of the GRIN lens fixed.

Animal surgeries

Anesthesia and surgeries were performed in accordance with Cornell University–approved animal use protocols. The mice were maintained under ketamine (76 mg/kg)/xylazine (5 mg/kg) anesthesia during the surgery and the following imaging session. To reveal the cortex for GRIN rod penetration, we used a dental drill to create a circular craniotomy (5 mm diam) centered above the dorsal parietal cortex. The dura visible beneath was cut at the edge of the field and carefully removed. For analgesia at the site of the craniotomy, lidocaine (2%) was applied topically. The body temperature was maintained at ∼36° celsius using a feedback-controlled heating pad during both surgery and imaging. Prior to the microangiography, the blood serum was labeled by a tail-vein injection of a 100-μl bolus of either fluorescein-labeled dextran (20 mg/ml, 500 kDa, Molecular Probes) or quantum dots (35 μM, 608 nm emission, Quantum Dot, Hayward, CA) in physiological saline.

GRIN lens and imaging parameters

The use of GRIN lenses for in vivo multiphoton microscopy is shown in Fig. 1. One end of the GRIN lens is positioned close to the focal plane of the objective lens of a standard multiphoton laser scanning microscope, with the opposite end inserted inside the specimen under study. The GRIN lens refocuses the laser light tens of micrometers from the opposite end of the lens, inside the specimen, where the scanned beam excites fluorescence that is collected back through the GRIN lens and the objective. Dispersion compensation is not required because pulse dispersions in the GRIN lenses used here are ∼2,700 fs2, similar to typical high-NA objective lenses.

fig. 1.

Experimental arrangement for using gradient index (GRIN) lenses in conjunction with multiphoton microscopy. A hydraulic micropositioning system locates the GRIN lens relative to the objective lens. A piezoelectric focus control performs fine focusing without moving the GRIN lens inside the animal.

To image neurons and capillaries in mouse brain, we glued two custom-fabricated 0.6-NA GRIN lenses of pitch 0.22 to either end of a 0.1-NA GRIN lens of pitch 1 (NSG America, Somerset, NJ) using a UV curable optical adhesive (Norland Products, Cranbury, NJ) as shown in Fig. 2. The lenses were 350 μm in diameter, and the total length of the glued lens was ∼16 mm. All but the last 900 μm at either end of the composite GRIN lens was protected from mechanical damage by a metal sheath with an outside diameter of 600 μm. While high NA end pieces determine the resolution of the composite lens, it was necessary to use lower NA material for the central length of lens to avoid focusing the excitation laser pulse inside the high NA material, which can produce a fluorescent background and may result in nonlinear optical effects such as self-phase modulation. This choice, however, reduces the total field of view to a 58-μm-diam circle. Other configurations may be used when larger fields of view are required. The pitch of the 0.6-NA material was chosen to give a focal distance of 35 μm in air and 47 μm in water. Changing the position of the focal plane of the objective results in a change in the working distance (WD) from the end of the GRIN lens in the sample. This allows z-scanning in the sample, from 0 to 95 μm from the surface of the lens, without moving the GRIN lens.

fig. 2.

Composite GRIN lens used for in vivo neural imaging. A: schematic of composite lens with path of excitation light shown in red. B: image of top portion of composite lens next to a penny.

Coupling to the GRIN lens was via a Zeiss Fluar 20×, 0.75 NA air immersion objective (Carl Zeiss, Thornwood, NY). The NA-mismatch between the coupling objective and the GRIN lens resulted in a 59% coupling efficiency. The transmission efficiency of the GRIN lens was ∼66%, resulting in an overall excitation efficiency of ∼40%. The full width half-maximum (FWHM) of the lateral point spread function (PSF), measured using sub-resolution fluorescent particles, was 825 nm, less than twice the diffraction limited value of 481 nm and similar to previous reports (Jung and Schnitzer 2003). The axial PSF was estimated to have a FWHM of 15 μm by measuring the fluorescence as a function of depth of penetration into a fluorescent plastic slide. The large axial PSF is likely due to spherical aberration.

In all cases, the GRIN lens was inserted approximately perpendicularly to the cortical surface through the cranial opening in steps of a few tens of micrometers. Pausing between steps allowed tissue to physically readjust to the presence of the lens. It has been suggested that rapid penetration by blunt tips in neural tissue minimizes tissue damage, while slow penetration minimizes tissue compression (Edell et al. 1992). Pausing between rapid steps should therefore minimize both tissue damage and compression. Care was taken to avoid inserting the lens at locations with considerable blood from surgery on the cortical surface. Once inserted, most trials produced useful images, with little, if any, apparent accumulation of blood on the lens surface during or after penetration.

For the images of blood vessels, the excitation wavelength was 830 nm in Fig. 3, A–C, and 780 nm in Fig. 3D, with ∼50 mW of power at the sample. Figure 3, A–C, shows 1-s scans taken at a WD of 50 μm. Figure 3D is a projection of five images taken with WDs of 13–40 μm from the end of the GRIN lens. Each image in the stack is the average of two scans lasting 3 s each.

fig. 3.

In vivo microangiography in wild-type mouse. A: blood vessels containing fluorescent quantum dots ∼800 μm below the surface of cortex. B: line scan of small capillary in A as indicated by dotted red line. C: zoom of section in B; Δx is the spatial dimension of the line scan, Δt is the time dimension, and the hypotenuse is the velocity. D: capillaries containing fluorescein-dextran more than 2 mm below the surface of cortex. Scale bars are 10 μm.

For the images of neurons in Fig. 4, A–E, the excitation wavelength was 920 nm with 20 mW of power at the sample. All scans were 3 s, with no averaging. The WDs were 43 and 42 μm for 4A and 4B, respectively. Figure 4C is a projection of four images with WDs from 46 to 62 μm. Figure 4, D and E, had WDs of 24 and 60 μm, respectively.

fig. 4.

In vivo images of brain in anesthetized in THY1-YFP line H mice. A–C: images of yellow fluorescent protein (YFP)-containing cell bodies of layer V neurons 700–800 μm below the surface of cortex. D: image of axon bundles in the external capsule ∼1 mm below the surface of cortex. E: image of neuropil in hippocampal region CA1, ∼1.5 mm below the surface of cortex. Scale bars are 10 μm.


Microangiographies of capillaries and larger blood vessels are shown in Fig. 3. Figure 3A shows a capillary along with larger blood vessels ∼800 μm below the surface of wild-type mouse cortex after administration of fluorescent quantum dots by tail vein injection. Fluorescent quantum dots have high two-photon cross-sections and are excellent indicators for in vivo angiography (Larson et al. 2003). Individual blood cells do not take up the quantum dots and are apparent as dark spots in the capillary. Figure 3B shows a line scan as indicated by the horizontal red line in Fig. 3A. The horizontal coordinate in Fig. 3B represents space along the line, and the vertical coordinate represents time as successive scans pass the same region. The dark streaks are from the passage of blood cells and can be used to estimate the velocity of blood flow, as indicated in the portion of Fig. 3B shown in larger scale in Fig. 3C. Taking into account the angle of the capillary with respect to the line scan, the blood velocity was estimated to be ∼0.6 mm/s, similar to the velocity measured in rat cortex (Chaigneau et al. 2003; Kleinfeld et al. 1998). Figure 3D shows blood vessels more than 2 mm below the surface of cortex in wild-type mouse after administration of fluorescein-dextran by tail vein injection. An ∼5-μm-diam capillary appears on the left, while a larger, ∼15-μm-diam blood vessel is on the right. Although blood flow was observed, the orientation of the capillary was not conducive to velocity estimates using the line scan technique. Previous attempts at fluorescence microangiography in brain were limited to vessels <600 μm below the surface of cortex (Kleinfeld et al. 1998). The ability to perform deep in vivo microangiography in brain and other tissues provides a new approach for visualization and chronic studies of microangiopathies, which are typical complications of diseases such as diabetes, hypertension, and Alzheimer's disease.

THY1-YFP line H mice (Jackson Laboratory, Bar Harbor, ME) express YFP in cortical layer V neurons and in the hippocampus (Feng et al. 2000), making them ideal candidates for deep in vivo imaging. Figure 4, A–C, shows layer V neuron cell bodies 700–800 μm below the surface of the cortex and represents the typical range of image quality in cortex for these lenses. Axonal bundles in the subcortical white matter ∼1 mm below the surface were also visible (Fig. 4D). Figure 4E is an image of neuropil from the CA1 region of the hippocampus ∼1.5 mm below the surface of cortex. Figure 4, A–D, demonstrates subcellular resolution, with clearly visible individual neuronal processes.

There are three primary aspects of possible tissue trauma due to GRIN lens penetration in solid organs such as the brain. First, the inevitable destruction of tissue along the penetration path with possible bleeding, tissue edema, and elevated intracranial pressure can be potential complications. Second, the compression of tissue of interest at the front end of the lens, and third, the subsequent formation of scar tissue and inflammation must also be considered.

The maximal thickness of the penetration path of 600 μm is well within the range of established acute and chronic brain implants such as push-pull cannulae (Philippu 1984). By avoiding major veins along the perforation path, it was possible to achieve penetration such that bleeding was not detected at the front surface of the lens. However, cleavage of blood vessels along the path of penetration is inevitable and must be taken into account when interpreting the results of angiography.

Parietal cortex tolerated the implantation of the GRIN lens without clinical signs of elevated intracranial pressure. However, regions with closer proximity to the brain stem (occipital cortex, cerebellum) may require caution to avoid transtentorial herniation. The penetration depth of multiphoton microscopy allows imaging ∼60 μm from the end of the GRIN lens, where the inevitable tissue compression is substantially alleviated. Accordingly, we did not observe gross morphological changes indicating compression in our images. Although blood flow velocity can vary greatly even within the same region of cortex (Chaigneau et al. 2003), our measured velocity of 0.6 mm/s was within the expected range of values and is therefore consistent with a lack of tissue compression. The accumulation of compressed tissue at the front of the GRIN lens during insertion was observed to be limited to ∼10 μm and did not interfere significantly with image acquisition.

It is likely that chronic implantation of GRIN lenses may induce gliosis and inflammation; however, studies of gliosis in response to electrode penetrations found responses limited to within the WD of the GRIN lenses (Edell et al. 1992; Turner et al. 1999) and independent of probe dimensions (Szarowski et al. 2003). The feasibility of multiphoton microscopy in conjunction with cranial windows has been consistently demonstrated in chronic imaging studies (Bacskai et al. 2001; Christie et al. 2001).

In addition to choosing lenses of the smallest practicable diameter, the impact of tissue damage in the brain may be minimized by working in larger animals and by using the lenses to image in previously inaccessible sulci without actually penetrating brain tissue. Use of intrinsic signals (Dombeck et al. 2003; Zipfel et al. 2003) or alternative labeling methods, such as calcium indicators (Svoboda et al. 1997) or dyes that stain β-amyloid plaques (Christie et al. 2001), with this deep imaging technique could enable many new physiological experiments both in basic research and in animal models of neurodegenerative diseases. This imaging modality may also be easily combined with conventional in vivo electrophysiological techniques.

GRIN lenses used in conjunction with multiphoton microscopy open up the possibility of performing deep in vivo fluorescence imaging in brain. We have demonstrated the compatibility of GRIN lenses with in vivo imaging. Potential complications such as motion artifact and the accumulation of damaged tissue at the front of the lens during penetration have been shown to be negligible. Scanned fiber optic coupling to long lenses may eventually enable work in large animals and in awake behaving animals. Considerable progress in fiber optic scanning for in vivo imaging of awake behaving animals has already been made by Helmchen and co-workers (Helmchen et al. 2001; Ouzounov et al. 2002). We anticipate that in vivo multiphoton imaging with GRIN lenses will prove to be a valuable tool for both biological research and clinical applications.


We thank NSG America for their cooperation and for custom GRIN lens fabrication and K. Hodgson for Fig. 1. We also thank Quantum Dot for the quantum dots.


This work was supported by the facilities of the National Institutes of Health (NIH) Biomedical Resource funded by the National Center for Research Resources-NIH/National Institute of Biomedical Imaging and Bioengineering. Additional support was received from the U.S. Department of Energy, the National Science Foundation, and from an NIH Training Grant in Molecular Biophysics.


  • The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.


View Abstract