The human brain is a structural engineering marvel where hundreds of thousands of miles worth of tightly packed axons and dendrites connect billions of neurons into a gigantic electrochemical network that generates memory, thought and action. Its main animal model, the mouse brain, is about three orders of magnitude smaller, but its anatomical structure is similarly dense and complex1.
Fluorescence microscopy is the method of choice for multiscale imaging of living brain tissue with high resolution. However, it typically relies on labeling sparse sets of brain cells, providing a fragmented and partial view of tissue anatomy. Electron microscopy and magnetic resonance imaging are practically label-free and unbiased approaches, yet they either offer high spatial resolution or non-invasive live imaging, but not both together. Moreover, they cannot be combined in situ with other powerful neuro-technologies, such as Ca2+ imaging, electrophysiology and optogenetics.
Getting a more detailed and broader view of brain tissue is not only important for mapping the functional connectivity of neuronal circuits2, it can also unearth useful information on anatomical context and tissue viability to assist in experiments and their interpretation.
Breaking this impasse, super-resolution shadow imaging (SUSHI) introduced a new paradigm to visualize the anatomy and extracellular space (ECS) of living brain tissue with nanometric spatial resolution3. It is based on fluorescence labeling of the interstitial fluid and super-resolution stimulated emission depletion (STED) microscopy4, 5, casting all membrane-enclosed cellular structures as sharply contoured ‘shadows’ against a bright background of extracellular fluorescence.
Imaging with inverted contrast is immune to bleaching because the interstitial fluid provides an inexhaustible reservoir of fresh dye molecules. It is potentially also much less invasive and harmful because the fluorophore does not need to be introduced into the cells and any phototoxic by-products do not accumulate inside of them, but can dissipate in the ECS. The most minute cellular and extracellular compartments (e.g., axon shafts, peri-synaptic astrocytic processes, spine necks, synaptic clefts) can be discerned in super-resolved shadow images3, 6, 7.
A recent study using the SUSHI approach and machine learning generated precise 3D reconstructions of brain tissue microstructures, even though the spatial resolution was limited to 140 nm8, suggesting that the neuroanatomical ground truth can be established by light microscopy when augmented by advanced computational image analysis.
Hampered by high technical demands and limited availability of super-resolution technology, SUSHI has not yet been widely adopted. Moreover, the technique has been used mostly in organotypic brain slices6, 7, which offer optimal imaging conditions, and not yet in acute brain slices or intact brains in vivo, where labeling and image contrast for shadow imaging is harder to come by.
The aim of this study was to mitigate these difficulties and make shadow imaging more versatile, outlining a straightforward (and adoptable) method for capturing more fully the complexity and dynamics of the anatomical structure of living brain tissue even without super-resolution microscopy.
To this end, we have worked out solutions for achieving innocuous, high-contrast fluorescence labeling of the interstitial fluid in acute brain slices and the intact brain in vivo. We show that the inverted signal can be read out by regular microscopy techniques, such as confocal, 2-photon and light sheet microscopy, providing fine-grained yet expansive views of the anatomical scenery in these major experimental preparations.
We started out by imaging organotypic hippocampal brain slices using an inverted confocal microscope (Fig. 1A). We chose the more popular Muller slices9, rather than Gähwiler10, for which the SUSHI technique had originally been developed. Muller slices are grown on a light-scattering membrane support, but by turning it upside down and placing a metal ring on top, it is possible to get a direct and stable view from below.
We used a 93X glycerol-immersion objective (NA 1.3) equipped with a motorized correction collar to reduce spherical aberrations caused by the mismatch in refractive index between medium (n ~ 1.46) and brain slice (n ~ 1.37). The spatial resolution of the microscope was around 212 nm in x-y and 550 nm in z (data not shown).
To generate morphological contrast, we added a small but membrane-impermeant organic fluorescent dye (Calcein, 100 mM, Fig. 1B) to the ACSF bath solution in which the brain slices were submerged, as described before3. After optimizing the correction collar, we could acquire images of the tissue with high contrast and resolution (Fig. 1C), offering a macroscopic perspective of the anatomical layout replete with crucial structural details, such as dendrites and axons.
Despite the high density of the fluorescent label, it was possible to acquire confocal shadow images (COSHI) and z stacks at least 50 mm below slice surface (SI Video 1A, B, C), before image quality decreased due to out-of-focus blur, aberrations and light scattering.
Given the diffusional replenishment of dye molecules, it was possible to acquire a high number (>100) of time-lapse frames (Fig. 1D; SI Video 2) with little or no drop in signal-to-noise ratio (SNR) or signs of phototoxicity (data not shown).
Next, we checked the utility of COSHI for studying microglia and their morphological dynamics in living brain tissue. Microglia are tissue-resident macrophages that are critical for the immune defense of the brain. They have highly branched and motile processes, which touch and seem to influence dendritic spines during brain development and neuroplasticity11, 12. However, it remains unclear, which other cellular structures they come into contact with in the surrounding neuropil.
To this end, we used organotypic brain slices from transgenic mice (CX3CR1-EGFP) where microglia are highlighted by GFP and labeled the ECS with a red dye (Alexa Fluor 594, 200 mM). Indeed, COSHI made it possible to reveal the complex arborization of microglial processes amidst their fully visible anatomical context (Fig. 1E and SI Video 3). Moreover, after inflicting a local laser lesion, which triggers a rapid and orchestrated immune reaction, it was possible to see how microglial processes navigate through the dense anatomical landscape towards the lesion site (Fig. 1F, SI Video 4). Thus, the shadow imaging approach paired with positive labeling may facilitate the study of cell growth and motility in various tissue micro-environments.
As a point-scanning technique, it typically takes several seconds to acquire a single confocal image, which is often too slow for imaging Ca2+ transients and other dynamic biochemical activities. In contrast, light-sheet microscopy reconciles high spatial with high temporal resolution because fluorescence excitation and detection are orthogonal to each other, enabling fast widefield imaging without out-of-focus blur (Fig. 2A)13.
To explore if shadow imaging is compatible with light-sheet microscopy, we used a custom-built microscope to image organotypic brain slices, where a thin and homogenous light sheet was created by a lattice excitation pattern and an oscillating mirror, as described previously14, 15. The spatial resolution of the system was around 273 nm in x-y and 524 nm in z (data not shown).
Because of the efficient signal detection scheme in light-sheet microscopy, it was possible to use much lower dye concentrations (Calcein, 20 mM, Fig. 2B). We could acquire high-contrast shadow images of the tissue, revealing fine details of its cellular microarchitecture (Fig. 2C), almost as well as COSHI but with >100X higher imaging speeds (Fig. 2D, SI Video 5 and 6). Thus, light-sheet shadow imaging (LISHI) of cellular structures can in principle be performed alongside high-speed imaging of biochemical dynamics such as synaptic release of glutamate and its spread in the ECS.
2-photon microscopy is the main technique used for imaging acute brain slices or in vivo, offering superior SNR and optical sectioning deep inside light scattering tissue (Fig. 2E)16. However, shadow imaging in these popular experimental preparations poses unique challenges for fluorescence labeling and image contrast. Inevitably, there are many dead and cut open cells on the surface of an acute slice, which will take up the fluorescent dye if it is bath-applied, diminishing image contrast between cellular compartments and ECS.
To circumvent this problem, we spritz-injected the dye inside of the brain slices (>50 mm below surface), where cells are mostly intact, via a pressurized patch pipette (Fig. 2F). Using a commercial 2-photon microscope with a 40X water-immersion objective (NA 1.0, spatial resolution x-y = 350 nm and z = 1.5 mm; data not shown), it was also possible to obtain shadow images with high contrast over large fields of view, temporarily lighting up the anatomical layout of the slices (Fig. 2G).
To increase and prolong fluorescence contrast, we used a dextran-conjugated dye (Alexa Fluor 488-Dextran, 500 mM), which slowed the dispersion of the dye in the ECS, making it possible to inject a relatively small volume (<1 mL) at low pressure with minimal tissue disturbance.
To demonstrate the ability of 2-photon shadow imaging (TUSHI) to reveal the anatomical context of a specific set of fluorescent cells, we spritzed the dye into acute brain slices prepared from mice implanted with YFP-labeled GBM tumor cells, which is a mouse model of glioblastoma of the mesenchymal subtype17, 18. With two fluorescence detection channels (for YFP and Alexa Fluor 488), it was possible to image the proliferating tumor cells and appreciate their spatial integration in the tissue (Fig. 2H). Of note, this is the first study where the shadow technique is applied to tumor tissue.
Taken together, the ‘spritz-shadow imaging’ technique, which is a variant of ‘shadow-patching’ for targeted electrophysiological recordings in vivo19, can on the fly enrich slice physiology studies with visual information on anatomical context.
Finally, we set out to extend the shadow imaging concept to the mouse brain in vivo to pave the way towards longitudinal neuroanatomical studies in mouse models of neuroplasticity and brain diseases.
We used a home-built 2-photon microscope with a 60X silicone oil objective (NA 1.3, WD 0.3 mm) that was equipped with a programmable spatial light modulator (SLM) to help reduce optical aberrations when imaging deep inside brain tissue (Fig. 3A). The spatial resolution of the microscope was around 320 nm in x-y and 925 nm in z (data not shown).
In addition to the usual challenges of in vivo imaging, such as brain motion and limited optical access, shadow imaging in vivo requires generating fluorescence contrast inside a large and sealed-off compartment. The labeling of the interstitial fluid should be strong, long-lasting and minimally invasive. Instead of injecting the dye directly into the tissue, which proves to be too disruptive and unreliable, we did stereotaxic injections into the ipsilateral ventricle controlled by a high-precision syringe pump, delivering the dye (~8 mL of 50 mM of Alexa Fluor 488) over an extended period of time (~10 min) (Fig. 3B). This amount theoretically corresponds to a dye concentration of a few millimolar in the interstitial fluid, if averaged over the entire mouse brain ECS. However, because of clearance, the effective concentration likely was much lower.
In this way, it was possible to achieve reproducible and strong brain-wide fluorescence ECS labeling that peaked shortly after the injection (<1 hour) but persisted at workable levels for at least four hours (Fig. 3C). The procedure was optimized for the animals to recover and live on normally.
Following standard protocols for craniotomy and window implantation, leaving the dura mater intact, we could create stable and clear optical access to the mouse cortex. An anesthesia protocol based on intraperitoneal injections of ketamine and xylazine, and careful physical positioning of the animal under the microscope using ear-bars for head fixation prevented almost all motion artifacts from inspiratory muscle contractions, enabling image acquisitions with minimal motion blur.
Applying these technical measures, we could acquire high-contrast 2-photon shadow images of the superficial layers of somatosensory cortex (Fig. 3D). The resolution and contrast of the images were lower for in vivo than for organotypic brain tissue, which can be explained by the use of longer wavelength light 2-photon excitation (l = 920 nm versus 488 nm), residual brain motion and presumably lower dye concentrations in vivo due to clearance from the interstitial fluid. Nevertheless, the TUSHI approach can reveal the outlines of neuronal cell bodies and blood vessels, where finer details like astrocytic endfeet, pericytes and perivascular spaces become readily visible.
Moreover, by taking z stacks of up to 100 mm in depth, it was possible to survey sizable volumes of brain tissue (Fig. 3E; SI Video 7). Because the dye is diffusible in the ECS, it was possible to acquire a high number of time-lapse images unaffected by bleaching (data not shown).
The volume fraction (VF) indicates the relative amount of ECS in the brain, estimated to be about a fifth of the entire brain. It is an important structural parameter of the tissue, which can be measured precisely by biophysical techniques but typically only over large tissue volumes20. By contrast, the VF can be determined directly anywhere in a field of view at micron scale by distinguishing ECS from cellular structures in shadow images simply by image binarization (Fig. 3F). We estimated the VF of cortical neuropil to be around 20-25% based on two different algorithms for image binarization (Fig. 3G). These values are much higher than what EM images of chemically fixed brain tissue suggest, but in line with the biophysical literature.
To confirm that TUSHI in vivo can be combined with imaging positively labeled cells, we again turned to the mouse model of glioblastoma17, 18. The approach showed YFP-labeled tumor cells infiltrating a cluster of neuronal cell bodies in motor cortex four weeks after their implantation into the mouse brain (Fig. 3H).
The technique may be valuable for studying various primary or secondary brain tumors in mouse models21, how tumor cells interact with tissue microenvironment and impact brain structures22. It also is uniquely suited for ex vivo tissue samples where positive labeling is not an option, such as fresh human pathology specimen from the clinics, as a diagnostic tool or for monitoring therapy. In principle, TUSHI could be combined with any other form of positive cell labeling, ie. circuit tracing, cell type reporters, or fluorescent biosensors to explore brain structure and function in view of the complete cellular context in living animals.
Taken together, our study presents a strong case for shadow imaging of living brain tissue using common diffraction-limited microscopy techniques. It can help reveal the anatomical context of labeled cellular players in a variety of patho-physiological settings, in addition to providing ECS structural information. In view of ongoing advances in ECS labeling, super-resolution microscopy and computational tools, shadow imaging is bound to become an indispensable tool for live-tissue microscopy.