Rapid, label-free identification of cerebellar structures using multiphoton microscopy

2.1 Rat cerebellar specimen preparation

Five 8-week-old experimental Sprague-Dawley rats (250–300 g) were killed with cervical dislocation and decapitated. The whole cerebellum was obtained, embedded in OCT compound, and kept frozen at −20 °C until needed. Consecutive coronal and sagittal sections (10-μm thickness) were obtained with a Leica CM3050 freezing microtome. To keep the tissue structurally intact, we also prepared paraffin sections from formalin-fixed and paraffin-embedded cerebellar tissue. During the imaging process, the unstained specimens were perfused with a small amount of phosphate-buffered saline (PBS) to avoid tissue shrinkage. Immediately afterward, a cover slip was placed on the PBS-moistened specimens for imaging. For histological comparison, three consecutive sections were directly cut by cryostat microtome without standard histology procedures. Two of them were used for multiphoton microscopic imaging and the middle section was subjected to standard hematoxylin and eosin (H&E) staining to histological comparison to the multiphoton microscopic results. All experimental preparations were performed in accordance with the animal welfare guidelines of the First Affiliated Hospital of Fujian Medical University, China.

2.2 Human cerebellar specimen preparation

A total of seven human cerebellar hemangioblastomas specimens were obtained from the Department of Neurosurgery at the First Affiliated Hospital of Fujian Medical University. Patients were written informed consent for obtaining prior to study participation and approved the Fujian Medical University Clinical Research Screening Committee for Studies Involving Human Subjects. During preparation process, a portion of the tissue for routine diagnosis was transferred to the pathology laboratory with ice. Tissue section and MPM imaging was treated as described above. The pathological diagnoses of cerebellar tumor specimens were reaffirmed independently by two experienced neuropathologists.

2.3 Multiphoton microscopy

Multiphoton microscopic imaging system used in this experiment has previously been described in detail 22. Imaging was performed with high-throughput scanning inverted Axiovert 200 microscope (Zeiss LSM 510 META, Jena, Germany). The laser system used a mode-locked femtosecond Ti:Sapphire laser (Coherent Mira 900-F, Coherent, Inc., Santa Clara, CA, USA), tunable from 700 to 980 nm (Verdi-10; Coherent, Inc., Santa Clara, CA, USA). A Plan-Apochromat oil immersion objective (63×, NA=1.4, Zeiss) was employed to focus the excitation beam and collect the backwards signals. All images were obtained with ∼6 mW (at 810 nm excitation) average power at the objective. The META detector consists of an optimized 32-channel photomultiplier tube (PMT) array and high quality reflective grating to detect all signals in backscattered geometry.

Our experiments used two independent channels of the META detector to capture the TPEF and SHG signals from the specimen. One channel detected TPEF signals at 430 to 716 nm (color-coded red); whereas another channel detected SHG signals at 387 to 419 nm (color-coded green). The different independent channels have independent optimal parameters to adjust the image contrast and brightness, such as laser intensity, detection gain, amplifier offset, and amplifier gain. An optional HRZ 200 fine-focusing stage (HRZ 200 stage; Carl Zeiss, Jena, Germany) and automated tiling functions procedure were used to achieve large-scale images. The image acquisition time is 1.97 sec/frame with a resolution of 512 pixels×512 pixels. Acquisition times for the large-scale images are 27 min 3 sec (Figure 1), 24 sec (Figure 2), 18 sec (Figure 3), 2 min 6 sec (Figure 4), and 3 min 42 sec (Figure 5), respectively. Acquisition, reconstruction, and analysis of total images were acquired within 35 min. All images had a 12-bit pixel depth in our system.

2.4 Image analysis methods

3.1 Visualizing normal rat cerebellar structures

We start with the imaging of normal, intact, unstained tissue sections from the rat cerebellum. The large-scale mosaic MPM images of the entire cerebellum in Figure 1 A show a representative coronal section, containing the vermis, hemispheres, medulla, and ventricle, demonstrating the ability of MPM to detect large-scale histoarchitectural features. Meanwhile, magnified cerebellar lobules in both coronal and sagittal views can also display a tightly folded cortex, a central core of white matter, fissure, and the pia mater, which covers the cerebellum (Figure 1C and E). We effectively observe the striking contrasts between highly cellular regions within the gray matter and tightly packed fiber tracts within the white matter. In cellular regions, we also clearly identify the three-layered cerebellar cortex, namely, the granular layer (GL), Purkinje cell layer (PL), as well as molecular layer (ML).

At higher magnification, increasing detailed information about various cerebellar structures is shown in Figure 2. Linear, myelinated axon bundles and single myelinated axon fibers (white arrows) in white matter are clearly identified. They appear to be color-coded yellow because the uniform polarity microtubule arrays inside myelinated axons could both emit comparable SHG and TPEF signals (Figure 2 A) 25. The magnified granular layer, Purkinje cell layer, and molecular layer show visible clarity in Figure 2C, E and G. The cell nuclei are lacking fluorescent signals and appear as dark spots. By contrast, the cellular outline can clearly identify by the strong TPEF signals of NAD(P)H and FAD in the cytoplasm 26. In Figure 2C, densely packed small granule cells and Golgi cells appear as oval or round structures, which concentrate in cytoplasm-rich inclusions in the granular layer. Figure 2E displays a single row of Purkinje cells with large cell bodies (circles) and numerous fluorescent granules encircling neuronal nuclei in the Purkinje cell layer. However, glial cells and basket cells of the molecular layer present no fluorescent granules in the cytoplasm (Figure 2G). These results indicate that the cells in the cerebellar cortex have a highly regular arrangement and distinct composition of the cytoplasmic content. On the other hand, capillaries and the pia mater of the cerebellum could be recognized by the SHG signal of collagen fibers (pink arrows in Figure 2I and pink arrowheads in Figure 2 K) 26, which can be identified by the cerebellar lobules outlined and depicted in the collagen deposition. Thus, MPM can offer detailed information to visualize the normal cerebellar structures. The tissue architecture details are in perfect agreement with the corresponding H&E-stained images (Figure 1B, D, and F, Figure 2B, D, F, H, J, and L).

3.2 Automated differentiation of gray and white matter

The gray and white matter regions of the cerebellum can be used to divide the layers of the cerebellum. Differentiating between these two regions could aid the neurosurgeon in improving surgical precision. Thus, we obtained representative TPEF/SHG images of the gray-white matter junction in the cerebellar lobule (Figure 3). The overlaid image clearly display that large amounts of axons condense into thread-like axon bundles. Numerous white matter tracts were identified using the TPEF signal and comparable SHG signal, presenting a yellowish color (position 4 in Figure 3C). While the adjacent cortex gray matter included a molecular layer (position 1 in Figure 3C), the Purkinje cell layer (position 2 in Figure 3C) and granular layer (position 3 in Figure 3C) appeared red in color due to the strong TPEF signal. Molecularly, white matter and gray matter have distinct differences in composition. Thus, these distinct contrasts in MPM images provide an effective means to differentiate between the white matter and surrounding gray matter.

To further evaluate the contrasts, we next applied a program (See Materials and Methods 2.4.1) for the automated differentiation of gray and white matter in MPM images. In Figure 3D and E, representative images of gray and white matter segmentation are converted to grayscale images. Meanwhile, the impurity and noisy areas are also removed. We observe the striking contrast in the segmentation results and identify axon fibers more obviously. The average of images evaluation time required 4.8 sec with Figure 3D and E. Furthermore, the MPM images, segmentation results, and H&E-stained image (Figure 3F) show clear correlation at the subcellular level. These results indicated that the combined MPM and automated image analysis have further differentiated between the gray and white matter, as well as discerned a precise delineation of the border of the gray-white matter junction.

3.3 Automated detection of the Purkinje cell layer

Purkinje cells are the central functional unit of the cerebellar cortex 27. Generally, Nissl staining is a standard histological method for visualizing Purkinje cell bodies, while in the original MPM image, the middle Purkinje cell layer, the most superficial molecular layer, and the deepest granular layer of three-layered cerebellar cortex can be seen in the representative cerebellar lobule. Figure 4 A depicts that the Purkinje cell layer mainly contains a monolayer of Purkinje cells, showing stronger TPEF signals than background signals.

To rapidly evaluate the Purkinje cell layer in MPM images, we developed another program (See Materials and Methods 2.4.2) to automatically detect the location of the Purkinje cell layer throughout all subdivisions of the cerebellar cortex. The segmentation results of the Purkinje cell layer display clearly the background and enhanced boundaries (Figure 4B). In Figure 4C, the Purkinje cell layer is highlighted in blue and is superimposed, revealing the Purkinje cells are typically presented in a single row at the boundary between the molecular and the granular layer. Figure 4E and F illustrate magnified images with a precision boundary comparable to those of the MPM images. Representative Purkinje cells are marked with white arrows (Figure 4E, G, and I) and arrowheads (Figure 4F, H, and J). Compared with corresponding H&E-stained images (Figure 4D), the contrast of the segmentation results is obviously enhanced, and the Purkinje cell layer can be rapidly identified by highlighting. To assess the accuracy of the Purkinje cell layer detection, the segmentation results of the Purkinje cell layer (Figure 4B) and corresponding H&E-stained images were compared and confirmed by a certified pathologist. The average of images evaluation time required for evaluating the Purkinje cell layer is 1.9 sec with each image (Figure 4C).

3.4 Rapid identification of cerebellar tumors

Visualizing the normal cerebellar structures, including the gray matter, white matter, and Purkinje cell layer, is of utmost importance as a reference to identify cerebellar tumors. Figure 5 A shows human cerebellar hemangioblastomas, which focus on the tumor core and peritumor regions, revealing the extent of the tumor and secondary effects of the surrounding tissue. Tumor-induced ischemia and hypoxia of peritumoral tissue lead to cytoplasmic rarefaction of the granule cell layer, as well as Purkinje cell loss and vacuoles of the Purkinje cell layer. Figure 5B and C clearly depict an abnormal granule cell layer and vacuolated cells of the Purkinje cell layer by combining the image analysis methods, which are used for discriminating normal tissue from peritumor regions. In the tumor core, a high concentration of tumor cells and increased cell densities are readily detected in MPM images, showing densely packed tumor cells and large irregular shape of tumor cell nuclei compare to peritumoral cells, which can be efficiently differentiated tumor core from peritumor regions based on striking nuclear atypia.

Again, we delineated the boundary between the tumor core and peritumor regions in white lines through the image analysis method (See Materials and Methods 2.4.3), and a precision boundary could greatly aid in tumor resection decision making. The average of images evaluation time required 21.6 sec with Figure 5C and D. In future work, we will also use this edge as a mask to label the tumor cells in both the tumor and normal regions. Comparison of the corresponding H&E-stained images (Figure 5E) revealed that MPM images could rapidly discriminate cerebellar tumors from peritumor and normal tissue. In particular, the secondary effects in the peritumor regions are hallmark features of cerebellar hemangioblastomas, a finding that could be of tremendous help for predicting prognosis.