Fast and precise targeting of single tumor cells in vivo by multimodal correlative microscopy

Abstract

Intravital microscopy provides dynamic understanding of multiple cell biological processes, but its limited resolution has so far precluded structural analysis. Because it is difficult to capture rare and transient events, only a few attempts have been made to observe specific developmental and pathological processes in animal models using electron microscopy. The multimodal correlative approach that we propose here combines intravital microscopy, microscopic X-ray computed tomography and three-dimensional electron microscopy. It enables a rapid (c.a. 2 weeks) and accurate (<5 µm) correlation of functional imaging to ultrastructural analysis of single cells in a relevant context. We demonstrate the power of our approach by capturing single tumor cells in the vasculature of the cerebral cortex and in subcutaneous tumors, providing unique insights into metastatic events. Providing a significantly improved throughput, our workflow enables multiple sampling, a prerequisite for making correlative imaging a relevant tool to study cell biology in vivo. Owing to the versatility of this workflow, we envision broad applications in various fields of biological research, such as cancer or developmental biology.

Keywords: Correlative microscopy; Electron microscopy; Extravasation; Intravital imaging; Invasion; Metastasis.

Fig. 1.

Workflow for multimodal correlative microscopy. Multimodal imaging of metastatic events observed in vivo requires specific sample and image processing methods. First, the event of interest is captured by using IVM (time, ∼1–2 days). The position of the ROI is marked at the tissue surface with NIRB (1 h). Based on this macroscopic mark, a biopsy containing the ROI is dissected and processed for electron microscopy analysis (1 day+4 days). The resin-embedded sample is then imaged with microCT (2 h). The imaged volume obtained from the IVM is registered to the microCT volume by matching correlated pairs of landmarks in Amira software (1–2 days). 3D registration allows determination of the position of the resin-embedded ROI relative to the surface of the block. The resin block is accurately trimmed to expose the tumor cell for electron microscopy imaging (2 h). Finally, 3DEM of the ROI is performed (4–5 days). If all the steps are performed without interruption, the average duration of this workflow is thus roughly 2 weeks.

Fig. 2.

IVM and microCT imaging of tumor cells in the mouse brain cortex, followed by 3D registration of both datasets. (A) A chronic cranial window is implanted 3 weeks before intravital imaging is performed. JIMT1 GFP-expressing cells are injected into the left ventricle of the mouse heart. For IVM analysis, the cranial window is mounted onto a customized stage that allows reproducible orientation, and thus imaging, of the same ROI over multiple days. (B) IVM analysis of an arrested tumor cell in the brain cortex vasculature. Dorsal views (top panels), z-projection of a large field of view around the ROI at 2 days post injection (top left panel) and of a smaller field of view, imaged 3 days post injection (top right panel). Side views (bottom panels), x,z-projection of the ROI, 2 days (left bottom panel) and 3 days (right bottom panel) post injection. Scale bars: 100 μm in left panels, 50 μm in right panels. (C) After IVM acquisition and perfusion fixation, the position of the ROI is marked by using NIRB at the surface of the brain, producing autofluorescence in both the green and red channel (yellow). The x,z projection shows the vasculature (red) and how the NIRB landmark is confined to the surface of the brain, distant (blue arrow) to the tumor cell (green). A cartoon and an image of the mouse brain show the relative positions of several NIRB landmarks. Aside from the landmark positioned above the ROI (orange box), three additional marks were created as references (green boxes) to facilitate targeting of the selected ROIs when dissecting the biopsies. Scale bars: 100 μm (left panel), 1 mm (right panel). (D) The microCT dataset shows the tissue biopsy (brown) within the resin block (yellow) and the blood vessels (gray). Scale bar: 100 μm. (E) 3D registration of the vasculature as segmented from the two imaging modalities – IVM (red) and microCT (gray). Corresponding points in both datasets are located and marked (yellow spheres for IVM and blue spheres for microCT). Scale bars: 100 μm. (F) Based on the landmarks shown in E, the IVM volume is registered into the microCT dataset with Amira software, which enables precise determination of the position of the tumor cell (green) within the resin block and relative to its surface (bottom panel in F). Scale bars: 100 μm.

Fig. 3.

FIB-SEM imaging of a tumor cell arrested in the vasculature of the brain. (A) Automated 3DEM. The resin block was precisely trimmed to expose the tumor cell for acquisition of data using FIB-SEM. During the course of the FIB-SEM data acquisition, low-magnification ‘key frames’ were obtained (z spacing ∼1 μm, size 50×50 μm). (B) The FIB-SEM z-stack (6350 frames, 8 nm isotropic voxel size) was used to study the fine organization of the tumor cell within the blood capillary. Frames of the FIB-SEM acquisition were segmented to show the tumor cell (TC, green), its nucleus (N, blue) and the capillary's endothelial cells and basal lamina (red). Scale bars: 10 μm (3D models), 5 μm (segmented sections). (C) Multimodal correlation – combining IVM, microCT and electron microscopy imaging to retrieve the tumor cell. Top panel, docking the IVM model in the microCT dataset enabled us to predict the position of the tumor cell (green, from IVM), relative to the nuclei (light gray, from microCT) and to a blood vessel (red, microCT). The nuclei and the vessel could be detected in the microCT scan because of their lower density relative to the surrounding tissue. Bottom panel, segmentation of the FIB-SEM key frames results in a 3D model of the nuclei (white), the vessels (red) and the tumor cell (green). Two views of the models are shown, rotated 90° with respect to each other. Comparing both models enabled us to correlate the positions of the nuclei that were visible with microCT and electron microscopy imaging. Matching nuclei are numbered in the top and bottom panels. Scale bars: 10 μm.

Fig. 4.

3DEM imaging of an arrested tumor cell in brain capillaries. 3DEM analysis reveals tumor cell (TC) extensions pointing towards and into the endothelial cell layer (EC) (first and second columns from the left, 2D and 3D views of the cell extensions, respectively; indicated with an arrowhead). Electron microscopy analyses revealed the junctions between endothelial cells (right-most column, arrowheads) and the basal lamina (right column, blue arrows). These ultrastructural features were observed in the FIB-SEM z-stacks obtained from two different tumor cells that were arrested in the vasculature of the brain (ROI1 and ROI2). The dotted lines and boxes in the 3D models of the FIB-SEM z-stack indicate from which regions of the tumor cells these images were obtained. Scale bars: 500 nm (electron microscopy images); 5 μm (3D models).

Fig. 5.

3DEM imaging allows dissection of the complex structural organization of the vessel containing an arrested tumor cell. (A) Each individual cell lining the tumor-cell-containing vessel, as well as the basal lamina and a perivascular cell we segmented. In the top left image (‘segmentation’), a detailed image of a FIB-SEM slice is shown where two neighboring endothelial cells (EC, pink and red), a perivascular cell (PC, purple), the basal lamina (blue) and the tumor cell (TC, green) are segmented. The bottom left panel shows a 3D representation of all the cells lining the vessel that contains the lower part of the ROI2 tumor cell (Fig. 4). In the columns and rows on the right side, each cell is shown individually. Scale bar: 5 μm. (B) A 3D representation of the basal lamina (blue) is shown (left panel), together with the plasma membranes of endothelial cells (various colors), of the perivascular cell (purple) and the lower part of the ROI2 tumor cell (green; Fig. 4). The three panels on the right depict different sections through the model that reveal the contours of the basal lamina, the endothelial cells and the tumor cell. At the level of the vessel bifurcation, the organization of the endothelial cells and of the basal lamina displays complexity, highlighting potential endothelial remodeling in this area. Scale bar: 5 μm.

Fig. 6.

IVM imaging of tumor cell protrusions of subcutaneous xenografts. (A) D2A1 LifeAct-YPet-expressing cells were injected subcutaneously into the mouse ear. After 2 weeks of tumor growth, the mouse was anesthetized and positioned on a custom-built stage. The ear slit in the stage kept the ear flat and still during intravital imaging. (B) z-projection of the IVM z-stack showing the tumor mass (Ypet, green), its invasive front and the vasculature (Evans Blue, red). The right panel shows a magnified view of the area boxed in the left panel and reveals invasive cells with distinct morphologies. Two cells have extended protrusions (arrows), whereas another one has smoother contours (arrowhead). Scale bars: 100 μm (left panel); 10 μm (right panel). (C) After IVM analysis, the area of interest was marked by using NIRB (white dotted line) at the skin surface. Bottom panel, x,z projection of the z-stack. The green arrowhead points to a tumor cell protrusion. The NIRB markings (orange arrowheads) are confined to the surface of the skin and are distant from the invasive front of the tumor mass. Following perfusion fixation, the NIRB markings (orange box) remain visible on the skin biopsy. Scale bars: 100 μm (left panels); 50 μm (right panel). (D) The resin-embedded sample was scanned by using microCT. The microCT dataset shows the skin tissue biopsy (brown), with the hair follicles (gray), nerves (purple) and blood vessels (red), within the resin block (yellow). Scale bar: 100 μm. (E) 3D registration of the tissue features as segmented from the two imaging modalities – IVM (top left panel) and microCT (bottom left panel). Corresponding points in both datasets were located and marked (yellow spheres for IVM and blue spheres for microCT, left and middle panels). Scale bars: 100 μm. (F) Top panel, based on the reference points shown in E, the IVM model (shown in blue and green) is registered to the microCT dataset (yellow and orange), which enabled us to determine the position of the ROI inside the resin block (bottom panel) and relative to the block surface (415 µm). Scale bars: 100 μm.

Fig. 7.

Serial TEM imaging and correlation of tumor cell protrusions in skin tissue. (A) Large fields of view comprising stitched TEM micrographs from 130-nm thick sections of the volume of interest shown in Fig. 6. The blood vessels (red) and tumor cells (green) are highlighted. Four different depths (marked in μm relative to the first selected section) are shown to illustrate specific ultrastructural hallmarks of the tumor cells and their microenvironment. Three cells from the invasive front imaged with IVM have been retrieved and highlighted (numbered arrowheads). The electron-lucent area in the lower left panel, marked with an asterisk, is a lymphatic vessel. Scale bars: 20 μm. (B) 3D model of the tumor cell protrusions (green) and vessels (red), segmented from 330 serial TEM sections. Corresponding tumor cell protrusions in A and B are indicated with numbered arrowheads. 3D models and TEM images of the cells indicated with arrowheads 1 and 3 are shown in Fig. 8. The inset shows a 3D model of the same region, imaged by using IVM. Scale bars: 20 μm.

Fig. 8.

Serial TEM and electron tomography analysis of tumor cell protrusions from subcutaneous xenografts. (A) 3D model obtained from serial TEM imaging of the cell body of the tumor cell 1 highlighted in Fig. 7. The nucleus (yellow) and plasma membrane (transparent green) of the tumor cell are segmented. Dashed boxes indicate where the higher magnification images shown in D,E were obtained. Scale bar: 5 µm. (B,C) The tumor cell (TC) is flanked by one stromal cell (B, ‘SC’, purple) and a bundle of collagen fibers (C, Col, blue) at a position where the ultrastructure indicates that the nucleus (‘N’) is slightly confined. The left panels show zoomed views of the 3D model, its orientation is shown in the insets. The middle panels show TEM images, and the right panels show color-coded maps of the middle panels. Scale bars: 5 µm. (D,E) Electron tomography analysis of the boxed areas in A. Scale bars: 5 µm (left panels); 200 nm (right panels). Electron tomography imaging revealed the presence of multiple cytoskeletal filaments (arrows) in a protrusion (D) and underneath the plasma membrane at the cell body (E). Moreover, groups of vesicles associated with the plasma membrane (arrowheads) can be recognized in the region where the tumor cells are in close proximity with collagen fibers (Col), a feature that can be observed in multiple regions. (F) 3D model from serial TEM imaging of tumor cell 3 shown in Fig. 7. Dashed boxes indicate where the higher-magnification images shown in I,J were obtained. Scale bar: 10 µm. (G,H) The tumor cell and its cellular and acellular microenvironment. The left panels show zoomed views of the 3D model, the orientations are shown in the insets. The middle panels show TEM images, and the right panels show color-coded maps of the middle panels. Scale bars: 5 µm. (G) A rounded blebbing protrusion of the tumor cell (green) is in close proximity to a neighboring stromal cell (purple). Scale bars: 5 µm. (H) One side of the tumor cell (green) is in close apposition to the skeletal muscle cell (Mu, orange). Scale bar: 5 µm. (I,J) Electron tomography analysis of the boxed areas in F. Electron tomography images reveal the filament network (arrow) at the base of the bleb-like protrusion (I), collagen fibrils (Col, I) and small intracellular vesicles reminiscent of caveolae (arrowheads, J). Scale bars: 5 µm (left panels); 500 nm (right panels).