Microcirculation in the human spleen is completely open as shown in 3D models in virtual reality

Histology

sample and sections

A sample from a 22-year-old male accident victim obtained in 2000 was fixed in 3.7% formaldehyde in tap water for 24 hours at 4°C, embedded in paraffin and used to cut 21 serial sections in 2020. Acquisition was carried out in accordance with ethical regulations (which means the patient’s informed consent) at the time of obtaining the sample. In 2000, ethical voting was not mandatory for working with human subjects at the University of Marburg Medical School. This practice was retrospectively approved by the Ethics Committee of the University of Marburg Medical School.

Serial sections were cut with an N35 blade (Feather Safety Razor Co. Ltd., Osaka, Japan) on a Leica RM2255 microtome with a blade tilt angle of 2.5° using silanized slides. The mean section thickness was 7 µm. High temperature antigen retrieval was used for CD34 and CD271 immunostaining, but not for CD141.

Triple staining procedure

Sections were triple stained using the antibodies and methods described in Steiniger et al.7. One difference was that instead of α-smooth muscle actin, CD141 was first detected in the sinuses and other endothelium (excluding capillaries) using mAb TM 1009 (Pharmingen/DAKO, Hamburg, Germany, No. M0617) at a 1:800 dilution by peroxidase complex technique. Avidin biotinylated. Sections were then autoclaved and CD34 was detected in the capillary endothelium using mAb QBend 10 (Dianova, Hamburg, Germany, No. DLN-09135) at a final dilution of 1:1000 mixed with mAb EP1039Y (GeneTex via Biozol, Eching, Germany, No. GTX61425) at 1:2000 for capillary sheaths. Finally, Bright Vision anti-mouse IgG was detected using Enzo High Def Blue for AP followed by Bright Vision anti-rabbit IgG with Perma Red chromogen (Biodiagnostic Systems, Pleasanton, USA via Zytomed Systems, Berlin, Germany, No. ZUC-001). 125). This chromogen also differed from the method used before. Perma Red chromogen solution was always freshly prepared according to the manufacturer’s recommendation by adding 4 μl of chromogen solution to 250 μl of buffer. The incubation continued for 30 minutes at room temperature with one change of coloring solution. All slides were covered in Mowiol (Sigma Aldrich, No. 324590).

All antibodies were carefully titrated for use in triple staining procedures. Omission of both antibodies previously showed that non-specific background staining by detection systems did not occur. However, we accepted a faint blue color for better orientation in the red core.

Visualization (Supplementary Figure S2)

acquisition

Sections were acquired using a Zeiss AxioScan.Z1 scanning microscope (Carl Zeiss Microscopy GmbH, Jena, Germany) with a × 20 lens at 0.22 μm/pixel. Scanner-generated files were extracted as full-size TIFF images using BioFormats (version 6.4.0)16.

Registration and normalization

The general processing pipeline follows the Lobachev . scheme17However, there were several issues and extensions of the procedure, as detailed below. In general, bioprocessing and image acquisition are followed by coarse registration, ROI selection, normalization, fine registration, interpolation, volume filtering, network building, mesh filtering, VR visualization, visual analytics and wire drawing, and finally, visualization of the final result of the representation. two dimensional.

Coarse scoring with our usual approach18,19 It was impossible with whole partitions. Image size exceeds OpenCV size limits20, 21 library, and thus feature-wide discovery was impossible on available devices.

A solution was adapted for reading the raw image, resizing it for feature detection, and manipulating features on the resized images to create a strict correspondence i.e. Lobachev et al.22. Transformations have been scaled up to original size and applied to full-section images with CImg (version 2.9.8)23 and ITK (version 5.2.0)24,25 libraries.

Then, 20 k × 20 k regions corresponding to larger portions of the section, but small enough to process in OpenCV, were coarsely selected. We did our usual way21. After the initial non-strict recording, a smaller ROI can be determined. To reach 4545 x 4545 pixels (corresponding to 1 mm2), larger regions, typically 8 k × 8 k pixels, were identified. Since non-rigid recording was not possible at the section level, eliminating larger interfacial distortions was a major problem, even after initial non-rigid operation in 20 k regions. We used a feature-based method21 As well as the recording based on Gauss – Seidel26 In the particularly challenging regions of ROI 1 and ROI 3. After the final recording of the fine grains, smaller areas (about 6 K × 6 K) were used for further processing. Not cropped to final size (4545 x 4545) before the following steps.

Before recording the fine grains, all sections were straightened for a single sample. We used the implementation of Khan et al.27 The method of Reinhard et al.28. Support for fine-grain recording normalization. Regarding normalization, color dissociation coefficients were determined. However, it was difficult to distinguish the blue color in the resulting images (CD34 .).+ capillaries) and blue-brown (CD141+CD34+ sinuses near the follicles). Red Detection (CD271+ capillary sheaths) were complicated by weak cationic fibroblasts ubiquitous. After recording, the resulting images were normalized again, using a different one-section sample to adjust for the above problems. In the following, we call the initial normalization “process normalization” and the second normalization is the “final” normalization.

Color separation and interpolation

Coloring colors were separated using the chromatic deconvolution method available in Fiji29. CD34 . channels+ and CD141+ (blue and brown immunofluorescence staining) obtained from processing normalization. After initial experiments, the CD271 . channel+ Cells (immunofluorescent blue staining) were obtained from final normalization.

The separate channels were converted to 8-bit grayscale images in the following way. CD271 staining was interpreted as the purple channel of the CMYK color space. The CD34 channel was the red discarded channel, and the CD141 channel was the blue discarded channel for the default RGB color space. This conversion was done with ImageMagick (version 6.9.7)30.

Then, the 8-bit discrete images were subjected to custom interpolation19, are performed separately for each staining. Serial sections are anisotropic. The resolution of the acquired images was 0.22 µm/pixel in x y– Size level, while precision is along a file zThe size axis was assumed to be 7 µm. We interpolated this contrast based on the dense optical flow31, as implemented in OpenCV. Interpolation resulted in anisotropic volumes with a resolution of 0.22 × 0.22 × 1 μm/voxel.

Size filtering and mesh filtering

Then, the series of images representing the sizes were cropped to 1 mm2 Flip side (4545 x 4545 pixels) and convert it to a single size file using Fiji. From now on, volumes have been handled in 3D Slicer (version 4.10.2)32,33, performed individually for each ROI and each staining channel. Automated sinus extraction did not produce satisfactory results for two reasons. First, the large diameter of the sinuses resulted in double curves for the surface models, in which the sinus wall is moved by the walking cube algorithm first from outside to wall and then from wall to inside. Second, the usual automatic segmentation methods, such as the watershed, were not expected to be successful because the sinus walls have many uncontaminated or poorly contaminated areas that appear as interruptions. Thus, we abandoned automatic processing and used manual sinusoidal segmentation in ITK-SNAP34. The volumes generated by the annotations were processed as detailed below to produce network representations.

The following procedures were applied to:

CD141+ Sinus lining (brown immune staining, ITK-SNAP annotations).

  • The “upper” limit for removing auxiliary tags

  • ITK ثنائي Dual Fill Hole Filter

  • Intensity normalization to 0-255

  • interpolation section19

  • Gaussian camouflage with a value of 1 sigma.

CD34+ Capillaries (immuno blue):

  • Grayscale closure process with kernel size 14-14-3

  • Gaussian blur with sigma = 1.

    For a detailed view of CD34+ capillaries, volume filtration was not applied.

CD271+ Capillary sheaths (red immunoglobulins):

  • Grayscale closing process with a kernel size 10-10-2

  • The grayscale is stretched with a nucleus of size 20-20-5, shaped like a ball

  • Gaussian camouflage with a sigma 2 value.

For capillary sheaths and capillaries, a 3D Slicer was used to construct the network. The iso values ​​were 205 (out of 255) for sheaths, 100 for capillaries, 80 for capillaries with venous walls, 117 for capillary details, and 180 for sinuses. The resulting networks were too large for practical use and were also open at size limits. Hence, we used PolyMender35 to repair the network. The PolyMender ‘qd’ variant was used, in which biological structures do not have straight lines and square angles. A quaternary tree depth of 9 was used for sinus and capillaries. For the sheaths, we used depth 8.

All grids, except for the capillary details, were smoothed using Taubin smooth36 With 10 default iterations, as implemented in MeshLab (version 1.3.2 under Linux, version 2021.10 under Windows)37. We also applied the removal of small, unconnected components, that is, the typical ‘litter’ from the network construction and uncontaminated erythrocytes in the background of our sections.

In the capillary sheaths we removed components smaller than 3% of the main diameter (43 µm). In capillaries this value was 2% (28 μm). In the sinuses we removed 3% of the small components.

This step ends the usual network processing steps, and the resulting networks are viable for scanning in virtual reality. Some networks have been modified in a more specific way, as detailed below.

custom processing

In virtual reality, we used a wireframe (see below), to highlight different parts of the models. Differently colored portions of the grids were separated using custom software, based on the VCG library37 and PyMesh38.

The networks (triangle networks, forming surface models) were crucial for further processing. When we used VR, we needed to represent our data that could easily be viewed on a VR headset in real time using commodity graphics hardware. Size-based rendering is avoided for VR as it consumes resources and may negatively affect frame rate.

Virtual Reality

To examine network models in virtual reality, we used our custom visualization software1. Briefly, it allows viewing reconstructions with original partitions. The user can comment on the models or paint parts of the models in different colors. Classification of objects (for example, capillary sheaths) is also possible.

Using user input, 3D models of capillary ends (Fig. 4a–d, Supplementary Files S1, S2, S3, S4) were drawn in VR and controlled for potential sinusoidal connections. The particularly useful capillary ends in the red pulp were marked in a second round for a separate view with the surrounding sinusoids (Fig. 5a–f, Supplementary Videos S3, S4). These capillary ends were obtained from the ‘detailed’ 3D reconstruction without additional retinal filtering. Finally, all open ends in ROI 3 were visualized in this way (Supplementary File S5). In addition, the capillary network around the bursa was manually marked (Fig. 9a,b, Supplementary Files S6, S7).

Leave a Comment