The micro CT scan slices of iodine stained brains from transgenic and wild type samples are shown in the Fig. 1a. Besides the basic anatomical structures visualized equally across both brains, the transgenic brain expressed a large amount of dark grey spots – most likely suspect Aβ plaques that were condensed widely in cortical (neocortex, hippocampus) and some subcortical areas.
Beta amyloid is the main component of amyloid plaques. The golden standard for its detection is its immunodetection by binding to a specific antibody tagged by fluorescent probe. Here, we used this approach to validate the identity of lower density loci detected by micro CT in the transgenic samples. To do this, the iodine stained brain samples were washed in ethanol solution after the micro CT scanning, cut into histological sections and finally specifically stained for the amyloid protein. For selected histological slices we identified their respective sections in the micro CT data. The resulting comparison is displayed in Fig. 1b-d which shows the corresponding slice as a micro CT image (Fig. 1b) and as a histological section (Fig. 1c, detailed view in Fig. 1d). The placements of the suspect Aβ plaques overlapped with the immunodetected Aβ plaques positions (arrows in Fig. 1d). This match strongly supports the fact that the introduced micro CT protocol sufficiently detects individual amyloid plaques in ex vivo whole brain.
In the industrial micro CT setups that use the cone beam geometry, the dimension of the sample is one of the main factors that determine the resulting voxel resolution of the obtained dataset. To acquire a 3D distribution of Aβ plaques in the best possible details, we decided to image an isolated part of the brain and to scan it again with a higher resolution. We focused on the dorsal hippocampus as this structure is severely impaired by the amyloid accumulation and cellular loss in AD. The dissected dorsal part of the hippocampus is showed in Fig. 2a,b. While the whole brain scan delivered a voxel size of 9 µm, in the case of isolated hippocampus we achieved a voxel dimension of 3 µm. The comparison of both scans (Fig. 2c,d) convincingly shows that the readability of the large amyloid plaques’ borders and especially the visibility of small plaques was compromised, even though they were distinguishable in the whole brain scan. On the contrary, the dissected sample yielded a considerably higher level of detail, enabling to detect a large amount of plaques of various sizes.
The next step was designed to assess the sensitivity of amyloid plaques detection using micro CT in contrast to the standard immunohistological staining. After the scanning, we sliced the isolated hippocampal tissue sample and immunostained it for the Aβ presence. Then the micro CT section corresponding to the given histological slice was found and their match was evaluated (Fig. 3a,b). The area of Aβ deposits on each image was marked by independent observers. Then we measured the areas with values of 597.9 µm2 median, IQR 862.3 µm2 in CT data, and 28.9 µm2 median, IQR 105.4 µm2 in histological sections. The respective plaque size histograms are depicted in Fig. 3c. For both methods, the distributions of observed Aβ plaque areas showed that their sensitivity had differed mainly in the lowest size category (0-500 µm2). The immunodetection was more than 10 times more sensitive for the smallest plaques’ detection. The comparison of detailed images in Fig. 3 shows that the border of the plaque is more difficult to read in micro CT. This might have caused an overestimation of the sizes of some of the smallest deposits. Consequently, they might have been marked as larger and so they were scored within the category of 501–1000 µm2.
The data obtained from the micro CT scan of the dissected hippocampus was chosen for a subsequent 3D analysis. Since the Aβ plaques had similar contrast values as other tissue structures, they could not be detected with global thresholding methods. Hence, a manual segmentation of the plaques was performed. After defining the region of interest, all segmented plaques were counted and measured. In the dissected hippocampus (total volume = 13.88 mm3) we identified in total 1666 individual plaques. The volume of the smallest individual deposit was 895 µm³, indicating the lower limit for amyloid plaque identification in present dataset of micro CT data. The biggest identified Aβ plaque had a volume of 721 552 µm3, the dataset of measured volumes was characterized by median value 38 423 µm³ and IQR of 57 512 µm³. The distribution of the Aβ plaques in 3D space with their color coded volumes is shown in Fig. 4a. Next, we assessed the shapes’ variability of Aβ plaques by measuring their compactness. The volume of the plaque was divided by the volume of the sphere circumscribed to the plaque. The values ranged between 0–1 where score of one represented a perfect sphere (Fig. 4b). The identified plaques had median of compactness at 0.396 with interquartile range 0.135. Relating both parameters of plaques compactness and size we found a negative correlation (r=-0.239, p < 0.001) between them, indicating the bigger plaques tended to express more irregular shape than the small sized ones.
A precise 3D model of the plaque occurrence in the hippocampal sample allowed to quantify their spatial distribution including the relation to other labeled structures. We evaluated the distances of plaques to the nearest blood vessel and the inter-plaque distance (Fig. 5). The reconstructed 3D model returned the median distance between the plaques and the nearest vessel of 64.5 µm with IQR 62.3 µm (Fig. 5b). We then investigated whether their relation followed a non-random pattern. For each of the 1666 plaques we generated a random coordinate within the dissected part of hippocampus, leaving out the detected blood vessels. We measured the respective distances of generated “plaques” to a nearest vessel with median of 89.1 µm and IQR 103 µm. The comparison between the sets of experimental and randomly generated data returned the plaque-vessel distance significantly shorter in experimental data than in the random sample (Z = 10.95, p < 0.001, Mann Whitney U test), indicating the plaques tended to appear closer to the blood vessels than if their distribution was random. Finally we tested whether the plaques aggregated together irrespectively of the vessels. The same method of generating dataset with random positions was applied, and the distances between the two closest plaques were measured within the experimental and the random positions datasets, respectively (Fig. 5b). The median of distances from the tissue sample data was 101.1 µm with IQR = 46.5 µm, whereas the randomly generated dataset returned a distribution shift towards larger values (median distance 123.4 µm and IQR 62.7 µm). The Mann Whitney U test confirmed a statistically significant difference between the both measurements (Z = 11.97, p < 0.001).