Traditional cancer studies rely heavily on 2D tissue section analysis, which fails to capture the complex 3D architecture of tumors. Tumor cells interact with surrounding cells, extracellular matrix, and stromal components like blood capillaries and immune cells. Previous observations in hepatoma cells showed lineage-specific arrangements, even in controlled environments, highlighting the variability in tumor tissue organization. This variability suggests that structural organization is influenced by factors such as tissue origin, differentiation state, genetic programming, mutations, activated molecular pathways, and surrounding stromal components. This study aimed to investigate the ultrastructural patterns of cancer tissue using advanced 3D electron microscopy techniques to understand this complex organization. 3D electron microscopy (EM) technologies, such as serial block-face (SBF)-SEM, offer the possibility to analyze large volumes of biological tissue in 3D at high resolution, overcoming the limitations of traditional 2D EM techniques. Hepatoblastoma (HB), the most common childhood liver cancer, provided a suitable model using patient-derived xenografts (PDXs) that recapitulate the histological, genetic, and biological characteristics of parental HB. The study employed SBF-SEM to image HB PDX samples, followed by quantitative analysis using 3D imaging, mathematical algorithms, and deep-learning approaches to measure cell and subcellular component sizes, planar alignment, spatial orientation, and distances to blood capillaries. The integrated workflow combining advanced imaging with computational approaches aims to provide a more comprehensive understanding of HB tissue architecture and spatial organization of tumor cells.

The literature review section emphasizes the limitations of 2D imaging in understanding tumor architecture and highlights the need for 3D imaging techniques to capture the complex interactions between tumor cells and their microenvironment. The authors cite previous studies demonstrating the variability in tumor cell arrangements even within the same cancer type, suggesting that environmental and genetic factors play a crucial role in shaping tumor tissue organization. They also mention the development and application of 3D EM technologies, including SBF-SEM, in biological research and its potential in studying cancer. Specific references are provided to support the classification of hepatoblastomas, therapeutic options, and existing models of the disease. The authors establish the context for their study by emphasizing the lack of detailed 3D bioarchitectural information regarding tumors and the potential of the proposed technique, onconanotomy, to address this gap. The use of hepatoblastoma patient-derived xenografts (PDXs) as a suitable model is also justified based on their ability to recapitulate features of the parental tumors.

The study used an integrated workflow combining sample preparation, SBF-SEM imaging, computational approaches, and infographic tools. Four HB PDX samples were initially analyzed by transmission electron microscopy (TEM) to assess the quality of tissue preservation after fixation and staining. Two samples were subsequently imaged using SBF-SEM, acquiring high-resolution 3D image stacks. The 3D image stacks were then analyzed using a combination of manual and semi-automatic segmentation techniques. Manual segmentation was used to accurately segment key structures, including blood capillaries and their components, and individual tumor cells and their subcellular components. The manual segmentation was performed with VAST-Lite and Ilastik software. However, manual segmentation is time-consuming and unsuitable for analyzing large volumes of data, which made it necessary to develop a semi-automatic segmentation process. A semi-automatic segmentation strategy involved manual segmentation on a subset of the images and using an optical flow algorithm to propagate those segmentations to neighboring slices to segment hundreds of cells. This process involved the use of Ilastik and VAST-Lite software. To overcome the challenges of manually segmenting numerous small mitochondria, a deep learning algorithm (3D U-net architecture) was trained on manually segmented mitochondria from a single cell. The algorithm was subsequently used for semi-automatic segmentation, followed by manual cleanup. Once segmented, various measurements were taken for cells and subcellular components, including size, alignment relative to the best-fit plane, and distance from the blood capillary. The main axes for analysis were determined using principal component analysis (PCA). A vectoral approach was applied to determine cell polarity in relation to a bile canaliculus-like structure using a 3D ray-casting approach with the Bresenham’s line algorithm. Statistical analyses (Spearman's correlation) were performed to investigate relationships between different parameters. The study involved a detailed sample preparation protocol to ensure optimal tissue preservation for EM analysis. Specialized software was used for image alignment, processing, and segmentation. The entire workflow is carefully outlined, ensuring the reproducibility of the methodology. Infographics were created using Blender to aid in visualization and interpretation of the 3D data.

The study revealed several key findings about the 3D organization of hepatoblastoma tissue: 1. **Correlation between cell size and subcellular components:** The size of hepatoblastoma cells was strongly correlated with the size of their nuclei, cytoplasm, and mitochondrial mass. Larger cells had larger nuclei, cytoplasm, and more extensive mitochondrial networks. 2. **Influence of blood capillaries on cell arrangement:** Tumor cells exhibited a tendency to align themselves with the orientation of nearby blood capillaries. The alignment was particularly evident for a significant subset of cells and nuclei (76.6% and 74.5% respectively, within 0-20° angle from the best alignment plane). This suggests that blood capillaries may influence the spatial arrangement of tumor cells. 3. **Polarization of cells towards bile canaliculus-like structures:** A cluster of tumor cells displayed polarization towards a bile canaliculus-like structure, a remnant of liver tissue organization. This suggests that these structures may play a role in guiding cell organization, similar to their role in normal liver tissue. 4. **Relationship between cell size and distance from blood capillaries:** Tumor cells located closer to blood capillaries were significantly larger than those farther away, with the mitochondrial network showing the strongest correlation. This suggests that proximity to blood vessels, and consequently oxygen availability, might influence cell size and mitochondrial content.

The findings of this study demonstrate that the 3D organization of hepatoblastoma tissue is not random but is structured by several factors. The strong correlation between cell size and the size of subcellular components is consistent with known scaling principles in cell biology. The observed influence of blood capillaries on tumor cell alignment and size suggests that vascularization plays a critical role in shaping tumor architecture and could influence the growth and metabolic activity of tumor cells. This is supported by the observed relationship between distance from capillaries and cellular components. The finding that tumor cells polarize towards bile canaliculus-like structures suggests that residual normal tissue structures can influence the overall tissue organization within a tumor. This pilot study supports the potential of onconanotomy as a powerful approach for analyzing the 3D organization of tumors, providing insights that may not be accessible through traditional 2D methods. Future studies with a larger dataset and analysis of more diverse tumor types are needed to further validate these findings and broaden our understanding of tumor biology.

This study successfully developed and applied onconanotomy, a novel 3D EM approach, to analyze the ultrastructural organization of hepatoblastoma. Key findings revealed correlations between cell size, subcellular components, and proximity to blood vessels. The results showed that tumor cells exhibited alignment along blood capillaries and polarization towards bile canaliculus-like structures. This research demonstrates the potential of onconanotomy in uncovering critical structural details that might inform cancer diagnosis, prognosis, and therapeutic strategies. Future research should expand the scope of this approach to analyze diverse tumor types and investigate the dynamic changes in tumor architecture during treatment. The creation of open-source 3D image databanks is suggested to facilitate collaborative research and advance our understanding of cancer.

The study's limitations include the relatively small sample size (two HB PDXs for SBF-SEM analysis), which might affect the generalizability of the findings. The manual segmentation, while providing accuracy, is time-consuming and limits the feasibility of analyzing larger volumes. The automation of organelle segmentation, particularly for mitochondria, was challenging despite utilizing deep learning; further improvement of this process is needed. Although the authors extensively describe their approach to addressing this limitation, the number of cells and organelles fully segmented remained limited. Future work should focus on expanding the dataset and optimizing the automatic segmentation process to address these limitations and obtain more robust results.