Article Text
PDF
Abstract
Background Analysis of the three-dimensional (3D) landscape of the tumor microenvironment (TME) has the potential to transform diagnosis and drug development. Current 2D histology techniques introduce sampling error, interobserver variability, and fail to capture the biology contained within the entire tissue. To-date, practical barriers have hampered widespread adoption of 3D imaging and quantification. We have developed a suite of technologies, the Alpenglow Biosciences Aurora™ 3D Spatial Biology platform, to stain, chemically clarify, image, and analyze entire, intact tissue samples, addressing these barriers. Herein, we present these methods for comprehensive 3D assessment of cancer tissues.
Methods Five de-identified human colorectal cancer (CRC) FFPE blocks were deparaffinized and stained with nuclear (TO-PRO-3) and general protein (eosin) fluorescent dyes, and optically cleared using a modified iDISCO protocol. Samples with a volume of 200–500 mm3 were imaged at 2 µm/pixel resolution with a hybrid open-top light-sheet microscope, the 3D i ™. Smaller regions of interest (ROIs), 0.5 mm3, were re-imaged at higher resolution, 0.33 µm/pixel. Nuclear segmentation and 3D spatial analysis were performed using our 3D ai ™ tools consisting of a U-Net, CytoMAP, and custom python scripts capable of scaling to the 10–20 billion pixel image datasets the 3D i ™ can produce. Tumor infiltrating lymphocytes (TILs) were identified and classified spatially into tumor associated stromal or parenchymal. The variation in stromal to parenchymal TILs was then quantified in 3D and compared to virtual 2D sections of the 3D images.
Results All tissue samples were imaged in 3D, and computational approaches were successfully scaled to 3D and used to segment and classify tumor stroma, parenchyma, and TILs within the high-resolution ROIs. Across the 5 samples analyzed here, 641,800 TILs were identified and spatially characterized. We found significant variation in the ratio of stromal to parenchymal TILs across the 3D volumes (figure 1) and showed as many as 85 traditional 4 μm slices would be needed to accurately quantify the spatial organization of the TILs within some samples.
Conclusions We successfully implemented 3D machine learning analysis pipelines using images from intact CRC tissue samples to identify and quantify key histological features such as tumor regions and TILs within the TME. In addition, quantification of the 3D spatial heterogeneity of key biological metrics highlights that 3D imaging enables a more accurate assessment of disease state. Further investigation is ongoing to link these results to prediction of patient outcomes and treatment response.
Ethics Approval The specimens used in this study were de-identified, were not collected specifically for the study, and the study team does not have access to the subject identifiers linked to the specimens or data, thus this study meets the criteria for ‘not human subjects’ research, exempt from the IRB approval process.
3D quantification of the tumor parenchyma, TIL distances, and spatial heterogeneity. (A) These positional plots show the 3D structure of the tumor region in red for a representative sample. The top middle plot shows the positions of TILs as black dots. The rightmost plot and bottom row show a thin section of the 3D data from various depths. (B) single 2D slice from the same ROI showing the nuclear channel. (C) A positional plot from CytoMAP of the same section shown in B. Each dot corresponds to a cell color-coded by 3D distance to the tumor parenchyma border. This border is highlighted in orange. Distances are split into the tumor parenchyma (distances > 0) and the tumor stroma (distances < 0). (D) 3D positional plot from CytoMAP showing the positions of all cells color coded by distance to the tumor parenchyma. (E) Plot of the variation in the stroma to parenchymal TIL distribution as a function of the number of slices sampled from the full 3D dataset. Each dot represents a simulated experiment with N randomly selected slices of the 3D dataset
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See http://creativecommons.org/licenses/by-nc/4.0/.