Skip to main content
2023 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Back

    Data Analysis and Deep Learning

    (G-257) CHARACTERIZING CELLULAR HETEROGENEITY OF IN SITU TUBULOINTERSTITIAL NEIGHBORHOODS IN LUPUS NEPHRITIS

    Thursday, October 12, 2023
    9:30 AM – 10:30 AM PDT
    Location: Exhibit Hall - Row G - Poster # 257
    Introduction:: In human lupus nephritis (LN), tubulointerstitial inflammation (TII) and fibrosis are associated with worse prognoses, yet their molecular and cellular mechanisms remain elusive. Investigating the in situ complex cell-cell interactions of the innate-adaptive immune axis and mapping the kidney’s cellular and structural heterogeneity can provide valuable clues and insights to the origin, progression, and interplay of TII and fibrosis in LN and facilitate novel therapeutic development. However, tools such as routine histochemical assessments and RNA sequencing of kidney biopsies either have limited numbers of markers to capture the in situ complexity or lack spatial resolution.

    Materials and Methods:: To elucidate the in situ patterns associated with TII and fibrosis in human renal tissues, we developed a 43-marker panel that allows for comprehensive mapping of different immune cell populations and renal structures. We then used co-detection by indexing (CODEX) coupled to spinning disk confocal microscopy to interrogate the spatial organization of cells and structures in whole renal biopsy sections from 20 lupus nephritis, 22 allograft rejected, and 5 normal kidney samples (pixel size: 0.1507 < ![if !msEquation] >< ![if !vml] >< ![endif] >< ![endif] >m). Given such data, there is a need for a robust quantitative approach to reveal meaningful biological insights on tissue immunity and pathological patterns. To address this challenge, we built a modular computational pipeline for analyzing highly multiplexed microscopy data with the goal of extracting quantitative spatial and biological features by identifying single cells, structures, and cell neighborhoods through novel computer vision methods. The pipeline includes separate modules of preprocessing, segmentation, classification, and spatial analytics.

    Results, Conclusions, and Discussions::

    The preprocessing module implements Ashlar for tiles stitching and alignment of multiple fluorescent images to achieve whole-composite mosaics of the samples with single-cell accuracy. In addition, to address imaging noises and artifacts such as autofluorescence and spectral bleed-through, we perform background subtraction and spectral normalization across imaging cycles. The segmentation module implements CellPose for instance segmentation of cell nuclei, ilastik for segmentation of tubules, and a U-Net deep convolutional neural network for glomerulus segmentation. To maintain accuracy such as challenging renal regions with high cell density, dead cells, red blood cells, we incorporate human-in-the-loop approach to correct and retrain the segmentation networks. The classification module extracts quantitative image features from each detected object output from the segmentation module to categorize cells and renal structures. Together, they reveal the diverse co-residence of multiple classes of immune cells and renal cells in the kidney and a spectrum of tubular inflammatory states across samples. In the spatial analytics module, we demonstrate the differential presences of different immune cell subsets and their spatial proximity with respect to other cell subsets and renal structures. We can examine all these molecular, cellular, and structural features to further characterize TII and fibrosis in both local and global tissue contexts, with the goals to map the relationships among the cells and structures and correlate features and disease outcomes beyond conventional analyses of only marker expressions. We are actively optimizing the current workflow, developing to tackle more granular segmentation and classification tasks, and validating these modules to quantitatively map cells and tissues with high precision and accuracy. We anticipate that high-resolution spatial mapping, in combination with other spatial modalities such as transcriptomics, will advance our understanding of the complex molecular basis of LN.



    Acknowledgements (Optional): :

    This study was funded by the NIH Autoimmunity Centers of Excellence (AI082724). We thank The University of Chicago Human Disease and Immune Discovery core and Integrated Light Microscopy Core Facility for their kind assistance with image acquisition and technological consultation.



    References (Optional): :