(A-24) Spatial morpho proteomics analysis in lymphoid follicles

Introduction:: Follicles and germinal centers (GCs) form into stages where the antigens can either exist in the form of soluble antigens or presented through antigen-presenting cells (APCs) including follicular dendritic cells (FDCs), macrophages, or dendritic cells. After the antigen encounter, B cells upregulate the expression of the chemokine receptor CCR7. This further drives B cells migration towards the ligand of CCR7 including CCL19 and CCL21 exist at the T-cell and B-cell border of the primary follicles. B cells then bind to CD4+ T-cells which further initiate B cell’s division and differentiation into either short-lived plasma cells or memory B cells or ultimately initiate a GC response1.


Materials and Methods::
Several multiplex protein datasets were used from both public data repositories and published papers such as Imaging Mass Cytometry (IMC)2, Cyclic Immunofluorescence (CycIF)3, and CO-Detection by indEXing (CODEX)4. Single-cell from tissues were segmented using a two-step semi-supervised pipeline. The primary follicles and the GCs in these datasets were manually annotated using CD20, CD3, and Ki67 signals, and a PCA model was fitted to extract outline embedding of the follicle and GC outlines. Single cells with their corresponding spatial were transformed into a spatial graph modeling the spatial neighboring of cells (Figure 1 a). A 3-layer graph neural network (GNN) is used with various graph layers and graph pooling and tested model accuracy in predicting Covid status from patients (Figure 1 b).




Results, Conclusions, and Discussions:: We compared the graph neural networks with baseline multi-layer perception models in a multi-instance learning framework5 for predicting COVID infections of pediatric tonsil tissues stained with 8 multiplexed markers. The MLP6 model obtained an accuracy of 0.92 using a mean pooling layer whereas the GNN7 achieved a model accuracy of 0.972 using an attention pooling layer. Next, we utilized the same framework for cell-level prediction of inside germinal centers, inside follicles, and outside of follicles. The model obtained an accuracy of 0.94 using GNN compared to 0.88 accuracies when using the MLP model with cell multiplex intensities only suggesting that the GNN framework successfully captures spatial information from the graph improving cell level prediction.


We proposed a GNN framework leveraging multiplexed technologies and tissue morphological features to predict patient conditions from Follicles and GCs in lymphoid tissues. It achieved high accuracy in predicting COVID infection in pediatric tonsil tissues and cell level prediction compared to the MLP model. This suggests that the spatial context of each single cell is important in both cell-level and instance-level prediction. Moreover, the GNN framework can be used as a low-dimensional embedding method by extracting the last layer of the GNN framework for downstream analysis tasks such as cell phenotyping, tissue structure (follicles and GCs) embedding, and tissue similarity analysis. Combined with model interpretability tools such as attention pooling and GNNExplainer, the GNN framework would show individual biomarker importance in prediction. With the advancement of spatial multiplex proteomics imaging resolution, the proposed GNN framework helps systematic prediction and computational analysis of the immune microenvironment for heat and disease.


Acknowledgements (Optional): : AFC holds a Career Award at the Scientific Interface from Burroughs Wellcome Fund and a Bernie-Marcus Early-Career Professorship. A. F. C. was supported by start-up funds from the Georgia Institute of Technology and Emory University. This work was performed in part at the Materials Characterization Facility (MCF) at Georgia Tech. The MCF is jointly supported by the GT Institute for Materials (IMat) and the Institute for Electronics and Nanotechnology (IEN), which is a member of the National Nanotechnology Coordinated Infrastructure supported by the National Science Foundation (Grant ECCS-1542174). This work was performed in part at the Georgia Tech Institute for Electronics and Nanotechnology, a member of the National Nanotechnology Coordinated Infrastructure (NNCI), which is supported by the National Science Foundation (Grant ECCS-2025462). Part of the work was conducted at the Center for Nanophase Materials Sciences, which is a DOE Office of Science User Facility, and using instrumentation within ORNL's Materials Characterization Core provided by UT-Battelle, LLC under Contract No. DE-AC05-00OR22725 with the U.S. Department of Energy. IEN also provided financial support in the form of a Facility Seed Grant. Supported in part by Winship Cancer Institute # IRG-21-137-07 -IRG from the American Cancer Society and American Lung Association Innovation Award.


References (Optional): : 1. Stebegg, M. et al. Regulation of the Germinal Center Response. Front. Immunol. 9, 2469 (2018).

2. Allam, M. et al. Spatially variant immune infiltration scoring in human cancer tissues. Npj Precis. Oncol. 6, 60 (2022).

3. Lin, J.-R., Fallahi-Sichani, M. & Sorger, P. K. Highly multiplexed imaging of single cells using a high-throughput cyclic immunofluorescence method. Nat. Commun. 6, 8390 (2015).

4. Black, S. et al. CODEX multiplexed tissue imaging with DNA-conjugated antibodies. Nat. Protoc. 16, 3802–3835 (2021).

5. Foulds, J. & Frank, E. A review of multi-instance learning assumptions. Knowl. Eng. Rev. 25, 1–25 (2010).

6. Pal, S. K. & Mitra, S. Multilayer perceptron, fuzzy sets, and classification. IEEE Trans. Neural Netw. 3, 683–697 (1992).

7. Zhou, J. et al. Graph Neural Networks: A Review of Methods and Applications. ArXiv181208434 Cs Stat (2021).