Background Adaptive immunity is driven by antigen-restricted cell:cell interactions. In mice, two-photon excitation microscopy (TPEM) has revolutionized our understanding of immune cell architectures. However, TPEM has several limitations: most notably it can only be used to study manipulated animal model systems and not human disease. Previously, we demonstrated that by quantifying the distance between B cells and T cells in multichannel confocal images of human tissue (Cell Distance Mapping, CDM) we could identify cognate interactions. However, CDM used fixed filters and could not accurately capture cell shape. We postulated that this might be important as T cells adopt different shapes when scanning for antigen and after recognizing MHC class II-restricted peptides.

Methods We implemented a deep convolutional neural network (DCNN) that accurately identified both cell position and shape. The DCNN output was then analyzed with a tuned convolutional neural network (TNN) to identify distance and cell shape features that best discriminated between different T cell populations relative to dendritic cells (DCs). We refer to this analysis pipeline as CDM³.

Results In mice, CDM³ discriminated between cognate and non-cognate T cell interactions with DCs with a sensitivity and specificity similar to most TPEM measures. In human lupus nephritis, CDM³ both confirmed that myeloid DCs present antigen to CD4⁺ T cells in situ and identified plasmacytoid DCs as an important antigen presenting cell in severe inflammation.

Conclusions CDM³ provides a novel tool for quantifying in situ adaptive immune cell networks broadly applicable to the study of human diseases including autoimmunity and cancer.

Acknowledgements Funded by the NIH including the Autoimmunity Centers of Excellence.