Deep learning-aided inter-species-comparison reveals shared and distinct molecular patterns in cynomolgus monkey and humans following non-specific T cell activation

Vincent D Friedrich^1,2*Kari Neier³Kristina Müller^1,2Birgit Fogal³Zuzana Loncova⁴Michael Rade⁵Muhammad Shoaib⁶Ulrike Köhl^2,5,7,8Kathleen Hoyt³Parimal Pande³Lily Blanchard³Ernest Raymond³Markus Scholz^1,9Kristin Reiche^2,5,8*Holger Kirsten^1,2*

• ¹University of Leipzig, Institute for Medical Informatics, Statistics, and Epidemiology, Leipzig, Lower Saxony, Germany
• ²Center for Scalable Data Analytics and Artificial Intelligence (ScaDS.AI), Leipzig, Lower Saxony, Germany
• ³Boehringer Ingelheim Pharmaceuticals, Inc., Ridgefield CT 06877, United States
• ⁴Biocenter, Institute of Bioinformatics, Medical University of Innsbruck, 6020 Innsbruck, Austria
• ⁵Fraunhofer Institute for Cell Therapy and Immunology (IZI), Leipzig, Lower Saxony, Germany
• ⁶Luxembourg Centre for System Biomedicine, University of Luxembourg, Luxembourg, Luxembourg, Luxembourg
• ⁷Cancer Center Central Germany (CCCG) Leipzig-Jena, University Hospital of Leipzig, Leipzig, Germany
• ⁸Institute for Clinical Immunology, Leipzig University, Leipzig, Germany
• ⁹University of Leipzig, Faculty of Mathematics and Computer Science, Leipzig, Germany

The early phase of drug development relies on the examination of the efficacy and safety of therapeutic agents in animal models. Due to their close genetic and physiological relation to humans, cynomolgus monkeys (Macaca fascicularis) are a promising animal model in preclinical studies investigating the immune system. However, the shared and divergent characteristics of the immune response at the molecular level are not yet fully understood, which makes transferring findings from these studies to human conditions challenging. Here, we demonstrate a cross-species analysis pipeline using single-cell transcriptomics (scRNA-seq) data from peripheral blood mononuclear cells (PBMCs), investigating the transcriptomic response in cynomolgus monkeys and healthy humans following anti-CD3/anti-CD28 T cell activation. For this, PBMCs were collected at baseline, stimulated in vitro, and measured at 0 hours, at 6 hours and at 24 hours post-stimulation, with two biological replicates per species. The analysis integrates Variational Autoencoder (VAE)-based deep learning, cell-cell communication, differential gene expression, and pathway enrichment for an in-depth data exploration. We observed shared molecular patterns across species in the transition from innate to adaptive immune response, such as the increase of CD4+ T cell proportion and the reduction of CD14+CD16- and CD14lowCD16+ monocytes. Specific transcriptional clusters related to metabolic reprogramming emerged in CD8+ T cells and related to inflammatory and antiviral programs in NK cells at 24 hours post-stimulation in both species, with stronger regulation of pathways related to cell cycle progression, DNA replication, and GPCR signaling in the emerging CD8+ T cell cluster in monkeys than in humans. Cross-species overlap in activated pathways increased from 6 to 24 hours post-stimulation, with pathway co-enrichment and shared foreground genes becoming more similar across species at 24 hours, including Regulation Of Natural Killer Cell Chemotaxis and Interleukin-27-Mediated Signaling Pathway. Across time, we observed a consistent decline in the expression of receptors and ligands involved in cell-cell communication in most cell types, however, the initial levels were higher in humans and the decline more pronounced. Our proposed computational framework enables systematic cross-species time series analyses, advancing translational research and contributing to improved development of immunomodulating therapies.