Longitudinal single-cell profiling in canine DLBCL reveals regulatory networks driving chemotherapy resistance.

Abstract Diffuse Large B Cell Lymphoma (DLBCL) is the most common lymphoid malignancy in both humans and dogs, characterized by clinical and biological heterogeneity. While standard chemotherapy induces clinical remission in many patients, the majority of them eventually relapse, and effective treatment options for refractory disease are limited. The underlying biological mechanisms driving therapeutic resistance remain poorly defined, in part due to the scarcity of longitudinal samples needed to track disease progression and clonal changes over time. To address this challenge, we leveraged naturally occurring DLBCL in pet dogs, a spontaneousand clinically relevant model that closely mirrors human DLBCL in molecular features, histopathology, treatment response and disease progression. A key advantage of this model is the ability to obtain serial lymph node biopsies as part of standard veterinary care, enabling longitudinal sampling across the disease course. While similar human studies may take more than five years for sample collection due to slower disease kinetics and limited biopsy access, comparable longitudinal data can be collected in dogs within a year. This makes the canine model a uniquely powerful platform for investigating tumor evolution and the emergence of therapy resistance in vivo. Using this approach, we collected matched samples from eight dogs at three clinically relevant timepoints: initial diagnosis, complete remission post-treatment, and relapse. This design allowed us to track the clonal dynamics of tumor subpopulations within the same host, enabling direct insight into how therapy-resistant states develop over time. We performed single cell RNA sequencing on hundreds of thousands of cells from these biopsies and applied a modern analytical pipeline integrating machine learning tools. Specifically, we used scVI for robust batch correction and latent space modeling, CellTypist for precise immune cell annotation, and PySCENIC to infer cell-specific gene regulatory networks. Together, these tools enabled us to build a comprehensive, multidimensional map of the DLBCL tumor microenvironment and its dynamic response to treatment. Our analysis uncovered a transcriptionally distinct population of B cells consistently present across all three disease stages. Although rare at diagnosis and nearly undetectable during remission, this population expanded dramatically at relapse, strongly suggesting it functions as a reservoir of therapy-resistant cells. Not aligning with any known B cell subtype, these cells exhibit a unique transcriptional identity defined by a complex regulatory program. Through PySCENIC, we identified a dense network of regulons enriched for transcription factors associated with stem-like properties, cellular plasticity, and immune evasion, including IKZF2, PBX1, LEF1, and IRF8. This suggests that chemotherapy resistance in DLBCL may arise from a shared phenotypic state maintained through distinct transcriptional programs rather than fixed genetic alterations. Among the key pathways implicated, Wnt signaling stands out as particularly compelling. This pathway has been extensively studied in human DLBCL, where it contributes to chemotherapy resistance by promoting stem-like characteristics in malignant B cells. The enrichment of LEF1 regulons within the resistant population raises the possibility that Wnt signaling is similarly active in resistant clones in canine DLBCL. LEF1 regulons also suggest activation of Akt signaling, a pathway linked to resistance in human DLBCL. These findings highlight the importance of longitudinal sampling at diagnosis, remission, and relapse to uncover the cellular and molecular basis of treatment resistance, an approach rarely feasible in human studies alone. Our single cell approach, powered by advanced machine learning and comparative oncology, identifies a persistent, adaptable subpopulation likely responsible for relapse. This work not only provides mechanistic insights into tumor evolution under therapeutic pressure but also opens new avenues to target resistant subclone, reinforcing the translational potential of canine models in advancing precision medicine for human DLBCL.