Studying a human genetic model of lung squamous cell carcinoma with organotypic cultures and xenografts uncovers distinct advantages of each system and implicates NOTCH1 loss in tumour development

Selecting appropriate experimental systems is crucial in cancer research, where factors such as model relevance, cost, and resource availability guide decisions. A detailed understanding of the strengths and limitations of each model helps ensure their optimal use. We recently developed a human lung squamous cell carcinoma (LUSC) model using genetically engineered human bronchial epithelial cells (hBECs). These were studied through organotypic air-liquid interface (ALI) cultures and standard in vitro assays, including proliferation, invasion, and anchorage-independent growth. However, we did not evaluate whether the same mutant hBECs behaved similarly in vivo, or if in vivo models offered distinct advantages. To address this, we conducted a comparative phenotypic analysis of mutant hBECs derived from the same donor in both ALI cultures and xenografts in immunocompromised mice. Both models followed a similar oncogenic trajectory, involving squamous differentiation and activation of Nrf2 and PI3K/Akt pathways, characteristic of the classical LUSC subtype. However, some transcriptomic differences related to an increase in microtubule formation and cell motility in xenografts emerged. Additionally, xenograft gene expression profiles more closely matched classical LUSC tumours than ALI cultures. Importantly, we observed spontaneous squamous differentiation in the absence of SOX2 overexpression and detected selection for NOTCH1 mutations in specific in vivo mutants. Truncation of NOTCH1 promoted squamous differentiation and suppressed mucociliary features in ALI cultures, underscoring its role as a potential LUSC driver. In summary, mutant hBECs in vitro and in vivo showed largely consistent phenotypes, validating both systems. However, in vivo models can enable the unbiased discovery of new genetic LUSC driver genes. This highlights the complementary value of integrating both model types in LUSC research.