Dissecting drug resistance in serial uveal melanoma biopsies using integrated, multi-modal single-cell profiling and novel machine learning tools.

PROJECT SUMMARY Uveal melanoma (UM) is a rare melanoma subtype with an estimated annual incidence of approximately 2000 in the United States. While most patients have excellent rates of local disease control with surgery or radiotherapy, nearly half develop metastatic disease, most frequently to the liver. Metastatic UM (MUM) is very treatment resistant and shows no significant responses to conventional chemotherapies or immune checkpoint inhibitors (ICI). UM is molecularly characterized by canonical mutations of the G? protein subunits (GNAQ/11), which result in hyperactivation of the MAPK pathway. Targeting this pathway with MEK inhibitors (MEKi) results in significant anti-tumor activity in vitro, and response rates of up to 14% in patients with MUM, thereby exhibiting significantly higher activity compared to other available systemic therapies. However, there remains significant potential to improve the efficacy of MEKi. To better define modifiers of MEKi sensitivity and resistance, it is important to consider the fact that most UM harbor mutually exclusive GNAQ/11co-mutations, including inactivating mutations or bi-allelic loss of BAP1 (~33%) or deleterious mutations in SF3B1 (~23%) or EIF1AX (13%), thus define distinct genomic subtypes of UM. These alterations likely provide dependencies that are not abrogated with MEKi alone, yet they may represent synthetic lethal vulnerabilities in the context of MEKi. Furthermore, there has not been a systematic evaluation of how MEKi (or any other therapy) alters cancer cell autonomous and cell non-autonomous mechanisms that could confer drug resistance. This is in part due to technical barriers and lack of in vivo models that faithfully recapitulate human MUM. In this proposal, we build on several innovations to systematically determine the impact of MEKi on the MUM ecosystem and define synthetic lethal dependencies across the UM genomic landscape and in the context of MEKi. We will achieve this in two specifim aims: In Aim 1, we will perform single-nuclei RNA-sequencing (snRNA-seq) in patients with MUM who underwent therapy with MEKi selumetinib and had serial biopsies (pre-, on- and off-therapy), and analyze these with several established analytical methods. Second, building on recent developments, we will build machine learning tools for the analysis of sequential single-cell data sets. In Aim 2, we will perform patient- informed CRISPR-screens with multi-modal single-cell RNA/protein readouts across the genomic spectrum of UM. Finally, we will perform genome-scale CRISPR-screens across multiple models to define genotype-shared and -unique modifiers of MEKi responses. Together, these approaches will provide the a comprehensive sequential single-cell analysis in solid tumors, develop tools for temporal single-cell analyses that can be referenced against a ground truth, and define genotype-dependent synthetic lethal vulnerabilities with concurrent MEKi therapy.