PDF, English Restricted access: Repository staff only until 6 May 2026. Login+Download (83MB) | Terms of use

Abstract

Metastatic melanoma is a cancer with poor prognosis and rising global incidence. Although immune checkpoint inhibition (ICI) has revolutionized treatment in advanced disease, most patients fail to achieve durable responses. The biological basis of this variability is insufficiently understood, and the identification of reliable biomarkers remains an unmet clinical need. As metabolic and functional profiles of immune cells are tightly intertwined, I hypothesized that metabolic profiling could complement functional immune cell characterization within the tumor microenvironment (TME). Specifically, I hypothesized that the presence of specific metabolic states at treatment baseline could inform ICI responses. Combining single-cell metabolic regulome profiling (scMEP) and multiplexed ion beam imaging (MIBI), I, for the first time, mapped expression profiles of rate-limiting metabolic enzymes and transporters across twelve cell lineages in tumors from 27 metastatic melanoma patients at spatial, single-cell resolution. Alongside a comprehensive structural characterization that identified compositional hallmarks of ICI response, I could identify several metabolic immune cell states associating with clinical prognosis. Hypoxic CD8+ T cell and macrophage states that displayed high levels of oxidative marks characterized future non-responders, while metabolic states with preferential expression of lactate dehydrogenase (LDH) differed in both their functional phenotypes and their association with response, indicating metabolic flexibility within lineages. Importantly, I could show that metabolic states associate, but not fully recapitulate functional states, suggesting that they represent a complementary layer of immune biology. To leverage spatial context, I developed a computational framework to quantify zonation patterns of metabolic regulator expression, metabolic niches. Metabolic niches were conserved within and across patients, and significantly associated with future ICI response. Using hundreds of image-derived features, I trained machine learning models to accurately predict ICI response. Crucially, the addition of metabolic features significantly improved model performance. Finally, to enable mechanistic follow-up, I established experimental and computational frameworks for multi-modal spatial metabolic profiling of 3D model systems. This bridges the gap between observational studies on clinical material and perturbable model systems, providing a basis for mechanistic studies of metabolic reprogramming in cancer.