Feeding the wrath with myelin

Abstract

Kloosterman and colleagues study molecular and cellular changes during radiation therapy and disease recurrence across glioblastoma molecular subtypes. They uncover a distinct immune-cancer cell metabolic crosstalk during pro-neural/oligodendrocyte-progenitor-cell-like to mesenchymal-like transition, wherein macrophages feed on cholesterol-rich myelin debris to provide lipids to mesenchymal tumor cells, thereby fueling glioblastoma growth.

Glioblastoma is one of the most lethal tumors. The tumor cells are characterized by marked molecular and cellular diversity as well as spatial heterogeneity; they are also capable of rapid growth and sinuous intracranial spread. When the disease is treated by radiation and chemotherapy, tumor cells unleash their wrath by acquiring additional molecular changes [1-3]. This plasticity and clonal evolution allows the disease to recur within a short time. It has long been known that innate immune cells such as microglia and macrophages are a large component of glioblastoma tumors. In the late 1970s, the use of EA rosette assay (monocyte rosette Ripley antibody-coated human erythrocytes (EA)) demonstrated that the percentage of ‘macrophages’ in intracerebral tumors of two experimental rat brain tumors were 6.5±2.4 and 21±2.7 [4]. Extending a similar approach to patient tumors also indicated a macrophage range of 8% to 78% (mean 45%) [5]. Yet, innate immune diversity and function in abetting tumor cells has remained incompletely understood.

The challenge of experimentally uncovering an ever-shifting landscape of a motley collection of innate immune cells in crosstalk with chameleonic tumor cells was taken up by Kloosterman and colleagues in their recent publication [6]. The authors utilize an RCAS-TVA gene delivery model of a proneural cell molecular subtype (PN; Verhaak nomenclature [7]) i.e. its equivalent oligodendrocyte progenitor cell (OPC)-like (Neftel nomenclature [8]) molecular subtype, that subsequently shifts towards a mesenchymal or mesenchymal-like (MES) molecular subtype upon mouse glioblastoma recurrence after fractionated radiation therapy [6]. In a tour de force, the authors exhaustively analyzed this model using various methods, including single cell transcriptomics, spatial transcriptomics, ATAC-seq on sorted innate immune cell types, and lipidomics. Both monocyte-derived macrophages (MDMs) and ontologically distinct microglia were binned into a spectrum of “pre-activated” to “inflammatory”, to “metabolically active”, and “proliferative” states. Pre-activated microglia were abundant at the infiltrative margins. MDMs, by contrast, inhabited regions of microvascular proliferation, a classic histopathological feature of glioblastoma. Upon disease recurrence coinciding with MES transition, there was increased MDM proliferation in tumors relative to the primary PN/OPC tumor. The pre-activated state of MDMs also shifted to a metabolic state. Vis-a-vis a school of conventional thinking suggesting that inflammation fuels glioblastoma growth, the authors reported the rather surprising finding that inflammatory microglia and MDMs were less abundant relative to metabolically active innate immune cells in MES glioblastoma. These results imply that the metabolically active state and not the inflammatory state is especially malevolent in the context of disease recurrence. Dubbed the “GPNMB⁺” cluster, these cells shared the transcriptional profile of “lipid-associated” and immunosuppressive macrophages. Pathways upregulated in these lipid-laden macrophages (LLMs) were associated with phagocytosis, synapse-pruning, and lipid transport. These LLMs were abundant in primary (NF1-driven primary MES glioblastoma mouse model) as well as in recurrent glioblastoma of the MES molecular subtype and, consistent with their pernicious nature, this cell state was associated with the hypoxic, pseudo-palisading necrosis regions (another histopathological distinction of glioblastoma). What was the source of lipids that these tumor-promoting macrophages fed on? Lipidomics showed a preponderance of cholesterol and cholesterol precursors. Yet, cholesterol accumulation occurred despite the downregulation of genes encoding cholesterol biosynthesis enzymes in these macrophages. This is interesting because it suggests that MES glioblastoma-associated innate immune cells engorge themselves on a feast of lipids, primarily myelin. CD36 seems to be the receptor involved in the feeding frenzy. Targeting LLMs by longitudinal treatment with a blood-brain barrier-permeable CD36 inhibitor following radiation therapy in the recurrent setting improved mouse survival compared to radiation alone. Accordingly, the proportion of LLMs was reduced but macrophages in an inflammatory state were increased in the tumor microenvironment, suggesting that feeding on lipids via CD36 drives the LLM transcriptional state. The investigators next established ex vivo co-culture models to study the interaction between MDMs, microglia, and PN or MES glioblastoma tumor cells. Exposure of macrophages to MES-like glioblastoma cells together with myelin debris instructed the uptake of cholesterol and led to an LLM profile. These macrophages also upregulated lipid transporters, which is interesting because it suggests that the macrophages might not be satisfied with just feasting on myelin debris, but that they also transport the lipids to the tumor cells. Consistent with this idea, MES tumor cells themselves were unable to take up myelin. In fact, these cells suffered from lipotoxicity and cellular stress when exposed to myelin. The LLMs not only prevented lipotoxicity by consuming myelin debris, but also fueled the proliferation of tumor cells by supplying them with lipids (Figure 1). Of note, MES cells were poor in synthesizing their own cholesterol and survived by depending on LLMs for growth. This concept was further validated by experiments demonstrating that the pharmacological inhibition of Liver X receptor (LXR) (a transcription factor involved in cholesterol homeostasis), or the genetic ablation of Abca1 (a gene involved in cholesterol efflux) in macrophages, adversely affected the viability of MES tumor cells.

Figure 1. Macrophage-tumor metabolic crosstalk in mesenchymal glioblastoma.

The schematic represents the model proposed by Kloosterman and colleagues [6], where tumor-associated macrophages phagocytose myelin debris to become “lipid-laden”, acquiring a unique transcriptional state in which lipids are transferred to mesenchymal tumor cells, leadinf to tumor cell proliferation and glioblastoma growth. Complementing this interaction, the high glycolytic state of mesenchymal cancer cells supports the induction of the “lipid laden” state of the macrophages [6].

To probe the clinical relevance of these observations, the authors queried the TCGA (The Cancer Genome Atlas) and GLASS (Glioma Longitudinal Analysis) Consortium datasets to determine whether the LLM signature was associated with disease prognosis. Indeed, the lipid-associated innate immune cell transcriptional signature correlated with significantly poorer prognosis and also primarily with MES molecular subtypes. Neoadjuvant immune checkpoint blockade (ICB) with anti-PD-1 antibody has been investigated in small, randomized trials in glioblastoma [9,10], but is not US FDA approved. Examination of transcriptome data from one study [11] that included 20 patients with recurrent glioblastoma treated with neoadjuvant anti-PD-1 antibody showed a lower percentage of LLMs compared to those in newly diagnosed glioblastoma (27 patients) or recurrent glioblastoma not treated with immunotherapy (22 patients). Data were not available to enable a correlation with ICB responsiveness. The authors speculated that the LLM high transcriptional signature associated with glioblastoma might account for the lack of ICB success for this disease. It remains to be determined if selecting glioblastoma patients with low LLM, but high inflammatory innate immune cell profiles might uncover a more promising response to ICB.

Leila Akkari’s research group further compared their primary PN/OPC-like mouse model to a primary classical (CL; Verhaak nomenclature) i.e. astrocyte (AC)-like (Neftel nomenclature) mouse RCAS-TVA model subject to recurrence after radiation therapy. Unlike the PN/OPC model, the CL/AC-like primary model did not transform to MES upon recurrence. Instead, the molecular subtype shifted to that of an PN/OPC-like subtype. The CL/AC-like glioblastoma was not enriched in LLMs. Importantly, both primary PN/OPC-like and primary and recurrent MES tumor cells could impart the LLM signature in tumor innate immune cells. Yet, why CL/AC-like glioblastoma lacked LLMs, despite a presumed abundance of myelin debris in the context of all subtypes of glioblastoma, remains an open question. Additionally, only MES and not PN/OPC-like tumor cells received lipids from LLMs. Why PN/OPC-like cells did not partake in lipid transfer from LLMs remains elusive. The absence of expected events – such as the importance of the dog’s silence for Sherlock Holmes in The Adventure of Silver Blaze [12]- might hold additional clues to unraveling the mysteries of glioblastoma-immune cell interactions, and help identify vulnerabilities to vanquish this disease.