Ladies and gentlemen, my name is Joel Zador. I'm one of the neurosurgery fellows at St. Michael's Hospital, University of Toronto, and today I'm going to talk about the regional differences in macrophage recruitment and activation in glioblastoma. Glioblastoma is the most common adult primary brain tumor and is invariably lethal. Despite extensive research, our standard of care remains maximum safe surgical excision followed by radiotherapy and chemotherapy. This leads a limited 15 to 20 month median survival. One of the promising avenues of new treatment modalities to turn this battle around is the modulation of the tumor immune system. So why are immune cells an appealing target? Well firstly, they're widely abundant as shown on these initial micrographs from the 70s and subsequent immunohistochemistry have confirmed perivascular as well as pericellular infiltrates shown in the middle panels. And when these tumors are transplanted into animal models, half the tumor mass was found to consist of immune cells. And when these immune cells were let to get on with their work, they actually ended up eating up the tumor in some of the models. This inspired some initial immunization experiments where patients were injected back their own radiated tumor, but unfortunately there was no benefit from this intervention shown on the growth curves on the bottom right. Nevertheless, this was an important step. So the field became interested in what these immune cells, specifically macrophages, actually did within the tumor. On the left, we have the initial experiments from the 1990s where peritoneal macrophages were stimulated with interleukin-4. And the activation state of these cells was very different from what the field knew as macrophages at that stage. They had increased endocytic clearance and had very little in the way of pro-inflammatory cytokine production. And two to three decades down the line, we know that these so-called M2 or alternatively activated macrophages promote tumor growth and survival. And their function is stimulated by a number of agents, including osteopontin, for example. So this is the current model of macrophage recruitment and activity in glioblastoma. These immune cells are either derived from the bloodstream, so-called bone marrow-derived macrophages, or they are recruited from resident microglia. And the process of this recruitment involves the glioma cells secreting a storm of cytokines which lure in the macrophages, presumably into an M2 alternative activation state. And in turn, these altered macrophages secrete a series of cytokines, shown in the bottom right here, which promote glioma growth and invasion. There are two important layers of complexity to this model, both of which we'll try to address in our study. Firstly, as summarized here on the left, that GBM is anatomically a very heterogeneous tumor. Different parts of the tumor have different cytoarchitecture and presumably different biology. And the other important aspect is that the M1-M2 macrophage activation state doesn't map well to in vivo settings. So by expanding the various stimuli from the microenvironment, as shown here on these gene co-expression network graphs, the macrophage response is also expanded, and it's not captured well by the M1-M2 dichotomy. So there's a lot of suggestion that in different parts of the tumor, the immune system behaves differently. Therefore, a research question was, firstly, what is the identity of the immune cells in different parts of the tumor, and what is their biological activity? For this, we adopted an open-source data set, specifically the single-cell cohort of Darmaris et al. We derived cell identity based on established cellular markers and inferred macrophage activity by combining the inflammation score and the maturity score for each cell, adopting the model of Li et al. And we further mapped the developmental trajectories or the biological journey, to speak, for these cells using a technique called pseudotime. This is our first set of results. We can see from the representative PCA plot in the top left that the tumor cells in red separate well from the immune cells in green and blue. Overlaying these results with the geographical labels from the inset in the top right, we can infer that in the tumor core, majority of cells are derived from a bone marrow source, whereas in the periphery, they are originating from a microglia. Imitating the gene expression profile of these cells enriched in cell adhesion for the core and chemotaxis for the periphery. These findings are in line with the results of Darmaris et al. We further refined the activation state of our macrophages by mapping it onto this two-by-two grid demonstrated in panel B. Pre-activation cells were labeled as M0 with low inflamed and low maturity state, and from this pool emerged the classic deactivated macrophages in the high inflamed and high mature state, whereas the M2 alternative macrophage population was interpreted as a highly mature but less inflamed state. Analysis of the numerics here on the right has shown that in the tumor core, the pre-activation pool, which we considered as the freshly recruited macrophages, were almost exclusively derived from a bone marrow pool in the tumor core, whereas in the tumor periphery, they were almost exclusively derived from a microglia source. Please note that the M1 and M2 activation states have varying proportions of microglia versus bone marrow derived labels. Next, we wanted to validate our results from single cell data to bulk gene expression data. For this, we used the IV data set, where the GBM samples are micro-dissected into five different anatomical regions based on their site architecture, and we mapped these regions onto our core versus periphery samples as shown on the header of the heat map in the middle of the slide. We inferred cell abundance based on metagene signatures and found that the pre-activation bone marrow derived macrophages were more abundant in the tumor core, and the pre-activation microglia cells were more abundant in the tumor periphery in keeping with our single cell results. We next analyzed the plausible receptor-ligand interactions that could drive this recruitment. So we've taken all the receptor-ligand pairs that were previously shown to drive macrophage activation and created these HIF plots. The purple color code is for pre-activation cells. To make sense of these complicated plots, we ranked each receptor based on its rate of synthesis and also its connectivity and found that there was a lot of biological redundancy, lots of parallel mechanisms in both core and periphery, and there was a lot of overlap in the driving mechanisms. For example, integrin-related processes were highly abundant. However, activation of CD44, for example, was higher in the core versus CCR1 activation was higher in the periphery. We finally wanted to follow the temporal changes in gene expression that drives tumor-associated macrophage activation. To do this, we implemented a technique called pseudotime analysis. This technique takes single cells in different biological states from their cross-section gene expression data and orders them to reconstitute a logical sequence, like a wave of gene expression, that drives a biological process. So this is how the analysis would work on our own data. Firstly, we isolate the tumor-associated macrophages based on their gene expression profile and then derive the developmental pathways where each branch represent a distinct cell state. Pseudotime analysis of the tumor core showed increased inflammation as the cells matured, except for the side branch highlighted by the purple arrow. These cells had lower inflammation scores. Investigating the gene dependencies of this branch has shown strong correlation with the PD-1 signaling of one of the targets of our currently trialed immune checkpoint inhibitors. The implications of this are that this presumably pro-tumor state is likely to be resistant to immune checkpoint blockade. Interestingly, the same analysis showed a completely different picture in the tumor periphery. Here, increasing maturation scores were paralleled by decreased inflammation scores, suggesting the evolution of a pro-tumor environment. When analyzing the genes associated with the most mature branch, indicated by the blue arrow here, we found strong enrichment in NF-kappa-b signaling, demonstrated by this scatter plot in the bottom right. This made biological sense, as NF-kappa-b has been previously shown to associate with the glioblastoma invasion and also tumor growth. Our pilot data shows important regional differences in tumor-associated macrophage biology in glioblastoma. We demonstrate that in the tumor core, majority of immune cells are derived from a bone marrow source versus microglia in the periphery. In addition to the published data in our abstract, we further demonstrate that in the core, immune cells develop towards a more inflamed state, whereas there's a small population of presumably pro-tumor immune cells that are possibly resistant to immune checkpoint blockade. In the periphery, we show that the main evolutionary trajectory is towards a non-inflammed state, and this associates with NF-kappa-b signaling. Our results highlight the need for a multi-targeted approach to GBM therapeutics. Finally, a few words about our team. We're a group of multidisciplinary researchers from computer science, physics, background, and some of us are also neurosurgeons. And our aim is to help the application of machine learning in the field of neuroscience. Finally, I'd like to thank the Michael and Amir Adan Fellowship for sponsoring me, and thank you for your attention.

macrophage recruitment

glioblastoma

treatment outcomes

immune cells

multi-targeted approach