Microglia and metastases to the central nervous system: victim, ravager, or something else?

Few data are available on the immune composition of the microenvironment of brain metastases from lung cancer. Some of these studies used animal models treated with exosomes derived from lung cancer cells and co-culture cell systems to mimic the microglia vascular niche. They demonstrated that a suppressive signal is transferred from the brain endothelium to microglia through the release of endogenous Dickkopf-related protein 1. This causes the switch from the M1 to the M2 phenotype and the gain of pro-tumorigenic features by microglia in the pre-metastatic niche [21]. Among the many factors involved in brain metastases from lung cancer, interleukin (IL)-6 and colony stimulating factor 1 and cell migration-inducing hyaluronidase (CEMIP) are implicated in the repolarization and over-activation of myeloid cells and microglia, respectively, to improve the transmigration of metastatic cancer cells across the BBB and to create a pro-inflammatory metastatic niche [20, 22, 23]. Interestingly, some works showed that human tumor cells escape the microglia immune surveillance by upregulating CD47 and SIRP ⍺ on tumor and microglial cells, respectively (Fig. 1A). This promotes their interaction, reduces microglia phagocytic activity and increases their secretion of trophic factors that might promote the progression of brain metastases from lung cancer [24].

In brain metastases from breast cancer, several immunohistochemistry-based studies have shown a strong expression of granulocyte–macrophage colony-stimulating factor (GM-CSF), C-X-C motif chemokine ligand 12 (CXCL12) and its receptor 4 (CXCR4), and CX3CL1. These factors promote microglial cell proliferation and attraction to the tumor microenvironment, respectively (Fig. 1A). In turn, microglia secrete tumor necrosis factor (TNF) ⍺ that stimulates brain endothelial cells. This increases BBB permeability and immune cell infiltration (including macrophages) that might support the metastatic process [25, 26]. Studies in animal models demonstrated that invading breast cancer cells modulate the microglia activation state and topography. Indeed, microglial cells infiltrate the tumor mass, accumulate in gliosis zones, and are detected in direct contact with tumor cells after extravasation. These microglial cells are heterogeneous, as indicated by the presence of activated (i.e., expressing only major histocompatibility complex (MHC) I, with stellate morphology), hypertrophied (i.e., hypertrophic stellate appearance), and reactive (i.e., expressing MHC I and MHC II, with amoeboid morphology) cells [27]. For instance, in a breast cancer xenograft model, loss of the long non-coding RNA X-inactive specific transcript (XIST) reprograms microglial cells toward the suppressive M2-like phenotype to promote brain metastasis formation [28]. This anti-inflammatory M2 phenotype has been detected around the tumor site and is characterized by the expression of arginase-1, mannose receptor 1, inducible nitric oxide synthase, and cyclooxygenase 2 [29].

In addition, gene expression analyses of microglial cells co-cultured with carcinoma cells showed alterations of the Toll-like receptor (TLR), WNT/β-catenin, stromal cell-derived factor 1 (SDF1)-CXCR4, phosphoinositide 3-kinase (PI3K), and chemokine ligand 2/chemokine receptor 2 (CCL2-CCR2) signaling pathways. These signaling cascades are important for the microglia-cancer crosstalk and their inhibition prevents breast cancer metastasis formation and invasion [10, 16].

To better understand the metastatic niche, Simon et al. developed a mouse model to investigate metastatic breast cancer cell-microglia interactions using intravital imaging combined with ex-vivo electrophysiology. To this aim, they implanted an optical window on the parietal bone to facilitate cell behavior monitoring in situ in the outer cortex of heterozygous Cx3cr1^GFP/+ mice. After breast cancer cell grafting, they observed a significant accumulation of activated microglia that surrounded the invading tumor cells. The inflammation resulted in significant cortical disorganization and abnormal local field potential spike events at the tumor site [30]. These changes could partly explain the epileptic seizure and cognitive damage observed in patients with brain metastases. Besides tumor cells, microglia can also be modulated by other cell populations of the brain microenvironment. For example, reactive astrocytes with activated signal transducer and activator of transcription (STAT)3 surround metastatic brain lesions and promote the expansion of CD74-expressing microglia that become highly enriched within macro-metastases and produce the growth-promoting factormidkine [31].

The communication between metastatic melanoma cells and microglia may influence the secretion of factors that promote vascularization, such as angioprotein-2 (ANG2), by melanoma cells, and of growth differentiation factor 15 (GDF-15, also called macrophage inhibitory cytokine-1, MIC-1) and other pro-inflammatory cytokines by microglia. All these factors promote the metastatic process [32, 33]. Moreover, microglia and melanoma cells reciprocally modulate their gene expression, cell signaling, and cytokine secretion. For instance, melanoma cells significantly affect microglia morphology, proliferation, migration, and matrix metalloprotease (MMP)-2 activation. In turn, microglial cells facilitate phenotypic changes in melanoma cells, resulting in increased proliferation, migration, and MMP-2 activation that promote their aggressiveness [34].

In mouse models of melanoma, IL-17A-STAT3 signaling plays an important role in the metastatic melanoma cell-microglia interaction. IL-17A stimulates angiogenesis and leads to IL-6 synthesis. In turn, IL-6 induces STAT3 activation in melanoma cells that upregulates genes involved in tumor angiogenesis and survival [34, 35]. Moreover, transforming growth factor (TGF)- β expression is increased in microglia, causing tolerance of metastatic melanoma cells by anti-tumor cytotoxic T cells. In a mouse model of melanoma metastases in the brain, TGF- β2 expressed by melanoma cells plays a crucial role in the establishment of metastases specifically in the CNS [36]. In another mouse model of spontaneous melanoma brain metastases, activated microglial cells are recruited to the tumor-brain interface. Depletion of microglia and macrophages using macrophage colony stimulating factor-1 receptor and MMP-3 inhibitors, respectively, drastically reduces the total number and mean size of brain metastases [17]. This suggests that these cell types are key players in melanoma metastases to the brain. Moreover, in vitro and in vivo experiments in immune-deficient mice bearing xenografts of human melanoma brain metastases showed that the extracellular cysteine protease inhibitor cystatin C is involved in the melanoma cell-microglia interaction. Cystatin C secretion is increased in both melanoma and microglial cells. This factor promotes melanoma cell migration and inhibits microglia migration to the melanoma tumor site, thus supporting melanoma brain metastases [37]. Furthermore, pre-clinical data showed an abnormal interaction between microglia and metastatic melanoma cells. This was caused by altered JNK and p38 signaling in microglia that decreases their phagocytic function and supports the metastatic niche [34] (Fig. 1A).

These results in three different cancer types highlight that the interplay between microglia and metastatic cells from melanoma and lung and breast cancer significantly influences the disease outcome. Indeed, microglial cells can have anti- or pro-tumorigenic roles in animal models in function of their activation state and microenvironment. Microglial cells have been detected in human brain metastatic samples, but their association with the clinical outcome is still not clear and requires further validation [38, 39].

Melanoma, lung and breast cancers can also spread in the leptomeninges of brain and spinal cord or in the cerebrospinal fluid, causing leptomeningeal metastases (LM). They concern ~ 23% of patients with melanoma, 9–25% of patients with lung cancer, and 5% of patients with breast cancer, and are associated with very poor survival [4]. Although activated microglia have been found in human samples and mouse models of LM, very little is known about their role in LM development and growth [40, 41].

Classically, microglial cells have been classified into the M1 and M2 subtypes that have different functions and are induced by different microenvironmental stimuli (Fig. 1B). M1 microglial cells are induced by pro-inflammatory cytokines (e.g., TNFα, interferon gamma, and lipopolysaccharide) [42, 43]. By producing reactive oxygen species that increase STAT1 expression, M1 microglia trigger the secretion of inflammatory cytokines, such as inducible nitric oxide synthase, IL-1β, IL-12 and IL-23. This explains their anti-tumor immune role [44]. Specifically, by expressing MHC II, CD74, CD80/CD86 and CD11c [45], they function as tumor antigen-presenting cells, and activate T cells, leading to tumor cell death and clearance [17, 46, 47]. Conversely, M2 microglial cells are induced by anti-inflammatory cytokines (e.g., IL-4, IL-13, IL-10, and TGFβ) [48]. By overexpressing vascular endothelial growth factor, CD204, CD163, MMPs, and arginase-1 immunomodulatory molecules and releasing immunosuppressive molecules, they promote STAT3 expression and tumor cell growth, and inhibit their antigen presentation function [42, 43, 49, 50] (Fig. 1B).

Activated microglia are an abundant source of inflammatory and anti-inflammatory molecules that can affect tumor progression and also brain metastasis formation. The M2 microglia phenotype results in disturbance of CNS homeostasis. Therefore, M2 cells contribute to support tumor progression and metastasis development, due to the local immunosuppression [28]. Metastatic tumor cells induce the M1 to M2 phenotypic shift that supports their growth. However, it is not clear how they can evade the cytotoxic effect of M1 microglia and what is the exact relationship between M1-M2 microglia balance and tumor metastases during brain invasion by metastatic cancer cells.

Microglia role in CNS diseases, including brain metastases, seems to be complex and goes beyond the oversimplified binary M1-M2 definition or the mixed M1 and M2 functions that cannot explain the myeloid compartment heterogeneity and plasticity. Indeed, when the BBB integrity is affected during tumor progression, the M1-M2 classification is not adequate because of the increase in macrophage number due to the homing of peripheral immune cells in the brain that can enhance the myeloid infiltrate heterogeneity. This makes more difficult to decipher the specific function and phenotype of each subpopulation in this complex microenvironment [51].

To better understand their specific roles, several studies using human and mouse models tried to explain the different expression profiles of resident microglia and infiltrating macrophages in physiological contexts and in brain malignancies. The first hypothesis is based on their distinct microenvironments of origin that facilitate and direct their differentiation [52, 53]. This leads to a specific expression profile of tumor-associated microglia compared with infiltrating macrophages. Specifically, their profile is enriched in cytokines, chemokines and complement components, and is correlated with antigen presentation and immune suppression functions. The second hypothesis is based on specific markers expressed by differentiated microglia compared with infiltrating macrophages. These markers include transmembrane protein 119 (TMEM119, a homeostatic microglia marker), P2Y12 (expressed by human microglia during development), Sal-like 1 (a transcriptional regulator that defines microglia identity and function), and sialic acid-binding immunoglobulin-type lectin (a specific microglia activation marker involved in tumor recognition and engulfment) [9].

The major challenge in these studies is the lack of specific, stable and standardized experimental systems. Consequently, the microglia heterogeneity and functional diversity, the contribution of infiltrating macrophages, and the specific roles of all these different subpopulations in healthy and disease/metastatic brain cannot be completely understood using classical methods and concepts.

Microglial cells constantly sense changes in the environment and adapt to them. This is possible because they express a specific set of genes that are called the microglia sensome and encode proteins that sense extracellular signals (e.g., purinergic receptors) (Fig. 2A). As first defined by Hickman et al., the microglia sensome is composed of the top 100 genes, mainly receptors, that are highly expressed on the microglial cell surface and that are involved in sensing potential pathological conditions. Abels et al., applied the approach developed by Hickman et al. to study the microglia sensome in other scRNA-seq datasets (i.e., data by Gosselin et al., [57] for the mouse microglia sensome and data by Gosselin et al., and Galatro et al., [57, 58] for the human microglia sensome). This allowed them to provide new transcriptome information on microglia and to identify similarity and differences between the murine and human sensomes. They found a significant overlap, including 57 genes that are highly expressed in both species and that they called the microglia core sensome (Fig. 2A). These genes are present in at least 75% of all analyzed microglia sensomes. To determine the microglia core sensome function, these 57 genes were ordered in eight different groups in function of their differential expression between microglia and cortex using the expression data from the study by Gosselin et al. [57]. These gene groups included purinergic and related receptors, cytokine receptors, chemokine and related receptors, Fc receptors, pattern recognition and related receptors, extracellular matrix receptors, endogenous ligands receptors/sensors and transporters, proteins involved in cell–cell interactions, and potential sensors without known ligands. Next, analysis of the specific ligands recognized by the identified sensome genes showed an overlap between the human and mouse ligands. This suggests that the mouse and human microglia can sense the same ligand groups. Then, the ligands recognized by the sensome receptors were classified in different ligand groups (i.e., glycoproteins, cytokines, immunoglobulins, amino acids, carbohydrates, electrolytes, lipopeptides, chemokines, neuraminic acids, nucleic acids, receptors, lipids, fatty acids, leukotrienes, hormones, steroids, and phospholipids). These ligands are involved in the most important physiological pathways necessary for cell function (this will be discussed in more details further in this section). If deregulated, these molecules can contribute also to brain tumor and metastasis development. For instance, it has been shown that cytokines, chemokines and their receptors, and TLRs (e.g. IL-6, IL-6R, CSF1R, CX3CR1, TLR3) are directly or indirectly linked to brain injuries, tumorigenesis, and metastasis [59‐63]. Compared with the mouse sensome, the human core sensome includes a higher number of genes that encode extracellular matrix receptors, endogenous ligand receptors, sensors, and transporters.

Additional datasets were included to test the impact of this core sensome on CNS disorders (such as Alzheimer’s disease and amyotrophic lateral sclerosis) and during aging. Similar changes were identified in different datasets concerning the same disease or condition. Importantly, in conditions of brain damage, the microglia sensome was deregulated. In human microglia datasets, different microglia clusters were identified. They were characterized by the expression of CCL2, CCL4, EGR2 and EGR3, suggesting a more activated state of microglia. This might be due to environmental factors and to epigenetic differences between human and mouse microglia. Differences were detected also in genes involved in phagocytosis, complement, and susceptibility to neurodegenerative diseases [64].

Among the genes identified in the human core sensome, some might have a role in brain metastasis. For instance, IL-6 trans-signaling via the soluble IL-6 receptor (IL-6R) is crucial for microglia repopulation to robustly support neurogenesis, specifically by enhancing the survival of newborn neurons that directly support cognitive functions. This neuroprotective and pro-regenerative microglia phenotype can contribute to repair brain injuries and fight diseases, such as brain metastasis [63]. On the other hand, PD-L1 (CD274) has a pro-metastatic role in brain metastasis. It has been reported that recurrent brain metastases after radio-immunotherapy are partly due to the accumulation of PD-L1⁺ myeloid cells. Their presence indicates the establishment of an immune suppressive environment that counteracts the re-activated T-cell responses [65]. In addition, TLRs are transmembrane components that in physiological conditions sense danger signals, connect the innate (e.g., microglia) and adaptive immune systems (e.g., T cells) and contribute to tissue homeostasis. However, in cancer, TLR roles are contradictory depending on the cancer type/stage and immune microenvironment context. For instance, before metastasis initiation, TLR3 signaling promotes tumor cell death in breast and lung cancer and also in head and neck squamous cell carcinoma. TLR3 stimulation (e.g., by interferon type I signaling) results in cancer cell apoptosis in human and mouse models, or in the suppression of cancer cell migration, depending on the tumor stage. However, after the metastatic process initiation, TLR3 activation enhances tumor migration [61, 66, 67].

Moreover, microglia phagocytic activity, which is crucial for tumor and metastatic cell clearance, relies on specific receptors expressed on the cell surface (e.g., TLRs, and triggering receptor expressed on myeloid cells 2 (TREM-2)) and their downstream signaling pathways (e.g., NF-kB and IRF3, 4, 5, 7 and 8) [61, 68]. Changes in the microglia phagocytic state (e.g., increase in cell body size and decrease in process length) and increased microglia abundance in hippocampus could lead to abnormal brain pathologies. For instance, in a mouse model of Parkinson’s disease, increased microglial phagocytic activity and cell density induce synapse loss and upregulation of CSF1R and CSF2RA (microglia proliferation), CD68, ICAM1, and ICAM2 (microglial cell engulfment), and IL-6, IL-1β, CD11b, and TNFα (pro-inflammation molecules) in hippocampus [60]. Similarly, in humans, deregulation of CSF2RA (included in the human core sensome) might abnormally increase microglia phagocytosis and proliferation, thus leading to an exacerbation of chronic inflammation-associated brain metastasis.

Lastly, cell migration-inducing and hyaluronan-binding protein (CEMIP; not listed in the microglia sensome genes) is elevated in tumor tissues and exosomes from patients with brain metastases and predicts brain metastasis progression and decreased patient survival. Uptake of CEMIP⁺ exosomes by brain endothelial and microglial cells induces inflammation in the perivascular niche by upregulating TNF, and CCL/CXCL cytokines, known to promote brain vascular remodeling and metastasis [23].

Altogether, a decrease in sensome component expression could be associated with neurodegenerative disease development or tumor growth [69‐72] and even with brain metastases. Importantly, when translating mouse results to humans, the similarities and differences between these species must be taken into account.

Recently, in a xenograft lung-to-brain metastasis model, bulk and scRNA-seq data confirmed the functional heterogeneity of microglia and infiltrating macrophages, based on their origin. This suggests that several immune subsets coexist in the same diseased brain, but they exhibit different functions [18, 19, 73]. Other scRNA-seq analyses performed in primary brain tumor samples shed light on the phenotypic heterogeneity of brain tumor-infiltrating microglia. For instance, the first scRNA-seq analysis of tumor-infiltrating myeloid cells in dehydrogenase (IDH)-mutant adult glioblastoma samples found a microglia to macrophage-like cell phenotypic spectrum based on the gradual expression of their markers [74]. Using marker genes identified in murine glioma models, another scRNA-seq analysis described different signatures of microglia and macrophage subpopulations with a predominance of the M2 phenotype [75]. Moreover, using a multimodal single-cell analysis of the tumor microenvironment, Guldner et al. identified a heterogeneous, but spatially defined CNS myeloid response during brain metastasis growth, mostly promoted by microglia with a typical signature in which the homeostatic markers CX3CR1 and TMEM119 are downregulated. This leads to the enrichment of the interferon response and to CXCL10 upregulation that promote the pro-metastatic state maintenance and the immune suppressive niche via the recruitment of resident VISTA^hi and PD-L1⁺ myeloid cells to the metastatic site [76]. This spatial transcriptomic study allowed exploring the spatial location of microglia at high resolution in the context of brain metastases. Using a human brain metastasis tissue array, hypertrophic Iba1⁺ myeloid cells (first identified as microglia) were observed in the periphery of and within the human brain metastasis samples, independently of the primary tumor origin. The spatially defined morphological patterns of Iba1⁺ myeloid cells in brain metastases was also confirmed in a mouse model of brain metastases, irrespective of the brain topography and disease stage. Compared with control (healthy) brain, most Iba⁺ myeloid cells in brain metastases were hypertrophic with enlarged cell bodies and reduced protrusion features, indicating activation and a potent response. Principal component analysis of eight morphological features correlated with the distance between microglia and tumors (body volume, cell volume, distance to the nearest cell, number of protrusions, roundness, protrusion volume, protrusion width, total protrusion length) defined morphology scores that differentiated two distinct groups: naive myeloid cells with low morphology scores and brain metastasis-associated myeloid cells with higher morphology scores. Morphology scores increased in myeloid cells close to the brain metastatic lesion borders and were highest in myeloid cells within these lesions, suggesting their activation [76, 77].

In addition, single-cell proteomics and protein expression of human microglia have been evaluated by CyTOF that has larger panels compared with flow cytometry. Recently, CyTOF analysis of 74 immune cell parameters to describe leukocytes in the microenvironment of human glioma and brain metastases showed a clear distinction between glioma and brain metastasis samples. The glioma microenvironment presented predominantly reactive microglia. Conversely, tissue-invading leukocytes accumulated in brain metastases, with a preferential localization of infiltrating macrophages within the tumor core and of microglia in the tumor periphery [78] (Fig. 2B). Thorough investigations are needed to determine whether this regional specificity reflects a site-specific function or a particular vulnerability. Altogether, single-cell omics have given a clearer and at higher resolution picture of microglia functional and regional specialization in the context of brain tumors and metastases to the CNS.