Microglia-mediated synaptic elimination in neuronal development and disease

Abstract

It has recently become clear that microglia, the immune cells of the central nervous system, are far more active in the healthy brain than previously thought. Microglia facilitate many stages of brain development by shaping neuronal connectivity via synaptic elimination. Dysfunction of these same processes likely underlies a wide range of neurological and neurodevelopmental disorders.

microglia are the resident immune cells of the central nervous system (CNS), acting as phagocytic scavengers (Kettenmann et al. 2011). Because microglia become activated in response to pathogenic insults and form the brain’s initial immune response to injury, the role of microglia in disease processes has been extensively studied (Kettenmann et al. 2011). However, relatively little is known about microglia in nonpathological states. Furthermore, microglia are exquisitely sensitive to perturbations in the surrounding environment (Kettenmann et al. 2011). As a consequence, the in vivo phenotype of microglia may not be adequately recapitulated in model systems. For these reasons, one major question remained unanswered: how do microglia function in the intact brain? The use of novel in vivo techniques revealed that microglia are much more active in the noninjured CNS than previously thought (Davalos et al. 2005; Nimmerjahn et al. 2005). Further studies have revealed that microglia actively shape developing neuronal networks, likely through phagocytosis of synapses (Paolicelli et al. 2011; Schafer et al. 2012; Sipe et al. 2016). These findings have yielded new hypotheses about the role of microglial dysfunction in the pathogenesis of diverse neurological and neurodevelopmental diseases. However, relatively little is known about the specific molecular mechanisms underlying these processes.

The identification of three major steps in microglia-mediated synaptic elimination may help to provide a framework to better understand this process: 1) How are synapses selected for elimination? 2) How do microglia find those synapses? 3) How do microglia engulf and digest those synapses? This review describes our current progress toward answering each of these questions and discusses their relevance to neural development and disease.

Synapse Selection in Development and Plasticity

During normal hippocampal development, extensive synaptic pruning is required to form the synaptic architecture necessary for mature function. Recent studies have identified microglia as critical mediators of this process due to their ability to phagocytize synapses (Paolicelli et al. 2011; Zhan et al. 2014). Using a combination of immunolabeling and two separate imaging techniques with high spatial resolution (electron microscopy and superresolution fluorescent microscopy), Paolicelli et al. (2011) directly visualized engulfed synaptic material within microglia of the developing hippocampus. Furthermore, mice lacking a critical protein implicated in microglial migration and phagocytosis (CX3CR1) display hallmarks of immature connectivity within the hippocampus, including defective synaptic pruning (Paolicelli et al. 2011). Examination of the ultrastructure of hippocampal synapses in CX3CR1 knockout (KO) mice during adulthood revealed a decreased density of multisynaptic boutons (MSBs), specifically those connected to the same dendrite, which serve to increase the synaptic input strength of a given axon (Zhan et al. 2014). One possible explanation for this finding is that microglia eliminate excess weaker unisynaptic boutons to promote the formation of the stronger MSBs. The structural and electrophysiological changes seen at the synaptic level in CX3CR1 KO mice correspond to reduced functional connectivity between different brain regions (as measured by functional MRI) and decreased social interest, qualities reminiscent of neurodevelopmental disorders such as autism (Zhan et al. 2014). These changes could underlie the observed developmental defects in CX3CR1 KO mice, but more work is needed to uncover the precise role of CX3CR1 in hippocampal development.

Similar patterns of microglial function have been identified in the lateral geniculate nucleus (LGN), the thalamic relay station for the visual system. In this system, microglia phagocytize synapses in an activity-dependent manner (Schafer et al. 2012). Retinal ganglion cells (RGCs) synapse onto neurons in the LGN, which in turn project to the visual cortex. Within the first postnatal week in mice, the number of axonal inputs to each LGN neuron decreases, and eye-specific domains begin to form in the LGN (for references see Stevens et al. 2007). Microglia in the LGN engulf RGC inputs during this window, as shown by the identification of RGC axon terminals within microglial lysosomes (Schafer et al. 2012). When RGC activity is pharmacologically stimulated or suppressed unilaterally, inputs from the “less active” eye are preferentially engulfed by LGN microglia (Schafer et al. 2012). These findings suggest that the plasticity of the visual relay system requires activity-dependent microglial phagocytosis.

Ocular dominance plasticity (ODP), a form of experience-dependent plasticity in the visual cortex, also requires microglial function, although it occurs through a different mechanism than synaptic refinement in the LGN (Sipe et al. 2016). ODP refers to the shift in the responsiveness of the binocular primary visual cortex (V1b) to inputs from each eye following monocular deprivation (MD). Normally, V1b receives visual inputs from both eyes, albeit with a bias toward the contralateral eye. During a critical period surrounding postnatal day 28 in the mouse, MD leads to changes in connectivity in the contralateral V1b (Smith et al. 2009). The input strength from the deprived eye decreases in a long-term depression (LTD)-like process by day 4 of MD, followed by an increase in input strength from the nondeprived eye (Smith et al. 2009). Unlike the plasticity in the LGN, simply suppressing RGC activity with tetrodotoxin does not induce the LTD-phase of ODP in the visual cortex (Frenkel and Bear 2004). Rather, it is the residual but uncoordinated electrical activity in the closed eye that is key to this process (Frenkel and Bear 2004). Despite differences in the mechanisms that underlie synaptic selection in each of these systems, it is clear that impairments of microglial function negatively impact neuronal plasticity in all three critical functional networks.

Microglial Chemotaxis

Before microglia can phagocytize selected synapses, they must find them. Similar to peripheral immune cells, microglia are responsive to many chemotactic agents, including ATP (Kettenmann et al. 2011). Microglia respond within minutes to focal laser-ablation injury in vivo by extending their processes toward the site of injury, possibly to wall off the damage (Davalos et al. 2005). The microglia-specific P2Y12 receptor, a Gi-coupled ADP receptor, is required to sense the ATP gradients generated from local CNS injury and thus mediates the chemotactic response (Haynes et al. 2006). Astrocytes may also play a role by amplifying the initial signal via ATP-induced ATP release, suggesting a coordinated glial response to acute neuronal injury (Davalos et al. 2005). Given that P2Y12 is also required for ODP in the visual cortex, it is plausible that similar mechanisms are present in this form of synaptic plasticity (Sipe et al. 2016). Theoretically, a relatively small amount of ATP released from neuronal synapses could be greatly amplified by astrocytes, making ATP an ideal long-range chemotactic molecule.

Recent work has shed light on the signaling cascades downstream of P2Y12 in microglia. P2Y12 activation triggers an outward potassium current (I_K), which is required for microglial chemotaxis (Swiatkowski et al. 2016; Wu et al. 2007). While the role of this I_K is not clear beyond its requirement for chemotaxis, ion currents in migrating cells are tightly coupled to changes in the cytoskeleton to facilitate redistribution of cellular volume (for references, see Kettenmann et al. 2011). Another important target downstream of P2Y12 and other Gi-activating chemoattractants is phosphatidylinositide 3-kinase gamma (PI3Kγ), which is activated by the Gβγ subunit of the Gi heterotrimer (Irino et al. 2008). PI3Kγ phosphorylates the membrane phospholipid PIP₂, forming PIP₃, which regulates cytoskeletal dynamics and many other processes. An extracellular chemokine gradient results in an asymmetric distribution of PIP₃ in the membrane, allowing for the subsequent polarization of migrating cells (for references, see Schneble et al. 2017). It has been shown in vitro that norepinephrine, acting on Gs-coupled β-adrenergic receptors, antagonizes the chemoattractive effects of both ATP and the complement fragment C5a (Gyoneva and Traynelis 2013; Schneble et al. 2017). This occurs via PKA-dependent phosphorylation and inhibition of PI3Kγ (Schneble et al. 2017). The balance between Gi and Gs signaling likely contributes to the regulation of microglial motility in vivo, considering ATP and norepinephrine are abundant neurotransmitters in the brain and their receptors are expressed on microglia (Kettenmann et al. 2011).

To the author’s knowledge, the only study examining microglial process chemotaxis in normal development was that of Sipe et al. (2016), which showed that P2Y12 is required for ODP. Within 12 h of MD, before any changes in neuronal activity can be observed, microglial processes become markedly elaborated (Sipe et al. 2016). Additionally, microglia make more contacts with synaptic clefts and contain an increased number of engulfed synaptic receptors following MD, a phenomenon observed in WT but not P2Y12 KO mice (Sipe et al. 2016). These results are consistent with the hypothesis that microglial processes locate synapses selected for degradation in ODP by ATP-mediated chemotaxis. A role for P2Y12-mediated microglial chemotaxis has also been suggested in seizures (Eyo et al. 2014). Excessive excitatory neurotransmission in the hippocampus leads to microglial process extension toward neuronal cell bodies in a P2Y12-dependent manner (Eyo et al. 2014). In fact, P2Y12 KO mice have an increased seizure burden following kainate-induced seizures, suggestive of a neuroprotective role for microglial P2Y12 in this context (Eyo et al. 2014). Going forward, it will be important to determine the signaling mechanisms both upstream and downstream of P2Y12 activation in vivo as well as the developmental and pathological conditions that activate microglial chemotaxis, via ATP or other chemotactic agents.

Phagocytosis and Degradation

After selecting and migrating toward the appropriate synapse, the final step in the process is phagocytosis. The complement system provides a mechanism by which synapses could be marked for microglial phagocytosis. The complement cascade is a component of the immune system that allows for opsonization and subsequent clearance of invading pathogens and foreign antigens. In the classical complement activation pathway, C1q protein binds to the surface of the target and a sequence of proteolytic reactions leads to the covalent bonding of C3b proteins to the same target (for references, see Stevens et al. 2007). Complement receptors (such as CR3) expressed on the surface of phagocytes can then bind C3b and promote phagocytosis of the opsonized pathogen or antigen. Microglia are the only cells in the CNS that express the complement receptor CR3 and thus could theoretically bind and phagocytize synapses marked by C3 (for references, see Schafer et al. 2012). In fact, genetic disruption of components of the complement cascade blocks microglial phagocytosis of synapses and eye-specific segregation in the LGN and leads to more numerous and weaker inputs into the LGN (Schafer et al. 2012; Stevens et al. 2007).

Complement-dependent synaptic elimination has also been implicated in the pathogenesis of Alzheimer Disease (AD; Hong et al. 2016; Shi et al. 2017). In AD, cognitive decline is most closely associated with synapse loss, rather than the pathological hallmarks of the disease (i.e., amyloid-beta (Aβ) plaque deposition and neurofibrillary tangles (Terry et al. 1991). In a mouse model of familial AD, synapse loss is observed by 3–4 mo of age, before detectable plaque deposition (Hong et al. 2016). In these mice, C1q expression is observed by 1 mo of age, a time point that precedes the loss of synapses (Hong et al. 2016). Intraventricular injection of soluble oligomeric Aβ (oAβ) in healthy mice also produces an AD-like phenotype with extensive loss of synapses (Hong et al. 2016). In this model, C1q expression is precipitated by the injection of oAβ (Hong et al. 2016). Consistent with findings in the LGN, the synaptic loss observed in multiple models of AD is decreased by blocking the function of C1q, C3, or CR3, suggestive of impaired microglial phagocytosis (Hong et al. 2016). In a mouse model of later-stage AD, genetic deletion of C3 abrogated the cognitive decline, synapse loss, gliosis, and inflammation associated with control animals despite an increased plaque burden (Shi et al. 2017). Together, these studies suggest that activation of the complement cascade is a key step in the progress of synapse loss and, by extension, the clinical symptomatology of AD. Unlike in the developing LGN, in which C1q is expressed by RGCs, C1q mRNA localizes to microglia in mouse models of AD (Hong et al. 2016). Abundant expression of C1q in microglia, rather than specific subsets of neurons, could theoretically account for nonspecific loss of synapses in AD as compared with a precisely coordinated process during development. Alongside this pioneering work in AD, similar work has emerged implicating complement-dependent synaptic pruning in the pathophysiology of frontotemporal lobar dementia (Lui et al. 2016), West Nile virus-associated neurodegeneration (Vasek et al. 2016), schizophrenia (Sekar et al. 2016), and epilepsy (Chu et al. 2010). Aberrant regulation of the complement cascade may therefore represent a shared etiological pathway in many common yet poorly understood neurological diseases.

After phagocytosis, synaptic material must be digested by microglia. The autophagy regulator Atg7 is important in microglial digestion of synaptic material during development (Kim et al. 2016). Using a transgenic mouse in which Atg7 is specifically deleted in microglia, Kim et al. (2016) showed that the lack of Atg7 leads to impaired digestion of synaptosomes by microglia in vitro and impaired synaptic phagocytosis by microglia in neuronal-microglial cocultures. In both systems, excessive synaptic proteins accumulated within the microglial cell body (Kim et al. 2016). Correspondingly, microglia-specific Atg7-deficient mice in vivo have an increased number of dendritic spines and synapses in cortex and display altered social and repetitive task behaviors, suggestive of an autistic-like phenotype (Kim et al. 2016). This study demonstrates that microglial autophagy is critical for proper synaptic development, opening up a new direction for future research. Furthermore, this study provides another mouse model in which impaired microglia-mediated synaptic pruning leads to an autism-like phenotype.

Conclusion

It has become clear that microglia play a much larger role in neuronal development and pathology than previously thought. Specifically, microglia directly shape synaptic networks through the elimination of synapses, with implications for both normal development and disease. A focus on microglia as possible etiologic agents in a wide range of diseases of the CNS may lead to novel therapeutic targets in these diseases. In parallel, much more work is needed to delineate the basic mechanisms involved in microglia-mediated synaptic elimination.

GRANTS

B. S. Whitelaw is supported by the Medical Scientist Training Program at the University of Rochester School of Medicine and Dentistry (NIH T32 CM07356).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author.

AUTHOR CONTRIBUTIONS

B.S.W. drafted manuscript; B.S.W. edited and revised manuscript; B.S.W. approved final version of manuscript.