Synaptic plasticity
Most synaptic plasticity occurs at glutamatergic synapses, where transmission is mediated by ionotropic receptors (GluN, GluA and GluK) and metabotropic receptors (mGlu). Transporters, expressed on neurons and astrocytes, limit the activation of glutamate receptors. All these receptors are subject to many post-translational modifications, amongst which two of the most important are protein/protein interactions and phosphorylation.
One main role of cytokines in synaptic transmission is their ability to modulate the induction of synaptic plasticity. IL-1β, IL-2, IL-6, IL-8, IL-18, IFNα, IFNγ and TNFα have all been shown to be able to inhibit LTP and to induce changes in hippocampal dependent learning and memory tasks [
40‐
42]. LTD can also be inhibited either directly by cytokines like IL-1β or during the inflammation process [
40,
43]. Conversely, under physiological conditions, activation of the JAK2/STAT3 pathway is required for the induction of GluN-dependent LTD in the hippocampus [
12]. Further the induction of LTP increases the expression of IL-6 which acts in a negative feedback manner to limit the magnitude of potentiation [
44,
45]. These results demonstrate that cytokine signaling not only acts in a metaplastic manner to modulate bidirectional changes in synaptic efficacy, but is also part of the physiological mechanism.
The IL-1β receptor has been shown to physically interact with GluN receptors enabling the rapid regulation of GluN activity, via Src-dependent phosphorylation events [
46]. The IL-1β receptor can also decrease GluA surface expression [
47]. TNFα can promote GluA dependent activity in hippocampal neuron and may induce GluA internalization in striatal GABAergic neurons (reviewed in [
48]).
The mechanisms by which cytokines or inflammation alter synaptic function is complex as microglia by themselves can directly facilitate synaptic strength independently of any changes in synaptic activity. Once activated, they can, for example, induce STP via IL-1β dependent mechanisms [
49] or LTD in pathological context [
50].
In addition to cytokines, major histocompatibility complex (MHC) class I molecules also play a role in modulating the induction of synaptic plasticity. MHC class I molecules are a group of proteins which, within the immune system, translocate cytosolic peptides generated by proteasome-mediated degradation to the cell surface for recognition and subsequent cell elimination by cytotoxic T cells [
51]. MHC class I molecules have also been found to be expressed by neurons in the CNS and localize to synapses, axon terminals and dendrites [
52‐
56]. In the dorsal lateral geniculate nucleus (dLGN), the MHC class I molecule H2-D
b was found to be necessary in limiting the synaptic incorporation of calcium-permeable AMPARs and thus permit the induction of LTD [
57]. In the hippocampus, β
2M−/− TAP−/− mice (which lack cell surface expression of MHC class I molecules) have a shift in LTP threshold in area CA1, such that low-frequency stimulation, which usually induces LTD, instead causes LTP, and LTP induced by high-frequency stimulation is larger in magnitude [
54,
58]. Finally, MHC class I acts as a negative regulator of synapse density in both the cortex and hippocampal area CA3 [
59,
60], which, in area CA3, is the result of a MHC class I-insulin receptor complex which constrains basal insulin receptor signaling [
60].
Finally, the complement system, which is part of the innate immune system, and in particular the complement component C3, could also play a direct role by modulating the efficiency of glutamatergic synaptic transmission in the absence of any inflammation process, by a mechanism not yet explored [
61], but that could involve synaptic stripping [
62], a process initially defined as the removal of dysfunctional synapses by activated microglia [
63].
Structural plasticity
Organisms are born with an excess number of synapses throughout the CNS, and during development superfluous connections are removed in an experience-dependent manner, a process known as synaptic pruning [
64]. Within the CNS synaptic pruning has been extensively studied in the visual system, specifically the LGN and the striate cortex. Initially, LGN neurons receive inputs from multiple retinal ganglion cells (RGCs), however during postnatal development inputs are selectively removed such that eventually each LGN neuron only receives input from one or two RGCs [
65]. Additionally, monocular deprivation (MD) during the developmental critical period can lead to reduced responsiveness to the deprived eye in the LGN and striate cortex as deprived eye inputs are weakened [
66]. Stevens and colleagues [
67] found that the classic complement cascade mediated the elimination of RGC inputs onto LGN neurons via microglia phagocytosis early in postnatal development (P5). A subsequent study found that complement-mediated synaptic pruning in the LGN is regulated by neuronal activity, as inhibiting activity in one eye with TTX increased microglia engulfment of that eye’s inputs, while the converse occurred when RGC activity in one eye was stimulated with forskolin [
68]. The developing visual system model has also revealed the involvement of other immune pathways in synaptic pruning including purinergic signaling with microglia via the P2Y12 receptor [
69] and MHC class I proteins [
57].
In the developing hippocampus it has been demonstrated that puncta of the postsynaptic protein PSD-95 are contained within microglia, providing evidence of microglia-mediated synaptic pruning. Further, knocking out the microglia-specific fractalkine receptor CX3CR1 resulted in an increased spine density in neonatal mice [
70]. It was also found that CX3CR1 KO mice had impaired synapse maturation [
70,
71], assayed by measuring the number of release sites per neuron-neuron connection, and thus the authors proposed that that synapse elimination allows for subsequent strengthening of remaining synapses [
71]. However, CX3CR1 KO mice were also found to have enhanced IL-1β levels which resulted in a specific impairment of LTP [
72], suggesting that the effects of deficient synaptic pruning and synapse maturation observed in CX3CR1 KO mice may not be directly linked.
It has been proposed that weakened synapses are subsequently ‘tagged’ with complement proteins to induce microglia phagocytosis [
73]. Synapse elimination has been found to occur in the hippocampus following both mGlu- and GluN-dependent LTD [
74‐
79]. Thus, future studies should seek to directly examine whether there is an interaction between synaptic depression and microglia phagocytosis.
In the adult brain, activation of microglia leads to the displacement of inhibitory synapses from the soma of neurons and is neuroprotective by a mechanism involving GluN activation [
80‐
82]. Synaptic stripping by microglia participates in network remodeling but its exact role in pathology remains to be fully demonstrated [
83].
Synaptic pruning can also occur independently of any physical interaction with glia. This mechanism has been well described for neurons with axon injury and occurs concomitantly with glial cell activation. Astrocytes as well as microglia release numerous factors (cytokines, chemokines, thrombospondins, etc.) that directly influence synapse integrity [
84,
85].
Though research has mainly focused on microglia in relation to synapse elimination, recent advances have shed light on the role microglia also play in spine formation. The generation of CX3CR1-CreER mice allowed Parkhurst and colleagues [
86] to conditionally deplete microglia or knockout microglial BDNF. Subsequent in vivo two photon imaging revealed that these manipulations impaired spine formation in the motor cortex following motor learning. Furthermore, in vivo two photon imaging of the somatosensory cortex of developing mice revealed that microglia contact of dendrites frequently led to spine filipodia formation [
87]. Therefore, it appears that microglia can cause bidirectional modifications of dendrite spine structure but the underlying mechanisms (see also the review of Kettenman et al. [
63]) involved in this process need more investigation by future studies.
Synaptic scaling
At glutamatergic synapses, when extended periods of elevated or depressed neuronal activity occur, homeostatic mechanisms can be activated. They alter strength across all synapses to return activity to an optimal range, a process known as synaptic scaling [
88]. Chronic (24–48 h) blockade of neuronal action potential firing or glutamatergic synaptic transmission results in a large increase in mEPSC amplitude (a putative measure of post-synaptic sensitivity to glutamate) as well as the number of surface AMPARs [
89]. Stellwagen and Malenka [
90] found that TNFα is both necessary and sufficient for scaling up of post-synaptic AMPARs. Interestingly, although both neurons and glia are capable of producing TNFα, the authors found that it is the glia-released TNFα that is critical for scaling up of synapses. A subsequent study found that β3 integrins are also necessary for scaling up of synapses, and that TNFα application increases the surface expression of β3 integrin [
91]. Further, the authors found that levels of β3 integrin control surface AMPARs and thus mEPSC amplitude, suggesting a model in which glial cells, in response to reduce network activity, release TNFα leading to an increase in β3 integrin surface expression and subsequent accumulation of AMPARs at the synapse [
92].
To address the physiological significance of TNFα-meditated synaptic scaling, Kaneko and colleagues [
93] examined visual system. In the striate cortex, similar to the hippocampus, LTP was normal in TNFα−/− mice, however scaling up of AMPAR-mediated mEPSCs was absent. In vivo, TNFα−/− mice had impaired ocular dominance plasticity following MD, specifically having a complete deficit in the increase of non-deprived eye cortical response, despite a normal decrease in the cortical response for the deprived eye. Thus, synaptoimmunological factors are critical for both phases of ocular dominance plasticity, with the complement cascade mediating the loss of cortical responsiveness to the deprived eye, while TNFα mediates the compensatory, homeostatic increase in non-deprived eye cortical responsiveness.