Modulation of information processing by AMPA receptor auxiliary subunits

AMPA‐type glutamate receptors (AMPARs) are key molecules of neuronal communication in our brain. The discovery of AMPAR auxiliary subunits, such as proteins of the TARP, CKAMP and CNIH families, fundamentally changed our understanding of how AMPAR function is regulated. Auxiliary subunits control almost all aspects of AMPAR function in the brain. They influence AMPAR assembly, composition, structure, trafficking, subcellular localization and gating. This influence has important implications for synapse function. In the present review, we first discuss how auxiliary subunits affect the strength of synapses by modulating number and localization of AMPARs in synapses as well as their glutamate affinity, conductance and peak open probability. Next we explain how the presence of auxiliary subunits alters temporal precision and integrative properties of synapses by influencing gating kinetics of the receptors. Auxiliary subunits of the TARP and CKAMP family modulate synaptic short‐term plasticity by increasing anchoring of AMPARs in synapses and by altering their desensitization kinetics. We then describe how auxiliary subunits of the TARP, CKAMP and CNIH families are involved in Hebbian and homeostatic plasticity, which can be explained by their influence on surface trafficking and synaptic targeting. In conclusion, the series of studies covered in this review show that auxiliary subunits play a pivotal role in controlling information processing in the brain by modulating synaptic computation.


Introduction
Synaptic transmission via chemical synapses is the main route of neuronal communication and forms the backbone of information processing in the central nervous system. Plastic changes in strength and mode of single synapses alter the way information is integrated and computed in our brain. Fast excitatory synaptic transmission in the central nervous system is mostly mediated by AMPA-type glutamate receptors (AMPA receptors). Their central position in excitatory synapses makes these receptors key targets for the regulation of excitatory synaptic communication. Consequently, alteration in neuronal communication is often mediated by a change in number and function of AMPA receptors.
Structurally, AMPA receptors are heterotetramers that assemble from different combinations of the four AMPA receptor subunits GluA1-GluA4 (AMPA-type glutamate receptor subunit 1-4) (Traynelis et al. 2010). The combination of the subunits in the final receptor determines its basic electrophysiological parameters. Moreover, AMPA receptors are embedded into networks of interacting proteins several of which strongly impact AMPA receptor function (Schwenk et al. 2009 (CNIHs), 2012 (CNIHs, TARPs, GSG1L); Engelhardt et al. 2010 (CKAMP44); Shanks et al. 2012 (GSG1L); Greger et al. 2017 (TARPs)). The importance of these auxiliary subunits becomes evident if one looks at the broad variety of AMPA receptor-mediated currents in various neuron types in the brain. These currents only partially resemble the ones from heterologous expressed receptors. This discrepancy can be explained by the interaction of auxiliary subunits with AMPA receptors in the brain. In fact, we know today that AMPA receptor function depends to a large extent on the interaction with auxiliary subunits. They influence AMPA receptor trafficking to the cell surface, subcellular localization and gating of the receptors (see Table 1).
Currently, the group of auxiliary subunits of AMPA receptors includes the families of TARPs (transmembrane AMPA receptor regulatory proteins), CNIHs (cornichon homolog proteins), CKAMPs (cysteine-knot AMPA receptor modulating proteins; aka Shisas) and GSG1L (germ cell-specific gene 1 like protein) ( Figure 1, Table 1). Auxiliary subunits differ in their influence on AMPA receptor function and, additionally, display distinct regional and developmental expression profiles. Interestingly, the expression of some auxiliary subunits is activity dependent. This indicates an important role of auxiliary subunits in homeostatic synaptic scaling and neuronal adaptation processes. Several recent reviews discuss in detail the influence of auxiliary subunits on AMPA receptor trafficking, assembly, receptor composition, and structure (Eibl & Plested, 2017;Greger et al. 2017;Jacobi & Engelhardt, 2017, 2018Bissen et al. 2019;Chen & Gouaux, 2019;Kamalova & Nakagawa, 2021). This review will specifically focus on how AMPA receptor auxiliary subunits influence function and localization of synaptic AMPA receptors and the consequences of this influence on computation of synapses.

Synaptic strength
The strength of a synapse depends on the number of synaptic AMPA receptors (Fig. 2). Throughout the brain, Opposing effects are probably due to different AMPA receptor subunits or expression systems (e.g. oocytes, HEK293 cells, cultured neurons or acute brain slices). N/A, not available. References: TARPs: Tomita et al. 2003;Yamazaki et al. 2004;Priel et al. 2005;Rouach et al. 2005;Cho et al. 2007;Milstein et al. 2007;Kott et al. 2009;Shi et al. 2010;Kato et al. 2008Kato et al. , 2010Khodosevich et al. 2014;CKAMPs: von Engelhardt et al. 2010;Khodosevich et al. 2014;Farrow et al. 2015;Klaassen et al. 2016;Schmitz et al. 2017;CNIHs: Schwenk et al. 2009;Coombs et al. 2012;Kato et al. 2010;Herring et al. 2013;Boudkkazi et al. 2014;GSG1L: Shanks et al. 2012;McGee et al. 2015;Gu et al. 2016;Mao et al. 2017. synaptic AMPA receptor density varies from synapse to synapse and is subject to constant changes. The control of synaptic AMPA receptor number is perhaps one of the most obvious effects of auxiliary subunits. However, the strength of a synapse depends not only on the number of synaptic receptors, but also on their gating properties. Thus, strength of synaptic communication is additionally influenced by glutamate affinity, conductance and peak open probability of AMPA receptors (Fig. 2;Clements, 1996;Kullmann et al. 1999;MacGillavry et al. 2013). Finally, anchoring of AMPA receptors in subsynaptic nanodomains has been shown to influence synaptic strength (Nair et al. 2013). All these receptor properties are differentially influenced and/or controlled by auxiliary subunits, making these proteins key regulators of synaptic strength (Table 1). The regulation of AMPA receptor density by auxiliary subunits does not depend on a single, uniform mechanism. Some auxiliary subunits affect the trafficking of AMPA receptors to the cell surface, others (and sometimes the same ones) control the anchoring of AMPA receptors at the postsynaptic density. Yet others are thought to affect removal of AMPA receptors from the synapse. To date, the mechanisms of how AMPA receptor auxiliary subunits control receptor trafficking to the cell surface are not fully understood (reviewed in Jacobi & Engelhardt, 2018). Genetic deletion of many auxiliary subunits leads to a significant reduction in surface AMPA receptor levels. This reduction of AMPA receptor number is especially dramatic in TARP γ-2-deficient stargazer mice, where the deletion of TARP γ-2 leads to a total loss of surface AMPA receptors on cerebellar granule cells in postnatal day 14 mice (Chen et al. 2000). Deletion of other auxiliary subunits shows a less severe impact on AMPA receptor surface levels, presumably due to the high functional and spatial redundancy of large Additionally, highlighted are locations of specific domains of the different protein families: the PDZ binding motifs for TARPs and CKAMPs, the C-terminal phosphorylation site of TARPs (that interacts with the cell membrane and influences synaptic anchoring) and the characteristic region of CKAMPs (cysteine-rich region) and CNIH-2/3 (adapted from Kamalova &Nakagawa, 2021). J Physiol 599.2 groups of auxiliary subunits. Hence, deletion of CKAMP44 or TARP γ-8 reduces somatic AMPA receptor currents by approximately 50% in dentate gyrus granule cells. However, the deletion of both auxiliary subunits reduces somatic AMPA receptor currents to approximately 8% (Khodosevich et al. 2014). Based on the numbers, the deletion of both subunits seems to be additive. In contrast, in cerebellar Golgi cells synaptic AMPA receptor number is not affected by deletion of TARP γ-2 or TARP γ-3 alone, but is virtually absent in TARP γ-2/γ-3 double knockout mice (estimated from synaptic current amplitudes). This suggests that the two auxiliary subunits are functionally redundant and can compensate for the loss of the other auxiliary subunit (Menuz et al. 2008). On the other hand, TARP γ-3 is not expressed in cerebellar granule cells Fukaya et al. 2005), which explains why it cannot compensate for the loss of TARP γ-2 in this cell type. Interestingly, cerebellar granule cells express TARP γ-7. Knockdown of TARP γ-7 in stargazer mice rescues synaptic currents, suggesting that TARP γ-7 prevents synaptic localization of AMPA receptors in the absence of TARP γ-2 (Bats et al. 2012). Hence, the effect of the genetic deletion of an AMPA receptor auxiliary subunit depends on the presence of other functionally redundant subunits.
The amplitude of AMPA receptor-mediated currents depends not only on the number of postsynaptic receptors, but also on their affinity to glutamate. AMPA receptors show a relatively low glutamate affinity compared to other glutamate receptors, especially the NMDA receptor-type (Liu et al. 1999). Peak synaptic glutamate concentration has been estimated to be in the millimolar range Jonas, 2000). However, peak glutamate concentration rapidly decreases with increasing distance from the presynaptic release site. Simulations and experimental data show that synaptic glutamate concentration is non-saturating, i.e. results in the opening of only a fraction of the postsynaptic AMPA receptors (Liu et al. 1999;Rusakov et al. 1999;Jonas, 2000;McAllister & Stevens, 2000;Wu et al. 2007). Consequently, synaptic strength should depend on glutamate affinity of AMPA receptors. As a corollary, one would assume that auxiliary subunits may influence synaptic strength not only by controlling the number of synaptic AMPA receptors, but also by modulating their glutamate affinity. Several auxiliary subunits of the CKAMP-, CNIH-and TARP-families increase glutamate affinity (Yamazaki et al. 2004;Tomita et al. 2005a;Coombs et al. 2012;Khodosevich et al. 2014;Farrow et al. 2015). An exception is TARP γ-5, which decreases glutamate affinity of GluA2-containing AMPA receptors (Kato et al. 2008). Although there is no direct experimental evidence, it is therefore likely that the change in synaptic strength in mice with genetic deletion of auxiliary subunits results not only from an alteration in AMPA receptor number but also from a change in glutamate affinity (Rouach et al. 2005;Tomita et al. 2005a;Menuz et al. 2008;Coombs et al. 2012;Herring et al. 2013;Khodosevich et al. 2014;Chen et al. 2018).
The strong decline of the glutamate concentration with growing distance from the vesicle release site explains the relevance of the precise subsynaptic position of AMPA receptors for synaptic strength. In fact, AMPA receptors do not distribute uniformly in the synapse, but cluster in nanodomains (Masugi-Tokita et al. 2007;MacGillavry et al. 2013). Interaction of AMPA receptors with membrane-associated guanylate kinases (MAGUKs) such as PSD95 explains low diffusion rates and hence long dwell times of AMPA receptors in nanodomains (Nair et al. 2013). In addition, the N-terminal domain is important for synaptic localization of AMPA receptors (Díaz-Alonso et al. 2017;Watson et al. 2017). Importantly, the nanodomains are in close proximity with presynaptic vesicle release sites (Tang et al. 2016), a structural organization that is most likely mediated by trans-synaptic protein-protein interaction, e.g. between Neuroligin and Neurexin. This alignment of a postsynaptic nanodomain with a presynaptic release site in, so called, nanocolumns ensures high glutamate concentration at the postsynaptic site of AMPA receptor anchoring. Consistently, disruption of nanocolumns by expression of a truncated form of NLG1 reduces synaptic strength (Haas et al. 2018). Auxiliary subunits of the CKAMP and TARP families interact with MAGUKs, and in particular with PSD95 (Dakoji et al. 2003;Khodosevich et al. 2014;Klaassen et al. 2016;Schmitz et al. 2017). CKAMPs and TARPs therefore increase synaptic strength not only by augmenting synaptic number of AMPA receptors but most likely also by anchoring AMPA receptors in nanodomains in close vicinity of presynaptic vesicle release sites.
Finally, the strength of a synapse depends also on the conductance and peak open probability of its receptors. Most auxiliary subunits increase AMPA receptor conductance and/or peak open probability (Tomita et al. 2005a;Cho et al. 2007;Schwenk et al. 2009;Pierce & Niu, 2019). One exception is GSG1L, which decreases synaptic strength in cerebellar and hippocampal neurons, presumably by decreasing the synaptic AMPA receptor density and channel conductance of calcium permeable receptors (McGee et al. 2015;Gu et al. 2016;Mao et al. 2017).

AMPA receptor kinetics
The computation of excitatory synapses depends not only on the peak size of the depolarizing current but also on its shape. Hence, rise time, deactivation and desensitization kinetics of synaptic AMPA receptors determine the charge transfer and timing of synaptic currents and therefore directly affect synaptic communication. Core subunit composition of AMPA receptors (i.e. GluA1-4, flip/flop) determines gating kinetics (Mosbacher et al. 1994;Traynelis et al. 2010). However, especially deactivation and desensitization rates of AMPA receptors are more strongly affected by the presence of auxiliary subunits. Most known auxiliary subunits decrease the deactivation rate of AMPA receptors, with the exception of TARP γ-5, which increases the deactivation rate. The desensitization rate and/or the steady-state desensitization of AMPA receptors is decreased by TARPs and CNIHs but increased by CKAMPs and GSG1L (Menuz et al. 2008;Schwenk et al. 2009Schwenk et al. , 2012Shanks et al. 2012;Straub & Tomita, 2012;Boudkkazi et al. 2014;Khodosevich et al. 2014). The deactivation time constants of heterologously expressed AMPA receptors without auxiliary subunits are in the range of 0.7 ms (homomeric GluA2o) and 1.3 ms (homomeric GluA1o), i.e. a difference of 600 μs. In contrast, incorporation of TARP γ-8 or CNIH-2 into AMPA receptors increases the deactivation time constant of homomeric GluA1 receptors to ca 5 ms and 9 ms, respectively (Kato et al. 2010). Similarly, desensitization time constants are several-fold larger in TARP γ-8 or CNIH-2-containing GluA1 receptors compared to pure homomeric GluA1 receptors (Kato et al. 2010). Deactivation and desensitization time constants of AMPA receptors are, in most neurons, considerably slower than those of heterologously expressed receptors. For example, principal cells of the hippocampus display deactivation time constants in the range of 2.3 ms (dentate gyrus) and 3 ms (CA1, Colquhoun et al. 1992). The most likely explanation for these slow kinetics is the presence of AMPA receptor complexes that contain auxiliary subunits. Indeed, genetic deletion of TARP γ-8, CKAMP44, CKAMP52, GSG1L and CNIH-2 and 3 decreases deactivation time constants of AMPA receptors in CA1 neurons and dentate gyrus granule cells (Rouach et al. 2005;Herring et al. 2013;Boudkkazi et al. 2014;Khodosevich et al. 2014;Gu et al. 2016;Klaassen et al. 2016). Importantly, the magnitude of the influence of auxiliary subunits on gating kinetics strongly depends on the AMPA receptor composition and presence of the flip/flop cassette (Turetsky et al. 2005;Tomita et al. 2005aTomita et al. , 2007Kato et al. 2007Kato et al. , 2010Kott et al. 2007;Morimoto-Tomita et al. 2009;Dawe et al. 2019) Experiments with fast application of glutamate onto patches of neurons therefore indeed showed that auxiliary subunits influence gating properties of AMPA receptors in the brain, such as deactivation and desensitization. In most neurons, decays of synaptic currents are mainly dictated by deactivation kinetics of AMPA receptors. This suggests that auxiliary subunits shape decays of synaptic currents. There are a few neuron types that express AMPA receptors with gating kinetics similar to those of homomeric AMPA receptors. Particularly fast deactivation (0.5 ms) has been observed for neurons of the auditory system (Raman et al. 1994;Raman & Trussell, 1995). It is likely that AMPA receptors with such fast kinetics contain few auxiliary subunits that slow deactivation such as TARPs or CNIHs. Importantly, decay time constants of synaptic currents are also extremely fast in auditory neurons (<1 ms). Axosomatic synapses and short membrane time constants ensure little filtering of synaptic currents and explain negligible temporal summation of auditory inputs (Rothman et al. 1993;Raman et al. 1994). Several auditory neurons form giant synapses with a high number of J Physiol 599.2 AMPA receptors. The fast kinetics and large amplitude of AMPA receptor-mediated currents in auditory neurons are crucial for conveying temporal information with high fidelity (Rothman et al. 1993). Very fast EPSC decays were observed additionally in amacrine, basket and stellate cells (Geiger et al. 1997;Crowley et al. 2007;Osswald et al. 2007), suggesting that in these neurons AMPA receptor auxiliary subunits are also not strongly expressed or exert only a small influence on EPSC decays.
However, compared to the extremely fast kinetics in the aforementioned cells, AMPA receptor decay kinetics are considerably slower in most synapses. For example, AMPA receptor decay rates are around 2-7 ms in hippocampal synapses (Geiger et al. 1995;McGee et al. 2015;Klaassen et al. 2016;Schmitz et al. 2017). Consistent with the hypothesis that slow decay kinetics depend on the influence of auxiliary subunits, the deletion of GSG1L, CKAMP52 or CKAMP59 decreases decay rates in these synapses (McGee et al. 2015;Klaassen et al. 2016;Schmitz et al. 2017). Auxiliary subunits therefore alter computational properties of synapses by influencing decay kinetics. Synapses of hilar mossy cells display comparably slow decay kinetics (ca 12 ms), which is explained by a high expression of CNIH-2 in this cell type (Boudkkazi et al. 2014). The slow kinetics make these synapses less suitable for transmission of information with high temporal precision, but ideal for the integration of information.

AMPA receptor Ca 2+ permeability
Ca 2+ permeability and conductance of AMPA receptors depends on their subunit composition. Thus, GluA2-containing AMPA receptors are Ca 2+ impermeable, and GluA2-lacking AMPA receptors are Ca 2+ permeable. Moreover, the presence of the subunit GluA2 decreases the conductance (Verdoorn et al. 1991;Burnashev et al. 1992). Consequently, the composition of AMPA receptors directly affects the computation of synapses.
The regulation of AMPA receptor composition by auxiliary subunits is mainly based on the specific interaction with certain receptor subunits and takes place on different functional levels. These different levels of regulation on one hand, and the presence of more than one auxiliary subunit per cell on the other, leads to a complex and sometimes not uniform effect of auxiliary subunits on receptor composition throughout the brain. TARP γ-2, for example, specifically protects GluA1-containing receptors from lysosomal degradation and thereby alters receptor composition in CA1 neurons (Kessels & Malinow, 2009). On the other hand, enhancement of cytoplasmic polyamine block of AMPA receptors in stellate cells of TARP γ-2 knockout mice or in Golgi cells of TARP γ-2/γ-3 knockout mice suggests that TARP γ-2 and/or γ-3 increase the number of GluA2-containing AMPA receptors in these cell types (Menuz et al. 2008;Bats et al. 2012). However, since TARPs themselves attenuate the polyamine block (Soto et al. 2014;Brown et al. 2017), it is also reasonable that the change in polyamine block in TARP γ-2/γ-3 knockout mice is not only due to a decrease in GluA2-containing AMPA receptors, but also due to the loss of the direct influence of TARPs on polyamine block. TARP γ-8 controls, together with the auxiliary subunits CNIH-2 and CNIH-3, the surface levels of GluA1-containing receptors in CA1 pyramidal neurons. The presence of TARP γ-8 may prevent the interaction of CNIH-2/-3 with subunits other than GluA1 and, thus, selectively promotes the trafficking of these receptors to the cell surface (Herring et al. 2013; but see also Boudkkazi et al. 2014). Yet another mechanism of the regulation of subunit composition by auxiliary subunits has been described by McGee and colleagues (McGee et al. 2015). The auxiliary subunit GSG1L specifically supresses currents of calcium permeable AMPA receptors by decreasing their Ca 2+ -permeability and conductance. Thus, in contrast to the other auxiliary subunits, GSG1L directly supresses the function of certain receptors, rather than promoting it.

Short-term plasticity
The main determinant for synaptic short-term plasticity in many synapses is the release probability of presynaptic vesicles. However, in synapses that frequently have two consecutive vesicle releases in a short period of time, AMPA receptor desensitization is relevant for short-term plasticity in addition to the presynaptic factors. High release probability, but also spill-over of glutamate from one release site to a neighbouring release site and slow diffusion of glutamate out of a synapse favour the influence of AMPA receptor desensitization on short-term plasticity (Blitz et al. 2004). Hence, the specific geometry of a synapse and the proximity of neighbouring release sites affects short-term plasticity via the desensitization of AMPA receptors. For example, AMPA receptor desensitization alters short-term plasticity in retinogeniculate synapses of relay neurons in the lateral geniculate nucleus (Chen et al. 2002;Hauser et al. 2014). Retinogeniculate synapses are very large synapses that contain many release sites (Rafols & Valverde, 1973). This allows spill-over of glutamate from active to non-active release sites. In addition, the geometry of retinogeniculate synapses precludes fast diffusion of glutamate out of the synaptic cleft (Rafols & Valverde, 1973;Budisantoso et al. 2012). Hence, presynaptic release of glutamate effectively desensitizes AMPA receptors in active and non-active neighbouring release sites of the same synapse (Budisantoso et al. 2012).
Examining the effect of receptor desensitization on short-term plasticity more closely shows that it is especially the time course of the recovery from desensitization of AMPA receptors that affects synaptic short-term plasticity. AMPA receptor desensitization and the recovery from desensitization are differentially modulated by different auxiliary subunits (Boudkkazi et al. 2014;Khodosevich et al. 2014;Farrow et al. 2015;McGee et al. 2015). For example, while heterologously expressed AMPA receptors that contain no auxiliary subunits recover very quickly from desensitization (time constants between 16 and 44 ms for GluA2-4; GluA1i = 151 ms, GluA1o = 105 ms; Kessler et al. 2008), the presence of the two auxiliary subunits CKAMP39 and CKAMP44 strongly slows recovery from desensitization. Thus, CKAMP39 and CKAMP44 increase the time constant of recovery from desensitization of AMPA receptors ca 10-fold. The time constants from heterologously expressed CKAMP44-containing AMPA receptors is consistent with the comparably slow recovery from desensitization of AMPA receptors found in neurons with high CKAMP44 expression (Khodosevich et al. 2014;Chen et al. 2018). TARPs on the other hand, decrease the time constant of recovery from desensitization (Priel et al. 2005;Khodosevich et al. 2014). Importantly, the genetic deletion of CKAMP44 or TARP γ-8 affects short-term plasticity in the hippocampus. Thus, short-term depression is stronger in TARP γ-8 knockout mice and less pronounced in CKAMP44 knockout mice (Khodosevich et al. 2014). Short-term plasticity experiments in the presence of cyclothiazide, a potent blocker of AMPA receptor desensitization, proved that the influence of the two proteins on short-term plasticity indeed results from their effect on the rate of recovery from desensitization of AMPA receptors (Khodosevich et al. 2014).
AMPA receptor diffusion mitigates the effect of desensitization on short-term plasticity. Thus, desensitized AMPA receptors diffuse out of the synapse and are replaced by non-desensitized AMPA receptors (Heine et al. 2008;Constals et al. 2015). Auxiliary subunits such as TARPs and CKAMPs reduce AMPA receptor diffusion by anchoring receptors to scaffolding proteins at the postsynaptic density (PSD) (Bats et al. 2007;Opazo et al. 2010;Sumioka et al. 2010;Klaassen et al. 2016). Auxiliary subunits should therefore in principle prolong the time that a desensitized AMPA receptor dwells in the synapse. However, synaptic anchoring by auxiliary subunits seems to depend on the conformation of the receptors and is weakened upon AMPA receptor desensitization (Constals et al. 2015). Thus, desensitized AMPA receptors show a higher mobility than non-desensitized AMPA receptors. Consequently, desensitized receptors diffuse out of the synapse and can be replaced by non-desensitized receptors. This can be explained by an unbinding of desensitized AMPA receptors from TARPs (Constals et al. 2015). The dwell time of AMPA receptors in synapses depends not only intracellular anchoring but also on their N-terminal domain (Díaz-Alonso et al. 2017;Watson et al. 2017) via which AMPA receptors may interact with extracellular or presynaptic proteins. It remains to be shown whether the N-terminal domain influences short-term plasticity by affecting diffusion of AMPA receptors. In addition, the N-terminal domain is highly mobile (Dawe et al. 2019) and it is possible that the agonist induced compression of the N-terminal domain alters synaptic anchoring and diffusion of AMPA receptors.
Klaassen and colleagues showed that auxiliary subunits can also influence short-term plasticity by affecting the decay kinetics of synaptic AMPA receptors (Klaassen et al. 2016). Short-term plasticity may be influenced by decay kinetics if firing frequency of presynaptic cells is high, i.e. when the activation of AMPA receptors occurs during the decay phase of a previous activation. CKAMP52 (aka shisa6) reduces the rate of AMPA receptor deactivation. Consistently, genetic deletion renders the decay of synaptic currents in CA1 neurons faster. This explains the decrease in short-term facilitation in CKAMP52 knockout mice when CA1 neurons are stimulated with high frequency of 50 Hz (Klaassen et al. 2016).
Short-term plasticity is usually tested in acute brain slices with artificial stimulation protocols. Neuronal firing patterns, but also release probability, glutamate diffusion and reuptake may be different in vivo. To understand whether AMPA receptor auxiliary subunits influence synaptic computation also in vivo, we recorded firing rates of lateral geniculate nucleus relay neurons in head-fixed non-anaesthetized mice in response to visual stimuli. The magnitude of On-and Off-responses was increased in CKAMP44 knockout mice compared to wild-type mice. These findings confirmed the data from acute brain slice experiments showing that CKAMP44 influences computation of synapses by affecting short-term depression (Fig. 3). Interestingly, CKAMP44 reduces relay neuron responses by this mechanism despite the fact that it increases the number of synaptic AMPA receptors (Chen et al. 2018). The data also imply that the influence of CKAMP44 on relay neuron firing is relevant in particular when presynaptic retinal ganglion cells fire at high frequency (Chen et al. 2018).

Long-term plasticity (LTP/LTD)
Long lasting alterations of the strength of synapses are believed to be the foundation of learning and memory. Since the first description of LTP by Bliss and Lomo in 1973, it has become clear that long lasting synaptic plasticity is not uniform, but exists in many different variations (reviewed in Huganir & Nicoll, 2013). A fundamental mechanism underlying LTP and LTD in many synapses is a change in the number of synaptic AMPA receptors. Several AMPA receptor auxiliary subunits including TARP γ-2 and γ-8, J Physiol 599.2 CKAMP59, CNIH-2/-3 and GSG1L influence synaptic long-term plasticity (Rouach et al. 2005;Tomita et al. 2005b;Herring et al. 2013;Khodosevich et al. 2014;Gu et al. 2016;Schmitz et al. 2017). This effect of auxiliary subunits on long-term plasticity is perhaps not too surprising, considering their role in the control of trafficking and subcellular localization of AMPA receptors. The mechanisms of how auxiliary subunits influence LTP have been extensively investigated for TARPs. For example, hippocampal LTP depends on phosphorylation of TARP γ-2 (Tomita et al. 2005b). Hippocampal and cerebellar LTD on the other hand require dephosphorylation of TARP γ-2 (Tomita et al. 2005b;Nomura et al. 2012). Similarly, phosphorylation of TARP γ-8 by CaMKII is needed for expression of LTP in hippocampal neurons (Park et al. 2016). Phosphorylation of TARPs initiates diffusion and synaptic trapping of AMPA receptor complexes via interaction with PDZ-domain containing proteins such as PSD95 . Besides TARP γ-2 and γ-8, the auxiliary subunits CNIH-2/-3 and GSG1L also affect LTP. However, while CNIH-2/-3, like the TARPs, is needed for normal LTP expression, GSG1L seems to suppress LTP (Herring et al. 2013;Gu et al. 2016).

Homeostatic plasticity
In order to integrate a broad spectrum of synaptic input and, at the same time, maintain a relatively stable output, neurons adjust the strength of their synapses by altering synaptic AMPA receptor density (O'Brien et al. 1998;Turrigiano et al. 1998). This form of long-term plasticity is termed homeostatic scaling and has to be separated from the Hebbian forms of synaptic plasticity that have been described above. However, similarly to Hebbian plasticity, changes in AMPA receptor density are the underlying mechanisms of up-and down-scaling of synapses. Additionally, other processes, such as alterations in the subunit composition of synaptic AMPA receptors and their phosphorylation pattern play a role during homeostatic scaling (Siddoway et al. 2013;Soares et al. 2013;Diering et al. 2014;Kim & Ziff, 2014). The influence of auxiliary subunits on AMPA receptor trafficking suggests that they may play a role in homeostatic scaling. Indeed, visual deprivation or TTX treatment increases the expression and phosphorylation of TARP γ-2 in the lateral geniculate nucleus or in cortical cultures, respectively (Louros et al. 2014). Importantly, synaptic up-scaling in response to visual deprivation depends on A, short-term depression of synaptic currents is less strong in retinogeniculate synapses of lateral geniculate relay neurons of CKAMP44 −/− mice than in wild-type mice. B, the weaker short-term depression in CKAMP44 −/− mice explains the augmented postsynaptic depolarization in response to train stimulation of retinogeniculate synapses. C, relay neurons fire action potentials when the same stimulus train as in B is delivered with stronger stimulation strength. Relay neurons show increased firing probability in response to this stimulus train than relay neurons of wild-type mice. This difference is also explained by the difference in short-term plasticity (adapted from Chen et al. 2018).
the phosphorylation of TARP γ-2 (Louros et al. 2014). Additionally, dephosphorylation of TARP γ-2 mediates downscaling of cortical synapses (Louros et al. 2018). It is not known whether synaptic properties, and in consequence synaptic computation, are altered during homeostatic plasticity due to changes in auxiliary subunit expression.

Conclusion
AMPA receptor auxiliary subunits provide neurons with a versatile tool to adjust their synaptic function according to their computational needs. The composition of AMPA receptor complexes influences EPSC kinetics, and synaptic short-term and long-term plasticity. Changes in the expression of auxiliary subunits in the context of homeostatic plasticity may therefore increase or decrease synaptic strength. In addition, homeostatic changes in AMPA receptor composition could affect how neurons compute excitatory inputs by altering EPSC kinetics, short-term plasticity and long-term plasticity rules. A detailed knowledge of the effects of AMPA receptor auxiliary subunits in physiological but also in pathological conditions is crucial for an understanding of their role in neurological or psychiatric diseases. Genetic linkage analyses suggest that TARP γ-2 and γ-3 may be implicated in familial epilepsy, Alzheimer's disease, schizophrenia and bipolar disorders (Wilson et al. 2006;Everett et al. 2007;Knight et al. 2008;Liu et al. 2008;Silberberg et al. 2008;Ament et al. 2015;Savas et al. 2017). This knowledge would also be relevant for the development of novel drugs that target specific AMPA receptor compositions, such as the recently published LY3130481, which efficiently reduces epileptic activity in rodents by blocking specifically TARP γ-8-containing AMPA receptors. Hippocampal and cortical neurons, but not, for example, cerebellar neurons, display high expression levels of TARP γ-8. This may explain why LY3130481 has considerably fewer motor side effects than perampanel, an antiepileptic drug that is an unspecific AMPA receptor antagonist (Kato et al. 2016).