Long-Term Potentiation and Long-Term Depression at Glutamatergic Synapses

The ability of synapses to change their strength in response to stimulation has been known since the 1970s, when it was demonstrated that repetitive stimulation of a sensory organ changes in the efficacy of a synaptic connection in the invertebrate’s nervous system (Castellucci, Pinsker, Kupfermann, & Kandel, 1970; Kupfermann, Castellucci, Pinsker, & Kandel, 1970). This form of long-term plasticity is known as long lasting synaptic facilitation (Brunelli, Castellucci, & Kandel, 1976; Castellucci & Kandel, 1976). Its signaling mechanisms have been worked out extensively (Armitage & Siegelbaum, 1998; Bailey & Chen, 1983, 1988a, 1988b; Cai, Pearce, Chen, & Glanzman, 2011; Esdin, Pearce, & Glanzman, 2010; Ezzeddine & Glanzman, 2003; Fulton, Condro, Pearce, & Glanzman, 2008) and its functional relevance is well understood (Carew & Kandel, 1973; Pinsker, Hening, Carew, & Kandel, 1973). Synaptic facilitation is involved in driving the relaxation of the gill, the respiratory organ of the sea slug, in response to stimulation of the syphon, a sensory organ and passage for oxygenated water to flow through the gill, when the animal learns that a stimulus that does not represent a threat (Kupfermann, Carew, & Kandel, 1974; Pinsker, Kupfermann, Castellucci, & Kandel, 1970; Walters, Carew, & Kandel, 1979). Studies in the sea slug directly related mechanisms for changes in synaptic efficacy to learning (Kandel et al., 1983; Walters, Byrne, et al., 1983a, 1983b; Walters, Carew, et al., 1981), and set the groundwork for investigating the presence and role of synaptic plasticity in the mammalian brain.

In the mammalian brain, evidence for long-term changes in synaptic strength becomes available in the early 1970s (Bliss & Lomo, 1973). In the first report about activity-dependent long-term plasticity it was shown that high frequency stimulation of the perforant path produced a long lasting increase in the field potential response and in the number of population spikes in the dentate gyrus of the hippocampus. This form of long lasting plasticity became known as long-term potentiation (LTP) (Bliss & Lomo, 1973). The activity-dependent changes in synaptic responses were reminiscent of the learning-induced changes in synaptic efficacy in the sea slug (Carew, Castellucci, & Kandel, 1971). Furthermore, high frequency activity, albeit in bursts within the theta oscillation range, had been associated with learning and memory (Berry & Thompson, 1978; Winson, 1978). These factors contributed to the initial excitement toward investigating long-term changes in evoked responses as synaptic mechanisms for storing memories in mammalian neural circuits (Bliss & Collingridge, 1993). Since the first reports, much effort has been devoted to determining how long-term plasticity is induced, what mechanisms underlie plasticity induction and expression, and how plasticity may contribute to neural network function.

Much of the work on the mechanistic aspects of long-term plasticity focused on hippocampal glutamatergic synapses (Bashir, Alford, Davies, Randall, & Collinridge, 1991; Castillo, Weisskopf, & Nicoll, 1994; Collingridge, Blake, Brown, Bashir, & Ryan, 1991; Manzoni, Weisskopf, & Nicoll, 1994; Nicoll, Castillo, & Weisskopf, 1994; Thiels, Xie, Yeckel, & Berger, 1996; Weisskopf, Castillo, Zalutsky, & Nicoll, 1994; Weisskopf & Nicoll, 1995; Xie, Barrionuevo, & Berger, 1996; Xie, Berger, & Barrionuevo, 1992). However, it became evident that these forms of long-term plasticity are not unique to glutamatergic synapses (Kano, 1995; Korn, Oda, & Faber, 1992) and not only inducible in the hippocampus (D’Angelo, Rossi, Armano, & Taglietti, 1999; Ennis, Lister, Aroniadou-Anderjaska, Hayar, & Borrelli, 1998; Kirkwood, Rioult, & Bear, 1996; Morris, Knevett, Lerner, & Bindman 1999; Winder, Egli, Schramm, & Matthews, 2002). There is now extensive evidence that long-term plasticity can be induced at glutamatergic (Kirkwood et al., 1996) and GABAergic synapses in neocortex (Holmgren & Zilberter, 2001; Maffei, Nataraj, Nelson, & Turrigiano, 2006), cerebellum (D’Angelo et al., 1999; Hansel & Linden, 2000), and subcortical circuits (Nugent, Penick, & Kauer, 2007; Winder et al., 2002). The activity requirements and cellular mechanisms may differ depending on specific connection and developmental stage (Feldman, Nicoll, Malenka, & Isaac, 1998; Lefort, Gray, & Turrigiano, 2013; Wang, Fontanini, & Maffei, 2012). The widespread capacity for long-term plasticity and the diversity of mechanisms indicate that the ability to change synaptic efficacy in an activity-dependent manner is a fundamental property of synapses throughout the brain. In the next sections, current knowledge regarding the activity requirement for plasticity induction and the diversity of cellular mechanisms involved will be reviewed. Recent evidence regarding the developmental regulation of plasticity, and how plastic changes may be integrated in local circuits will be reported. Current theories regarding the functional role of long-term synaptic plasticity will also be discussed.

Induction of Long-Term Plasticity: Rate Versus Timing

Soon after the publication of the first study on LTP, a new report showed that stimulation of the afferent fibers onto CA1 pyramidal neurons in the hippocampus induced LTP is stimuli were delivered at high frequency (100 stimuli at 100 Hz), while the same number of stimuli delivered at low frequency (1 Hz) induced a long-term depression of the evoked response (LTD) (Dunwiddie & Lynch, 1978). This study demonstrated that synapses in the mammalian hippocampus can changes their efficacy bidirectionally in response to stimulation and that the sign of the plasticity depends on the frequency of the stimulation. These results were recapitulated and expanded by a later study that thoroughly examined the frequency dependence of changes in synaptic efficacy and showed that, when stimulated below a threshold frequency, glutamatergic synapses in CA1 undergo LTD. If the frequency of repetitive stimulation used for induction was higher than the threshold frequency, LTP was induced instead (Rick & Milgram, 1996).

The study of LTD persisted in using low frequency stimulation. Repetitive stimulation at 1 Hz has been quite effective at inducing LTD in most of the pathways in which it has been tested (Doyere, Errington, Laroche, & Bliss, 1996; Hess & Donoghue, 1996; Kirkwood & Bear, 1994a, 1994b; Trommer, Liu, & Pasternak, 1996). Differently, the dependence of LTP induction of the frequency of stimulation was explored further. In parallel to studies using trains of high frequency stimuli (2–3 trains of 1s stimuli at 100 Hz; also known as tetanic stimulation), the analysis of the rate dependence of LTP proceeded to determine physiologically relevant patterns of stimulation that would effectively potentiate synapses. As activity in the theta frequency range was associated with learning (Gray & Ball, 1970) and memory (Landfield, McGaugh, & Tusa, 1972), and LTP was thought to be a synaptic correlate for memory (Levy & Steward, 1979), investigators devised induction paradigm that mimicked theta frequency activity that could be used either in vivo in anesthetized animals or in acute slice preparation to study the effect of this pattern of activity network activity and synaptic strength respectively. In the CA1 region of the hippocampus high frequency stimuli (100 Hz) organized in bursts delivered repetitively in the theta frequency range (4–7 Hz; theta burst stimulation, TBS) effectively induced LTP (Larson, Wong, & Lynch, 1986). Tetanic stimulation was an effective induction paradigm in the hippocampus (Bliss & Lomo, 1973), but not in neocortex (Kirkwood & Bear, 1994a, 1994b). Differently, TBS stimulation effectively induced LTP in many regions of the brain including the hippocampus (Larson & Lynch, 1988), visual cortex (Kirkwood & Bear, 1994a, 1994b), olfactory cortex (Jung, Larson, & Lynch, 1990), and at the mossy fiber input onto cerebellar granule cells (D’Angelo et al., 1999). Variation of rate-dependent induction paradigms were developed over the years to explore the parameter space for LTP induction. As neurons are spontaneously active and show oscillatory transitions in membrane potential (von Krosigk, Bal, & McCormick, 1993), experiments were designed to assess whether patterned stimulation on the depolarizing phase or on the hyperpolarizing phase of a neuronal oscillation would produce different forms of plasticity. Indeed, trains of stimuli delivered on the depolarizing phase on an oscillation induced LTP, differently patterned activity at the through of the oscillation induce LTD (Huerta & Lisman, 1995). This finding demonstrated that plasticity can be induced by activity in a physiological range. The need for depolarization for successful LTP induction indicated that both the afferent input and the postsynaptic neuron needed to be active for potentiation to occur (Jaffe & Johnston, 1990), while LTD was induced if the activity of the pre-and postsynaptic components were uncorrelated (Debanne, Gähwiler, & Thompson, 1994; Jaffe & Johnston, 1990). The results obtained in this seminal study fit well with Donald Hebb’s theory about learning, (Hebb, 1949). Thus, LTP became also known as Hebbian plasticity (Jaffe & Johnston, 1990). Recent studies challenged the view that high frequency stimulation is necessary for LTP induction, while low frequency stimulation is necessary for LTD. It was reported that low frequency stimulation in a number of circuits can induce a form of LTP that develops with a slower time course than high frequency LTP, but engages similar cellular mechanisms (Habib & Dringenberg, 2010a, 2010b). The conditions that would favor LTP over LTD induction at low frequencies have not been fully explored; however, there is evidence that the developmental stage at which plasticity is induced may affect the outcome of low frequency stimulation (Lante, Cavalier, Cohen-Solal, Guiramand, & Vignes, 2006).

The dependence of the sign of plasticity on the phase of the oscillation suggested that timing of pre- and postsynaptic activity played an important role in regulating synaptic efficacy (Debanne, Shulz, & Fregnac, 1995; Fregnac et al., 1994). This idea was tested in experiments in which for the first time a single connection between two identified neocortical pyramidal neurons was recorded, so that the activity of the presynaptic neuron and its postsynaptic target could be directly controlled by the experimenter (Debanne et al., 1996; Markram, Lübke, Frotscher, & Sakmann, 1997). If the action potential of the presynaptic neuron preceded the excitatory postsynaptic potential (EPSP) the efficacy of the monosynaptic connection between the two neurons was increased. When the EPSP/spike order was reversed the strength of the connection was decreased (Markram et al., 1997). These results highlighted the importance of timing for the induction of plasticity, providing the stepping stone for a form of long-term synaptic modification now known as spike timing dependent plasticity (STDP). STDP can be effectively induced at glutamatergic synapses in several brain regions and is often characterized as time-dependent LTD (t-LTD) if the postsynaptic EPSP is reduced after pairing, typically when the postsynaptic EPSP precedes the presynaptic action potential (Bi & Poo, 1998). Time-dependent LTP (t-LTP) refers to LTP of the postsynaptic responses and is typically induced when presynaptic spikes precede the postsynaptic EPSP or spike (Bi & Poo, 1998; Magee & Johnston, 1997; Markram et al., 1997; Sjostrom, Turrigiano, & Nelson, 2001). Further studies demonstrated that the STDP rule can be affected by the location of the activated presynaptic input along the dendritic harbor of the postsynaptic pyramidal neuron (Froemke, Poo, & Dan, 2005; Letzkus, Kampa, & Stuart, 2006; Sjostrom & Hausser, 2006).

STDP can also show variations to the pre-post timing rule if the frequency of pre-post pairing or the number of pre- and postsynaptic spikes is changed (Buchanan & Mellor, 2010). For instance, in the hippocampus, pairing of single pre-and postsynaptic spikes STDP at or above 10 Hz induces t-LTP independently of pre-and postsynaptic order (Buchanan & Mellor, 2010); while t-LTD only is induced if the pairing is below 10 Hz (Wittenberg & Wang, 2006). In the primary thalamo-recipient layer of somatosensory and visual cortex (layer 4) t-LTD is induced independently of the order of pre- and postsynaptic activity either using single spike pairing or pairing of bursts of pre- and postsynaptic spikes (Egger, Feldmeyer, & Sakmann, 1999; Wang et al., 2012).

Much of the work on the parameter space for the induction of plasticity was obtained in acute slices preparations, which allow for precise access to sites of stimulation and recordings, as well as high resolution on synaptic responses. Evidence t-LTD in vivo has been provided for the barrel cortex, in experiments in which it was induced when spontaneous action potentials in barrel cortex excitatory neurons was paired with whisker deflection mimicking a post- before pre- pairing (Jacob, Brasier, Erchova, Feldman, & Shulz, 2007). Regarding t-LTP, a new study showed that in this form of plasticity can be induced in anesthetized rodents (Fung, Law, & Leung, 2016).

A Wide Variety of Cellular Mechanisms for LTP and LTD

In parallel to the analysis of the activity requirements for LTP and LTD induction, much work has been done to identify the cellular/molecular mechanisms engaged in these forms of plasticity (Figure 1, LTPe and LTDe). Initially focusing on hippocampal glutamatergic synapses, it was shown that the induction of LTP requires the activation of a special type of glutamatergic receptor, the NMDA receptor (Harris, Ganong, & Cotman, 1984), which is sensitive to simultaneous glutamate binding and membrane depolarization (Mayer, Westbrook, & Guthrie, 1984; Nowak, Bregestovski, Ascher, Herbet, & Prochiantz, 1984). This dual sensitivity makes the NMDA receptor a coincidence detector, able to sense the activity of pre- and postsynaptic neurons. The requirement for coincident pre- and postsynaptic activity is also a good fit with Donald Hebb’s postulate for memory formation (Tsien, 2000). Hence, NMDA receptors signaling became the focus of studies aimed at determining the mechanisms for Hebbian plasticity (Shimizu, Tang, Rampon, & Tsien, 2000).

Open in new tab

Figure 1. Summary of mechanisms for LTP and LTD of GABAergic (Left) and Glutamatergic (Right) forms of synaptic plasticity. The variety of signaling mechanisms for presynaptically and postsynaptically expressed LTP and LTD for inhibitory (LTPi and LTDi) and excitatory (LTPe and LTDe) is represented schematically based on data from current literature.

NMDA receptors activation allows calcium inflow in the neuron that expresses them (Bading, Ginty, & Greenberg, 1993; Malenka, Kauer, Zucker, & Nicoll, 1988), and calcium acts as a second messenger that activates a number of molecules (Poncer, Esteban, & Malinow, 2002) involved in trafficking receptors at synapses (Takahashi, Svoboda, & Manilow, 2003), as well as signaling cascades that can activate gene transcription (Ghosh, Ginty, Bading, & Greenberg, 1994). The work of several groups has contributed to elucidating the complexity of the intracellular mechanisms involved in the induction and expression of LTP and LTD in the hippocampus. For an extensive description of detailed signaling cascades involved in hippocampal LTP and LTD, I refer to the comprehensive review by Herring and Nicoll (2016).

The involvement of the NMDA receptor in the induction of LTP and LTD is not unique to the hippocampus, as NMDA-dependent forms of long-term plasticity have been reported in many circuits (Artola & Singer, 1987; D’Angelo et al., 1999; Jung et al., 1990; Kauer, Malenka, & Nicoll, 1988; Kombian & Malenka, 1994; Lovinger, Tyler, & Merritt, 1993). Differently, the mechanisms of expression of LTP and LTD spurred a heated debate about whether LTP and/or LTD depend on presynaptic or postsynaptic mechanism (Granger & Nicoll, 2014; Lisman, 2009; MacDougall & Fine, 2014; Padamsey & Emptage, 2014). A long-standing model supports the hypothesis that long-term plasticity is expressed through postsynaptic mechanisms (Herring & Nicoll, 2016). According to this model, LTP and LTD are both postsynaptically expressed (Malinow & Malenka, 2002). Strong activation of postsynaptic NMDA receptors by tetanic or TBS stimulation increases intracellular calcium rapidly (Lisman, 2001). Such rapid change in intracellular calcium triggers activation of the calcium calmodulin kinase II (CAMKII) (Lisman, 1994) and other second messenger cascades that ultimately lead to the stabilization and increase in the number of postsynaptic AMPA receptors (Takahashi et al., 2003). Differently, the low frequency stimulation used for inducing LTD increases intracellular calcium concentration slowly and to a lesser extent than a high frequency paradigm (Lisman, 2001). This slow increase in calcium activates signaling pathways leading to the removal of AMPA receptors (Kameyama, Lee, Bear, & Huganir, 1998; Lee, Barbarosie, Kameyama, Huganir, & Bear, 2000; Lee, Kameyama, Huganir, & Bear, 1998). While there are solid experimental results in support of this model, there is also strong evidence to demonstrate that this is just one of the ways long-term plasticity can be induced (Habib & Dringenberg, 2010a, 2010b).

For example, not all forms of LTP and LTD require the activation of NMDA receptors. NMDA-independent forms of plasticity have been reported both in the hippocampus (Grover, 1998; Kessey & Mogul, 1997; Sokolov, Rossokhin, Kasyanov, & Voronin, 2003), in neocortex (Egger et al., 1999; Huemmeke, Eysel, & Mittmann, 2002; Wang et al., 2012), in the amygdala (Weisskopf, Baue, & LeDoux, 1999), and in the spinal trigeminal interpolaris (Kim, Weon, & Youn, 2016).

In addition, presynaptically expressed forms of LTP and LTD have been reported (Bekkers & Stevens, 1990; Errington, Galley, & Bliss, 2003; Kullmann, 1994; Skrede & Malthe-Sorenssen, 1981; Sokolov et al., 2002; Zakharenko, Zablow, & Siegelbaum, 2001). The initial findings regarding the involvement of presynaptic changes were debated as not truly involving changes in the presynaptic neuron, but depending on the activation of silent synapses, synapses that contain only NMDA receptors (Isaac, Nicoll, & Malenka, 1995; Liao, Hessler, & Malinow, 1995). The involvement of silent synapses, however, is prominent on juvenile animals, when silent synapses are abundant, less so in adult subjects (Durand, Kovalchuk, & Konnertn, 1996).

As the number of studies on this topic increased, more and more evidence accumulated in support of the possibility that some forms of LTP and LTD depend, or involve, presynaptic mechanisms of expression. In the cerebellum changes in the magnitude of the presynaptic current following LTP induction were reported (Maffei, Prestori, & D’Engelo, 2002, 2003). Furthermore, at many synapses LTP depends on the activation of presynaptically located NMDA receptors (Bouvier et al., 2016; Corlew, Wang, Ghermazien, Erisir, & Philpot, 2007; Costa, Mizusaki, Sjostrom, & van Rossum, 2017; Sjostrom et al., 2003; Wang et al., 2012). Finally, forms of long-term plasticity engaging a variety of presynaptic mechanisms have been reported in several circuits including the hippocampus (Humeau, Shaban, Bissière, & Lüthi, 2003; Wang et al., 2014). The debate about pre- or postsynaptic expression for LTP stemmed primarily from the conviction that the model initially proposed for the expression of this form of plasticity would generalize to all circuits in the brain. As evidence that glutamatergic synapses in many circuits in the brain are plastic, that long-term plasticity can be induced by different induction paradigms, and engage a variety of mechanisms, it is likely that the postsynaptic expression model represents one of many signaling pathways through which synapses can change their strength in response to patterned activity.

Under what conditions different mechanisms of expression may be engaged is not well understood. However, there has been some progress in this line of research. An elegant study demonstrated that in the CA1 region of the hippocampus LTP relying on presynaptic and/or postsynaptic mechanisms of expression can be engaged at the same synapse, depending on the state of the connection (Ward et al., 2006). In layer 5 of neocortex, recurrent glutamatergic synapses have the capacity for multiple forms of presynaptically and postsynaptically expressed LTP and LTD that can be co-induced by high frequency pre- and postsynaptic firing (Sjostrom, Turrigiano, & Nelson, 2007).

Another important regulator of the capacity and mechanisms for plasticity of cortical synapses is postnatal development. For example, in layer 4 of rodent primary visual cortex, recurrent glutamatergic synapses show two forms of LTD in the third postnatal week, both expressed presynaptically. One is induced by an STDP paradigm and depends on the activation of metabotropic glutamate receptors (mGluR); the other is induced by repetitive presynaptic bursting and depends on presynaptic NMDA receptors (Wang et al., 2012). The same induction paradigms, applied to the same synaptic connection later in development have significantly different outcomes. The STDP paradigm is not effective at changing synaptic strength after postnatal day 25 (P25), while the repetitive burst paradigm engages presynaptic NMDA receptors and postsynaptic voltage gated calcium channels to induce a presynaptically expressed form of LTP (Wang et al., 2012).

All of these studies indicate that LTP and LTD should not to be considered two forms of plasticity with unique features that are generalizable across brain circuits. Rather, both LTP and LTD constitute families of activity-dependent long-term plastic changes, including a wide range of mechanisms that can be engaged depending on the pattern of activity, state of the neuron, and developmental stage. Each of these forms of plasticity may serve a distinct role in the neural circuits in which it is induced.

Synaptic Plasticity in Local Circuits

The conditions for successful plasticity induction have been extensively addressed in acute slice preparations, in which it is possible to explore a wide set of conditions and patterns of activity. When considering synaptic plasticity in the context of a circuit, one important additional level of regulation to be considered depends on the presence of neuromodulators (Calabresi et al., 2000; Cho, Jang, Jo, Singer, & Rhie, 2012; DeBock et al., 2003; Hopkins & Johnston, 1988; Otmakhova & Lisman, 1996, 1998, 1999; Otmakhova & Lisman, 2000; Otmakhova, Lewey, Asrican, & Lisman, 2005; Prestori et al., 2013; Rinaldo & Hansel, 2013). Signaling from neuromodulators can affect the state of a neuron so that: 1- a pattern of activity that was ineffective at inducing plasticity can successfully lead to changes in synaptic strength (He et al., 2015; Huang, Huganir, & Kirkwood, 2013); 2- a pattern of activity that potentiated a synapse induces depression (or vice versa) (Matsuda, Marzo, & Otani, 2006; Seol et al., 2007). Release of neuromodulators is associated with behavioral states (Berry, Cervantes-Sandoval, Chakrabort, & Davis, 2015; Drachman, 1977; Major, Vijayraghavan, & Everling, 2015) and/or can act as reward signals (Chubykin, Roach, Bear, & Hussain Shuler, 2013; Schultz, Apicella, & Ljungberg, 1993). The involvement of these signaling pathways supports the interpretation that synaptic plasticity can be induced in an experience-dependent manner, with experience defined broadly as the effect of behavior or the association of sensory stimuli to reward. In addition, these results indicate that the capacity for plasticity of glutamatergic synapses is not uniquely dependent on the incoming pattern of activity of a specific input, but may also depend on other network components.

Another important factor to consider when looking at plasticity in the context of circuit interactions is that patterns of activity that drive the induction of plasticity at one input (homosynaptic plasticity) can also affect neighboring inputs (heretosynaptic plasticity). This effect was evident in results obtained in very early studies of plasticity in the mammalian hippocampus in extracellular multiple electrodes were positioned to independently activate distinct inputs onto pyramidal neurons (Dunwiddie & Lynch, 1978). This important study showed that LTP was induced the input that received the induction paradigm, while the other inputs either remained unchanged or showed depression (homosynaptic LTP and heterosynaptic LTD). Differently, if in the same experimental setting the LTD-induction paradigm was delivered to one of the inputs, the amplitude of the evoked responses decreased both in the induced pathways and in the pathways that did not receive the induction paradigm (homosynaptic and heterosynaptic LTD) (Dunwiddie & Lynch, 1978). Later studies showed that LTP induction can be associative: a weakly activated input neighboring an input that is activated with a high frequency paradigm can also be potentiated (Barrionuevo & Brown, 1983). The mechanisms thought to be involved in heterosynaptic forms of plasticity range from spillover of glutamate (Tsvetkov, Shin, & Bolshakov, 2004) that activate receptors at neighboring synapses (Humeau et al., 2003), to the engagement of distinct cellular mechanisms (Chen, Tan, Zeng, & Duan, 2013; Lange, Doengi, Jörg, & Jüngling, 2012; Nugent et al., 2007). All of these experiments are important because they address questions regarding how the induction of plasticity at one input may affect other elements of the circuit (Artola & Singer, 1993).

A third important factor that can affect the capacity for plasticity of glutamatergic synapses is the history of a synapse (Abraham & Bear, 1996; Otmakhova, Otmakhov, Mortenson, & Lisman, 2000): whether signaling pathways were activated prior to the induction of LTP (Tenorio et al., 2010) or LTD (Mellentin & Abraham, 2001). Experiments in which the induction paradigms were applied multiple times during the same recordings showed that multiple steps of LTP (or LTD) can be induced at the same input (Rioult-Pedotti, Friedman, & Donoghue, 2000). However, following a few inductions, the change in synaptic strength reaches a plateau and further delivery of the induction paradigm will produce either no change or a change with the opposite sign. These findings suggested that depending on the previous history of a synapse, the threshold for inducing plasticity may have shifted (Abraham, Mason-Parker, Bear, & Tate, 2001). Thus, if a behavioral paradigm or manipulation had affected the strength of a specific input, it would be possible to infer whether and what form of plasticity was induced by attempting additional inductions (Kirkwood et al., 1996; Rioult-Pedotti et al., 2000). This type of experiment, known as an occlusion test, has been extensively used to begin to investigate the functional aspects of plasticity and provides the basis for a theoretical framework that was designed to determine how LTP and LTD may coexist in a neural network. The theory is known as the sliding threshold for plasticity (Abraham et al., 2001; Clothiaux, Bear, & Cooper, 1991), and is extrapolated from a theory known as Bienenstock Cooper Munro (BCM) after the theorists who first proposed it (Bienenstock, Cooper, & Munro, 1982). In the motor cortex it was shown that the capacity for LTP and LTD of extracellularly evoked local field potentials was affected by motor learning. Specifically, less LTP could be induced in acute slices prepared from the hemisphere contralateral to the limb that had been trained to retrieve a pellet in a reaching task (Rioult-Pedotti et al., 2000). Local field potentials however measure responses of groups of neurons, not of single inputs; therefore from this experiment it was difficult to determine whether the effect on plasticity was dependent on uniform changes at a large number of synapses or to connection-specific changes. In principle, if a single connection between two neurons could be recorded, a reasonable extrapolation from the sliding threshold theory is that the initial amplitude of an evoked response would correlate with the magnitude of plasticity that can be induced. In the primary visual cortex, paired recordings have been used extensively to determine the effects of monocular deprivation on specific connections. Here, the results do not fully support the expectations of the sliding threshold theory. On the one hand, in the deprived hemisphere application of an LTP-inducing paradigm produced LTD, suggesting that the deprivation may have maximally potentiated this connection and shifted the induction for LTP (Wang et al., 2012). This effect was specific to a burst induced LTP, as STDP remained unaffected by the deprivation (Maffei et al., 2006; Wang et al., 2012). On the other hand, comparison of the baseline amplitude of monosynaptic connections in the hemisphere contralateral to the open eye (control) to that of the connections recorded in the hemisphere contralateral to the closed eye (deprived) indicated no changes in baseline transmission, pointing to lack of preexisting changes in synaptic strength (Maffei et al., 2006, Wang et al., 2012). In addition, no correlation between the initial amplitude of the monosynaptic response and the magnitude of plasticity was observed in either hemisphere (Wang et al., 2012). The only study to date reporting a correlation between magnitude of plasticity and baseline amplitude of synaptic responses comes from a study of LTP and LTD in the guinea pig in which a unique bust-spike timing paradigm of induction was used (Saez & Friedlander, 2009), suggesting that the experimental conditions (induction paradigm and species) may influence the results. Overall, these data indicate that the capacity for plasticity of glutamatergic synapses can be affected by behavioral manipulations and is sensitive to a number of factors within local circuits. These data also suggest that to fully understand the role of synaptic plasticity in the brain it is necessary to go beyond glutamatergic transmission and neuromodulation. An important and poorly addressed issue regards the integration of plastic changes and the effects of such changes on neurons’ output. These issues are particularly important in complex circuits in which synaptic plasticity can be induced at different synapses, and are thought to influence network function.

Long-Term Plasticity at GABAergic Inhibitory Synapses

Glutamatergic synapses have been the focus of studies of LTP and LTD; however, long-term forms of plasticity are not unique to excitatory synapses. Evidence that inhibitory GABAergic synapses also undergo activity-dependent long-term modifications begun to emerge from the early 1990s from studies in goldfish (Korn et al., 1992) and the mammalian cerebellum (Kano, Rexhausen, Dreesen, & Konnerth, 1992). Stimulation of glutamatergic climbing fibers onto Purkinje neurons produced a calcium-dependent long lasting potentiation of inhibitory postsynaptic currents (IPSCs) onto Purkinje neurons (Hashimoto, Ishii, & Ohmori, 1996; Kano et al., 1992). This form of heterosynaptic plasticity was later shown to rely on the coincident activation of NMDA and GABA_A receptors onto the postsynaptic neuron and require the activation of calcium calmodulin kinase II (Kano, Fukunago, & Konnerth, 1996). Soon after these pioneering studies it became evident that inhibitory synapses in other circuits are plastic (Adermark, Talani, & Lovinger, 2009; Chevaleyre & Castillo, 2003; Komatsu & Iwakiri, 1993; Maffei et al., 2006; Mapelli, Gandolfi, Vilella, Zoli, & Bijiana, 2016; Melis, Camarini, Ungless, & Bonci, 2002; Morishita & Sastry, 1993; Nugent et al., 2007; Woodin, Ganguly, & Poo, 2003).

The activity requirements and mechanisms for GABAergic synaptic plasticity are quite diverse. Initial studies focused on heterosynaptic forms of GABAergic plasticity induced by high frequency or TBS stimulation of axon terminals. Using extracellular stimulating electrodes, in most preparation both glutamatergic and GABAergic axons were activated by the stimulus, even if inhibitory currents or potentials were isolated pharmacologically (Chevaleyre & Castillo, 2003; Komatsu & Iwakiri, 1993; Mapelli et al., 2016). Using this experimental approach, in the primary visual cortex, high frequency stimulation of layer 4 inputs onto layer 5 pyramidal neurons induced LTP of GABAergic inhibitory postsynaptic potentials (IPSPs) (Komatsu & Iwakiri, 1993). This form of GABAergic LTP shows some interesting similarities to LTP of excitatory synapses in the CA1 region of the hippocampus: synapse specificity, associativity, dependence on postsynaptic NMDA receptors activation, and calcium signaling (Komatsu & Iwakiri, 1993). In the hippocampus both heterosynaptic LTP and LTD of GABAergic synapses have been reported (Chevaleyre & Castillo, 2003). These forms of inhibitory plasticity require either the additional activation of NMDA receptors or of metabotropic glutamate receptors (mGluRs) on the postsynaptic neuron. A number of signaling mechanisms have been implicated in heterosynaptic LTP and LTD of GABAergic inputs (for a detailed review, see Nugent & Kauer, 2008; Younts & Castillo, 2014), including signaling by retrograde diffusible molecules such as endocannabinoids (Chevaleyre & Castillo, 2003), nitric oxide (Mapelli et al., 2016), or brain derived neurotrophic factor (Xu, Kotak, & Sanes, 2010) and co-activation of receptors for neuromodulators (Komatsu, 1996). All of these forms of inhibitory plasticity are induced by patterns of activity similar to those used for long-term plasticity at glutamatergic synapses and are developmentally regulated (Komatsu & Iwakiri, 1993; Yoshimura, Ohmura, & Komatsu, 2003).

Homosynaptic inhibitory plasticity can also be induced at GABAergic synapses, although the mechanisms for this form of plasticity differ significantly from the ones described above. In the hippocampus, activity-dependent shift in the chloride transporter change the driving force for GABA-evoked synaptic currents (Woodin et al., 2003). In the primary visual cortex, paired activation of the presynaptic GABAergic neuron and its postsynaptic pyramidal neuron can induce plasticity at inhibitory synapses (D’Amour & Froemke, 2015; Holmgren & Zilberter, 2001; Maffei et al., 2006, Wang & Maffei, 2014). Depending on the pattern of activity of the pairing, calcium-dependent mechanisms (Holmgren & Zilberter, 2001) or calcium independent (Wang & Maffei, 2014) signaling pathways can be engaged.

Recent work focusing on the mechanistic aspects of calcium-dependent forms of GABAergic inhibitory plasticity was fundamental to identify some of the signaling cascades involved and highlighted possible signaling overlaps between excitatory and inhibitory forms of plasticity (Figure 1, LTPi and LTDi). LTP of GABAergic synaptic responses is associated with an increase in the expression of gephyrin (Petrini & Barberis, 2014; Petrini, Ravasenga, et al., 2014), the scaffold protein for GABA_A receptors, whose availability at synapses is in turn regulated by its state of phosphorylation (Tyagarajan et al., 2011). Mechanisms regulating activity-dependent changes in GABAergic transmission include calcium signaling, phosphorylation by CAMKII, GABA receptors associated protein (GABARAP), and glutamate receptor interacting protein (GRIP) (Marsden, Beattie, Friedenthal, & Carroll, 2007), as well as signaling cascades involved in the trafficking of GABA_A (Stellwagen, Beattie, Seo, & Malenko, 2005). A calcium independent but GABA_B receptor-dependent form of inhibitory LTP has also been reported (Wang & Maffei, 2014). GABAergic plasticity is also bidirectional, and some of the mechanisms for LTD of inhibitory transmission include signaling via cannabinoid receptors (Chevaleyre & Castillo, 2003) and calcium-calcineurin signaling (Bannai et al., 2009). Current work is ongoing to identify additional mechanisms and signaling cascades involved in the induction and expression of GABAergic synaptic plasticity. The emerging picture from experimental studies indicates that the diversity and complexity of the processes involved in inhibitory synaptic plasticity is just beginning to being unraveled.

In the mammalian brain the functional role of GABAergic inhibitory synaptic plasticity is not well understood. Recent results demonstrated that changes in sensory experience affect the efficacy of inhibitory synapses (Gainey, Wolfe, Pourzia, & Feldman, 2016; Kannan, Gross, Arnold, & Higley, 2016; Maffei et al., 2006; Takesian, Kotak, Sharma, & Sanes, 2013), suggesting a possible role for this form of plasticity in experience-dependent circuit refinement. In addition, the induction of inhibitory plasticity was shown to affect the capacity for plasticity excitatory synapses (D’Amour & Froemke, 2015; Ormond & Woodin, 2009, 2011; Wang & Maffei, 2014), indicating that this form of plasticity can influence excitatory synapses within local circuits. These studies suggest that GABAergic inhibitory synaptic plasticity contribute to learning processes.

Evidence in favor of this possibility comes from studies in Drosophila demonstrating that GABAergic plasticity is determinant for gustatory (Paranjpe, Rodrigues, VijayRaghavan, & Ramaswami, 2012) and olfactory (Das et al., 2011) habituation. Mutants of learning and memory also show alterations in inhibitory synaptic plasticity (Ganguly & Lee, 2013).

Conclusions

The ability to induce LTP and LTD in response to patterned activity allows excitatory and inhibitory synapses in the brain to modify their efficacy and keep a trace of these modifications as a form of long-term storage within local circuits. Our understanding of how these forms of plasticity can be induced and the mechanisms they engage has increased tremendously over the past decades. However, much of the focus has been on one form of plasticity or another, studied in isolation from the rest of the circuit. Excitatory and inhibitory synaptic drive is thought to be in balance in healthy functioning neural circuits, thus mechanisms must be in place to co-regulate or coordinate experience-dependent and learning-dependent changes in both components. To fully determine the functional relevance of LTP and LTD of both excitatory and inhibitory synapse the effort needs to shift toward looking at them as integrated properties of a circuit that are coordinated and co-regulated. Expanding the study of plasticity in this direction will likely advance our understanding of how experience and learning can be encoded in neural circuits and networks.