Information

Plasticity between excitatory and inhibitory neurons?

Plasticity between excitatory and inhibitory neurons?



We are searching data for your request:

Forums and discussions:
Manuals and reference books:
Data from registers:
Wait the end of the search in all databases.
Upon completion, a link will appear to access the found materials.

All that I've learned about synaptic plasticity only concern the synapses between excitatory neurons. For example, all pyramidal neurons (excitatory) in the cortex have plastic synapses between them, but as far as I know there are no plastic synapses between pyramidal neurons and their inhibitory interneurons.

So I am curious, are there any plastic synapses between inhibitory and excitatory neurons?


Various forms of plasticity has been reported for inhibitory-excitatory synapses as well. See below figure from [Woodlin et al. 2003].

  • Holmgren CD, Zilberter Y (2001) Coincident spiking activity induces long-term changes in inhibition of neocortical pyramidal cells. J Neurosci 21: 8270-7

  • Woodin MA, Ganguly K, Poo MM (2003) Coincident pre- and postsynaptic activity modifies GABAergic synapses by postsynaptic changes in Cl- transporter activity. Neuron 39: 807-20


Functional consequences of inhibitory plasticity: homeostasis, the excitation-inhibition balance and beyond

Computational network models increasingly incorporate inhibitory synaptic plasticity.

Inhibitory plasticity can stabilize network dynamics in spite of excitatory plasticity.

Inhibitory plasticity can maintain a balance of excitation and inhibition.

Inhibitory plasticity shapes the selectivity and temporal response profile of sensory neurons.

Computational neuroscience has a long-standing tradition of investigating the consequences of excitatory synaptic plasticity. In contrast, the functions of inhibitory plasticity are still largely nebulous, particularly given the bewildering diversity of interneurons in the brain. Here, we review recent computational advances that provide first suggestions for the functional roles of inhibitory plasticity, such as a maintenance of the excitation-inhibition balance, a stabilization of recurrent network dynamics and a decorrelation of sensory responses. The field is still in its infancy, but given the existing body of theory for excitatory plasticity, it is likely to mature quickly and deliver important insights into the self-organization of inhibitory circuits in the brain.


Specialized inhibitory cluster gates plasticity in fear learning

Has your heart ever started to race at the thought of an upcoming deadline for work? Or has the sight of an unknown object in a dark room made you jump? Well, you can probably thank your amygdala for that.

The small almond-shaped brain structure is central to how we perceive and process fear. As we start to learn to associate fear with cues in our environment, neuronal connections within the amygdala are dynamically altered in a process called synaptic plasticity. Although this physiological mechanism is important for facilitating fear learning, it has mostly been studied in the context of excitatory neurons within the amygdala. Far less is known about the role inhibitory cells play.

In a recent publication in Cell Reports, MPFI scientists from the Bolton Lab delve deeper into a specialized portion of inhibitory circuitry in the amygdala, known as the apical intercalated cell cluster (apITC). Characterizing this small but distinctive cluster of cells, the Bolton Lab has discovered rich connectivity and a rather unique ability to modulate plasticity in the amygdala.

"What really grabbed our attention was the fact that relatively little was known about apITC function or connective circuitry," explains Douglas Asede, Ph.D., first author and former postdoc in the Bolton Lab. "When working with a relatively unknown brain area, it's a game of inputs and outputs. First, you have to identify what connects with the neuron cluster and what it connects to, then evaluate what functional role that circuitry plays."

The Bolton Lab began its investigation by characterizing and functionally testing the incoming connections to the apITC. First, the team utilized a highly specialized technique called monosynaptic tracing to selectively identify the upstream presynaptic partners. Once identified, the researchers used a combination of presynaptic optogenetic stimulation (light activation) and postsynaptic electrophysiology to verify that the connections were functional.

"We were able to unravel a number of diverse inputs for this unique cell cluster, ranging from areas important for memory such as the entorhinal cortex to sensory processing regions such as the thalamus," explains Dr. Asede. "Among this diversity, two notable inputs from the thalamus stood out because of their relative strength compared to other connections we tested as well as their origin in thalamic regions known for their involvement in fear learning."

The strong connections to the apITC originated from two areas of the thalamus, the medial geniculate nucleus (MGm) and the posterior intralaminar nucleus (PIN). Previous work has shown the MGm and PIN to be important processing centers for auditory and somatosensory information, respectively. In the context of fear learning, inputs from the thalamus send fear-related sensory information to the amygdala, which then integrates and associates fear with particular cues from the environment.

To examine whether this sensory information flow through the apITC was important for fear learning, MPFI scientists studied the changes in these synaptic connections in mice directly after behavioral training. A group of mice underwent classical fear conditioning and behavior-driven changes and were then evaluated using pre and post-synaptic markers for plasticity. Interestingly, the team found significant signs of synaptic strengthening in the sensory inputs to the apITC after fear learning when compared to control animals.

"Typically, when synapses are important to a particular behavior, their connections are strengthened during learning, so our results really highlighted the importance of these sensory connections in fear learning," notes Dr. Asede.

The LA is a region of the amygdala that is strongly associated with fear learning, fear-related sensory integration, and the formation of fear-based memories. The Bolton Lab used simultaneous electrical stimulation of thalamic sensory inputs and optogenetic stimulation of apITC cell inputs to the LA to reveal that activation of apITC acts as a gate to reduce incoming sensory-driven responses in the LA.

"Armed with the understanding that apITC is important for sensory gating and fear learning, we next looked at what type of downstream connections the apITC makes to give us a clue about possible functions the cluster has in the amygdala fear circuitry."

Classically it's been thought that inhibitory cells within the brain make very short-range, downstream connections, acting to dynamically modify circuits within their own local environments. Using axonal reconstruction, the Bolton Lab identified that while most apITC connections are local axon collaterals to neighboring apITCs or project to a close region within the amygdala called the lateral amygdala (LA). Surprisingly, they also identified a subset of relatively long-range connections to more distant brain structures, challenging classical thinking on inhibitory circuits.


Neurobiologists able to induce brain plasticity using neuron transplantation

Your brain changes continuously over the course of your life. It is a phenomenon known as brain plasticity, or “the ability of the brain to modify its connections or re-wire itself”[1]. At a neuronal level, it is characterized by changes in the number of neurons and in the connections between these neurons.

Brain plasticity is important, because it is what allows the brain to develop , to learn and to repair itself after it suffers damages from things as severe as strokes or from blows to the head. It fluctuates over time. One of the periods during which it is heightened is a period of ocular dominance plasticity that occurs once both in humans and mice.

Usually the neuron population in the left visual cortex is more responsive to information from the right eye and the opposite is also true. During this critical period of ocular dominance plasticity, however, the deprivation of vision in one eye, such as when the eye is closed for an extended length of time, leads to an important change in responsiveness of the part of the brain usually more responsive to the closed eye towards the open eye. In mice, this period reaches its peak approximately four weeks after birth.

In 2010, researchers from a laboratory at the University of California, San Diego found a way to induce another period of ocular dominance plasticity in mice after this unique window of times, through the transplantation of an embryo’s neurons .

The primary author from the paper on this research, Dr. Derek Southwell, who at the time of the study was completing his PHD program, noted that, prior to the beginning of the study, it had been known that a certain type of inhibitory neuron could survive and disperse in the brain of the host when transplanted. These transplanted neurons had furthermore been shown to be able form connections with the host’s neurons. Dr. Southwell described the purpose of this study, as trying to understand “the effects of these transplanted cells on plasticity in the recipient brain”.

The researchers found, by an assessment of ocular dominance plasticity before and after four days of visual deprivation in one eye, that the transplantation of an embryo’s inhibitory neuron in mice was able to create a new period of ocular dominance plasticity in those same mice. This occurred once the transplanted cells reached the age at which the period normally takes place, when they were about 33 to 35 days old, even when the host was older. The research team also found through intrinsic signal imaging, a technique used to record neural activity, that the transplanted inhibitory neuron had migrated in the brain’s cortex after the transplantation and that they were able to develop the properties of mature inhibitory neurons.

Those transplanted inhibitory neurons were also shown to be three times more likely than the host’s inhibitory neurons to make connections with the host’s excitatory neurons, even if those connections were only about one-third as strong as host-to-host connections.

Representation of the connection between a transplanted inhibitory neuron an host excitatory neurons (left) compared with connections between host excitatory and inhibitory neurons (right).

The researchers noted that this ability of the transplanted neurons, by widely modifying the inhibitory signaling in the host’s brain, might be the reason why they were able to induce a new period of ocular dominance plasticity.

The conclusions from this paper were that inhibitory neuron transplantation could be used as a way to induce brain plasticity, maybe even as a new experimental preparation for research on the subject. Dr. Southwell furthermore highlighted that these findings have given rise to a long lasting research direction for all those involved in the research. More work is currently undergoing about the therapeutic usage of inducing plasticity in order to facilitate brain repair to restore the normal function of brains, such as for example in the treatment of PTSD.

Work this article is based on:

Southwell DG, Froemke RC, Alvarez-Buylla A, Stryker MP, Gandhi SP. Cortical plasticity induced by inhibitory neuron transplantation. Science (2010)


Differences in multiple forms of short-term plasticity between excitatory and inhibitory hippocampal neurons in culture

Synaptic transmission is highly dynamic, especially during periods of repetitive activity. This short-term synaptic plasticity, elicited by either pairs or short trains of action potentials at moderate frequencies (1-10 Hz), may give rise to either depression or facilitation of synaptic transmission. We analyzed these processes in isolated, synaptically coupled pairs of inhibitory or excitatory neurons grown in low-density cultures of hippocampal neurons. Most inhibitory and excitatory synapses in these cultures displayed paired pulse depression, although the responses of excitatory synapses were more variable and occasionally facilitation was seen. With tetanic stimuli, inhibitory synapses showed depression, but excitatory synapses showed a much richer repertoire of behaviors, including depression and facilitation. While many inhibitory synapses showed posttetanic depression following short trains of action potentials, excitatory synapses instead showed posttetanic facilitation. This facilitation is accompanied by an increase in paired pulse ratio, suggesting that it is the result of presynaptic mechanisms. Finally, excitatory synapses often displayed paired pulse and tetanic facilitation of asynchronous release, a process not seen in inhibitory synapses in these cultures. These similarities and differences in short-term plasticity exhibited by inhibitory and excitatory cells are likely to be critical for information processing and the control of neuronal excitability, under both normal and pathological conditions, such as epilepsy.


Methods

Network model

The network consisted of 400 PCs grouped into four subpopulations of 100 neurons each. Each subpopulation coded for a given orientation. We simulated 120 PV interneurons, 120 SST interneurons (30 in each subpopulation), and 50 VIP interneurons.

Neuron model

Neurons were modelled as conductance-based spiking leaky integrate-and-fire neurons. Their membrane potential evolves according to:

where (_<< m>>) is the membrane capacitance, (_) is the membrane potential of neuron (i) , (_<< m>>) is the leak reversal potential. (_<< m>>) and (_<< m>>) are the excitatory and inhibitory reversal potentials. (_<< m>>) , (_^<< m>>) and (_^<< m>>) are the leak, excitatory and inhibitory conductances. (_^<< m>>) and (_^<< m>>) are increased by (_) upon a spike event in a presynaptic excitatory or inhibitory neuron (j) , and decay exponentially with time constants (< au >_<< m>>) and (< au >_<< m>>) , respectively:

(xi) is zero-mean Gaussian white noise. Parameters defining the Ornstein-Uhlenbeck process are (sigma) = 2 mV and correlation time ( au) = 5 ms.

When the membrane potential reaches a threshold (_< heta >) , a spike event is recorded and the membrane potential is reset to its resting value (_<< m>>) (Table 1).

PVs were additionally connected via gap junctions which contribute a current (_<< m>,i>) to the RHS of Eq. (1) 54 . The gap junction current from neuron (j) to neuron (i) is the sum of a spikelet current and a subthreshold current. The subthreshold current is proportional to the difference in membrane potential between neurons (i) and (j) , with proportionality constant (_<< m>>) . The spikelet current is increased by (_<< m>>) upon a spike event in a presynaptic neuron and decays to 0 with timescale (< au >_<< m>>) . Mathematically,

Connectivity

Neurons from different cell classes (E: PC, P: PV, S: SST, V:VIP) are connected with chemical synapses as follows:

The connection probability (

_) from population (J) to (I) where (I,Jin <, ext,>) was chosen based on data from Pfeffer et al. 21 for connections from inhibitory populations, and on data from Hofer et al. 20 for connections from the excitatory population. Pfeffer et al. 21 provide connection probabilities and strengths for connections between the pairs and individual neuronal contributions (INCs) estimated from optogenetic stimulation of entire cell populations for the other connections. In cases where the connection probability was not measured directly, we chose the probability based on the INCs as follows: For (

_<< m

>< m>>) , (

_<< m>< m>>) , (

_<< m>< m>>) and (

_<< m>< m

>>) , the INC was very low, namely 0 (.06) (or 0 (.07) for (

_<< m>< m

>>) ). The connection from VIPs to PCs (EV) had the same INC of 0 (.06) , therefore we chose the same connection probability as for (

_<< m>< m>>) , which was 12.5%. For the remaining connections, we set the connection probability to 100%.

The synaptic weights (_) from a neuron (j) in population (J) to a neuron (i) in population (I) determine how much the synaptic conductances (^) and (^) increase upon a spike in neuron (j) . We initialised the synaptic weights based on the connectivity data from Pfeffer et al. 21 for connections from inhibitory populations, and based on data from refs. 20,22,23,24 for connections from excitatory to inhibitory populations. The number of neurons in each population was taken into account, when determining the connection strength. Connections between excitatory neurons were initially small and sampled from a Gaussian distribution truncated at 0.

For the gap junctions between PVs, (< au >_<< m>>) was 9 ms, except in Supplementary Fig. 4, where it was 3 ms. (_<< m>>) was 13 pA, unless otherwise stated. Subthreshold currents mediated by gap junctions were modelled only in Supplementary Fig. 9, where (_<< m>>) was 0.4 nS.

Inputs

PCs and SSTs received one of four inputs (corresponding to layer 4 (L4) inputs coding for four different orientations). Each L4 input produced a Poisson-distributed spike train with a rate of 4 kHz during its preferred stimulus, and 0 Hz otherwise. One of four stimuli was shown for 50 ms followed by a stimulus gap of 20 ms. During the stimulus gap, all L4 inputs produced spikes at the same rate of 1.6 kHz. The conductance of synapses from L4 to PCs was 0.28 nS. The conductances of synapses from L4 to SSTs was 0.15 nS during the stimulus and 0.165 nS during the stimulus gap. In addition, PCs and PVs received a baseline input from a Poisson process with a rate of 4 kHz. The weights to PCs were 0.13 nS, and to PVs 0.01 nS. VIPs received a top-down input during the vertical bar stimulus from a group of 100 neurons with connection strength 0.2 nS, which receive input from layer 4 with a connection strength of 0.3 nS.

Plasticity

For both excitatory and inhibitory plasticity we chose the simple classical STDP model 55,56,57

In the online implementation of this rule, the synaptic weight (_) from neuron (j) to neuron (i) is updated when either the pre- or the postsynaptic neuron spikes according to:


Background

Nervous systems face a constant challenge: how to maintain flexibility and stability at the same time. Neural circuits must stay flexible to allow for changes in connectivity and synaptic strength during development and learning. As changes in connectivity push neural circuits away from equilibrium, they need to maintain activity within a working range and avoid extremes of quiescence and saturation. Functional stability is maintained by homeostatic plasticity, which is defined broadly as a set of neuronal changes that restore activity to a setpoint following perturbation [1,2,3]. Recent studies have identified diverse homeostatic plasticity mechanisms triggered by a variety of perturbations. These mechanisms regulate dendritic and axonal connectivity of a neuron, as well as its intrinsic excitability (Fig. 1). In addition to maintaining the activity of individual neurons, homeostatic plasticity can act at a network level to coordinate changes in connectivity and excitability across multiple neurons to stabilize circuit function [4] (Fig. 2). Several recent reviews have covered the function of homeostatic plasticity in the mature nervous system [5,6,7,8]. Here, we focus on homeostatic plasticity in developing circuits.

Diverse homeostatic plasticity mechanisms stabilize the activity of developing neurons. When the activity of individual neurons decreases below (1 and 2) or increases above (3 and 4) a setpoint, homeostatic regulation of synaptic strength (1 and 3) and/or intrinsic excitability (2 and 4) acts to restore normal activity. By increasing (1) or decreasing (3) synaptic input (e.g., changes in mEPSC amplitude or frequency), a neuron’s output firing rate can be shifted up or down to the target activity (grey area). By increasing (2) or decreasing (4) intrinsic excitability (e.g., changes in the length and location of AIS), a neuron’s input/output relationship can be modified

Network-level homeostatic plasticity stabilizes activity of developing circuits. Network activity homeostasis is achieved by balancing excitation (red) and inhibition (blue). Synaptic strength and connectivity can be regulated in a cell-type-specific manner to maintain network homeostasis. Upward/downward red arrows: increased/decreased excitatory drive upward/downward blue arrows: increased/decreased inhibitory drive

Homeostatic regulation of intrinsic excitability

Neuronal intrinsic excitability is determined by the density, distribution, and function of ion channels, and controls how synaptic inputs are converted into action potential outputs [9]. Several studies have found a reciprocal relationship between intrinsic excitability and synaptic inputs across development, which stabilizes activity [10,11,12]. As synaptic inputs increase in developing Xenopus retinotectal circuits, Na + currents decrease, reducing intrinsic excitability [12]. Conversely, silencing synaptic inputs to developing Xenopus tectal neurons and Drosophila motorneurons increases Na + currents and intrinsic excitability [10, 12, 13]. Several mechanisms mediate homeostatic changes in Na + currents. Translational repression and post-translational phosphorylation reduce the density and open probability, respectively, of voltage-gated Na + channels in Drosophila motorneurons and rat cortical neurons in response to elevated synaptic activity [11, 14,15,16,17].

Multiple ion channels in the same neuron can balance each other to stabilize activity [2, 18, 19]. For example, the A-type K + channels shal and shaker are reciprocally regulated in motorneurons of Drosophila larvae: shaker is up-regulated in shal mutants, and shal is up-regulated in shaker mutants [20]. However, compensatory expression is not always a two-way street in Drosophila mutants of the delayed rectifier K + channel shab, increased expression of the Ca 2+ -dependent K + channel slo prevents motorneuron hyperactivity, but, loss of slo does not increase expression of shab [21]. Neurons can synergistically regulate ion channels with opposite effects on excitability to restore activity. Silencing of pyramidal neurons cultured from visual cortex of rat pups with TTX increases Na + currents and decreases K + currents [22]. Finally, neurons of the same type with similar excitability can vary significantly in their membrane conductances, which may reflect the complex homeostatic interactions between ion channels [23,24,25] (for more discussion, see [26, 27]).

Detailed examination of the distribution of ion channels revealed an important role of the axon-initial-segment (AIS) in intrinsic homeostatic plasticity. Changes in length and location of the AIS, a specialized region with clusters of voltage-gated Na + and K + channels involved in spike generation, can counter the effects of sensory deprivation or photostimulation [28,29,30,31]. In mice, eye opening at postnatal day 13–14 shortens the AIS of pyramidal neurons in visual cortex [32, 33]. Together, adjustments in ion channel density, distribution, and function, resulting from changes in transcription, translation, post-translational modifications, and trafficking, can alter intrinsic excitability and balance changes in synaptic input to maintain activity homeostasis [9, 34,35,36].

Homeostatic regulation of synapse strength and number

Homeostatic plasticity can regulate synaptic strength pre- and postsynaptically, and its dominant expression site can shift during development. In the early stages of network formation, miniature excitatory postsynaptic current (mEPSC) amplitudes increase when spike generation is blocked in cortical and hippocampal neuron cultures (i.e., suppression of intrinsic excitability), indicative of postsynaptic changes in AMPA receptor accumulation [37]. At later stages, presynaptic regulation of vesicle release and recycling is added, and mEPSC frequencies increase along with mEPSC amplitudes when spike generation is blocked [37, 38]. This suggests a developmental shift in the capacity for pre- and postsynaptic homeostatic plasticity [37]. Homeostatic control of synaptic strength has also been observed in vivo [39, 40]. The extent and expression site of this control depends on circuit maturation [41,42,43,44,45]. Homeostatic synaptic plasticity in layers 4 and 6 of primary visual cortex elicited by visual deprivation is restricted to an early critical period (postnatal day 16 to 21) [42, 43]. Later, homeostatic regulation of mEPSC amplitudes shifts to layers 2/3, where it persists into adulthood [42, 44]. The purpose of this shift in homeostatic plasticity across cortical layers remains unknown [41]. Chronic activity suppression by intracranial infusion of the Na + channel blocker TTX or NMDA receptor blockers increases spine densities of developing thalamocortical neurons in the dorsolateral geniculate nucleus of cats and ferrets [46, 47]. Thus, homeostatic plasticity can regulate synapse number as well as strength [48,49,50].

In addition to homeostatic synaptic changes elicited by experimental perturbations, Desai et al. showed that across development, mEPSC amplitudes in layers 2/3 and 4 of rat primary visual cortex decrease as mEPSC frequencies and synapse numbers increase [42]. Retinogeniculate circuits provide another example of developmental homeostatic co-regulation [51,52,53]. Initially, many retinal ganglion cells converge onto thalamocortical cells, each forming weak connections. Then, for up to 3 weeks after eye opening, thalamocortical cells prune inputs, retaining synapses from fewer ganglion cells, which strengthen their connections [53, 54]. Thus, presynaptic neurotransmitter release, postsynaptic receptor abundance, and synapse number are homeostatically co-regulated during normal development and after activity perturbations. In several systems, the expression sites and the combination of mechanisms engaged shift across development [2, 3, 55,56,57].

Homeostatic regulation of network activity

Homeostatic plasticity can stabilize the activity of individual neurons [54, 58, 59]. Neurons connect to each other in a cell-type-specific manner, forming circuits that perform specific functions. In the following sections, we discuss how homeostatic mechanisms are coordinated across neurons to stabilize circuit function [4, 60].

Homeostatic regulation of network excitation and inhibition

Network activity is determined by the ratio of excitation and inhibition (E/I ratio) [1, 4, 61]. In response to perturbations, developing circuits can differentially adjust inhibitory and excitatory connectivity to alter the E/I ratio and restore activity [62,63,64,65]. In developing hippocampal and organotypic cerebellar cultures, TTX or glutamate receptor antagonists decrease inhibitory synapse densities and strengths, whereas blocking GABAergic transmission with bicuculline increases the density of inhibitory synapses. Similarly, brain slice recordings in barrel cortex layer 4 showed that sensory deprivation selectively reduces inhibitory input to layer 4 spiny neurons in young but not in adult animals [66, 67]. Activity-dependent changes in inhibitory synaptic transmission appear to be regulated non-cell autonomously, as activity suppression of individual presynaptic or postsynaptic cells failed to elicit compensatory changes observed after global application of TTX in neonatal cultured hippocampal neurons [65]. It has been suggested that inhibitory interneurons may sacrifice their own firing rate homeostasis to stabilize spiking of cortical pyramidal neurons after global activity blockade [4, 68]. Another example of network homeostasis comes from studies of monocular deprivation during the critical period [4]. Here, homeostatic plasticity adjusts recurrent and feedforward connections between layer 4 circuits and layer 2/3 circuits in primary visual cortex. Visual deprivation via intraocular TTX injection increases the excitatory drive and reduces inhibitory drive from layer 4 to layer 2/3, compensating for the lost excitatory sensory input [4, 69, 70]. Intriguingly, in another deprivation paradigm (i.e., lid suture), increased intrinsic excitability and decreased E/I ratios stabilize activity in layer 2/3, indicating the same circuit can use different combinations of homeostatic mechanisms to compensate for sensory deprivation.

In addition to regulating excitatory and inhibitory synapse strength and number, homeostatic plasticity can switch the transmitter phenotype of neurons from glutamate to GABA or vice versa to adjust the E/I ratio of developing circuits [71,72,73]. In the embryonic Xenopus spinal cord, the fractions of neurons expressing excitatory transmitters increase and decrease, respectively, when network activity is pharmacologically suppressed and enhanced. These switches in transmitter phenotype occur without changes in the expression of cell identity markers [74]. Similar to homeostatic regulation of inhibitory synapses, the activity-dependent transmitter switch is non-cell autonomous and depends on network activity, evidenced by the reciprocal relationship between the number of silenced cells and the ratio of neurons expressing GABA vs. glutamate [75]. Whether switches in transmitter phenotypes contribute to network homeostasis during normal development remains to be investigated [71].

Homeostatic regulation of cell-type-specific connectivity

Recent advances in single-cell RNA sequencing together with large-scale morphological and functional surveys have revealed a great diversity of excitatory and inhibitory cell types, which serve distinct circuit functions [76,77,78,79]. This raises the questions whether, beyond categorical differences between excitatory and inhibitory neurons, homeostatic plasticity may act in a cell-type-specific manner to stabilize circuit function [80]. In the developing dentate gyrus, loss of excitatory drive by tetanus toxin expression results in reduced inhibitory input to granule cells [81]. This reduction is cell-type specific, affecting somatic innervation by parvalbumin-positive basket cells, but not dendritic innervation by calretinin- and somatostatin-expressing interneurons. Selective reduction of somatic inhibition efficiently restores the firing of granule cells [82, 83]. Similarly, monocular deprivation during a pre-critical period was shown to regulate feedback but not feedforward inhibition to layer 4 pyramidal cells in rat primary visual cortex [84] and early hearing loss weakens inhibitory synapses from fast-spiking interneurons but not from low-threshold spiking interneurons onto pyramidal cells [85, 86].

Homeostatic regulation of excitatory connectivity can also be cell type specific [87]. In the developing mouse retina, following removal of their dominant B6 bipolar cell input, ONα retinal ganglion cells up-regulate connectivity with XBC, B7, and rod bipolar cells, but leave input from B8 bipolar cells unchanged. This cell-type-specific rewiring not only maintains the sustained activity of ONα retinal ganglion cells, but also precisely preserves their light responses. Thus, homeostatic plasticity can regulate inhibitory and excitatory connectivity in a cell-type-specific manner to maintain the activity and sensory function of developing circuits.

Homeostatic regulation of patterned spontaneous activity

Throughout the nervous system, developing circuits spontaneously generate activity patterns that help refine their connectivity [88, 89]. Before eye opening, waves of activity originating in the retina propagate through the visual system and dominate activity up to primary visual cortex [90,91,92]. Retinal waves mature in three stages (I-III), in which different circuit mechanisms generate distinct activity patterns that serve specific functions in visual system refinement [88]. In mice, stage I waves, which are mediated by gap-junctional coupling of retinal ganglion cells, were first observed at embryonic day 17. Around birth, the wave generation switches to networks of cholinergic amacrine cells (stage II, postnatal day 1–10) followed in the second postnatal week by glutamatergic input from bipolar cells (stage III, postnatal day 10–14). The transitions between stages appear to be homeostatically regulated. When stage II (i.e., cholinergic) waves are disrupted by genetic deletion or pharmacological blockade of ß2 nicotinic acetylcholine receptors nAChRs, stage I waves persist until premature stage III waves take over [93,94,95,96]. Similarly, in VGluT1 knockout mice, in which stage III waves are abolished, stage II waves persist until eye opening [97]. Studies of developing spinal networks revealed an important role of excitatory GABAergic currents in homeostatic regulation of patterned spontaneous activity [98]. During development, GABA switches from excitatory to inhibitory as initially high intracellular Cl − concentrations are lowered by the developmentally regulated expression of cation-chloride cotransporters [99, 100]. When spontaneous network activity in chick embryos was reduced by injection of a sodium channel blocker, excitatory GABAergic mEPSC amplitudes were found to increase because of an increased Cl − driving force due to intracellular Cl − accumulation [101, 102].

Although homeostatic mechanisms can restore spontaneous activity patterns following perturbations, the extent to which these activity patterns support normal circuit refinement varies depending on age and means of perturbation and needs to be further investigated [103,104,105].


Inferior colliculus

The IC is a mandatory relay station on the ascending auditory pathway(Oliver and Heurta, 1992 Pollak et al., 2002 Malmierca, 2003 Morest and Oliver, 1984). Age-related changes of inhibition within the IC would likely impair the ability of the animal to further refine the localization of an environmental sound source from information received from the SOC, nuclei of the lateral lemnisci, and DCN (Vater et al.,1992 Litovsky and Delgutte,2002 Escabi et al.,2003 Pecka et al.,2007 Palmer et al.,2007). In addition, inhibition plays a role in processing acoustic delay information as well as strict temporal processing(Pollak et al., 2002). Delay coding is critical for echolocation in bats and may play a role in processing periodic vs aperiodic segments in communication signals. IC circuits utilizing both GABAergic and glycinergic inhibition have been shown to be important in coding selective communication calls in animals and are critically involved in delay circuits in bats(Yan and Suga, 1996 Portfors and Wenstrup, 2001 Klug et al., 2002). The IC receives excitatory glutamatergic inputs directly from the DCN as well as a major ascending projection from the SOC (for a review, see Kelly and Caspary, 2005). Extrinsic GABAergic projections to the IC arise bilaterally from the dorsal nuclei of the lateral lemniscus, while glycinergic inputs originate from the ventral nucleus of the lateral lemniscus and the LSO. In addition, intrinsic GABAergic neurons are located throughout both the central nucleus and the shell nuclei of the IC. IC neurons also receive a major excitatory descending projection from the auditory cortex(Winer et al., 1998 Winer et al., 2002 Winer, 2006).

As is the case for age-related changes described below, it is not known whether age-related inhibitory changes in IC are the result of de novo aging changes within the central nervous system or are the direct result of a gradual loss of peripheral input or both. In response to superthreshold acoustic stimulation, noise-exposed animals (modest damage to the auditory periphery) show altered evoked responses in the IC and auditory cortex, providing a functional picture suggestive of hyperexcitability(Willott and Lu, 1982 Popelár et al., 1987 Salvi et al., 1990 Gerken et al., 1991 Syka et al., 1994 Szczepaniak and Møller,1995 Wang et al.,1996 Syka and Rybalko,2000 Aizawa and Eggermont,2007). Neurochemical findings in support of these functional changes reveal that damage to the auditory periphery results in a selective down-regulation of normal adult inhibitory GABAergic function in the IC. Surface-recorded evoked potentials from the IC of noise-exposed rats show reduced sensitivity to bicuculline blockade(Szczepaniak and Møller,1995). Deafness resulted in decreased GABA release in vivo and decreased numbers of IC neurons showing electrically evoked suppression of unit activity (Bledsoe et al., 1995). IC GAD levels were reduced 2–30 days following noise exposure (Abbott et al.,1999 Milbrandt et al.,2000). GABA uptake and release following ossicle removal or cochlear ablation resulted in complex long-term changes in GABA and glycine neurochemistry (Suneja et al.,1998). Collectively, these studies suggest that decreased acoustic input at the auditory periphery results in significant changes in GABA neurotransmission in normal adult IC.

The effects of aging on protein levels of GABAA receptor subunitα 1, β2 and γ1 in the IC of FBN and F344 rats. The y-axis represents subunit protein percentage difference from young adult animals (3–4 months) of middle aged(18–20 months) and aged (28–32 months) in rat IC. Note that GABAA receptor γ1 subunit protein significantly increased in aged rats of both FBN and F344. While significantly decreasedα 1 subunit protein was found in the IC of aged FBN rats, it was not found in aged F344 rats ( * P<0.05). (Modified from Caspary et al.,1999.)

The effects of aging on protein levels of GABAA receptor subunitα 1, β2 and γ1 in the IC of FBN and F344 rats. The y-axis represents subunit protein percentage difference from young adult animals (3–4 months) of middle aged(18–20 months) and aged (28–32 months) in rat IC. Note that GABAA receptor γ1 subunit protein significantly increased in aged rats of both FBN and F344. While significantly decreasedα 1 subunit protein was found in the IC of aged FBN rats, it was not found in aged F344 rats ( * P<0.05). (Modified from Caspary et al.,1999.)

Age-related changes in inferior colliculus

Single unit recordings from the IC of aged rats show a significant decrease in the level of inhibition within the excitatory response area, an increase in the breadth of the excitatory response area at 30 dB above threshold, and less precise temporal processing of modulated sounds(Palombi and Caspary, 1996a Palombi and Caspary, 1996b Palombi and Caspary, 1996c Palombi and Caspary, 1996d Shaddock-Palombi et al.,2001). Similar physiological changes occur in C57 and CBA mice(Willott, 1986 Willott et al., 1988 Willott et al., 1991 McFadden and Willott, 1994 Walton et al., 1998 Walton et al., 2002 Simon et al., 2004).

A number of measures of presynaptic GABA neurotransmission show age-related changes in the mammalian IC. GABA levels, GABA immunostaining, GAD activity and GABAA and GABAB receptor binding all decrease in the aged rodent IC (Banay-Schwartz et al.,1989a Banay-Schwartz et al.,1989b Caspary et al.,1990 Gutiérrez et al.,1994 Raza et al.,1994 Milbrandt et al.,1994 Milbrandt et al.,1996). The IC neuropil shows an age-related rearrangement of synaptic endings onto soma and proximal dendrites(Helfert et al., 1999).

Possibly in response to age-related presynaptic changes, age-related postsynaptic changes occur in the mammalian IC GABAA receptor. The GABAA receptor is a heterogeneous family of ligand-gated Cl – ion channel receptors, which receive input from GABA-releasing inhibitory circuits. GABAA receptors exist as pentameric subunit complexes made up of combinations of 19 possible GABAA receptor subunits, which can be activated/allosterically modulated by numerous pharmacological agents(Sieghart, 1992a Sieghart, 1992b Wafford et al., 1993 Sieghart, 1995 Rabow et al., 1995 Möhler et al., 2002). Thus, changes in the make-up of the GABAA receptor would alter the function of sensory coding in the aged IC. In the IC, both GABAAreceptor subunit message and protein levels show age-related changes, with a significant down-regulation of the adult α1 subunit in favor of an up-regulation of the α3 GABAA receptor subunit (Gutiérrez et al.,1994 Milbrandt et al.,1997 Caspary et al.,1999). In situ hybridization and western blot studies show significant age-related up-regulation of the γ1 subunit(Fig. 3) suggestive of a compensatory age-related increase in the affinity for GABA(Milbrandt et al., 1997 Caspary et al., 1999). Coexpression of the γ1 subunit with α1 andβ 2 subunits in oocytes produces a GABAA receptor complex, which fluxes more Cl – ions per mmol GABA than wild-type γ2 subunit containing receptor constructs(Ducic et al., 1995). Receptor binding studies found significant age-related enhancement of GABA's ability to modulate binding at the picrotoxin GABAA receptor site(Fig. 4)(Milbrandt et al., 1996). Modulation of GABAA receptor binding at this site using bath-application of GABA resulted in an age-related, dose-dependent shift to the left in the GABA modulation curve (Fig. 4) (Milbrandt et al.,1996). This dose–response shift in the binding assay further supports the observed age-related subunit changes.

GABA (10 nmol l –1 –10 mmol l –1 )modulation of 3 [H]-TBOB (t-butylbicycloorthobenzoate)binding in the CIC of young and aged F344 rats. The dose–response curve is shifted to the left. These data have functional implications since the aged GABAA receptor must be more sensitive to GABA than the young GABAA receptor for the channel to be open allowing TBOB to bind to the picrotoxin binding site. (Modified from Milbrandt et al., 1996.)

GABA (10 nmol l –1 –10 mmol l –1 )modulation of 3 [H]-TBOB (t-butylbicycloorthobenzoate)binding in the CIC of young and aged F344 rats. The dose–response curve is shifted to the left. These data have functional implications since the aged GABAA receptor must be more sensitive to GABA than the young GABAA receptor for the channel to be open allowing TBOB to bind to the picrotoxin binding site. (Modified from Milbrandt et al., 1996.)

In addition, a direct measure of age-related subunit efficacy was obtained by examining the ability of GABA to flux Cl – ions into microsac/synaptosome preparations from rat IC. GABA influx was significantly increased in samples from aged rat IC, confirming oocyte expression studies noted above (Caspary et al.,1999). These findings differ with previous whole brain synaptosome chloride uptake studies, which found reduced Cl – uptake with aging (Concas et al., 1988). Age-related GABAA receptor changes may reflect a partial postsynaptic compensation for the significant age-related loss in presynaptic GABA release.

FBN rat layer V neurons exhibit two major types of receptive field maps.(A) 32% showed the classic V/U-shape with young and aged neurons showing similar responses to current pulse stimulation (not shown). (B) 47% of pyramidal neurons demonstrated a more complex, dynamic response map. (C) Aged complex receptive field neurons responded more vigorously than young neurons to 200 ms current pulses, suggesting altered inhibitory control. Such increased excitability to current would be consistent with reduced GAD67 immunostaining around layer V (LV) somata (see insets scale bars, 5 μm). (Modified from Turner et al., 2005c.)

FBN rat layer V neurons exhibit two major types of receptive field maps.(A) 32% showed the classic V/U-shape with young and aged neurons showing similar responses to current pulse stimulation (not shown). (B) 47% of pyramidal neurons demonstrated a more complex, dynamic response map. (C) Aged complex receptive field neurons responded more vigorously than young neurons to 200 ms current pulses, suggesting altered inhibitory control. Such increased excitability to current would be consistent with reduced GAD67 immunostaining around layer V (LV) somata (see insets scale bars, 5 μm). (Modified from Turner et al., 2005c.)

Taken together, these changes suggest a net down-regulation from normal levels of adult inhibitory function in the aged animals leading to a degradation of temporal and binaural coding in the aged IC.


Inhibitory and excitatory synapse dynamics in the brain

The brain adapts to the environment in part by persistently modifying and rearranging the diverse synaptic connections between neurons. These changes include strengthening or weakening existing links, as well as forming and eliminating synapses — long-term adjustments that are required for learning and memory.

Since excitatory synapses on excitatory neurons are localized to small protrusions called dendritic spines, earlier studies have used dendritic spine dynamics to monitor excitatory synaptic remodeling in vivo. However, the lack of a morphological surrogate for inhibitory synapses has precluded their observation, and although the interplay between excitatory and inhibitory transmission maintains a critical role in brain plasticity, the inability to monitor inhibitory synapse dynamics has prohibited examination of how they correspond with excitatory changes.

Sensory input impacts synapse activity

A new study co-authored by Elly Nedivi, Picower Institute for Learning and Memory, her students Jerry Chen and Katherine Villa from the Department of Biology, and colleagues Jae Won Cha and Peter So from the Department of Mechanical Engineering at MIT, as well as collaborator Yoshiyuki Kubota from the National Institute for Physiological Sciences in Japan, characterizes the distribution of inhibitory synapses across the brain’s neurons and shows that they are divided into two populations, one on dendritic spines adjacent to an excitatory synapse, the other on the dendritic shaft. They then measured the remodeling kinetics of the two populations during normal and altered sensory experience.

Researchers simultaneously monitored inhibitory synapses and dendritic spines across brain neurons using high-resolution dual-color two-photon microscopy. Their findings indicate that inhibitory spine and shaft synapses respond differently during normal and altered visual sensory experience, and when the inhibitory synapses and dendritic spines of cortical neurons are rearranged, they are locally clustered, based on sensory input. This work is slated to appear in the April 26 issue of Neuron.

To date, the distribution of inhibitory synapses on cell dendrites was estimated via volumetric density measurements. The new MIT study, however, demonstrates uniform distribution of inhibitory shaft synapses — versus spine synapses, which are twice as abundant along distal apical dendrites — suggesting that the two types of synapses have different roles in shaping dendritic activity. The differential distribution of inhibitory spine and shaft synapses may reflect their influence on the integration of calcium input from various sources.

Kinetic and clustering distinctions between synapse types

The research team also discovered that both synapse types are dynamic, but inhibitory spine synapses are fourfold more dynamic than their shaft counterparts. Monocular deprivation (MD), a visual paradigm for plasticity, resulted in a significant but transient loss of inhibitory spine synapses during the first two days of MD, while loss of shaft synapses persisted for at least four days. This demonstrates the impact of altered sensory experience and the kinetic distinction between the two synapse populations.

Next, the scientists looked for evidence of local clustering between excitatory and inhibitory synaptic changes during normal visual experience by performing analyses on dynamic and stable inhibitory synapses and spines. They found that inhibitory synapse changes occur in closer proximity to dynamic dendritic spines as compared to stable spines, and dendritic spine changes occur in closer proximity to dynamic inhibitory synapses as compared to stable ones. Researchers also demonstrated that this clustering pattern between dynamic inhibitory synapses and dendritic spines was enhanced by MD.

The researchers also proved that the percent of clustered dynamic spines and inhibitory synapses in response to MD is significantly higher than would be expected simply based on an increased presence of dynamic inhibitory synapses. “This suggests that while MD does not change the overall rate of spine turnover on cortical neurons, it leads to a greater coordination of these events with dynamics of nearby inhibitory synapses,” Nedivi explains.

New insights reveal potential impact on long-term memory

The ability of the MIT researchers to distinguish inhibitory spine and shaft synapses provides new insight into inhibitory synapse dynamics in the adult visual cortex. The inhibitory synapse losses that occur during altered visual experience, noted above, are consistent with findings that visual deprivation produces a period of disinhibition in the visual cortex.

In addition, the results of this MIT study provides evidence that experience-dependent plasticity in the brain is a highly orchestrated process, integrating changes in excitatory connectivity with the active elimination and formation of inhibitory synapses. This sheds new light on the importance of coordinating excitatory and inhibitory circuitry to help nurture long-term memory.


Imaging Plasticity

Many CPGs are capable of modifying neuronal structure, suggesting that neuronal structure can be modified by activity even in the adult brain. To investigate the molecular mechanisms underlying structural plasticity in the mammalian brain we have a long-standing collaboration with Peter So’s lab in the Department of Mechanical Engineering at MIT to develop multi-photon microscopy for large volume, high resolution imaging of dendritic arbor and synaptic structural dynamics in vivo. Using this system we have imaged and reconstructed the dendritic trees and axon collaterals of neurons in visual cortex of thy1-EGFP transgenic mice. These transgenic mice express EGFP in a random subset of neurons sparsely distributed within the superficial cortical layers that are optically accessible through surgically implanted cranial windows. The images show an abundance of detail, with the basal and apical dendrites instantly recognizable and spines clearly visible. This enables dendritic branch dynamics in individual neurons to be examined over several months. We were the first to reveal that dendritic arbor remodeling in the adult is restricted to inhibitory interneurons (Lee et al., PLoS Biology 2006), that interneuron remodeling is not a feature determined by cell lineage, but rather is imposed by the laminar cortical circuitry (Lee et al., PNAS 2008), and is elicited by sensory experience in an input-specific manner so as to facilitate or attenuate experience-dependent plasticity (Chen et al, Nature Neuroscience 2011). Further we showed that inhibitory dendrite remodeling is a common element of structural plasticity mechanisms built into the neocortical module (Chen et al., J. Neurosci. 2011).

Once interneurons were recognized as key players in activity-dependent circuit plasticity, a major obstacle to addressing questions regarding inhibitory synapse plasticity as well as how it may be coordinated with excitatory inputs at the dendritic level, has been the inability to visualize inhibitory synapses in vivo. We therefore expanded our high-resolution large volume imaging capabilities with incorporation of two-color multiphoton microscopy to simultaneously monitor both inhibitory synapse and dendritic spine remodeling across the entire dendritic arbor of cortical L2/3 pyramidal neurons in vivo during normal and altered sensory experience. This has allowed us to address, for the first time, fundamental questions about the interplay between excitatory and inhibitory synaptic transmission during normal adult brain function as well as experience-dependent plasticity (Chen et al., Neuron 2012). Currently we are working to integrate additional colors into our imaging set up to enable simultaneous tracking of multiple synaptic and cellular components. We are also interested in integrating Ca+2 imaging with our structural markers.

Molecular Mechanism of Inhibitory Synaptic Rearrangements

Work from our lab and others (Cane et al., 2014 Chen et al., Neuron 2012 Kelsch et al., 2008 van Versendaal et al., 2012), has shown that both excitatory and inhibitory synapses are extremely plastic in vivo. While many of the molecular players that underlie the formation of excitatory synapses are well characterized, little is known about the molecular regulation of inhibitory synapses. My project uses molecular biology techniques as well as multi-color two photon microscopy to probe specific interactions that regulate the formation and elimination of inhibitory synapses in vivo.

In Vivo Two-Photon Imaging of Dendritic and Synaptic Structural Changes during Early Development

The ultimate structure and function of our brains are directly influenced by our experiences and interactions with the world around us. The critical period for ocular dominance plasticity is a time during development where a change in visual input to the visual cortex can cause profound changes in the connectivity of this area of the brain. For several years, our lab has developed techniques to image the entire dendritic arbors of neurons with synaptic resolution over time. In the adult, we have studied the plasticity of various types of neurons and characterized the dynamics of this plasticity during normal experience and during experimental manipulation. My project makes use of the genetic labeling and high resolution imaging techniques developed in the lab to investigate the formation of inhibitory and excitatory connections and their activity-dependent selection during the cortical critical period for ocular dominance plasticity in mice.

Mapping specificity of afferent inputs distribution across L2/3 pyramidal cell arbors

Surprisingly, we know little at the single cell level regarding the distribution of particular types of inputs onto any given cortical cell type. What is the cortical vs subcortical input source of excitatory synapses located across the dendritic arbor, and do their synaptic dynamics differ depending on this source? What is the specific interneuron type targeting inhibitory synapses located across the dendritic arbor, and do their synaptic dynamics differ depending on this source? Do inhibitory shaft synapses receive different inputs than inhibitory spine synapses? Their very different kinetics and response to visual deprivation, suggest that may be the case (Chen et al., Neuron 2012). Are inhibitory synapses receiving specific types of afferent inputs more likely to exhibit coordinated dynamics with excitatory synapses? My project uses in vivo multicolor two-photon microscopy and transgenic mouse lines to identify the afferent sources to synapses at different dendritic locales, determining whether synapses targeted by certain afferents are less or more dynamic, and whether their dynamics are coordinated.

Experience Dependent Stabilization of Synapses in Vivo

We find that in the cpg15 knockout mouse (cpg15 KO) there is abnormal postnatal development of excitatory connectivity in the hippocampus (Fujino et al., Genes & Dev. 2011), as well as visual cortex (Picard et al., J. Neurosci. 2014). In the dentate gyrus of the hippocampus as many as 30% of spines initially lack synapses. Chronic in vivo imaging of cortical pyramidal neurons through a cranial window shows that while dendritic spine dynamics in cpg15 KO mice are comparable to controls, fewer of the dynamic events in these mice are stabilized, and thus favor persistent spine loss (Fujino et al., Genes & Dev. 2011). These results suggest that the in vivo developmental deficits in the cpg15 KO mouse derive from lack of a synaptic stabilization signal. My project makes use of new in vivo synaptic labeling methods and multi-color two-photon microscopy to examine how CPG15 regulates synapse stabilization in vivo and explores whether CPG15 could link experience driven neuronal activity and new synapse stabilization.

In Vivo two Photon Imaging of Modulatory Cortical Afferent Inputs

Widespread projections of neuromodulatory neurons influence the remodeling of local neural circuits and modify their response to stimuli. Despite the critical importance of modulatory cortical afferents on plasticity, little is known about their connections in the cortex at the cellular level. I will utilize multiphoton in vivo microscopy to characterize the distribution of modulatory afferent inputs to the cortex and how they are associated with structural plasticity. By monitoring the interaction between defined modulatory inputs and local neural synaptic components we can begin to address the local influence of neuromodulators on structural plasticity of inhibitory and excitatory synapses in sensory cortex.


Discussion

We explored the plastic changes in network structure that can be induced by transcranial direct current stimulation (tDCS), exploiting the homeostatic response of synaptic growth and decay. We demonstrated that weak subthreshold DC stimulation induces changes of neuronal firing rates and, thus, triggers network remodeling and cell assembly formation. Depolarized neurons first reduce the number of excitatory input synapses during stimulation, but then create new excitatory synapses predominantly with other stimulated neurons after stimulation is off. Interestingly, hyperpolarization also causes new synapses being formed preferentially among stimulated neurons. Stimulation triggers a profound and sustainable reorganization of network connectivity and leads to the formation of cell assemblies. With the help of our model, we explored different parameters of tDCS stimulation and found that strong and focused stimulation generally enhances the newly formed cell assemblies. We also observed that repetitive stimulation with well-chosen duty cycles can boost the induction of structural changes, and repetitive stimulation with alternating polarization may induce even higher connectivity changes.

We used network connectivity as a direct readout of stimulation effects, which is possible in model simulations, but cannot easily be done in experiments. However, the factors that we found to amplify the overall impact of stimulation are not unheard of in tDCS practice. Strong and focused stimulation, for example, which results from a high-definition electrode montage, does indeed lead to a stronger readout (MEP) and potentiates the therapeutic effects as compared with a conventional montage (Kuo et al., 2013). While applying the same total current, a high-definition montage induces stronger electric fields in smaller brain volumes (Edwards et al., 2013). Moreover, a high-definition montage narrows down the most affected brain region. We also found in our model that both factors indeed contribute to the induction of higher connectivity. Moreover, repetitive stimulation can boost connectivity, provided the duty cycles are chosen right. In fact, it has been demonstrated in experiments (Monte-Silva et al., 2013) that two 13-min stimulations interrupted by a 20-min pause yields stronger MEP aftereffects than a single, uninterrupted 26-min stimulation, while a repetition with a 24-h pause in between could not accumulate the aftereffects at all. In our model, we likewise found that multiple stimulation episodes with properly chosen pauses can achieve better effects than a single, uninterrupted stimulation.

Other computational approaches have been employed previously to analyze the neuron-scale mechanisms underlying tDCS or DCS. Most notably, Bikson et al. (2006) has explored several aspects of this: extracellular potassium concentration, polarization of the axonal terminal, action potential timing, and inhibitory neurons. Joucla & Yvert (2009) provided an estimate of membrane potential changes for large axons exposed to an electric field, and Aspart, Ladenbauer, & Obermayer (2016) conceived the influence of the electric field on neuronal dendrites as external input to the soma. Another computational approach based on modern neural imaging methods has shed light on the question of how strong the stimulation effects actually are. Spherical head models were first used to estimate the 3-D current flow for any given electrode montage (Miranda, Lomarev, & Hallett, 2006). Later, fMRI-based modeling was employed to devise individualized treatment of stroke or depressive patients (Datta et al., 2009 Ho et al., 2014 Huang et al., 2017). Our present work adopted insight and parameters from both approaches. In addition, we developed a new and original computational model to explore the impact of structural plasticity at the level of networks. This provides a bridge between the level of single neurons and the level of large-scale networks. Although our model contributes new explanations for some core observations in tDCS practice, there are still important issues left that cannot be appropriately addressed with our highly simplified model lacking relevant features of brain geometry. Also, the exact rules of growth and the timescales involved in homeostatic structural plasticity remain to be elucidated in experiments. To treat the influence of tDCS on network dynamics and structural plasticity of multiple brain regions would require a “network of networks” approach, which is, however, beyond the scope of our current study.

What are the actual effects of tDCS on network activity and function? Although robust and sustained effects of tDCS using relatively weak stimulation currents (1–2 mA) have been demonstrated (Nitsche et al., 2009 Nitsche & Paulus, 2000), Horvath et al. (2015) pointed to the difficulty reproducing positive results. Recently, Vöröslakos et al. (2018) have shown that the amount of membrane polarization due to tDCS depends on the strength of the applied current, and that there should be indeed no effect expected for very low intensities. Our simulation results suggest, however, that repetition could boost the impact on connectivity. The peak connectivity reached after sufficiently many repetitions, however, depends on stimulus intensity. Very weak stimulation cannot achieve high connectivity changes, even if repeated ad infinitum. Strong stimulation within a safe range could achieve higher connectivity, but too strong stimulation may lead to unfavorable network dynamics. Our model predicts very clearly that the accumulated effect achieved by stimulation depends not only on the exact repetition pattern, but also on stimulation intensity. On the other hand, a quantitative assessment of the aftereffects is difficult. In our work, the effect of tDCS on the network is quantified by measuring anatomical connectivity among stimulated and nonstimulated neurons. Such measurement is currently not possible in experiments, neither in vivo nor in vitro. Transcranial stimulation perturbs neuronal firing rates transiently and leads to the formation of cell assemblies, which persist after tDCS has been switched off and neuronal activity is back to baseline. Therefore, considering the homeostatic nature of structural plasticity, it is actually impossible to measure the effect of tDCS using simple neuronal activity measures. The question is, what are the effects of altered connectivity on the activity and the function of neuronal networks, and how can these effects be measured? This is a very interesting question, and the answer is complicated. Even if newly formed cell assemblies do not affect spontaneous activity as the firing rate of the neurons may be homeostatically regulated, they might still influence the evoked responses of neurons. Interestingly, Horvath et al. (2015) reviewed many tDCS studies and found that stimulation has a reliable effect only on the MEP amplitude, out of many potential biomarkers that were tested. The debate about the effects of tDCS on network function should, therefore, include the measures to quantify the outcome of a stimulation.

Another important issue raised by our work is that the total effect of stimulation might be too weak for detection. The connectivity changes triggered by a single cycle of polarization at ΔVm = 0.1mV can only be detected if the full connectome is available for quantification. While possible in simulations, such a scenario is unrealistic in an experimental setting. Our simulation results suggest, however, that the outcome should increase upon repetitive stimulation and, therefore, possibly becomes easier to measure. The measurement time window of tDCS effects adds another puzzle to this question. The connectivity of the stimulated plastic network undergoes constant changes. During and after stimulation, for instance, total connectivity decreases and increases fast, constituting the homeostatic response. In contrast, the newly formed cell assembly persists for much longer periods and decays only with a slower time constant. It is not yet clear, however, which parameters influence this time constant, and it might be that different current intensity and electrode size have an impact on it. In fact, Jamil et al. (2017) recently observed in experiments that the current intensity might interact with the duration of stimulation needed for the homeostatic reversal of plasticity. If the exact stimulation protocol indeed influences the timescale of the aftereffect, naively comparing tDCS effects under different stimulation conditions “before” and “after” does not provide sufficient information regarding its outcome. In view of this, using a measure that takes the dynamics of the changes triggered by stimulation into account, such as the IG measure introduced in this work, could quantify the effects of stimulation much more reliably.

In general, one needs to interpret the results and predictions of our work on network remodeling induced by tDCS with due caution. Our current work, however, could be a first step toward the goal of understanding and optimizing tDCS performance. More experiments addressing the impact of tDCS in human and in animal brains are definitely needed, and the results of our simulation study might indicate some new directions.