Electrical synapses between inhibitory neurons shape the responses of principal neurons to transient inputs in the thalamus: a modeling study

As multimodal sensory information proceeds to the cortex, it is intercepted and processed by the nuclei of the thalamus. The main source of inhibition within thalamus is the reticular nucleus (TRN), which collects signals both from thalamocortical relay neurons and from thalamocortical feedback. Within the reticular nucleus, neurons are densely interconnected by connexin36-based gap junctions, known as electrical synapses. Electrical synapses have been shown to coordinate neuronal rhythms, including thalamocortical spindle rhythms, but their role in shaping or modulating transient activity is less understood. We constructed a four-cell model of thalamic relay and TRN neurons, and used it to investigate the impact of electrical synapses on closely timed inputs delivered to thalamic relay cells. We show that the electrical synapses of the TRN assist cortical discrimination of these inputs through effects of truncation, delay or inhibition of thalamic spike trains. We expect that these are principles whereby electrical synapses play similar roles in regulating the processing of transient activity in excitatory neurons across the brain.

rhythms associated with slow-wave sleep or absence epilepsy 19,20 . However, the role of electrical synapses in TRN on the transient, stimulus-evoked TC activity is relatively underexplored.
Here we examine the role of electrical synapses within TRN on the activity patterns of TC neurons. We use a reduced model of four cells: two pairs of reciprocally connected TC-TRN neurons, with an electrical synapse between the two TRN neurons. We deliver closely-timed inputs, mimicking inputs of similar temporal, spatial, or frequency arriving from sensory surround, to the TC cells. We examine how TC spiking is impacted by the inhibition delivered from the coupled TRN neurons. Our results demonstrate that electrical synapses can either fuse or further separate input-generate spiking, and we predict that these effects ultimately impacting the ability of recipient cortical cells to discriminate between the inputs.
Chemical synapses include fast inhibitory GABA A (E GABA = −75 mV) and excitatory AMPA (E AMPA = 0 mV) synapses. Synaptic conductance kinetics is implemented with a pair of fall and rise time constants (with τ rise = 0.1τ fall for both synapses); τ fall,GABA = 5 ms and τ fall,AMPA = 2 ms. Implementation of synapses resembles NEURON's implementation of 22 Our simulated network consists of 2 TRN cells connected via a single electrical synapse, and 2 TC cells receiving external input (Fig. 1A), simulated for 250 ms. Within the network, TRN cells each send inhibitory input to TC cells via GABA A synapses, and TC cells send excitatory inputs to TRN cells via AMPA synapses. Since the model was easily excitable, no additional DC current was sent to TRN neurons to reach subthreshold excitation. External inputs are AMPAergic excitatory inputs and provided only to the two TC cells.
Model electrical synapses (gap junctions) are linear and symmetrical. Values of electrical synapse conductance (G elec ) varied from 0 to 0.025 mS/cm 2 , which converts to a coupling coefficient of roughly 0.2883. An applied current, iDC, was used from 0 to −0.1 µA/cm 2 to quantify membrane conductance G m ≈ 0.0551 mS/cm 2 and the effective coupling coefficient values, which do not differ significantly with the theoretical values of G m /(G m + G elec ) (data not shown).
The maximal GABAergic conductance (G GABA ) varied from 0 to 0.05 mS/cm 2 , and the maximal AMPAergic conductance (G AMPA ) between TC and TRN cells was fixed at 0.05 mS/cm 2 to make sure that at least two spikes SCientifiC REPORTS | (2018) 8:7763 | DOI:10.1038/s41598-018-25956-x from a TC cell were required to elicit response from a TRN cell in an uncoupled network. External input to TC 1 (in) stays constant at maximal conductance (G in1 ) of 0.06 mS/cm 2 , and arrives (t in1 ) at 60 ms. The arrival times (t in2 ) of external input to TC 2 (in 2 ) varied between 10 and 110 ms. The maximal conductance (G in2 ) of the input to TC 2 varied from 0.02 to 0.1 mS/cm 2 .
Each simulation was simulated in parallel using MATLAB R2016 Parallel Toolbox and Lehigh University High Performance Computing Resources (Sol), solved with MATLAB's ode23 Runge-Kutta implementation using a maximum timestep of 0.01 ms.

Analysis.
For each simulation, the number of spikes in each TC neuron was extracted. Temporal independence (ψ) and spiking separation (φ) were defined as outlined below.
Thalamocortical spiking window and normalized temporal independence. The spiking window (σ) of a neuron was defined to be the temporal duration between its first spike and last spike, with addition of a 5-ms window following the last spike used to allow for EPSP decay in a cortical cell (Eq. 4). If a neuron does not fire, its spiking window is considered to be empty σ = . ( From each TC spike train, relative independence of the corresponding TC (Eqs 5 and 6) was calculated and normalized to values between 0 (complete overlap) and 1 (no overlap) as described in the following equations. If either of TC neurons does not spike, we consider them to be completely independent (Eq. 6*).

]: Time interval between a, b; |x|: length of x
Thalamocortical temporal separation. The separation of TC 1 and TC 2 spike trains (Eq. 7) was computed as the time difference between the spiking window termination of the leading neuron and the spiking window beginning of the following neuron. Separation is considered to be positive when both TCs spike and are independent of each other (Eq. 7a). In overlapping cases, this measure is artificially set as negative, as loss of separation, with the magnitude as the temporal amount of overlap (Eq. 7b). In cases of either TC 1 or TC 2 not spiking, this measure is undefined (Eq. 7c).
Gain in independence and temporal separation. For each pair of (G elec , G GABA ) we represented independence (or separation) in the parameter space of input differences between in 2 and in 1 For each pair (G elec ≠ 0, G GABA ), we calculated the gain in independence (or separation) as the difference of the quantitative measure from its value in the uncoupled case (G elec = 0) for each G GABA . In other words, the representation becomes in in , we calculated a normalized combination for input differences as

Results
To examine the impact of electrical synapses of the TRN on thalamocortical transmission, we constructed a 4-cell model comprising two pairs of reciprocally connected TC and TRN cells, with an electrical synapse between the two TRN cells (Fig. 1A). We used values of G elec from zero to 0.025 mS/cm 2 , corresponding to coupling coefficients measured in our model TRN neurons ranging from 0 to 0.3, mimicking the physiological range of cc values in TRN 14,23 . Input delivered to TC 1 was fixed at a constant arrival time and strength, and the inputs to TC 2 were varied in arrival time and input strength. We used input strengths that resulted in a burst of spikes that is typical of thalamic cells, lasting for tens of ms (e.g. Fig. 1B). To characterize the impact of electrical synapses on the output of the TC cells, we quantified independence (ψ) as the percentage as inversely related to the temporal intersection of the two TC spike trains, normalized to the duration of spiking. We also quantified separation (φ) as the time interval between the termination of one spike train and the onset of spiking in the other TC cell; negative values of separation result from overlapping trains. Higher values of independence and positive values of separation, we believe, increase the chances that a cortical cell that receives inputs from both of these TC cells will be able to discriminate the input as arising from different sectors of the sensory environment (e.g. whiskers).
With no electrical synapse between the pairs, the inhibitory feedback from TRN to TC cells is the sole influence that acts to separate spike trains for our inputs, and thus the inputs that TC relays to cortex (Fig. 1B). More specifically, increases in inhibitory strength between TRN and TC cells result in earlier termination of spiking in the TC cells ( Fig. 1B-D). This resulted in increases in both separation, starting from trains that overlapped by −32 ms at G GABA = 0 to a separation of 9.5 ms at G GABA = 0.05 ms/cm 2 , and increases in independence (43 to 100% over the range of G GABA ) of the trains from each other (Fig. 1E). However, the spike times of TCs themselves remain unchanged in this case.
For increases in strength of the electrical synapse between TRN cells, we observed three effects that ultimately impacted the separation and independence of TC spiking in our simulations: latency of spiking in the TC that received inputs later, truncation of spike trains, and prevention of any spiking.
An example of electrical synapse-mediated changes in latency is shown in Fig. 2. In this case, input was delivered 40 ms to TC 2 after the input to TC 1 , and both inputs were of the same strength (0.06 mS/cm 2 ). Inhibitory strength between TRN and TC cells was relatively low in this example simulation (G GABA = 0.02 mS/cm 2 ). While the spike trains are always independent in this set of simulations, the initial separation (without electrical synapses) is 9.2 ms ( Fig. 2A i ), and we note that a cortical neuron that receives and sums input trains from both TC 1 SCientifiC REPORTS | (2018) 8:7763 | DOI:10.1038/s41598-018-25956-x and TC 2 may not differentiate between one spike train that starts 9.2 ms after the termination of the previous train. Increases in electrical synapse strength within the model resulted in systematically delayed spiking of TC 2 from its input, from 8.5 to 51 ms ( Fig. 2A i-iv ) with corresponding increases in separation between trains up to 55 ms ( Fig. 2A iv ), and ultimately acted to prevent spiking in TC 2 ( Fig. 2A v ). In these cases, separation values follow spike latencies and show a strong dependence on both electrical and inhibitory synapse strengths (Fig. 2B,D). We note that electrical synapse strength acts synergistically with inhibitory synapse strength (Fig. 2E), such that increases in both parameters result in large latency changes, or ultimately prevention of spiking in TC 2 (whitespace in Fig. 2D,E).
Spike train truncation arises for stronger values of inhibition within the network, and this effect modulated by the electrical synapse (Fig. 3). We illustrate this effect with a simulation in which a weaker (0.05 mS/cm 2 ) input was delivered to TC 2 20 ms after the input to TC 1 . In this set, as electrical synapse strength increases, spiking in TC 2 decreases (Fig. 3A). Independence of the TC spike trains covaries with increases in electrical synapse strength, as the spike train of TC 2 diminish in spike count (Fig. 3D), and separation also increases (Fig. 3E). In this example, the baseline condition, with no or weak electrical synapses, is features moderately overlapping TC spike trains. However, the stronger inhibition in this example (G GABA = 0.045 mS/cm 2 ) acts in concert with increasing values of electrical synapse strength to terminate any prolonged spiking of TC 2 , from an initial 3 spikes to 1 spike (Fig. 3A i-iii ), effectively decreasing overlap to eventually result in complete independence (from 55 to 100%) and increases in separation (−12.7 to 0.2 ms). For varied values of inhibitory strength within the network, the relationships between independence, separation and electrical synapses become more complex, resulting from the dependence of rate (Fig. 3D) on both electrical and inhibitory synapse strength.
In general, spiking rate in TC 2 depends on both inhibitory and electrical synapse strength, as well as on the details of the input, specifically on its arrival time and strength. The dependence of spiking rate in TC 2 on electrical and inhibitory synapse strength is shown in Fig. 4. Generally, we see that electrical synapses between TRN cells are more effective in terminating trains for more temporally separated inputs (Fig. 4A-C), and require a minimal amount of electrical coupling (G elec > 0.01) and inhibitory coupling (G GABA > 0.015 mS/cm 2 ) to separate those inputs. The delays for input integration in TRN cells and TC cells determine the timescale at which electrical synapses influence the circuit. On the other hand, when inputs arrive more closely timed (Fig. 4G-I) and are more similar in strength (Fig. 4H), it takes both higher G elec and G GABA to cause any effects. In other words, the more different the inputs are, either in time or amplitude, the larger the effects are from the electrical and/or the inhibitory synapse.
Beyond spike number, electrical and inhibitory synapses also act together to regulate spike train separation (Fig. 5). Similar to the case for spiking rate, inputs arriving at the same time lead to overlapping TC spiking even for maximal electrical coupling and feedback inhibition (Fig. 5G-I), and the biggest impact of electrical synapses on separation is seen for inputs that are different in arrival time ( Fig. 5A-C). Interestingly, weaker inputs require less inhibitory strength for the electrical synapse to increase temporal separation (Fig. 5A,D,G), and can prevent spiking when the inputs are far enough apart in time (Fig. 5A). This is a result of increased thalamocortical spiking delay, which results in larger separation values, and more time for weak or medium inhibition to act through electrical synapse and exert effects on the other TC cell.
For closely timed and similar inputs, electrical synapses are relatively ineffective in separating the inputs; in fact, for similar inputs, we observed that electrical synapses, acting through inhibitory synapses, instead result in temporal fusion of the inputs. This effect is shown in the progressive decrease in independence, indicated by the increased area of blue shading across sets of panels, in Fig. 6, where the center of each panel represents inputs of identical size and arrival time. Increases in electrical synapse strength, in this context, broaden the set of input differences for which spike trains are non-independent, seen when comparing sets of panels within Fig. 6. (% ψ < 0.8 increases from 32 to 46%), Fig. 6C (29 to 35%), and Fig. 6D (21 to 33%).
Finally, we evaluated the net gain in independence and separation created by the presence and strength of electrical synapses within the circuit over all values of G elec and G GABA used in our simulations. Here, we use radius of each concentric circle to represent a normalized combination of input strength and timing differences (Fig. 7). Identical inputs are thus at the center of each subpanel. In Fig. 8, each subpanel shows the difference in spike train independence or separation between that set of simulations (each subpanel represents a fixed pair of G GABA and G elec ) and the baseline set of simulations, in which electrical synapses are absent. Finally, we separated gains in values, shown in the upper halves of each circle, and decreases in values, which are shown in the lower half.
From this figure, we see that as electrical synapse strength is increased (left to right), the gains in independence (Fig. 8A) and separation (Fig. 8B) from baseline of spike trains in the two TC cells can both increase or decrease (green or blue). Overall, we observe that there is an interaction between the strength of electrical and inhibitory synapses, where for each increase in G GABA , a larger value of G elec is necessary to influence the ultimate  2 rate, plotted against all values of G elec and G GABA . Input strengths to TC 2 were varied between panels from left to right (G in2 = 0.04, 0.06 and 0.08 mS/cm 2 ; input to TC 1 was always G in1 = 0.06 mS/cm 2 ), and input timing was varied from bottom to top (t in2 = 60, 80 and 100 ms; t in1 was always 60 ms).
SCientifiC REPORTS | (2018) 8:7763 | DOI:10.1038/s41598-018-25956-x independence or separation of TC spike trains. This effect reflects the circuitry, in which the effects of the electrical synapse are imparted through the actions of the inhibitory synapses. For small values of G GABA , increases in TC independence dominate the results, while for larger values of G GABA , decreases become the more dominant effect. As above, effects (increase or decrease) are larger for larger radius, such that inputs with more differences are modified by the electrical synapses.
Changes in separation follow a similar trend. For each value of G GABA, as electrical synapse strength increases within the circuit, gains in separation become larger, especially for more-different inputs (larger radii). Thus the presence or increase in the electrical synapse acts to separate TC spike trains, and with that the possibility for the cortex to discriminate between them.

Discussion
Noting that most of the experimental demonstrations or computational simulations of circuits containing electrical synapses focus on the relationship between electrical synapses and synchrony, we set out to explore the impact of electrical synapses on transient spike train processing. Using a minimal 4-cell model of paired thalamic relay and thalamic reticular nucleus cells with a single electrical synapse to connect the pairs, here we have shown that through feedback inhibition, electrical synapses of the TRN influence both timing and rate of thalamic spiking. Ultimately, the electrical synapses of the TRN modulate both the temporal independence and separation of the spike trains that thalamic cells send on to cortex, thus impacting whether cortical cells receive spike trains that can be discriminated as arising from separate sensory receptors or receptive fields of the sensory surround.
The results of our simulations are important in the context of plasticity of electrical synapses in the TRN 23,24 . Our results show that changes in strength of electrical synapses can shift the character of a relay response, from one that separates its inputs to one that fuses inputs, for instance. Changes in the strength of electrical synapses are represented here by a shift along an axis, and show that even smaller changes in electrical synapse strength, such as the ~15% changes measured as a result of synchronized burst firing in TRN, have the potential to determine output rate, timing, independence and separation.
Changes in separation, latency and rate due to embedded electrical synapses within the thalamic relay circuitry might also increase efficient coding, by modulating the sparsity of sensory relay. Decorrelation in lateral geniculate nucleus 25 and retinal ganglion neuronal activity 26 in response to naturalistic stimulation, while additional correlation for white noise stimuli is achieved, is believed to improve sparse coding in the visual system and is predicted to be driven by the presence of inhibitory interneurons 27,28 . Additionally, it is worth noting that sparseness is not only formed spatially (a few active neurons in a population), but also temporally (involving restricted amount of spiking of active neurons) 29 . Modulation of separation and rate may affect not only sparsity in the relay system, but also on cortical neurons (or any higher order neuronal structures) via postsynaptic summation effects.
Our model is the simplest core unit, or motif, of intrathalamic connectivity, and our model assumed these four cells to be identical in term of intrinsic properties. Our motif of two coupled TC-TRN pairs further shows that electrical synapse not only affects thalamocortical rates and latencies, thus contributing to basic coding of spatial-temporal sensory inputs (for example, whisker inputs 4,5 ), but also regulates temporal independence and separation. However, Figure 5. Electrical synapses between TRN cells increase separation of TC spike trains. Each panel (A-I) is a heat map for separation between TC spike trains, plotted against all values of G elec and G GABA . Input strengths to TC 2 were varied between panels from left to right (G in2 = 0.04, 0.06 and 0.08 mS/cm 2 ; input to TC 1 was always G in1 = 0.06 mS/cm 2 ), and input timing was varied from bottom to top (t in2 = 60, 80 and 100 ms; t in1 was always 60 ms). Whitespace in separation means there is no spiking in either TC cell.   Fig. 6A, with the black horizontal line representing inputs arriving simultaneously (t in1 = t in2 ) and the black vertical line representing inputs of equal strength (G in1 = G in2 ). Colors represent the 5 different strata of normalized combinations used to construct Fig. 8, with the darkest blue being the least different and yellow being the most different. Figure 8. Change in spiking independence and separation between the two TC cells relative to uncoupled baseline (Gelec = 0). Each circle represents the difference of TC spike train independence or similarity, computed between results of simulations with G elec > 0 (one set per panel) and the uncoupled case (G elec = 0). From left to right, electrical synapse conductance increases across panels (G elec = 0.001 to 0.025 mS/cm 2 ). From top to bottom, inhibitory synapse conductance increases (G GABA = 0.005 to 0.050 mS/cm 2 ). Within each circle, radius represents input timing and strength similarity in 5 different strata, with the inner arcs representing the inputs to the two TC cells that were most similar in timing and strength, and outer arcs representing inputs that were most different (Fig. 8) connections within thalamus and between thalamus and cortex are more substantially complex in vivo. Thalamic cells receive convergent inhibitory inputs from multiple TRN cells, and TRN cells receive input from multiple thalamic relay cells. While we have focused on straightforward thalamic relay of singular inputs, mimicking POm, other thalamic subsectors receive inputs from broad areas of the sensory surround. Cortical feedback is also not represented in the present model, and neuromodulation is also unaccounted for. Thus, there is much future work to be done to thoroughly explore the role of electrical synapses in transient signal processing within thalamus.
The underlying circuit -inhibitory neurons connected by an electrical synapse, and providing feedback inhibition to the principal neurons that excite them -is one that we expect may be embedded within retina, where AII amacrine cells regulate retinal ganglion cell spiking; and within cortex, where inhibitory neurons regulate principal cell firing, as well as in the cerebellum. We expect that the general principles seen here -that electrical synapses act through inhibitory synapses to increase latency, decrease spike rates, and modify the independence and separation of spike trains in principal cells -will also impact information processing in the many areas that contain electrical synapses.
Our results make specific predictions that can be tested in experiments. For example, we predict that the spiking responses of nearby PoM neurons with overlapping receptive fields to peri-threshold inputs will be modulated by either altering or abolishing electrical synaptic strength in TRN by electrical 24 , pharmacological 30,31 or optogenetic means. We expect to see, especially in cases of separated input arrival timings that potentiated electrical coupling leads to decreased spiking and increased separation, while depressed or blocked electrical coupling leads to increased spiking and decreased separation.