Altered hippocampal interneuron activity precedes ictal onset

Although failure of GABAergic inhibition is a commonly hypothesized mechanism underlying seizure disorders, the series of events that precipitate a rapid shift from healthy to ictal activity remain unclear. Furthermore, the diversity of inhibitory interneuron populations poses a challenge for understanding local circuit interactions during seizure initiation. Using a combined optogenetic and electrophysiological approach, we examined the activity of identified mouse hippocampal interneuron classes during chemoconvulsant seizure induction in vivo. Surprisingly, synaptic inhibition from parvalbumin- (PV) and somatostatin-expressing (SST) interneurons remained intact throughout the preictal period and early ictal phase. However, these two sources of inhibition exhibited cell-type-specific differences in their preictal firing patterns and sensitivity to input. Our findings suggest that the onset of ictal activity is not associated with loss of firing by these interneurons or a failure of synaptic inhibition but is instead linked with disruptions of the respective roles these interneurons play in the hippocampal circuit.


Summary 23
Although failure of GABAergic inhibition is a commonly hypothesized mechanism underlying 24 seizure disorders, the series of events that precipitate a rapid shift from healthy to ictal activity 25 remain unclear. Furthermore, the diversity of inhibitory interneuron populations poses a 26 challenge for understanding local circuit interactions during seizure initiation. Using a combined 27 optogenetic and electrophysiological approach, we examined the activity of two identified 28 hippocampal interneuron classes during seizure induction in vivo. We identified cell type-29 specific differences in preictal firing patterns and input sensitivity of parvalbumin-and 30 somatostatin-expressing interneurons. Surprisingly, the impact of both sources of inhibition 31 remained intact throughout the preictal period and into the early ictal phase. Our findings 32 suggest that the onset of ictal activity is not due to a failure of inhibition, but is instead 33 associated with a decoupling of inhibitory cells from their normal relationship with the local 34 hippocampal network. 35

Introduction
Here we used optogenetic tools to identify, track, and probe two distinct populations of 87 hippocampal interneurons, the PV-and SOM-expressing cells, in two models of acute seizure 88 initiation in vivo. We find cell type-specific differences in the preictal activity of PV and SOM 89 cells and in the evolution of their sensitivity to input. However, the inhibitory influence of 90 interneuron firing on nearby neurons remains largely intact throughout the preictal and early ictal 91 periods, suggesting that seizure does not arise from a failure of GABAergic inhibition.  In an initial series of experiments, we assessed the spontaneous activity of these three 113 cell classes during a baseline period and four preictal periods of equal duration leading into 114 PTZ-induced seizure ( Figure 1D). PV, SOM and RS cells exhibited increased firing rates 115 following PTZ administration, but showed markedly different firing rate trajectories ( Figure 1E). 116 Strikingly, we found that most PV cells strongly increased their firing rate in the last preictal 117 period as compared to the first (94.4%, p<0.001, Binomial test) or third (83.3%, p<0.001; Figure  118 S3). In contrast, this progressive, late increase in firing rate was not observed in SOM or RS 119 cells ( Figures 1E and S3). Increased PV cell firing was independent of the latency to ictal onset 120 ( Figure S4) and was observed in the absence of significant changes in spike waveform 121 amplitude over time ( Figure S5). We next explored whether preictal firing rate changes were 122 accompanied by changes in the temporal spike pattern. Immediately preceding ictal onset, PV, 123 but not SOM or RS, cell firing became significantly more regular (i.e., less bursty) (Figure 2A). In 124 addition, PV cells showed an increased tendency to fire spikes separated by short (<10ms) 125 inter-spike-intervals (ISI; Figure S6). Unidentified cells with narrow spikes did not exhibit 126 changes in firing rate, firing regularity, or ISI statistics ( Figure S7). 127 In a separate series of experiments, we observed a similar increase in PV cell firing 128 rates during the late preictal period preceding Pilocarpine-induced seizures ( Figure S8), 129 suggesting that this is not a unique feature of PTZ-induced seizures but rather may be a general 130 feature of preictal activity in CA1. PV cells also showed increased regularity and shorter ISIs 131 during preictal periods preceding Pilocarpine-induced seizures, in the absence of changes in 132 spike waveform amplitude ( Figure S8D). 133 The progressive changes in PV interneuron firing rate and temporal pattern suggest that 134 the relationship between these cells and the surrounding local network may be altered prior to 135 seizure initiation. We therefore computed the mean spike field coherence (SFC) for PV, SOM 136 and RS cells during baseline activity and across the four preictal periods ( Figure 2B). We Together, these findings highlight cell type-specific changes in PV interneuron activity 146 leading up to the onset of ictal activity. To examine whether preictal changes in interneuron 147 output were accompanied by changes in sensitivity to input, we tested the responses of PV and 148 SOM cells to optogenetic stimulation during each preictal period. We measured the probability 149 of interneuron spiking in response to light pulses of varying intensity ( Figure 3A). PV cells 150 showed no progressive change in the slope or maximal response (Rmax) of the input-output 151 function ( Figure 3B-C). In contrast, SOM cells showed a significant increase in slope and a 152 significant decrease in Rmax across the preictal periods, suggesting a progressive preictal 153 alteration in their sensitivity to inputs. 154 To assay whether the observed changes in interneuron activity were associated with 155 altered inhibition of their targets, we measured the impact of ChR2-evoked interneuron spiking 156 on the firing rate of nearby RS cells. During baseline activity, we found that the firing rate of RS 157 cells decreased following ChR2-evoked PV and SOM cell spiking ( Figure 4A-B). We compared 158 the impact of ChR2-evoked inhibition during the four preictal periods and an additional period 159 immediately following ictal onset. RS firing suppression was not significantly changed across 160 the preictal and early ictal periods as compared to baseline when the PV cells were driven at 161 either moderate or high light intensities (see Methods; Figure   Thus, both PV-and SOM-mediated inhibition appear largely intact preceding ictal onset and 164 remain so during the transition to ictal activity. 165 At the ictal transition, changes in interneuron activity mainly appeared to be at the level 166 of firing rates, rather than postsynaptic impact on nearby RS cells. We next examined whether 167 interneuron firing remained elevated during the ictal period. Rigorous spike waveform 168 identification after ictal onset is highly challenging. However, we were able to track a subset of 169 recorded neurons through an initial 60-second ictal period. We found that after ictal onset, PV 170 and SOM firing decreased from the preictal peak back to baseline levels ( Figure 5A ChR2-evoked inhibition on local RS cells. Overall, these results suggest that seizure did not 185 arise from either an overall failure of GABAergic inhibition or from unconstrained excitation, but 186 rather from a decoupling of excitatory and inhibitory activity. We examined interneuron activity in two models of acute induction of status epilepticus, 235 which may provide insight into the mechanisms by which normal, healthy neural circuits 236 transition to pathological patterns of activation. We used both PTZ, thought to be a competitive 237 antagonist of the GABA A receptor (Huang et al., 2001), and Pilocarpine, a nonselective 238 muscarinic acetylcholine receptor agonist (Turski et al., 1989). Despite distinct pharmacological 239 mechanisms, we found similar trajectories for PV interneuron activity in both models, suggesting 240 that preictal increases in PV activity may be a common element of acute seizure initiation.  Signals were digitized and recorded with a DigitalLynx 4SX system (Neuralynx, Bozeman MT). 360 All data were sampled at 40kHz and recordings were referenced to the cerebellum. LFP data 361 were recorded with a bandpass 0.1-9000Hz filter and single-unit data was bandpass filtered 362 between 600-9000Hz. 363 364

Optogenetic manipulations 365
Light activation of ChR2-expressing cells was performed using a 473nm laser (OptoEngine LLC, 366 Midvale UT). A 200µm fiber was positioned on the cortical surface next to the electrode array 367 and lowered slowly into the cortical tissue directly above dorsal hippocampal CA1. To avoid 368 heating of the brain, we calibrated the light power (<75mW/mm2) during ChR2 unit tracking 369 experiments in order to ensure a mean spike probability of ~1 spikes per 5 ms light pulse in the 370 targeted population. Real-time output power for each laser was monitored using a photodiode 371 and recorded continuously during the experiment. During baseline periods, we identified ChR2-372 expressing interneurons using short (5 ms) pulses of blue light, relying on the short latency of 373 ChR2-evoked spikes and the high degree of temporal precision of the evoked spikes. In a 374 subset of experiments (Fig 3-4), we measured the input-output function of ChR2-identified 375 interneurons in response to a calibrated range of light intensities. 376 377

Spike sorting 378
Spikes were clustered semi-automatically using the following procedure. We first used the 379

Analysis of waveform parameters 396
For each isolated single unit, we computed an average spike waveform for all channels of a 397 tetrode. The waveforms were manually inspected and the channel with the largest peak-to-398 trough amplitude was used to measure the peak-to-trough duration values (Fig. 1 Supplement 399 1) as well as mean spike amplitude (Fig. 1 Supplement 5). We also computed the repolarization 400 value of the normalized (between -1 and +1) waveforms at 0.9 ms (similar to Vinck et al., 2015). corresponded to the LFP trace crossing an absolute z-score value >5 as compared to baseline. 408 Sustained, elevated z-scores were generally observed after ictal onset, and ictal onset was 409 typically coincident with the first ictal spike. We used spectrograms to validate that there were 410 no consistent LFP changes prior to ictal onset (Fig 1A). Spectrograms of LFP power around ictal 411 onset were computed using a wavelet transform with 7 cycles for each frequency and a Hanning 412 taper. LFP power was normalized by dividing by the summed power across the entire trace and 413 taking the base-10 logarithm. 414 415

Definition of analysis periods 416
Seizure latency varied across mice ( Figure S4). For each experiment, the preictal period from 417 injection to ictal onset was therefore divided into four equal periods. To characterize the 418 progressive changes in hippocampal CA1 preictal activity (Figs 1-2), we computed the change 419 in firing activity parameters as compared to baseline for the four preictal periods. For the 420 analysis of spontaneous activity, we only used baseline and preictal periods that did not contain 421 epochs of laser pulses and selected cells with baseline firing rates greater than 0.1Hz. For the 422 analyses of evoked activity in Figures 3-4, all cells were used. For the analysis of RS 423 suppression by ChR2-evoked inhibition, we defined an additional early ictal period as the 60s 424 following ictal onset. For analysis of ictal spiking, we used only the subset of interneurons 425 whose ictal spike activity could be resolved and we divided the 60s ictal period into four periods 426 of 15s each. 427 428

Firing rate and bursting 429
The mean firing rate per analysis period was computed as the number of spikes in that period 430 divided by the duration of that period in seconds. Changes in the temporal patterning of preictal 431 firing were detected using two metrics. First, we quantified the propensity to engage in irregular 432 burst firing using the coefficient of local variation (LV ; Fig 2A), which has been shown to be 433 robust against non-stationarities in firing rates. LV values greater than 1 indicate irregular firing, 434 whereas LV values smaller than 1 indicate sub-Poisson regular firing (Shimokawa and 435 Shinomoto, 2009) . Second, we computed the log fraction of ISIs between 2 and 10 ms over the 436 fraction of ISIs between 10 and 100 ms, i.e., Log(ISI short /ISI long ), as in Vinck et al. (2015). Here, p(I) is the fitted probability of a spike in the 2-15ms following laser pulse onset, I is the 476 laser intensity, S is a scaling factor, c is the c50, and 1/A is the slope. The Rmax was defined as 477 the value of p(I) at the maximum laser intensity tested. We fitted these curves by minimizing the 478 absolute deviation between fit and data (i.e., the L1 norm) using MATLAB's fminsearch function. We computed this modulation separately for pulses of medium and high intensity (Fig 4C-D). 493 The medium intensity level was defined as the level at which the simultaneously driven PV cells 494 had, on average, a 50% firing probability. The highest intensity level was the level at which the 495 simultaneously recorded PV cell spiking reached its maximum spike probability.