Giant Depolarizing Potentials Trigger Transient Changes in the Intracellular Cl- Concentration in CA3 Pyramidal Neurons of the Immature Mouse Hippocampus

Giant depolarizing potentials (GDPs) represent a typical spontaneous activity pattern in the immature hippocampus. GDPs are mediated by GABAergic and glutamatergic synaptic inputs and their initiation requires an excitatory GABAergic action, which is typical for immature neurons due to their elevated intracellular Cl- concentration ([Cl-]i). Because GABAA receptors are ligand-gated Cl- channels, activation of these receptors can potentially influence [Cl-]i. However, whether the GABAergic activity during GDPs influences [Cl-]i is unclear. To address this question we performed whole-cell and gramicidin-perforated patch-clamp recordings from visually identified CA3 pyramidal neurons in immature hippocampal slices of mice at postnatal days 4–7. These experiments revealed that the [Cl-]i of CA3 neurons displays a considerable heterogeneity, ranging from 13 to 70 mM (average 38.1 ± 3.2 mM, n = 36). In accordance with this diverse [Cl-]i, GDPs induced either Cl--effluxes or Cl--influxes. In high [Cl-]i neurons with a negative Cl--driving force (DFCl) the [Cl-]i decreased after a GDP by 12.4 ± 3.4 mM (n = 10), while in low [Cl-]i neurons with a positive DFCl [Cl-]i increased by 4.4 ± 0.9 mM (n = 6). Inhibition of GDP activity by application of the AMPA receptor antagonist CNQX led to a [Cl-]i decrease to 24.7 ± 2.9 mM (n = 8). We conclude from these results, that Cl--fluxes via GABAA receptors during GDPs induced substantial [Cl-]i changes and that this activity-dependent ionic plasticity in neuronal [Cl-]i contributes to the functional consequences of GABAergic responses, emphasizing the concept that [Cl-]i is a state- and compartment-dependent parameter of individual cells.


INTRODUCTION
Spontaneous neuronal activity transients are a hallmark of developing neuronal systems and play an essential role for several developmental processes like neuronal migration, myelinization, cortical regionalization, or the establishment of neuronal connectivity (for review Spitzer, 2006;Ben-Ari and Spitzer, 2010;Griguoli and Cherubini, 2017;Kirischuk et al., 2017). Such spontaneous activity transients can be generated in the sensory periphery, but are also intrinsic to developing neuronal networks (for review Ben Ari, 2001;Khazipov and Luhmann, 2006;Blankenship and Feller, 2010;Kilb et al., 2011). The fact that even synaptic networks of cultured neurons develop correlated activity transients (Rolston et al., 2007;Sun et al., 2010) suggests that the formation of circuits capable of mediating recurrent activity is probably an innate feature of the neuronal nature. One striking example for spontaneous, repetitive activity transients are hippocampal giant depolarizing potentials (GDPs) (Ben-Ari et al., 1989). Such GPDs have also been observed in the immature neocortex (Allene et al., 2008), thalamus (Pangratz-Fuehrer et al., 2007), and the embryonic spinal cord (Czarnecki et al., 2014).
In accordance with these observations, GDPs are characterized by an initial phase of depolarizing GABAergic post-synaptic currents (PSCs) and subsequent AMPA mediated PSCs which synergistically drive the depolarization (Khalilov et al., 2015). When the depolarization surpasses the GABA equilibrium potential (E GABA ) the GABAergic currents become outwardly directed, promoting an inhibitory action of GABA that limits the amount of depolarization/activation and thus the synchronization between pyramidal neurons (Khalilov et al., 2015). Finally, activation of Ca 2+ -dependent K + channels and post-synaptic GABA B receptors decreases the excitability and the firing frequency of pyramidal neurons (Sipila et al., 2006a;Khalilov et al., 2017), which reduce AMPA and GABA post-synaptic potentials (PSPs) and thus terminates GDPs.
Since GABA A receptors are ligand-gated anion-channels with a high permeability for Cl − ions (Farrant and Kaila, 2007), Cl − -fluxes through activated GABA A receptors can influence [Cl − ] i on a shorter time scale (Kaila et al., 1989;Bracci et al., 2001;Isomura et al., 2003;Jedlicka and Backus, 2006;Jedlicka et al., 2011;Lillis et al., 2012;Sato et al., 2017). The resulting activity-dependent [Cl − ] i increase upon massive GABAergic stimulation, in combination with the HCO 3 − permeability of GABA A receptors, leads to a shift from hyperpolarizing/inhibitory to depolarizing/excitatory GABAergic action (Staley et al., 1995;Sun et al., 2001;Isomura et al., 2003, but see Kaila et al., 1997 for more complex events involved in this hyperpolarizing-depolarizing shift). This process has been termed ionic plasticity (Blaesse et al., 2009;Raimondo et al., 2012;Kaila et al., 2014a). The relation between Cl − influx, dendritic volume/morphology and the capacity of Cl − extrusion systems determines the size of such activity-dependent Cl − transients (Staley and Proctor, 1999;Wright et al., 2011;Mohapatra et al., 2016). In the immature system, with high [Cl − ] i and depolarizing GABAergic events, the GABA-mediated Cl − efflux causes a transient decline in [Cl − ] i that temporarily attenuates the amplitude of subsequent GABAergic responses, thereby reducing or omitting possible excitatory effects (Achilles et al., 2007;Gonzalez-Islas et al., 2010;Kolbaev et al., 2011b). However, it has remained unclear whether GDPs lead to detectable [Cl − ] i changes.
In order to elucidate whether the massive GABAergic activity during a GDP influences [Cl − ] i in the immature hippocampus, we performed whole-cell and gramicidin-perforated patch-clamp recordings from visually identified CA3 pyramidal neurons in hippocampal slices from mice at postnatal day 4-7. These experiments revealed that the [Cl − ] i in pyramidal neurons is altered following a GDP.  (Wright et al., 2011).

Slice Preparation
All experiments were conducted in accordance with EU directive 86/609/EEC for the use of animals in research and the NIH Guide for the Care and Use of Laboratory Animals, and were approved by the local ethical committee (Landesuntersuchungsanstalt RLP, Koblenz, Germany). All efforts were made to minimize the number of animals and their suffering. Time pregnant C57Bl/6 mice were obtained from Janvier Labs (Saint Berthevin, France) and housed in the local animal facility. Newborn pups of postnatal days [P] 4-7 were deeply anesthetized with enflurane (Ethrane, Abbot Laboratories, Wiesbaden, Germany). After decapitation, the brains were quickly removed and immersed for 2-3 min in icecold standard artificial cerebrospinal fluid (ACSF, composition see below). Horizontal slices (400 µm thickness) including the hippocampus were cut on a vibratome (Microm HM 650 V, Thermo Fischer Scientific, Schwerte, Germany). The slices were stored in an incubation chamber filled with oxygenated ACSF at room temperature before they were transferred to the recording chamber.

Data Acquisition and Analysis
Whole-cell and gramicidin-perforated patch-clamp recordings were performed as described previously (Kyrozis and Reichling, 1995;Kolbaev et al., 2011b) at 31 ± 1 • C in a submergedtype recording chamber attached to the fixed stage of a microscope (BX51 WI, Olympus). Pyramidal neurons in the stratum pyramidale of the CA3 region were identified by their location and morphological appearance in infrared differential interference contrast image. Patch-pipettes (5-12 M ) were pulled from borosilicate glass capillaries (2.0 mm outside, 1.16 mm inside diameter, Science Products, Hofheim, Germany) on a vertical puller (PP-830, Narishige) and filled with the pipette solutions (composition see above).
Signals were recorded with a discontinuous voltageclamp/current-clamp amplifier (SEC05L, NPI, Tamm, Germany), low-pass filtered at 3 kHz and stored and analyzed using an ITC-1600 AD/DA board (HEKA) and TIDA software. All voltages were corrected post hoc for liquid junction potentials of −9 mV for 10 mM [Cl − ] p , −6 mV for 50 mM [Cl − ] p , and −3 mV for the perforated-patch solution (Achilles et al., 2007). Input resistance and capacitance were determined from a series of hyperpolarizing current steps. Action potential amplitude was calculated from the threshold (as determined by eye) and action potential duration was measured at half-maximal amplitude. Spontaneous post-synaptic currents (sPSCs) were detected and analyzed according to their amplitude and shape by appropriate settings using Minianalysis Software (Synaptosoft, Fort Lee, NJ, United States). Charge transfer is the total amount of charges that flow during an event. For PSCs it was determined in Minianalysis by integration of the currents between calculated onset and termination time points. Charge transfer of GDPs was determined in TIDA by integration of the current deflection from the holding current between the starting and endpoint of a GDP, as defined by eye.
The GABA reversal potential (E GABA ) was determined from the GABAergic currents induced by focal pressure application of 100 µM GABA via a micropipette (tip diameter ca. 1-2 µm, placed 50-100 µm from the soma in the stratum radiatum, puff duration 5-10 ms) during a voltage ramp protocol (from −3 to −63 mV; Figures 3A,B). For this purpose the current of a control voltage ramp was subtracted from the current of the voltage ramp delivered in the presence of GABA. The voltage ramp was applied during a quasi-stationary phase of the GABAergic response ( Figure 3B). The E m value at which this differential current reverses was considered as E GABA (Figures 3C-E). Because for the central aim of this study it was not possible to pharmacologically block voltage-dependent Na + currents with TTX, the voltage ramp protocol was preceded by a 100 ms long depolarizing phase at −3 mV to inactivate voltage-dependent Na + currents ( Figure 3B).
In order to take the contribution of HCO 3 − ions to the reversal potential of GABA A receptors into account (Farrant and Kaila, 2007), we calculated [Cl − ] i from E GABA with the Goldman-Hodgkin-Katz equation:  (Ruusuvuori et al., 2010), a Henry coefficient of 0.318 and a pKs of 6.128 (Mitchell et al., 1965). A relative HCO 3 − permeability (P HCO ) of 0.44, which has been determined for GABA A receptors in hippocampal neurons (Fatima-Shad and Barry, 1993), was used and P Cl was defined as 1. The driving-force of Cl − (DF Cl ) was calculated from the difference between the average E m during a GDP and E Cl (DF Cl = E m -E GABA ). To calculate GABAergic (g GABA ) and glutamatergic conductances (g Glu ) the peak amplitudes of GDP associated currents (I GDP ) were divided by the estimated driving force at the given holding potentials using Ohm's law [for GABA: g GABA = I GDP /(E m − E Rev GABA ); for glutamate: All values are given as mean ± SEM. If not explicitly noted, Student's t-test was used for statistical analysis (Systat 11). Significance was assigned at p-levels of 0.05 ( * ), 0.01 ( * * ), and 0.001 ( * * * ).

Microcontroller-Based GDP Detection
To enable the online determination of [Cl − ] i at distinct latencies after a GDP we used a microcontroller (Arduino Uno 1 ) connected to the inputs and outputs of the NPI amplifier. A threshold crossing algorithm comparing a floating average of 50 E m datapoints (to avoid triggering by single action potential) with a manually preset threshold potential was used to detect a GDP ( Figure 3F). The time point when E m falls below the threshold was defined as end of a GDP. After a defined interval (0.1-20 s) the microcontroller provided a signal that switches the amplifier to voltage-clamp mode and delivers the ramp protocol used for the determination of E GABA (Figures 3A-E). The used program code for the Arduino microcontroller is available at https://forum.arduino.cc/index.php?topic=564489.0.

Compartmental Modeling
For morphological reconstruction some CA3 pyramidal cells were filled with biocytin (0.5-1%, Sigma-Aldrich) under whole-cell conditions as described in detail before (Horikawa and Armstrong, 1988;Schröder and Luhmann, 1997). Reconstruction and morphological analysis of the biocytin-labeled neurons were performed from 60x oilimmersion images using Fiji 2 . A reconstructed CA3 pyramidal cell was imported into the NEURON simulation program 3 (Figures 5A,B). The following passive parameters were used: R a (specific axial resistance) = 34.5 cm; R m (specific membrane resistance) = 2 k cm 2 ; C m (specific membrane capacitance) = 1 µFcm −2 .
GABA A synapses were simulated as a post-synaptic parallel Cl − and HCO 3 − conductance with exponential rise and exponential decay (Jedlicka et al., 2011): where P is a fractional ionic conductance that was used to split the GABA A conductance (g GABA ) into Cl − and HCO 3 − conductance. E Cl and E HCO3 were calculated from Nernst equation. The GABA A conductance was modeled using a two-term exponential function, using separate values of rise time (0.5 ms) and decay time (80 ms) (Santhakumar et al., 2005). Parameters used in our simulations were as follows: (Fatima-Shad and Barry, 1993).
For the experiments used to model the impact of access resistance (R s ) or dendritic filtering on the determination of the reversal potential we considered static [Cl − ] i and and implemented a single-electrode voltage-clamp process to the soma, using R s values of 0.5, 5, 10, 20, and 40 M . A GABA synapse with a decay time of 100 ms and a peak conductance of 10 nS was used to emulate the GABA application protocol.
For the modeling of the GDP-induced [Cl − ] i and [HCO 3 − ] i changes we calculated ion diffusion and uptake by standard compartmental diffusion modeling (De Schutter and Smolen, 1998;De Schutter, 2010;Mohapatra et al., 2014Mohapatra et al., , 2016. To 2 www.fiji.sc 3 neuron.yale.edu simulate intracellular Cl − and HCO 3 − dynamics, we adapted our previously published model (Jedlicka et al., 2011). Longitudinal Cl − and HCO 3 − diffusion along dendrites was modeled as the exchange of anions between adjacent compartments. For radial diffusion, the volume was discretized into a series of four concentric shells around a cylindrical core (De Schutter and Smolen, 1998) and Cl − or HCO 3 − was allowed to flow between adjacent shells (Hines and Carnevale, 2000). The free diffusion coefficient of Cl − inside neurons was set to 2 µm 2 /ms (Kuner and Augustine, 2000). Since the cytoplasmatic diffusion constant for HCO 3 − is to our knowledge unknown, we also used a value of 2 µm 2 /ms. To simulate transmembrane transport of Cl − and HCO 3 − , we implemented an exponential relaxation process for To simulate the GABAergic activity during a GDP, 534 GABA synapses with a peak conductance of 0.789 nS and a decay of 80 ms were randomly distributed in the dendritic compartment of the reconstructed neurons. This number of GABAergic inputs generates a charge transfer of 88 pA (at initial [Cl − ] i of 10 mM and under VC conditions at 0 mV) similar to the charge transfer recorded experimentally under this condition (see section "Properties of GDPs"). To simulate AMPA synapses, additional 107 Exp2syn processes with a peak conductance of 0.509 nS, a reversal potential of 0 mV, a rise time of 0.1 ms, and a decay of 11 ms were randomly distributed. GABA and AMPA inputs were activated stochastically using a normal distribution that emulates the distribution of glutamatergic/GABAergic PSCs observed in the present study. We analyzed the mean [Cl − ] i and [HCO 3 − ] i of all dendrites to simulate the experimental procedure for E GABA determination, in which GABA was applied to the dendritic compartment.

Properties of the Recorded Cells
The average resting membrane potential (RMP) of the CA3 pyramidal cells under whole cell condition was −52.1 ± 0.8 mV (n = 116), their input resistance (R in ) at RMP was 0.9 ± 0.1 G and their membrane time constant amounted to 86 ± 8.3 ms, corresponding to a membrane capacitance of 152 ± 24.3 pF. Upon depolarization above a membrane potential (E m ) of −42.4 ± 0.68 mV, these cells were capable to fire action potentials with an amplitude of 54.9 ± 1.3 mV (n = 72) and a duration of 2.7 ± 0.4 ms. Comparable results were observed under gramicidin-perforated patch conditions. In these experiments the average resting membrane potential was −53.6 ± 0.24 mV (n = 24), the input resistance was 1.75 ± 0.13 G and the membrane time constant amounted to 105.9 ± 11.4 ms, corresponding to a membrane capacitance of 65 ± 9.1 pF. Upon E m depolarization above −46.2 ± 1.2 mV these cells fired action potentials with an amplitude of 36 ± 2 mV (n = 22) and a duration of 3.6 ± 0.4 ms.

Properties of GDPs
Giant depolarizing potentials were present in 96 of in total 111 cells (corresponding to 86.5%) from 73 slices/44 animals investigated with different recording conditions. In the first set of experiments we used a high Cl − pipette solution ([Cl − ] p = 50 mM) to characterize the properties of spontaneous network events. Under this condition in all 13 cells (from 13 slices/6 animals) spontaneous massive depolarizing events with an average amplitude of 24.9 ± 1.2 mV (n = 130 GDPs in 13 cells) and a duration of 1.24 ± 0.04 s were observed (Figures 1A,B). While these depolarizing events were subthreshold in three cells, they reliably evoked 6.4 ± 0.4 APs (n = 104 GDPs in 10 cells) in the remaining cells. Since the amplitude, duration and appearance of these depolarizing events mimic the properties of GDPs (Ben-Ari et al., 1989;Khalilov et al., 2015), we considered them as GDPs. These GDPs occurred at a rate of 1.3 ± 0.13 min −1 (n = 13 cells).
To investigate the contribution of synaptic events underlying these GDPs, we next performed voltage-clamp recordings using a [Cl − ] p of 10 mM. For this purpose GABAergic and glutamatergic synaptic events were isolated at 0 mV (close to the reversal potential of glutamatergic currents) and at −60 mV (close to the calculated reversal potential of GABAergic currents at a [Cl − ] p of 10 mM), respectively. At a holding potential of 0 mV in total 125 GDPs associated inward currents (I GDP ) were identified (n = 17 cells), occurring at a rate of 1.35 ± 0.15 min −1 (Figure 1C). These I GDP had an average amplitude of 309.5 ± 3.5 pA (n = 125), a duration of 0.91 ± 0.02 s and conveyed an average charge transfer of 88 ± 3.3 pC (Figure 1D). At a holding potential of −60 mV in total 88 I GDP from 15 cells could be recorded. They had an average amplitude of 79.3 ± 3.7 pA (n = 88), a duration of 0.7 ± 0.02 s, conveyed an average charge transfer of 14.6 ± 0.7 pC, and occurred at a rate of 1.3 ± 0.2 min −1 (Figures 1C,D). These peak currents correspond to conductances of 5.1 ± 0.1 nS for the GABAergic and 1.3 ± 0.06 nS for the glutamatergic GDP component.
In order to analyze the temporal relation between GABAergic and glutamatergic synaptic inputs during a GDP, we performed voltage-clamp recordings at −30 mV, which in theory would enable identification of glutamatergic inward and GABAergic outward currents. However, as from the raw current traces GABAergic outward currents and glutamatergic inward current could not be separated (Figure 1E), we had to identify the onset of GABAergic and glutamatergic events from the first derivative of the current trace ( Figure 1E). The cumulative probability distribution of those identified GABAergic (n = 4003 events) and glutamatergic events (n = 699) revealed that glutamatergic events start slightly delayed to GABAergic synaptic inputs and also terminated significantly earlier (p < 0.001, Kolmogorov-Smirnov test, 2351 events) (Figure 1F), although these events reflect only a fraction of all synaptic events during a GDP.

Relation Between GDPs and Spontaneous Synaptic Events
In order to estimate how much synaptic inputs underlie a GDP, we next analyzed the properties of spontaneous postsynaptic events (sPSCs) that occurred clearly outside of GDPs (Figure 2A). Voltage-clamp recordings using a [Cl − ] p of 10 mM revealed that at −60 mV (close to the calculated reversal potential of GABAergic currents) sPSCs had an average amplitude of 18.3 ± 1.3 pA (n = 17 cells), a rise time of 3.5 ± 0.2 ms, a decay time of 10.8 ± 1.5 ms and a charge transfer of 131.5 ± 14.4 fC. These putatively glutamatergic sPSCs occurred at a frequency of 5.8 ± 1.2 Hz (Figure 2B).
The average amplitudes of GABAergic and glutamatergic sPSCs correspond under these conditions to a unitary conductance of 0.78 and 0.3 nS, respectively. Comparison of charge transfer at holding potentials of 0 mV suggests that 101 GABAergic synaptic inputs mediate the same charge transfer as the GABAergic component of an average GDP. And the comparison of charge transfer at holding potentials of −60 mV suggests that 107 glutamatergic synaptic inputs mediate the same charge transfer as the glutamatergic component of an average GDP.

GDPs Induced Transient [Cl − ] i Alterations
As a variety of studies demonstrated that excessive GABAergic stimulation can alter [Cl − ] i (Kaila et al., 1989;Staley et al., 1995;Jedlicka and Backus, 2006;Achilles et al., 2007;Kolbaev et al., 2011b), we assumed that the Cl − -fluxes underlying the GABAergic currents during a GDP may be sufficient to induce significant alterations in [Cl − ] i . To address this question, we determined the [Cl − ] i shifts after a GDP had occurred. In order to maintain the [Cl − ] i undisturbed the experiments were performed under gramicidin-perforated patch-clamp conditions (Kyrozis and Reichling, 1995). These experiments demonstrated that the average [Cl − ] i of CA3 pyramidal neurons amounted to 38.1 ± 3.2 mM (n = 36), however with a considerable scatter between ca. 13 and 70 mM. Application of the NKCC1 inhibitor bumetanide (10 µM) significantly (p < 0.001) reduced the [Cl − ] i to 21.4 ± 2.7 mM (n = 9). In the presence of bumetanide a slow [Cl − ] i decline with a time constant of 280s occurred, indicating that passive Cl − -fluxes are rather small. In addition, In order to quantify these different responses, we next estimated the driving-force of Cl − (DF Cl ) during a GDP from the difference between the average E m during a GDP and E Cl . Regression analysis for the [Cl − ] i changes against the DF Cl revealed a significant (R 2 = 0.713, F = 49.74, p < 0.001) correlation between DF Cl and the [Cl − ] i changes after a GDP ( Figure 4C). Therefore we subdivided neurons in groups with positive DF Cl and negative DF Cl , respectively. In both groups a significant [Cl − ] i change occurred after a GDP (Figures 4D,E). In the neurons with positive DF Cl the maximal [Cl − ] i decrease in the first 2 s after a GDP amounted to 12.4 ± 3.4 mM (n = 10, p = 0.0057, Figure 4D), while GDPs induced in the neurons with negative DF Cl a relatively long lasting [Cl − ] i increase (Figure 4E), which 2 s after a GDP amounted to 4.4 ± 0.9 mM (n = 6, p = 0.0039). Further analysis suggested that neither the initial [Cl − ] i , nor the distribution of neurons showing Cl − influx/efflux or the amount of GPD-induced [Cl − ] i changes depend on the age of the animals (data not shown).

Inhibition of GDP Activity Reduces [Cl − ] i
Since these results demonstrated that GDPs lead to long lasting [Cl − ] i changes, which in particular in case of [Cl − ] i increases showed no evident back regulation in the first 10 s, we postulated that ongoing GDP activity can contribute to the initial [Cl − ] i of immature hippocampal CA3 pyramidal neurons. Indeed, after inhibition of GDP activity for ≥30 min by bath application of 10 µM CNQX (Bolea et al., 1999), the average [Cl − ] i amounted to 24.7 ± 2.9 mM (n = 8), which is significantly (p = 0.0035) smaller than the initial [Cl − ] i under control conditions (38.1 ± 3.2 mM, n = 36). In addition, the CV in the [Cl − ] i droped from 0.51 under control conditions to 0.33 in the presence of CNQX, indicating a more homogenous distribution of [Cl − ] i . In summary, these results indicate that ongoing GDP activity led in general to an increased [Cl − ] i , suggesting that the association of a high Cl − conductance and a depolarized E m during a GDP in combination with slow passive efflux rates of Cl − favors an accumulation of Cl − in these neurons. In order to evaluate whether the frequency of GDP activity influences [Cl − ] i of CA3 pyramidal neurons under control conditions, the average occurrence of GDPs was plotted against the initial [Cl − ] i in these cells ( Figure 4F). This analysis revealed that initial [Cl − ] i and the occurrence of GDPs are not correlated (R 2 = 0.036, F = 0.72, p = 0.407). This observation indicates that the [Cl − ] i of an individual cell cannot directly be related to the frequency of GDP activity in the slices. In addition, neither the amplitude of GDPs not their duration are significantly correlated with [Cl − ] i (data not shown).

Compartmental Modeling
The observation that R in was higher and the membrane capacitance was lower in the gramicidin-perforated patchclamp recordings suggests that the access resistances (R s ) was under perforated-patch conditions higher than in whole-cell recordings. To investigate to which extend such differences in R s can affect the determination of [Cl − ] i we employed compartmental modeling using a reconstructed immature CA3 pyramidal neuron (Figures 5A,B, see section "Materials and FIGURE 3 | Determination of E GABA using a voltage ramp protocol. (A) Voltage (black trace) and current (green trace) traces of a typical clamp protocol used to determine E GABA . Voltage ramps between -3 and -63 mV were given from a 100 ms long depolarizing phase to inactivate fast Na + currents. The second voltage ramp was delivered during the quasi-stationary phase of GABAergic response evoked by a short application of 100 µM GABA pulse (gray arrowhead). The dark blue trace in the bottom depicts the current response of the same cell without a voltage ramp protocol to illustrate the stationarity of the GABAergic current during the voltage ramp. (B) The ramp interval as shown in (A) at a higher temporal resolution. (C) Current traces from (B) recorded in the absence (red) and presence (blue) of GABA. (D) The difference between both current traces reverses at 13 ms. (E) Voltage traces recorded in the absence (red) and presence (blue) of GABA. Note that both traces are virtually identical. From these traces the reversal potential was determined. (F) Voltage and current traces of an experiment illustrating the procedures of a GDP triggered ramp protocol. From a continuous recording in current-clamp mode the microcontroller identified a GDP that crossed the threshold (E Thr ), the end of this GDP was determined as the time point when E m recovers to E Thr . After a defined latency ( t) the microcontroller provided a pulse to the amplifier to switch to VC mode and run the ramp protocol.
Methods" for details) . Therefore we estimated the influence of R s on the determination of E GABA using a voltage-ramp protocol. Implementation of a voltage ramp protocol via a simulated single-electrode process with defined R s values of 0.5, 5, 10, 20, and 40 M on a neuron with a [Cl − ] i of 30 mM led to E GABA values of −28.6, −28.7, −28.8, −28.9, and −29.0 mV, respectively (Figures 5B-D). This simulation suggests that E GABA is only marginally affected by R s .
Next we used the compartmental modeling to estimate to which extend space-clamp problems will affect the determination of [Cl − ] i by E GABA . These simulations revealed that the E m values during the voltage ramp are attenuated in distant dendrites (Figures 5E,F). Accordingly, E GABA, determined by a voltage ramp applied via a simulated single-electrode (Rs = 5 M ) was shifted in the distal dendrites (Figures 5G,H) Figure 5K). In summary, these results suggest that in addition to [Cl − ] i , also [HCO 3 − ] i shows a considerable ionic plasticity, leading to a decreased DF HCO3 .

DISCUSSION
The main aim of the present study was to reveal, whether GDPs lead to detectable [Cl − ] i changes. The main findings of this study can be summarized as follows: The [Cl − ] i in CA3 neurons recorded by perforated patch recordings revealed an obvious large scatter in the present study. From these values it could be assumed that both, inhibitory and excitatory GABAergic responses coexist in CA3 neurons at the same developmental stage, as has been reported in other brain structures (Chavas and Marty, 2003). However, this large scatter of [Cl − ] i is in line with a series of previous publications that reported similar distributions in the initial [Cl − ] i of immature hippocampal (Dzhala et al., 2012), but also neocortical neurons (Glykys et al., 2009;Sato et al., 2017). On the other hand, Valeeva et al. (2013) described considerably smaller [Cl − ] i using non-invasive electrophysiological measurements. In accordance with the established role of NKCC1 for the maintenance of a high [Cl − ] i in neurons (Rohrbough and Spitzer, 1996;Blaesse et al., 2009), bath application of the loop diuretic bumetanide significantly decreased [Cl − ] i and also reduced the scatter in [Cl − ] i . We could not exclude, that part of the rather large scatter in the initial [Cl − ] i observed in the present publication may to some extent be related to perforated-patch recordings from neurons that have been affected by the slicing procedure (Dzhala et al., 2012). In addition, our modeling studies demonstrate that [Cl − ] i determined from E GABA are critically affected by spaceclamp problems within the dendritic compartment. Thus we cannot exclude that a considerably part of the scatter in [Cl − ] i and in particular the high [Cl − ] i in some neurons are caused by insufficient voltage-clamp conditions at the site of GABAergic stimulation. However, while this obstruction influences the absolute [Cl − ] i values, it doesn't affect the main observation of this study, that GDPs induce considerable [Cl − ] i shifts. Because we used identical stimulation parameters before and after a GDP, the influence of space-clamp errors should be identical for both [Cl − ] i values. In contrast, R s seems to have only a marginal effect on the determination of E GABA , which is mainly caused by the fact that during a voltage ramp protocol I GABA , and thus the voltage-error caused by R s , is minimal at E GABA . In any way, the mean [Cl − ] i value of our recordings is in good agreement with the values expected to be recorded in neurons residing sufficiently deep below the slice surface to be unaffected by the slicing procedure (Dzhala et al., 2012) and the in-toto, unsliced hippocampal preparation shows a comparable wide distribution of [Cl − ] i between 5 and 50 mM (Dzhala et al., 2012). In addition, we considered the first [Cl − ] i value that was determined in a cell as initial [Cl − ] i . However, we cannot exclude that this cell experienced a GDP before the recording interval started (about 30-60 s before [Cl − ] i determination). Finally, our results revealed that inhibition of GDP activity reduced not only the initial [Cl − ] i , but also the variance in these values, indicating that probably also different frequency or different amounts of GABAergic inputs during GDPs may contribute to the scatter in From the observation that E GABA and [Cl − ] i was increased in superficial neurons of the traumatized slice surface, it has been suggested that part of the excitatory GABAergic action may represent a "slicing artifact" (Dzhala et al., 2012). However, GDPs have also been observed in whole-hippocampal preparations (Leinekugel et al., 1998), which excludes that they represent "slicing artifacts" caused by artificially increased [Cl − ] i in traumatized neurons (Valeeva et al., 2013). In fact, GDPs are also present under in vivo conditions (Khazipov et al., 2001;Leinekugel et al., 2002), suggesting that GABA A receptors may provide the excitatory drive required to trigger GDPs also in vivo. These observations in the immature hippocampus are in contrast to recent in vivo recordings from immature neocortical neurons, which revealed that GABA A receptor-mediated responses are indeed depolarizing, but most probably provide an inhibitory effect on immature neocortical neurons (Kirmse et al., 2015;Valeeva et al., 2016).
While the GABAergic conductance underlying GDPs determined in our study is in agreement with published values (Khalilov et al., 2015), the glutamatergic component is considerably higher than the published values. In addition, there is an obvious discrepancy between our conclusion that comparable numbers of GABAergic and glutamatergic synaptic inputs underlie a GDP (drawn from the comparison of charge transfer), while the detection of synaptic inputs from the first derivative of the current trace at −30 mV revealed much less identified glutamatergic synaptic events. However, the latter method detects clearly not all events, but, in particular during the phase of maximal inputs, only the most prominent ones. Therefore we assume that a considerable portion of GABAergic events and a larger fraction of the glutamatergic events remained undetected in this analysis. Nevertheless, we still consider that this analysis provides a good estimate for the temporal relation between GABAergic and glutamatergic inputs. Our findings are also in good agreement with previous studies demonstrating that GABAergic inputs precede the onset of glutamatergic inputs (Mohajerani and Cherubini, 2005;Khalilov et al., 2015).
The main observation of the present study is that GDPs induced considerable [Cl − ] i shifts. Such ionic plasticity in [Cl − ] i transients has been found after a variety of pathophysiological impacts (Lillis et al., 2012;Kaila et al., 2014b;Sato et al., 2017) and physiological processes (Bracci et al., 2001;Chub et al., 2006;Gonzalez-Islas et al., 2010;Kolbaev et al., 2011b). While in the present study the overall effect was partially masked by the variability in initial [Cl −  − ] i decrease will lead to a decreased DF HCO3 and thus affects E GABA . This observation supports the concept that also [HCO3 − ] i changes contribute to ionic plasticity after GABAergic stimulation (Staley and Proctor, 1999;Blaesse et al., 2009;Raimondo et al., 2012), in particular in the immature brain where the replenishment of [HCO 3 − ] i via carbonic anhydrase VII is reduced (Rivera et al., 2005).
The GDP-induced shifts in [Cl − ] i affect the DF Cl and thus the functional consequences of subsequent GABA A receptor activation. However, it is important to consider that the [Cl − ] i of the arbitrary neuron investigated by patch-clamp is most probably not decisive for the generation of GDP activity in a slice, but that GDPs are initiated in few hub neurons (Bonifazi et al., 2009;Wester and McBain, 2016). While the [Cl − ] i increase in low [Cl − ] i neurons attenuates the inhibitory effect of GABA in these neurons (Staley and Proctor, 1999;Raimondo et al., 2012), the [Cl − ] i decrease in high [Cl − ] i neurons reduced depolarizing GABAergic responses and their putative excitatory effect (Kolbaev et al., 2011a,b). Although the small GDP-induced [Cl − ] i decline in the whole population of recorded neurons suggests that these [Cl − ] i shifts may only marginally reduce the global excitatory effect of GABA on network excitability, the [Cl − ] i decline in the high [Cl − ] i neurons will clearly attenuate the excitatory potential of GABA in these cells. In addition, the GDP-induced [HCO 3 − ] i decline, which was suggested from our modeling studies, reduces the inwardly directed DF HCO3 and would thus additionally attenuate an excitatory effect of GABA.
As the initiation of GDPs (Sipila et al., 2005) as well as the positive feedback during the "onset" phase (Khalilov et al., 2015) depend on excitatory GABAergic inputs, we propose that these high [Cl − ] i neurons may be particularly relevant for the establishment of GDPs. Therefore we assume that the transient, GDP-mediated [Cl − ] i decline in CA3 pyramidal neurons with high [Cl − ] i will considerably attenuate the excitatory effect of GABA during a GDP. This effect adds to the dynamic model of Khalilov et al. (2015), which demonstrated that during the peak (or "catharsis") phase of the GDP DF Cl becomes positive, thus providing an inhibitory effect that is essential to limit excitatory influences.
Interestingly, it has been observed that interneurons of the mature hippocampus are much more susceptible to activitydependent shifts in E GABA than pyramidal neurons (Lamsa and Taira, 2003), suggesting that ionic plasticity during a GDP is more pronounced in these cells. Therefore the typically shunting depolarization provoked by GABA A receptors in hippocampal interneurons (Banke and McBain, 2006), might be shifted to excitatory effects during a GDP. And since a subpopulation of GABAergic interneurons serve as hub neurons that orchestrate GDP activity (Bonifazi et al., 2009;Picardo et al., 2011), the resulting decline in the GABAergic inhibitory drive on GABAergic interneurons will probably impact the frequency of GABAergic inputs to pyramidal neurons during GDPs.
In addition, the [Cl − ] i dynamics and the resulting decreased excitatory drive of GABAergic inputs may contribute to the termination phase of the GDP, in addition to other processes that have been shown to terminate GDPs (Sipila et al., 2006a;Khalilov et al., 2017). The calculation from Khalilov et al. (2015) proposed depolarizing GABAergic current during the repolarization phase, because their assumption of a constant [Cl − ] i resulted in stable E GABA values. Therefore DF GABA will become negative when E m falls below E GABA during the repolarization. The collapse of [Cl − ] i during a GDP observed in the present study may explain a reduced GABAergic depolarization/excitation during this phase, which will support the termination of a GDP.
In the developing spinal cord GABA-dependent spontaneous neuronal activity induces a collapse of the Cl − gradient, which temporarily attenuates the excitatory drive provided by GABA A receptors, and the subsequent slow Cl − accumulation that reestablishes GABAergic excitation underlies the low frequency of this recurrent activity (Chub and O'Donovan, 2001;Chub et al., 2006). Thus the dynamics of hippocampal [Cl − ] i homeostasis after a GDP may also determine the frequency of GDPs. The breakdown of [Cl − ] i in the high [Cl − ] i neurons and in hub neurons will reduce excitatory GABAergic responses (Kolbaev et al., 2011a), which not only contributes to the shutoff of GDPs, but also prevents the generation of new excitatory network events. With the [Cl − ] i re-accumulation via NKCC1, GABAergic responses regain their excitatory potential. At the time point when the net GABAergic effect becomes sufficiently excitatory again, it can contribute to the generation of the next GDP. We assume that neurons with a low [Cl − ] i do not contribute to the initiation of GDP activity, thus the GDP-induced [Cl − ] i -increase in these neurons probably doesn't play an essential role in the generation of GDPs.
The observation that the peak of [Cl − ] i changes occurred with several seconds delay to the end of the GDP most probably reflects the time required for a diffusional equilibration within the dendrites (Kuner and Augustine, 2000) and is in line with previously reported delays of 2-5 s between electrical stimulation and the peak of dendritic [Cl − ] i responses (Berglund et al., 2008, see also Jedlicka et al., 2011). While this rather slow propagation of [Cl − ] i transients limits the spatial extend of ionic plasticity, it doesn't exclude that GDPs induces in the vicinity of GABAergic synapses sufficiently high [Cl − ] i alterations to affect GABAergic transmission.
The KCC2-dependent back-regulation of [Cl − ] i after an activity dependent increase in mature neurons occurred within several seconds (Lamsa and Taira, 2003;Jin et al., 2005). However, as the Cl − -loader NKCC1 probably modulates a less effective transmembrane transport, [Cl − ] i back regulation upon activity-dependent Cl − loss in immature neurons seems to take longer time, even up to some minutes (Achilles et al., 2007). Thus the observed slow recovery of the GDP-induced Cl − -loss in high [Cl − ] i neurons is in agreement with the relatively slow time constant of NKCC1 mediated Cl − uptake. The recovery upon GDP-induced [Cl − ] i increases in the low [Cl − ] i group appeared to be even slower, indicating that it is mainly mediated by [Cl − ] i diffusion toward the soma and the ineffective passive transmembrane fluxes observed in immature hippocampal neurons. This observation was supported by the slow [Cl − ] i decline observed in the presence of the NKCC1 inhibitor bumetanide.
The fact that recovery rates after [Cl − ] i increases were slower than after [Cl − ] i losses also suggests that ongoing GDP activity may increase basal [Cl − ] i . Indeed, we observed that inhibition of GDP activity significantly reduces initial [Cl − ] i . This result demonstrates that the [Cl − ] i in immature neurons depends substantially on GABAergic activity due to the slow kinetics of activity dependent [Cl − ] i transients. However, GDPs may also effect [Cl − ] i homeostasis by post-translational modification of KCC2 and NKCC1 transporters (Blaesse et al., 2009;Kaila et al., 2014a).
In summary, our results indicate that the [Cl − ] i and thus the physiological effects of GABAergic inputs critically depend on prior GABAergic activity. Therefore, in immature hippocampal neurons ionic plasticity must be considered to explain [Cl − ] i and thus the functional state of GABAergic transmission at each single time point. These observations also foster the concept that neuronal [Cl − ] i has to be considered as a state-and compartment-dependent parameter of individual cells (Wright et al., 2011).

AUTHOR CONTRIBUTIONS
WK and HL designed this study. AL performed all recordings. AL and WK analyzed the data. WK, AL, and PJ performed computational modeling. AL, HL, PJ, and WK wrote the manuscript.

FUNDING
This study was supported by funding from the Deutsche Forschungsgemeinschaft (DFG) to WK and HL and a FTN stipend to AL. This manuscript is part of the Ph.D. thesis of AL.