Conditioning by Subthreshold Synaptic Input Changes the Intrinsic Firing Pattern of CA3 Hippocampal Neurons

Firing Pattern of CA3 Hippocampal Neurons Saray Soldado-Magraner13*, Federico Brandalise23, Suraj Honnuraiah1, Michael Pfeiffer1, Urs Gerber2, Rodney Douglas1 1Institute of Neuroinformatics, University of Zurich and ETH Zurich, Switzerland 2Brain Research Institute, University of Zurich, Switzerland 3These authors contributed equally to this work *Correspondence: ssaray@ini.uzh.ch, rjd@ini.uzh.ch Abstract 1

However, there are substantial reasons to doubt that firing patterns are static properties of neurons. The discharge 34 dynamics depends on the distribution and activations of the membrane conductances that it expresses (Hille, 2001; it is a generic property of CA3 cells. Using a conductance-based neuron model and the channel blockers XE991 and 5 Figure 1. Firing pattern transitions occur in CA3 neurons after subthreshold paired-pulse stimulation of afferents. A-C) Three examples of neurons in the CA3 area presenting different morphologies and different firing patterns in control conditions. The discharge patterns were measured by injection of step currents of increasing amplitude. Control measurements (grey traces, left) were followed by stimulation of the mossy fibers. The upper trace shows all voltage traces elicited upon different levels of current injection on that cell. Two sample traces of this set are shown below. EPSPs (middle panel) were evoked in response to a stimulation with double current pulses, separated by 20 ms and repeated 500 times at 1 Hz. The series of repeated pulses are shown superimposed. The median trace is highlighted in red. The inset shows the configuration of recording and stimulating electrodes (on the CA3 region of the hippocampus and on the dentate gyrus, respectively). Below, the morphology obtained by labelling the cells with biocytin is shown. After the conditioning, patterns were measured again (blue traces, right). A) Pyramidal cell switches from non-adapting burst to intrinsic burst firing. B) Pyramidal cell switches from delay accelerating to intrinsic burst continuous pattern. C) Bipolar cell switches from non-adapting continuous to adapting continuous firing (scale bars = 50µm). D) Mean fraction of spikes for the population in the first and second half of the voltage trace (see green and yellow rectangle below the trace in A for an example) for both control and conditioned cases. A significant redistribution on the fraction of spikes is observed after the conditioning, where the fraction of spikes on the first half is increased while it decreases in the second half (n=50, p=1.92e-6, two-sided Wilcoxon signed rank test). E) Empirical cumulative distribution function for the data shown in D. Every individual cell, for both control and conditioned cases, is represented as the number of spikes for the first half of the trace minus the spikes for the second half (n=50) 7 Figure 2. The expression of the conditioning effect is gradual over the course of the stimulation. Firing patterns were assessed every 100 conditionings until 500 trials were completed. A) Representative cell whose firing pattern is non-adapting in control conditions (grey). After 100 stimulations the cell shows an adapting pattern (red), after 200 the adaptation gets stronger and an intrinsic burst pattern emerges after the successive conditionings (orange to blue). Conditioning protocols are showed on the insets. Red line shows the median. B) Mean fraction of spikes for the population in the first and second half of the voltage trace during the successive conditionings. A significant redistribution on the fraction of spikes is observed during the course of the conditioning. The fraction of spikes on the first half continuously increases in favor of the second half (n = 8, p=1.24e-5, repeated measures ANOVA) C) Empirical cumulative distribution function for the data shown in B. The number of spikes for the first half of the trace minus the spikes for the second half is shown for every cell (n=8) D) Average resting membrane potential (Vm) and input resistance (Rin) of the cells during the course of the conditioning. Vm was measured as the baseline voltage before the depolarization caused by the conditioning, Rin was measured by injecting a negative current step after each conditioning trial. Circles indicate mean, bars indicate SEM (n=8). In the cases where Vm was above -60 mV a small holding current was injected to keep the cell stable during the conditioning.
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. transition in firing, in which more regular or accelerating patterns changed towards different degrees of adaptation and intrinsic burst responses. For example, the cell shown in Figure 2A presents a non-adapting or regular pattern in 114 control conditions. After 100 repetitions of the conditioning protocol the spike distribution changes to an adapting 115 continuous pattern. This adaptation gets reinforced after 200 conditionings, with the cell later adopting an intrinsic 116 burst pattern. Other cells showed a progression for more regular to only adapting patterns. As an average, the smooth 117 transition is reflected in the mean fraction of spikes shown in Figures 2B and C. As it can be observed, successive 118 conditionings translate into a stronger redistribution of this fraction in favor of the first half of the trace, a reflection 119 of the cells adopting a stronger adapting or intrinsic burst responses (n=8). The progression of the resting membrane 120 potential (Vm) and input resistance (Rin) during the course of the conditioning was also monitored, and it is shown in 121 Figure 2D. A progressive increase in Vm and a progressive decrease in Rin is observed (Rin, from 166.4 ± 54.2MΩ 122 to 115.3 ± 53.3MΩ, two-sided Wilcoxon signed rank test, p=0.031, n=7) (Vm, from -64.6 ± 6.2mV to -58.7 ± 5.7mV, 123 two-sided Wilcoxon signed rank test, p=0.047, n = 7).

124
Changes in intrinsic properties of neurons have been previously reported to be long-lasting, similar to synaptic forms 125 of long-term plasticity (Turrigiano and Nelson, 2004; Titley et al., 2017). We thus assessed whether the firing pattern 126 plasticity was stable in time or whether the changes were transient. We conditioned the cells and followed their firing 127 response up to 30 minutes after conditioning. In all cases, cells presented a post-conditioned characteristic change in 128 firing pattern towards adapting and intrinsic burst patterns (as in Figure 1). This same pattern was assessed every 10 129 minutes and showed to be persisting up to the whole recording period (see Figure 3), indicating a long-lasting change 130 in intrinsic excitability. In two cells the pattern persisted up to 50 min of recording. A representative cell is shown affect the outcome of the conditioning, as shown by previous reports of subthreshold synaptic plasticity in the CA3 circuit (Brandalise and Gerber, 2014; Brandalise et al., 2016). However, no differential effect on the pattern was found 139 (Figure 3), indicating that the mechanism of intrinsic firing plasticity is not sensitive to these two different time scales.

140
Firing pattern transitions are independent of synaptic input and are blocked by protein kinase A and C inhibitors 141 We attempted to resolve whether synaptic input was necessary to elicit the changes, or whether they could be induced 142 by direct stimulation of the soma. To this end, we used intra-somatic injection of paired step current pulses whose 143 parameters were chosen to elicit a similar somatic voltage response compared to that generated by the mossy fiber 144 stimulation ( Figure 4). This direct subthreshold somatic stimulus evoked changes in discharge pattern that were 145 similar to those elicited by the indirect mossy stimulation. For example, the cell in Figure 4A  for mossy fiber stimulation. This result suggests that the mechanism underlying the changes in firing pattern is not 151 localized to synapses, but rather acts at a more central, probably somatic or proximal dendritic level.

152
The fact that the firing pattern transitions could be reproduced by this direct depolarization of the soma raised the 153 question of whether the somatic depolarization elicited by mossy fiber activation is necessary to elicit the observed 154 changes in firing. We thus repeated the mossy conditioning experiment (Figure 1) while artificially hyperpolarizing 155 the neuron (see Figure S2). The hyperpolarization did not abolish the effect of the conditioning, since a significant 156 redistribution of spikes in favor of the first half of the trace was observed after the conditioning (see Figure S2B and . The effect of the conditioning is persistent over time. Cells where followed for 40 minutes to assess whether the firing pattern plasticity was long-term. A) Example cell with an accelerating firing pattern in control conditions (grey). The cell was conditioned subthresholdy with a double mossy fiber current pulse separated by 60 ms and given at a frequency of 1 Hz (protocol is shown in the middle, red line indicates the median). A change in pattern to intrinsic burst is elicited (blue). B) The same cell was followed every 10 min after conditioning until reaching 40 min of recording (orange, purple and blue). The pattern remained stable. C) Mean fraction of spikes for the population in the first and second half of the voltage trace before, after conditioning and every 10 min thereafter. A significant redistribution on the fraction of spikes is observed after the conditioning (n=6, p=0.031, two-sided Wilcoxon signed rank test). No significant change in this fraction was observed over 30 min after conditioning (n=6, p=0.4, repeated measures ANOVA). D) Empirical cumulative distribution function for the data shown in C.
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint . CA3 firing pattern transitions occur upon somatic conditioning and are blocked by kinase inhibitors. A) Example of an intrasomatic conditioned cell that switched from delay accelerating (grey traces) to intrinsic burst firing (blue traces). The conditioning protocol is shown in the middle column. The red line shows the median. EPSPs were evoked by injection of paired current steps, of 50 ms in duration and separated by 20 ms. The double steps were repeated 500 times at 1 Hz. The series of repeated pulses are shown superimposed. A sample trace is shown in red. B) Mean fraction of spikes for the population in the first and second half of the voltage trace for both control and conditioned cases. A significant redistribution on the fraction of spikes occurs after the conditioning. The fraction of spikes on the first half is increased while it decreases in the second half (n=12, p=0.0024, two-sided Wilcoxon signed rank test). C) Empirical cumulative distribution Function for the data shown in B. Every individual cell is represented as the number of spikes for the first half of the trace minus the spikes for the second half (n=12). D) Example of a mossy fiber conditioned cell (as described in Figure 1) under the presence of H-89 and Go 6983 (PKA and PKC inhibitors) on the intracellular pipette. The cell expresses a delay accelerating pattern in control conditions and remains under such pattern after the conditioning protocol is applied. E) Mean fraction of spikes for the population in the first and second half of the voltage trace for both control and conditioned cases. The redistribution of the fraction of spikes was not significant after the conditioning (n=13, p=0.266, two-sided Wilcoxon signed rank test). F) Empirical cumulative distribution function for the data shown in D. Every individual cell is represented as the number of spikes for the first half of the trace minus the spikes for the second half (n=13). pattern switched to intrinsic burst. These results suggest that a transient depolarization, such as the intrasomatically 162 injected stimulus, is sufficient but not necessary to elicit the effect. The residual effect may indicate that the mechanism 163 is localized near the MF synapse, in which case somatic hyperpolarization could be insufficient to prevent depolarization 164 there. Note however that in most neurons a handful of depolarizing trials were accidentally elicited while adjusting 165 the magnitude of the current MF pulse. Additionally, some cells presented occasional rebound spiking caused by the 166 hyperpolarization, while an increase in stimulation amplitude due to the increase in driving force was also frequent.

167
This could all potentially contribute to the observed effect.  Figure 4D shows a cell whose firing pattern in  Figure 1D, no significant redistribution of the spikes was found on the presence of the inhibitors (n=13). These results 179 suggest that phosphorylation is implicated in the mechanism of firing pattern transition.

180
Using Dynamic Time Warping (DTW) and a conductance based model to infer firing transitions and parameter 181 changes after plasticity 182 We observed that the conditioning induced firing pattern changes from more regular patterns towards early bursting 183 and adapting patterns. We sought to quantify these changes using hierarchical clustering methods (Druckmann et al., CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint every response, and to quantify the frequencies of transitions between them. For our clustering method, we obtained instantaneous firing rate vectors of the experimental voltage traces and estimated pairwise distances using the DTW 187 algorithm. DTW operates directly on the action potential temporal sequence rather than relying on a pre-defined set can be interpreted as the subthreshold voltage envelope in which the discharge response of each cell rides, and this 190 envelope is the essence to catalogue similar firing patterns. Once the distances between the vectors were calculated, 191 Wards linkage was applied in order to obtain a hierarchical tree to reveal the classes.

192
The results of the cluster analysis of discharge patterns are shown in Figure 5A. We set the threshold of the clustering  In particular, many of the traces belonging to the delayed spiking cluster are derived from cells whose traces at low 210 13 . CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. Every experimental trace is matched to a model database of traces. Using the DTW distance on the instantaneous firing rate vectors the best matches are selected (best match is depicted). A conductance estimate for the experimental trace is obtained (average of 10 best matches are shown). D) Transitions observed between firing patterns before and after conditioning. Each cell is assigned to a single cluster (represented as a box) for both the control and conditioned cases. Arrows indicate transitions between types whenever a cell changed cluster. Self-loops indicate that the firing pattern was retained after conditioning. Numbers indicate percentages of observed transitions. The number of cells in each category under control conditions is displayed next to each pattern type. Cells tend to transition towards adapting and bursting patterns following conditioning (n = 43). Seven cells were assigned as unclassified. A conductance road map showing the key conductances responsible for a transition in firing pattern are represented on the edges. The main channels implicated are gCa, gKd slow and gKm. E) Average conductance composition for matched experimental cells in control (grey) and conditioned cases (blue). There is a significant increase in gKm, gCa and a decrease in gKd slow (gCa p=0.015, gKm p= 0.0084, gCaK p= 0.2203, gKd p= 0.2501, gKd slow p=2.01e −8 , two-sided Wilcoxon signed rank test, n = 485).
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint current injections were assigned to the accelerating cluster, or belonged to non-adapting cells with spiking delay. The  We next aimed to infer which underlying parameters could be responsible for the systematic transitions. Our results

224
showing that phosphorylation inhibition blocks the conditioning effect support the hypothesis that the prime candidate 225 for this plasticity is a change in the profile of active conductances. We explored this possibility using simulations of 226 action potential discharge in a conductance-based single compartment neuron model containing 10 voltage and calcium 227 gated ion channels (see Methods). The densities and kinetics of these channels were derived from experimental 228 measurements of CA3 pyramidal neurons (Hemond et al., 2008).

229
A database of representative ranges of maximal conductances that could plausibly explain the discharge patterns 230 observed experimentally was generated using the single compartment model. To do this, the maximal conductances of 231 the different channels were swept through ranges that would likely encompass the experimentally observed patterns.

232
The spiking conductances were left constant, whereas we varied the conductances with longer time constant, which are 233 responsible of the discharge dynamics: gCa , gCaK , gKm , gKd and gKd slow (see Table 1 for the exact ranges). In this 234 manner a total of 100'000 conductance profiles were generated. We obtained the discharge response to different levels of current injection for each conductance profile, giving a total of 800'000 voltage traces with their associated maximal conductance profiles. Every single experimental trace (coming from both, control and conditioned cases) was matched 237 against the collection of traces in the model database using the DTW algorithm on the instantaneous firing rate vectors 238 (see Figure 5C for an example). The best fits were then selected, allowing us to obtain an estimate of the maximal 239 conductance profile likely to be present in the experimental neuron ( Figure 5C). The key to infer the parameters is thus 240 to recognize, via the DTW algorithm on the rate vectors, the subthreshold voltage envelope generated by the long-time 241 constant conductances, and not the precise spike times.

242
The diagram of Figure 5D represents the crucial conductances determining the transitions between discharge patterns 243 in firing pattern space. These are gKm, gCa and gKd slow . For example, to move to the intrinsic burst cluster (yellow) a 244 characteristic enrichment in gKm and gCa is needed, which allows for the generation of the burst (given the presence 245 of basal levels of gCaK) and the spacing of further spikes. For the accelerating and delayed patterns (blue and purple), 246 the presence of gKd is important for a delayed onset of the spiking, and the slow inactivation of gKd slow is necessary 247 for generating the continuous acceleration of the spike rate. In the case of the adapting patterns (green), the inclusion 248 of gKm is necessary for the slowing down of the action potentials after the initial discharge.
249 Figure 5E shows the average conductance content of the matched experimental traces in control and conditioned cases.

250
The shift towards adapting and intrinsic bursting behaviors after the conditioning corresponds to a significant increase 251 in gKm and gCa, and a decrease in gKd slow (gCa p=0.015, gKm p= 0.0084, gCaK p= 0.2203, gKd p= 0.2501, gKd slow 252 p=2.01e −8 , two-sided Wilcoxon signed rank test, n = 485). This correspondence of firing patterns and biophysical 253 parameters offers an interpretation of the causes of transitions between firing behaviours induced by the conditioning.

254
Inhibition of Kv7 and calcium channels abolishes the effect of firing pattern plasticity 255 Having unraveled via the modeling study that the changes in firing pattern could correspond to an increase in gKm and 256 gCa conductances, we decided to test this prediction via pharmacological blocking of the corresponding channels. We 257 thus repeated the original experiment ( Figure 1) and administrated the blockers NiCl 2 and XE991 via the perfusion 258 system after the conditioning. NiCl 2 is known to block T,N and L type calcium channels, with a stronger effect on the 259 16 . CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint (Brown and Randall, 2009). The results of this experiment are shown in Figure 6. Administration of the drugs blocked 261 the effect of the conditioning, with the cells loosing the change in pattern 20 min after perfusion (note that the drug took 262 3-4 min to reach the bath and start diffusing). For example, Figure 6A shows a cell that switched from accelerating to 263 intrinsic burst after conditioning. Drugs were administrated immediately after checking the conditioning effect and the 264 cell was followed for 30 min since that point ( Figure 6C). 20 minutes after starting of drug perfusion the cell presented  Figures 6B and D), the model estimate of its conductance 269 distribution is shown, as explained in Figure 5C. An increase in gCa and gKm conductances is observed after the 270 conditioning, which then decreases after application of their corresponding channel blockers. 271 We noted that a residual adaptation remained in some neurons after the drug administration. This could likely be 272 due to the low concentration of XE991 employed (10 µm), although it could also be due to the small acceleration 273 at the beginning of the trace given by the delay present in these neurons. We decided to identify whether gKd was 274 responsible for such delay as hinted by the model (Figure 6D). D-type currents, caused by Kv1 channels, can be 275 blocked by low concentrations of 4-aminopyridine (4-AP). We thus repeated the experiment using 4-AP and NiCl 2 276 as conductance blockers (see Figure S3). After successful conditioning of the cells (see Figure S3A for an example), 277 the drugs were administered via the perfusion system. As in the previous experiment, the burst response presented 278 by the neuron, likely caused by calcium channels, was abolished. 15 min after perfusion the delay presented by the 279 neuron was also removed, with the cell adopting an adapting continuous pattern. Middle panel of Figure S3B shows 280 the effect of 4-AP in removing the delay of the population of cells. There is a significant reduction of the timing of 281 the first spike comparing to the XE991 experiment. For this subset of cells, no effect on the fraction of spikes was 282 found after drug perfusion (see Figure S3E and F). This is likely due to the cells keeping adaptation profiles with no CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint  This study was performed on organotypic cultures, derived from brain slices of newborn rats that were incubated 290 for three weeks using the roller-tube technique (Gähwiler, 1981). Organotypic cultures have been used extensively  Figure 4). A subset of cells were also condioned via the mossy fiber 299 pathway. In general, the conditioning elicited an increase in burstiness at the end of the voltage trace, which was higher 300 in frequency of that encountered in the organotypic case. For example, the cell in Figure 7A presented a non-adapting 301 pattern in control conditions. After conditioning the cell showed a delay burst pattern. A subset of cells also presented 302 an increase in adaptation ( Figure 7B) as observed in the organotypic slice case ( Figure 5D). A significant change in the 303 fraction of spikes was found at the population level ( Figures 7C and D). These results indicate that conditioning elicits 304 a change in firing pattern in the acute slice preparation, suggesting the firing pattern plasticity is a generic property of 305 CA3 cells.

19
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint Figure 7. Conditioning changes the firing pattern of CA3 neurons in the acute slice preparation. A) Example of a cell with a non-adapting firing pattern in control conditions (grey). The cell is conditioned via somatic current injections (as in Figure 4). Protocol is shown in the inset. The firing pattern of the cell changes towards delay bursting (blue). B) Cell presenting a non-adapting burst firing pattern in control conditions. After somatic conditioning the cell presents an adapting firing response. C) Mean fraction of spikes for the population in the first and second half of the voltage trace during successive conditionings. There is a significant redistribution of the fraction of spikes after the conditioning (control-conditioned, n = 31, p=0.003, two-sided Wilcoxon signed rank test) (control-conditioned-reconditioned, n = 31, p=0.04, repeated measures ANOVA) D) Empirical cumulative distribution Function for the data shown in C. The number of spikes for the first half of the trace minus the spikes for the second half is shown for every cell.
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint

307
The diversity of firing patterns upon step current injection that neurons present have been studied and catalogued for CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint in our study significantly adapt their spiking dynamics, adding an extra timing dimension to the previous reports (in accordance to Grasselli et al. (2016) and Campanac et al. (2013)).

333
Most of the work on intrinsic plasticity require that the cell fire during the conditioning, whereas we observe that  However, our changes were induced after direct subthreshold somatic conditioning, ruling out a synaptic cause.

342
The studies cited here were performed both in acute slices and in organotypic cultures. We find that the expression of 343 the conditioning differed between these two preparations. Although induction of adaptation was found both cases, the

366
One of the typical transitions that we observe is the switch towards bursting behaviors. We emphasize that this is 367 not the only transition induced, but rather that special attention should be given to this bursting mechanism. It is 368 known that some neurons present this dual behavior. For example, relay cells on the thalamus become bursty upon 369 hyperpolarization because of T-type conductance inactivation (Sherman, 2001). In our case, the cells depolarized 5 370 mV in average, while kinase inhibitors blocked the effect, ruling out this hyperpolarizing cause.   CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint were converted into a time series of the instantaneous firing rates. The instantaneous firing rate at each spike was taken as 1/Inter-spike-Interval (ISI). Then the instantaneous rates where linearly interpolated across the spike times at down-sampling the interpolated rate traces by a factor of 10, in order to make them computationally tractable to the 498 similarity measurement.

499
Similarity distances between pairs of traces were calculated using the Dynamic Time Warping (DTW) measure (Berndt 500 and Clifford, 1994). DTW takes into account that two similar signals can be out of phase temporarily, and aligns them 501 in a non-linear manner through dynamic programming (Keogh and Ratanamahatana, 2005). The algorithm takes two 502 time series Q = q 1 , q 2 , . . . , q n and C = c 1 , c 2 , . . . , c m and computes the best match between the sequences by finding 503 the path of indices that minimizes the total cumulative distance 504 DTW(Q,C) = min where w k is the cost of alignment associated with the k th element of a warping path W . A warping path starts at q 1 505 and c 1 respectively, and finds a monotonically increasing sequence of indices i k and j k , such that all elements q i in 506 Q and c j in C are visited at least once, and for the final step of the path i end = n and j end = m holds. The optimal 507 DTW distance is the cumulative distances y(i, j), corresponding to the costs of the optimal warping path q 1 , . . . , q i 508 and c 1 , . . . , c j . This distance can be computed iteratively by dynamic programming: where d(q i , c j ) is the absolute difference between the elements of the sequence. The optimal warping path is obtained 510 by backtracking from the final element y(n, m), and finding which of the three options (increasing i only, increasing j 511 only, or increasing i and j simultaneously) led to the optimal warping distance, until i = 1, j = 1 is reached. A warping 512 29 . CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint window constraint of 10% of the vector size was chosen (Keogh and Ratanamahatana, 2005). 513 The pairwise DTW distances were used to perform hierarchical clustering by Ward's algorithm (Ward Jr, 1963). The 514 number of classes increases with the level of the hierarchy. We choose to cut the tree at a level that provided sufficient 515 structure to interpret the hierarchy in terms of recognized response types (Ascoli et al., 2008).

516
Every recording for a given cell was treated as an independent observation, and could in principle be assigned to any 517 cluster. If the electrophysiological state of the cell is expressed in all of its responses, then we expect that all the 518 independent observations derived from that cell should be assigned to the same cluster. However, traces derived from 519 current injections to the same cell in different conditions (pre-or post-stimulation) are expected to be assigned to 520 different clusters if there is significant change in the underlying electrophysiological state.

521
In fact the independent traces did not cluster perfectly. Instead, the majority of independent observations derived 522 from a given state clustered together and there were a few that fell into other clusters. Therefore, we chose to label CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint properties were modeled by including appropriate voltage and calcium gated ion channels whose density and kinetics 536 were obtained from experimental recordings performed in CA3 neurons (Hemond et al., 2008). All the conductances 537 included in the model where obtained from this work, except for gKd slow , which had to be added in order to match the 538 accelerating traces. We found that a 10 fold increase in the time constant of inactivation of gKd significantly improved 539 the accelerating index. A similar slow gKd current matching these kinetics has actually been found in CA3 neurons 540 (Luthi et al., 1996). The faster Kd current, has been previously reported both in cortex and hippocampus (Storm, 1988; Each current I x is described by the equation whereḡ is the maximal conductance, m and h are activation and inactivation terms, V is the membrane potential, and 547 E the reversal potential of the channel. The reversal potentials for Na + and K + were E Na = 50 mV and E K = -85

31
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint The first term of the above equation describes the change caused by Ca 2+ influx into a compartment with volume v. F 554 is the Faraday constant, I Ca is the calcium current and τ Ca is the time constant of Ca 2+ diffusion.

555
The occasional decrease in spike amplitude seen in some of the experimental traces is probably due to sodium and ranges used to generate the database see Table 1. A total of 100'000 conductance vectors were generated by 567 combining the different conductances. The firing pattern of every conductance vector was produced at 8 different 568 levels of step-current injection, obtaining a total of 800'000 voltage traces. An integration step of 0.2 ms was used.

569
After generating the database non-spiking traces were removed, together with traces with saturating spikes. This

32
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under which includes all the observed experimental firing patterns, was generated by varying 5 maximal conductances (gCa , gCaK , gKm , gKd and gKd slow ) over a given range. Different ranges of step current I were also needed to reveal the different firing types. A total of 100'000 conductance vectors were generated by combining the different conductances. The firing pattern of every conductance vector was produced at several levels of step-current injection, obtaining a total of 800'000 voltage traces. Note that gCaT , gCaN and gCaL are englobed under the single parameter gCa.
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint the time scale of our experimental traces, which were unraveled at 3-5 seconds of step current. Although generation 576 of the traces for this longer duration was possible, the resulting firing patterns did not reproduce faithfully all the 577 spiking dynamics encountered in the experiments. A change in channel kinetics (Hemond et al., 2008), an additional 578 conductance, or a dendritic load could possibly solve the issue. The objective however was to gain an intuition on the 579 possible conductance distribution changes induced by the conditioning. This together with computational reasons to 580 generate the database led us to proceed with the simulations using a 1 second current step. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint Figure S1. CA3 firing patterns are stable over time as well as to changes in resting membrane potential. Firing pattern transitions are not elicited by step current injection alone. A) Examples of two cells whose firing pattern have been measured by step-wise current injection (protocol showed in the inset). The cells do not show changes in firing pattern after 15 min of recording. B) Mean fraction of spikes for the population in the first and second half of the voltage trace for both control and conditioned cases. No significant redistribution on the fraction of spikes is observed (n = 15, p=0.583, two-sided Wilcoxon signed rank test). C) Empirical cumulative distribution Function for the data shown in B. Every individual case is represented as the number of spikes for the first half of the trace minus the spikes for the second half. D) Firing pattern transitions are not elicited by sustained shifts in membrane potential. Examples of two cells that have been hold at different membrane potentials through steady current injection (-70, -80 and -60 approximately). After changing the holding potential of the recorded neuron the firing pattern was measured by step-wise current injection (protocol showed in the inset). No transitions of firing pattern were observed at any of the different holding potentials. E) Mean fraction of spikes for the population in the first and second half of the voltage trace for every condition. No significant redistribution on the fraction of spikes is observed (Vm 60 vs 70, p=0.652; Vm 60 vs 80, p=0.084; Vm 70 vs 80, p=0.695) (n = 10, two-sided Wilcoxon signed rank test)). F) Empirical cumulative distribution function for the data shown in E. Every individual case is represented as the number of spikes for the first half of the trace minus the spikes for the second half.
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint Figure S2. The effect of mossy fiber conditioning does not require somatic depolarization. Conditioning was performed while hyperpolarizing the cell with a negative current pulse. A) Cell that presents an accelerating pattern in control conditions (grey). Conditioning (MF double pulse, delta 20 ms given at 1 Hz) was elicited under the presence of an hyperpolarizing current step (protocol is shown in the inset, red line indicates the median). The cell changes its firing pattern to an non-adapting burst response. Thereafter, the cell is reconditioned via only the mossy fiber double pulse. After this conditioning the cell presents an intrinsic burst pattern. B) Mean fraction of spikes for the population in the first and second half of the voltage trace during the successive conditionings. A significant redistribution on the fraction of spikes is observed after the conditioning under the hyperpolarizing pulse (n=9, p=0.016, two-sided Wilcoxon signed rank test). After a second conditioning, without hyperpolarization, a stronger effect is found (n=9, p=0.008, two-sided Wilcoxon signed rank test). C) Empirical cumulative distribution Function for the data shown in B. The number of spikes for the first half of the trace minus the spikes for the second half is shown for every cell (n=9).
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint Figure S3. Inhibition of Kv1 and calcium channels reduces the delay of the traces but does not abolish the effect on the fraction of spikes. A) Cell with an adapting burst firing pattern in control conditions (grey) that switches to intrinsic burst after conditioning (blue). C) After 15 min of bath application of 4-AP and NiCl 2 (at 30 and 200 µM) the cell switches to an adapting continuous pattern, with no delay (purple and blue traces). B) and D) Conductance model fits for every voltage trace. The middle panel shows the mean delay for the first spike when 4-AP is administrated with NiCl 2 via the perfusion system, in comparison with XE991. A decrease on this delay by 4-AP is observed (n=8, p=0.02, two-sided Wilcoxon signed rank test). E) Mean fraction of spikes for the population in the first and second half of the voltage trace during the experiment. A significant redistribution on the fraction of spikes in favor of the first half is observed after conditioning (n=8, p=0.015, two-sided Wilcoxon signed rank test). Application of the drugs did not have an effect on the fraction of spikes (n = 8, p=0.98, repeated measures ANOVA) C) Empirical cumulative distribution function for the data shown in E.
. CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019.   CC-BY-NC-ND 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted August 7, 2019. ; https://doi.org/10.1101/084152 doi: bioRxiv preprint in nerve and muscle cells. Physiological reviews 85:883-941.