KCNQ5 K+ channels control hippocampal synaptic inhibition and fast network oscillations

KCNQ2 (Kv7.2) and KCNQ3 (Kv7.3) K+ channels dampen neuronal excitability and their functional impairment may lead to epilepsy. Less is known about KCNQ5 (Kv7.5), which also displays wide expression in the brain. Here we show an unexpected role of KCNQ5 in dampening synaptic inhibition and shaping network synchronization in the hippocampus. KCNQ5 localizes to the postsynaptic site of inhibitory synapses on pyramidal cells and in interneurons. Kcnq5dn/dn mice lacking functional KCNQ5 channels display increased excitability of different classes of interneurons, enhanced phasic and tonic inhibition, and decreased electrical shunting of inhibitory postsynaptic currents. In vivo, loss of KCNQ5 function leads to reduced fast (gamma and ripple) hippocampal oscillations, altered gamma-rhythmic discharge of pyramidal cells and impaired spatial representations. Our work demonstrates that KCNQ5 controls excitability and function of hippocampal networks through modulation of synaptic inhibition. Several K+ channels control neuronal excitability, but the function of KCNQ5 (Kv7.5), which displays wide expression in the brain, is not known. Here the authors show that KCNQ5 controls excitability and function of hippocampal networks through modulation of synaptic inhibition.

K CNQ genes encode five voltage-gated potassium channels in mammals (KCNQ1-5 or K v 7.1-K v 7.5) and mutations in KCNQ1-4 underlie human diseases 1,2 . KCNQ1 mutations entail cardiac arrhythmias and deafness, mutations in KCNQ2 and -3 lead to benign familial neonatal convulsions [3][4][5] and KCNQ4 mutations can cause progressive hearing loss 6 . KCNQ5 has not yet been linked directly to human disease, although KCNQ5 is deleted together with other genes in some patients with mental retardation 7,8 and dysfunction of KCNQ3/KCNQ5 heteromers has been associated with autism disorders 9 .
In the brain, KCNQ subunits 2, -3 and -5 show wide neuronal expression, whereas KCNQ4 is restricted to specific brain stem nuclei including parts of the auditory pathway 10 . KCNQ4 is also found in the inner ear 6 and a subset of touch receptors in the skin 11 . KCNQ2 and -3 localize to axon initial segments and nodes of Ranvier 12,13 . KCNQ2/KCNQ3 heteromers are considered to be a molecular correlate of the M-current 14,15 . These currents are partially active at resting membrane potential and their inhibition by G-protein-coupled receptors increases neuronal excitability. In addition, KCNQ2 and -3 mediate components of the medium and slow afterhyperpolarization (AHP) currents in hippocampal neurons [16][17][18] .
Compared with KCNQ2, -3 and -4, little is known about the function of KCNQ5. KCNQ5 is widely expressed throughout the brain 13,19,20 . In the hippocampus, KCNQ5 contributes to AHP currents in the CA3 area 13 . KCNQ5 was reported to localize mainly to neuronal cell somata and dendrites in human cortex and hippocampus 21 , to glutamatergic synapses in several rat brain stem nuclei 22,23 and to the postsynaptic membrane of vestibular calyx terminals 24 . Furthermore, KCNQ5 was proposed to be located at the presynaptic membrane of the calyx of Held and to control its properties 23 .
Here we show that in the mouse hippocampus, where it is most highly expressed in the CA3 area, KCNQ5 localizes to both pyramidal cells and all investigated subtypes of inhibitory interneurons. Immunolabelling reveals KCNQ5 clustering at postsynaptic membranes of GABAergic synapses. To elucidate the global physiological function of KCNQ5, we used Kcnq5 dn/dn mice that carry a dominant negative mutation (G278S) in the pore loop 13 , resulting in a non-conducting mutant protein. This functional ablation of KCNQ5 results in alterations of synaptic inhibition and excitability in the hippocampus in vitro and in vivo. Fast network oscillations, gamma-rhythmic firing of pyramidal cells and spatial representations are impaired in freely moving Kcnq5 dn/dn mice. Our work shows that KCNQ5 modulates hippocampal synaptic inhibition and network synchrony and unveils an important role of KCNQ5 in hippocampal information processing.

KCNQ5 expression in pyramidal cells and interneurons.
KCNQ5 is widely expressed in many brain regions 19,20 . In the hippocampus, mRNA hybridization (Allen brain atlas, http:// www.brain-map.org/) and immunohistochemistry revealed higher KCNQ5 expression in the CA3 than CA1 hippocampal region 13 . Immunolabelling with our KCNQ5 antibody was intense in strata oriens and radiatum, regions that harbour extensive synaptic contacts (Fig. 1a). In CA3, there was a gradient of KCNQ5 expression from CA3a (close to CA2) towards the CA3c area (close to dentate gyrus). In the absence of mice lacking KCNQ5 expression, the specificity of KCNQ5 labelling was confirmed with Kcnq5 dn/dn knock-in mice 13 , which express a dominant-negative (dn), non-conducting pore mutant (G278S) of KCNQ5 (ref. 20). In these mice, the pattern of KCNQ5 labelling of the hippocampus was changed (Fig. 1b). Labelling was reduced in the CA1 area, suggesting that the mutant protein might be less stable. In CA3, KCNQ5 immunoreactivity was shifted from the stratum (st.) radiatum and st. oriens to pyramidal cell bodies (Fig. 1b,c).
We neither detected compensatory changes in the expression of other KCNQ family members 13 or of several Kir K þ channels ( Supplementary Fig. 1a). The overall normal brain histomorphology of Kcnq5 dn/dn mice 13 and the unchanged expression of several synaptic proteins in these mice ( Supplementary Fig. 1b) further suggested that the reduced KCNQ5 labelling of st. radiatum and oriens is not owed to neurodegeneration. As KCNQ4 channels with an equivalent pore mutation (G285S) 6 show diminished plasma membrane localization in heterologous expression systems 25 and in vivo 11,24 , we rather suspected abnormal protein trafficking as the underlying cause. Indeed, when expressed in Xenopus oocytes, KCNQ5 G278S subunits carrying an extracellular HA-tag displayed reduced plasma membrane localization compared with similarly tagged wild-type (WT) KCNQ5 ( Supplementary Fig. 1c,d). This effect was observed either when KCNQ5 G278S was expressed alone or together with KCNQ3, which can form heteromers with KCNQ5 (refs 19,20). Hence, the altered staining pattern of Kcnq5 dn/dn hippocampi likely resulted from impaired trafficking of the mutant non-conducting protein. Instead of reaching their normal peripheral position on neurites, KCNQ5 G278S mutant channels remain in somata of pyramidal cells, similar to the recently described retention of this mutant in somata of vestibular ganglion cells 24 .
In addition to confirming the specificity of our immunolabelling experiments, the trafficking defect of the KCNQ5 G278S mutant facilitated the identification of cell types expressing KCNQ5. The shift of immunoreactivity from st. oriens and st. radiatum to pyramidal cell somata (Fig. 1a-c) indicated that a sizeable portion of KCNQ5 is present on neurites of pyramidal cells. In Kcnq5 dn/dn , but not in Kcnq5 þ / þ hippocampi, immunostaining revealed KCNQ5-positive cell somata also outside of the pyramidal layer, for example, in the st. radiatum (Fig. 1b). This preferential labelling of Kcnq5 dn/dn somata likely results from a retention of the mutant protein in cell bodies, where, in contrast to the WT situation, it can be distinguished from diffuse background labelling. Indeed, in situ hybridization revealed similar labelling outside the pyramidal cell layer in Kcnq5 þ / þ and Kcnq5 dn/dn mice ( Supplementary Fig. 2a,b). To examine whether these cells corresponded to inhibitory interneurons, we crossed Kcnq5 dn/dn mice to GAD67-GFP mice that specifically label GABAergic neurons 26 (Fig. 1d). Indeed, 12.3% (98/781 cells from three mice) of GFP-positive cells in Kcnq5 dn/dn Â GAD67-GFP þ / À mice were robustly labelled for KCNQ5. KCNQ5labelled interneurons were found in the st. radiatum, oriens and lacunosum-moleculare (Fig. 1b,d-f). The KCNQ5 signal in Kcnq5 dn/dn mice overlapped with markers for various types of interneurons like parvalbumin and somatostatin (Fig. 1e,f) and calretinin and neuropeptide Y ( Supplementary Fig. 3a-c). A sizeable fraction of interneurons stained with any of the interneuron markers was unambiguously labelled for KCNQ5 ( Supplementary Fig. 3d). Likewise, combining in situ hybridization with immunohistochemistry, parvalbumin-and somatostatin-positive interneurons were found to express Kcnq5 mRNA in both Kcnq5 þ / þ and Kcnq5 dn/dn hippocampi ( Supplementary Fig. 2c-f). Hence, in addition to hippocampal pyramidal cells, KCNQ5 is widely expressed in hippocampal interneurons of different subtypes. exerts dominant-negative effects on co-expressed KCNQ3, -4 and -5 subunits 13,20,24 . Kcnq5 dn/dn mice, which carry the G278S mutation on both alleles, are therefore expected to lack currents from homo-or heteromeric channels carrying at least one KCNQ5 subunit. By contrast, the consequences of a deletion of KCNQ5 in Kcnq5 À / À mice are difficult to predict. In some cells, the number of KCNQ2/KCNQ3 channels may increase because KCNQ5 no longer competes with KCNQ2 for KCNQ3 binding. As KCNQ2/KCNQ3 heteromers yield large K þ -currents 27 , Kcnq5 À / À mice may display phenotypes resulting from both a decrease and increase of specific K þ currents. Our approach, that is, the use of Kcnq5 dn/dn mice avoids all these complications and allowed us to specifically investigate the role of KCNQ5 in hippocampal slices and in vivo.
The identification of KCNQ5 in interneurons prompted us to investigate intrinsic excitability of interneurons in the CA3 after blocking synaptic transmission. To identify interneurons, we crossed Kcnq5 dn/dn mice with mice expressing GFP under control of the GAD67 promotor. Recorded cells were filled with biocytin by the patch pipette and 59 cells could be subsequently classified morphologically into basket, bistratified and orienslacunosum-moleculare interneurons. All three types of interneuron (Fig. 2a-c)   response to depolarizing current injections ( Fig. 2d-i). We likewise observed an increased excitability in non-classified interneurons in st. radiatum ( Supplementary Fig. 4a). In contrast to interneurons, the excitability of CA3 and CA1 pyramidal cells appeared unchanged ( Supplementary Fig. 4b,c). Resting membrane potential was not altered in either cell type ( Supplementary Fig. 4d,e).
Enhanced hippocampal GABAergic inhibition in Kcnq5 dn/dn mice. The increased excitability of interneurons suggested that the functional loss of KCNQ5 would boost synaptic inhibition in the CA3 area. Indeed, we observed an increase in the frequency of spontaneous inhibitory postsynaptic currents (sIPSCs) in Kcnq5 dn/dn CA3 pyramidal cells and a shift of their mean amplitudes to larger values ( Fig. 3a-d). By contrast, neither frequency nor amplitude of spontaneous excitatory postsynaptic currents (sEPSCs) were altered in Kcnq5 dn/dn CA3 pyramidal cells ( Supplementary Fig. 4f). Moreover, inhibition of extrasynaptic GABA A receptors in voltage-clamp experiments suggested that ambient GABA (g-aminobutyric acid) was increased in Kcnq5 dn/dn CA3 pyramidal cells ( Supplementary Fig. 4g,h). Because KCNQ channels can be inhibited by muscarinic M1 receptors 14,28 and since hippocampal circuits are modulated by cholinergic input 29,30 , we tested the effect of the muscarinic agonist carbachol. Transient application of carbachol (20 mM) markedly increased both sIPSC amplitude and frequency in WT CA3 pyramidal cells ( Fig. 3a-d). This effect was absent in Kcnq5 dn/dn slices, indicating a major contribution of KCNQ5 to acetylcholine-mediated control of inhibition in CA3. In addition to sIPSCs, we also investigated effects on inhibitory postsynaptic currents (IPSCs) of pyramidal cells that were evoked by extracellular stimulation in st. radiatum close to the recorded cell. Consistent with our results on sIPSCs, also evoked IPSCs were significantly larger in Kcnq5 dn/dn mice compared with controls ( Fig. 3e).
KCNQ5 localizes to inhibitory synapses. As KCNQ5 was expressed in layers rich in synaptic contacts (Fig. 1a), we asked whether this protein might be found at synapses. Indeed, KCNQ5 co-localized with gephyrin ( Fig. 4a,f), a postsynaptic marker for GABAergic and glycinergic synapses, and partially overlapped with presynaptic markers like the vesicular GABA transporter vGAT and the GABA-synthesizing enzyme GAD65/67 (Fig. 4b,c). Co-localization with markers for inhibitory synapses was not only seen in st. oriens or radiatum, but also on pyramidal cell bodies that were decorated by KCNQ5-positive puncta (Fig. 4c)  were co-labelled with gephyrin (748 and 641 synapses, respectively, from three mice). KCNQ5 localized to inhibitory synapses also in other brain regions like the cerebellum (not shown) and the brainstem ( Supplementary Fig. 5a,b). KCNQ5 was not found at glutamatergic terminals that were labelled for the glutamate receptor subunit GluR2 or the vesicular glutamate transporter vGlut1 (Fig. 4d,e). As described previously 22,23 , KCNQ5 immunoreactivity was found at calyces of Held ( Supplementary Fig. 5a,b). Closer examination, however, revealed that KCNQ5 did not co-localize with vGlut1 at the presynaptic membrane ( Supplementary Fig. 5a), but rather with gephyrin ( Supplementary Fig. 5b) that marks inhibitory synapses, which contact this giant excitatory synaptic bouton 31 .
High-resolution stimulated emission depletion (STED) microscopy of CA3 synapses showed that KCNQ5 labelling overlapped with postsynaptic gephyrin, but not with presynaptic bassoon (Fig. 4f,g). Evaluation of over 200 synapses for each pair revealed a significantly shorter distance of KCNQ5 to the postsynaptic markers gephyrin and homer than to presynaptic bassoon (Fig. 4h). Subcellular fractionation ( Supplementary Fig. 5c) showed the presence of KCNQ5 in the postsynaptic density fractions (PSD) I and II, further supporting a postsynaptic localization. The reduction of KCNQ5 levels in synaptic fractions from Kcnq5 dn/dn relative to WT mice, which exceeded the reduction in total brain homogenate ( Supplementary Fig. 5c), agreed with the shift of the KCNQ5 G278S mutant from synaptic regions to cell bodies ( Fig. 1a-c). The overall reduction of KCNQ5 in Kcnq5 dn/dn brain ( Supplementary Fig. 5c) suggested an increased degradation of the mutant protein and agreed with the reduced KCNQ5 labelling in the CA1 area of Kcnq5 dn/dn mice (Fig. 1b).
Synaptic transmission in Kcnq5 dn/dn hippocampal slices. The postsynaptic localization of KCNQ5 ( Fig. 4a-c,f,g) suggested that it might shunt postsynaptic currents. To check for this possibility, we recorded miniature inhibitory postsynaptic potentials (mIPSPs; current-clamp configuration, B À 60 mV) from CA3 principal neurons and indeed found that both amplitudes and  Supplementary Fig. 7b). In control experiments performed at B À 90 mV, a voltage at which KCNQ5 channels are closed 33 , mIPSPs were comparable in WT and Kcnq5 dn/dn mice ( Supplementary Fig. 7c,d) supporting our hypothesis of KCNQ5 involvement in shunting. In voltage-clamp recordings, which methodologically rule out potential effects of electric shunting, both amplitude and frequency of miniature inhibitory postsynaptic currents (mIPSCs) were unchanged across genotypes ( Supplementary Fig. 7a). Unchanged mIPSCs also 0 *** *** suggested that KCNQ5 had no effect on vesicle filling or release. Together, these experiments demonstrate that KCNQ5 contributes to the shunting of postsynaptic inhibitory activity.
To examine the impact of KCNQ5 on synaptic transmission at the network level, we determined the input/output properties of the CA3 area in hippocampal slices. The slopes of field excitatory postsynaptic potentials (fEPSP), which reflect global postsynaptic responses of pyramidal cells upon stimulation of both glutamatergic and GABAergic inputs, were significantly smaller in Kcnq5 dn/dn than in WT mice (Fig. 4k). This difference was abolished in the presence of the GABA A receptor antagonist gabazine (Fig. 4l), demonstrating that the reduced postsynaptic response in Kcnq5 dn/dn mice was due to enhanced inhibition rather than decreased excitation of pyramidal cells.
Hippocampal network oscillations and neuronal firing in vivo. We next investigated how the changes in excitability and synaptic transmission as observed in brain slices translate into neuronal discharge and network oscillations in freely moving Kcnq5 dn/dn mice. The activity of putative pyramidal cells and interneurons was recorded using movable silicon probes and tetrodes in CA3 and CA1 areas during exploration, rest and sleep. The activity of Kcnq5 dn/dn CA3 pyramidal cells was changed as indicated by the reduced number of bursts ( Fig. 5a,b, Table 1). In line with our in vitro data demonstrating unaltered excitability of CA3 principal cells ( Supplementary Fig. 4b,c), but increased excitability of CA3 interneurons ( Fig. 2 and Supplementary Fig. 4a), overall average firing rates of CA3 pyramidal cells were not different between genotypes, whereas the average firing rates of fast-firing CA3 interneurons were significantly increased in Kcnq5 dn/dn mice (Fig. 5b, Table 1). In contrast to unchanged average rates, the firing of Kcnq5 dn/dn CA3 pyramidal cells was reduced during gamma oscillations ( Fig. 5b and Table 1).
Network oscillations coordinate the activity of neuronal populations and are believed to be crucial for cognitive processes 34 . They occur in different frequency ranges and can be recorded as changes in the extracellular local field potential (LFP). Gamma frequency band (35-85 Hz) oscillations of area CA3 shape gamma oscillations in the CA1 region in rats 35 . We now found a similar synchronization between CA3 and CA1 LFP in mice (Fig. 5c). In Kcnq5 dn/dn mice, the power of gamma oscillations (35-85 Hz) was decreased in both CA3 and CA1 areas ( Fig. 5d and Supplementary  Table 1) and the distribution of the preferred discharge phases of individual pyramidal cells was altered (Fig. 5e). Whereas in control mice B42% of neurons preferentially fired during the ascending phase of the gamma cycle ('RisPyr' cells 36 ), in Kcnq5 dn/dn mice only B25% of cells fired preferentially during this phase (Fig. 5e). During theta oscillations, WT pyramidal cells showed a strong preference for the depolarizing part of the theta cycle during which CA3 area-generated gamma oscillations 37 and the maximal discharge of RisPyr 36 are typically observed. In contrast, the discharge of Kcnq5 dn/dn pyramidal cells was more uniformly distributed along the theta cycle ( Supplementary Fig. 8a). Theta oscillations (5-10 Hz) did not differ between genotypes (Fig. 5d and Supplementary Table 1).
The CA3 and CA1 networks also participate in the synchronization underlying high-frequency oscillations (140-200 Hz, 'ripples') in CA1 during rest and sleep 38,39 . Whereas the power spectra of ripples were similar in WT and Kcnq5 dn/dn mice ( Supplementary Fig. 8b and Supplementary Table 1), their occurrence was markedly reduced in mutants (Fig. 5f). Thus, fast network oscillations were attenuated in Kcnq5 dn/dn mice, whereas synchronization at lower frequencies (theta, 5-10 Hz) remained intact.
Changes in spatial representations in Kcnq5 dn/dn mice. In freely moving mice, CA1 pyramidal cells displayed no significant changes of average firing rates, number of spikes per burst or intraburst discharge frequency between the genotypes (Table 1). However, Kcnq5 dn/dn CA1 pyramidal cells fired more bursts (Table 1), consistent with the apparent reduction of RisPyr cells, which display lower burst probability 36 . Firing of CA1 pyramidal cells as a function of the spatial position of the mouse was analyzed during exploration of a circular arena. Peak firing rate and spatial coherence of pyramidal cells' activity were largely similar between genotypes (Fig. 6a,b and Supplementary Table 2). However, Kcnq5 dn/dn mice displayed larger place field sizes and increased firing rates averaged over the entire arena ( Fig. 6b and Supplementary Table 2). Thus, our results indicate unaltered overall excitability but impaired spatial representations in the CA1 area in Kcnq5 dn/dn mice. By contrast, spatial firing of CA3 pyramidal cells did not differ between genotypes except for reduced peak firing rates in the mutant (Supplementary Fig. 9).

Discussion
We have used a genetic mouse model expressing a nonconducting dominant negative, trafficking-impaired KCNQ5 mutant to disrupt the function of KCNQ5-containing K þ channels in the brain and to investigate its physiological role using electrophysiological techniques both in vivo and in vitro. We found that KCNQ5 is an important modulator of synaptic inhibition in the CA3 area. Loss of KCNQ5 function changed fast network oscillations and modified spatial discharge of pyramidal cells in freely moving mice.
Constitutive gene disruption may entail secondary changes that can sometimes be compensatory. We cannot exclude that the loss of KCNQ5 channel function, or even the mislocalization of the nonfunctional protein, may have changed the expression of some genes in Kcnq5 dn/dn mice or may have had other secondary effects. However, the expression of KCNQ2 and KCNQ3 was unchanged in Kcnq5 dn/dn brain 13 , and here we found neither changes in Kcnq5 mRNA levels, nor in those of several Kir K þ ARTICLE channels. Moreover, the levels of several synaptic proteins were unaltered in Kcnq5 dn/dn brain and no change in CNS morphology could be detected. Importantly, our 'dominant-negative' strategy, in which all heteromeric channels containing one or more KCNQ5 subunit(s) are inactivated, avoided the possibility of the formation of more efficient KCNQ2/3 heteromers that may occur in Kcnq5 À / À mice. KCNQ2, -3 and -5 are broadly expressed in the brain and show robust expression in the hippocampus 13,19,32,33 . All three isoforms, as well as KCNQ2/3 and KCNQ3/5 heteromers, mediate K þ currents that slowly activate upon membrane depolarization. These channels are ideally suited to control neuronal excitability because they are partially open at resting voltages and can be inhibited by various G-protein-coupled receptors. Despite these similarities, KCNQ isoforms have different physiological roles, as illustrated by epilepsy which was described for mutations in KCNQ2 or KCNQ3, but not KCNQ5 (ref. 40).
Whereas slightly different biophysical properties may play a role in determining the divergent roles of KCNQ isoforms,  13 where we demonstrated its localization in pyramidal cells and interneurons. Even more important may be the divergent subcellular localization. KCNQ2/KCNQ3 channels are targeted to axon initial segments and nodes of Ranvier, structures involved in action potential generation and conduction, respectively. Neither subunit was detected at hippocampal synapses 32 . The localization of KCNQ2 and -3 at axon initial segments of excitatory neurons agrees excellently with their role in epilepsy. By contrast, we could not detect KCNQ5 at axon initial segments, but rather observed intense KCNQ5-positive puncta at inhibitory synapses in several brain regions. The relocation of the trafficking-deficient KCNQ5 mutant to pyramidal cell bodies, together with subcellular fractionation and STED microscopy strongly suggested a postsynaptic localization of WT KCNQ5 on pyramidal cells.
Others have described KCNQ5 in the presynaptic calyx of Held 22,23 . However, we found that KCNQ5 did not co-localize with markers for excitatory synapses at the presynaptic membrane of this giant synaptic terminal, but rather with gephyrin, a postsynaptic protein of inhibitory synapses, at the outer face of the calyx that directly receives inhibitory input 31 . However, KCNQ5 is also found in the postsynaptic membrane of vestibular calyx synapses 24 that receives excitatory input from ensheathed vestibular hair cells. Hence, in principle, KCNQ5 might be expressed in postsynaptic membranes of both inhibitory and excitatory synapses. We cannot rule out that excitatory synapses in the hippocampus express KCNQ5 at levels that are below the detection threshold of our labelling technique.
Likewise, the sensitivity of our antibody may not be high enough to detect the protein when it is uniformly distributed over  ARTICLE neuronal plasma membranes. Unambiguous detection of KCNQ5 in interneurons was only possible in Kcnq5 dn/dn mice in which the mutant protein accumulated in cell bodies. We suspect that KCNQ5 expression in interneurons is not limited to the rather small subset of cells identified by robust KCNQ5 accumulation because this fraction seems too small to account for the general increase in CA3 interneuron excitability in Kcnq5 dn/dn mice.
KCNQ5 dampens synaptic inhibition at both main cellular sites of expression in CA3, that is, at interneurons and pyramidal cells. Interneurons lacking functional KCNQ5 were hyperexcitable as evident from current injection experiments and from increased frequency of spontaneous IPSCs of their target cells. This hyperexcitability is most likely a direct, cell-autonomous effect of the loss of a KCNQ5-mediated plasma membrane conductance since excitability was assessed in the absence of synaptic transmission. This increase in excitability, which was found for all classes of interneurons investigated here, is consistent with the role of KCNQ channels in stabilizing the membrane potential and in spike-frequency adaptation 15 . Although KCNQ5 may additionally dampen synaptic inhibition by shunting inhibitory postsynaptic currents, our findings suggest that the increased excitability of interneurons is the predominant mechanism that increases inhibition in our mouse model. Ambient GABA is predominantly derived from synaptic vesicles released by interneurons 42 , which are more excitable in Kcnq5 dn/dn mice. Indeed, tonic GABA currents, which have been described previously in the CA3 region 43 , were enhanced in our mutant mice, suggesting that both tonic and phasic GABA contributes to the overall increase in inhibition in Kcnq5 dn/dn mice. Importantly, modulation of KCNQ5 activity appears to be physiologically relevant as muscarinic stimulation increased the frequency of spontaneous IPSCs in WT, but not in Kcnq5 dn/dn slices.
Unlike interneurons, pyramidal cells of Kcnq5 dn/dn mice showed no increased excitability, suggesting that KCNQ5, in contrast to KCNQ2/3 in CA1 neurons 17 , contributes little to the K þ conductance at the spike initiation zone or that the excitability of CA3 pyramidal cells is under homeostatic control. Importantly, increased firing rates of Kcnq5 dn/dn CA3 interneurons, but not of pyramidal cells, were also observed in vivo.
The activity of large neuronal populations is organized by hippocampal network oscillations, which are thought to be crucial for cognitive processes 34,37,44 . Loss of KCNQ5 function selectively impaired gamma (35-85 Hz) and ripple (140-200 Hz) oscillations without affecting slower oscillations in the theta band (5-10 Hz), a finding differing from those of mouse models targeting other proteins involved in hippocampal excitability [45][46][47][48] . During 35-85 Hz gamma oscillations CA1 networks follow the rhythm generated in the CA3 area 35 . Therefore, the changes in CA1 gamma oscillations of Kcnq5 dn/ dn mice likely result from changed synchronization in the CA3 area, where KCNQ5 expression is higher (see ref. 13 and present work). Network oscillations emerge from the coordinated interaction of neuronal excitation and inhibition 49,50 . In computational models, excitatory cells drive the activity of interneurons resulting in delayed feedback inhibition. Modulation of this inhibitory feedback, as well as the magnitude and temporal pattern of the excitatory drive, are key factors in shaping gamma oscillations 49 . These aspects of excitability directly depend on network inhibition, which is modified in Kcnq5 dn/dn mice. Furthermore, reduced AHP currents in the CA3 pyramidal cells in Kcnq5 dn/dn mice 13 might have also contributed to impairment of gamma-rhythmic firing 51 . Less irregular spiking of CA3 pyramidal cells (reduced burst firing, as observed here) counteracts gamma synchronization in a computational model 49 , in line with reduced gamma oscillations in Kcnq5 dn/dn mice. Optogenetic silencing of interneurons increases CA1 pyramidal cell burst firing in vivo 52 . Conversely, the increased excitability of CA3 interneurons might lead to decreased burst firing of Kcnq5 dn/dn CA3 pyramidal cells.
Increased inhibition in the CA3 area could also be a cause of the reduced occurrence of otherwise intact ripple oscillations in the Kcnq5 dn/dn CA1 area. Collective discharge of large neuronal populations in the CA3 area, so called sharp-wave bursts, provide excitatory drive required for generation of ripples in the CA1 area 39 (but see ref. 53). Increased inhibition in the CA3 of Kcnq5 dn/dn mice could lead to less frequent generation of sharpwaves, which, like other forms of population bursts in the CA3 area, rely on the excitability of the CA3 network 39,54 . Hence, both decreased gamma and ripple oscillations as well as decreased burst firing of CA3 pyramidal cells and reduced peak firing rates in CA3 place fields in Kcnq5 dn/dn mice likely arise from increased synaptic inhibition in the CA3 area.
Signalling from CA3 to CA1 area during gamma oscillations is believed to influence positional firing of pyramidal cells 36,55 . Chemical ablation of the CA3 area 56 or its functional genetic disconnection from the CA1 area 57 lead to less sharp 56 and enlarged 57 CA1 place fields. Kcnq5 dn/dn mice display enlarged CA1 place fields suggesting that gamma-frequency synchronization in the CA3 and CA1 areas may be crucial for precise spatial coding in CA1. In line with our results, enlarged place fields accompanied by memory deficits were observed after blocking the CA3 output 57 . Although our present in vitro findings point out functions of hippocampal KCNQ5 channels, other brain regions, which also express KCNQ5, might to some extent also contribute to the observed change in place fields. Although it seems likely that KCNQ5 dampens excitation not only in hippocampus, but also in other brain regions, such a generalization will require experimental verification.
Using a genetic approach we identified specific roles of KCNQ5 that set this isoform apart from KCNQ2 and -3, both of which are also widely expressed in brain. A major difference between these KCNQ isoforms is their global role in network excitability. Both KCNQ2 and -3 are mutated in human neonatal epilepsy 3,4,58 , and mouse models with reduced KCNQ2 or KCNQ3 function display reduced seizure thresholds, epileptic discharges or overt epilepsy 17,59,60 . By contrast, Kcnq5 dn/dn mice neither displayed seizures 13 nor epileptic discharges in vivo. This difference may be explained by differences in expression patterns. Loss of KCNQ2/3 function increases network excitability because these channels are prominently localized at spike initiation zones of excitatory neurons, whereas loss of KCNQ5 rather dampens hippocampal excitability because it is expressed in interneurons and at inhibitory synapses. As a consequence, Kcnq5 dn/dn mice did not exhibit the strong neurological phenotype observed in KCNQ2/3 mouse models.
Immunohistochemistry. Immunohistochemistry was performed on paraffin sections as described previously 13 . Nuclei were stained with 4 0 ,6-diamidino-2phenylindole (DAPI STED microscopy. For distance measurements between KCNQ5 and bassoon, gephyrin and homer, respectively, immune-labelled paraffin sections (8 mm) were mounted in 2 0 ,2 0 -thiodieethanol (TDE). The sections were immersed with increasing TDE concentrations in PBS buffer (10, 25, 50, 97%, 10 min each) and finally embedded in 97% TDE. STED imaging was performed by using a commercial two-channel STED microscope (Leica Microsystems) as described previously 61 . Briefly, the system consists of an inverted microscope (DMI 6000 CS with a Tandem Scanning System SP5) equipped with two pulsed excitation lasers of 532 and 635 nm (E70 ps pulse width, 80 MHz repetition rate; PicoQuant) and a Titanium saphire laser (Chameleon ultra II (E100 fs pulse width, 80 MHz repetition rate), Coherent) as depletion light source. STED-stacks, consisting of at least six slices were acquired sequentially with a Â 100HCX PL APO CS oil objective (Leica Microsystems). STED-depletion was done at 730 nm (Abberiordye) and at 750 nm (Atto-dye) Scanning-format of 1,024 Â 1,024 and zoom of six resulted in a pixel size of 25, 2 nm (x,y) and 126 nm (z). Scan speed was set to 700 Hz with line averaging of 48.
Fluorescence signals were detected sequentially by avalanche photo diodes (Perkin Elmer Inc.) within two spectral regions (589-625 nm for Abberior STAR 580) and 655-685 nm for Atto 647N) separated by a dichroic beam splitter at 650 nm. To enhance focus stability and minimize thermal drift, the microscope was housed in a light tight customized heatable incubation chamber (Pecon). Following image acquisition, the raw data were deconvolved using Huygens Professional 4.4 (Scientific Volume Imaging B.V) software. Deconvolution was done by using the pulsed STED optimized function to calculate a theoretical point spread function out of the specific microscope parameter. Default deconvolution parameters were applied. To measure the distances between KCNQ5 and Bassoon, Gephyrin or Homer, respectively, slices from the centre of the stack were chosen by using the Huygens Twin Slicer. A line profile plot from peak to peak was created manually and distance values registered. Subsequent statistical analysis was done with GraphPad Prism 5 (GraphPad Software, Inc.). Further image processing was done by using ImageJ (National Institutes of Health) and Photoshop CS 5 (Adobe System Inc.).
Tissue preparation for electrophysiology. Mice of age 4-6 weeks were deeply anaesthetized with isoflurane, killed by decapitation and brains were removed. Horizontal hippocampal slices (400 mm) were cut using a Leica VT1200S vibratome. Different preparation solutions were used. For all recordings, the tissue was prepared in ice-cold, saccharose-based artificial cerebrospinal fluid (ACSF) (in mM): NaCl 87, Na 2 PO 4 1.1, NaHCO 3  Field potential recordings. Recording and stimulation electrodes (2-4 MO) were filled with ASCF and placed in the st. radiatum of the CA3 area. fEPSPs were evoked at 0.1 Hz. Only experiments with clearly detectable fibre volleys were used for analysis. During input-output measurements, the stimulus strength was adjusted and at least three independent measurements for every stimulation step were performed in order to achieve the desired fibre volley amplitudes. To minimize the error caused by potential population spike contamination in all experiments, we analyzed the slope of the fEPSP instead of the fEPSP amplitude. Furthermore, all traces with unusual time course of the initial slope were excluded from analysis.
Patch-clamp recordings. Recordings were performed in the whole-cell configuration. Hippocampal pyramidal neurons (CA3/CA1) were identified with differential interference contrast microscopy. Interneurons were identified by GFP fluorescence in mice expressing the GAD67-GFP allele 26 in a heterozygous state. Recordings were performed in interneurons from st. oriens, st. pyramidale and st. radiatum. In recordings from st. oriens and st. pyramidale, interneurons were filled with biocytin (0.3-0.5%) to allow post-hoc anatomical reconstruction of axonal arborization and subsequent classification into oriens-lacunosum-moleculare cells, bistratified cells and basket cells (see below). Interneurons from st. radiatum were not filled with biocytin and pooled into one group for analysis.
Pipettes were pulled from borosilicate glass (Hilgenberg). Pipette resistances were B4-6 MO. Patch-clamp recordings were performed with MultiClamp 700B, a Digidata 1,440 A interface and PClamp software (Molecular Devices). Signals were low-pass filtered at 3 kHz and digitized at 10 kHz. Recordings with series resistance 430 MO were discarded.
Statistical analysis of in vitro data. Unless otherwise stated, Student's t-test (paired and non-paired) or analysis of variance (SPSS) were used where appropriate. Statistical significance level was set to Po0.05 and marked in figures by asterisks (*). All data were analyzed using Clampfit (Molecular Devices), Prism (GraphPad Software) and Matlab (MathWorks). Spontaneous events were analyzed with MiniAnalysis (SynaptoSoft Inc.) using a threshold-based search paradigm and visual post hoc inspection. Histograms of synaptic events were derived from at least 200 events per cell.
Slice processing and reconstruction of biocytin-filled cells. For anatomical reconstruction, cells were loaded with 0.3-0.5% biocytin. After finished recordings, slices were transferred to 4% paraformaldehyde for overnight fixation and were then maintained in 0.1 M phosphate buffer. For histological processing, the slices were re-sectioned (100 mm). Biocytin-filled cells were subsequently visualized with 3,3 0 -diaminobenzidine tetrahydrochloride (0.015%) using a standard ABC kit (Vectorlabs, Burlingame, CA, USA) and reconstructed on a light microscope at Â 40 or Â 60 with a Neurolucida 3D reconstruction system (MicroBrightField).