A limited role of NKCC1 in telencephalic glutamatergic neurons for developing hippocampal network dynamics and behavior

Significance

Neuronal circuits develop in an activity-dependent manner. In vitro data suggest that the intracellular accumulation of chloride—a universal characteristic of immature nerve cells—drives network maturation through a depolarizing action of the neurotransmitter GABA. We here demonstrate that deletion of the chloride cotransporter NKCC1 from forebrain pyramidal cells severely impairs hippocampal network synchrony in vitro. In contrast, NKCC1 has weak and event type-dependent effects on spontaneous network activity in vivo, and loss of NKCC1 leaves the maturation of synaptic properties, network dynamics, and hippocampus-dependent behaviors largely unaffected. Our data reveal a subtle neural network function of NKCC1 in hippocampal glutamatergic neurons in vivo, but challenge the assertion that NKCC1 in this major cell type is central to hippocampal development.

Abstract

NKCC1 is the primary transporter mediating chloride uptake in immature principal neurons, but its role in the development of in vivo network dynamics and cognitive abilities remains unknown. Here, we address the function of NKCC1 in developing mice using electrophysiological, optical, and behavioral approaches. We report that NKCC1 deletion from telencephalic glutamatergic neurons decreases in vitro excitatory actions of γ-aminobutyric acid (GABA) and impairs neuronal synchrony in neonatal hippocampal brain slices. In vivo, it has a minor impact on correlated spontaneous activity in the hippocampus and does not affect network activity in the intact visual cortex. Moreover, long-term effects of the developmental NKCC1 deletion on synaptic maturation, network dynamics, and behavioral performance are subtle. Our data reveal a neural network function of NKCC1 in hippocampal glutamatergic neurons in vivo, but challenge the hypothesis that NKCC1 is essential for major aspects of hippocampal development.

Intracellular chloride concentration ([Cl⁻]_i) is a major determinant of neuronal excitability, as synaptic inhibition is primarily mediated by chloride-permeable receptors (1). In the mature brain, [Cl⁻]_i is maintained at low levels by chloride extrusion, which renders γ-aminobutyric acid (GABA) hyperpolarizing (2) and counteracts activity-dependent chloride loads (3). GABAergic inhibition in the adult is crucial not only for preventing runaway excitation of glutamatergic cells (4) but also for entraining neuronal assemblies into oscillations underlying cognitive processing (5). However, the capacity of chloride extrusion is low during early brain development (6, 7). Additionally, immature neurons are equipped with chloride uptake mechanisms, particularly with the Na⁺/K⁺/2Cl⁻ cotransporter NKCC1 (8–12). NKCC1 contributes to the maintenance of high [Cl⁻]_i in the developing brain (13), favoring depolarization through GABA_A receptor (GABA_AR) activation in vivo (14, 15).

When GABA acts as a depolarizing neurotransmitter, neural circuits generate burst-like spontaneous activity (16–20), which is crucial for their developmental refinement (21–24). In vitro evidence indicates that GABAergic interneurons promote neuronal synchrony in an NKCC1-dependent manner (10, 12, 25–28). However, the in vivo developmental functions of NKCC1 are far from understood (29, 30). One fundamental question is to what extent NKCC1 and GABAergic depolarization supports correlated spontaneous activity in the neonatal brain. In the neocortex, GABA imposes spatiotemporal inhibition on network activity already in the neonatal period (14, 25, 31, 32). Whether a similar situation applies to other brain regions is unknown, as two recent chemo- and optogenetic studies in the hippocampus yielded opposing results (25, 33). Manipulations of the chloride driving force are potentially suited to resolve these divergent findings, but pharmacological (34–36) or conventional knockout (10, 11, 37) strategies suffer from unspecific effects that complicate interpretations.

Here, we overcome this limitation by selectively deleting Slc12a2 (encoding NKCC1) from telencephalic glutamatergic neurons. We show that chloride uptake via NKCC1 promotes synchronized activity in acute hippocampal slices, but has weak and event type-dependent effects in CA1 in vivo. Long-term loss of NKCC1 leads to subtle changes of network dynamics in the adult, leaving synaptic development unperturbed and behavioral performance intact. Our data suggest that NKCC1-dependent chloride uptake is largely dispensable for several key aspects of hippocampal development in vivo.

Results

Behavioral Performance of Mice in the Conditional Absence of NKCC1 from Telencephalic Glutamatergic Cells.

To investigate the role of depolarizing GABA in telencephalic glutamatergic neurons, we deleted Slc12a2 in Emx1-lineage cells of embryonic mice (Fig. 1A). A comparison with NKCC1^flox/flox (wild-type [WT]) littermates showed that Cre-dependent recombination had been widespread in the hippocampus (Fig. 1B) and profoundly reduced hippocampal Slc12a2 mRNA levels in Emx1^IREScre:NKCC1^flox/flox (KO^Emx1) mice at postnatal days (P) 2 to 3 (Fig. 1C and SI Appendix, Table S1). Fate mapping of Emx1-lineage cells at P54 to 63 confirmed that Cre expression was restricted to glutamatergic neurons (GAD67^–/NeuN⁺), whereas cells expressing a pan-GABAergic (GAD67) or subtype-selective GABAergic interneuron (parvalbumin or somatostatin) marker were reporter negative (SI Appendix, Fig. S1 A–J). WT and KO^Emx1 mice were born at Mendelian ratios (P = 0.41, χ² = 0.69, n = 71 mice, χ² test), displayed a normal postnatal gain in body weight (SI Appendix, Fig. S1K), similar brain growth (measured by lambda-bregma distance, SI Appendix, #S14 in Table S1), and unperturbed developmental reflexes, including cliff avoidance behavior and righting reflex (SI Appendix, Fig. S1 L and M). Unlike conventional NKCC1 knockout mice (38), adult KO^Emx1 mice exhibited unaltered locomotion and exploration in the open field (Fig. 1 D and E). Spontaneous alternations in a Y maze, a measure of spatial working memory, were unaffected (Fig. 1 F and G). Spatial learning and memory in a Morris water maze (Fig. 1 H–K), as well as cue- and context-specific fear memory (Fig. 1 L–N), were not significantly altered either. In sum, KO^Emx1 mice lack the severe phenotype of NKCC1 null mice, which is due to NKCC1 loss from tissues other than the brain (37), and display a normal development of several hippocampus-dependent and -independent behaviors.

Fig. 1.

Behavioral performance of NKCC1 KO^Emx1 mice. (A) In the targeted Slc12a2 locus, exons 8 to 10 are flanked by loxP sites. Cre-dependent recombination causes a frameshift (dotted) and introduces a stop codon in exon 12. (B) Confocal image demonstrating Cre-reporter expression (GCaMP3) in a horizontal brain slice (P4). (C) Hippocampal Slc12a2 mRNA levels compared with the geometric mean of WT, normalized to Gapdh and Hmbs. (D) Sample trajectories in the open field. (E) Total distance covered and relative time spent in the center. (F) Sample trajectories in the Y maze. (G) Spontaneous alternations (dotted, chance level). (H) Morris water maze. (I and J) Distance to platform for cued trial (visible platform) and acquisition (hidden platform). (K) Time in target quadrant during probe trial (no platform). (L–N) Freezing time during fear conditioning (L), reexposure to cue (M), and context (N). (C–K) Open circles represent single animals. (E–N) Mean ± SEM; CS, conditioned stimulus; US, unconditioned stimulus; ***P < 0.001. See SI Appendix, Fig. S1 and Table S1.

Loss of NKCC1 Attenuates the Depolarizing Action of GABA.

To investigate how NKCC1^Emx1 loss alters cellular GABA actions in the neonatal hippocampus, we used low-gramicidin perforated-patch current-clamp recordings that allow monitoring the membrane potential (V_m), while minimizing the risk of membrane breakthrough (39, 40). Measurements were performed in slices (P3 to 4) in the presence of antagonists of ionotropic glutamate receptors (10 µM 6,7-dinitroquinoxaline-2,3-dione [DNQX], 50 µM DL-2-amino-5-phosphonopentanoic acid [APV]) and voltage-gated Na⁺ (0.5 µM tetrodotoxin [TTX]) and Ca²⁺ (100 µM CdCl₂) channels so as to block recurrent excitation and amplification of voltage changes by intrinsic conductances. Resting V_m of CA3 pyramidal cells (PCs) was similar in WT and KO^Emx1 mice (Fig. 2 A and B and SI Appendix, Table S2). However, NKCC1 deletion significantly reduced the peak membrane potential (V_peak) induced by a saturating puff of the GABA_AR agonist isoguvacine (100 µM, 2 s) by almost 10 mV (Fig. 2B). Because V_peak will approximate the reversal potential of GABA_AR-mediated currents (E_GABA) under these conditions, our data imply lower [Cl⁻]_i in NKCC1 KO^Emx1 mice. Isoguvacine-induced V_m changes were consistent across trials (Fig. 2C). Moreover, amplitudes of somatic Ca²⁺ transients (CaTs) evoked by bath-applied GABA (100 µM) were attenuated in putative PCs of KO^Emx1 mice (P2 to 4), whereas glutamate-induced CaTs were unaffected (SI Appendix, Fig. S2). Furthermore, cell-attached recordings in the presence of DNQX (10 µM) and APV (50 µM) revealed that puff-applied isoguvacine (100 µM) increased action potential (AP) frequency in WT, but not in KO^Emx1, mice at P3 to 4 (Fig. 2 D–F). We conclude that NKCC1 deletion attenuates GABA_AR-mediated depolarization and virtually abolishes excitatory GABA effects in vitro.

Fig. 2.

Loss of NKCC1 attenuates the depolarizing action of GABA. (A) Sample gramicidin perforated-patch current-clamp recordings in response to puff-applied isoguvacine (Iso). (B) Resting (V_rest) and peak isoguvacine-induced (V_peak) membrane potential. (C) Correlation of isoguvacine-induced membrane potential changes (ΔV_m) in two successive trials. ρ, Spearman’s rank correlation coefficient. (D) Sample cell-attached recordings in response to puff-applied isoguvacine. (E) Number of action currents (300-ms intervals) immediately before (base) and after (Iso) puff onset. (F) Reduced fraction of responsive cells (resp) in KO^Emx1 mice. Open circles represent single cells. Mean ± SEM; ns, not significant; *P < 0.05, ***P < 0.001. See SI Appendix, Fig. S2 and Table S2.

PC-Specific NKCC1 Deletion Impairs Spontaneous Hippocampal Network Activity In Vitro.

Based on experimental and theoretical results (10, 27, 35), we hypothesized that a PC-specific attenuation of GABAergic depolarization would impair network synchrony. We tested this prediction in the CA3 of acute slices (P1 to 4) using whole-cell recordings of spontaneous GABAergic postsynaptic currents (sGPSCs) isolated by their reversal potential (Fig. 3A). sGPSC frequencies did not significantly differ between genotypes (Fig. 3B and SI Appendix, Table S3). In WT slices, sGPSCs typically occurred in bursts (known as giant depolarizing potentials [GDPs]) (41). However, this temporal structure was largely absent in cells from KO^Emx1 mice. For quantification, a burst index was computed for measured data and compared with a Poisson point process in which sGPSCs occur randomly. Burst indices of WT, but not of KO^Emx1 cells were consistently higher than those expected for a random case (Fig. 3C). Controlling for the effect of sGPSC frequency, we found that burst indices were significantly reduced by NKCC1^Emx1 loss (Fig. 3D), indicating impaired neuronal synchrony. This conclusion was confirmed by single-cell confocal Ca²⁺ imaging revealing a dramatic reduction in GDP frequency in CA3 of KO^Emx1 mice at P1 to 4 (SI Appendix, Fig. S3). Of note, the effects of NKCC1^Emx1 loss were region dependent because the kinetics of single-cell recruitment in CA1 during GDPs was significantly slowed down in KO^Emx1 mice, while GDP frequency was similar (SI Appendix, Fig. S3).

Fig. 3.

NKCC1^Emx1 deletion impairs spontaneous hippocampal activity in vitro. (A) Sample whole-cell recordings of sGPSCs isolated by reversal potential. (B) sGPSC frequency. (C) Burst index vs. sGPSC frequency. The dotted line represents a Poisson point process. (D) Burst indices. (E) GCaMP-based confocal imaging of spontaneous activity in corticohippocampal slices. Top: GCaMP3 fluorescence overlaid with ROIs for analysis (Left) and sample ΔF/F₀ images for time periods indicated below. (Scale bars, 500 µm.) Bottom: ΔF/F₀ sample traces. (F and G) Total active time (F) and mean area under the curve per event (AUC, G). Open circles represent single cells (B–D) or slices (F and G). Mean ± SEM; ns, not significant; *P < 0.05, **P < 0.01, ***P < 0.001. See SI Appendix, Figs. S3 and S4 and Table S3.

In large-scale confocal Ca²⁺ imaging experiments (GCaMP3^Emx1) at P1 to 4, large-amplitude network events were detected in slices from WT (Emx1^IREScre:NKCC1^wt/wt:GCaMP3^LSL) mice, often involving multiple hippocampal subfields and/or both hippocampal and adjacent cortical areas (Fig. 3E). When CA1 was coactive, it stereotypically followed activity in CA3 or the medial entorhinal cortex (SI Appendix, Fig. S4). While large-scale activity was also present in KO^Emx1 (Emx1^IREScre:NKCC1^flox/flox:GCaMP3^LSL) slices, the area under the curve of network events was substantially reduced (Fig. 3G). Additionally, a reduced active time was found in CA3 (Fig. 3F). We thus conclude that PC-specific NKCC1 deletion leads to a profound attenuation of neuronal synchrony in vitro.

NKCC1 Effects on Network Dynamics Are Not Accounted for by Alterations in Intrinsic Excitability or Basic Synaptic Properties.

The question of how GABA instructs neurodevelopment remains largely unanswered (29). Some data indicate that a lack of NKCC1 interferes with the normal maturation of synaptic (9) or intrinsic (11) conductances. We therefore examined whether altered network dynamics in NKCC1 KO^Emx1 mice reflect an acute shift in E_GABA or developmental changes following NKCC1 disruption in embryonic life. As an overall measure of intrinsic excitability, we first assessed spontaneous AP firing in the presence of synaptic receptor antagonists (10 µM DNQX, 50 µM APV, 10 µM bicuculline methiodide [BMI]) using cell-attached recordings from CA3 PCs (P3 to 4). In contrast to results from NKCC1 null mice (11), AP frequencies were unaffected by conditional NKCC1 loss (Fig. 4 A and B and SI Appendix, Table S4). Moreover, NKCC1^Emx1 disruption did not significantly alter neuronal input–output relationships (Fig. 4 C and D). Accordingly, passive membrane properties (SI Appendix, #S10–S12 in Table S4) and AP threshold (Fig. 4E) were unaffected.

Fig. 4.

NKCC1 effects on hippocampal network dynamics are not accounted for by alterations in intrinsic excitability or basic synaptic properties. (A) Sample cell-attached recordings in the presence of ionotropic glutamate and GABA receptor antagonists. (B) Mean spontaneous AP frequency. (C) Sample current-clamp recordings in response to current injections. (D and E) Current–frequency relationship (D) and AP threshold (E) are unaltered. (F) Sample voltage-clamp measurements of mEPSCs. (G) mEPSC frequency, median mEPSC amplitude, and mean mEPSC decay-time constant. (H) Sample voltage-clamp measurements of mGPSCs. (I) mGPSC frequency, median mGPSC amplitude and mean mGPSC decay-time constant. Open circles represent single cells. Mean ± SEM; ns, not significant; ***P < 0.001. See SI Appendix, Table S4.

We next investigated whether basic properties of glutamatergic/GABAergic synapses were altered upon NKCC1^Emx1 disruption. Whole-cell recordings from CA3 PCs were performed at P3 to 4 and P14 to 16, covering a time period of intense synaptogenesis and subunit reorganization of postsynaptic receptors. AMPA receptor-dependent miniature PSCs (mEPSCs) displayed a profound developmental increase in frequency, accompanied by an increase in (quantal) amplitude and a deceleration in decay kinetics (Fig. 4 F and G) (42). These alterations were independent of genotype (SI Appendix, #S4–S6, Table S4). GABA_AR-mediated miniature postsynaptic currents (mGPSCs) also underwent a substantial developmental increase in frequency (Fig. 4 H and I, Left), along with an acceleration in their decay kinetics (Fig. 4 I, Right) (43), but no change in quantal amplitude (Fig. 4 I, Middle). These developmental trajectories were indistinguishable between WT and KO^Emx1 mice (SI Appendix, #S7–S9, Table S4). In summary, both intrinsic excitability and basic synaptic properties were unaffected by NKCC1^Emx1 loss and, thus, are unlikely to account for the alteration of in vitro hippocampal network dynamics in KO^Emx1 mice.

Hippocampal Sharp Waves In Vivo Persist in the Conditional Absence of NKCC1.

Next, we investigated the effect of NKCC1^Emx1 deletion on in vivo hippocampal network multiunit activity (MUA) at P4. We recorded local field potentials (LFPs) along the CA1–dentate gyrus axis using linear 32-site silicon probes in awake head-fixed neonates. We observed similar activity patterns in P4 hippocampi of both WT and NKCC1 KO^Emx1 mice: sharp waves (SPWs) with a phase reversal in CA1 stratum pyramidale (s.p.) and short-lasting oscillations in the 10 to 30 Hz (beta) frequency range (hippocampal network oscillations [HNOs]) (Fig. 5 A and B) (36, 44). Depth profiles of SPWs were comparable to those in rats (45) (“rad/lm SPW” in Fig. 5C). Additionally, we found inverted SPWs with their maximum current sink in stratum oriens (s.o.) (“oriens SPW” in Fig. 5C). Both SPW types elicited distinct MUA patterns in the CA1 pyramidal layer that were similar for both genotypes. None of these SPW event properties such as occurrence frequency, amplitudes, or event durations differed between WT and KO^Emx1 animals (Fig. 5D and SI Appendix, Table S5). Likewise, SPW properties did not differ between genotypes at P7 or P12 (SI Appendix, Fig. S5). MUA counts in 50-ms bins either centered around, or aligned to, 0 ms for oriens and rad/lm SPWs, respectively, also did not differ by genotype (Fig. 5D and SI Appendix, Table S5). Furthermore, HNO rate and length were similar between the genotypes at P4 (Fig. 5E). HNO MUA shape was also comparable between genotypes, and MUA spike counts in the first 500-ms bin—although highly variable across animals—did not differ between WT and KO^Emx1 mice (Fig. 5E).

Fig. 5.

Hippocampal sharp waves in vivo at P4 persist in NKCC1 KO^Emx1 mice. (A) Example SPW (red arrow) and beta-frequency (∼10- to 30-Hz range) hippocampal network oscillation (HNO, blue line) recorded by linear silicon probes. LFP traces (black) are superimposed over the current source density (CSD). 0 µm represents the field reversal of s.p. (B) Example wavelet spectrograms of CSD in s.l.m. in median absolute deviation (MAD) units per frequency. Events marked as in A and B. Note that some HNOs identified in the LFP may be of too small amplitude to be clearly visible in the wavelet of the CSD. (C) Median LFP and CSD depicted from SPWs with sinks either below (“rad/lm,” Left) or above (“oriens,” Right) s.p. Left two panels are from a WT, and Right two are from a KO^Emx1 mouse. Lower panels display the SPW trough-aligned average MUA for all sets with five or more events. (D) Frequency, LFP amplitude, and MUA spike count of rad/lm and oriens SPWs. Displayed MUA counts are from 50-ms bins centered around the peak at 0 [−25, 25] ms for oriens SPWs, and after the peak from 0 [0, 50] ms for rad/lm SPWs. (E, Left) HNO start-aligned average MUA. Note the −200-ms peak in MUA due to (the ∼30 to 60% of) HNO events having a preceding SPW (c.f. A and B). Note also that the MUA decay results from averaging HNOs of varying length. Right three panels, HNO occurrence rate, duration and MUA count in the [0, 500]-ms bin. Open circles represent single animals. Mean ± SEM. See SI Appendix, Fig. S5 and Table S5.

NKCC1 Promotes In Vivo Correlated Network Activity in the Hippocampus, but Not in the Visual Cortex.

We further characterized in vivo network dynamics using wide-field Ca²⁺ imaging, which primarily reflects AP firing and allows recording from a large field of view (FOV). For optical access to CA1 in vivo, a hippocampal window preparation (46) was adapted for use in neonatal mice (P3 to 4; Fig. 6A). LFP recordings using a tungsten electrode positioned in stratum radiatum (s.r.) confirmed that SPW frequencies measured through the intact cortex or after focal cortex aspiration did not differ significantly (SI Appendix, Fig. S6 A–C and Table S6). Ca²⁺ imaging was then performed in mice expressing GCaMP3 in Emx1-lineage cells (Fig. 6B). Under N₂O anesthesia, network activity occurred in the form of spatiotemporal clusters of CaTs, frequently recruiting large parts of the FOV (∼0.6 mm², n = 22; Fig. 6 B–E). Simultaneous LFP recordings revealed that approximately two-thirds of Ca²⁺ clusters were temporally unrelated to SPWs (−SPW; Fig. 6 C–E and G and SI Appendix). About one-fourth of −SPW-related LFP epochs had a relative beta (12 to 30 Hz) power significantly higher than that of cluster-free LFP periods (SI Appendix, Fig. S6D and #S18 in Table S6), suggesting that HNOs and −SPW Ca²⁺ clusters constitute partially overlapping event classes. However, most −SPW-related LFP epochs either lacked a clear electrophysiological signature or displayed inconsistent LFP changes different from SPWs (Fig. 6 C–E and SI Appendix, Fig. S6D). Of note, many beta-negative −SPW Ca²⁺ clusters were spatially confined and did not overlap with the LFP electrode (SI Appendix, Fig. S6D). In vivo two-photon Ca²⁺ imaging in WT mice revealed that only 13.1 ± 1.0% of somatic CaTs in CA1 PCs cooccurred with SPWs. Moreover, 14.2 ± 0.8% of CaTs followed SPWs with short (<3 s) latency, presumably reflecting SPW tails (16), whereas 72.8 ± 1.2% of CaTs were temporally unrelated to SPWs (SI Appendix, Fig. S6 G–I). Importantly, wide-field imaging confirmed that both +SPW and −SPW Ca²⁺ clusters were entirely AP dependent (SI Appendix, Fig. S6 E and F).

Fig. 6.

NKCC1 promotes correlated network activity in an event type- and region-specific manner in vivo. (A) Experimental design. EC, extracellular electrode. (B) GCaMP fluorescence and ROIs for analysis. (C) Time-aligned GCaMP fluorescence (ROIs indicated in B), LFP (0.5 to 100 Hz) and respiration/movement (resp) signal. Ca²⁺ clusters are indicated by gray, SPWs by red arrows. Note CaTs with or without a concurrent SPW. (D) Boxed regions in C at higher magnification. +SPWs CaTs display faster rise kinetics and higher amplitudes. (E) Raster plots indicating CaT times per ROI. Insets: GCaMP fluorescence (F₀) and binary area plots of Ca²⁺ clusters indicated on the Left. (F) Reduced CaT frequencies per ROI. (G) NKCC1^Emx1 deletion reduced the frequency of –SPW Ca²⁺ clusters. (H–J) Rise kinetics (H), cluster area (I), and amplitude (J). SI, sampling interval. (K) Sample ΔF/F₀ traces (mean of all analyzed ROIs), LFP (0.5 to 100 Hz), and raster plots before/after superfusion of gabazine (40 µM). Red arrows indicate detected SPWs (control) and SPW-like events (gabazine). (L) +SPW and –SPW Ca²⁺ cluster frequencies. (M) Sample raster plots demonstrating cluster activity in the visual neocortex. Insets: Transcranial GCaMP fluorescence overlaid with binary area plots of three events indicated on the Left. (Scale bars, 200 µm.) (N–P) Average CaT frequency per ROI (N), mean cluster area (O), and cluster frequency (P). Open circles represent single animals. Mean ± SEM; ns, not significant; *P < 0.05, ***P < 0.001. See SI Appendix, Figs. S6–S8 and Table S6.

For quantification of Ca²⁺ imaging data, the FOV was subdivided into a regular grid of regions of interest (ROIs) (Fig. 6B). We found that the mean frequency of spontaneous CaTs per ROI was significantly lower in KO^Emx1 mice than in WT littermates (Fig. 6F and SI Appendix, Table S6). Likewise, Ca²⁺ cluster frequency was reduced from 7.9 ± 0.4 min⁻¹ in WT to 6.3 ± 0.4 min⁻¹ in KO^Emx1 mice (SI Appendix, #S2 in Table S6). Thus, our data identify an excitatory network effect of NKCC1^Emx1. While there was no significant interaction between the independent variables (genotype × SPW), the frequency of −SPW, but not of +SPW events was significantly reduced (Fig. 6G). The frequency of LFP-SPWs was not altered either, in line with the observation that >90% of SPWs resulted in a Ca²⁺ cluster (SI Appendix, #S3 in Table S6). As compared with +SPW Ca²⁺ clusters, −SPW events had slower rise kinetics (Fig. 6H), smaller cluster areas (Fig. 6I), and lower CaT amplitudes (Fig. 6J), pointing to a separate class of events. Interestingly, no genotype-dependent difference was found in any of these parameters (SI Appendix, #S4–S6 in Table S6), demonstrating that network dynamics in NKCC1 KO^Emx1 mice qualitatively resembled those observed in WT mice.

Depolarizing GABA can mediate both excitation and inhibition in a context-dependent manner (47–49). We therefore examined the contribution of endogenously released GABA by locally blocking GABA_ARs. Superfusion of the hippocampal window with gabazine (40 µM) induced a significant increase in LFP-SPW frequencies, amplitudes, and durations (SI Appendix, #S9–S11 in Table S6), indicating that SPW-like events in the presence of gabazine may be mechanistically different from SPWs in control. Furthermore, gabazine increased the frequency of +SPW Ca²⁺ clusters, but decreased that of −SPW events (Fig. 6 K and L). Additionally, amplitudes of both event classes increased following GABA_AR block, but this effect was significantly stronger for +SPW Ca²⁺ clusters (SI Appendix, #S8 in Table S6). Effects of GABA_AR inhibition on −SPW Ca²⁺ clusters were similar to those on network events in vitro, where gabazine (20 µM) induced a profound increase in amplitude, but not frequency (SI Appendix, Fig. S7 A–D). Collectively, our data suggest that GABA can facilitate −SPW Ca²⁺ clusters, while additionally imposing inhibitory constraints on spontaneous network activity.

Systemic application of the NKCC1 inhibitor bumetanide was previously found to abolish SPWs (35) or leave SPWs unaffected (36). Bypassing the blood–brain barrier (34), we aimed at blocking NKCC1 by superfusing the hippocampal window with bumetanide (50 µM) for a prolonged time period (60 min) (SI Appendix, Fig. S7E). Surprisingly, we observed moderate increases in the frequencies of CaTs as well as +SPW and −SPW Ca²⁺ clusters. These changes occurred in both genotypes (SI Appendix, Fig. S7 F–H and #S23–S25 in Table S6). While the activity increase might reflect cell-type-specific effects of NKCC1 and/or time-dependent alterations, this dataset supports the view that NKCC1 in PCs is not instrumental for SPW generation.

In contrast to our results in hippocampus, previous electrophysiological (36, 50) or optical (14) measurements did not reveal a contribution of NKCC1 to network activity in neocortex. As this might imply brain region-dependent differences in the role of NKCC1 and, thus, depolarizing GABA (33), we performed additional experiments in the visual cortex (P3 to 4). Cell-attached recordings in acute slices confirmed that puff-applied isoguvacine induced AP firing in PCs from WT, but not from KO^Emx1 mice (SI Appendix, Fig. S8 A–C). Moreover, Slc12a2 mRNA levels in the visual cortex of neonatal NKCC1 KO^Emx1 mice were strongly reduced (SI Appendix, Fig. S8D). Using in vivo wide-field Ca²⁺ imaging, we found that spontaneous network activity was unaffected by NKCC1 loss at P3 to 4 (Fig. 6 M–P) and P9 to 10 (SI Appendix, Fig. S8 E–H). Furthermore, multisite LFP recordings revealed that spindle bursts, the main electrophysiological signature of Ca²⁺ clusters in the neonatal visual cortex (14), were unaltered in KO^Emx1 mice (SI Appendix, Fig. S8 I–K).

Collectively, our data indicate that PC-specific expression of NKCC1 has a minor effect on in vivo correlated spontaneous activity (−SPW) in the hippocampus, whereas network activity in visual cortex is independent of NKCC1.

Long-Term Alterations of In Vivo Hippocampal Network Dynamics upon NKCC1^Emx1 Deletion.

As correlated spontaneous activity contributes to the maturation of neuronal circuits (51, 52), we next investigated the long-term developmental effects of NKCC1^Emx1 deletion onto in vivo hippocampal network dynamics in urethane-anesthetized 13- to 14-wk-old mice. We recorded depth profiles of spontaneous LFPs along the CA1–dentate gyrus axis. Hippocampal LFPs from both WT and NKCC1 KO^Emx1 mice showed paradoxical (rapid eye movement [REM]) sleep-like activity, i.e., theta/gamma oscillations and non-REM slow wave sleep (SWS)-like LFP epochs with the characteristic SPW-ripple (SPW-R) complexes in CA1. These activity patterns alternated spontaneously throughout the recording session (Fig. 7A). LFP analysis of theta power depth profiles during REM-like epochs with similar theta phase depth profiles (SI Appendix, Fig. S9 A and B) revealed increased theta amplitudes in stratum lacunosum moleculare (s.l.m.) and stratum moleculare (s.m.), whereas amplitudes of low-gamma and MUA frequency bands were unaffected (Fig. 7 B and C and SI Appendix, Table S7). Analysis of SPW-Rs during SWS-like LFP epochs revealed lower ripple spectral frequencies, unchanged SPW amplitudes, and decreased occurrence rates (Fig. 7D). Cross-frequency coupling between theta and gamma oscillations, i.e., the modulation of gamma amplitudes by theta phase, was unaltered in KO^Emx1 mice (SI Appendix, Fig. S9 C and D). In addition, peak spectral frequencies in the theta (3 to 7 Hz) or gamma (30 to 100 Hz) range were similar in both genotypes (SI Appendix, Fig. S9E). Together, our data indicate that NKCC1 deletion from forebrain PCs altered the properties of theta oscillations in hippocampal layers receiving entorhinal input as well as intrahippocampal dynamics during ripple oscillations.

Fig. 7.

Long-term alterations of in vivo hippocampal network dynamics upon NKCC1^Emx1 deletion. (A) Top: Representative example of REM- and SWS-like activity classification of the hippocampal LFP in adult urethane-anesthetized mice. SWS- and REM-like activity in s.l.m. showed characteristic inverted peaks in the delta band (blue) and theta/delta ratio (red), respectively. Bottom: Z-score normalized wavelet spectrogram showing differences in theta and delta power between REM- and SWS-like oscillations. (B) Band-limited power analysis during REM-like activity. Theta amplitude (Left) in s.l.m. and s.m. was higher in KO^Emx1 compared with WT mice. Low-gamma (Middle) and MUA (0.7 to 3 kHz, Right) frequency bands were unaffected, as were the remaining frequency bands analyzed (see SI Appendix). Shaded areas represent ± SEM across animals. (C) Power spectra from s.p. and s.l.m. (D) SPW-R events occurring during SWS-like activity showed lower spectral frequency (Left) and incidence (Right) in KO^Emx1 mice. Ripple-associated SPWs showed no significant amplitude difference between the genotypes (Middle). (E) Experimental arrangement and sample OGB1 fluorescence images. (F, Left) Two-photon fluorescence images; (F, Middle) sample ΔF/F₀ traces; (F, Right) Pearson correlation matrices for the same FOVs. (G) Single-cell CaT frequencies (averaged across animals). (H) Mean CaT frequencies. (I) Mean pairwise Pearson correlation coefficients for measured and randomly shuffled CaT trains. (J) Distributions of Pearson correlation coefficients. (K) Mean spike-time tiling coefficients (STTC) for measured data and randomly shuffled CaT trains. Open circles represent single animals. Mean ± SEM; ns, not significant; *P < 0.05, **P < 0.01, ***P < 0.001. See SI Appendix, Fig. S9 and Table S7.

We further analyzed single-cell network dynamics using two-photon Ca²⁺ imaging in s.p. at P45 to 50 under isoflurane anesthesia (Fig. 7E). As previously reported (53), CaT frequency distributions were strongly skewed to the right in either genotype (Fig. 7 F and G). Moreover, mean CaT frequencies did not differ significantly between WT and KO^Emx1 mice (Fig. 7H). Based on pairwise Pearson correlations we found that, by the age investigated, CA1 network activity had transitioned to a largely desynchronized state in both genotypes (54). Quantitatively, however, Pearson correlations were significantly lower in KO^Emx1 mice (Fig. 7 I and J). Likewise, the spike-time tiling coefficient, an independent affinity metric (55), was significantly lower in the conditional absence of NKCC1 (Fig. 7K). These differences were abolished by randomly shuffling CaT times, confirming that they had not arisen from an inequality of CaT frequencies.

Collectively, our data demonstrate subtle long-term alterations in both intrahippocampal dynamics and entorhinal cortex–hippocampus communication due to PC-specific NKCC1 deletion.

Discussion

NKCC1 Dependency of Network Dynamics In Vivo.

NKCC1 mediates chloride accumulation in immature neurons, giving rise to a mainly depolarizing action of GABA in vivo (13–15). However, the in vivo network functions of NKCC1-dependent GABAergic depolarization remained elusive (reviewed in ref. 29). Conflictingly, systemic application of the NKCC1 inhibitor bumetanide was found to either abolish SPWs (35) or leave SPWs unaffected (36). Technical limitations of in vivo pharmacological approaches, as of local application used here (SI Appendix, Fig. S7 E–H), include the lack of cell-type specificity and poor concentration control. By conditionally deleting Slc12a2 from telencephalic glutamatergic cells, we overcome these limitations and demonstrate that NKCC1 facilitates synchronized network activity in the neonatal CA1 in vivo. Specifically, NKCC1 deletion selectively reduced the frequency of −SPW Ca²⁺ clusters (by ∼20%), while SPWs were unaffected (Figs. 5 and 6). Developmental compensations cannot be excluded from our data, including those that act at the level of [Cl⁻]_i and/or [HCO₃⁻]_i (10, 56, 57). However, if the latter are present at all, they are partial, as we demonstrate that NKCC1 shifted E_GABA by on average ∼8 mV in hippocampal slices. As our transgenic approach ensures that the manipulation of NKCC1 is equally operative in slices and intact brains, our data point to a differential impact of NKCC1 in vivo vs. in vitro. Additionally, no compensatory alterations in intrinsic/synaptic properties were detected in NKCC1 KO^Emx1 mice that could explain the observed minor effect on network dynamics in vivo. Moreover, the impact of NKCC1 is region specific, as its deletion failed to affect spindle bursts in the visual cortex. Conceptually, this conclusion is supported by a recent chemogenetic study revealing that GABA is excitatory in CA1, but inhibitory in V1 (33). Mechanisms underlying this region specificity are unknown, but it is worth noting that NKCC1 did contribute to the maintenance of GABAergic excitation in both CA1 (Fig. 2) and V1 (SI Appendix, Fig. S8) in vitro.

The +SPW and −SPW Ca²⁺ clusters were differentially modulated by local GABA_AR inhibition, as gabazine increased the frequency of +SPW Ca²⁺ clusters, but decreased that of −SPW events (Fig. 6L). This observation might explain discrepant results of two well-designed recent studies, in which either inhibitory (25) or excitatory (33) GABA effects were found to dominate. Hence, we argue that GABA_AR signaling has a dual (excitatory–inhibitory) role in the neonatal CA1. Which mechanism(s) could underlie the excitatory vs. inhibitory GABA effects at this age? GABAergic depolarization may either increase or decrease neuronal firing—depending on the GABA_AR conductance, E_GABA, and the spatiotemporal patterning of synaptic inputs (47–49). Spindle bursts and SPWs are driven by synchronized inputs from the lateral geniculate nucleus (18) and entorhinal cortex (45), respectively, while our LFP data suggest that this may not be the case with −SPW events. Though speculative, such differences in synaptic inputs might explain why −SPW Ca²⁺ clusters are NKCC1 dependent, whereas spindle bursts and SPWs are not. Whether different classes of GABAergic interneurons (58) serve differential functions in vivo, remains unknown.

Two Classes of Network Events in the Neonatal CA1 In Vivo.

Employing Ca²⁺ imaging with LFP recordings, we identified two classes of network events in the neonatal CA1: +SPW and −SPW Ca²⁺ clusters, which differed in several basic parameters (Fig. 6 H–J). As compared with neonatal V1 (18, 19), CA1 Ca²⁺ clusters were less spatially confined. This was most prominent for +SPW events, which tended to activate the entire FOV (59) and frequently followed myoclonic limb/body twitches (45, 60) (SI Appendix, Fig. S6C). Thus, SPWs may provide a pan-CA1 input signal representing, e.g., somatosensory feedback. Of note, −SPW events, of which one-fourth had a significant beta power, constituted about two-thirds of all Ca²⁺ clusters detected at P3 to 4 (Fig. 6G). Since GCaMP signals mainly reflect neuronal firing (61), our data indicate that a considerable fraction of APs in CA1 PCs is not driven by SPWs (45). This conclusion is supported by two-photon Ca²⁺ imaging (SI Appendix, Fig. S6 G–I) and electrophysiological data from freely moving rat pups (P4 to 6) showing that ∼60% of CA1 MUA bursts were not associated with SPWs (16). Some −SPW Ca²⁺ clusters followed +SPW events with short delays and, hence, might reflect SPW tails (Fig. 5E) (16, 36). Given their descriptive differences and differential dependence on NKCC1 (Figs. 5 and 6G) and GABA_AR signaling (Fig. 6L), we argue that +SPW and −SPW events are mechanistically distinct. Based on their differential kinetics and NKCC1 dependence, we further propose that GDPs do not represent the in vitro analog of SPWs recorded in vivo. We revealed that GDPs share pharmacological properties with −SPW events, as both are NKCC1 dependent and disinhibited in amplitude, but not frequency, by gabazine. Given that GDPs are generated in CA1 even if deafferented from CA3 and cortical inputs (62), it is tempting to speculate that −SPW Ca²⁺ clusters might emerge from and reflect a property of local intrahippocampal circuits, whereas SPW generation requires cortical input (45).

Developmental Functions of NKCC1-Dependent GABAergic Depolarization.

We found that several passive and active (Fig. 4 A–E) membrane properties were unaffected by NKCC1^Emx1 loss. This suggests that compensatory changes in intrinsic excitability, observed in conventional NKCC1 mice (11) or after E15 to P7 pharmacological NKCC1 blockade (63), may be due to loss of NKCC1 function in other cell types, or reflect secondary alterations related to NKCC1 deletion outside the brain. Rigorous approaches will be required to clarify potential functions of NKCC1 in GABAergic interneurons. Furthermore, basic characteristics of AMPAR- and GABA_AR-mediated transmission were unaltered in NKCC1 KO^Emx1 mice (Fig. 4 F–I). This challenges the hypothesis of the importance of NKCC1-dependent GABAergic depolarization as a universal requirement for the NMDAR-dependent developmental unsilencing of glutamatergic synapses (9, reviewed in refs. 29, 63)—although depolarizing GABA may facilitate NMDAR activation by attenuating their voltage-dependent Mg²⁺ block (64). Collectively, our data indicate that network effects induced by NKCC1 disruption in telencephalic PCs directly result from changes in E_GABA (i.e., [Cl⁻]_i). Thus, we confirm previous in vitro (8, 12, 35) and recent in vivo (13–15) evidence indicating that NKCC1 maintains high [Cl⁻]_i in immature neurons. Of note, NKCC1 deletion did not completely abolish depolarizing responses to GABA in acute slices, suggesting that factors other than NKCC1 also contribute (10). Additionally, optical and electrophysiological measurements demonstrated subtle alterations of network dynamics in adult NKCC1 KO^Emx1 mice (Fig. 7). Whether these changes reflect developmental adaptations or acute effects due to altered E_GABA, e.g., in the axon initial segment (65, 66), remains unknown. Importantly, they were not associated with performance deficits in hippocampus-dependent behavioral tasks (Fig. 1). These findings may be clinically relevant, as NKCC1 has emerged as a potential target for the treatment of neurodevelopmental disorders (36, 67, 68). While our study does not rule out more specific long-term developmental functions of NKCC1 and/or roles of NKCC1 under pathophysiological conditions, our data converge to suggest that basal physiological NKCC1 activity in telencephalic glutamatergic neurons is largely dispensable for several key aspects of cortical network maturation.

Materials and Methods

Animals.

Mouse strains used in this study are listed in SI Appendix, Table S8. All animal procedures were performed with the approval of the local governments (Thüringer Landesamt für Verbraucherschutz, Bad Langensalza, Germany, and Landesamt für Natur-, Umwelt- und Verbraucherschutz, NRW, Recklinghausen, Germany) and complied with European Union norms (Directive 2010/63/EU). Emx1^IREScre (#005628), GCaMP3^LSL (Ai38, #014538), GCaMP6s^LSL (Ai96, #024106), tdTomato^LSL (Ai14, #007908), and C57BL/6J (#000664) mice were originally obtained from The Jackson Laboratory. NKCC1^flox (Slc12a2^flox) mice were generated as previously described (38). The age of the animals is indicated in Results. Mice of either sex were used. Experiments were performed blinded to genotype. Further details are provided in SI Appendix.

Electrophysiological, Optical, and Behavioral Experiments.

A description of all experimental and analytical procedures is provided in SI Appendix.

Statistical Analysis.

Unless otherwise stated, the statistical parameter n refers to the number of 1) animals for in vivo experiments, 2) slices for in vitro Ca²⁺ imaging, and 3) cells in patch-clamp recordings, respectively. All data are reported as mean ± SEM, if not stated otherwise. The Kolmogorov–Smirnov test or Shapiro–Wilk test was used to test for normality. Parametric testing procedures were applied for normally distributed data; otherwise nonparametric tests were used. In the case of two-sample t tests and unequal group variances, Welch’s correction was applied. For multigroup comparisons, ANOVA was applied. Analysis of covariance (ANCOVA) was used to control for covariate effects. P values (two-tailed tests) <0.05 were considered statistically significant. Details of the statistical tests applied are provided in SI Appendix, Tables S1–S7.

Data Availability

All data discussed in the paper are available in the main text and SI Appendix.