|Home | About | Journals | Submit | Contact Us | Français|
Electroclinical dissociation of neonatal seizures refers to electrographic seizure activity that is not clinically manifest. Dissociation increases after treatment with Phenobarbital, which increases the GABAA receptor (GABAAR) conductance. The effects of GABAAR activation depend on the intracellular Cl− concentration ([Cl−]i) that is determined by the inward Cl− transporter NKCC1 and the outward Cl− transporter KCC2. Differential maturation of Cl− transport observed in cortical vs. subcortical regions should alter the efficacy of GABA-mediated inhibition. In perinatal rat pups, most thalamic neurons maintained low [Cl−]i, and were inhibited by GABA. Phenobarbital suppressed thalamic seizure activity. Most neocortical neurons maintained higher [Cl−]i, and were excited by GABAAR activation. Phenobarbital had insignificant anticonvulsant responses in the neocortex until NKCC1 was blocked. Regional differences in the ontogeny of Cl− transport may thus explain why seizure activity in the cortex is not suppressed by anticonvulsants that block the transmission of seizure activity through subcortical networks.
The neurotransmitter GABA, by binding to its ionotropic receptor (GABAAR), mediates the flux of Cl− and HCO3− in neurons. The opening of these receptors produces an inhibitory or excitatory response depending on the reversal potential for GABA (EGABA), mediated mainly by the intracellular Cl− concentration ([Cl−]i), and its relation to the neuron’s resting membrane potential (RMP). If EGABA is negative to RMP, GABAAR activation will lead to an influx of Cl− and inhibition. However, if EGABA is positive to RMP, GABAARs activation will mediate the efflux of Cl−. This efflux depolarizes the membrane potential and can trigger mechanisms of neuronal excitation such as activation of voltage-gated cation channels.
Immature neurons maintain a high [Cl−]i so that GABAAR activation leads to Cl− efflux and membrane depolarization, while mature neurons maintain a low [Cl−]i so that GABAAR activation leads to hyperpolarization (Ben-Ari et al., 2007; Kahle et al., 2008). The neuronal [Cl−]i is set by the balance of Cl− fluxes across the neuronal membrane. These fluxes include Cl− movement through GABAAR channels (Huguenard and Alger, 1986; Staley and Proctor, 1999; Thompson and Gahwiler, 1989), voltage gated Cl− channels (Staley, 1994), the Cl−/HCO3− exchanger (Gonzalez-Islas et al., 2009), and in particular, two cation-chloride co-transporters: NKCC1, an electroneutral importer of Cl−, and KCC2, an electroneutral extruder of Cl− (Payne et al., 2003). NKCC1 is highly expressed in the hippocampus and neocortex at birth in rodents and humans (Dzhala et al., 2005; Plotkin et al., 1997; Wang et al., 2002). High levels of NKCC1 expression are correlated with an elevated [Cl−]i (Brumback and Staley, 2008; Shimizu-Okabe et al., 2002). During early post-natal development, as NKCC1 expression decreases, and KCC2 expression increases. This change in transporter expression has been demonstrated to underlie the lowering of neuronal [Cl−]i in the hippocampus (Dzhala et al., 2005; Rivera et al., 1999). However, the transition from NKCC1 to KCC2 expression does not occur synchronously throughout the whole central nervous system (Stein et al., 2004).
The ontogeny of KCC2 mRNA expression follows a caudal-rostral pattern. Spinal cord and subcortical neurons begin to express KCC2 early during embryogenesis, while KCC2 expression in cortical neurons increases after birth (Stein et al., 2004; Wang et al., 2002). These expression patterns of NKCC1 and KCC2 suggests that at birth, GABA should have a more inhibitory effect in spinal and subcortical neurons compared to cortical neurons, but there is no direct evidence for a differential effect of GABA on cortical versus subcortical structures.
We considered the possibility that the clinical manifestation of cortical seizure activity is influenced by differences in cortical vs. subcortical ontogeny of the Cl− transporters that determine the polarity of GABAA currents. Convulsions are the muscle contractions arising as a consequence of cortical electrographic seizures. However, cortical seizure activity is neither necessary (Kreindler et al., 1958) nor sufficient (Gale, 1992) for convulsive activity. For example, absence seizures involve the most of the cortex but do not cause convulsions (Hughes 2009). Convulsions can be blocked by selective inhibition of subcortical structures (Iadarola and Gale, 1982). 2-deoxyglucose, Fos, and single positron emission computed tomography studies indicate that cortical seizure activity that does not produce convulsive activity activates only the cortex, whereas seizure activity that produces convulsions also activates subcortical structures including the thalamus (White and Price, 1993; Blumenfeld et al., 2009). These studies support the idea that convulsions arise as a consequence of both cortical and subcortical activity (Gale 1992; Blumenfeld 2002), so that inhibition of activity in either area could suppress convulsions.
Electrographic neonatal seizures frequently have no clinical manifestations, a phenomenon referred to as electroclinical dissociation or uncoupling (Boylan et al., 2002; Castro, Jr. et al., 2005; Connell et al., 1989; Painter et al., 1999; Rennie and Boylan, 2003). Phenobarbital, an allosteric modulator of GABAAR (Twyman and Macdonald 1989), is the drug of choice for treating neonatal seizures. Video-EEG studies have demonstrated that phenobarbital inhibits electrographic seizure activity much less effectively than clinically apparent convulsive activity. This differential efficacy exacerbates electroclinical dissociation (Clancy et al., 1988; Connell et al., 1989; Rennie, 1997; Scher et al., 2003) so that the incidence of electroclinical dissociation in neonates is 80% after anticonvulsive treatment (Boylan et al., 1999; Boylan et al., 2002; Murray et al., 2008; Scher et al., 2003; Weiner et al., 1991).
To gain insight into mechanisms of electroclinical dissociation and its exacerbation by phenobarbital, we tested the hypotheses that the neocortex and subcortical structures have different [Cl−]i during postnatal development using the genetically expressed Cl-sensitive dual wavelength fluorescent protein Clomeleon (Kuner and Augustine, 2000; Berglund et al. 2006). We then tested whether these regions have oppositely directed responses to GABAAR agonists and phenobarbital, and whether bumetanide, an NKCC1 blocker, preferentially alters the responses to GABA in cortical areas expressing high levels of NKCC1.
The differences in KCC2 and NKCC1 expression between neurons in the thalamus vs. neocortex suggest that resting neuronal [Cl−]i will also be different in these two structures (Wang et al., 2002; Yamada et al., 2004). Gramicidin perforated patch recording is routinely used to measure [Cl−]i. However, the membrane pores formed by gramicidin, while impermeable to Cl−, allow the flux of Na+ which can alter NKCC1 activity because this transporter relies on the transmembrane Na+ gradient for its function (Brumback and Staley, 2008). In addition, this technique is time consuming which limits the number of neurons that can be sampled. To circumvent these problems, we tested whether [Cl−]i is higher in the neocortex than the ventroposterior (VP) thalamus during early development by non-invasively measuring neuronal [Cl−]i in hundreds of cells with single cell resolution using two-photon imaging of Clomeleon, an optogenetic Cl− indicator. In CML-1 mice, Clomeleon is expressed in both the neocortex and thalamus of mice (Berglund et al., 2006; Kuner and Augustine, 2000). Clomeleon is a dual wavelength fluorophore, and the fluorescence ratio provides a quantitative readout of ionic concentration that is independent of indicator intensity (Bright et al., 1989). These features made it feasible to measure [Cl−]i in hundreds of cells simultaneously per slice.
We imaged the neocortical layers IV and V and VP thalamus at P5-6, the earliest age at which there was sufficient expression of Clomeleon. Two main findings were observed: first, [Cl]i was widely distributed in both structures at P5-6, P10-11, P15-16 and P20 (Fig. 1A and B). Second, neurons in the VP thalamus always had a significantly lower [Cl−]i than in the neocortex (Fig. 1C; Table 1). At P5-6, thalamus [Cl−]i was 15.1±1.01 mM (n=8,950 cells, 13 slices) compared to 22.1±1.03 mM in neocortex (n=968 cells, 11 slices; p<0.001, Wilcoxon-Mann-Whitney test; Table 1). This difference between neocortex and thalamus [Cl−]i persisted at P10-11, at P15-16 and P20 (Fig. 1C; Table 1). [Cl−]i progressively decreased in both structures during development (Fig. 1C). In addition to the [Cl−]i being different in these two structures, the raw data (the measured YFP/CFP ratios) on which the [Cl−]i values were calculated and that are independent of Cl− calibration, were also statistically different between thalamus and neocortex during the different ages studied, (p<0.001 for each age; Wilcoxon-Mann-Whitney test). The differences in [Cl−]i are consistent with the different patterns of KCC2 and NKCC1 expression in both structures (Wang et al., 2002), and support the idea of a caudal-rostral pattern of maturation of neuronal Cl− transport (Stein et al. 2004).
Our results indicate a lower [Cl−]i in thalamus than neocortex, yet GABA actions depend on the relation between RMP and EGABA, which is mainly determined by [Cl−]i. This relationship is surprisingly difficult to measure (Tyzio et al., 2003) due to the aforementioned difficulties with the influence of direct recording techniques on the neurons’ internal ionic milieu. To avoid this problem, we used extracellular multi-unit activity (MUA) recordings that do not alter the intracellular milieu yet provide a direct measure of the net effect of GABAAR activation on the frequency of network action potentials (Cohen and Miles, 2000; Dzhala and Staley, 2003). MUA were recorded in rat VP thalamus at early ages (P3-4 and P7-8) in the presence of 2 mM kynurenic acid to block all ionotropic glutamate receptors. Isoguvacine (ISO; 10 μM), a GABAAR agonist, induced a significant initial decrease in MUA frequency at P3-4 in 86% of the slices (6/7 slices; Fig. 2). This decrease indicates that the net effect of GABAAR activation was inhibitory in the thalamus. The inhibitory effects of isoguvacine persisted in 100% of the slices at P7-8 (n=5; p=1 Fisher’s exact test between P3-4 and P7-8; Fig. 2). In contrast, isoguvacine induced an initial increase in MUA in the rat neocortex at P9-10 in 89% of the slices (8/9 slices), an increase that is consistent with an excitatory net effect of GABAAR activation (Fig. 2). We were unable to record spontaneous MUA in the rat neocortex at earlier ages, which precluded the determination of GABAAR agonist effects on MUA frequency. However, we were able to measure [Cl−]i directly in neocortical neurons at these ages.
Figure 2 also demonstrates a late decrease in MUA in some of the neocortical recordings. We have previously demonstrated that ion flux due to persistent GABAAR activation by exogenously applied agonists rapidly overcomes NKCC1 transport capacity (Brumback and Staley, 2008), so that ECl equilibrates near RMP. Under these conditions, GABAAR activation is shunting and inhibitory (Mann and Paulsen, 2007). For this reason, we only analyzed the initial response to isoguvacine.
In summary, the MUA experiments demonstrate that the actions of GABA during early development are inhibitory in the thalamus while mainly excitatory in the neocortex. These results are consistent with the [Cl−]i measurement obtained through Clomeleon. Therefore, if GABA has more inhibitory effects in the thalamus, we hypothesized that phenobarbital, a GABAAR positive modulator (Twyman and Macdonald 1989), will be more effective in decreasing epileptiform activity in this structure than the neocortex. We turned to the thalamo-cortical preparation as an ideal preparation to study a synaptically connected sub-cortical (VP thalamus) and cortical (neocortex layer IV/V) structure.
Thalamo-cortical slices have been used to study the effect of anticonvulsive drugs in the adult brain (Zhang et al., 1996; Zhang and Coulter, 1996). Because of the observed differences in neuronal [Cl−]i in the thalamus and neocortex during early development, we took the advantage of thalamo-cortical slices to study the effects of phenobarbital, an allosteric GABAAR modulator (Macdonald and Twyman, 1992) and the most commonly used anticonvulsive drug used in neonatal seizure treatment (Carmo and Barr, 2005), in these interconnected brain structures. This preparation allowed us to measure epileptic activity in synaptically linked cortical and subcortical structures as a test of our hypothesis regarding the mechanisms of electroclinical dissociation. Epileptiform activity was recorded from rat thalamo-cortical slices (P9-10) prepared using a modification of the Agmon and Connors (1991) procedure (See methods, Fig. 3A). Epileptiform activity was not sufficiently stable at earlier postnatal ages, precluding their study. Synchronized epileptiform activity in the neocortex and thalamus was recorded during perfusion of 0 MgCl2 (low-Mg2+) aCSF (Fig. 3B and C). The broadband (1–500 Hz) power of the extracellular field potential recordings was Fig 4C and D, as well as Fig. 3D for a shorter time span of signal power calculation). This approach allowed us to measure the power of the epileptiform activity as a function of time (Dzhala et al. 2005). At P9-10, epileptiform activity simultaneously recorded in thalamus and neocortex was stable for 90 minutes, and an analysis of variance every 10 minutes confirmed no change in the signal power over the entire recording (Fig. 3D and E). Therefore, we used this preparation to study the effects of phenobarbital in the synaptically connected thalamus and neocortex as early as P9-10.
Steady state thalamo-cortical differences in neuronal [Cl−]i (Figure 1) and steady-state thalamo-cortical differences in the effects of GABAAR activation (Figure 2) support the hypothesis that there are corresponding differences in the effects of positive allosteric modulators of GABAAR such as phenobarbital (Twyman and Macdonald, 1989). However, activity-dependent increases in neuronal [Cl−]i have been well observed in a number of preparations (Misgeld et al., 1986; Huguenard and Alger, 1986; Staley and Proctor, 1999) and sustained increases in neuronal [Cl−]i have been observed in immature hippocampal neurons following trains of action potentials (Fiumelli et al., 2005; Brumback and Staley, 2008) and seizure activity (Khalilov et al. 2003). To ascertain that the relationship between thalamic and cortical neuronal [Cl−]i were not changed during seizure activity, we repeated the Clomeleon imaging studies illustrated in Figure 1 during seizure activity induced by low Mg2+ aCSF in P9-10 thalamo-cortical slice preparations as illustrated in Figure 3. Extracellular field potential recordings and two-photon confocal imaging of Clomeleon were performed in the neocortex layer IV/V and ventro-posterior thalamus either separately (n=5) or simultaneously (n=3). [Cl−]i was measured every 10 min before and after onset of epileptiform activity. Recurrent interictal and ictal-like epileptiform discharges were detected in all neocortical recordings (n=5) and in two out of four recordings in the thalamus. In two more recordings from the thalamus application of low Mg2+ aCSF induced only a progressive increase in action potential frequency (not shown). During seizure activity, [Cl−]i in cortical neurons (Fig. 4A, C) increased more than thalamic [Cl−]i (Fig. 4B, D) and the thalamocortical differences were stably maintained throughout one hour of recurrent interictal and ictal-like epileptiform activity (Fig. 4E). Twenty minutes after onset of recurrent epileptiform discharges [Cl−]i was 14 ± 1 mM above the resting chloride level (p < 0.05 vs baseline [Cl−]i, ANOVA followed by Tukey’s means comparison test, n = 268 neurons in 3 experiments) in the neocortex and 0.5 ± 0.2 mM (p > 0.05; n= 348 cells in two slices) above the resting chloride level in the thalamus. One hour after onset of seizures, corresponding [Cl−]i rise was 26 ± 1.3 mM (p < 0.05) in the neocortex and 4.2 ± 0.2 mM (p < 0.05) in the thalamus. Thus seizures enhance rather than diminish thalamo-cortical differences in neuronal [Cl−]i.
Extracellular MUA recordings were recorded after a period of stable epileptiform activity (low-Mg2+) of at least 30 minutes duration. Baseline control recordings consisted of a stable epileptiform activity for at least 20 min after which phenobarbital (100 μM) was perfused (Fig. 5A, B and C). The efficacy of phenobarbital (PB) was determined by comparing the mean signal power during the same time windows between thalamus and neocortex and equal length epochs between control and drug perfusion periods. Consistent with the regional differences in [Cl]i and the effects of GABAAR activation described above, phenobarbital statistically decreased epileptiform power in the thalamus (CON: 0.246±0.03 mV2· Hz, PB: 0.151±0.02 mV2· Hz; n=9; p<0.001, paired t-test; PB/CON: 0.62±0.04) but failed to do so in the neocortex (CON: 0.944±0.22 mV2· Hz, PB: 0.915±0.23 mV2· Hz; n=9; p=0.838 paired t-test; PB/CON: 1.01±0.16; Fig. 5D and E). A decrease in ictal power occurred in 9/9 thalamic recordings and only in 3/9 neocortical recordings.
4-Aminopyridine, a voltage gated K+ channel blocker, is another convulsant frequently used in in vitro seizure models. We turned to this seizure model to confirm the differential effects of phenobarbital between thalamus and neocortex recorded in the low-Mg2+ model. Results in the presence of 50–100 μM 4-AP (P9-10) were similar to those found in the low-Mg2+ model. Phenobarbital reduced the epileptiform power in the thalamus (CON: 0.43±0.03 mV2· Hz, PB: 0.34±0.03 mV2· Hz; n=5; p=0.007, paired t-test; PB/CON: 0.77±0.04) but not in the neocortex (CON: 1.03±0.45 mV2· Hz, PB: 0.81±0.26 mV2· Hz; n=5; p=0.313, Wilcoxon Signed Rank test; PB/CON: 0.90±0.07; Suppl. Fig. 1).
Next, we studied the effect of phenobarbital on the neocortex and thalamus when the structures were electrically disconnected from each other. We first recorded from coronal brain slices, some of which had their subcortical structures trimmed, in low-Mg2+ aCSF. Consistent with the ineffectiveness of phenobarbital on the neocortex in the thalamo-cortical slices, this drug did not change the epileptiform power in isolated rat neocortical slices at P8 (CON: 1.03±0.33 mV2· Hz, PB: 1.21±0.3 mV2· Hz; n=4; p=0.224, paired t-test; PB/CON: 1.25±0.11; Fig. 6A to C). In addition, phenobarbital also failed to decrease epileptiform activity induced by low-Mg2+ recorded in mouse coronal brain slices from P11 Clomeleon mice (CON: 1.66±0.4 mV2· Hz, PB: 2.13±0.35 mV2· Hz; n=5; p=0.106, paired t-test; PB/CON: 1.40±0.20; Fig. 6D to F). We next recorded from the isolated VP thalamus. Epileptiform activity in the disconnected VP thalamus was obtained using horizontal brain slices from P9-10 rat, with an additional cut through the area of thalamo-cortical fibers in the presence of low-Mg2+ plus 100 μM 4-AP. The electrical activity of the neocortex was simultaneously recorded with the thalamus to verify that there was no synchronous activity between both structures. Using this preparation, the thalamus showed independent epileptiform activity from the neocortex. Consistent with the results in thalamo-cortical slices, the electrically disconnected thalamus showed a significant reduction in epileptiform power in the presence of 100 μM phenobarbital (CON: 1.16±0.45 mV2· Hz, PB: 0.72±0.27 mV2· Hz; n=6; p=0.031, Wilcoxon Signed Rank test; PB/CON: 0.65±0.04; Fig. 7). Again, the electrically disconnected neocortex showed no decrease in the presence of phenobarbital in these same slices (CON: 0.82±0.32 mV2· Hz, PB: 0.75±0.26 mV2· Hz; n=6; p=0.563, Wilcoxon Signed Rank test; PB/CON: 1.03±0.11).
Therefore, phenobarbital failed to reduce epileptiform activity in the neocortex in both connected thalamo-cortical preparations and when disconnected from thalamus, yet this drug reduced thalamic epileptiform activity in the thalamo-cortical slices and in thalami that were disconnected from the neocortex. This suggests that phenobarbital is ineffective in the neocortex at this early age of development, whether or not the thalamus is connected. Consequently, in the post-natal period, phenobarbital could have regionally specific effects determined by the local neuronal [Cl−]i.
There is almost a complete lack of KCC2 expression in the neocortex in early development while the expression of NKCC1 persists as late as the second and third post-natal week (Dzhala et al., 2005; Plotkin et al., 1997). High NKCC1 expression correlates with a higher neuronal [Cl−]i during early development (Achilles et al., 2007; Brumback and Staley, 2008; Dzhala et al., 2005; Shimizu-Okabe et al., 2002). Bumetanide is an effective blocker of NKCC1 at low micromolar concentrations (Hannaert et al., 2002; Payne et al., 2003) and has been shown to significantly decrease epileptiform activity in neonatal hippocampus in vitro and in vivo when combined with phenobarbital (Dzhala et al., 2005; Dzhala et al., 2007) but see (Kilb et al., 2007). Thus, we tested whether bumetanide would decrease the power of the epileptiform activity in the neocortex without altering thalamic epileptiform activity in the combined thalamo-cortical slice preparation.
After eliciting synchronized epileptiform activity in the thalamo-cortical preparation with low-Mg2+, 100 μM phenobarbital plus 10 μM bumetanide (PB+BUM) was perfused. In contrast to phenobarbital alone, this drug combination statistically decreased the epileptiform power in the neocortex (CON: 1.3±0.26 mV2· Hz, PB+BUM: 0.41±0.10 mV2· Hz; n=8; p=0.010, paired t-test; PB+BUM/CON: 0.30±0.04), while continuing to decrease the epileptiform power in the thalamus (CON: 0.18±0.03 mV2· Hz, PB+BUM: 0.09±0.01 mV2· Hz; n=8; p=0.011, paired t-test; PB+BUM/CON: 0.58±0.07; Fig. 8). However, in the thalamus, phenobarbital plus bumetanide did not show a further decrease in total power ratio compared to phenobarbital alone (PB/CON: 0.62±0.04, n=9; PB+BUM/CON: 0.58±0.07, n=8; p=0.619, unpaired t-test), as would be expected based on the lower [Cl−]i in these neurons. In contrast, phenobarbital plus bumetanide was more effective in the neocortex than phenobarbital alone (PB/CON: 1.01±0.16, n=9; PB+BUM/CON: 0.30±0.04, n=8; p=0.002, Wilcoxon-Mann-Whitney test). Consistent with this result, phenobarbital plus bumetanide statistically decreased the epileptiform power elicited in the neocortex isolated from the thalamus of Clomeleon mice [P9] in the presence of low-Mg2+ (CON: 2.46±0.28 mV2· Hz, PB+BUM: 1.1±0.14 mV2· Hz; n=8; p<0.001, paired t-test; PB+BUM: 0.46±0.05; Supp. Fig. 2). The effectiveness of phenobarbital plus bumetanide in reducing epileptiform activity in the neocortex vs. thalamus further supports a difference in GABAergic maturation in cortical versus subcortical structures.
To provide additional evidence that phenobarbital is effective depending in part on the neuron’s [Cl−]i, we repeated the experiments described for neocortex and thalamus in two different regions, the hippocampus and amygdala. The ontogeny of Cl− transport in the amygdala is similar to the thalamus: they both express KCC2 at very early ages during embryonic development with an almost complete absence of NKCC1. In contrast, the ontogeny of hippocampal Cl− transport closely resembles the neocortex (Wang et al., 2002), and the CA3 hippocampal area is well characterized regarding its GABAergic developmental profile and response to anticonvulsants (Ben-Ari et al., 2007; Dzhala et al., 2007; Dzhala and Staley, 2003; Quilichini et al., 2003; Rivera et al., 1999; Tyzio et al., 2007). Therefore, we investigated the effects of isoguvacine on MUA frequency recorded in the amygdala (n. medialis) at early post-natal days in the presence of kynurenic acid (2 mM) using the hippocampal CA3 region as an internal control when simultaneous recordings in the amygdala and CA3 region could be performed. Similarly to the thalamus, isoguvacine (10 μM) decreased MUA frequency in the rat amygdala at post-natal day 3–4 [P3-4] in 9 out of 11 slices recorded (82%; Suppl. Fig. 3). In contrast, 10 μM isoguvacine increased the frequency of action potentials recorded in the CA3 region of the hippocampus at the same post-natal age in 5 out of 6 slices (83%; Suppl. Fig. 3).
The inhibitory effect of isoguvacine was similar in the amygdala at P7-8 (100%, n=7; p=0.5 Fisher’s exact test between P3-4 and P7-8; Suppl. Fig. 3) while the excitatory effects of isoguvacine persisted in the CA3 region during this same age in 5 out of 6 slices (83%; p=1 Fisher’s exact test between P3-4 and P7-8; Suppl. Fig. 3).
Epileptiform activity was induced in rat P4-6 bran slices by perfusing aCSF with low-Mg2+. After a control baseline period, 100 μM phenobarbital was perfused. Phenobarbital had a modest, yet significant, decrease in the signal power of the CA3 region (CON: 0.41±0.09 mV2· Hz, PB: 0.31±0.06 mV2· Hz; n=11; p=0.032, Wilcoxon Signed Rank test; Suppl. Fig. 4). In comparison, phenobarbital statistically decreased the epileptiform signal power in the amygdala (CON: 0.25±0.04 mV2· Hz, PB: 0.14±0.02 mV2· Hz; n=8; p=0.002, paired t-test; Suppl. Fig. 5). Moreover, the phenobarbital effect in the amygdala was more effective than in the CA3 region when the PB/CON of the signal power in both structures was compared (Amygdala: 0.60±0.06, n=8; CA3: 0.84±0.07, n=11; p=0.021, unpaired t-test) which correlates with the inhibitory effect of isoguvacine in the amygdala but not the hippocampus.
These results support a caudal-rostral development of GABAergic activity as thalamus and amygdala share a similar KCC2-NKCC1 expression pattern and hence a more mature GABAergic signaling than the neocortex and hippocampus, which also share a similar NKCC1-KCC2 pattern and more immature GABA effects.
CA1 pyramidal neurons of females rat pups were shown to have a lower [Cl−]i than male pups during early development (Galanopoulou, 2008). We decided to test if sex differences in GABAergic effects were evident in the CA3 region, thalamus and amygdala. The effect of isoguvacine recorded in female rats (P4-6) was comparable to the ones recorded in males (female: [CA3]: 100% of slices increased MUA frequency (n=6 slices); [Amygdala]: 100% of slices decreased frequency (n=7 slices). [Thalamus]: 83% of slices decreased MUA frequency (5/6 slices); Suppl. Fig. 6). Thus, the effects of isoguvacine did not differ between male and female brains in the recorded regions.
Our main findings are: 1) [Cl−]i varies substantially between neighboring neurons in both the developing thalamus and neocortex, but the average [Cl−]i is significantly lower in thalamic vs. cortical neurons. 2) Phenobarbital is an effective anticonvulsant in the thalamus but not in the neocortex, due to a net inhibitory effect of GABA in the thalamus but an excitatory effect in the neocortex. 3) The combination of bumetanide and phenobarbital is effective in decreasing epileptiform activity in the neocortex while it is not different from phenobarbital alone in the thalamus. Our results support the idea that caudal-rostral [Cl−]i maturation determines neuronal responses to GABA and therefore the effects of allosteric modulators of GABAR function. Taken together, these differences in [Cl−]i comprise a candidate mechanism of electroclinical dissociation of neonatal seizures and the exacerbation of dissociation by GABAergic anticonvulsants.
The best methods for measuring neuronal Cl− are still being established. Approaches include gramicidin perforated patches, which form membrane pores impermeable to anions while allowing the passage of cations (Kyrozis and Reichling, 1995). A disadvantage of this widely used technique is diffusion of Na+ from the intracellular pipette to the neuron, which alters [Na+]i, and the driving force for NKCC1-mediated Cl− transport (Brumback and Staley, 2008; Shimizu-Okabe et al., 2002). Cell-attached patches have also been used to estimate EGABA from the reversals of single-channel currents (Tyzio et al., 2007). These methods are less invasive but they have limited time resolution, and may activate stretch-activated potassium channels near the patch (Hammami et al. 2009). Both patch techniques are time-consuming, so that only limited numbers of neurons can be assayed. Measurement of spontaneous action potential frequency and Clomeleon dye measurements are noninvasive and have the added benefit of recording a significant number of neurons simultaneously (Cohen and Miles, 2000; Dzhala and Staley, 2003). The large sample sizes reported here provide a new insight into the heterogeneity of the ontogeny of neuronal Cl− transport not only between but also within regions of the brain. However, action potential measurements do not provide a direct measure of EGABA. Disadvantages of Clomeleon include pH sensitivity, the need for calibration, potential differences in tissue transmission of yellow vs. cyan light, and limitation to neurons that express the dye. For example, [Cl−]i in the amygdala was not imaged due to a null or low expression during early development in CML-1 mice.
Intracellular ionic concentrations are affected by the extracellular concentration of ions as well as substances present in the cerebrospinal fluid. Compared to our results, cortical neurons in brain slices supplied with ketone bodies demonstrated more negative values for EGABA that were dependent on intact HCO3−/Cl− exchange (Zilberter et al., 2009). However, perfusion of high concentrations of a weak acid is a well-established method of acidifying the intracellular compartment (Roos and Boron 1981; Dulla et al., 2009). Independently of metabolic effects, the perfusion of 4 mM of the ketone 3-hydroxybutyrate (pKa 4.7) by Zilberter et al. would be expected to acidify the neuronal cytoplasm, reducing intracellular HCO3− (Roos and Boron 1981). HCO3−/Cl− exchange would then reduce [Cl−]i (Romero et al. 2004), producing the observed negative values of ECl and EGABA and dependence on HCO3−/Cl− exchange.
The expression of KCC2 is high in the thalamus during the perinatal period, while NKCC1 expression is low (Li et al., 2002; Stein et al., 2004; Wang et al., 2002). In the neocortex, NKCC1 expression is high in the perinatal period and decreases progressively the first two postnatal weeks; conversely, embryonic and early perinatal KCC2 expression is low and increases post-natally (Clayton et al., 1998; Plotkin et al., 1997; Yamada et al., 2004). The expression of KCC2 is negatively correlated with [Cl−]i, while the expression of NKCC1 is positively correlated with [Cl−]i (DeFazio et al., 2000; Hubner et al., 2001; Owens et al., 1996; Rheims et al., 2008a; Sung et al., 2000; Yamada et al., 2004; Zhu et al., 2005). While NKCC1 and KCC2 appear to be the main transporters setting [Cl−]i in neurons (Sung et al., 2000; Zhu et al., 2005), other transmembrane chloride pathways may also be involved including NDCBE (Na+ Driven Chloride Bicarbonate exchanger, also known as SLC4A8; involved in pH regulation) and CLC-2 (Conductive Cl− channel) (Chen et al., 2008; Gonzalez-Islas et al., 2009; Romero et al., 2004; Staley, 1994; Staley et al., 1996). However, the robust effects of NKCC1 inhibition on the efficacy of GABA-mediated inhibition (Figure 8; Dzhala et al. 2005; Sipila et al. 2006; Brumback and Staley 2008; Rheims et al. 2008b) indicate that NKCC1 is the primary Cl− accumulator in neocortical and hippocampal neurons. Layer IV/V neocortical [Cl−]i was higher than the VP thalamus during early development in agreement with our MUA recordings. The [Cl−]i developmental profile in neocortex and thalamus is reminiscent of the development of the visual cortex and the dorsal lateral geniculate nucleus (dLGN) (Ikeda et al., 2003).
While [Cl−]i measurements in vivo are lacking, bumetanide blocked sharp waves recorded in the hippocampus in vivo (Sipila et al., 2006). Sharp waves are analogous to the giant depolarizing potentials (GDPs) recorded in this area in vitro, and GDPs are also blocked by bumetanide (Dzhala et al. 2005). This suggests that the higher [Cl−]i recorded in vitro during early development (Tyzio et al. 2007; Figure 1) is also present in vivo. Neonatal seizures occur most commonly several hours after an acute cerebral injury (Tekgul et al., 2006). Hypoxia has been shown to cause both a late increase NKCC1 expression (Yan et al., 2003) and a delayed increase in neuronal [Cl−]i (Pond et al. 2006). This increase in [Cl−]i may be sufficient to invert the action of GABA and trigger seizures. Seizures themselves may further increase [Cl−]i, because trains of action potentials have also been demonstrated to cause sustained increases in Cl−i (Fiumelli et al. 2005; Brumback and Staley 2008; Figure 4). Thus in the setting of post-hypoxic seizures in human neonates that express high levels of NKCC1 in cortical neurons (Dzhala et al. 2005), the balance of evidence suggests that GABAAR activity is excitatory in the cortex during neonatal seizures.
Our measurements revealed a wide distribution of neuronal [Cl−]i. Our neocortical [Cl−]i measurements are similar to previous [Cl−]i measurements obtained from different techniques (Owens et al., 1996; Yamada et al., 2004), yet we demonstrate a wide intra-area variation that is difficult to observe with direct recording methods because of the number of neurons that must be recorded (Ebihara et al. 1995). The remarkable inter-neuron variance in [Cl−]i implies that very large numbers of neurons should be sampled in order to accurately assess neuronal [Cl−]i. This [Cl−]i variation suggests that some neurons will have EGABA above RMP while in others it will be below RMP. Therefore, the net effect of drugs that modulate GABAergic transmission will depend on the proportion as well as the classes of neurons having excitatory and inhibitory actions of GABA.
The EEG preferentially records neocortical activity (Marks et al., 1992). As described in the introduction, convulsions are the motor manifestation of cortical seizure activity, however this activity, recoded electrographically, is neither necessary nor sufficient to produce convulsive activity in human neonates (Mizrahi and Kellaway, 1987; Connell et al., 1989; Scher et al. 2003; Murray et al. 2008). Thus many neonates have convulsions with no cortical EEG correlate (Mizrahi and Kellaway, 1987), and neonates frequently manifest electrographic seizure activity without convulsive activity (electroclinical dissociation or uncoupling; Scher et al. 2003). The prominent role of subcortical networks in the production of convulsive activity (Gale, 1992; White and Price, 1993; Blumenfeld, 2009) frame the hypothesis that selective inhibition of subcortical structures could suppress convulsive activity but not electrographic cortical seizure activity in neonates, and this selective inhibition could arise from the differential neuronal Cl− transport (Stein et al. 2004), [Cl−]i (Figure 1), and consequent effects of GABAAR activation (Figure 2) in these structures.
The low efficacy of phenobarbital in suppressing cortical vs. thalamic epileptiform activity in our experiments comprises a candidate mechanism for the exacerbation of electroclinical dissociation by phenobarbital. Phenobarbital exhibited the same efficacy in the perinatal thalamus as in adult in vitro thalamo-cortical preparations (Zhang and Coulter, 1996) as well as in vivo (Hosford et al., 1997), and this efficacy was not improved by bumetanide, supporting the idea that in the perinatal period, thalamic neurons have developed mature patterns of neuronal Cl− transport (Stein et al. 2004), GABAA signaling (Figure 2), and anticonvulsant responses to phenobarbital (Figure 5). We used the thalamus as a model subcortical structure because of the high correlation between thalamic activation and convulsive activity (White and Price 1993; Blumenfeld 2009) as well as the high expression of Clomeleon in the perinatal CML1 mouse (Berglund et al. 2008). However, electroclinical dissociation could also arise as a consequence of a higher efficacy of phenobarbital in any subcortical network involved in transmission of convulsive activity, including the spinal cord, a region with the earliest KCC2 expression (Stein et al. 2004).
Although the anticonvulsant effects of phenobarbital in human neonates are poor when measured electrographically, they are not absent (Painter et al. 1999). Similarly, we observed modest but significant anticonvulsant effects in our experiments in area CA3, even though GABA appeared excitatory in this region based on the net increase in action potential frequency in response to perfusion a GABAAR agonist. There are several reasons why phenobarbital could exhibit anticonvulsant activity in structures with excitatory responses to GABAAR activation. During seizure activity, paroxysmal depolarizing shifts in the membrane potential (PDS) result in membrane potentials that are very positive to the GABA reversal potential, even in immature neurons. Thus, GABAAR activation can repolarize the membrane potential relative to the PDS even though EGABA is more positive than RMP. Additionally, high concentrations of GABA agonists or allosteric modulators may result in Cl− fluxes through the GABAAR that are sufficiently large and prolonged to change [Cl−]i and so shift EGABA toward RMP (Brumback and Staley, 2008; Proctor and Staley, 1999), in which case GABAA receptor activation would provide effective shunting inhibition. Further, in a network of neurons, a variety of EGABAs are apparent (Fig. 1). Thus, the effect of phenobarbital on the network is even less predictable than in an individual neuron. Finally, phenobarbital may have other anticonvulsant actions besides allosteric modulation of the GABAA receptor (Kitayama et al. 2002). These issues may underlie the sometimes conflicting results of different investigators regarding the anticonvulsant efficacy of allosteric modulators of GABAAR (Dzhala et al., 2005; Dzhala et al., 2007; Kilb et al., 2007; Wells et al., 2000). Our results provide evidence that phenobarbital’s anticonvulsant efficacy parallels theefficacy of GABAAR-mediated inhibition (Dzhala et al., 2005; Quilichini et al., 2003).
To our knowledge, this is the first report to show the differential effect of phenobarbital in two synaptically connected structures that also demonstrate differences in [Cl−]i. We recorded the epileptiform activity of a thalamo-cortical slice for the first time, as far as we know, from P9-10 rat pups. The different effects of phenobarbital on both structures correlates to the opposite effects of isoguvacine on MUA frequency, the differences in neuronal [Cl−]i, and the previously reported differences in KCC2 vs. NKCC1 expression in these two areas. This combined preparation provides experimental support for the hypothesis that exacerbation of electroclinical dissociation arises from the selective anticonvulsant efficacy of phenobarbital in subcortical vs. cortical regions.
Drugs that reduce the [Cl−]i by blocking NKCC1 activity, e.g. bumetanide, improve the electroencephalographic anticonvulsive effects of phenobarbital in vitro and in vivo (Dzhala et al., 2005; Dzhala et al., 2008). In Swiss mice, bumetanide alone decreased seizure activity in the neocortex during early development (Rheims et al., 2008b) and the effect of bumetanide is synergistic with phenobarbital (Dzhala et al., 2008). This drug combination could be of significant impact in the human neonate, and therefore we studied these two drugs in conjunction. This synergistic effect statistically decreased the epileptiform activity in the neocortex, whereas phenobarbital alone did not. The anticonvulsant effects of phenobarbital were not enhanced by bumetanide in the thalamus. Thus the anticonvulsant effect of bumetanide would be more evident electrographically rather than clinically: neonates whose convulsions were already suppressed by selective effects of Phenobarbital on subcortical networks could not manifest additional reductions in convulsions, but electrographic seizure activity could be reduced.
Neuronal NKCC1 is upregulated in adult patients suffering from chronic temporal lobe epilepsy, so bumetanide might be helpful in this population (Huberfeld et al., 2007; Palma et al., 2006). Post-hypoxic increases in NKCC1 expression (Yan et al. 2003) and neuronal chloride are seen in adult neurons experimentally (Pond et al. 2006), and electroclinical dissociation has recently been observed in critically ill adult patients (Oddo et al. 2009), but it is not known whether pharmacological inhibition of NKCC1 is useful in this setting.
KCC2 and NKCC1 expression pattern in the developing human parietal cortex is similar to the levels found in rat neocortex (Dzhala et al., 2005), although the pattern of expression of these transporters in subcortical structures has not been studied in humans. Bumetanide has been extensively studied as a diuretic in human neonates (Lopez-Samblas et al., 1997; Sullivan et al., 1996). More recent experimental indicate that long-term suppression of NKCC1 activity and chronic KCC2 over-expression produce alterations in migration patterns of neocortical principal neurons and their dendritic length (Ge et al. 2006; Cancedda et al., 2007; Wang and Kriegstein, 2008). However, long-term changes in brain activity due to NKCC1 block during seizures have not been studied. Most human neonates would not require long-term treatment with bumetanide; the majority of neonatal seizures are an acute reaction to brain injury (Tekgul et al. 2006), and under these circumstances anticonvulsant treatment can usually discontinued after the first week of life (Carmo et al. 2005). Recently, bumetanide treatment was reported to decrease seizure duration and frequency in a 6 week baby girl with anticonvulsant-resistant status epilepticus secondary to meningitis (Kahle et al., 2009). A pilot trial of bumetanide for neonatal seizure treatment is currently under way (Clinicaltrials.gov # NCT00830531).
In conclusion, our results support the hypothesis of a caudal-rostral functional maturation of GABA actions and provide a candidate pathophysiological mechanism of neonatal electroclinical dissociation as well as a pharmacotherapy that could be beneficial in this condition.
Post-natal (ages: P3–10) male (female where indicated) Sprague-Dawley rats and CLM1 Clomeleon mice (P5-P20; C57bl/6 background) were anaesthetized and decapitated according to a protocol approved by the Massachusetts General Hospital Center for Comparative Medicine. The brain was removed and placed in ice-cold artificial cerebrospinal fluid (aCSF) containing (mM): NaCl (120), KCl (3.3), CaCl2 (1.3), MgCl2 (2), NaH2PO4 (1.25), NaHCO3 (25) and D-glucose (10) with pH 7.3–7.4 when bubbled with 95% O2 and 5% CO2. Coronal and horizontal brain slices, 450–500 μm thick, were cut with a Leica VT1000S Vibratome (Leica Microsystems, Wetzlar, Germany) in aCSF containing 2 mM kynurenic acid (Sigma; St. Louis, MO). In a subset of coronal slices, subcortical structures were trimmed. Thalamo-cortical brain slices were obtained by modifying the procedure by Agmon and Connors (1991) as follows: after exposing the brain by removing the parietal, temporal and occipital bones, a 45° angle coronal cut was performed before removing the brain from the base of the cranium and its rostral end was glued to the vibratome dish. Slices were placed into an interface holding chamber kept at room temperature and stored for at least 1 h before being transferred to the recording chamber.
Brain slices were placed in an upright microscope (Zeiss, Axioskop; Thornwood, NY) perfused with aCSF (0.5 mM MgCl2 to increase spontaneous activity when recording MUA) held at 32–34 °C and aerated with 95% O2-5% CO2 (flow rate of 5–8 ml/min). Tungsten coated electrodes were placed in either neocortex layer IV/V, CA3 region of the hippocampus, amygdala (n. medialis), and thalamus (ventroposterior nucleus). Recordings were obtained through a CyberAmp 320 amplifier (Molecular Devices, Sunnyvale, CA) with a x10,000 gain, digitized at 10 KHz with a DigiData 1321A (Molecular Devices) and recorded with pClamp (Molecular Devices). Epileptiform activity was induced by perfusing the brain slices with aCSF containing 0 MgCl2 (low-MgCl2) or 4-Aminopyridine (4-AP) and recordings were started after 30 min of consistent epileptiform activity. Drugs were perfused as indicated.
Two-photon imaging was performed using a Fluoview 1000MPE with pre-chirp optics and a fast acousto-optical modulator (AOM) mounted on a Olympus BX61WI upright microscope (Olympus Optical) using a 20x water immersion objective (NA 0.95; Suppl. Fig. 7A). A mode-locked Ti:Sapphire laser (MaiTai, Spectra-Physics, Fremont, CA) generated two-photon fluorescence with 860 nm excitation. CFP and YFP filters (340–480 and 500–540 nm respectively) and two photomultiplier tubes (Hamamatsu Photonics) were used to simultaneously acquiring CFP and YFP signal. Three dimensional stacks (3D) were imaged. Slices were perfused with aCSF held at 32–34 °C and aerated with 95% O2-5% CO2.
Recordings were high-passed filtered at 10 Hz. Events were detected by Minianalysis (Synaptosoft, Inc., Fort Lee, NJ). Isoguvacine effect was determined by using a peak function running under Origin (OriginLab Corporation, Northampton, MA) based on the following equation:
where y0 is the offset, A is amplitude, xc is the center and w is the width. The first peak (either negative or positive) after perfusion of isoguvacine was taken as the effect of this drug.
A custom-written macro running under IGOR Pro v6 (Lake Oswego, OR) was used. Briefly, the complete trace was loaded and down sampled to 2 KHz. For each epoch of 30 sec, its mean value was subtracted and Fast Fourrier Transformed using Hann window apodization. Next, the FFT was smoothed with a median window of 7 points, divided by the total number of points, and the signal power area (1–500 Hz) was calculated (Supp. Fig. 4B and C). The wide-band power is mathematically identical to the square of the field potential, and thus represents an unbiased measure to quantify seizure activity (e.g. Sup. Fig. 2; Dzhala et al. 2005, 2008). The wide-band power is proportionately affected by the frequency of discharge and the fraction of the population participating in the discharge. Finally, the mean signal power during equal control and drug condition epochs was calculated. This method allowed us to address the combined effect of drugs on the frequency and power of the epileptiform activities. Graphs display the signal power in the EEG band (1–160Hz) (Bragin et al., 1999; Bragin et al., 2004; Khosravani et al., 2005).
Quantitative measurements on 3D stacks were performed using Image J (National Institutes of Health, freeware) offline. The CFP and YFP images were loaded and their respective background value was subtracted for the 3D volume. Next, a median filtered was applied to all of the 3D planes. Cells were visually identified and a region of interest (ROI) was drawn around the cell bodies and the ratio of the YFP/CFP fluorescence intensity was measured. Each cell’s YFP/CFP ratio was converted into [Cl−] by the following equation:
where K’D is the apparent dissociation constant, Rmax is the ratio when Clomeleon is not bound by Cl− and Rmin when it is completely quenched by F− (Berglund et al., 2008; Kuner and Augustine, 2000). These values were calculated from the calibration of [Cl−]i in acute slices (Suppl. Figure 7B). The mean of the logarithmic value of the [Cl−]i was used in every case, as the ion concentration follows a log-normal distribution. ECl was calculated using the Nernst equation setting the temperature variable at 33 °C.
Acute brain slices of the neocortex (P11-15) were used to calibrate the YFP/CFP intensity to [Cl−] (Berglund et al., 2008; Krapf et al., 1988). Briefly, acute brain slices were exposed to the K+/H+ ionophore nigericin (50 μM) and the Cl−/OH− antiporter tributyltin chloride (100 μM) in the presence of 20, 80 and 123 mM [Cl−]o. Neurons that experienced a change in YFP/CFP intensity to each [Cl−]o were used for the calibrations. The data points obtained with the different [Cl−]o are described by the ratiometric function:
Rmax and K’D were free parameters, while Rmin was determined by quenching Clomeleon with 123 mM F− (Duebel et al., 2006; Kuner and Augustine, 2000). With single-cell resolution, we calculated Rmax and K’D for 47 individual neurons that we were able to follow across all calibration conditions. The K’D was 91±5.43 mM, Rmax was 1.026 and Rmin was 0.268 (Suppl Fig. 7B). To determine if the Cl− calibration is consistent with the expected [Cl−]i values, we perfused 500 μM GABA onto P31 neocortical pyramidal cells. We observed an increase in [Cl−]i in cells with low initial [Cl−]i while the opposite was recorded in neurons with high initial [Cl−]i (Suppl. Fig. 7D,E). While Clomeleon is sensitive to intracellular pH (pHi) at high [Cl−] (150 mM), it is not pH sensitive at lower [Cl−] (10 mM). Changes of 0.2 pH units would cause the apparent [Cl−]i to change a few millimolars for [Cl−] smaller than 50 mM and produce large deviations at 150 mM (Kuner and Augustine, 2000). However, this sensitivity of Clomeleon to pH should not affect our [Cl−]i determination as pH is tightly regulated in neurons (Bonnet et al., 1998) and we performed all of our experiments in resting neurons bathed in CO2-HCO3− buffered solution which prevents pHi changes. The pHi could be of concern in in vivo preparations, where the pH is not directly controlled; or in experiments known to change pHi. Moreover, there were no exchanges of extracellular solutions or drug perfusion that could alter pHi. Even though we were careful in the calibration of Clomeleon, we also included the statistics on the actual YFP/CFP ratios, which are independent of Cl− calibration.
The Shapiro-Wilk test was use to determine if the data followed a Gaussian distribution. The parametric Student t-test (paired and unpaired, two-tail) was used if the data were normally distributed while the Wilcoxon-Mann-Whitney test (unpaired data, two-tail) and Wilcoxon Signed Rank test (paired data, two-tail) was used for non-Gaussian distributions. Analysis of variance (one-way ANOVA) was performed when measuring multiple conditions. Statistical analysis of the linear regression slopes was done using a Student t-test (Zar, 1999). Fisher’s exact test (two-tail) was used when comparing proportions. Statistical significance was set to p<0.05. Values are expressed as mean ± standard error of the mean (SEM).
Isoguvacine was obtained from Tocris (Ellisville, MO). Kynurenic acid, phenobarbital, nigericin, tributyltin chloride, 4-AP and bumetanide were obtained from Sigma (St. Louis, MO). Bumetanide was dissolved in 100% ethanol and made fresh every day.
To the Staley lab for helpful discussion. This work was supported by the American Epilepsy Society (JG), NIH NS580752 (KVK), NIH NS040109 (KJS) and NIH EB000768 (BJB).
Author contribution: JG: design, electrophysiology, two-photon imaging, analysis, writing; KK, DV: two-photon imaging and analysis; GF, TK and GA: mice generation; BJB: two-photon imaging; KJS: design, writing.