The GABA_A Reversal Potential

The anions that flow through the GABA receptor channel are chloride and bicarbonate (Eccles et al., 1977; Bormann et al. 1987). Each of these anions diffuses along its electrochemical gradient. For chloride, this flow is most often inward. The chloride concentration is high in the extracellular space and lower in the cytoplasm of neurons. This difference in concentration produces a strong inward diffusion gradient that usually overcomes the repulsive force of the negative membrane potential on the negatively charged chloride anions, resulting in inward flux. Bicarbonate, on the other hand, is at a slightly lower concentration in the cytoplasm than the extracellular space, as a consequence of the differences in pH: about 7.2 in the cytoplasm and 7.4 in the extracellular space (Roos and Boron 1981; Chesler, 2003). From these pH values, the Henderson-Hasselback equation (1) predicts a bicarbonate ion concentration of 16 mM in the cytoplasm and 25 mM in the extracellular space.

where pK, the negative logarithm of the dissociation constant for H₂CO₃, is 6.1 and a physiological H₂CO₃ concentration in 5% CO₂ is 1.2 mM (Leader 1979). Using the Nernst equation (2), the calculated reversal potential for bicarbonate is about –12 mV.

where EA is the reversal potential for membrane currents carried by ion A (in this case HCO₃⁻), A_i and A_o are the intracellular and extracellular concentrations of A, R = the gas constant 8.3145 J / (mol × K), T = temperature in Kelvin, Z = the valence of the ionic charge, and F = Faraday’s constant, 96,485 C/mol.

This means that at potentials more negative than –12 mV, such as a resting membrane potential of –70 mV, bicarbonate will flow out of the open GABA_A channel. This is the opposite direction from the inward chloride flux. The permeability of chloride through the GABA_A channel is about five times higher than bicarbonate (Bormann et al. 1987), and the concentration of chloride on the side of the channel that it enters is about five times the corresponding bicarbonate concentration, so at first glance, Cl⁻ would seem to carry 25 times more charge than HCO₃⁻ through the open GABA_A channel. But in addition to diffusive gradients, the electromotive driving force for ion flux must also be considered. The driving force is the difference between membrane potential and the reversal potential. Near the chloride reversal potential, the driving force for HCO₃⁻ efflux can be as much as 25× the driving force for Cl influx, so the two ions may carry equal amounts of charge through the channel under these conditions. The Goldman-Hodgkin-Katz current equation (3) is used to predict these sometimes complex relationships between permeability and driving forces for the permeant anions (Goldman 1943; Hodgkin and Katz 1949).

where I = membrane current carried by ion A, P = the membrane permeability of that ion, V is the membrane potential, and F, R, and T have the values specified in equation (2). A more complex equation describes the currents carried by two permeant ions such as Cl⁻ and HCO₃⁻ (Staley and Proctor 1999).

Because of the higher concentration and GABA_A permeability of Cl⁻ versus HCO₃⁻, the reversal potential for GABA_A receptor-gated anion currents, E_GABA, is close to the reversal potential for chloride. The reversal potential for mixed chloride and bicarbonate currents can be calculated from the Goldman-Hodgkin-Katz potential equation (4), which extends the Nernst equation to the case of two permeant ions:

where the symbols and constants are the same as for equation (3), Z = 1, and the permeability ratio of Cl⁻ to HCO₃⁻ is 5 (Bormann et al. 1987).

In other words, the direction of the current through the open GABA_A channel is primarily determined by the size of the concentration gradient for chloride diffusion and the repulsive force of a membrane potential more negative than E_GABA, or the attractive force of a membrane potential more positive than E_GABA. Because the concentration of chloride in the extracellular space is so much larger than the concentration of chloride in the intracellular space, the driving force for chloride diffusion is not altered very much by changes in the extracellular concentration of chloride in the range of a few millimoles. Using round numbers for an example, equation 2 predicts that if the extracellular chloride concentration was 100 mM and the intracellular chloride concentration was 5 mM, then a 5 mM increase in the extracellular chloride concentration would only change the chloride reversal potential by the natural log of a 5% change times RT/F, or about 1 mV. However, because the intracellular chloride concentration is so low, the same 5 mM change to the intracellular chloride concentration represents a 100% change in concentration. The chloride reversal potential changes by the natural log of that 100% change times RT/F, or about 18 mV. Using equation (4) to account for HCO₃⁻, the same changes in chloride produce changes in E_GABA of 1 mV for extracellular Cl⁻ shift and 12 mV for the same shift in intracellular Cl⁻: from –73 to –61 mV. These examples demonstrate that the GABA_A reversal potential (E_GABA) is strongly influenced by the chloride reversal potential. Thus, changes in the intracellular chloride concentration of only a few millimoles can result in large changes in E_GABA relative to the resting membrane potential.

E_GABA versus RMP

Chloride concentrations of 5–10 mM in the cytoplasm of neurons and ~110 mM in the extracellular space, together with physiological bicarbonate concentrations, result in E_GABA that is close to the resting membrane potential (RMP) of the neuron (Eccles et al. 1977; Tyzio et al., 2008). The difference between E_GABA and the RMP of the neuron is the driving force for anion flux through the GABA receptor when there are no other synaptic or voltage-gated conductances affecting the membrane potential, for example, concurrent depolarizing glutamatergic postsynaptic potentials. This driving force for GABA currents is often small, only a dozen millivolts or less, and usually directed such that E_GABA < RMP; that is, anion flux through the open GABA channel will be inward and hyperpolarizing when the neuron’s membrane potential is near RMP (Coombes et al. 1955; Andersen et al. 1980; Thompson and Prince 1988). As we saw in the preceding paragraph, however, a few millimoles shift in the cytoplasmic chloride concentration can result in a shift of E_GABA of a dozen mV, so that E_GABA could become positive to RMP. In this case, opening the GABA_A receptor will result in a net outward anion current. This outward anion flux would cause the cytoplasm to lose negative charge, so the membrane becomes depolarized; that is, the membrane potential becomes more positive (Lux 1971; Eccles et al. 1977; Misgeld et al. 1986; Huguenard and Alger 1986; Staley and Proctor 1999). Thus, the direction of the current through the open GABA_A receptor, and the effect of the open GABA_A receptor on the membrane potential, can be reversed by a few millimoles shift in the cytoplasmic chloride concentration.

When Are GABA Currents Excitatory?

If GABA_A receptor-gated currents depolarize the membrane, does that mean that the action of the GABA receptor is excitatory? The answer is sometimes yes, and sometimes no (Doyon et al. 2016). Many conductances are activated in the voltage range over which GABA_A receptors have been observed to drive the membrane potential, that is, over the observed range of E_GABA. For example, the magnesium block of the NMDA receptor begins to be removed at membrane potentials positive to –70 mV (Konnerth et al. 1989). Low-threshold calcium currents begin to activate at potentials positive to –60 mV (Chemin et al. 2002), and somatic sodium channels begin to activate at –50 mV (Colbert and Pan 2002). On the other hand, inactivation of low-threshold calcium channels is rapid in this potential range (Chemin et al. 2002), and sodium channels also inactivate in this range of membrane potentials (Ulbricht 2005).

Is E_GABA a Monolithic Number?

I spent the first several years of my research career searching for the true E_GABA of hippocampal CA1 pyramidal cells but found that E_GABA was different in every neuron. I considered this effort a failure and never reported the heterogeneity of E_GABA. Twenty recent studies have measured E_GABA in these same neurons, using the same experimental techniques at the same rodent ages. These 20 studies of perforated patch recordings of CA1 pyramidal cells in acute slice preparations reported a range of E_GABA from –85 to –60 mV (for details, see Rahmati et al. 2021). The range of the group mean values reported for E_GABA in these studies supports the idea that there is no one true value for “E_GABA.” Rather, E_GABA varies from neuron to neuron, so that the average E_GABA did not converge toward a “true” value.

What Determines E_GABA?

This inherent interneuronal variance in EGABA raises questions as to how E_GABA is determined. As we said earlier, both bicarbonate and chloride permeate the GABA_A receptor. Bicarbonate concentrations are set by the pH and pCO₂ (equation 1), and although these are not thought to vary much, recent measurements suggest intercellular variance in pH (Boffi et al. 2018). Even with this intercellular variance in pH, the HCO₃⁻ potential will still be far from resting membrane potential, so that most of the variance in E_GABA will arise from variance in the transmembrane chloride gradient. The determinants of the transmembrane chloride gradient are controversial. The Nernst equation (equation 2) predicts that the chloride reversal potential is determined by the cytoplasmic chloride concentration, presuming that the extracellular chloride concentration is relatively constant.

The Role of Transporters in Chloride Homeostasis

Initially, the cytoplasmic chloride concentration was thought to be determined by cation-chloride cotransport (Rivera et al. 1999; Yamada et al. 2004; Dzhala et al. 2005; reviewed in Kaila et al. 2014). Cation-chloride cotransporters are expressed in every cell. In non-neuronal cells, the transporters function to import chloride salts and export chloride salts (Hoffman et al. 2009). Cation-chloride cotransporters that import chloride salts are driven by the transmembrane sodium gradient: there is much more sodium in the extracellular fluid than the cytoplasm, so sodium diffusion is inward (Russell 2000). Examples of these transporters are the NCC transporters of the renal tubule, and the electroneutral transporter NKCC1, which is expressed more widely, including in neurons (Plotkin et al. 1997; reviewed in Russell 2000). Electroneutral transporters move equal amounts of cations and chloride so that there is no change in the distribution of charge across the membrane. This means that the membrane potential is not changed by transport, and it also means that membrane potential does not alter transport rates. Electroneutral cation chloride cotransporters that export chloride are driven by the transmembrane potassium gradient: there are much higher concentrations of potassium in the cytoplasm than the extracellular space, so potassium diffusion is outward. Examples are KCC2, which is only expressed in neurons, and KCC3, which is widely expressed (Kahle et al. 2015). In non-neuronal cells, the cation-chloride cotransporters function to maintain cellular volume (Kahle et al. 2015; Hoffman et al. 2009). Uncontrolled loss of water due to exposure to hypertonic extracellular fluid rapidly induces apoptotic cell death, so cells need a mechanism such as NKCC1-mediated salt import to prevent that (Lang and Hoffman 2012). KCCX transporters prevent volume overload by exporting chloride salts. Because cells other than neurons express aquaporin water channels, the movement of salt is followed by osmotic flux of water (Hoffman et al. 2009). These transporters are linked to osmotic and cytoplasmic volume sensors via intracellular signaling pathways involving SPAK and WNK/OSR1 (Shekarabi et al. 2017).

In neurons, potassium-chloride cotransport was shown to be responsible for exporting chloride after a large GABA_A receptor-mediated chloride load (Misgeld et al. 1986; Thompson and Prince 1988). An idea regarding regulation of E_GABA was proposed based on the developmental profile of the inwardly directed cation chloride cotransporter NKCC1 (Plotkin et al. 1997b) versus the outwardly directed cation-chloride cotransporter KCC2 in neurons (Rivera et al. 1999). Immature neurons expressed NKCC1, the electroneutral sodium-potassium-chloride cotransporter, while mature neurons expressed KCC2, the electroneutral potassium-chloride cotransporter (Plotkin 1997b; Clayton et al. 1998; Rivera et al. 1999; Dzhala et al. 2005). The equilibrium point for these membrane transporters depends on the transmembrane gradients of the ions that they transport. For KCC2, the free energy change from ion transport is zero when:

For NKCC1, the equation is:

(Brumback et al. 2008). Using round numbers for the transported ion concentrations of K⁺_i = 120 mM, Na⁺_i = 10 mM, K⁺_o = 5 mM, Na⁺_o = 140 mM, and Cl⁻_o = 100 mM, NKCC1 would be at equilibrium at about 60 mM Cl⁻_i, while KCC2 comes to equilibrium at about 4 mM Cl⁻_i. This means that Cl⁻_i in immature neurons expressing NKCC1 would be 60 mM, so that from equation (4), E_GABA would be –18 mV, and GABA_A receptors would gate Cl⁻ efflux rather than influx. This would make GABA_A currents strongly depolarizing from resting membrane potential. GABA-mediated depolarization of developing neurons has been frequently observed (LoTurco et al. 1995; Dzhala et al. 2003; Tyzio et al. 2007; Kirmse et al. 2015). Adult neurons expressing KCC2 would have an E_GABA of –74 mV and a hyperpolarizing GABA_A response from a resting membrane potential of –70 mV. Hyperpolarizing GABAA responses have been frequently observed in adult CA1 pyramidal neurons, so this also makes sense.

So, to a first approximation, the shift from a depolarizing GABAA response in immature neurons to a hyperpolarizing response in mature neurons could be explained by the shift in transporter expression (Dzhala et al. 2005). However, there are significant problems with this explanation for the change in E_GABA with maturity. First, the cation-chloride transporters are volume regulators (Hoffman et al. 2009); if they are also tasked with regulating E_GABA, then E_GABA should be very strongly linked to neuronal size. Although E_GABA can be shifted with osmotic manipulations, the predicted relationship between E_GABA and neuronal size has not been observed (Glykys et al. 2019). Second, neuronal co-transport capacity is very high for both NKCC1 and KCC2-expressing CA1 pyramidal cells (Staley and Proctor 1999; Brumback et al. 2008; Hugenard et al. 1986; DeFazio and Hablitz 2001). This high rate of transport should clamp cytoplasmic chloride at the equilibrium point of the expressed cotransporter. Consequently, E_GABA should be fixed as well and vary little between neurons. However, this is not what has been observed (Huberfeld 2007). As reviewed in Rahmati et al. (2021) and discussed earlier, there is a wide range of E_GABA from neuron to neuron, and from study to study, even when those studies use identical techniques (perforated patch recordings) to study the same population of neurons (juvenile CA1 pyramidal cells). Several potential explanations have been proposed to the variance in E_GABA based on determination of cytoplasmic chloride by transport.

One possibility is that the equilibrium conditions of transport can be set by post-translational modifications of the transporters. NKCC1 transport of sodium, potassium, and chloride ions is generally considered to occur in an electroneutral 1:1:2 stoichiometry (Russell 2000), but that stoichiometry could be changed to alter the equilibrium chloride concentration without affecting electroneutrality. For example, a 1:2:3 ratio of sodium, potassium, and chloride transport would be electroneutral and would have a much lower equilibrium cytoplasmic chloride concentration (Brumback et al. 2008). If that stoichiometry was set by post-translational modifications, then E_GABA could be determined by these modifications. However, this would not apply to mature neurons expressing KCC2 (Rivera et al. 1999; Dzhala et al. 2005). This is because a 1:1 K:Cl ratio is the only electroneutral stoichiometry for potassium chloride cotransport. Thus, post-translational modifications in KCC2 cannot explain the variance in E_GABA in mature neurons (Huberfeld et al. 2007; Rahmati et al. 2021).

A second mechanism by which the observed variance in E_GABA can be explained is via active transport by both KCC2 and NKCC1 in the same neuron. Then there could be push-pull determination of the cytoplasmic chloride concentration, based on the number and activity of each transporter expressed in the membrane (Kaila et al. 2014). This could lead to precise determination of cytoplasmic chloride concentration and E_GABA. This mechanism could also explain inter-neuronal variance in E_GABA, if different numbers of NKCC1 and KCC2 were expressed in different neuronal membranes. However, this would entail constant, oppositely directed, nonproductive Cl transport, with NKCC1 importing chloride ions and KCC2 exporting the same chloride ions. This transport would rapidly degrade the Na⁺ and K⁺ gradients, which could only be restored by consumption of ATP by the action of the NaKATPase ion pump. This would impose a very heavy energy burden on neurons, the brain, and the organism. Neurons could avoid this energy cost by expressing both transporters, and using post-translational modifications such as phosphorylation to shut off either inward or outward chloride transport (Kahle). However, this would limit E_GABA to one of two values: that determined by NKCC1 equilibrium (about –20 mV) and that determined by KCC2 equilibrium (equal to the potassium reversal potential, about –100 mV). This does not solve the problem of continuously distributed values of E_GABA (Huberfeld et al. 2007; Rahmati et al. 2021).

The third mechanism is the pump-leak model (Dusterwald et al. 2018). This mechanism entails very modest chloride transport rates relative to chloride influx rates (Kay 2017). The idea is that transport rates are readily overwhelmed by chloride influx from GABA_A receptors and other sources of chloride influx, including membrane conductances, ion exchangers, and ion transporters. For KCC2, the cation-chloride cotransporters can be thought of as a bilge pump that is bailing out a leaky boat. The level of water in the boat, or concentration of chloride in the cytoplasm, represents the steady-state balance between chloride leaking in and transport pumping chloride out. The leak and pump fluxes would be oppositely directed for neurons expressing NKCC1. If different neurons have different chloride influx and cotransport rates, this model can explain interneuronal variance in cytoplasmic chloride and E_GABA. However, there are significant problems with this model, which is based on a model of red blood cells developed in 1960 (Kay and Blaustein 2019). This was well before the actual values of neuronal membrane resistance and GABA conductances were discovered. The first problem is that measured cotransport rates for KCC2 and NKCC1 in CA1 pyramidal cells greatly exceed the GABA-gated chloride flux (Lux 1971; Misgeld et al. 1986; Huguenard and Alger 1986; Staley and Proctor 1999; DeFazio and Hablitz 2007; Brumback et al. 2008). This would correspond to a boat with a really big pump and a very small leak, so there is no reason to expect much water in the boat. Pump-leak proponents have suggested that the measurement of membrane GABA-gated chloride flux in deafferented ex vivo brain slices is low relative to in vivo flux, and they are probably right (Doyon et al. 2016). However, low chloride influx would make the chloride concentration dependent only on the equilibrium condition of the pump. In other words, at low levels of chloride influx, the pump-leak model reduces to the “pump-only” model discussed earlier, so chloride would be distributed according to the equilibrium conditions of the transporter (equations 5 and 6). There would be no neuron-neuron variance in E_GABA in hippocampal slices with low rates of chloride influx. However, this is not what is seen; there is a great deal of interneuronal variance in neuronal chloride content (Huberfeld et al. 2007; Glykys et al. 2014; Sato et al. 2017) and E_GABA (Huberfeld et al. 2007; Rahmati et al. 2021). The second problem with this model is that the cytoplasmic chloride concentration and E_GABA should vary with time, as GABAergic activity ebbs and surges, and the pump is more-or-less overwhelmed. However, this is also not seen. For example, Rahamati et al. (2021) found that neuronal cytoplasmic chloride concentrations and E_GABA were stable with respect to time for at least 1 hour.

There are, however, some well-established situations where E_GABA is not stable with respect to time, and where the leak-pump model is relevant. These situations occur at very high rates of GABA_A receptor activation and correspondingly high rates of transmembrane chloride influx through GABA_A channels (Huguenard et al. 1986; Staley et al. 1995, 1999). Under these conditions, E_GABA becomes depolarized as cytoplasmic chloride accumulates faster than it can be transported out by cation-chloride cotransporters (Staley et al. 1995; Staley and Proctor 1999). Similar conditions can overwhelm NKCC1-mediated chloride transport (Brumback et al. 2008). Over the 40 years during which this phenomenon has been described, all reported cases entail pathological GABA_A activity driven by exogenous application of pharmacological GABA_A agonists, tetanic stimulation of GABAergic presynaptic neurons, or applications of convulsants (Barker and Ransom 1978; Alger and Nicoll 1979; Andersen et al. 1980; Wong and Watkins 1982; Scharfman and Sarvey 1987; Avoli and Perreault 1987; Cherubini et al. 1990; Perkins and Wong 1996; Kaila et al. 1997; Dallwig et al. 1999; Jin et al. 2005; DeFazio and Hablitz 2007). Under these conditions, E_GABA changes with time. This has been modeled by several investigators (Staley and Proctor 1999; Lewin et al. 2012; Doyon 2016). Under physiological conditions, transport capacity is sufficient to stabilize E_GABA, and the pump-leak model does not apply. This issue is addressed in more detail in the pathology section.

All GABA Signaling Is Local

If cation-chloride transporters do not determine the chloride concentration in neurons, what does? We have proposed that chloride is displaced by large, immobile polyanionic macromolecules such as nucleic acid polymers, actin, and tubulin (Glykys 2014, 2017; Rahmati 2021). The steady-state chloride concentration in a neuron is ~10 mM, but the potassium concentration is 135 mM, so there are 125 mM of anions that are not chloride in the cytoplasm. HCO₃⁻ accounts for 25 mM, but that still leaves 100 mM of anions that cannot be explained by monomeric sulfate and phosphate ions, which account for only a few millimoles (Morawski et al. 2015). Phosphates in nucleic acid polymers explain some of the 100 mM, but most of the anions are the amino acid moieties of proteins that are deprotonated at physiological cytoplasmic pH (Gianazza and Righetti 1980). Many proteins such as actin and tubulin are comprised of large numbers of such amino acids and, as a consequence, have very high negative surface charges at physiological pH (Sanabria et al. 2006; Janke et al. 2008). Not all amino acids in proteins are deprotonated at physiological pH, and the number of amino acids that are deprotonated vary depending on the amino acid composition of different proteins, and the number of any particular protein varies from cell to cell (Gut et al. 2018; Chen 2020), so the exact amount of anionic charge that is comprised of the deprotonated amino acid components of proteins will vary from neuron to neuron. Thus, there will be a corresponding variance in the concentration of mobile anions, HCO₃⁻ and Cl⁻, from neuron to neuron, depending on the quantities of each species of anionic macromolecule in individual neurons (Glykys et al. 2014). This variation in cytoplasmic chloride concentration would lead to a corresponding variance in E_GABA in each neuron. In the compilation of reported values for E_GABA reported in Rahmati et al. (2021), the group mean values for E_GABA determined by perforated patch recordings in juvenile CA1 pyramidal cells varied over a range of 25 mV. There was no correlation between the number of neurons sampled and the value of E_GABA, supporting the idea that inherent variance rather than inadequate sampling drove the variance in reported values.

Cytoplasmic Chloride Microdomains

The spatial distribution of proteins in the cytoplasm is very heterogenous (Gut et al. 2018; Chen et al. 2020). Presumably, the anionic amino acid moieties of cytoplasmic proteins have a correspondingly heterogenous distribution. This would lead to spatial inhomogeneity of charge distribution. But the cytoplasm is an ionic solution that acts as a conductor. Gauss’s law predicts no accumulations of charge in a conductor. That is to say, mobile cations such as potassium will be attracted to areas where proteins have high negative charge densities, and mobile anions such as chloride will be repelled from those areas. Similarly, chloride will be attracted to areas with higher concentrations of cationic amino acids, and potassium will be repelled. Cations and anions will continue to move until charge distribution is homogenous in the cytoplasm. Although this process evenly distributes charge within the conductive ionic milieu of the cytoplasm, it results in unequal local subcellular concentrations of mobile cations and anions.

The induction of spatial inhomogeneities in the distribution of mobile ions due to the distribution of immobile ions has been well described. Donnan was one of the first to apply these ideas to the cytoplasm (Donnan 1911). Donnan used the ability to cross the cell membrane to define mobile (permeant) versus immobile (nonpermeant) ions. He demonstrated a large effect of the distribution of immobile anions on the distribution of mobile anions. However, a membrane is not necessary to immobilize ions; large anionic crosslinked polymers are sufficiently immobile to induce similar effects (Procter, 1914; Fatin-Rouge et al., 2003). For example, these polymers and their effects on the distribution of mobile ions form the basis for water-softening and ion exchange technologies (Marinsky 1985; Helfferich 1995). In ion exchange chromatography, immobilized anionic charges can be used to attract mobile proteins that have positive surface charge at the selected pH. The proteins replace inorganic cations. Similarly, in water softening, polyvalent cations replace sodium bound to immobile anionic macromolecules. In both cases, the fixed anionic charges are altering the local distribution of mobile ions, both organic and inorganic. Ionic polymers do not need to be covalently crosslinked to be sufficiently immobile to induce inhomogeneities in the distribution of mobile ion concentrations; they just need to be much less mobile than the mobile ions (Marinsky 1985).

There are many anionic polymers in the cytoplasm with limited or no mobility. The cytostructural polymeric proteins actin and tubulin are two examples; these proteins have surface charge densities comparable to or greater than chloride (Sanabria et al. 2006). The distribution of these and other anionic polymers in the cytoplasm is clearly heterogenous (Koleske 2013; Morawski et al. 2015; Gut et al. 2018; Chen et al. 2020). Thus, it is reasonable to expect that the local concentrations of mobile cytoplasmic ions will vary inversely with the concentrations of these anionic polymers.

The physical dimensions over which the mobile ions vary is not known. It is easy to imagine that inhomogeneities in charge distribution could be limited to the Debye layers of the immobile anions. On the other hand, if this were localized to only the ions in the water of the hydration layers of proteins, the mean value for the cytoplasmic chloride concentration (~10 mM) should be closer to the bulk value for the potassium concentration (~135 mM). One solution to this problem is that most of the water in the cytoplasm may be in the hydration layers of macromolecules; that is, the cytoplasm is a gel of heterogenous macromolecules with little free water (Ellis 2001).

Role of Transporters in the Chloride Distribution by Displacement Model

If local cytoplasmic chloride concentrations are determined by displacement by large immobile charged polymers, what is the role of cation-chloride transport? The transporters maintain the distribution of chloride concentrations. Consider a low-chloride cytoplasm. If chloride salts cross the membrane through transport or the coactivation of glutamate and GABAA synapses, then the cytoplasm will either become hypertonic or, if water follows salt influx, swell. Then the action of KCC2 and KCC3 will be the same in neurons as in other cells: by exporting that salt, they maintain cytoplasmic osmotic strength and volume. Similar considerations would hold for NKCC1 in a high-chloride cytoplasm. An analogy would be an automobile: the driver determines the direction and speed, but the action of an energy-consuming motor is required to maintain speed in the determined direction. The number of motors, their size, and modifications of the motor(s) would only be important at extreme speeds.

Experimental Analysis of Chloride Distribution by Displacement

Rahamti et al. (2021) evaluated the distribution of chloride in the cytoplasm of neurons using two-photon microscopy of the ratiometric chloride-sensitive transgenic fluorophore SuperClomeleon (Grimley et al. 2013), which is based on the chloride sensitivity of yellow fluorescent protein emission, as well as fluorescence lifetime imaging (FLIM) of the single-wavelength fluorophore 6-methoxy-N-ethylquinolinium (MEQ; Verkman et al. 1989), and electrophysiological measures of E_GABA. Rahamti et al. found that there was a continuous distribution of cytoplasmic chloride concentration microdomain sizes that almost certainly included volumes that were below the 0.4 µm XY resolution of the two-photon microscopy techniques employed. The chloride concentrations in the microdomains of pyramidal cells in organotypic slice cultures varied from 2 to >50 mM, with 10 mM being the most common value. These organotypic slice cultures are considered a model of posttraumatic epilepsy (Lillis et al. 2015), and the slices from which these images were taken were likely undergoing periodic seizures. Images were obtained in 1 µM tetrodotoxin to block seizure-induced changes in dendritic chloride.

The continuously varied chloride concentrations in the cytoplasmic microdomains are difficult to reconcile with high-velocity equilibrative membrane chloride transport. As discussed above, cytoplasmic chloride values near 5 mM are readily explainable based on KCC2 transport, and cytoplasmic chloride values near 60 mM are easy to explain based on NKCC1 transport, but the vast majority of recorded subcellular chloride values were substantially different from both of these values. How can equilibrative cation-chloride transporters with defined chloride equilibrium chloride concentrations be in equilibrium with so many values of cytoplasmic chloride? We had previously proposed that the differences could be explained on the basis of the free energy of water transport in the calculation of the equilibrium conditions for NKCC1 and KCC2 cotransport (Delpire and Staley 2014). Neurons have no aquaporin water channels (Papadopoulos and Verkman 2013). Thus, transporters could be the only means of water flux across the membrane. If so, then very small differences of osmotic pressure across the membrane could result in free energy changes for water transport that are comparable to K Na and Cl free energy changes. This is because the salts are transported along with water in roughly isotonic proportions, so that there are hundreds of water molecules transported with each pair of ions (MacAulay et al. 2004).

Another possibility is easier to grasp conceptually once the existence of cytoplasmic chloride microdomains is appreciated. If the local concentration of mobile chloride ions in the cytoplasm is affected by the distribution of charges on large immobile macromolecules, the same should be true for any mobile ion in the cytoplasm, including potassium and sodium. This would result in potassium and sodium microdomains that could dramatically alter the local concentration gradients for these ions across the cell membrane. Under these conditions, the wide range of chloride concentrations observed by Rahmati et al. would reflect the local equilibria of cation-chloride transmembrane transporters. Thus, transporters could be at equilibrium over the wide range of chloride concentrations observed in neuronal cytoplasms (Fig. 71–2).

Figure 71–2.

If the concentrations of both potassium and chloride ions covary with the local surface charge density on immobile ions, KCC2 can be in equilibrium with local cytoplasmic chloride concentrations that range from 5 to 125 mM, each of which satisfies equation (more...)

A last point to make before moving to pathologies: Rahmati et al. also found that, as a consequence of these varied concentrations of cytoplasmic chloride, the value of E_GABA varied locally. In fact, for every presynaptic interneuron she studied, there was a unique value for E_GABA at the interneuron’s GABAergic synapses on the postsynaptic pyramidal cell. For individual pyramidal cells, the range of E_GABA for different presynaptic interneurons was up to 20 mV. This means that GABAergic synapses, which are considered “untunable” in computational analyses (Neill and Scanziani 2021), are in fact individually tunable based on their values for E_GABA. The reversal potential at each synapse will determine the magnitude and direction of GABA_A currents under different conditions, for example, from resting membrane potential. Just as membrane current at glutamatergic synapses are determined by their plastic conductances, membrane currents at GABAA synapses could be determined by their plastic driving forces. While plasticity of these E_GABA values has not yet been demonstrated, the wide range and stability (over the 1-hour course of a perforated patch recording) suggest that E_GABA at individual synapses provide a rich substrate for signal processing and information storage.

Seizures Associated with Acute Brain Injury

A wide variety of brain injuries, including trauma, focal and global hypoxic-ischemic injury, infection, and hemorrhage, are complicated by cytotoxic cerebral edema and acute seizures (Vespa et al. 1999; Hantus et al. 2019; Griffith et al. 2020; Zhou et al. 2021). Why are edema and seizures so consistently associated with such a wide variety of injury mechanisms? One possibility is that salt and water influx into neurons after injury results in neuronal swelling while the chloride moiety of the salt shifts the E_GABA to more positive values. Such a shift in E_GABA could result in compromise in inhibition to the point of generating seizure activity (Glykys et al. 2017). Such seizures would not be expected to respond to most anticonvulsants that are predicated on an inhibitory effect of GABAA receptor activation (Dzhala et al. 2009). Instead, treating these seizures would require a reduction in the salt influx into the neurons. This might have the added benefit of reducing cerebral edema. Of course, reducing salt influx into neurons requires an understanding of the mechanisms of salt influx. Unfortunately, our understanding of these mechanisms is incomplete, but it is improving.

Salt Flux Associated with Neuronal Injury

The canonical explanation for salt influx into neurons after injury is compromise of the NaKATPase due to loss of ATP production by oxidative phosphorylation (Simard et al. 2007; Stokum et al. 2016). A strength of this explanation is that in hypoxic-ischemic injury there is neither oxygen nor carbon for oxidative phosphorylation. However, there are problems with this explanation. First, compromise of NaKATPase happens within minutes, but the time course of cerebral edema (Loubinoux et al., 1997; Leinonen et al. 2017) and seizures (Vespa et al. 1999; Wusthoff et al. 2011; Lynch et al. 2012) after injury is much slower, starting hours after injury and peaking at 24–72 hours after injury. Second, seizures and edema are reperfusion phenomena; that is, they occur after reperfusion, when NaKATPase should have full access to oxygen and carbon substrates. Third, when NaKATPase is blocked, intracellular sodium goes up, but cells do not swell (Glykys et al. 2014). This is likely because NaKATPase is a sodium-potassium exchanger, so potassium concentrations drop as intracellular sodium increases. Finally, NaKATPase dysfunction does not explain the accumulation of chloride; instead, sodium and potassium concentrations change.

Other Mechanisms of Salt Influx

A second potential mechanism of neuronal swelling and seizures after brain injury is a change in membrane permeability. Neuronal membrane permeability increases after injury (Coulter et al., 1992; Pisani et al. 1998; Farkas et al., 2006; Whalen et al., 2008). Pannexins are known to be inserted into the cell membrane after neuronal injury, and these channels are nonspecifically permeable to cations and anions (Thompson et al. 2006; Weilinger et al. 2012). Consider a pannexin-mediated nonspecific cation and anion conductance inserted into the membrane near a low-chloride, high-cation microdomain. We presume that prior to the pannexin insertion, the action of NaKATPase maintains a high cytoplasmic potassium concentration and low sodium concentration. Using round numbers, under these conditions, E_K and E_Cl could be equal and ~ –100 mV, while Ena will be ~ +100 mV. When the pannexin conductance is added, the equilibrium conditions for K and Cl flux change from equation (5) to the Donnan equilibrium condition for nonspecific cation and chloride membrane permeability in the presence of immobile anions:

(Delpire and Staley 2014). This change in permeability results in a large influx of sodium and chloride as well as an egress of K, so that E_K = E_Cl = E_Na = ~ –10 mV at steady state. At steady state, the neuron will be inexcitable because resting membrane potential will also be depolarized by the positive shift in E_K, and action potentials are no longer possible when E_Na = E_K. However, if the pannexin conductance increases gradually as pannexin channels are added, the neuron will transition from the normal state through a period where GABA_A responses in this part of the neuron are strongly depolarizing and could contribute to seizure initiation and anticonvulsant resistance, before reaching the unexcitable state (Thompson et al. 2008).

Equation (7) provides an important insight into cytotoxic cerebral edema. If we take our round number approach to normal physiology, with 4 mM chloride in the local cytoplasm and 100 mM chloride in the nearby extracellular space, then the reversal potential for chloride from equation (4) is –69 mV. To get to –10 mV, the intracellular chloride needs to increase to almost 70 mM. This increase in the concentration of chloride must be accompanied by a cation, so the sum of potassium and sodium concentrations must also increase by almost 70 mM. This, in turn, implies a massive increase in the cytoplasmic volume, because all these changes must be isotonic; that is, water will accompany the influx of sodium and potassium chloride salts. This influx would handily explain the slow neuronal volume increases observed in cytotoxic edema.

Not surprisingly then, perhaps the most effective experimental strategies for preventing cytotoxic neuronal swelling and death are limiting the influx of chloride from the extracellular space (Goldberg and Choi, 1993; Ko et al. 2014). This is most convincingly done by removing chloride from the extracellular fluid (Goldberg and Choi, 1993; Ko et al. 2014). Many investigators have also tried to block chloride entry pharmacologically, with varying success (Wei et al. 2004; Kumar et al. 2006; Madry et al. 2010; Weillinger et al. 2012). Pharmacological approaches are never as successful as removing extracellular chloride, likely because of the heterogeneity of chloride pathways that may appear in a damaged neuronal membrane. Nevertheless, pharmacological blockade of chloride entry is a more feasible translational strategy than removal of chloride from the brain’s interstitial fluid.

Bumetanide Trials for Neonatal Seizures

NKCC1 is a promising target for limiting chloride influx into damaged neurons. NKCC1 is a high-velocity transporter that, by virtue of the cotransported sodium ion, has an equilibrium cytoplasmic chloride concentration that is ~60 mM, as discussed in the physiology section. Thus, NKCC1 will transport chloride into damaged neurons until E_Cl reaches the steady-state value of –10 mV discussed above. Blocking NKCC1 would reduce chloride influx into damaged neurons, as well as into high-chloride cytoplasmic microdomains of undamaged neurons. The end effect should be less excitatory GABA signaling. Accordingly, several investigators have demonstrated that the NKCC1 antagonist and loop diuretic bumetanide reduces seizure activity in vitro (Dzhala et al. 2005, 2008).

The activity of bumetanide in vivo is a more complex story. Although we initially considered bumetanide as a treatment for neonatal seizures based primarily on the in vitro data (Dzhala), several investigators subsequently questioned whether bumetanide crossed the blood–brain barrier sufficiently to reach brain concentrations that would block neuronal NKCC1. Although it would seem simple to test this in animals to determine whether bumetanide is effective, the pharmacokinetics of bumetanide in rodents is quite different than in humans: the half-life of bumetanide is 10 minutes in adult rats, compared to 90 minutes in adult humans (Halladay et al. 1978; Pentikainen et al. 1977). Bumetanide levels were low in the brains of neonatal rodents exposed to hypoxia sufficient to induce seizures but not brain injury (Cleary et al. 2013). But bumetanide was effective in vivo in a broad spectrum of brain injuries: in perinatal rat pups (Dzhala et al. 2005) and adult rats (Sivakumaran and Maguire 2016) in whom seizures were triggered by kainic acid, or in rat pups after electrical kindling (Mazarati et al. 2009), and adult rats after global hypoxic-ischemic injury (Liu et al. 2012) and stroke with reperfusion (Yan et al. 2003; reviewed in Tao et al. 2019) and trauma (Lu et al., 2008). The only experimental situations in which bumetanide was not successful in vivo was stroke in rat pups that was not followed by reperfusion (Kang et al. 2015) or direct injury to the blood–brain barrier in adult rats (Wilkinson et al., 2019). These findings, together with the robust experimental finding that reduced extracellular chloride prevents neuronal swelling due to chloride salt accumulation (Goldberg and Choi 1993), raises the possibility that bumetanide limits the influx of salt into the brain and that this was consequently limiting chloride flux into neurons. NKCC1 is present at the blood–brain barrier (Odonnell et al., 2004), and bumetanide serum concentrations of approximately 1 µM (Cleary et al. 2013) would be sufficient to block NKCC1 (Gillen et al., 1996) at the blood–brain barrier. Whether this is the true site of action of bumetanide in vivo, or whether the blood–brain barrier in the setting of immaturity, brain injury, and seizures is more permeable to bumetanide, remains to be seen. In the meantime, two human trials of bumetanide for neonatal seizures have been reported. Most neonatal seizures are due to acute brain injury, so we will discuss these trials in this section.

Human Bumetanide Trials

The first trial of bumetanide to be published was the NEMO trial (Pressler et al., 2015). This was an open-label, dose escalation, electrographically defined seizure trial with no control group. The trial was halted after enrolling 14 patients because of concern of possible ototoxicity: 3 of 11 surviving participants developed hearing loss. Although hearing loss is also associated both with the underlying hypoxic-ischemic brain injury as well as the aminoglycosides that were administered to two of the three infants (Smit et al., 2013), in the absence of a control group (hypoxic ischemic injury and aminoglycoside exposure but no bumetanide), it was most prudent to halt the trial. The published report also concluded that bumetanide was not effective, although this conclusion was based, in large part, on absence of effect in 5 of the 14 patients who had no seizures during the baseline period, and therefore could not have responded to bumetanide (Thoresen and Sabir 2015). When these patients were excluded, bumetanide appeared effective, but without a control group, the effect of bumetanide could not be differentiated from the natural history of neonatal seizures after brain injury (Wusthoff et al. 2011; Lynch et al. 2012).

Fortunately, the other trial of bumetanide for neonatal seizures included a placebo control group as well as randomization and treatment blinding (Soul et al. 2021). To ethically include a control group, controls were treated using current best therapy (escalating phenobarbital doses) while the experimental group received the same phenobarbital therapy, but with the addition of bumetanide if the initial phenobarbital doses failed to control electrographic seizure activity. This phase I–II trial, although not designed to test efficacy, nevertheless demonstrated a strong efficacy signal. Higher doses of bumetanide were associated with larger reduction in seizure burden (amount of time spent seizing). However, the severity of seizures was not equivalent in the control and bumetanide arms; by chance, the bumetanide arm had much more severe seizures. Some have questioned whether the effect of bumetanide might then simply reflect regression toward the mean in a population with initially more severe seizures. This is very unlikely for two reasons. First, neonatal seizures do not regress toward the mean; more severe seizures respond much more poorly to anticonvulsant therapy (Painter et al. 1999; Glass et al. 2019). Second, because this was a phase I–II trial, only a single dose of bumetanide was given, so that seizure activity was unaffected several hours later. The total seizure burden (all recorded seizures before and for the duration of the long-term electroencephalographic [EEG] monitoring after bumetanide) was also more severe in the bumetanide treatment group than the control group, indicating that there was no regression toward the mean.

Should bumetanide be used for neonatal seizures now? There are several caveats. First, bumetanide was used only in patients who did not respond to the standard loading doses of phenobarbital. The majority of enrolled patients did respond to the initial phenobarbital doses. Whether bumetanide would be more effective in patients who responded to phenobarbital is not known. In a recent levetiracetam trial for neonatal seizures, levetiracetam was given at the first seizure, that is, before phenobarbital failure, in patients randomized to the levetiracetam arm, and in that trial, phenobarbital was much more effective than levetiracetam (Sharpe et al. 2020). Thus, it is important to recognize that in the bumetanide trial, bumetanide was given in conjunction with phenobarbital, and only in those patients in whom standard phenobarbital therapy had failed. These are likely the most severely brain-injured patients (Painter et al. 1999; Glass et al. 2019). Second, the controlled bumetanide trial was small, and seizure severity was significantly different in the two groups. Although this should have biased the trial against bumetanide efficacy, it also demonstrates that efficacy trials require larger numbers of enrolled patients. Hopefully such a trial can be undertaken soon.

If bumetanide works at the blood–brain barrier to limit chloride influx into the brain and thence into neurons, would it work in acute brain injury in older children and adults? This is an intriguing question. Studies carried out in adult animals show efficacy of bumetanide in limiting cerebral edema after brain injuries (Yan et al., 2003; Lu et al., 2008, 2012; Tao et al. 2019), and seizures after kainate (Sivakumaran and Maguire 2016). A single case report suggested that bumetanide may be effective in older infants (Kahle et al. 2009), but appropriately powered human adult and pediatric trials will be necessary to address this question conclusively. It is also important to remember that bumetanide was ineffective in an experimental model in which the blood–brain barrier was extensively damaged (Wilkinson et al. 2019). When the blood–brain barrier is extensively damaged and passive permeability is increased, blocking NKCC1 may not limit NaCl flux into the brain. There may be clinical conditions where this is also the case, for example, conditions characterized by extensive enhancement of brain by the administration of intravenous contrast agents (Schneider et al. 2004; Durukan et al. 2009).

Temporal Variance in E_GABA

As discussed in the section on the pump-leak model, when GABA_A receptors are activated at pathologically high intensity by exogenous agonists, tetanic stimulation of afferents, or seizure activity, chloride influx can transiently exceed the transport rate of cation-chloride cotransporters (Staley et al. 1995, 1999; Jin et al. 2005; Brumback et al. 2008; DeFazio and Hablitz 2007; Lewin et al. 2012; Doyon et al. 2016; Lombardi et al., 2019), resulting in cytoplasmic chloride accumulation and a depolarizing shift in E_GABA that can be sufficient to trigger action potentials (Staley et al. 1995). Such activity-dependent positive shifts in the GABA_A reversal potential provide a potential mechanism for temporal decreases in the efficacy of inhibition that could contribute to ictogenesis (Lillis et al. 2012; reviewed in Staley 2015). Such activity- and time-dependent decreases in inhibitory efficacy are thought to underlie the seizures that occur in the presence of the potassium channel blocker 4 aminopyridine (4AP), which preferentially increases the firing of interneurons and subsequent GABA release (Lopanstev and Avoli 1998), and optogenetic chloride loading (Alfonsa et al. 2015).

Several anticonvulsant strategies are likely to suppress activity-dependent disinhibition. Acetazolamide is an anticonvulsant that antagonizes carbonic anhydrase, the enzyme that catalyzes the hydration and dehydration of CO₂ to carbonic acid (H⁺ and HCO₃⁻). The action of carbonic anhydrase, in combination with pH buffering, stabilizes the driving force for HCO₃⁻ efflux through GABA_A channels, even at very high levels of GABA_A receptor activation. Thus, E_HCO3, which is about –12 mV, is more stable than E_Cl at high rates of GABA_A receptor activation, so that E_GABA moves closer to E_HCO3 as cytoplasmic Cl accumulation drives to more positive values (Staley and Proctor 1999). Figure 71–1 By destabilizing E_HCO3, acetazolamide diminishes the depolarizing component of the GABA_A current, and this effect is most pronounced at high current densities, that is, high rates of GABA_A receptor activation. Without the depolarizing current, chloride accumulation is diminished because E_Cl = RMP and there is no movement of chloride across the membrane. Thus, acetazolamide can limit ictogenic activity-dependent degradation of inhibition due to positive shifts in E_GABA (Hamidi and Avoli 2015).

Figure 71–1.

Stabilization of the HCO₃⁻ transmembrane gradient by diffusion of CO₂ across the membrane and into the cytoplasm in concert with the action of carbonic anhydrase, which catalyzes the dehydration of HCO₃⁻ in the extracellular space as (more...)

An anticonvulsant that does not interfere with low-level GABAA activity, but reinforces high-level GABA activity, would seem to be the ideal way to remove activity-dependent disinhibition while minimizing side effects. However. acetazolamide is not an ideal anticonvulsant because the depolarizing effect of HCO₃⁻ efflux can be replaced by concurrent activation of ionotropic glutamate receptors, which occurs during seizures. Inward current through glutamate receptor-operated channels provides the depolarizing drive for cytoplasmic chloride accumulation and positive shifts in E_GABA even in the absence of HCO₃⁻ efflux. Nevertheless, acetazolamide should be kept in mind in conditions where intense GABA_A receptor activation is induced pharmacologically, for example, when continuous infusions of benzodiazepines or barbiturates are used to control status epilepticus (Tasker et al. 2016). Acetazolamide is available for intravenous administration and could be a useful adjunct to control seizures under these circumstances (Staley 2004; Hamidi and Avoli 2015).

Newer approaches to the problem to activity-dependent disinhibition have focused on enhancing the activity of KCC2, the potassium-chloride cotransporter. Human epilepsies due to loss-of-function mutations in SLC12A5, the gene coding the KCC2 protein, have recently been identified (Duy et al. 2019; Akita 2020). This supports experimental findings that reduced KCC2 function may be proconvulsant (Kelley et al. 2016; Chen et al. 2017; Dzhala et al. 2021). If reduced KCC2 function is proconvulsant, would enhanced KCC2 function be anticonvulsant? If the outward transport of chloride could be enhanced, then higher levels of GABA_A receptor-gated chloride influx could be sustained before cytoplasmic chloride accumulation began to shift E_GABA so far positive as to become excitatory. Strategies for KCC2 enhancement include genetic enhancement (Magloire et al. 2019) and pharmacological enhancers of potassium-chloride transport (Gagnon et al. 2013; Delpire 2021). These enhancers demonstrate anticonvulsant efficacy in experimental preparations (Moore et al. 2018; Dzhala et al. 2021). Unlike acetazolamide, the efficacy of KCC2 enhancers should not be compromised by depolarizing glutamate-gated cation influx. However, like the osmotic enhancement of KCC2 activity (Glykys et al. 2016), pharmacological enhancement of KCC2 activity is an incompletely effective anticonvulsant strategy. One important limitation of this approach is the effects on intracellular signaling pathways that regulate cytoplasmic volume. As stated earlier, loss of volume is a robust mechanism for triggering apoptosis (Hoffman et al. 2009), and some early KCC2 enhancers were toxic (Kikuchi et al. 2015). A second limitation is extracellular potassium accumulation. High rates of potassium-chloride export solve the problem of cytoplasmic chloride accumulation, but replace it with the problem of extracellular potassium accumulation (Vitanen et al. 2010). This potassium must either be buffered by astrocytes (Witthoft et al. 2013) or pumped back into neurons. Both processes require energy input in the form of the activity of the sodium-potassium exchanger NaKATPase. But energy is in short supply during seizure activity (Wasterlain et al. 2010). Thus, while enhancement of KCC2 activity is an intriguing anticonvulsant strategy, there remain significant hurdles still to be overcome prior to implementation.

Before closing the section on anticonvulsant manipulations of ion transport, it is worth mentioning the remarkable anticonvulsant activity of high concentrations of the loop diuretic furosemide. A dose of 1–5 mM furosemide completely blocks seizures while leaving evoked responses and brief interictal activity intact (Hochman et al. 1995; Dzhala et al. 2021). Both NKCC1 and KCC2 are blocked by furosemide, but at much lower concentrations. The anticonvulsant mechanism of furosemide has not been elucidated, in part because the high concentration likely entails nonspecific effects. Some of these nonspecific effects may involve GABA_A signaling, for example, block of some classes of GABA_A receptors (Banks et al. 1998), and some may not, for example, interference with glutamate reuptake (Roseth et al. 1995). Nevertheless, the anticonvulsant effects of high concentrations of furosemide remain an intriguing clue as to mechanisms of ictogenesis and anticonvulsant strategies.

Spatial Variance in E_GABA

If all GABA_A signaling is local, as implied by the Cl⁻ cytoplasmic microdomains described by Rahmati et al. (2021), then GABA_A receptors may be more tunable than originally thought (Neill and Scanziani 2021). E_GABA could provide a unique mechanism for tuning the strength of inhibitory synapses based on the degree of local membrane hyperpolarization or depolarization produced by activation of the GABA_A receptors at that synapse and the difference between the local E_GABA and membrane potential. If E_GABA is local, heterogeneous, and stable (Rahmati et al., 2021), and E_GABA affects local neuronal signal processing (Doyon et al., 2016), it would seem necessary to closely regulate E_GABA. Such regulation could encode information, including the computational rules of the local dendrite and associated synapses. Whether such regulation occurs, how that regulation is accomplished, and whether it is plastic and experience-dependent, remain to be discovered. The overall shift of E_GABA from early development to maturity indicates that E_GABA can be stably altered, but how that translates locally remains to be seen. Thus, the next section is speculative but provides new avenues of investigation into causes and treatments of epilepsy.

Chronic Epilepsy

Twenty years ago, an intriguing clue as to a cause of medically intractable epilepsy was discovered by Richard Miles and colleagues (Cohen et al. 2002). They recorded from human tissue resected for mesial temporal sclerosis and epilepsy control. Few pyramidal cells were found in the sclerotic hippocampi. However, a subset of neurons in the subiculum demonstrated depolarizing responses to synaptically released GABA. These recordings were performed using sharp electrodes filled with potassium acetate, and acetate has been reported to depolarize E_GABA (Eccles et al., 1977). However, this effect was not seen in later studies (Bormann et al. 1987). A subsequent study by Miles et al. (Huberfeld et al., 2007) in human subicular tissue resected for epilepsy control demonstrated a continuously distributed E_GABA from –50 to –80 mV, presaging the experimental distributions of E_GABA reviewed and measured in Rahamati et al. (2021). Huberfeld et al. (2007) struggled to explain this distribution based on the presence of KCC2 and NKCC1 in the recorded cells, but found that neurons with depolarizing GABA_A responses could express KCC2 based on immunohistochemical assays. In retrospect, these human data strongly support the distribution of neuronal cytoplasmic chloride via displacement by immobile anionic macromolecules, and the support of a range of cytoplasmic chloride concentrations by a single equilibrative transporter (Fig. 71–2).

Many studies from this era that sought to connect cation-chloride cotransporter expression to human (Munakata et al. 2007; Karlocai et al. 2016) and experimental epilepsy (Gonzalez 2016) with mixed results. The physiology section of this review provides an explanation for these mixed results: cation-chloride transporters serve to maintain the local E_GABA, rather than set or determine the local E_GABA. As shown in Figure 71–2, a hyperpolarizing GABA_A response can be maintained by KCC2, but higher chloride concentrations can also be maintained by KCC2 operating at its equilibrium condition. Thus, a perfect correlation between E_GABA and transporter species should not be expected (Huberfeld 2007). Rather, the local E_GABA is set by the local compliment of immobile anions and cations. The cation-chloride cotransporters serve a critical function: to maintain the local concentration of chloride after GABA-gated chloride flux.

If all GABA signaling is local, this creates a substantial pharmacological problem. GABA_A synapses that depolarize the local membrane may provide critical amplification in a circuit that ultimately inhibits neuronal firing. Examples could include inactivation of dendritic sodium conductances on pyramidal cells, excitation of a pyramidal cell that in turn activates many inhibitory interneurons, and excitation of interneurons. Any global pharmacological maneuver to block depolarizing GABA_A responses will also block such depolarizing responses that ultimately inhibit the network and serve as a critical anticonvulsant mechanism.

If all GABA_A signaling is local, then we need to be able to manipulate the mechanisms of plasticity that alter the GABA_A reversal potential. This may entail the turnover of not only intracellular macromolecular anions but also extracellular macromolecular anions that displace chloride. In the intracellular space, plastic mechanisms include addition of local immobile anions via polyglutamylation of actin (Janke et al. 2008). In the extracellular space, these immobile anions would be comprised of the sulfated glycosaminoglycan moieties of the extracellular matrix (Glykys and Staley 2017). The enzymes, transporters, and signaling mechanisms that enact these changes remain to be discovered, but provide a rich opportunity for our understanding of ictogenesis and novel anticonvulsant strategies.