Diuretics and epilepsy: Will the past and present meet?


Address correspondence to Edward H. Maa, 777 Bannock Street, Mail Code 4000, Denver, CO 80204, U.S.A. E-mail: edward.maa@dhha.org


Clinical studies from over half a century ago suggested efficacy of a variety of diuretics in focal and generalized epilepsies as well as in status epilepticus, but these findings have not been translated into modern epilepsy training or practice. Recent advances in our understanding of neuronal maturation and the pathophysiology of neonatal seizures provide fresh insight into the mechanisms by which diuretics might reduce susceptibility to seizures. In vitro and in vivo rodent studies and human epilepsy surgical cases have shown that specific diuretic agents targeting the cation-chloride cotransporters decrease neuronal synchrony and neuronal hyperexcitability. These agents are thought to convey their antiepileptic activity by either expanding the extracellular space or promoting a cellular chloride transport balance that reflects a more developmentally “mature,” less excitable state. It may be time to reexamine whether diuretics could serve as adjunctive therapies in the treatment of refractory epilepsies.

Despite advances in antiepileptic therapy, one-third of patients remain refractory to medical treatment (Kwan & Brodie, 2000; Mohanraj & Brodie, 2006). For mesial temporal sclerosis (MTS), long-term seizure control from surgical resection can be achieved in only half of patients (Tellez-Zenteno et al., 2005; Schmidt & Stavern, 2009). These data reveal a need for the development of novel antiepileptic drugs (AEDs), which is dependent on a better understanding of epilepsy pathophysiology.

Antiseizure effects of diuretics were first reported in 1938, when a patient was documented to exhibit improvement in absence seizure frequency after receiving sulfanilamide for the treatment of streptococcal tonsillitis (Yeoman). In 1952, Bergstrom et al. described their observations of the antiepileptic efficacy of another sulfonamide, acetazolamide. In 1959, Lombroso and Forxythe reported that acetazolamide was effective at controlling primary generalized epilepsies, including absence and generalized tonic–clonic seizures, but was of little benefit at treating focal epilepsies and catastrophic epilepsy syndromes; however, they noted that after prolonged therapy, acetazolamide lost most of its effectiveness. In 1955, Millichap et al. compared the antiseizure effects of sulfanilamide and acetazolamide, and concluded the increased antiepileptic efficacy of acetazolamide was due to its improved ability to inhibit carbonic anhydrase. Millichap later hypothesized that the mechanism of action of the ketogenic diet (KD) was probably similar to that of acetazolamide (1969).

Although these studies suggested a role for diuretics as a general class of AEDs, the longer-term effectiveness and broader coverage of seizure types offered by phenobarbital, phenytoin, and valproate, as well as the obvious side effects of diuretics, likely contributed to diuretic compounds falling out of favor. Renewed interest in diuretics as AEDs emerged when large epidemiologic studies from the Mayo Clinic revealed that a history of diuretic use appeared to decrease the incidence of first-time, unprovoked seizures in Olmsted County, between 1955 and 1984 (Hesdorffer et al., 1996). These findings were bolstered by animal studies that showed that furosemide and chlorothiazides were protective against the maximal electroshock model of seizures, (Hesdorffer et al., 2001).

As recent developments in our understanding of neuronal maturation and the pathogenesis of neonatal seizures, temporal lobe epilepsy, and cortical migration defects have reinvigorated the debate as to whether diuretics are effective antiepileptics, we present the research supporting two different hypotheses regarding the anticonvulsant effects of diuretics: (1) ephaptic effects and (2) effects on the chloride gradient that regulates γ-aminobutyric acid (GABA) signaling. We conclude that diuretics should be reexamined as adjunctive therapies in the treatment of refractory epilepsies.

Seizures and the Extracellular Space

Water movement in brain tissue

Human plasma osmolality is homeostatically maintained at 287 mosmol/kg H20 (mOsm), varying by ±1%. However, acute hyperosmolality (and the resulting “shrinkage” of cell volume) promotes the reactive intracellular accumulation of Na+, K+, and Cl via regulatory volume increase (RVI); a more chronic exposure to hyperosmolar conditions causes brain cells to generate “idiogenic osmoles” (primarily free amino acids) to combat cell shrinkage (Hoffmann et al., 2009) (Fig. 1). Acute hypoosmolality promotes the transport of Na+, K+, and Cl out of cells via regulatory volume decrease (RVD), and the loss of intracellular “idiogenic osmoles” in order to combat cell swelling. However, hyperacute osmotic changes that occur within minutes are not buffered by these adaptive mechanisms but instead allow rapid water shifts across membranes between cells (neurons and glia) and the extracellular space (ECS), resulting in rapid cell swelling in hypotonic ECS and cell shrinkage in hypertonic ECS (Andrew, 1991). Glial cells typically show faster rates of volume change because they contain aquaporins, whereas neurons do not (Amiry-Moghaddam & Ottersen, 2003; Tait et al., 2008).

Figure 1.

Mechanisms of cell volume regulation. RVI, regulatory volume increase; RVD, regulatory volume decrease. See text for details. From: Hoffmann et al., 2009 (with permission).

Cerebral hydration and seizures

In studies of water intoxication in 1926, Rowntree administered pituitary extract (adrenocorticotropic hormone, ADH) with increasing amounts of free water and induced generalized seizures in animals, which were prevented by injection of hypertonic NaCl in follow-up experiments.

Carter described the experience of his group using hypertonic urea to treat status epilepticus in institutionalized patients (1962). Based on a prior series of more than 1,000 cases of status epilepticus, their use of barbiturates controlled status within 10 min, but required an average of four injections or 24 h of treatment to maintain seizure control. Their barbiturate therapy success rate was 90% but with 10% mortality. In the hypertonic urea series, they infused 57 patients with intravenous urea (half of the patients had previously received a standard barbiturate dose but continued seizing). Status epilepticus was stopped in all 57 patients, 51 patients having stoppage between 3 and 6 min, the longest amount of time being 10 min. There was no mortality in this series (1962).

Urea, a water-soluble and poorly membrane-permeable compound, increased the electroshock threshold of rats in studies by Reed and Woodbury, who felt its mechanism “is probably related more closely to the shifts of water and electrolytes between the intracellular and extracellular compartments of the brain than to the gross changes in total brain water and electrolytes” (1964). They concluded from these findings that hyposmolar conditions and resultant cell swelling and ECS shrinkage appear to promote seizures, whereas hyperosmolar conditions and resultant cell shrinkage and ECS expansion appear to prevent seizures (1964).

Because of their interest in the brain edema that accompanies kainic acid (KA)–induced seizures, Baran et al. performed a series of experiments with mannitol administration at different points in time following KA injection (1987). They discovered that mannitol infused during the period of time associated with cerebral edema attenuated status epilepticus and prevented the typical irreversible postepileptmic brain damage (if given within 10 min after the onset of generalized convulsion). Not only did the seizures stop after mannitol treatment, but noradrenaline levels measured 3 h after KA injection were within the control range. Mannitol infused before or after this period had little effect (1987).

Traynelis and Dingledine abolished 48 of 56 electrographic seizures in slices using media made 5, 15, or 30 mOsm hyperosmotic by agents restricted to the ECS, including mannitol, sucrose, raffinose, l-glucose, and dextran (1989). They noted that seizure suppression was seen in all media; however, seizures were abolished more quickly at higher mOsm concentrations. Membrane permeable compounds including glycerol and d-glucose were ineffective (1989).

A look at possible mechanisms

With the discovery that effects on voltage and ligand-gated ion channels were responsible for the anticonvulsant activity of many of our first-generation antiepileptics, synaptic theories of epileptiform synchronization and epileptogenesis have prevailed. Some diuretics appear to act by modulating synaptic activity. For example, acetazolamide blocks carbonic anhydrase, an enzyme that catalyzes the hydration and dehydration of CO2 into carbonic acid (i.e., bicarbonate and a proton). Although Cl is the primary charge carrier through GABAA channels, bicarbonate also permeates the GABAA channel, causing the membrane response to be slightly more depolarizing than if only Cl were carrying the charge. During intense synaptic activity, such as occurs during seizures, the Cl currents through the GABAA channel lead to equilibration of the Cl concentrations on either side of the membrane, so that the driving force for Cl current is reduced (Huguenard & Alger, 1986; Misgeld et al., 1986). This is not the case for bicarbonate flux, which continues as long as carbonic anhydrase continues to catalyze the conversion of CO2 and water to and from bicarbonate and protons. The end result is a more pronounced depolarizing shift in the GABA response, which reduces the efficacy of GABAergic inhibition. Blocking extra- and/or intracellular carbonic anhydrase with acetazolamide or other inhibitors causes the bicarbonate concentrations to collapse along with the Cl gradients, thereby blocking the depolarizing response (Staley et al., 1995; Staley & Proctor, 1999). Because anticonvulsants such as barbiturates and benzodiazepines prolong the GABA response, this effect of acetazolamide may be synergistic with the effects of these GABAergic anticonvulsants (Staley, 2004).

Nonsynaptic mechanisms may also be important in epileptiform synchronization. Taylor and Dudek demonstrated that synaptic transmission may not even be necessary for ictogenesis (1982), and similar findings were reported by Jefferys and Haas (1982). Using a low Ca+ with Mn2+ bath, chemically evoked postsynaptic potentials were blocked in slices. Electrical stimuli applied to the slices produced large-amplitude action potentials synchronized across a large CA1 population despite complete synaptic blockade. They hypothesized a combination of mechanisms including electrotonic coupling via gap junctions, electrical field or “ephaptic” effects, and fluctuations of extracellular ions. After nearly 30 years, these studies remain as compelling as they are difficult to interpret because of the multitude of effects of calcium on transmembrane ion channels and intracellular signaling cascades.

Other investigators demonstrated that neural activation can induce shrinkage of the ECS by 20% (Dietzel et al., 1982; Ransom et al., 1985). Ransom et al. found that neural stimulation resulted in increased [K+]o, resulting in ECS shrinkage that was blocked by Cl free solution, SITS (4-acetamido-4-isothiocyanostilbene-2,2′-disulfonic acid), and furosemide. They suggested that a glial anion transport system was integral to activity-dependent ECS shrinkage. Dudek et al. later examined the effects of ECS size in the setting of total synaptic blockade and concluded that activity-dependent reduction of ECS and subsequent enhancement of nonsynaptic mechanisms of neuronal excitation may represent a positive feedback loop that contributes to the induction and maintenance of epileptogenesis (1990).

Uptake of K+ and Cl in cultured astrocytes was blocked using various combinations of furosemide and bumetanide, and K+ uptake in the presence of ouabain was sensitive to the omission of Na+/Cl medium (Kimelberg & Frangakis, 1985). This evidence supported the presence of an astrocytic K+ + Na+ + Cl cotransport system. Osmotic manipulations confirmed astrocytic regulatory volume decreases after exposure to hypotonic medium, regulatory volume increases after subsequent exposure to isotonic medium, and osmotic shrinkage only in the presence of hypertonic medium. Kimelberg and Frangakis hypothesized that the activity-dependent ECS shrinkage, as seen in the rat optic nerve study, may be correlated with astrocyte swelling due to stimulation of cation plus Cl cotransport systems (1985).

Shrinkage of the ECS is hypothesized to promote excitability, synchrony, and ultimately seizures (Lux et al., 1986; Traynelis & Dingledine, 1989). Initially ECS shrinkage can exaggerate the transient elevations of extracellular potassium associated with intense periods of neuronal activity, thereby increasing excitability. ECS shrinkage can also enhance electrical field effects that promote synchrony via two proposed mechanisms. As neighboring neurons move closer to one another they may adopt a similar refractory period during the momentary negative extracellular potential that occurs during action potentials. A synchronous firing of neurons will create a voltage gradient that results in electrical current preferentially traversing neighboring cells rather than ECS, as shrinkage of the ECS increases its electrical resistance. The current that does move through the ECS tends to depolarize, and, therefore, recruit silent neurons in the vicinity (Taylor & Dudek, 1984) Alternatively, the ECS voltage gradients can cause a transient redistribution of the major extracellular electrolytes (Na+ and Cl) near activated neurons. Traynelis and Dingledine concluded that ECS shrinkage is a “critical determinant of the conditions that enable CA3 interictal input to precipitate an electrographic seizure in CA1” (1989).

The evidence for diuretics as antiepileptics

Hochman et al. examined the antiepileptic effects of high (millimolar) concentrations of the chloride cotransporter antagonist, furosemide, in a broad range of in vitro epilepsy models (1995). Monitoring ECS volume changes by measurements of the intrinsic optical signal, a measure of light scattering by brain tissue (Lipton, 1973), MacVicar and Hochman blocked stimulus evoked changes in light transmission through the CA1 region of hippocampal slices with furosemide (1991). Building on this work, Hochman et al. used direct electrical stimulation of the Shaffer collaterals to demonstrate furosemide-blockage of afterdischarge activity as well as associated optical changes, which were reversed after furosemide washout (1995). In addition to blocking epileptiform activity, furosemide appeared to dissociate synchronized activity (spontaneous discharges decreased) and excitability (increased amplitude of evoked epileptiform discharges). Furosemide successfully blocked several different physiologic models of epilepsy: hyperkalemic bath (model of hyperexcitability), hypomagnesemic bath (anticonvulsant resistant model), 0 Ca2+ bath (synaptic blockade), 4-AP (K+ channel antagonist), bicuculline (GABAA antagonist), and kainic acid (status epilepticus and kindling model). They concluded that furosemide must act at a common mechanism necessary for the generation of synchronized bursting, and based on their optical signal data, they proposed that furosemide’s action was related to cell volume regulation due to glial swelling. Gutschmidt et al. confirmed similar findings, demonstrating furosemide blockade of seizures in slices bathed in solutions containing low Ca2+, low Mg2+, and 4-AP (1999).

Hochman et al. further tested the hypothesis that the antiepileptic effects of furosemide are mediated through actions on Cl transport by mimicking its anticonvulsant effect with low extracellular Cl (1999). Epileptiform discharges in slices perfused with high K+, bicuculline, or 4-AP were blocked by perfusion of low- Cl media. Furosemide causes a depolarizing shift in EGABA by blocking K+ and Cl cotransport, allowing equilibration between intra- and extracellular chloride concentrations, whereas low [Cl]o reduces inhibitory postsynaptic potential driving force by directly diminishing the transmembrane chloride gradient. Both conditions reduce the driving force for inward Cl flux, which diminishes the efficacy of GABAergic inhibition, resulting in stimulus-evoked hyperexcitability, as evidenced by evoked epileptiform bursts of increased magnitude. Because they demonstrated nearly identical effects on neuronal hypersynchronization, furosemide and low [Cl] o were felt to exhibit their antiepileptic properties on glial Na+/K+/2Cl cotransport. They also note that their data did not depend on the actual location of the Na+/K+/2Cl cotransporter and acknowledged the possibility of a neuronal Na+/K+/2Cl cotransporter (1999). Unresolved puzzles are the prolonged time required for the anticonvulsant effect of low Cl solutions (about an hour), and the very high concentration of furosemide needed for the profound anticonvulsant effect: Hochman et al. used 2.5, about 25 times the concentration needed to completely block potassium chloride cotransporter 2 (KCC2) and sodium potassium chloride cotransporter 1 (NKCC1) (Gillen et al., 1996) and five times the amount needed to block other potassium-chloride cotransporters (Gagnon et al., 2007). Without suppressing normal neuronal excitability, Hochman and Schwartzkroin confirmed that the antiepileptic effects of chloride cotransport antagonism are due to desynchronization (2000).

With their accumulated techniques and experience, Haglund and Hochman investigated the role of ECS modulation in human susceptibility to ictal events (2005). In adult patients undergoing resection of an epileptic focus (still on their baseline antiepileptic medications) investigators intraoperatively measured intrinsic optical signal and electrocorticography following a single intravenous bolus of furosemide (20 mg). Within 10–30 min after administration, the frequency of interictal spikes was reduced on average by 60% in the five patients tested. In eight patients, they examined furosemide-induced resistance to direct cortical electrical stimulation and reported complete blockade in 4, and >50% reduction of afterdischarge duration and amplitude in the remainder. If furosemide raised epileptic threshold through modulation of the ECS, it would be expected that osmotic agents would also have similar effects. They repeated the above studies in four patients with a single 50-g intravenous injection of mannitol and found the same 60% reduction of spontaneous spiking frequency and 42% reduction of afterdischarge duration following direct cortical stimulation. They concluded that (1) furosemide and mannitol suppressed highly resistant epileptic activity, (2) the effect was seen in all patients with both mesial temporal and neocortical epilepsy, and (3) concomitant EEG activity from nonepileptic sites was preserved, suggesting that epileptic suppression was not due to global suppression of neuronal activity (2005). The means by which mannitol is administered may alter its affect, as others have found that boluses of mannitol alter the blood–brain barrier and may provoke seizures (Marchi et al., 2007).

These data support the hypothesis that diuretics inhibit epileptic activity by inducing volume shifts. This might reflect an expansion of the extracellular space and concomitant reduction in ephaptic interactions, but there are many other potential mechanisms. For example, alterations in cellular volume trigger a cascade of intracellular signals designed in part to activate mechanisms of volume restoration (Kahle et al., 2006, 2008; Hoffmann et al., 2009), but the anticonvulsant effects of activating these signaling cascades are not known. Additional mechanisms of action may also be implicated through inhibition of the K+-Cl cotransporter KCC2 (Gutschmidt et al., 1999; Viitanen et al., 2010) and in their effect on chloride uptake mediated by NKCC1 at the axon initial segment (Khirug et al., 2008), or via inhibition of Cl/HCO3 exchange (Gutschmidt et al., 1999), although these mechanisms do not address the high concentrations of furosemide required in vitro (but perhaps not in vivo: Haglund & Hochman, 2005) or the curious time dependence of the low-chloride media. It has also been suggested that chloride-dependent steps in glutamate transport may be responsible for these anticonvulsant effects (Staley, 2002).

In addition to their broad safety profile, the attractiveness of diuretics as antiepileptics lies both in their broad efficacy in in vitro and in vivo models of epilepsy as well as the selectivity of their effects on epileptic versus normal activity, which unfortunately is not shared by most of our current antiepileptic medications.

Neonatal Development and Seizures

The latest narrative involving seizures and diuretics has evolved from our greater understanding of neuronal development and neonatal seizures beginning around the 1990s. Occurring in the first month of life, neonatal seizures are seen with a frequency of 1–2% of neonatal intensive care unit patients. The most common causes are infarction, hemorrhage, and hypoxic ischemic encephalopathy (HIE), with metabolic and dysgenetic causes accounting for the remainder (Volpe, 2001). The presence of neonatal seizures was associated with normal outcome in only 36 of 106 patients (Pisani et al., 2007). Conventional therapy is not only frequently inadequate but there is no compelling evidence for the use of any existing medication for the treatment of neonatal seizures (Booth & Evans, 2004). The only randomized control trial looking at first-line drugs demonstrated a 45% electrographic seizure-free rate with either phenobarbital or phenytoin alone, capturing another 15% if the other agent is used in conjunction with the first (Painter et al., 1999). Despite this, phenobarbital and benzodiazepines are the mainstay of clinical practice.

GABA is excitatory in neonatal development

In the immature neuron, a delicate equilibrium exists between excitation and inhibition. This balance is critical in these early stages of development, with failure of neuronal growth and synaptic maturation in the setting of excessive inhibition and seizure and excitotoxic death in the setting of excessive excitation (Ben-Ari, 2002). Ben-Ari addressed this mismatch dilemma through a series of experiments, first demonstrating that activation of GABA synapses in immature neurons was depolarizing as a result of a high intracellular concentration of chloride [Cl]i. (Fig. 2) Excitatory GABA signaling sequentially precedes the development of excitatory glutamate signaling. As more glutamate and GABA synapses are generated, a chloride extruding system becomes operative, and GABA begins to exert its conventional inhibitory influence.

Figure 2.

Intraneuronal chloride concentration and GABA activity. The GABAA receptor functions as a chloride channel. The electrochemical gradient for chloride across the neuronal membrane will determine whether chloride is transported into or out of the cell via the GABAA receptor. This, in turn, determines the strength and polarity of GABA neurotransmission. The intracellular level of chloride is established in part by the combined activities of the SLC12A cation-chloride cotransporters NKCC1 and KCC2, which mediate chloride influx and efflux, respectively. NKCC1 and KCC2 mediate secondary-active transport: KCC2 uses the energy stored in the potassium gradient to export chloride from the cytoplasm, and NKCC1 uses energy stored in the sodium gradient to import 1 potassium and 2 chloride ions into the cytoplasm. In adult neurons, KCC2 activity predominates, resulting in a relatively low intracellular chloride concentration; therefore, chloride moves into the cell via the GABAA receptor and elicits hyperpolarizing inhibition. However, in neonatal neurons and in several pathologic conditions such as trauma and various epilepsy syndromes, NKCC1 activity predominates. This results in chloride accumulation and an efflux of chloride through the GABAA receptor upon GABA binding, eliciting depolarizing excitation (right GABA channel).

The large calcium oscillations associated with this developmental stage appear to activate c-fos, and enhance expression of brain-derived neurotrophic factor (BDNF) mRNA and KCC2 protein. It is this last product that appears to be responsible for the molecular switch in GABA function, but prior to this switch, use of traditional antiepileptic agents that promote GABA activation may unintentionally be promoting an excitatory state.

Actually GABAergic signaling depends on intracellular chloride concentration

The GABAA receptor is a neuronal ligand-gated chloride channel that is opened during GABA binding. This results in a conformational change that allows the passive flow of chloride ions either into or out of the cell depending on its chloride equilibrium potential ECl (Kahle et al., 2008). When [Cl]i is high, ECl is more positive relative to the resting membrane potential (Vm), and chloride channel opening allows outwardly directed chloride current to depolarize the neuron. When [Cl]i is low, ECl is more negative relative to Vm, and chloride channel opening allows inwardly directed chloride current to hyperpolarize the neuron. It should be noted that GABA-mediated depolarization, if sufficiently large, facilitates excitation but does not guarantee action-potential generation. In fact, if ECl is depolarized relative to Vm but hyperpolarized relative to action potential threshold, GABA depolarization is still inhibitory through the mechanism of shunting inhibition. Through Ohm’s Law V = IR, as chloride conductance increases with channel opening, membrane resistance decreases, and greater current is required to change membrane potential.

The cation-chloride cotransporters

Chloride homeostasis is critical to a wide range of neurophysiologic processes, from ensuring appropriate electrical response to GABA signaling, to regulating cell volume in response to extracellular osmolar changes (Kahle et al., 2008). Strict control of [Cl]i is modulated by the SLC12 gene family of cation-chloride cotransporters (CCCs), including the inwardly-directed NKCC and outwardly directed KCC cotransporters. These CCCs are intrinsic membrane proteins that transport chloride across plasma membranes using the energetically favorable transmembrane gradients of sodium and potassium, respectively. NKCC1 and NKCC2 load chloride into cells by coupling the favorable inward sodium current to raise [Cl]i above its electrochemical equilibrium, whereas KCC1, KCC2, KCC3, and KCC4 couple chloride to the transmembrane potassium gradient, thereby lowering [Cl]i below its electrochemical equilibrium. NKCC1 is prominently expressed in neurons, glial cells, the choroid plexus, and vascular endothelium, whereas NKCC2 is primarily found in the kidney. KCC2 is exclusively expressed in mature neurons. In nonneuronal cells, hypotonic ECS, cell swelling, and elevated intracellular chloride activate protein phosphatases that dephosphorylate and thereby inhibit NKCC and stimulate KCC, resulting in reduced [Cl]i. At the same time, hypertonic ECS, cell shrinkage, and decreased intracellular chloride inhibit protein phosphatases and promote phosphorylation of the cotransporters to activate NKCC and inhibit KCC, resulting in increased [Cl]i (Payne et al., 2003). The associated serine/threonine protein kinase (WNK) and serine-threonine kinase 39 (SPAK) kinases coordinately and reciprocally regulate the activity of the NKCC and KCC cotransporters (Kahle et al., 2006).

Shared physiology of rodent and human neonatal seizures and implications for therapy

GABA excitation appeared to be a plausible explanation for the medical refractoriness of neonatal seizures, and Dzhala et al. hypothesized that neonatal seizures may be controlled by blockade of the cotransporter (NKCC1) responsible for the high intracellular concentration of chloride [Cl]i necessary for GABA-mediated excitation (2005). They demonstrated significantly higher human NKCC1 expression during postconceptional weeks (PCWs) 31 through 41, compared with 1 year (PCW 92) and older. During the first year of life, from PCW 54 through PCW 92, NKCC1 expression tapered to levels seen in the adult. Human KCC2 expression was present but significantly lower than the adult throughout the fetal and neonatal period, and between PCW 31 through 41, as NKCC1 levels were peaking, KCC2 levels began rising to reach adult levels by the end of the first year of life. Reflecting GABA’s switch from excitation to inhibition, they showed that phenobarbital was ineffective at controlling seizures in neonatal rodent slices, but successfully stopped ictal tonic–clonic discharges in older rats. Next they showed a hyperpolarizing shift in the effect of GABAA receptor activation compared with controls after application of bumetanide in neonatal slices, and followed this experiment by blocking electrographic seizure-like activity with this selective NKCC1 blocker. They confirmed the importance of this site by demonstrating the lack of bumetanide’s apparent antiepileptic activity in homozygous NKCC1-knockout mice (2005).

If bumetanide can pharmacologically mimic mature chloride conductance in the immature brain, theoretically, phenobarbital in combination with bumetanide might prove to be a powerful antiepileptic combination. In another series of experiments, their laboratory demonstrated this hypothesis. Phenobarbital was able to depress recurrent seizures in only 30% of low-Mg2+ rat hippocampi, whereas bumetanide in combination with phenobarbital totally abolished seizures in 70% of hippocampi and significantly reduced the frequency, duration, and power of seizures in the remaining 30% (Dzhala et al., 2008).

Other investigators (Kilb et al., 2007; Wahab et al., 2011) found more limited anticonvulsant effects of bumetanide in immature rodent brain. Dzhala et al. (2010) demonstrated that these differences are likely related to the duration of seizure activity preceding the application of bumetanide. Seizure activity resulted in increases in intracellular Cl that made the effect of GABAA receptor activation more strongly depolarizing, and bumetanide had much stronger effects on this increase than on baseline Cl.

Pathology recapitulates ontogeny

If the refractoriness of neonatal seizures can be explained by normal developmental patterns of chloride handling, is it possible that the refractoriness of some forms of epilepsy later in life may be explained by a regression to the neonatal phenotype? Is there something specific to seizure that may induce such a regression, or is it a phenomenon of neuronal injury in general? In fact, a diverse group of investigators examining multiple models of cellular injury have demonstrated parallel findings.

Building on previous work demonstrating reduced KCC2 expression in axotomy, nerve crush injury, cuff-induced chronic nerve pain, and interictal activity, with resultant depolarizing shift of GABA signaling, Wake et al. demonstrated reduced KCC2 activity triggered by oxidative stress (H2O2), seizure (BDNF), and hyperexcitability (0 Mg2+), which even preceded decreases in KCC2 expression (2007).

Oxygen-glucose deprivation (OGD) followed by reoxygenation, an in vitro model of ischemia, showed increases in [Cl]i, which was not simply a consequence of neuronal damage but causally responsible for the damage and was preventable with bumetanide (Pond et al., 2006). Yan et al. induced cerebral ischemia by occlusion of the left middle cerebral artery followed by reperfusion, demonstrating stimulation of NKCC cotransporter activity and increased NKCC expression in cortical neurons during reperfusion (2001). After application of bumetanide, they showed a 25% reduction of infarct volume and 70% reduction of ipsilateral cerebral water volume (Yan et al., 2001). In a follow-up study, bumetanide inhibition of NKCC1 activity abolished glutamate-mediated neurotoxicity and significantly attenuated OGD-induced neuronal death, suggesting that NKCC1 plays a critical initial role in excitotoxic death. Interestingly, NKCC1 was not involved in excitotoxic death in immature neurons probably owing to the lack of NMDA glutamate receptors (Beck et al., 2003).

Investigators studying endothelial-based NKCC cotransporter also demonstrated hypoxia-induced increase in blood–brain barrier NKCC activity within 30 min of exposure to a range of moderate (7.5% O2) to severe (0.5% O2) levels of hypoxia, which approximate levels seen in the penumbra and core of stroke beds, respectively; intracellular volume increase mirrored brain edema formation and occurred only after 4–5 h of hypoxia, which reflects a similar time frame of NKCC stimulation (Foroutan et al., 2005). Vascular endothelium placed under steady laminar fluid shear stress or exposed to a variety of inflammatory cytokines, including interleukin-1β and tumor necrosis factor-α, can induce and maintain elevated expression of NKCC in cultured human umbilical vein endothelial cells, and similar upregulation of NKCC mRNA is seen in vivo with intraperitoneal administration of bacterial endotoxin lipopoly saccharide (LPS) in mouse lung and kidney (Topper et al., 1997).

Temporal lobe epilepsy: pharmacological refractoriness explained?

Mesial temporal lobe epilepsy (MTLE) is the most common form of adult localization-related epilepsy and is often refractory to the sodium channel blocking or GABA-enhancing medications that dominate our current antiepileptic armamentarium. Gliosis and deafferentation have been shown to induce a depolarizing shift of GABA reversal potential and is associated with decreased KCC2 expression (Nabekura et al., 2002). In pilocarpine status epilepticus, rodent dentate granule cells exhibited a positive shift of EGABA, which altered synaptic integration, increased granule cell excitability, and compromised the “gating function” of the dentate gyrus for up to 2 weeks following the injury, and this shift was mediated by decrease in KCC2 function (Pathak et al., 2007). In vivo kindling of seizures leads to a massive upregulation of BDNF, which results in decreased levels of KCC2 mRNA and protein and a 50% increase in intracellular Cl (Rivera et al., 2002), which has already been shown to be associated with sustained propensity for excitability: seizures begetting seizures.

With this background, it was not a far stretch in 2002 for Cohen et al. to record interictal neuronal activity from slices prepared from postresective tissue in a series of human patients with MTLE. They demonstrated spontaneous extracellular spikes from the subiculum with relative silence from the other structures and maintenance of spontaneous discharges even after the subiculum was isolated from CA1 input. They further identified a population of pacemaker pyramidal cells that exhibited depolarizing GABA signaling that was instrumental in the synchronization of the spontaneous discharges emanating from the subiculum.

In 2006, Palma et al. expanded on these findings in another series of surgical specimens from human MTLE. Quantitative analysis of the mRNA extracted from these four patients showed upregulation of the NKCC1 mRNA and downregulation of the KCC2 mRNA in the hippocampal subiculum compared with other regions of the hippocampus or temporal lobe neocortex from the same patient. Next they isolated and injected cell membranes from the same brain regions into Xenopus oocytes, which quickly incorporated the human GABAA receptors into the oocyte membrane. In oocytes injected with subicular membranes, EGABA was significantly depolarized compared with oocytes injected with membranes from other sites in all four patients tested. When they inactivated the NKCC1 cotransporter by lowering bath temperature or by the addition of bumetanide, EGABA was similar between subicular and nonsubicular oocytes. Their findings appear to corroborate that the spontaneous subicular firing discovered by Cohen et al. is dependent on the depolarizing shift of GABA created by elevated levels of intracellular chloride. Munoz et al. reported that 20% of NKCC positive cells did not express KCC2 in the epileptogenic transition zone between the subiculum and sclerotic areas of CA1 (2007). Huberfeld et al. examined a series of 27 patients with MTLE and found that KCC2 mRNA was completely absent from the 30% of subicular pyramidal cells that showed depolarization. In a different experiment on slices of resected tissue, they blocked human interictal-like activity with bumetanide, which was then reversed with washout (2007). To demonstrate that bumetanide acted via GABAergic signaling rather than by a change in cellular volume as suggested by Hochman et al. (1995), they increased bath K+ in the presence of picrotoxin and showed that bumetanide was unable to suppress interictal-like activity when GABA pathways were simultaneously and completely blocked.

Cortical migration defects: arrested development?

Malformations of cortical development (MCDs) such as focal cortical dysplasia (FCD), glioneuronal tumors (ganglioglioma, GG), and hemimegalencephaly (HMEG) are commonly associated with pharmacologically refractory epilepsy. In postresective epilepsy surgical patients, Aronica et al. confirmed low NKCC1 expression and strong KCC2 expression in perilesional areas of brain tissue and high expression of NKCC1 in both neuronal and glial components of FCD, HMEG, and GG tissue (2007). Sen et al. reported similar histologic findings in a series of 8 patients with hippocampal sclerosis (HS) and 17 patients with focal cortical dysplasias (2007). Strong NKCC1 immunoreactivity was seen in the CA2 region and the few surviving neurons in the CA1 and CA3 regions of patients with HS, as well as in the dysplastic neurons in all cases of FCD (Sen et al., 2007). Hannan et al. examined subcortical and periventricular nodular heterotopias from four patients with a variety of MCDs and identified populations of GABAergic neurons, which appeared to be immature, in all tissue studied (1999). Furthermore, they demonstrated sparse reciprocal connections from the nodule to overlying clusters of abnormal neurons within the cortical plate, providing an explanation for both a mechanism of hyperexcitability and a route of amplification for the highly refractory seizures seen in MCD. These findings support the hypothesis that the pathogenesis of malformations of cortical development involves a delay of maturation, and the tendency toward hyperexcitability that accompanies this immature state.

Loop diuretics: adjunctive oral antiepileptics?

Despite long-standing, compelling evidence, bumetanide and furosemide have only recently garnered serious interest as potential antiepileptic agents. Bumetanide is a loop diuretic with a major site of action in the ascending loop of Henle. It is highly protein bound (>90%) and has a half-life between 60 and 90 min. Elimination in the pediatric population appears to be slower with half-life of 6 h at birth to 2.4 h at 1 month of age (Lopez-Samblas et al., 1997). The diuretic effect of bumetanide is 40 times more potent than furosemide, and maximum diuresis occurs within 15–30 min of intravenous administration. The mechanism of action of bumetanide is selective blockade of the cation-chloride cotransporter NKCC, unlike furosemide which also inhibits the KCCs. In edema related to congestive heart failure, there is good evidence of short and long-term safety and tolerability, with a side-effect profile mainly related to the intended diuresis (e.g., hypokalemia, water loss and dehydration, electrolyte depletion) (Dixon et al., 1981; Handler et al., 1981). Ototoxicity has been seen in animals tested at high relative doses of bumetanide, and may be potentiated by other known ototoxic drugs. Rare and spontaneous reports of thrombocytopenia are noted, and allergy to sulfonamides predisposes hypersensitivity to bumetanide. The drug is Pregnancy Category C but is not teratogenic in rodents at doses 3,400 times the maximum therapeutic human dose (MTHD).

As noted previously, NKCC comes in two isoforms, with NKCC2 uniquely found in the kidney and NKCC1 found in a variety of secretory epithelia and nonepithelial cells, such as neurons (Haas & Forbush, 2000). This anatomic isolation of isoforms offers a potential for targeted therapeutic action; however, selective neuronal NKCC1 inhibitors are not currently available. Utilizing knowledge of the renal metabolism of bumetanide and furosemide, it may be possible to benefit from the intended blockade of NKCC without the intense diuresis. These medications are processed by the renal organic anion transport system, but in the presence of the gout medication, probenecid, its preferential binding results in reduced renal excretion and decreased diuresis (Odlind et al., 1983; Kawaguchi et al., 2001). Escalating doses of probenecid showed an inverse relationship with urinary output when given prior to loop diuretic administration. However, it may be preferable to use the techniques of modern medicinal chemistry to develop more specific NKCC1 antagonists (Delpire et al., 2009).

Whether the loop diuretics inhibit activity-induced shrinkage of the extracellular space and hypersynchronous electrical activity, or rectify an aberrant chloride flux across a pathologic neuronal membrane, there appears to be an antiepileptic effect in the setting of refractory epilepsy that is worth investigating. In fact, in 1976, this study was performed (Ahmad et al., 1976). Fourteen long-term care residents of the National Hospitals-Chalfont Centre for Epilepsy were administered placebo or furosemide 40 mg every 8 h in a double blind, placebo-controlled, crossover study. Eleven of 14 patients completed the study and were highly refractory to the current therapy of the time, having electroencephalography (EEG)–verified complex partial seizures at a combined 281 events in the 4-week baseline period and 236 events in the 4 weeks after return to baseline. Placebo was provided for 4 weeks with a total of 224 events, and furosemide was provided for 4 weeks with a total of 130 events, representing an average reduction of 58%, including one patient who became seizure free. The subjects were divided into two groups with the first group receiving furosemide followed by placebo crossover and the second group receiving placebo followed by furosemide crossover. Analysis of variance showed patients had fewer “fits” during the active treatment period compared with the other periods (p < 0.001). Adverse events included drowsiness in 5 of 14 patients with 3 of the 5 withdrawing from the study due to severe drowsiness. There was no effect on serum phenytoin or primidone levels, but phenobarbital levels were increased from 24.8 (±10.8) to 27.4 (±8.8) μg/ml. Serum sodium was reduced from 142.6 to140.9 mm and potassium was reduced from 4.1 to 3.7 mm. Serum bicarbonate was increased from 30.1 to 32.5 mm, whereas serum urea and vital signs remained unchanged. These findings are remarkable when considering the highly refractory nature of the seizures in these patients. The 11 subjects who completed the study averaged 22.5 complex partial seizures a month, 7 of 11 had a 50% or greater reduction in seizures, 1 of 11 appeared to worsen, and 3 of 11 had marginal or no change. One of 14 (or 7%) became seizure free. These findings (50% responder rates of 63% and seizure freedom rates of 7%) are comparable and arguably better than most modern adjunctive therapy trials of the newer antiepileptics. Although this study involved only a small number of patients, the results are compelling.


With mounting support for the apparent antiepileptic effects of oral and intravenous diuretics in both animal models and humans, it is unclear why diuretics are not considered in current epilepsy treatment algorithms. Diuretics may disrupt both neuronal hypersynchrony and hyperexcitability via inhibition of activity-induced shrinkage of the extracellular space, and/or modulation of transmembrane chloride flux. In the abnormally excitable neuron, known to harbor intracellular Cl levels comparable to immature neurons, loop diuretics such as bumetanide appear to return chloride balance toward “mature” levels (Kahle & Staley, 2008). Investigation into these compounds as antiepileptic drugs, either alone or as adjuncts in rational polypharmaceutic regimens, seems warranted. In addition, the development of novel compounds that selectively target CCCs and induce less diuresis and systemic electrolyte alterations would also be useful.


Edward Maa wishes to thank his coauthors for their experience and insight; to Daryl Hochman for his comments and advice regarding probenecid, and his lively discussion at the poster platform at AES 2009, and to Phil Schwartzkroin for his editorial suggestions and continued interest in this fascinating topic.


The authors have no commercial financial interests to report in the presentation of this review. “We confirm that we have read the Journal’s position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.”