Novel Concepts for the Role of Chloride Cotransporters in Refractory Seizures

Epilepsy is associated with a multitude of acquired or genetic neurological disorders characterized by a predisposition to spontaneous recurrent seizures. An estimated 15 million patients worldwide have ongoing seizures despite optimal management and are classified as having refractory epilepsy. Early-life seizures like those caused by perinatal hypoxic ischemic encephalopathy (HIE) remain a clinical challenge because although transient, they are difficult to treat and associated with poor neurological outcomes. Pediatric epilepsy syndromes are consistently associated with intellectual disability and neurocognitive comorbidities. HIE and arterial ischemic stroke are the most common causes of seizures in term neonates and account for 7.5-20% of neonatal seizures. Standard first-line treatments such as phenobarbital (PB) and phenytoin fail to curb seizures in ~50% of neonates. In the long-term, HIE can result in hippocampal sclerosis and temporal lobe epilepsy (TLE), which is the most common adult epilepsy, ~30% of which is associated with refractory seizures. For patients with refractory TLE seizures, a viable option is the surgical resection of the epileptic foci. Novel insights gained from investigating the developmental role of Cl- cotransporter function have helped to elucidate some of the mechanisms underlying the emergence of refractory seizures in both HIE and TLE. KCC2 as the chief Cl- extruder in neurons is critical for enabling strong hyperpolarizing synaptic inhibition in the brain and has been implicated in the pathophysiology underlying both conditions. More recently, KCC2 function has become a novel therapeutic target to combat refractory seizures.

either epileptogenesis or the inherent age-dependent susceptibility. The discovery of new paradigms to investigate and efficiently rescue epileptogenesis soon after the emergence of seizures is the goal to prevent the onset of chronic epilepsy and long-term comorbidities. Even a single seizure event can have long-term effects and valuable insights have been gained using preclinical rodent models [11,12], many of which have focused on the role of GABAergic transmission in the CNS. This review will focus on the mechanisms that underlie seizure refractoriness in both HIE and TLE and provide an overview of current translational research focuses and proposed therapeutic paradigms.

Factors controlling tonic vs phasic inhibition
GABARs consist of two major categories: GABAARs are ionotropic receptors, whereas GABABRs are a subset of G protein-coupled receptors (GPCRs) and are metabotropic. GABAARs are composed of five subunits, of which many subunit variants have been identified including six α, three β, three γ, one δ, three ρ, one π, one θ, and one ε [13]. The subunit composition of the GABAARs changes developmentally and serves variable roles, such as the ability of δ-containing GABAARs to detect extrasynaptic GABA thereby mediating persistent tonic inhibition [14,15]. Tonic and phasic GABARmediated currents are essential for the regulation of different aspects of neurotransmission. Phasic GABAergic neurotransmission adapts rapidly to stimuli and is essential in synaptic integration through feedforward inhibition, thereby ensuring that excitatory inputs are summed only when synchronized [16]. In contrast, tonic inhibition generally reduces the magnitude and duration of excitatory neurotransmission, thereby inhibiting excess and pathological neuronal excitability. Beyond their role as a mediator of inhibitory neurotransmission in the CNS, GABABRs have been shown to promote the secretion of BDNF through the activation of phospholipase C, as well as promote the maturation of GABAergic synapses in the rodent hippocampus [17]. Activation of presynaptic GABABRs in the mouse hippocampus has been shown to decrease synaptic vesicle release [18]. The diversity of GABAergic signaling demonstrates the importance of its acute regulation throughout development and why it is often highlighted as an underlying mechanism in many neurodevelopmental and neuropsychiatric pathologies.
Although excitatory GABAergic neurotransmission is essential for neuronal development in the neonatal brain, acute insult can lead to pathological and irreversible disruption of these circuits. 20-60% of incidents of epilepsy are a result of acute insults to the CNS [19]. Among these insults, HIE is one of the most common to result in acute neonatal seizures [20]. Despite the introduction of new antiseizure medications (ASMs) into the clinic, refractory seizure incidence in neonates remains unchanged [21]. Two first-line ASMs like phenobarbital (PB) and valproate have shown little evidence that they reduce the subsequent incidence of epilepsy, a result which is tied to the regulation of the transmembrane Clgradient and the electroneutral ion transporters KCC2 and NKCC1.
Neonatal HIE can be devastating due to several mechanisms relating to postnatal neurogenesis [22]. GABA has been shown to regulate the maturation and migration of immature cortical neurons during development [23,24], as well as adult-born granule cells [25]. The effect of disrupted GABAergic circuits in the brain may be particularly impactful on granule cell migration, as preclinical studies in rodent models have shown that a greater rate of granule cell neurogenesis in induced by seizures in the adult hippocampus [26]. As HIE is a leading cause of TLE later in life, these factors illustrate the importance and urgency of developing new ASMs that impede epileptogenesis in the developing circuits after acute early-life insults.

Susceptibility to seizures
To stimulate synaptogenesis and development, excitation predominates over inhibition in the neonatal period and throughout early infancy [27,28]. Glutamate is the primary excitatory neurotransmitter in the brain and great focus has been placed on the ionotropic glutamate receptors and their effects of the electrochemical gradients of Na + , K + , and Clas they relate to development and seizure susceptibility. Ionotropic glutamate receptors are characterized into three subtypes: N-methyl-D-aspartate receptors (NMDARs), α-amino-3-hydroxy-5-methyl-4isoxazole propionate receptors (AMPARs), and kainate receptors.
The subunit composition of these receptors is developmentally regulated [29] and is an underlying mechanism of differential neurotransmission in the developing and adult brains. AMPARs are variably composed of the subunits GluA1, GluA2, GluA3, and GluA4, which contain highly conserved ligand binding and transmembrane domains [30]. The precise subunit composition of these ionotropic receptors has important consequences on the electrochemical homeostasis of the developing cells. In AMPARs, GluA2 plays a critical role in Ca 2+ permeability; AMPARs containing the GluA2 subunit will not permit Ca 2+ influx. In rodents, there are fewer GluA2-containing receptors in the developing brain compared to the mature brain, which leads to greater Ca 2+ influx and seizure susceptibility [31]. In P10-12 rats after hypoxia, GluA2 expression was lower in the hippocampus and neocortex [32].

Preclinical Models for Epileptogenesis
The most commonly used preclinical models of epileptogenesis are post-status epilepticus (SE) models of TLE [33]; subcortical and temporal lesions are thought to drive ictogenesis in SE models [34]. Importantly, acute CNS insults frequently result in neuronal network reorganization which leads to epileptogenesis. One critical mechanism underlying these disruptions is the plasticity of GABAergic mechanisms in response to the insult. Commonly, it is suggested that epilepsy is a result of compromised GABAergic inhibition and/or recurrent excitation [35]. Compromised GABAergic inhibition results from loss of susceptible interneuron cell populations [36,37] or compromised inhibition due to formation of new recurrent excitatory circuits [38]. Studies have demonstrated that certain populations of GABAergic cells are vulnerable in animal models of induced epilepsy [39]. Additionally, the GABAergic hypothesis of epileptogenesis suggests pathologies where GABAergic signaling results in depolarization instead of hyperpolarization; positive GABAR modulators have been shown to exacerbate seizures [40] instead of suppressing them [41], indicating that excitatory GABAergic neurotransmission can be ictogenic.

Chloride Cotransporter Function in the CNS
GABAergic neurotransmission in the brain has been a focus of neuroscience research for many decades and has been shown to undergo a developmental switch from excitatory to inhibitory [42]. GABA functions as a neurotransmitter by activating the GABAAR and permitting Clto passively flow across the open receptor according to its electrochemical transmembrane gradient. The direction of Clflow is dependent on the intracellular chloride concentration ([Cl -]i) of the neuron. Neuronal [Cl -]i is primarily controlled by two transporters at the plasma membrane: the primary neuronal Clexporter KCC2 and Climporter NKCC1. NKCC1 is encoded by the SLC12A2 gene in humans and serves an essential role in the secondary active transport of Na + , K + , and Clinto neurons. The relative developmental expression patterns of NKCC1 and KCC2 are dynamic as expression of KCC2 increases during development; the increased ratio of KCC2 to NKCC1 [43] that stabilizes at 2yrs of age is thought to underlie the developmental switch of GABAergic signaling from excitatory to inhibitory. In the neonatal brain, NKCC1 is highly expressed and accumulates Clintracellularly, whereas low KCC2 expression results in a lower rate of Clextrusion. The greater rate of Climport due to the lower KCC2 to NKCC1 ratio results in a higher [Cl -]i in neonatal neurons compared to mature neurons. In immature cortical neurons which highly express NKCC1 but not KCC2, the activation of the GABAAR by GABA results in Clefflux due to the higher [Cl -]i. Paradoxically, GABAAR activation in the neonatal brain results in depolarization of the cell instead of hyperpolarization due to Clefflux and underlies the refractoriness associated with neonatal seizures [44,45].
Two isoforms of NKCC1 have been studied extensively: NKCC1a and NKCC1b. Studies have shown that mRNA expression of the NKCC1a isoform occurs primarily in the brain [46]. NKCC1 activity has been shown to be stimulated by phosphorylation of residues Thr(203), Thr(207), and Thr(212) [47]. Inhibition of the WNK1 and WNK3 pathways have been shown to decrease NKCC1 phosphorylation and activity, thereby accelerating recovery after stroke [47] and reducing neuropathic pain [48]. Off-label use of the diuretic and NKCC1 antagonist bumetanide (BTN) was ineffective as an ASM [49]. An Nkcc1 -/mouse model showed impaired balance and deafness but no evidence of seizures [50]. Recently, a study showed a loss of function mutation of SLC12A2 in two sisters with encephalopathy and developmental abnormalities, but no seizures [51]. These studies suggest KCC2 plays a more critical role in ictogenesis than NKCC1.
Several isoforms of KCC2 have been identified in recent years, each with their own developmental expression patterns [52]. The most widely studied are the full-length isoforms KCC2a and KCC2b which differ in their N-termini [53]. In the gestational period and shortly after birth, KCC2a and KCC2b expression levels were similar in brainstem neurons in a mouse model, with KCC2b expression increasing with development [54]. Although the function of KCC2a is not fully understood [55], KCC2b is the predominant isoform in the CNS and functions as its primary neuronal Clextruder.
The cotransporter activity of KCC2 is dependent on its presence at the plasma membrane which is controlled through post-translational phosphorylation and dephosphorylation [56]. The capacity for Clextrusion by KCC2 is potentiated by phosphorylation of the residue serine 940 (S940) [57] which increases the cell surface stability of KCC2. Ca 2+ influx from increased NMDAR activity has been shown to decrease S940 phosphorylation through protein phosphatase 1 [58], decreasing its plasma membrane stability and Clextrusion capacity by promoting endocytosis. Another critical phosphorylation site of KCC2 is the residue threonine 1007 (T1007), which been shown to decrease membrane insertion [59,60], thereby decreasing Clextrusion capacity. These factors, along with the additional intracellular roles that Aging and Disease • Volume 12, Number 4, July 2021 1059 KCC2 plays demonstrate the intricate spatiotemporal control exhibited on its function and expression during development.  Downregulation of KCC2 has been shown to be deleterious. In a Kcc2 -/mouse model, KCC2 knockout mice showed early motor deficits which resulted in respiratory failure and death shortly after birth [61]. Aside from the effects of perturbed Clhomeostasis, KCC2 plays an essential role in the promotion of dendritic spine development through protein-protein interactions between the C-terminus of KCC2 and 4.1N [62], which mediates interactions between the cytoskeleton and plasma membrane. Spine formation is mediated by excitatory GABAergic neurotransmission during development [63], and aberrant spine morphology has been identified in multiple neuropsychiatric disorders including epilepsy and autism [64,65]. These interwoven mechanisms highlight the delicate control of KCC2 expression throughout development and acknowledge the importance of rescuing post-insult KCC2 hypofunction. Interestingly, preclinical models have shown differences in KCC2 expression patterns after insult. After HIE, KCC2 expression decreases, whereas after treatment with the chemoconvulsant pentylenetetrazol (PTZ), KCC2 expression goes up [66] (Fig. 1A and 1B).
Recent discoveries related to the complex CNS functions of KCC2 have provided novel insights. Significant understanding has been gained into both the developmental switch of GABA-mediated depolarization to hyperpolarization during early development and in newborn neurons in the neurogenic niches in the adult brain as related to KCC2 expression and function. During development, the BDNF-mediated modulation in KCC2 expression [67] and BDNF-TrkB-mediated modulation of KCC2 membrane insertion and function [44,68,69] have initiated the understanding of a new role for KCC2 in the emergence of refractory seizures in neonatal brains where KCC2 hypofunction has been documented [45]. Rescuing KCC2 hypofunction through BDNF-TrkB pathway inactivation [44] or KCC2 functional enhancement [45] has been shown to rescue refractoriness in P7 CD-1 mice. In humans, KCC2 was downregulated in patients with refractory TLE ( Fig. 2A-H) [70,71]. In vitro studies performed on epileptic tissue gathered from patients with refractory TLE have shown lower levels of KCC2 mRNA in subicular pyramidal neurons (Fig. 2E-H) [70]. Rescuing NKCC1 hyperfunction using the NKCC1 antagonist and new generation short-acting diuretic bumetanide (BTN) has been effective in its off-label use in a wide range of neurological and psychiatric pathologies including Rett syndrome, autism, chronic pain, and epilepsy [63,72]. These studies demonstrate that Clhomeostasis and the critical roles of KCC2 and NKCC1 underlie many of these neurological pathologies.
Phenobarbital is a positive allosteric modulator and agonist of the GABAA receptor (GABAAR), thereby increasing GABA functional activity when administered [73]. This mechanism of action may underlie the documented inefficacy of PB against refractory neonatal seizures due to the expression patterns of KCC2 and NKCC1 [74]. Positive allosteric modulation of the GABAAR when GABA function is excitatory rather than inhibitory is a possible mechanism underlying the emergence of refractory acute neonatal seizures [44,45,69]. Downregulation of KCC2 has been demonstrated to increase neuronal excitability and epileptogenesis in rodent models [69], highlighting its recovery as a possible therapeutic paradigm [75].
Although off-label use of BTN has been shown to increase quality of life and alleviate symptoms in developmental disorders [49,63], there has been growing concern that active efflux of BTN at the blood-brain barrier (BBB) by Oat3, Oatp1a4, and MRP4 is a limiting factor in the bioavailability of BTN in the brain [76]. In both invasive and non-invasive preclinical models of neonatal hypoxia, BTN failed to reduce seizure burden when given alone or concomitantly with PB [66,68,77,78]. In a rat kindling model, BTN prevented increased kindling-induced seizure susceptibility at P11 but not at P14 or P21 [79]. A clinical trial that examined BTN efficacy as an anti-seizure agent in HIE reported its failure and significant side effects involving ototoxicity [80]. Another major concern surrounding BTN as a potential ASM is its short half-life as potent diuretic and antagonist of NKCC1/2. A BTN trial for HIE seizures conducted in the United States recently reported significant diuresis after a single dose but showed no definitive proof of its anti-seizure efficacy despite longer half-life of BTN [normal t1/2 = 1.5h] due to concomitant hypothermia protocols [81]. In contrast, recent clinical trials where BTN has been tested in the autism patient population have shown significant efficacy for rescuing the neuropsychiatric phenotype [82][83][84][85][86][87]. In two recent studies, a clinical trial for tuberous sclerosis complex patients with autism reported that BTN ameliorated neuropsychiatric behavioral problems but had no effect on seizure frequency or severity [86,88], suggesting that NKCC1 antagonism with BTN is ineffective as a paradigm for reducing seizure burden.
To address the poor bioavailability of BTN due to the BBB, ester prodrugs of BTN have been developed and showed significantly higher brain penetrance than BTN, circumventing known obstacles involving ionization of BTN at physiological pH and tight binding of BTN to plasma proteins which underlie its poor brain penetrance [89] and low brain/plasma ratio. This has inspired the pursuit of novel BTN derivatives which selectively and potently inhibit NKCC1, although recent attempts have not been successful [90]. These factors confound NKCC1 as a target for the treatment of neonatal seizures and highlight the avenue of rescuing KCC2 hypofunction as a major focus of upcoming seizure and epilepsy research. Targeting ion cotransporters like NKCC1 and KCC2 carries impediments which include the temporal dynamics of plasma membrane protein stability, internationalization, protein turnover, and drug bioavailability. A thorough understanding of these factors will be essential in producing translationally viable research.

Refractory Neonatal Seizures
The first two weeks of postnatal development in rodents are the rough equivalents of the first year in human postnatal development [42,91] and this developmental window is the focus of many preclinical rodent models. Perinatal arterial ischemic stroke is the most common form of cerebral infarction in neonates, and results in symptoms such as neonatal seizures in 60% of incidents [92]. A rapid increase in endogenous BDNF following stroke have been reported [93][94][95]. Endogenous BDNF binding to the TrkB receptor has been shown to be responsible for the pathophysiological downregulation of KCC2 [96,97]. K252a, a TrkB receptor antagonist, was shown to produce a significant increase in KCC2 mRNA and protein levels in organotypic hippocampal slice cultures [97]. In neonatal seizures the inefficacy of firstline anticonvulsant PB has been proposed to depend on the reversal of the Clgradient which is maintained by KCC2. Age-dependent induction of BDNF after seizure activity may underlie the excessive release of BDNF in the immature brain. In a mouse model using unilateral carotid ligation without transection or infarct, it has recently been shown that the acute and transient poststroke antagonism of the TrkB receptor may be a novel and effective way of curbing the emergence of pharmaco-resistant seizures in the neonatal brain [68]. This transient block also rescued the acute and sub-acute downregulation of KCC2 which otherwise could be detrimental in a maturing brain given the important role of KCC2 in dendritic spine formation, AMPA receptor traffic, and formation of functional GABA synapses [62]. Ischemia-related transient downregulation of KCC2 may also underlie the transient nature of HIE related neonatal seizures. ANA-12 is a recently identified low-molecular weight antagonist of TrkB [98]. It binds to the extracellular domain of TrkB, prevents BDNF-induced TrkB activation, and abolishes the biological effects of BDNF on TrkB-expressing cells but not those of NGF or NT-3 on TrkA and TrkC expressing cells. More importantly, ANA-12 inhibited TrkB function in the brain after systemic administration without any known adverse effects even with multiple doses [69]. Paradoxically, similar results were documented using the TrkB agonists LM22A-4, HIOC, and deoxygedunin, which rescued neonatal refractory seizures at P7 in CD-1 mice [44]. The manipulation of BDNF pathway-related microRNA (miRNA) expression may be a promising strategy to prevent epileptogenesis [99]. miR-155 is a miRNA that involved in a multitude of biological functions, including inflammatory responses in the CNS [100]. It has also been shown to repress Bdnf by binding to its 3'UTR [101], which makes miR-155 a possible therapeutic target for anti-epileptogenesis.

Refractory TLE Seizures
In contrast to early-life refractory seizures associated with HIE, temporal lobe epilepsy (TLE) is an epilepsy syndrome associated with seizures predominantly initiated in the temporal lobe and is often accompanied by temporal lobe pathology like sclerosis, mossy fiber sprouting, and hippocampal degeneration [102][103][104][105]. The known causes for TLE are multi-pronged but often remain unknown [33,106]. Brain injury, infection, tumors, malformations, inflammation are some of the known conditions that can precede TLE. ~40% of TLE patients become refractory significantly affecting the quality-oflife despite being on a combination of 2 or more ASMs at the highest daily dose regimens. This cohort of refractory patients generally end up opting for resective surgery to remove the seizure initiating focus which is usually the anterior horn of the hippocampus in the temporal lobe [107]. However, patients who did not become seizure-free after elective surgery have shown accelerated and progressive memory impairment [108]. Subiculum tissue resected from refractory TLE patients with hippocampal sclerosis showed spontaneous interictal-like activity in vitro which correlated negatively with KCC2 expression [70]. Similar studies have shown decreased KCC2 transcription and increased NKCC1 transcription in resected TLE tissue [109]. TLE has been broadly associated with impaired Clhomeostasis and depolarizing GABAergic transmission [110]. Similar to inhibition of the BDNF-TrkB pathway in rescuing neonatal HIE refractoriness, sequestration or inhibition of the TrkB-pathway in an adult model of temporal lobe epilepsy have shown promising results [111,112].
One of the leading causes of TLE and hippocampal sclerosis is early-life seizures, especially febrile status epilepticus (FSE) [113,114]. Mutations in SLC12A5, the gene which codes for KCC2, have been reported in patients with febrile seizures [115] which suggests that mutations in its coding or regulatory regions may contribute to the genetic susceptibility of TLE. Hippocampal sclerosis is the most common pathology underlying TLE [116] detected on MRI. Coronal imaging orthogonal to the long axis of the hippocampus reveals reduced hippocampal volume and perturbed hippocampal cytoarchitecture. Hippocampal sclerosis is often accompanied by mossy fiber sprouting [117] which is thought to underly hippocampal hyperexcitability. Patients with refractory mesial TLE whose quality of life is greatly perturbed can elect to undergo anteromesial temporal resection to remove the ictogenic tissue [118]. Much of what has been learned about TLE in the human brain is a result of ex vivo studies performed on resected tissue from refractory TLE patients. Although this data has provided a wealth of insight on the pathology, it confounds the ability to make a causal connection between TLE and its underlying mechanisms such as perturbed Clhomeostasis caused by KCC2 hypoexpression or hypofunction. However, much has been learned by studying the sclerosed hippocampi in preclinical models of TLE, which have shown similar pathophysiology like mossy fiber sprouting in the dorsal hippocampi [119]. Attempts to increase PB efficacy in a pilocarpine-induced epilepsy adult rodent model showed that the NKCC1 inhibitors BTN, azosemide, or torasemide did not improve PB efficacy [120], suggesting that NKCC1 antagonism is not a promising avenue for rescuing refractory epilepsies.

The future of clinical trials aimed at antiepileptogenesis
Continuous EEG (cEEG) studies and clinical data have reported clustering of seizures in the acute and chronic phases of both HIE and TLE ( Fig. 3A and 3B). Targeting KCC2 hypofunction as a mechanism underlying the emergence of refractory seizures relies on the ability for a long-term (months rather than hours) reversal; this requires taking seizure clustering into consideration. With the peak-trough variation in seizure frequency, clinical trials need to be wary of false positives due to administration of an ASM at a time point where seizure frequency is already declining naturally and regression to the mean is variable between patients (Fig. 3C). Granular plotting of temporal seizure frequency and clustering over a 5-day period from a hypothermia study indicate that variability of clustering is a prominent characteristic of HIE seizures. Reviewing time points when drug interventions occurred in the study in relation to the clustering phenotype of the seizures is critical because there is a high risk of false positives or false negatives when data for ASM efficacy are binned over short durations of a couple hours. A similar concept applies to chronic refractory epilepsy like TLE (Fig. 3B-D) where a few days of monitoring would fail to accurately detect both severity of the epilepsy at baseline and the efficacy or failure of an ASM. The multi-day rhythms inherent in these pathologies [121], as well as the variable temporal patterns of seizure frequency between patients and variable baseline seizure burdens demonstrate the importance of appropriate statistical analysis and binning of data to show true efficacy of an intervention. An anti-seizure effect for HIE intervention would lower seizure burdens over 24h, which can only be reported through longer continuous EEG recording and binning durations that take into account the clustering nature of the ictal events. These insights will also permit the design of evidence-based guidelines for repeated dosing regimens since the goal is to significantly suppress all acute seizures in the first week of neonatal life. Similar strategies apply to targeting the antiepileptogenic process for the evolution of TLE pathogenesis; long-term cEEG monitoring will help determine if ASM treatment paradigms contribute to the reversal of both the refractory epilepsy and its severity. Studies designed to identify the efficacy of novel antiepileptogenic interventions would also need long-term cEEG using the latest invasive and non-invasive technology [121,122] to identify sub-clinical and nocturnal seizure clusters. This will help prevent false positive and false negative results.

Conclusion
Although HIE is neonatal and transient and TLE is chronic and progressive, the underlying pathophysiology related to emergence of refractoriness and long-term comorbidities in affected patients overlap. Understanding the novel mechanisms that underlie the emergence of seizure refractoriness such as Clcotransporter dysfunction can help to develop novel strategies for preventing refractoriness by modulating both seizure susceptibility and epileptogenesis. Preclinical research has shown great promise in manipulating KCC2 as a therapeutic target in rescuing refractoriness in HIE and TLE (Fig. 4). Manipulation of NKCC1 through diuretics like BTN and its analogues has been shown to be inefficacious in both preclinical research and clinical trials. Early intervention and prevention of epileptogenesis [123] remains an unmet need and is a focus of translational neuroscience research.