Neuronal Swelling: A Non-osmotic Consequence of Spreading Depolarization

As first described by Rowntree in 1923 [18], an acute reduction of plasma osmolality elicits disorientation and seizure activity in patients or experimental animals, independent of any physical brain damage noted above [13, 15]. Clinical examples include exercise-associated overhydration, hastened rehydration therapy, dialysis disequilibrium syndrome, or compulsive polydipsia. Each can become a clinical emergency if coupled with SIADH, the syndrome of inappropriate (i.e., low) antidiuretic hormone secretion [13, 14, 19‐21]. Studies using brain slices have determined that several mechanisms promote elevated excitability and synchronicity when plasma becomes acutely hypoosmotic: (1) Increased synaptic strength, gauged by increased evoked excitatory postsynaptic potential (EPSP) amplitude and duration (Fig. 2) [22, 23], resulting from a concentrating effect on released transmitter as astrocytes swell and reduce the ECS volume. (2) Increased spontaneous EPSP frequency [22‐24] synaptic responses, which are partly N-methyl-D-aspartate (NMDA) receptor-mediated [25]. (3) Increased field effects from reduced ECS volume, causing elevated extracellular resistance (R_e), meaning that a fixed current (I) generated by a synchronized burst discharge will produce increased extracellular negative voltage (V_e), from Ohm’s law (V_e = IR_e). This adds to transmembrane potential, the depolarizing “blip” recruiting nearby silent neurons [26‐28]. (4) Reduced ECS volume, which concentrates K⁺ and glutamate and further heightens excitation. (5) Altered intrinsic electrical properties of pyramidal neurons. Electrophysiological characteristics are subtly altered by acute osmotic change between − 80 and + 60 mOsm [22‐24]. The amplitude of active spike afterdepolarization and burst excitability are increased by hypoosmotic treatment (the latter shown in Fig. 3). These osmotic effects are mediated by a persistent Na⁺ current and increased membrane time constant [29]. Both are enhanced by hypoosmolality, adding to burst excitability [29]. Hypoosmotic change also increases the membrane time constant, slowing the spike afterdepolarization, and thereby promoting additional spiking (Fig. 3) (Both properties are counteracted by hyperosmolality, a crucial requirement for convincingly demonstrating osmoresponsiveness, adhered to by each study noted above). The excitatory actions result from osmolality dilutions by as little as 13%. Working together, they elevate intrinsic neuronal excitability (slightly) and network excitability (considerably), thereby promoting increased likelihood of hippocampal and neocortical interictal bursting and seizure initiation over many minutes. However, over hours and days, as neuronal swelling begins, excitability dramatically decreases (“Neuronal Recovery and Water Loss” section).

Does acutely reduced osmolality induce SD in vivo? The answer seems to be not at clinically relevant pathophysiological levels. Highly unphysiological hypoosmotic media promoted SD in one study [30] but not in three others [2, 7, 31]. One in vivo study reported SD generation under very low osmotic conditions [32]. We know of no clinical reports in which a patient with acutely lowered plasma osmolality showed SD symptoms. Nevertheless, osmotic gray matter swelling might promote regional SD propensity by increasing tissue excitability via the mechanisms noted above.

Subacute CSF osmolarity changes over hours or days are tolerated in vivo [9] via regulatory volume increases (RVI) or regulatory volume decreases (RVD). RVI counteracts initial passive hyperosmotic cell shrinkage, increasing cell volume to normal [3]. RVD involves the reverse pathway in response to hypoosmolality, acting to reduce cell volume [3, 33] through active ion and osmolyte efflux followed by water [33].

There is evidence supporting long-term RVD/RVI (hours and days) in experimental animals under osmotic stress [3, 34]. Brain cells can actively regulate volume in response to hypoosmotic stress over many hours by losing so-called organic solutes (also termed idiogenic osmoles), which are small molecules such as the amino acids glutamate and taurine [35]. In contrast, under hyperosmotic stress, organic solutes are synthesized to retain water and maintain volume. This volume regulation has been demonstrated by Arieff et al. [36] in animal models with hyperglycemia, in which brain tissue initially lost ~ 10% of its water (as measured by weight). However, after several hours, cerebral organic solute concentration was elevated to minimize water loss [36, 37], indicating adaptation through solute accumulation. It is unclear whether long-term RVI and RVD occur in glia, neurons, or both.

There is little evidence beyond isolated cell studies that brain cell RVI/RVD occurs over minutes following sudden osmotic challenge [2, 3]. This is important because short-term hypoosmotic conditions combined with SIADH can catastrophically swell the brain, which can be lethal due to the unyielding volume of the skull, spinal column, and meninges (Fig. 1) [3, 38, 39]. Worse, hypoosmolality promotes seizure activity (“An Acute Drop in Osmolality Increases Excitability of the Live Brain Slice” section), which in itself will promote brain swelling [3, 6, 40].

There is an inverse linear relationship between light transmittance (LT) by gray matter of a live slice and the degree of osmotic shift (Fig. 4) because astrocytes behave as excellent osmometers [2, 41]. The optics are not detailed here. In short, swollen cell membranes adapt a more planar configuration which scatters less light, increasing LT. However, dendritic beads, although swollen, are of an optimum diameter to scatter light, thereby decreasing LT [42]. Confirming how astrocytes and neurons change their volume is facilitated by the real-time imaging afforded by two-photon laser scanning microscopy (2-PLSM).

A study by Murphy et al. [43] claimed evidence of neocortical neuron swelling under extreme hypoosmotic stress. Their confocal imaging detected reduced fluorescence intensity, presumably from cytoplasmic dilution of the fluorophore, as a measure of neuronal swelling. This contradicted several studies using 2-PLSM [2, 7], and their interpretation is compromised by several issues. First, they detected hypoosmotic swelling in neurons close to the slice surface (30–40 µm) where cells are damaged from slicing (Fig. 5a, b). For instance, Andrew et al. [2] noted that many neurons shallower than 50 µm displayed swollen cell bodies and beaded dendrites (Fig. 5a). Second, Murphy et al. [43] saw no cell body swelling further from the slice surface (65–90 µm), supporting other reports of no hypoosmotic swelling of neurons detected in deeper, healthier regions [2, 7] of up to 200 µm [44]. Third, at depths > 40 µm, confocal microscopy loses significant resolution compared with 2-PLSM, preventing accurate measurement of dendritic and axonal volumes at depth. Because these regions represent > 99% of pyramidal neuron membrane surface, it is prudent to measure more than just the somata. Andrew et al. [2] measured all three regions. As with neuronal soma (and unlike adjacent astrocytes), dendrites and axons > 50 µm below the slice surface steadfastly maintained their volume between − 40 and + 80 mOsm. Then, under OGD, Andrew et al. [2] imaged each neuronal region that displayed constant volume under acute osmotic stress (“Neuronal Swelling in Response to Ischemia” section). In every case, there was rapid and dramatic swelling evoked by SD (somata shown in Fig. 6). Thus, neurons that were consistently unresponsive to osmotic shifts were clearly healthy. In fact, SD is the standard cause of neuronal swelling detected in 2-PLSM studies, whether induced by high K⁺ [2, 45], OGD [2, 7], or in vivo by ischemia [46].

Finally, the suggestion by Murphy et al. [43] that Andrew et al. [2] failed to detect neuronal swelling because of higher temperature (35 °C) or the aCSF not reaching deeper neurons is without merit because adjacent astrocytes at depth responded with appropriate osmotic volume changes. Likewise, Hirrlinger et al. [49] observed astrocytic osmotic swelling at depths > 40 µm, and swelling continued for 37 minutes, increasing almost linearly. Evidently, increased depth does not hinder aCSF perfusion or the observation of swelling.

In addition, no neuronal RVD or RVI was observed. Neither was RVD nor RVI detected in adjacent, osmotically-compliant astrocytes [2, 7]. We conclude that neurons do not exhibit detectable volume responses, nor short-term volume regulation under acute, physiologically relevant, osmotic shifts.

One type of mammalian neuron apparently displays osmoresponsiveness, although there is only minimal documentation that these neurons volume-respond in studies using nonisolated cells [50]. These neurons have long been theorized to intrinsically sense plasma osmolality changes by swelling and shrinkage during hypoosmotic and hyperosmotic conditions, respectively. For example, magnocellular neuroendocrine cells in the supraoptic and paraventricular hypothalamic nuclei increase firing with slightly elevated plasma osmolality [51], triggered by the shrinkage-induced opening of transient receptor potential vanilloid type-1 channels. This causes depolarization and discharge, promoting pituitary antidiuretic hormone (vasopressin) release, which signals renal water conservation. Other leaky BBB regions may also facilitate hyperosmolality detection by osmosensitive neurons, causing the perception of thirst [52, 53]. Similarly, astrocytic shrinkage for sensing hyperosmolality in these hypothalamic regions is important for thirst detection [54]. Notably, however, osmosensitive neurons have not been reported to express aquaporins.

Aquaporin-4 (AQP4) is expressed abundantly in glia, particularly astrocytic endfoot processes abutting blood vessels, ependymal cells, and the subarachnoid space lining [55, 56] ("Brain Cell Swelling with Osmotic Shifts" section). AQP4 expression is upregulated in several pathologies, including ischemia and TBI [57], and is known to significantly mediate cerebral edema through glial swelling [8, 57, 58]. Solenov et al. [57] observed reduced astrocytic swelling in AQP4-deficient mouse models. Although some glial osmoresponsiveness has been observed with AQP4 inhibition [59], this may be due to other membrane transporters passively moving water, providing residual permeability [8]. Although AQP4 expression is a major contributor to SD-associated glial swelling, it is not an absolute requirement, nor are aquaporins needed for neuronal osmosensitivity (“Osmosensory Neurons are Apparently Osmo-responsive” section).

Neuronal swelling during ischemia is classically considered osmotically driven. The textbook explanation of cytotoxic edema describes a premorbid process, whereby extracellular Na⁺ enters neurons through voltage-sensitive Na⁺ channels activated by ischemia. Na⁺ accumulates intracellularly, exacerbated by the Na⁺/K⁺ pump’s reduced ability to extrude Na⁺ [4]. Negatively charged intracellular proteins hold in K⁺ via Donnan forces. Na⁺ and K⁺ buildup, then somehow electrostatically draw in Cl⁻. Intracellular accumulation of all three ions then purportedly draws in water by osmosis through some undefined conduit and elicits neuronal swelling [65].

But is this scenario correct? Let us begin with the initial blood flow loss that lowers tissue ATP levels. This slows the Na⁺/K⁺ pump, causing some intracellular Na⁺ accumulation. Neurons may then respond with hyperpolarization for several seconds generated by an ATP-sensitive K⁺ current and/or by Na⁺/K⁺ pump upregulation [66]. Next follows a depolarizing phase (accompanied by spike inactivation presumably from inhibition of voltage-sensitive Na⁺ channel opening, although those channels contribute some inward Na⁺ current). As the Na⁺/K⁺ pump completely fails, a precipitous opening of a massive and nonspecific Na⁺/K⁺ conductance [67] initiates SD which resists glutamate receptor antagonists and tetrodotoxin (TTX) blockade. Prior to SD, the K⁺ transmembrane gradient ([K⁺]_i = 155 mM; [K⁺]_o = 4 mM) is opposite, although similar to the Na⁺ gradient ([Na⁺]_i = 12 mM, [Na⁺]_o = 155 mM). During SD, the membrane is freely permeable to Na⁺ and K⁺ which move down their concentration gradients. There is minimal net cation influx because Na⁺ influx is counteracted by K⁺ efflux. Dogma holds that Na⁺ influx exceeds K⁺ efflux because negatively charged intracellular proteins hold in K⁺, but entering Na⁺_i can play the same role, allowing K⁺ efflux. As well, voltage-sensitive K⁺ channels should open during SD, further increasing K⁺ conduction outward. So, there is no obvious cationic gradient drawing in Cl⁻. Although chloride could simply move inward down its concentration gradient ([Cl⁻]_i =  ~ 7 mM; [Cl⁻]_o = 140 mM) [68], Cl⁻ influx and water uptake during ischemia involves specific transporters (“Transport (Carrier) Proteins and Neuronal Swelling” section) rather than ligand-gated channels activated by γ-aminobutyric acid (GABA) or glycine. From experiments substituting extracellular Cl⁻ with other anions, Cl⁻ influx is not required for SD generation, but it is required for neuronal swelling and dendritic beading post-SD [31, 69]. Yet ischemic swelling and beading of neurons are considered independent of osmotic gradients [8, 62, 63] (“Osmosensory Neurons are Apparently Osmo-responsive” section).

This begs the question: What drives neuronal swelling in the wake of SD? Inhibiting the cotransporters NKCC1 and KCC2 (chloride transporters), anion exchange protein (AE3), and monocarboxylate transporter (MCT2) reduces dendritic beading in vivo, indicating that pronounced ion displacement during SD leads to transmembrane solute imbalances, dramatically reducing neuronal water influx in the seconds and minutes post-SD [31] (“Transport (Carrier) Proteins and Neuronal Swelling” section). Over the longer-term, the NKCC1 blocker bumetanide reduces cerebral edema in vivo post-TBI by decreasing cotransported water influx from stressed neurons [58]. Is this swelling osmotically driven? Unpublished experiments by Andrew and Kirov associated with a previous publication by Andrew et al. [2] show that post-SD slice swelling cannot be countered by simple exposure to hyperosmolar aCSF lasting 10–15 minutes (Fig. 7b). Post-SD, neuronal expansion is ongoing and continuous over a minute or more, to the point where cell bodies become swollen spheroids. Additionally, dendritic bead diameter can approach 5 µm (Fig. 7a).

Innumerable injured neurons near the cut slice surface do not lyse over many minutes (Fig. 8a) or even hours (Figs. 5a, b), indicating that a cytosolic water-osmolyte equilibrium must be reached. Neuronal volume can increase severalfold post-SD, so likely more than Na⁺_i, K⁺_i, and Cl⁻_i are holding in water. Degradation of nondiffusible proteins (or any large molecules) into their constituent parts may increase cytosolic osmolality according to Gibbs–Donnan equilibrium, which would hold water intracellularly [70]. However, it is unclear what percentage of such water would be “restricted.” Restricted water is constrained to the hydration layers of macromolecules (reviewed in [68]). The layers of restricted water are only a few molecules wide, but the density of macromolecules in the cytoplasm is high. Thus, while the amount of restricted water is ~ 50% of all cytoplasmic water, it cannot contribute to the intracellular osmolality [68]. So, it is unclear to what degree macromolecular breakdown would draw in water. Alternately, the role of membrane stretching and/or increased hydrostatic pressure on swelling dynamics is difficult to model, but it has been proposed that increased intracellular pressure from the stretched plasma membrane could reduce water influx [71], which might stabilize the grossly swollen state of cell bodies and dendrites.

An older definition of cytotoxic edema described intact capillaries as the source of water and electrolyte (termed ionic edema). However, it is apparent that a major source is CSF inflow to the quickly-expanding perivascular space (PVS) from SD-evoked propagating vasoconstriction [74]. This propagating ‘spreading ischemia’ immediately follows in the wake of SD, as do astrocytic and neuronal swelling. Astrocyte swelling is understood to result from K⁺ taken up from depolarized neurons, followed by osmotic Cl⁻ uptake, and selective water uptake via AQP4. Note again, AQP4’s recently-appreciated role in SD [8, 62].

Another contributing factor to neuronal swelling may be metabolic water, a product of oxidative metabolism (specifically, oxidative phosphorylation). Total mammalian neuronal metabolic water is estimated from different models to be 7–56%, although it was accurately measured by Li et al. [75] in bacteria to be 30–40%. We suggest that during ischemia, residual oxidative metabolism may generate intracellular water, enhancing swelling.

Neuronal swelling can also be artificially-elicited with neurotoxins that open Na⁺ channels, thereby pathologically increasing [Na⁺]_i. For instance, veratridine opens Na⁺ channels, eliciting Na⁺ influx and SD [76] which swells neurons in brain slices [65]. Glutamate receptor agonists like NMDA [65] or domoic acid [77] also swell neurons in slices without evoking SD. Blockers of the voltage-gated Cl⁻ channel, SLC26A11, reduced neuronal swelling from veratridine and NMDA in slices, but had little effect on SD onset time or neuron swelling [31, 65, 78]. These studies have more relevance to neurotoxicity than to ischemia or SD.

Except for neurotoxic poisoning (previous paragraph), ischemia-induced SD caused by stroke, TBI, subarachnoid hemorrhage, or sudden cardiac arrest promotes acute neurological damage. Specifically, neuronal swelling and beading is initiated within seconds of SD invading imaged gray matter [79]. Recently, it has become clear that SD itself expands the PVS, exacerbating brain swelling. In mouse stroke models, a single SD event evokes propagating vasoconstriction, expanding the PVS [74]. Water entering the PVS from CSF accesses astrocytes through endfoot AQP4, contributing to early poststroke cytotoxic edema. How neurons swell is more mysterious (“Neuronal Swelling in Response to Ischemia”, “Transport (Carrier) Proteins and Neuronal Swelling” sections). Importantly, cytotoxic edema is a direct result of SD, yet SD is not even mentioned in many recent reviews of neuronal death [80‐83].

Aquaporins are expressed in astrocytes, accounting for their osmoresponsiveness (“Short-Term Brain Cell Volume Regulation” section). Table 1 provides an overview of aquaporin isoform expression by brain cell type [10, 56, 57, 84‐90]. Astrocytes swell with hypoosmolality and shrink with hyperosmolality within seconds, while most neurons do not [2, 7, 16]. Neurons resist osmotic swelling (Figs. 5c, 7a) or shrinkage (Fig. 7a). So, astrocytes are the primary cellular contributor to cerebral edema in an acute dilutional situation (Fig. 1) commonly induced by a behavioral or iatrogenic event (Box 1).

In contrast, neurons start to swell within seconds following SD in situ induced by ischemia [2, 91, 92], or in slices induced by elevated [K⁺]_o (Figs. 4, 5B3), OGD [2] (Figs. 5b5, 6), hyperthermia, or hypothermia [93, 94]. In all cases, SD is necessary and sufficient to evoke neuronal swelling. The mechanism of this swelling is poorly-understood, although the common denominator is likely Na⁺/K⁺ pump dysfunction [95].

The intact brain becomes hyperexcitable in response to acute osmotic swelling of the astrocytes (“An Acute Drop in Osmolality Increases Excitability of the Live Brain Slice” section). This can induce generalized seizure in vivo over tens of minutes, particularly in cases of undiagnosed SIADH [13‐15]. This excitability arises at the cellular level and is not a result of brain compression. Over many hours following stroke or TBI, neurons may join astrocytes in swelling, but these neurons display lowered excitability [58, 99].

Twelve hours after middle cerebral artery occlusion (MCAo) in mice, there are subsets of pyramidal neurons near the infarct that are necrotic (red arrows, Fig. 9a), whereas others appear normal (white arrows, Fig. 9a). Many other pyramidal cells display intermediate disruption (i.e., swollen cell bodies) (S, Fig. 9a–c) and beaded dendrites (yellow circles, Fig. 9a–c) presumably the result of at least one SD event. These fields display reduced excitation electrophysiologically (not shown) [99]. Similarly, adjacent to a TBI lesion, swollen neurons are observed in cortical regions exhibiting reduced excitability when patch-clamped [58]. So, ischemic edema swells neurons and astrocytes as well as reducing excitability over many hours or even days. This contrasts with acute osmotic swelling where neuronal and network excitability increases without neuronal volume changing.

Ions may move with or without their hydration shells while being conducted through ion channels. Most K⁺ channels completely strip each K⁺ ion of its water shell before passage. In the past, it was thought that Na⁺ ions were completely stripped of their shells by the voltage-sensitive Na⁺ channel, but it may only be partial. Regarding glutamate receptor-mediated Na⁺ influx and K⁺ efflux, both ions remain partly hydrated as they pass through the channel. But the channel is non-selective, so Na⁺ and K⁺ counteract each other’s associated water flux. Indeed, in Biedermann et al. [100], appended video simulations do not reveal directional water movement through the α-amino-3-hydroxy-5-methyl-4-isoxazoleproprionic acid (AMPA) receptor channel. Likewise, neuronal swelling in response to NMDA receptor activation involves the cotransporters NKCC1 and swelling-activated KCC2 [101]. Directional water moving as hydration shells of ions conducted through glutamate receptor-activated channels has not been reported. This all supports the argument that water movement through the Na⁺ and K⁺ channels (i.e., Those channels most involved in cationic movement during neuronal activation) are not major conduits of transmembrane water.

Ion channels passively conduct ions. To maintain selectivity, ions are partially or entirely stripped of their water shells prior to channel entry. However, with cotransporters, water ‘rides along’ with substrates. Mammalian neurons express chloride cation cotransporters (CCCs), specifically Na⁺/K⁺/Cl⁻ (NKCC1) and K⁺/Cl⁻ (KCC2) transporters [62]. Both rely on cation gradients generated by the Na⁺/K⁺ pump as an indirect energy source [102] (Fig. 10A). NKCC1 couples the intracellular transport of two Cl⁻ with one K⁺ and one Na⁺ [102]. KCC2 uses the outward K⁺ gradient to couple efflux of one Cl⁻ and one K⁺. CCCs provide a pathway for transmembrane water transport, even against a prevailing osmotic gradient [62]. Of importance to neuronal swelling, NKCC1 can, along with Cl⁻, facilitate water influx against opposing gradients up to 50 mOsm [62, 103]. This finding was supported by Zhang et al. [102] using cryo-electron microscopy structural analysis of the NKCC1 inward-open state, which contains a continuous pathway for water [102]. Neurons can exhibit volume increases and decreases through NKCC1 and KCC2 transport respectively [58], the latter showing more expression in mature neurons [58]. Inhibiting CCCs and reducing [Cl⁻]_o in slices significantly reduces SD-induced dendritic beading [31], indicating the importance of Cl⁻ gradients and altered CCC activity in neuronal swelling. Over hours post TBI, the neuronal NKCC1:KCC2 expression ratio increases [58, 104, 105] and is associated with elevated [Cl⁻]_i [58].

Although these CCCs are involved with neuronal swelling, they lack sufficient water-carrying capacity alone to entirely mediate Cl⁻/water influx [102]. Considering the importance of Cl⁻ gradients to neuronal swelling, these CCCs are likely one of several Cl⁻ transporters responsible for water influx. For instance, anion exchange protein 3 (AE3) facilitates Cl⁻/HCO₃⁻ exchange and is another (less well understood) potential mediator of Cl^-/water movement [31, 108].

Monocarboxylate transporters (MCTs) also move water across membranes by facilitating proton-linked short-chain monocarboxylate (lactate, pyruvate, ketone body) transport [62, 107, 109]. There are four isoforms; MCT1 in BBB endothelium, MCT1 and 4 in astrocytes [110], and MCT2 in astrocytic endfeet and neurons [110, 111]. MCT2 displays a tenfold higher substrate affinity [107, 110]. MCT transport direction depends on the prevailing substrate and proton gradients [107]. The translocation cycle of MCT (Fig. 10b) when transporting L-lactate, involves one proton followed by lactate binding the outward-open conformation. A conformational change to an inward-open state releases substrates and associated water intracellularly.

Because glucose is the brain’s primary energy source, the BBB is relatively lactate-impermeable [110]. However, metabolism altered by hypoglycemia [111], ischemia, or TBI [112, 113] increases BBB monocarboxylate permeability [107] through enhanced MCT1 action. From this, there is MCT1- and MCT4-mediated astrocytic lactate uptake [114], and additional lactate production from glycogen stores to spare residual glucose for neurons [114, 115]. Astrocytic lactate is released into the ECS [112, 113, 116] for neuronal uptake via MCT2 [117]. This is thought to allow for neuronal ATP generation though the lactic acid cycle involving lactate conversion to pyruvate [114, 115]. This provides support for the debated astrocyte neuron lactate shuttle hypothesis.

MCT2 lacks passive water permeability [62], but can cotransport water with its substrate (~ 500 water/lactate) and is enhanced during ischemia [62]. Although MCT2 has potential to induce dendritic water accumulation with ischemia [62, 114], and combined inhibition of AE3 and MCT2 in vivo [31] yielded reduced post-SD beading, specific MCT2 inhibition with 4-CIN [31, 117] does not affect dendritic beading. The role of MCT2 these studies appears to lie in neuronal recovery, aiding in both neuronal water efflux [31] (“Neuronal Recovery and Water Loss” section) and metabolic substrate uptake for functional recovery [117].

Glucose is an essential cerebral energy substrate under normal metabolic conditions [118‐121]. Neuronal glucose transporter (GLUT3) facilitates glucose influx [119] and its expression is transiently and globally enhanced under ischemia [118, 120]. Its water cotransport properties as observed in Xenopus oocytes mimic glial isoform GLUT1 [122], meaning GLUT3 may cotransport ~ 330 water/glucose, in an analogous translocation cycle outlined in Fig. 10A [62]. Given its ischemic upregulation and water cotransport capacity, GLUT3 could mediate neuronal swelling [120, 122, 123]. The caveat is that glucose supply will drop during ischemia, although not entirely [124]. Given the graded transition of reduced cerebral blood flow radiating from the ischemic core, there may be enough residual glucose, particularly in distant regions, to briefly facilitate GLUT3-mediated water uptake [124, 125]. During ischemia, neurons are metabolically-stressed, leading to brief enhancement of GLUT3 glucose and water uptake, causing the accumulation and trapping of this water intracellularly (“N-Acetyl-L-Aspartate/N-Acetylaspartylglutamate Metabolism and Molecular Water Pump Cycle” section). This process has potential to contribute to neuronal swelling and may help keep the metabolically-stressed neurons alive for a short time following ischemic onset. However, the exact magnitude and duration of GLUT3-mediated water intake have not been investigated.

As mentioned, briefly-enhanced neuronal glucose and water uptake is possible through GLUT3. We propose that this process further swells neurons via increased oxidative glucose metabolism, causing the production and trapping of metabolic water. Because enhanced GLUT3 activity under conditions such as prolonged stimulation [126] corresponds to elevated metabolic water production [127], the brief period of enhanced GLUT3 activity as described above ("Glucose Transporter (GLUT3)" section) may also enhance intraneuronal metabolic water production and trapping, contributing to swelling. Similar to GLUT3 activity, this is likely brief, lasting only until circulating oxygen is entirely lost. Although cerebral blood flow and oxygen extraction fraction poststroke have been observed, each showing graded reductions moving away from the ischemic core [124, 125, 127], the exact duration and contribution of metabolic water production to neuronal swelling are unknown.

N-acetyl-L-aspartate (NAA) is a derivative of aspartic acid synthesized and stored in neurons [127‐129]. NAA synthesis is cyclic, thought to be associated with a transmembrane molecular water pump (MWP) system [127]. Normally, following synthesis, NAA leaves the neuron down its gradient, into the ECS (Fig. 11a) [127]. N-acetylaspartylglutamate (NAAG) is analogous, although less common, but follows the same cycle. Oligodendrocytes [127] and possibly astrocytes [130] take up NAA/NAAG for their hydrolysis, and the by-products are sent back to neurons for resynthesis [127, 128]. NAA/NAAG efflux is associated with the total efflux of ~ 85 metabolic water molecules [127, 128]. Assuming that most CNS NAA/NAAG production is neuronal, this system can theoretically remove up to 50% of neuronal metabolic water produced under normal conditions. With ischemia, NAA/NAAG synthesis is reduced [127], yet the rate of NAA/NAAG hydrolysis remains the same. This destroys the NAA/NAAG gradients (Fig. 11b) [127]. We hypothesize that metabolic water accumulates, promoting acute neuronal swelling during ischemia.

The functional threshold of decreased perfusion with ischemia is the minimum perfusion allowing neuronal function without irreversible damage [132]. The functional threshold of a cell during ischemia depends on cell type, region, infarct time, and magnitude [132, 133]. Generally, total brain death does not occur immediately following acute ischemia provided there is timely reperfusion [125], and there is a 3 to 6 hour window of opportunity post-insult where some brain tissue can be salvaged [29, 132]. During reperfusion and recovery, neurons must remove excess intracellular water [29, 132]. Similar to water accumulation, the mechanisms underlying neuronal water efflux with ischemic recovery are poorly-characterized.