OUP user menu

TGF-β receptor-mediated albumin uptake into astrocytes is involved in neocortical epileptogenesis

Sebastian Ivens, Daniela Kaufer, Luisa P Flores, Ingo Bechmann, Dominik Zumsteg, Oren Tomkins, Ernst Seiffert, Uwe Heinemann, Alon Friedman
DOI: http://dx.doi.org/10.1093/brain/awl317 535-547 First published online: 22 November 2006


It has long been recognized that insults to the cerebral cortex, such as trauma, ischaemia or infections, may result in the development of epilepsy, one of the most common neurological disorders. Human and animal studies have suggested that perturbations in neurovascular integrity and breakdown of the blood–brain barrier (BBB) lead to neuronal hypersynchronization and epileptiform activity, but the mechanisms underlying these processes are not known. In this study, we reveal a novel mechanism for epileptogenesis in the injured brain. We used focal neocortical, long-lasting BBB disruption or direct exposure to serum albumin in rats (51 and 13 animals, respectively, and 26 controls) as well as albumin exposure in brain slices in vitro. Most treated slices (72%, n = 189) displayed hypersynchronous propagating epileptiform field potentials when examined 5–49 days after treatment, but only 14% (n = 71) of control slices showed similar responses. We demonstrate that direct brain exposure to serum albumin is associated with albumin uptake into astrocytes, which is mediated by transforming growth factor β receptors (TGF-βRs). This uptake is followed by down regulation of inward-rectifying potassium (Kir 4.1) channels in astrocytes, resulting in reduced buffering of extracellular potassium. This, in turn, leads to activity-dependent increased accumulation of extracellular potassium, resulting in facilitated N-methyl-d-aspartate-receptor-mediated neuronal hyperexcitability and eventually epileptiform activity. Blocking TGF-βR in vivo reduces the likelihood of epileptogenesis in albumin-exposed brains to 29.3% (n = 41 slices, P < 0.05). We propose that the above-described cascade of events following common brain insults leads to brain dysfunction and eventually epilepsy and suggest TGF-βRs as a possible therapeutic target.

  • astrocytes
  • blood–brain barrier
  • epileptogenesis
  • neocortex
  • transforming growth factor beta receptors


Epilepsy, affecting 0.5–2% of the population worldwide, is one of the most common neurological disorders. While the characteristic electrical activity in the epileptic cortex has been extensively studied, the mechanisms underlying epileptogenesis are still poorly understood. Focal neocortical epilepsy often develops following traumatic, ischaemic or infectious brain injury. Under these conditions, vasculature damage is common and includes a local compromise of the blood–brain barrier (BBB; Tomkins et al., 2001; Neuwelt, 2004; Abbott et al., 2006). Ultrastructural studies of human epileptic tissue demonstrating increased micropinocytosis and fewer mitochondria in endothelial cells, a thickening of the basal membrane, and abnormal tight junctions further support the notion of lasting BBB dysfunction in at least some forms of epilepsy (Kasantikul et al., 1983; Cornford and Oldendorf, 1986; Cornford, 1999). Indeed, clinical and animal studies showed that vascular damage and, specifically, opening of the BBB is often observed in epileptic brain regions, but was generally believed to result from the seizure activity, rather than contribute to its generation (Cornford, 1999). However, conversely to that option, we have observed in some post-traumatic patients a long-lasting BBB opening corresponding to abnormal cortical function as revealed by EEG analyses (Korn et al., 2005). These observations led us to hypothesize that BBB dysfunction may have a direct role in the pathogenesis of epilepsy. Supporting this hypothesis of primary BBB lesions as an initial event leading to neocortical epilepsy, we demonstrated that opening of the BBB in the rat somatosensory cortex exposes the secluded brain microenvironment to serum components, resulting in the delayed development of epileptiform activity (Seiffert et al., 2004). However, the mechanisms underlying cortical dysfunction following BBB injury are unknown. Here we have set out to elucidate these mechanisms, in an animal model of BBB disruption.

Experimental evidence suggests that astrocytes display modified properties in epileptic tissue from human and animal (Pollen and Trachtenberg, 1970; Bordey and Sontheimer, 1998; Hinterkeuser et al., 2000; Jauch et al., 2002; Eid et al., 2005) and are likely to play a key role in the pathogenesis of epilepsy (Seifert et al., 2006). Since astrocytes are known to be contributors to BBB formation (Ballabh et al., 2004) and enhanced immunolabelling against the astrocytic marker, glial fibrillary acidic protein (GFAP), is observed to follow a breach in the integrity of the BBB (Seiffert et al., 2004), we hypothesized that these cells may play a role in epileptogenesis after BBB disruption.

In this study, we investigated the mechanisms underlying epileptogenesis induced by BBB opening. We demonstrate for the first time in a rat model the role played by astrocytes in epileptogenesis and propose a cascade of events that takes place during the window period of epileptogenesis, i.e. after BBB opening and before the development of epileptiform activity. Surprisingly, we have identified transforming growth factor β receptor (TGF-βR) as a key player in the cellular response, and demonstrate an effective blockade of the cascade, and the resulting epileptiform activity, by blocking TGF-βRs.

Material and methods

In vivo experiments

All experimental procedures were approved by the ethical committees dealing with experiments on animals at Charité University Medicine, Berlin and Ben-Gurion University of the Negev, Beer-Sheva. The in vivo experiments were performed as described previously in Wistar rats (Seiffert et al., 2004). For the ‘treated rats’, we added to the artificial CSF (ACSF) the BBB-disrupting agent deoxycholic acid sodium salt (DOC, 2 mM, Sigma-Aldrich, Steinheim, Germany) or bovine serum albumin (BSA, 0.1 mM, >98% in agarose-cell electrophoresis; Merck, Darmstadt, Germany, ordering number 1.12018.0025), corresponding to 25% of serum albumin concentration (0.4 mM determined for 10 rats, see also Geursen and Grigor, 1987; osmolarity 303–305 mOsmol/l). ACSF alone was applied to the sham-operated controls. The composition of the ACSF was (in mM): 129 NaCl, 21 NaHCO3, 1.25 NaH2PO4, 1.8 MgSO4, 1.6 CaCl2, 3 KCl, 10 glucose. In some experiments, the cortex was exposed (30 min) to ACSF containing the TGF-βR1 kinase activity inhibitor SB431542 (100 μM, Tocris, Bristol, UK) in dimethyl sulphoxide (DMSO, 0.1%, Merck) and TGF-βR2 antibody (50 μg/ml, Santa Cruz Biotechnology, Santa Cruz, USA) and subsequently exposed to BSA (0.1 mM) for 30 min. The control brains for these experiments were superfused with ACSF containing 0.1% DMSO, followed by ACSF with DMSO and BSA (0.1 mM).

In vitro slice preparation

Brain slices were prepared by standard techniques (Kaufer et al., 1998; Pavlovsky et al., 2003; Seiffert et al., 2004). To study albumin uptake, slices were incubated in a submerged chamber containing ACSF with 0.004–0.1 mM fluorescein isothiocyanate (FITC)-conjugated albumin (Sigma-Aldrich, Germany, osmolarity 311–312 mOsmol/l) for 5–60 min. In some experiments, slices were incubated with non-labelled BSA (0.004, 0.04 and 0.4 mM, osmolarity 308–311 mOsmol/l) in the presence of FITC-albumin, with 0.004–0.04 mM FITC-labelled dextran (70 kDa, Sigma-Aldrich) or with 0.04 mM Texas-Red-conjugated ovalbumin (Invitrogen, Karlsruhe, Germany). To block TGF-βRs, slices were incubated (60 min) in ACSF containing SB431542 (in DMSO) or with TGF-βR2 antibody. FITC-albumin (0.004 mM) was then added, and the slices were incubated for another 25 min. Following incubation, slices were washed with oxygenated ACSF (30 min) in a submerged chamber and prepared for histological analysis (see below).

Electrophysiological recordings

For electrophysiological recordings, we used brain slices prepared as mentioned above. Following the slicing procedure slices were transferred immediately to the recording chamber, maintained at 36°C, as reported previously from our laboratory (Seiffert et al., 2004). For detection of epileptiform activity we recorded field potentials from 10 positions along the treated/sham-treated region in cortical layer 4, stimulating on the border of white to grey matter. Extracellular potassium concentrations ([K+]o) were measured with ion-sensitive microelectrodes (ISMEs, Lux and Neher, 1973; Jauch et al., 2002). For K+-ionophoresis, double-barrelled theta glass electrodes with slightly angled tips were filled with 1 M KCl and 154 mM NaCl and glued to the ISME (tip distance: 50–80 μm). K+ was applied by ionophoresis (60 s, 150–1000 nA). Injections were repeated at least three times at 5 min intervals to confirm stability. Intracellular recordings were performed with sharp microelectrodes using standard techniques (Seiffert et al., 2004).

Drug application

Kir and K+ leak currents were blocked by BaCl2 (100 μM and 2 mM, respectively), dissolved in sulphate-free ACSF (Ransom and Sontheimer, 1995; Jauch et al., 2002). To isolate astrocytic currents, the following drugs were applied in combination before BaCl2 application: 30 μM 2-amino-5-phosphovaleric acid (APV), 30 μM 6-cyano-7-nitroquinoxyline-2,3-dione (CNQX), 10 μM bicuculline and 1 μM tetrodotoxin (TTX) (all from Tocris, Bristol, UK) were used to block N-methyl-d-aspartate (NMDA), AMPA/KA, and GABA receptors and voltage activated Na+-channels, respectively. Drugs were applied by addition to the ACSF.

Evaluation of BBB integrity

Two approaches were used to estimate BBB integrity: (i) ex vivo measurements following intra-peritoneal injection with 2 ml of 2% Evans blue (Sigma, St Louis, USA) (Friedman et al., 1996; Seiffert et al., 2004); and (ii) image analyses of in vivo MRI measurements by using a 7 tesla scanner (Pharmascan 70/16 AS, Bruker Biospin, Ettlingen, Germany) with a 16 cm horizontal bore magnet and a 9 cm (inner diameter) shielded gradient having a maximum strength of 300 mT/m. Rats were anaesthetized with 1.5% isoflurane delivered in 100% O2 via a face mask and then placed in the centre of a 38 mm RF coil on a heated pad. Respiration and pulse rate were continuously monitored (monitoring unit Model 1025; SA Instruments, Inc., Stony Brook, New York). Coronal slices were imaged (35 slices, slice thickness = 0.5 mm). The field of view was 3 × 3 cm, and the matrix was 256 × 256, resulting in an in-plane resolution of 117 μm. Two brain imaging sequences were performed: (i) T1-weighted 2D turbo spin echo with RARE factor 2 (TR 1141.7 ms, TE 13.2 ms, 8 averages, total scan time 19 min: 30 s), in which the sequence was repeated before and after the injection of the BBB non-permeable agent gadolinium diethylene triamine pentaacetate (Gd-DTPA, 0.5 mol/l, 0.5 ml/200 g body weight; Magnevist, Schering, Berlin, Germany); and (ii) T2-weighted sequence with RARE factor 4 (TR 5046.6 ms, TE 36.5 ms, 5 averages, total scan time 26 min: 54 s). Spatially matching T1 images were compared for statistically significant differences in signal enhancement, reflecting changes in BBB permeability (Tomkins et al., 2001).


For histological experiments, rat brains were fixed by transcardial perfusion with 4% paraformaldehyde in 0.1 M phosphate-buffered saline. After perfusion, brains were kept in the same fixative at 4°C overnight. Brains were then removed from the skull, dissected, and treated with 96% alcohol overnight and subsequently paraffin embedded in accordance with routine procedures. Eight to ten micrometre coronal sections were mounted. Immunohistochemistry was performed on 10 μm paraffin sections. Sections were incubated with primary antibodies at 4°C overnight. We used rabbit antibodies against GFAP (1 : 400, DakoCytomation, Glostrup, Denmark), microtubule-associated protein 2 (MAP2, 1 : 500, Sigma) and Kir 4.1 (1 : 200, Alomone Labs, Jerusalem, Israel). Signal detection was achieved by incubation with secondary antibody for 2 h at room temperature. Alexa Fluor 568 goat anti-rabbit antibody (1:200, MoBiTec, Göttingen, Germany) was used for red fluorescence, and biotinylated goat anti-rabbit antibody (1:250, Vector, Peterborough, UK) followed by a standard ABC-DAB development was used for non-fluorescent staining (Bechmann et al., 2000). To verify double-labelling throughout the entire extent of the cells, we examined them in orthogonal planes with a Zeiss Axiovert 510 confocal microscope (Thornwood, New York). Upper and lower thresholds were set with the range indicator function. We obtained optical stacks of 1 μm thick sections through putatively double-labelled cells.

Gene expression

mRNA levels were determined using quantitative RT–PCR by real-time kinetic analysis with an iQ5 detection system (Bio-Rad). Primer pairs specific to GFAP (forward: 1138+ 5′AGAAAACCGCATCACCATTC3′; reverse: 1287− 5′TCCTTAATGACCTCGCCATC3′), Kir 4.1 (forward: 150+ 5′GAGACGACGCAGACAGAGAG3′; reverse: 310− 5′CCACTGCATGTCAATGAAGG3′) and actin (forward: 1012+ 5′GGGAAATCGTGCGTGACATT3′; reverse: 1081− 5′GCGGCAGTGGCCATCTC3′) were used. Real-time PCR data were analysed using the Livak 2-Delta Delta C(T) calculation method (Livak and Schmittgen, 2001). Presented are percentages of gene expression level of the treated hemisphere, as compared with the non-operated contralateral hemisphere. Actin mRNA levels were used as internal controls for variations in sample preparation.

Data acquisition and analysis

Signals were amplified (SEC-10L, NPI, Tamm, Germany), filtered at 3 and 0.03 KHz (field potential, K+ signal, respectively), displayed on an oscilloscope, digitized on-line (CED-1401, Cambridge, UK) and stored for off-line analysis. Data and bar graphs are presented throughout as means ± SEM. Differences between treated and control slices were determined by the non-parametric Mann–Whitney test for independent samples. The effect of pharmacological agents was tested with the non-parametric Wilcoxon signed rank test for related variables. We performed all statistical tests using SPSS 12.0.1 for Windows. P < 0.05 was taken as the level of statistical significance.


Brain exposure to serum albumin leads to cortical dysfunction

In order to study the consequences of an open BBB on brain function, we employed an established in vivo model of disturbing BBB using DOC sodium salt. Using this model we produce a localized, highly reproducible perturbation in the BBB by opening a cranial window over the somatosensory region through which the exposed cortex is superfused with DOC solution (2 mM) for 30 min. In vivo MRI obtained 24 h after exposure to DOC confirmed BBB opening by showing local enhancement of the T1 signal following injection of Gd-DTPA. Both a local increase in cortical diameter and an increased T2 signal indicated local vasogenic brain oedema (Fig. 1A and B, n = 4). The MRI images also demonstrated that there was no penetrating injury or significant intracortical bleeding due to the treatment, supporting previous histological analyses (Seiffert et al., 2004).

Fig. 1

Focal BBB disruption causes prominent cortical dysfunction. (A) T2 sequence MRI of a rat brain 24 h following BBB opening. Note local brain swelling due to vasogenic oedema in the treated region (arrows). (B) Colour-coded T1 image showing areas of significant signal change after gadolinium-DTPA injection. Colour bar represents percentage of contrast enhancement. White arrows point to the intraventricular choroid plexus, normally lacking a BBB. (C) Electrophysiological recordings from sham-operated, DOC- and albumin (alb)-treated cortices one week after treatment. The filled bar graphs represent the averaged integral of the evoked field potential 50–500 ms after stimulation (marked with dotted line) and the empty bar graphs the percentage of slices with paroxysmal activity. (D) Simultaneous recordings from an albumin-treated cortex one week after treatment. Electrode numbers are displayed on the left. Inset shows a photograph of a treated slice; the dots represent typical locations of recording (white) and stimulating (black) electrodes. The line represents the region of the exposed cortex in the treated (T) or sham-operated (S) animals. (E and F) Voltage maps representing extracellular recordings from 11 electrodes positioned along a sham (E) and a treated (F) slice one week after surgery. Note the propagated evoked activity in the treated slice. x-axis represents time and y-axis distance along the cortex. Colour bar represents voltage amplitudes. Black arrows point out location and time of stimulation.

Treatment with either DOC or BSA induced indistinguishable hypersynchronized epileptiform activity in the treated region, as expected (see below and Seiffert et al., 2004). In the cortices of control sham-operated animals, brief electrical stimulation of the white matter evoked electrophysiological responses in a small cortical region (<2 mm in width). These responses were characterized by a short fixed latency (<7.5 ms), consistent with direct stimulation of nearby cortical columns (Fig. 1C and E). In contrast, in BSA- and in DOC-treated brains, these ‘early’, short latency responses, were followed by long-lasting paroxysmal field potentials that could be evoked in the entire treated region (∼3–5 mm, Fig. 1C, D and F). A tested slice was regarded to display abnormal epileptiform activity only when a clear, delayed all-or-none paroxysmal response was observed >50 ms after low intensity white matter stimulation (<2 × Ithreshold). The integral of the field potential was measured 50–500 ms after stimulation and was found to be similar in DOC- and albumin-treated slices but significantly lower in control slices (Fig. 1C). In both control and treated brains, the early evoked synaptic response was limited to a narrow band of cortex, while the paroxysmal prolonged activity was propagating along a wide cortical area within the treated region (Fig. 1D–F). Simultaneous recordings from treated slices using multiple electrodes revealed similar propagating epileptiform activity 1 week after treatment for both DOC and albumin-treated slices (6.18 ± 1.98 and 5.31 ± 1.75 mm/s, respectively, n = 4 for each group, Fig. 1D–F). Clear epileptiform activity was recorded in 72% of slices (DOC: n = 100 out of 139, BSA: 36 out of 50 slices) from over 90% of the treated rats (DOC: n = 47 of 51, BSA: 12 of 13 rats), but in only 9.1% of slices (2 of 22) from sham-operated rats (1 of 7) and in 16.3% of slices (8 of 49) from the non-treated contralateral hemisphere of treated rats (6 of 19 animals, Pearson χ2 test, P < 0.001). These findings point to reorganization of the BBB-disrupted cortex in a manner similar to the chronically injured (Prince and Tseng, 1993), undercut (Hoffman et al., 1994), or maldeveloped cortex (Jacobs et al., 1996). We observed spontaneous recurrent partial seizures in three of our treated animals, sometime followed by secondary generalization. The observed behavioural spontaneous seizures, while not investigated in detail under this study, support the notion that paroxysmal activity observed in the in vitro slice preparation indeed reflects abnormal epileptic network activity in vivo (video; available as supplementary material at Brain Online).

Albumin is selectively transported into astrocytes

Like BBB opening, direct application of serum albumin in vivo causes cortical dysfunction (Fig. 1). To confirm that our BBB opening protocol results in diffusion of serum albumin into the brain's extracellular space, we injected Evans blue intra-peritoneally and then traced the BBB non-permeable albumin–Evans blue complex (red fluorescence, Fig. 2A) in brain capillaries (Ehrlich, 1885; Friedman et al., 1996). While in control brains fluorescence was limited to the intra-capillary space, after BBB opening Evans blue–albumin complex was observed around the capillaries (Fig. 2A). Six to eight hours post-treatment, the albumin–dye complex was detected inside some cellular elements (Fig. 2A, arrows). In order to confirm the inclusion of albumin in the protein–dye complexes, we performed immunohistochemistry staining with an antibody directed against serum albumin. Albumin antibody staining produced a similar staining to the Evans blue dye distribution (data not shown), further verifying the penetration of serum albumin into the brain microenvironment.

Fig. 2

Albumin is preferentially transported into astrocytes. (A) Sections from control and BBB-treated animals 6 h following intra-peritoneal injection of Evans blue. Note the separation between intravascular Evans blue–albumin complex (red) and cellular elements of the brain (DAPI staining in blue) in controls, compared with the extracellular and intracellular staining under BBB opening. Arrows mark apparent membrane processes of stained cells. (B) Direct exposure of brain slices in vitro to FITC-albumin resulted in fast extranuclear (5 and 10 min), and nuclear staining (30 min) of cells. (C) Cellular elements labelled with FITC-albumin resembled astrocytes (arrows) and perivascular cells (open arrows). Inset: Co-localization of FITC-albumin and DAPI nuclear staining. FITC-dextran was taken up only by perivascular cells (right panel). (D) Co-administration of FITC-albumin and Texas Red-ovalbumin: Uptake of ovalbumin is limited to perivascular cells. Inset: Co-localization of albumin and ovalbumin in perivascular cells but not in a nearby parenchymal cell. (E and F) Confocal imaging of FITC-albumin labelled cells showing co-localization with cells positively immunolabelled for GFAP (astrocytes, E) but not for MAP2 (neurons, F). In all insets scale bar represents 10 μm. (G) Number of stained cells at different times after exposure to FITC-albumin (n = 15 sections, three slices at each time point). (H) Addition of non-labelled BSA resulted in a dose-dependent decrease in the number of FITC-albumin labelled cells (n = 15 sections, three slices at each time point).

To further explore the mechanisms underlying albumin uptake by brain cells, we directly exposed cortical slices to albumin labelled with FITC or biotin. Prominent intracellular staining was evident in the neocortex and hippocampus of all slices (242.2 ± 10.3 cells/mm2, n = 269 windows from 50 sections, 10 slices). The number of stained cells increased during the first 40 min of exposure (87.9 ± 20.0, 93.8 ± 60.5, 223.0 ± 25 and 270.0 ± 13.6 cells/mm2 for 5, 10, 30 and 40 min, respectively), during which staining clearly shifted from extranuclear sites (membrane and/or cytoplasma) to the nucleus (Fig. 2B–D and G). Many labelled cells exhibited processes that were directed towards blood vessels, resembling astrocytic end feet (Fig. 2C, left panel). In addition, labelled albumin was found in the cytoplasm (but not the nucleus) of other cells around blood vessels, probably perivascular cells (Bechmann et al., 2001). To control the specificity of albumin uptake, slices were exposed to either FITC-dextran (70 kDa), which has a molecular weight similar to that of albumin, or to ovalbumin labelled with Texas Red (45 kDa). In the latter experiments, labelling was limited to the cytoplasm of perivascular cells and was never found in parenchymal cells (Fig. 2C right panel, D). To identify the brain cells that take up FITC-albumin, immunohistochemical labelling for astrocytes and neurons was performed using antibodies directed against GFAP and MAP2 as markers, respectively. Confocal analysis of co-labelled cells revealed that most of the FITC-albumin-containing cells expressed GFAP, but none expressed MAP2 (Fig. 2E and F). In addition, a small number of labelled cells were both non-GFAP and non-MAP2 positive. These cells were not characterized in this study: they could be non-GFAP expressing astrocytes, pericytes and/or microglia. To test the nature of the uptake process a competition assay was performed in the presence of increasing amounts of non-labelled albumin in the bathing solution. FITC-albumin uptake was reduced in a dose-dependent manner as expected for a receptor-mediated process (Fig. 2H). Taken together, the selectivity of ligand uptake (albumin but not dextran and ovalbumin), the selectivity of cell-type uptake (astrocytes but not neurons), the sub-cellular localization of labelled albumin and the dose-dependency in the competition assay strongly suggest that the process of albumin uptake into the brain cellular compartments is mediated via a specific receptor.

Serum albumin induces epileptiform activity in vitro

In the in vitro brain slices, albumin uptake by astrocytes was faster and more efficient in comparison to that observed in the in vivo paradigm. If albumin uptake has a role in the development of BBB dysfunction, we would expect that the induction of cortical dysfunction by albumin may also be accelerated in vitro. We tested this hypothesis by continuous recordings of population activity (n = 26) or single neuron responses (n = 5) to albumin wash-in (0.1 mM). Albumin wash-in for 1–3 h resulted in one slice (10%, n = 10) showing abnormal, paroxysmal responses. Exposure to albumin for 4–6 h resulted in robust hypersynchronized, prolonged paroxysmal responses in 15 of 16 slices (94%; Fig. 3). All control slices washed with ACSF for a similar time showed normal field potentials (n = 5). A gradual transformation from normal, brief synaptic responses to epileptiform activity was also observed during continuous (>3 h) intracellular recordings from all five neurons exposed to albumin-containing ACSF (Fig. 3B). The epileptiform activity persisted despite >4 h of wash-out with albumin-free ACSF, pointing to a lasting cortical dysfunction.

Fig. 3

Serum albumin induces epileptogenesis in vitro. Extracellular (A) and intracellular (B) electrophysiological recordings in slices exposed to BSA. Numbers on the left indicate hours of exposure. Paroxysmal activity developed fully 5–6 h after exposure. (C) Bar graph shows the averaged integral (50–500 ms after stimulation) of the evoked responses in 31 slices (see text for details).

TGF-β receptors mediate albumin uptake into astrocytes

The experiments described so far suggested that the rapid transport of albumin is receptor mediated and that this transport is followed by a delayed and robust change in network neuronal responses. TGF-βR type 2 (TGF-βR2) has been recently found to function as an albumin-binding protein in lung endothelial cells (Siddiqui et al., 2004). To test the possibility that TGF-βRs mediate albumin uptake into brain astrocytes, we exposed cortical slices to either the TGF-βR type 1 (TGF-βR1) kinase activity inhibitor, SB431542, and/or to antibodies against TGF-βR2. SB431542 reduced the number of FITC-albumin-labelled-cells in a dose-dependent manner (Fig. 4A–C). Similarly, in the presence of anti-TGF-βR2 antibodies, the number of labelled cells was reduced and the labelled fraction showed mainly membrane staining, with no nuclear staining (Fig. 4B and D). These findings suggest that the transport of albumin into cells is dependent on TGF-βRs.

Fig. 4

TGF-β receptors mediate albumin uptake and epileptogenesis. (A and B) Microscopic sections of brain slices exposed for 30 min to FITC-albumin in the presence or absence of anti-TGF-βR2 antibodies. No nuclear staining is observed in the presence of anti-TGF-βR2 antibodies (see higher magnification in the inset and quantification in D). (C) Number of FITC-albumin labelled cells is reduced by the TGF-βR1 antagonist SB431542 in a dose-dependent manner. (D) Percentage of cells with nuclear FITC-albumin labelling in the absence (control) and presence (+Ab) of anti-TGF-βR2 antibodies. (E) Traces showing epileptiform activity recorded in vitro one week following in vivo exposure to albumin and a brief normal response in slices from a cortex exposed to albumin in the presence of TGF-βR blockers. (F) Bar graph representing percentage of slices showing paroxysmal epileptiform activity in brains treated with albumin in the absence (Alb) and presence (+blockers) of TGF-βR blockers. All recordings were obtained one week following treatment in the presence of ACSF (see text for details).

To probe the role of TGF-βRs in the generation of abnormal electrophysiological responses, we exposed rat cortices in vivo to albumin in the presence and absence of TGF-βR antagonists. Extracellular recordings in vitro one week after treatment revealed paroxysmal activity in 76.3% of the slices from albumin-exposed rats as compared with 29.3% of the slices from rats exposed to both albumin and TGF-βR antagonists (45 of 59 slices, n = 13 animals compared with 12 of 41 slices, n = 6 animals, Pearson χ2 test, P < 0.001, Fig. 4E and F). Thus, the application of TFG-βR antagonists at the time of exposure of the brain environment to serum albumin effectively blocks the consequent generation of epileptiform activity. These findings validate the involvement of TGF-βR-mediated albumin transport in the generation of abnormal brain activity following albumin exposure.

Extracellular buffering of K+ is impaired during epileptogenesis

Previous studies of the injured cortex, the BBB-disrupted cortex and the albumin-exposed cortex all show a window of at least several days before epileptiform activity can be recorded (e.g. Hoffman et al., 1994; Seiffert et al., 2004). Within this period of epileptogenesis an astrocytic reaction is established, as shown by enhanced immunostaining against the GFAP. The pioneer works of Kuffler and Potter (1964) established that astrocytes control the brain's extracellular environment, particularly by buffering rises in [K+]o during neuronal activity. Thus, we tested the hypothesis that during this window period of epileptogenesis, i.e. following treatment but prior to the onset of epileptiform activity, [K+]o buffering is impaired (Pollen and Trachtenberg, 1970; D'Ambrosio et al., 1999). Using ISMEs, we measured changes in [K+]o following neuronal activation 24 h after BBB disruption. In all slices, a low-frequency (20 s interval) stimulation of the white matter showed normal field responses for treated animals similar to that recorded from sham and non-operated control rats, thus excluding epileptiform activity at this early stage (field potential amplitude; stimulation at five times threshold intensity: 1.58 ± 0.29 mV, n = 9, 1.53 ± 0.28 mV, n = 7 and 1.59 ± 0.17 mV, n = 13, for BBB-treated, sham and non-operated rats, respectively, Fig. 5A). During repetitive stimulation (25 Hz, 2 s, one stimulation train), the rate of increase of [K+]o (time to 50% of maximum: 0.71 ± 0.04, 0.67 ± 0.03 and 0.71 ± 0.03 s) and the maximal increase in [K+]o (4.84 ± 0.67, 5.40 ± 1.07 and 4.34 ± 0.59 mM) were similar in treated and control brains. In contrast, the decay in [K+]o was slightly, but significantly, slower in treated than in sham-operated or control slices (decay time to 50% of maximal [K+]o: 1.91 ± 0.1 s, n = 11; 1.51 ± 0.06 s, n = 7; and 1.50 ± 0.07 s, n = 15, respectively, P < 0.01, Fig. 5B). Interestingly, the reduced [K+]o clearance following stimulation returned to control values within four weeks after treatment with DOC (1.91 ± 0.1 s, n = 11, 1.65 ± 0.14 s, n = 2, 1.51 ± 0.05 s, n = 4 for 1, 7 and 30 days, respectively; Fig. 5C).

Fig. 5

Abnormal [K+]o buffering 24 h following treatment is due to a transcriptional downregulation of Kir 4.1 channels. (A) Representative traces showing normal evoked field potentials recorded 24 h after treatment in sham-operated (S) and treated (T) cortices. (B) Two superimposed traces of the [K+]o signals in response to a 2 s, 25 Hz stimulation (marked as underlying bar). [K+]o signals were normalized to maximal increase (100%, 4.34 and 4.84 mM for control and treated, respectively). (C) [K+]o decay time to 50% of its maximal value 1 day (1d), 1 week (1w) and 1 month (1m) after treatment as well as 1 day after sham-operation (s) and in non-operated controls (c) (n = 11, 2, 4, 7 and 15, respectively). (D and E) Representative traces showing the effect of Ba2+ on ionophoretically induced [K+]o increase in treated and control slices before (D) and after (E) addition of 0.1 mM Ba2+. (F) Summary of Ba2+ effect on [K+]o increase. (G and H) Images of GFAP immunostaining in cortical sections 24 h after treatment compared with controls. Insets: Higher magnifications of GFAP (left) and Kir 4.1 (right) immunostaining in consecutive sections from the same brain. Note the enhanced GFAP and reduced Kir immunostaining in morphologically identified astrocytes. Black arrows point to astrocyte processes towards neighbouring vessels. (I) % change in mRNA levels for GFAP and Kir 4.1 in albumin or DOC-treated cortices compared with the contralateral, non-treated hemisphere (n = 5 for each group).

Reduced inward-rectifying K+ currents in the albumin-exposed cortex

The reduced clearance of [K+]o during epileptogenesis implied a down regulation of astrocytic K+ channels. It has been demonstrated that the inwardly rectifying K+ channels play a particularly important role in K+ buffering (Ransom and Sontheimer, 1995; D'Ambrosio et al., 1999). We performed local K+ application by ionophoresis and pharmacological manipulations to further explore the mechanisms underlying reduced [K+]o clearance. To abolish neuronal firing and synaptic responses, thus excluding neuronal contribution to [K+]o changes, these experiments were conducted in the presence of NMDA, AMPA/KA and GABA receptors as well as voltage activated Na+-channel blockers. We applied low (100 μM) and high (2 mM) concentrations of Ba2+ to differentially block inward-rectifying K+ currents (IKIR) and leak K+ currents (IKL), respectively (Ransom and Sontheimer, 1995; Jauch et al., 2002). Before the application of Ba2+ we performed a series of ionophoretic K+ applications to determine a stable baseline of [K+]o increases during injection (amplitude of [K+]o increase during injection: 1.52 mM ± 0.05, n = 41 injections). After wash in of low concentrations of Ba2+, [K+]o increase during ionophoresis was elevated in control brains by 77 ± 15% (n = 5 slices, 5 animals), whereas in brains of treated animals this increase was significantly lower (31 ± 5% increase, n = 5 slices, 5 animals, P < 0.05, Fig. 5D–F). No difference was found when Ba2+ concentrations were elevated to 2 mM to block IKL (68 ± 12% versus 66 ± 5%, control versus treated, respectively, Fig. 5F). These experiments indicate a reduction in IKIR in the presence of normal IKL 24 h following treatment. Six hours of in vitro exposure to serum albumin (Fig. 3) was similarly associated with a reduced effect of 100 μM Ba2+ on ionophoretically induced increases in [K+]o (31 ± 4 and 68 ± 12% in treated and controls, respectively, P < 0.05). Since the Kir 4.1 channel has been shown to be expressed in cortical astrocytes, especially in the processes of astrocytes wrapping synapses and blood vessels (Higashi et al., 2001; Hibino et al., 2004), we performed immunostaining experiments to reveal Kir 4.1 channel levels following treatment with DOC. We found that Kir 4.1 immunolabelling was indeed limited to morphologically identified astrocytes (Fig. 5G, also confirmed by GFAP immunostaining, data not shown) and to blood vessels. Twenty-four hours after treatment, Kir 4.1 channel immunolabelling was markedly reduced, whereas GFAP labelling was enhanced, as expected (Fig. 5G and H; Seiffert et al., 2004). Furthermore, quantitative real-time RT–PCR showed significant higher GFAP and lower Kir 4.1 mRNA levels 14–48 h following in vivo exposure to either DOC or albumin (Fig. 5I, n = 5 for each group). These accumulating results suggest an early transcriptional down regulation of Kir 4.1 channels, yielding a lower level of Kir 4.1 functional protein and resulting in reduced [K+]o buffering.

Activity-dependent [K+]o accumulation and subsequent neuronal hyperexcitability during epileptogenesis

Finally, we studied the effect of reduced [K+]o clearance on neuronal excitability. The observed slowing of [K+]o clearance suggested K+ accumulation during low-frequency stimulation. Indeed, while the increase in [K+]o during the first stimulus was similar for DOC-treated and control brains (0.093 ± 0.013 versus 0.095 ± 0.019 mM, respectively), during the 50th stimulus (0.67 Hz), [K+]o peak levels increased to 315 ± 39% of the first stimulus in treated brains, but only to 193 ± 11% in controls (n = 7, P < 0.05; Fig. 6A). Stimulus-induced [K+]o enhanced accumulation was associated with the appearance of all-or-none, paroxysmal, prolonged negative deflections in the field responses. The latter were associated with a further increase in [K+]o and were blocked by the NMDA-receptor antagonist, MK-801 (Fig. 6B). Intracellular recordings from identified pyramidal neurons confirmed that repetitive stimulation was associated with long depolarization shifts, which upon further membrane depolarization induced action potentials (Fig. 6C–D).

Fig. 6

Activity-dependent K+ accumulation and neuronal hyperexcitability 24 h following BBB disruption. (A) Recording of [K+]o increase during slow (0.67 Hz) repetitive stimulation shows excessive [K+]o accumulation in the treated cortex. (B) Superimposed (1st, 3rd and 5th) field potential responses during repetitive 0.67 Hz stimulation. Note the MK-801 sensitive, late negative deflection of the field potential in the treated slice. (C) Intracellular recording from a single identified pyramidal neuron in a treated slice showing depolarizing after-potentials induced by low-frequency (0.4 Hz) repetitive stimulation (resting potential = −75 mV). (D) Recording from the same cell as in C. Stimulus-induced depolarizing after-potentials led to action potential firing when the resting potential was set to −60 mV (overshooting action potentials are truncated).


In this study, we outlined a novel mechanism underlying epileptogenesis in the BBB-injured cerebral cortex. Our experiments were designed in light of accumulating clinical evidence supporting a causative role between lasting enhanced BBB permeability and epilepsy (Tomkins et al., 2001; Avivi et al., 2004; Korn et al., 2005). We confirmed that both in vivo and in vitro exposure to serum albumin can induce hypersynchronized responses to a single stimulation. The observed paroxysmal events recorded in the BBB/albumin-treated cortex were similar to those described in cortical slices from chronic animal models of epilepsies (e.g. the chronically injured cortex: Prince and Tseng, 1993; Jacobs et al., 1996; chemical kindling: Barkai et al., 1994; or pilocarpine treatment: Sanabria et al., 2002). Similar to the above-mentioned models of neocortical epilepsy, spontaneous interictal-like hypersynchronous activity was only rarely recorded in the BBB-treated cortex in vitro (seeFig. 5 in Seiffert et al., 2004). While EEG recordings are needed to characterize the in vivo correlates for the paroxysmal responses recorded in vitro, the clear spontaneous seizures observed in few of the BBB-treated rats support the relevance of this model in studying epileptogenesis.

Similar to lesional neocortical epilepsy in man and to the above-mentioned models in experimental animals the epileptogenic effect of albumin is delayed. This latent period suggests that the underlying mechanism cannot be explained solely in terms of simple binding to a channel/receptor, but rather that it involves a slower biological process, such as a transcriptional response. In contrast to the in vivo condition, in vitro exposure to albumin showed the development of paroxysmal activity within 4–6 h, perhaps reflecting a more efficient diffusional equilibration or uptake of albumin or additional injury-related processes occurring in the slice preparation. The rapid uptake of albumin in vitro also stresses the in vivo efficiency of the BBB in limiting the incursion of serum proteins (e.g. by the strong uptake capacity of the perivascular cells), even when the endothelial barrier is disrupted (e.g. in the presence of DOC).

On the basis of our findings, we now propose a new mechanism for the uptake of albumin and its effect on astrocytes. It is noteworthy that astrocytes are normally exposed to albumin only during brain development when the BBB is not yet fully developed (at this stage brain astrocytes do indeed display lower IKIR and reduced K+ buffering; Kressin et al., 1995). Hence, the presence of specific albumin uptake in adulthood is surprising but may serve to reduce vasogenic oedema following BBB disruption. The findings of this study—kinetics of albumin entry into astrocytes, the specificity of albumin (labelled dextran or ovalbumin were not transported) and the reduced transport of albumin in the presence of non-labelled albumin—all point to a receptor-mediated uptake. TGF-βR emerged as a possible candidate in light of the following pieces of evidence from previous studies: uptake of albumin is modulated by TGF-βRs in the kidney (Gekle et al., 2003) and lung endothelial cells (Siddiqui et al., 2004); brain astrocytes express TGF-βRs (Vivien et al., 1998); and TGF-βR expression is increased following brain injury (Morganti-Kossmann et al., 2002). The observation in this study that albumin uptake was inhibited by the TGF-βR1 kinase activity inhibitor, SB431542, suggests that the uptake depends on intracellular TGF-βR signalling. It is noteworthy that exposing brain slices to anti-TGF-βR antibodies blocked the uptake of albumin into the cells but did not prevent the surface membrane staining (Fig. 4B), suggesting that albumin binds to another site at the same receptor or to an additional surface receptor. It remains to be further studied whether albumin transport to the nucleus directly triggers altered gene expression and to what extent other intracellular signalling pathways are involved. Previous studies in cultured astrocytes from rat brains show that albumin induces calcium signalling (Nadal et al., 1995) and that TGF-βR activation causes a rapid down regulation of Kir channels (Perillan et al., 2002) further supporting a direct role for the interactions between albumin and TGF-β signalling systems in modulating astrocyte functions. Since TGF-β1 may be secreted in a latent form non-covalently bound to extracellular matrix proteins (Munger et al., 1997), we cannot entirely rule out an alternative hypothesis that albumin exerts its action by increasing the bioavailability of TGF-β1.

We confirmed that the astrocyte reaction (observed as increased GFAP expression) is associated with reduced K+ buffering capacity as early as 24 h after brain exposure to albumin in vivo. Slowing of [K+]o decay was observed both in the presence of apparently normal cortical excitability (to a single stimulation) and during the initial period of abnormal activity, but returned to control values within 4 weeks, despite the continuous presence of abnormal electrophysiological responses. The period of reduced K+ buffering was similar to the period observed in which an increase in the number of GFAP-labelled astrocytes was observed (Seiffert et al., 2004). Astrocytic reaction is prominent in a wide variety of brain insults, in both animals and man, and while it may be important in stabilization of the injured tissue (e.g. scar formation), it seems to interfere with neuronal regeneration (Silver and Miller, 2004). Abnormal K+ buffering in the injured brain has been reported previously (D'Ambrosio et al., 1999; Anderova et al., 2004), but this is the first report showing that abnormal K+ buffering precedes—and may be associated with—the development of epileptiform activity (see below). Furthermore, we showed that while K+ clearance gradually returns to normal values, hyperexcitability is maintained. We also demonstrated that reduced K+ uptake by astrocytes (as we employed sodium channel and synaptic receptor blockers to block neuronal activity; Jauch et al., 2002) is associated with reduced IKIR and not IKL. This conclusion was initially supported by the augmenting effect of low concentrations of Ba2+ on ionophoretically induced K+ signals. Ba2+ also affects a number of voltage- and calcium-dependent K+ channels in neurons. However, since we blocked transmitter receptors and voltage-dependent Na+ channel in these experiments, effects of Ba2+ on neuronal excitability are unlikely. The reduced IKIR in the BBB-treated cortex is consistent with previous studies showing a loss of IKIR in reactive cortical astrocytes in rats around freeze lesions (Bordey et al., 2001), ischaemic insults (Koller et al., 2000) and direct injuries (Schroder et al., 1999) as well as in epileptic Tsc1 knockout mice (Jansen et al., 2005) and human subjects with temporal lobe epilepsy (Bordey and Sontheimer, 1998; Hinterkeuser et al., 2000; Jauch et al., 2002). Interestingly, all these cortical insults are frequently associated with enhanced BBB permeability. Only Kir 4.1 and 5.1 channels have been shown to be expressed in the neocortex (Hibino et al., 2004) and their key role in buffering activity-dependent [K+]o increases is supported by studies showing that their expression is limited to the astrocytic membrane domains facing blood vessels or in the processes surrounding synapses (Higashi et al., 2001). Our immunolabelling experiments and the determination of mRNA levels suggest a rapid downregulation of Kir channel expression together with the upregulation of GFAP. At this point, however, we cannot rule out additional changes in Kir channels' rectification properties (Bordey et al., 2001) and/or their re-distribution in the cell membrane (Warth et al., 2005). A downregulation of Kir channels will not only affect potassium buffering but also lead to depolarization of astrocytes and thereby reduce the efficacy of glutamate transport into astrocytes, thus contributing to the facilitated emergence of epileptiform discharges.

We showed here that even a relatively small reduction in K+-buffering capacity may be functionally significant, since it augments activity-dependent K+ accumulation and consequent NMDA receptor activation (Fig. 6). The activation of NMDA receptors may be due to K+ accumulation at the synaptic cleft and consequent depolarization at the post-synaptic site (thus increasing the likelihood of NMDA receptor opening) and/or by reduced glutamate transport into depolarized astrocytes. A plausible hypothesis would be that the increased, repeated activation of NMDA receptors leads to non-specific synaptic plasticity, thus strengthening excitatory synapses and causing hyperexcitability (Li and Prince, 2002; Shao and Dudek, 2004). This premise would also explain the efficacy of NMDA-receptor antagonists in improving cortical functions after brain injury and in some neurodegenerative disorders—all conditions in which the BBB is frequently impaired (Hickenbottom and Grotta, 1998; Sonkusare et al., 2005).

In summary, we conclude that following brain insults, exposure of brain cells to albumin—the most abundant serum protein—leads to cortical dysfunction, recorded as epileptiform hypersynchronous activity. We suggest that the development of cortical dysfunction is mediated by TGF-βRs, which facilitate albumin uptake into astrocytes and down regulation of Kir currents. This, in turn, causes abnormal accumulation of [K+]o and consequent NMDA-receptor-dependent pathological plasticity. Since a wide spectrum of common neurological disorders is associated with BBB disruption, we propose that amelioration of neural injury in these conditions may be achieved via targeting the TGF-βRs.


The authors thank K. Froehlich, J. Mahlo and H. Levy for technical assistance. This study was supported by the Sonderforschungsbereich 507 and TR3 (AF and UH), the German-Israeli Foundation for Scientific Research and Development (AF) and the Mary Elizabeth Rennie Epilepsy Foundation research grant (DK). Funding to pay the Open Access publication charges for this article was provided by the Sonderforschungsbereich 507.

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/2.0/uk/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.


View Abstract