Research ArticleBLOOD-BRAIN BARRIER

Paroxysmal slow cortical activity in Alzheimer’s disease and epilepsy is associated with blood-brain barrier dysfunction

See allHide authors and affiliations

Science Translational Medicine  04 Dec 2019:
Vol. 11, Issue 521, eaaw8954
DOI: 10.1126/scitranslmed.aaw8954

Breaking barriers in aging and disease

The blood-brain barrier (BBB) regulates the communication between the vasculature and the brain. Aging and neurological disorders have been associated with BBB defects. Now, Milikovsky et al. and Senatorov et al. studied the consequences of BBB impairments in aging and disease. Milikovsky et al. found that in patients with epilepsy or Alzheimer’s disease, as well as in aging mice, BBB impairments were spatially associated with transient electroencephalographic abnormalities. Senatorov et al. extended the study at the molecular level, showing that BBB breakdown triggered transforming growth factor–β (TGFβ) signaling in astrocytes and cognitive impairments in aging rodents. Similar abnormalities were also found in brain tissue from aging individuals. TGFβ inhibition in aged mice reversed the pathological phenotype.

Abstract

A growing body of evidence shows that epileptic activity is frequent but often undiagnosed in patients with Alzheimer’s disease (AD) and has major therapeutic implications. Here, we analyzed electroencephalogram (EEG) data from patients with AD and found an EEG signature of transient slowing of the cortical network that we termed paroxysmal slow wave events (PSWEs). The occurrence per minute of the PSWEs was correlated with level of cognitive impairment. Interictal (between seizures) PSWEs were also found in patients with epilepsy, localized to cortical regions displaying blood-brain barrier (BBB) dysfunction, and in three rodent models with BBB pathology: aged mice, young 5x familial AD model, and status epilepticus–induced epilepsy in young rats. To investigate the potential causative role of BBB dysfunction in network modifications underlying PSWEs, we infused the serum protein albumin directly into the cerebral ventricles of naïve young rats. Infusion of albumin, but not artificial cerebrospinal fluid control, resulted in high incidence of PSWEs. Our results identify PSWEs as an EEG manifestation of nonconvulsive seizures in patients with AD and suggest BBB pathology as an underlying mechanism and as a promising therapeutic target.

INTRODUCTION

Alzheimer’s disease (AD) prevalence is constantly increasing as the world population ages. It is estimated that in 2050, about 80 million will suffer from the disease worldwide (1). Understanding the mechanisms underlying AD is crucial for the development of early diagnostics and treatment.

Recent data from patients with AD and animal models imply a potential role for microvasculopathy, specifically for blood-brain barrier dysfunction (BBBd), in AD and neurodegeneration (26). In addition, a growing body of evidence shows a substantial comorbidity of undiagnosed epilepsy among patients with AD (7, 8).

BBBd was shown to be a key component in the pathogenesis of epilepsy (epileptogenesis) in many studies (915) with a specific role attributed to the extravasation of serum albumin into the brain neuropil (12, 13, 16). In animal studies, induction of BBBd induces astroglial dysfunction, neuroinflammation, alterations in the extracellular matrix (17), excitatory synaptogenesis (16), pathological plasticity (18), and increase in neural excitability (10, 11, 16, 1921), which reduce seizure threshold and precede neurodegeneration (13, 22). We therefore hypothesized that BBBd that accompanies AD would cause epileptiform-like network changes that would be detected in electroencephalogram (EEG) recordings from patients and may constitute an independent mechanism of neural dysfunction and degeneration as suggested in the two-hit vascular hypothesis (4).

Slowing of EEG activity has been observed in AD and other types of dementia (2328) and may constitute a potential sign of underlying dysfunction in neural networks. However, there has not been a detailed investigation of the temporal characteristics of this generalized slowing, nor has there been an understanding of the biological mechanisms that are underlying it.

In this study, we analyzed EEG recordings from patients with AD and found that cortical slowing is composed of transient paroxysmal slowing of the cortical network [paroxysmal slow wave events (PSWEs)]. We then further searched for PSWEs in young patients with epilepsy and for their spatial correlation with BBBd. Using animal models, we investigated PSWE characteristics in aged mice, young 5x familial AD (5xFAD) mice, and young rats with epilepsy. Hypothesizing a causative link between BBBd and PSWE, we infused albumin into the lateral ventricle of rats to bypass the BBB. After a month of continuous infusion, PSWEs were detected in epidural recordings. Our results suggest PSWEs as an indicator for nonconvulsive seizure–like activity that reflects microvascular pathology and hence highlight a potential therapeutic target in disorders of the brain.

RESULTS

Slow network activity in AD is composed of distinct PSWEs

To explore changes in the function of brain networks, we analyzed EEG recordings from patients attending the Cognitive Neurology Clinic at Rabin Medical Center, Beilinson Hospital, Petach Tikva, Israel, and diagnosed with AD (n = 16; mean age, 72.05 ± 9.707 years) or mild cognitive impairment (MCI) (n = 12; mean age, 73.37 ± 4.765 years) compared with age-matched controls (n = 11; mean age, 76.73 ± 10.06 years; see further clinical details in table S1). Spectral analyses revealed significantly lower α relative power (8 to 12 Hz) among patients with AD compared with MCI (P = 0.003) or controls (P = 0.006). Consistent with previous reports (24, 25, 28), patients with AD showed significantly greater θ relative power (5 to 8 Hz) compared to age-matched controls (P = 0.0147; Fig. 1, A and B). By studying the temporal characteristics of EEG slowing, we found that it is composed of transient events, in which the network switches from apparently normal activity to brief periods of low-frequency activity, which we termed PSWEs (Fig. 1C). To further characterize PSWEs, we calculated the median power frequency (MPF) in 2-s-long signal epochs with 1-s overlap and repeated the detection of PSWEs in a variety of upper bounds (2 to 8 Hz with 2-Hz interval between iterations). For each iteration, we used a blind Gaussian mixture model (GMM) algorithm to cluster PSWE frequency (per minute) into high- and low-occurrence groups (hPSWE and lPSWE, respectively). ROC (receiver operating characteristic) analysis revealed that an MPF of 6 Hz is the frequency that best separates patients with AD and age-matched controls (table S2). Thus, an “event” was classified as a PSWE if the MPF is lower than 6 Hz for five consecutive seconds or more (Fig. 1D). The occurrence per minute of PSWEs was significantly higher in AD compared with age-matched controls (P = 0.001) and with patients with MCI (P = 0.0015; Fig. 1D and table S1). Nine of the n = 16 patients with AD were classified into the hPSWE group, whereas only 1 of the 12 patients with MCI and 1 of the 11 controls were classified into the hPSWE group. Thus, the prevalence of AD in the hPSWE group (90%) was significantly higher compared to their prevalence in lPSWE groups (41%; P = 0.0159; Fig. 1E). In addition, mini-mental state examination (MMSE) scores among the hPSWE group was lower compared to the lPSWE group (P = 0.0034; Fig. 1F), and MMSE score was significantly and inversely correlated with PSWE occurrence per minute (P = 0.006, Pearson’s correlation; Fig. 1G), suggesting that PSWEs might reflect abnormal brain function, associated with cognitive impairment. Further characterizing the PSWEs, we found that within the hPSWE group, they were longer in duration compared to those in the lPSWE group (P < 0.0001; Fig. 1H). Heat scalp mapping shows that PSWEs are recorded in most scalp electrodes (Fig. 1I). Last, ROC analyses revealed an area under the curve (AUC) of 0.86 (P = 0.0015; Fig. 1J) for patients with AD and age-matched controls and an AUC of 0.84 (P = 0.002; Fig. 1K) for AD and MCI.

Fig. 1 Dementia-related slow network activity is rather continuous but composed of distinct PSWEs.

(A) Spectral analyses of patients with AD, those with MCI, and age-matched controls. (B) Area under the curve of spectral analyses of δ (1 to 5 Hz), θ (5 to 8 Hz), α (8 to 12 Hz), and β (12 to 20 Hz) relative power bands in patients with AD, those with MCI, and age-matched controls. *P = 0.0147; **dementia vs. aged matched control P = 0.006; **dementia vs. MCI P = 0.003. (C) A PSWE detected in a patient with AD. Shown are traces from electrodes P3 (upper trace) and P4 (averaged as reference). The segment within the dashed rectangle of P3 is shown magnified. The median power frequency is presented below each trace. Segments below 6 Hz (dashed line) are marked in red. (D) Gaussian mixture model algorithm classified values of PSWE occurrence per minute from each patient into two groups, low PSWE and high PSWE (lPSWE and hPSWE, respectively). PDF, probability density function. (E) Prevalence of patients with AD or MCI in the lPSWE and hPSWE groups. *P = 0.0159. (F) MMSE scores of patients in lPSWE and hPSWE groups. (G) Correlation of MMSE and PSWE per minute. **P = 0.0034. (H) Distribution of PSWE duration in the control and lPSWE and hPSWE groups. (I) Mean heat maps of PSWE per minute in each electrode for control and lPSWE and hPSWE groups. ****P < 0.0001 for hPSWE vs. lPSWE or hPSWE vs aged matched control. (J and K) Receiver operating curve (ROC) for PSWE per minute as an indicator of AD when compared with age-matched controls or patients with MCI.

PSWEs overlap with BBBd in patients with epilepsy

Comorbidity of AD and epilepsy is attracting increased attention due to the promise of antiepileptic therapy for some patients with AD (7, 8). Because PSWEs are paroxysmal and transient changes in network activity, similar to the characteristics of epileptic seizures, we next analyzed EEG from patients with epilepsy (n = 17; mean age, 33.18 years; SD, 16.25; tables S3 and S4) in comparison with age-matched controls (n = 9; mean ag, 40.67 years; SD, 14.09) and found PSWEs more frequent in patients with epilepsy (P = 0.0052; Fig. 2A). When grouped to hPSWE (>mean controls + 2 SDs) and lPSWE groups, we found that patients with epilepsy comprise 90% of the hPSWE group (9 of 10) and 47% of the lPSWE group (8 of 17) (P = 0.0257; Fig. 2B). PSWE occurrence among the present cohort of patients with epilepsy was not age dependent (Pearson’s r = −0.28, P = 0.27; fig. S1). To challenge the hypothesis that PSWEs are associated with BBBd regardless of other age-related effects that can be found in the patients with AD, we implemented DCE-MRI (dynamic contrast-enhanced magnetic resonance imaging) to test BBBd in patients with epilepsy as previously described (3032). In 12 patients with epilepsy (mean age, 28.33; SD, 13.86) and 59 age-matched controls (mean age, 28.44 years; SD, 4.018; tables S4 and S5), BBBd was significantly greater among patients with epilepsy compared with controls (P = 0.0056; Fig. 2C). When extent of BBBd (percent brain voxels with pathological permeability) was grouped into low (LBBBd) and high (HBBBd) groups (below and above mean plus 2 SD of control values), we found that the prevalence of patients with epilepsy among the HBBBd group was significantly greater compared with the LBBBd group [50% (4 of 8) versus 12.5% (8 of 64), respectively; P = 0.0073; Fig. 2D]. ROC analyses revealed an AUC of 0.83 for PSWEs (P = 0.0065; Fig. 2E) and an AUC of 0.77 for BBBd (P = 0.001; Fig. 2F). In a subset of patients with epilepsy (n = 10), both DCE-MRI and EEG were conducted. The percentage of brain volume with BBBd was calculated for eight cortical lobes (left and right frontal, temporal, parietal, and occipital) and correlated with PSWE occurrence per minute as detected in electrodes corresponding to the same lobe. In 7 of 10 patients, positive Spearman’s r values were found (Fig. 2, G to I). The Spearman correlation was statistically significant for three of the patients (r = 0.93, P = 0.0022; r = 0.76, P = 0.037; r = 0.738, P = 0.0458; table S6).

Fig. 2 PSWEs overlap with BBBd in patients with epilepsy.

(A) PSWE occurrence per minute among patients with epilepsy compared with age-matched controls. **P = 0.0052. (B) Prevalence of epilepsy diagnosis among the hPSWE group compared with the lPSWE group. *P = 0.0257. (C) BBBd among patients with epilepsy compared with age-matched controls. **P = 0.0056. (D) Prevalence of epilepsy diagnosis among the high-BBBd (HBBBd) group compared with the low-BBBd (LBBBd) group. **P = 0.0073. (E) Receiver operating curve (ROC) for PSWE as an indicator of epilepsy. (F) ROC for BBBd as an indicator of epilepsy. (G) Correlation of BBBd value by lobe with respect to PSWE occurrence per minute within the same lobe. (H and I) A permeability map of BBBd (radiological view) and PSWE per minute heat map (radiological view) of the same patient, respectively. Voxels with high relative permeability are color-coded on T1 image (see Materials and Methods). Rt., right; Lt., left.

PSWEs and BBBd characterize aged mice

We next compared brain activity recorded from aged mice (n = 20, 18 to 22 months old), a natural model of age-related cognitive decline that shows BBBd and cognitive impairment (33) to that of young animals (n = 5, 12 weeks of age). Similar to the findings in humans, we detected PSWEs, and in a similar process to that mentioned above, we found that a frequency threshold of 5 Hz and a minimal duration of 10 s yield optimal separation between the groups. When applying a GMM, two groups were obtained: the hPSWE group composed of n = 8 old animals and the lPSWE group composed of n = 5 young animals and n = 12 old animals (Fig. 3A and table S7). Spectral analysis showed greater δ relative power (P = 0.0003) and lower α relative power (P = 0.0015) in recordings from old hPSWE compared with old lPSWE. Young animals had lower δ relative power (P = 0.045), greater α relative power (P = 0.0451), and greater β relative power (P = 0.0295) compared with hPSWE mice. No differences were found between young and old lPSWE (Fig. 3B). Spectral analysis after excluding PSWEs from the signal showed that the differences in α relative power were diminished, δ relative power was greater among hPSWE compared with lPSWE (P = 0.0041), and β relative power was lower when hPSWE compared with lPSWE mice (P = 0.0251; Fig. 3C). In line with the human findings in patients with AD, we found that PSWEs in the old hPSWE group are longer compared with PSWEs among young (P < 0.0001) or old lPSWE groups (P < 0.0001; Fig. 3D). Median frequency of PSWEs was lower among the old hPSWE group compared with the old lPSWE or young groups (P < 0.0001, Kolmogorov-Smirnov; Fig. 3E). Because slow activity is known to occur during deep sleep (34), we examined PSWE occurrence throughout a 24-hour cycle. The number of PSWEs among the old hPSWE group was higher regardless the time of the day (P < 0.002 for each hour, multiple t tests with Holm-Sidak multiple comparisons correction; Fig. 3F). Analysis of locomotor activity (21) confirmed that aged animals were moving for more than 75% of the time during PSWEs, suggesting that these are not sleep-related episodes (fig. S2). Last, recording from young 5xFAD mice (n = 6, 9 weeks of age), a well-known model for AD, shown to have BBBd (35), revealed higher (P = 0.007) occurrence of PSWEs compared to age-matched controls [n = 8, 8 to 11 weeks of age mice; of them, n = 4 animals were treated with artificial cerebrospinal fluid (ACSF), and n = 4 were treated with dextran dissolved in ACSF]. No difference was seen between the two control groups (P = 0.37; Fig. 3G and table S8)].

Fig. 3 Slowing of network activity in aged animals is composed of PSWE.

(A) After automatic detection of PSWEs in ECOG recordings from young and old mice, animals were blindly classified by Gaussian mixture model into two groups according to the PSWE occurrence per minute—hPSWE and lPSWE. (B) Spectral analysis of darkness time (3 cycles mean) in young, old lPSWE, and old hPSWE. The subplate shows area under the curve of the spectral analyses of δ (1 to 5 Hz) ***P = 0.0003, θ (5 to 8 Hz), α (8 to 12 Hz) **P = 0.0015, and β (12 to 20 Hz) *P = 0.0124 relative power bands in young, old lPSWE, and old hPSWE. (C) Spectral analysis was recalculated after PSWEs were excluded from the signal. The subplate shows area under the curve of the spectral analyses of δ (1 to 5 Hz) **P = 0.0041, θ (5 to 8 Hz), α (8 to 12 Hz), and β (12 to 20 Hz) **P = 0.0251 relative power bands in young, old lPSWE, and old hPSWE. (D) Distribution of PSWE duration in young, old lPSWE, and old hPSWE mice. (E) Distribution of PSWE median frequency in young, old lPSWE, and old hPSWE mice. (F) PSWE occurrence per hour throughout the night among young, old lPSWE, and old hPSWE mice. Gray background represents darkness hours. (G) PSWE occurrence per minute among young 5xFAD mice compared with controls. **P = 0.007.

PSWE and BBBd are prevalent in a status epilepticus model of epilepsy

We next tested contrast-enhanced MRI scans in young animals with status epilepticus (SE)–induced epilepsy. Consistent with our observation in patients with epilepsy, we found greater BBBd in animals with epilepsy compared with naïve animals (P = 0.0015; Fig. 4, A to C, and table S9) (14, 36, 37). Sticking to our hypothesis, we analyzed electrocorticogram (ECOG) data recorded from rats with SE-induced epilepsy and found PSWEs to be more frequent (P = 0.055; table S10) and to have greater duration (P = 0.02) with a lower median frequency (P = 0.03) compared with naïve animals (Fig. 4, D to F). ECOG spectral analysis revealed relative power greater in δ (1 to 5 Hz; P = 0.0159) and lower in α (8 to 12 Hz; P = 0.0159) and β (12 to 20 Hz; P = 0.0159) among rats with epilepsy compared with naïve animals (Fig. 4, G and H). Last, we compared frequency and duration of PSWEs with that of identified convulsive epileptic seizures [detected automatically as previously described (12, 16, 20, 21)]. Spontaneous seizures (n = 18; median, 7 s; min, 6 s; max, 24 s) were significantly shorter (P = 0.0007) than PSWEs (n = 4885; median, 12 s; min, 10; max, 64 s; Fig. 4I) and with a higher median frequency (P < 0.0001; spontaneous seizures: median, 7.813 Hz; min, 4.639 Hz; max, 13.18 Hz; PSWE: median, 3.418 Hz; min, 1.953 Hz; max, 4.883 Hz; Fig. 4J) compared with PSWEs.

Fig. 4 PSWE and BBBd are prevalent in a status epilepticus model of epilepsy.

(A) BBBd among epileptic rats compared with control. **P = 0.0015. (B and C) MRI scans of control and epileptic animals, respectively. Voxels with high permeability values are color-coded (see Materials and Methods). (D) PSWE occurrence per minute (averaged in five recording days) in animals with epilepsy compared with naïve animals. (E) Distribution of PSWE duration in naïve animals and animals with epilepsy. (F) Distribution of PSWE median frequency in naïve animals and animals with epilepsy. (G) Spectral analysis of darkness time ECOG (5 cycles mean) in animals with epilepsy compared with naïve animals. (H) Area under the curve of δ (1 to 5 Hz) *P = 0.0159, θ (5 to 8 Hz), α (8 to 12 Hz) *P = 0.0159, and β (12 to 20 Hz) *P = 0.0317 relative power bands in animals with epilepsy compared with naïve animals. (I) Comparison of seizures and PSWE duration recorded from the same rats with epilepsy. Plot shows median, min, and max values. ***P = 0.0007. (J) Comparison of seizures and PSWE median frequency recorded from the same rats with epilepsy. Plot shows median, min, and max values. ****P < 0.0001.

Brain exposure to serum albumin induces PSWE network activity

Because BBBd has been suggested to have a role in the pathogenesis of both AD (26) and epilepsy (12, 13, 16), we tested the causal role of BBBd and, specifically, the serum protein albumin in network dysfunction. We used osmotic pumps in rats to inject albumin into the right lateral ventricle intracerebroventricular (ICV) for 28 days and performed immunohistochemistry to confirm the accumulation of albumin in brain astrocytes (Fig. 5, A and B), similar to that observed in old mice (33). We next recorded ECOG from young rats (10 to 12 weeks of age) exposed to ICV albumin (n = 8) compared to animals exposed to ACSF (controls, n = 5). Four weeks after onset of perfusion (but not at 1 week), albumin-treated rats showed significantly greater δ relative power (1 to 5 Hz) band compared with ACSF-injected animals (P = 0.0016) and significantly lower α relative power (8 to 12 Hz, Mann-Whitney (MW); P = 0.0016) and β relative power (12 to 20 Hz, MW; P = 0.0031) bands (Fig. 5C) with a decrease in level of significance when detected PSWEs were excluded from the signal (Fig. 5D). The analysis revealed significantly higher occurrence of PSWEs in the albumin-injected hemisphere compared with the noninjected hemisphere (P = 0.0078, Wilcoxon) or ACSF-injected hemisphere in controls (P = 0.0070, Mann-Whitney; Fig. 5E and table S11). Consistently with the findings in patients and animal models, PSWEs in the albumin-treated group were longer (P < 0.0001; Fig. 5F) and slower (P < 0.0001; Fig. 5G) compared with the ACSF-treated group.

Fig. 5 Long-term albumin perfusion may serve as a model of chronic BBB disruption and induces PSWE comprised network activity.

(A and B) Staining of tissue taken from a rat intraventricularly treated with albumin for 28 days. (A) Untreated hemisphere. (B) Treated hemisphere. Scale bar, 50 μm. GFAP, glial fibrillary acidic protein. (C) Spectral analyses of darkness time ECOG recordings (3 cycles recording mean) from albumin-treated animals compared with ACSF-treated animals. The subplate shows area under the curve of the spectral analyses of δ (1 to 5 Hz) **P = 0.0016, θ (5 to 8 Hz), α (8 to 12 Hz) **P = 0.0016, and β (12 to 20 Hz) **P = 0.0031 relative power bands. (D) Spectral analysis was recalculated after PSWEs were excluded from the signal. The subplate shows area under the curve of the spectral analyses of δ (1 to 5 Hz) **P = 0.0016, θ (5 to 8 Hz), α (8 to 12 Hz) **P = 0.0031, and β (12 to 20 Hz) **P = 0.0031 relative power bands. (E) PSWE occurrence per minute in the ipsilateral hemisphere compared with the contralateral hemisphere of the albumin (alb)–treated animals **P = 0.0078 and compared to the ipsi- **P = 0.007 and contralateral hemispheres of ACSF-treated controls. ns, not significant. (F) Distribution of PSWE duration in the treated hemisphere of albumin-treated and ACSF-treated animals. (G) Distribution of PSWE median frequency in the treated hemisphere of albumin-treated and ACSF-treated animals.

DISCUSSION

We show that slowing in cortical network activity observed in EEG recordings from patients with AD is composed of well-defined, discrete PSWEs that can be automatically detected and quantified. Whereas PSWEs were recorded bilaterally in patients with AD, PSWEs were most often focal and colocalized with BBBd in patients with epilepsy. ECOG recordings from three animal models with diffuse (aging or 5xFAD) (35) and focal (epilepsy) BBBd confirm the occurrence of PSWEs. Last, we found that PSWEs are detected after infusion of albumin into the brain ventricles, suggesting that PSWEs represent a BBBd-related network dysfunction.

When quantitatively investigating EEG slowing in AD, we found that it is composed of discrete PSWEs. Similar results were obtained in patients with epilepsy and four animal models—aging, 5xFAD, epilepsy, and brain exposure to albumin. PSWEs were detected in scalp electrodes from humans and in epidural electrodes from implanted animals and were similar in occurrence, frequency, and duration, suggesting that these discrete events represent a transient pathological transformation of activity within cortical networks, which we show to underlie some of the differences in slowing found between the groups. In epilepsy models, BBBd and the induction of proinflammatory transforming growth factor–β (TGFβ) signaling by serum albumin were shown to underlie changes in neural circuits that induce epileptogenesis (1013, 15, 19, 37, 38). The colocalization of PSWEs with BBBd in patients with epilepsy and that BBBd and PSWEs are both present in epileptic animals suggest that these slow, transient events represent interictal pathology within (or in vicinity to) the epileptogenic zone.

BBB pathology has also been suggested as an early pathogenic event contributing to neurodegeneration in AD and MCI (2, 3, 6). Therefore, we postulated that there is a common, shared mechanism in which BBBd causes neural dysfunction that underlies the occurrence of PSWEs. Finding PSWEs after prolonged application of ICV albumin (preferentially in the treated hemisphere) that recapitulated similar features to those found in patients and rodents further supports the hypothesis that the extravasation of albumin into the brain neuropil is sufficient to induce neural dysfunction and that PSWEs are likely a reliable indicator for such comprised network. It is not yet known if BBBd is both sufficient and necessary for these network changes to occur, partly because there is no experimental injury model that does not involve BBBd. In the accompanying paper (33), the authors demonstrate that blocking TGFβ signaling by IPW improves cognitive function and reduces PSWE occurrence in aged mice, further supporting the notion that TGFβ signaling underlies cortical dysfunction characterized by PSWEs.

The potential role of BBBd in network dysfunction and PSWEs as a reliable sign that can be reversed with targeted treatment highlights therapeutic strategies for related conditions, including AD (3) and epilepsy (39). Blocking TGFβ signaling by the Food and Drug Administration–approved angiotensin receptor blocker losartan was shown to attenuate SE-induced BBBd (36) and to ameliorate epileptogenesis in models of BBBd, electrical kindling, and SE (12, 16, 40, 41). The potential benefit of losartan treatment in AD was observed in both animal models and humans (42, 43). Our finding suggests both a mechanism underlying these observations and a therapeutic and pharmacodynamic indicator to identify subgroups of patients with inflammation-induced network dysfunction, as well as follow-up treatment success.

Epilepsy is common among patients with AD (7, 8, 44, 45). Increase in synaptic release probability induced by amyloid-β was offered as a mechanism underlying hyperexcitability and seizures (46, 47). The silent nature of seizures in AD leads to multiple misdiagnosed patients (7, 8). The paroxysmal and discrete properties of PSWEs in AD and epilepsy suggest that it may be related to this enigmatic comorbidity. Seizures are rare paroxysmal changes in network activity, often localized to deep brain structures, and unlikely to be detected in routine EEG examination (48). In contrast, PSWEs are frequent interictal events and may reflect a private case of subclinical seizures. The diffuse nature of PSWEs in AD suggests generalization, another common characteristic of seizures in the elderly (49). Convulsive epileptic seizures seem to show a different pattern, are more likely to present with faster activity, and are shorter compared with PSWEs. Hence, we suggest that PSWE detection in routine EEG recordings may serve as an affordable and sensitive tool to diagnose subclinical seizures among patients with AD. The potential benefit of antiepileptic drugs to patients with AD has been suggested (7, 8, 50), supported by the observation that levetiracetam, a common antiepileptic drug, attenuates slow activity recorded from patients with AD (51).

In human, foci of BBBd were shown to colocalize with epileptiform slow EEG activity in patients diagnosed with postconcussion syndrome (52). Our results show that the focal occurrence of PSWEs among patients with epilepsy is colocalized with the corresponding BBBd, further supporting the association between these paroxysmal events and BBBd. In contrast, PSWEs in AD were recorded diffusely in most scalp electrodes. Although little is known on BBBd localization in AD, hippocampal BBBd has been reported in patients with MCI (3). Future studies are required to study BBBd localization in relation with source of PSWEs during aging and cognitive decline. Diffuse cortical dysfunction may be due to local changes in BBB integrity within the neocortex or remote changes in deeper networks such as the thalamus and basal forebrain that drive and modulate cortical activity (53, 54).

We show that albumin is sufficient to induce a PSWE comprised network (Fig. 5). Brain exposure to other serum-born proteins such as fibrinogen was shown to be associated with neuroinflammatory and neurodegenerative processes (5557), and we cannot exclude the idea that these may induce similar network transformations. To test whether any large molecule may cause such changes, we used two control groups in the 5xFAD experiment. ACSF or 70-kDa dextran dissolved in ACSF. We found no difference between the subgroups.

Slow waves are a hallmark of deep sleep (34), and therefore, some of the PSWEs we detect in animals could be sleep-related. However, locomotion analysis in recorded animals confirmed that, in most cases, PSWEs were recorded while the animals were moving. Because “zero movement” includes activities such as grooming, eating, staring, or freezing (animal in constant location), we argue that only a small portion of the detected PSWEs are actual sleep.

This study holds a diagnostic and therapeutic promise; yet, it is limited by its retrospective nature, relatively small number of participants, and the absence of DCE-MRI scans and EEG recordings in the same patients with AD. Future prospective trials are required to determine the sensitivity and specificity of this diagnostic measure. Combining EEG recordings with DCE-MRI, or other quantitative BBBd measure, is necessary to solidify the spatiotemporal relations between PSWEs and vascular pathology.

In summary, we propose that transient shifts in activity of the cortical network can be identified and quantitatively assessed using scalp EEG. These cortical events, termed PSWE, reflect BBBd and suggest a microvascular contribution to the pathology of epilepsy and AD in subgroups of patients. PSWEs may serve as an affordable diagnostic indicator for brain diseases and as a pharmacodynamic measure for BBB-targeted therapeutics or antiepileptic drugs.

MATERIALS AND METHODS

Study design

The aim of the presented study was to test the hypothesis that BBBd is associated with functional network changes that can be detected in scalp EEG. For that purpose, retrospective EEG and clinical data of patients with AD (n = 16), MCI (n = 12), and epilepsy (n = 17) or control subjects (n = 20) were collected. In addition, DCE-MRI scans were analyzed in patients with epilepsy (n = 12) and controls (n = 59). All procedures were approved by the ethics committees of Rabin, Wolfson, or Soroka Medical Centers as detailed below. Exclusion and inclusion criteria are detailed below. Sample sizes were derived from data availability. Outcomes were assessed by correlation of occurrence per minute of PSWEs and their electrode locations, with the amount of BBBd according to DCE-MRI analysis and with level of cognitive impairment of patients, namely, MMSE scores. To further test the hypothesis in animal models, ECOG signals were recorded in three BBBd models and their corresponding controls: old mice (n = 20) versus young mice (n = 5), young 5xFAD mice (n = 6) versus young mice treated with ACSF (n = 4) or dextran (n = 4), young rats with SE-induced epilepsy (n = 5), and young naïve rats (n = 3). Measures of quantitative ECOG were compared between the groups. Last, to challenge a causal link between BBBd and PSWEs, we simulated BBBd by infusion of albumin or ACSF (controls) into the brain ventricles of young rats (n = 8 and n = 5, respectively). Animals were randomly assigned to experimental groups. Researchers who analyzed the data were blind to patient or animal identity. Raw data are reported in data file S1.

EEG recordings from patients with AD and MCI

Routine EEG of patients referred to memory clinic and that of a control group were retrospectively analyzed. Recordings for 30 min were performed using the Nihon Kohden Neurofax-1200 with a standard 10-20 electrode system (19 electrodes; locations are detailed in table S10) in an awake state with open and closed eyes as well as photostimulation and a phase of hyperventilation. All EEGs were performed at the Rabin Medical Center, Petach Tikva, Israel. Subjects with MCI and dementia are being recruited and classified in the Dementia Clinic at Rabin Medical Center (A.G.). In the MCI group, only those patients meeting the criteria for MCI according to the updated guidelines from the National Institute on Aging and Alzheimer’s Association workgroup were included. Briefly, the diagnosis of MCI requires that (i) disease onset is insidious, (ii) there is impairment in one or more cognitive domains without an overt functional impairment, and (iii) the subject does not meet dementia criteria. Patients classified as suffering from AD were included if possible or probable, sporadic AD was present in accordance with the 2011 National Institute on Aging (NIA) guidelines. Patients with additional brain diseases were excluded.

EEG recordings from patients with epilepsy

EEG was recorded from patients with epilepsy in three medical centers: (i) Rabin Medical Center, n = 7 patients, n = 9 age-matched controls, standard EEG recordings in the international 10-20 system (19 electrodes), Nihon Kohden Neurofax-1200 system, sampling rate of 256 Hz. (ii) Wolfson Medical Center, n = 6 patients, standard EEG recordings in the international 10-20 system (19 electrodes), Micromed, sampling rate of 256 Hz. (iii) Soroka Medical Center, n = 4 patients, 61 electrodes cap with Ag/AgCl ring electrodes, Micromed, sampling rate of 1024 Hz. EEG was recorded for 20 to 30 min in an awake state with open and closed eyes as well. (Electrode coordinates are detailed in tables S12 and S13). Patients were randomly selected from the clinical recording pool of epilepsy outpatient clinics. None of the patients was seizure free under current medications, and none reported a seizure in the 48 hours before recordings. Controls are patients who were recorded for reasons other than epilepsy or dementia such as falls or headache. Patients or controls with additional brain diseases were excluded. Analyzers were blind to patients/subject’s clinical details.

Telemetric ECOG recording in rodents

All animal experiments were conducted after an ethical committee approval of the institution in which they were performed, Ben Gurion University of the Negev, Beer-Sheva, Israel, or the University of California, Berkeley, Berkeley, CA, USA. Electrodes and wireless transmitters were implanted as previously described (16, 21) in the following animal groups: group 1, 12-week-old (n = 5) mice; group 2, 18- to 22-month-old (n = 12) mice; group 3, 18- to 22-month-old saline ICV-treated (n = 8) mice; group 4, 9- to 11-week-old Wistar male rats implanted with a 28-day-long ICV perfusion osmotic pump [n = 8; 0.8 mM bovine serum albumin (BSA; Sigma-Aldrich) infusion, n = 5 ACSF (10) infusion]; group 5, Sprague-Dawley rats (n = 5; weighing 300 g) 3 weeks after SE (see below) and naïve rats (n = 3); group 6, 8- to 11-week-old (n = 4) 1-week-long ICV 70-kDa dextran (0.4 mM; Sigma-Aldrich)–treated mice; group 7, 8- to 11-week-old (n = 4) 1-week-long ICV ACSF-treated mice; and group 8, 9-week-old 5xFAD model (58) (n = 6) mice. Groups 1 to 3 electrode coordinates (relative to bregma): 0.5 mm rostral or 3.5 mm caudal and 1 mm lateral to each side (four screws, two in each hemisphere); group 4 coordinates: 4.8 mm posterior or 2.7 mm anterior and 2.2 mm lateral to each side (four screws, two in each hemisphere); group 5 coordinates: 3 mm caudal and 2.5 mm lateral to each side (two screws, one in each hemisphere). Groups 6 to 8 electrode coordinates: 3 mm caudal and 2 mm lateral (one screw in each hemisphere). In groups 3 and 6 to 8, a pump (1 μl/hour; model 2001, ALZET) cannula was inserted via a drill in coordinates 1 mm lateral, 0.5 mm caudal, and 1 mm ventral. In group 4, the pump (2.5 μl/hour; model 2ML4, ALZET) cannula was inserted in coordinates 1 mm caudal, 1.5 lateral, and 4 mm ventral. Continuous ECOG (sampling rate, 500 Hz for groups 1 to 4 and 1000 Hz for groups 5 to 8) was recorded wirelessly from freely moving animals in their home cage for the described duration of experiments.

ECOG/EEG analysis

Signal processing was performed offline. Preprocessing included high–pass filter (1 Hz), low–pass filter (100 Hz), and notch filter (45 to 55 Hz). EEG human recordings were analyzed as reference to average. Human data were preprocessed by EEGLAB (59). Animal data were preprocessed using home-developed MATLAB scripts. To detect PSWE, we buffered ECOG or EEG signals into 2-s-long epochs with 1-s overlap. Spectral analysis by fast Fourier transform (FFT) was applied, and the MPF was extracted for each epoch. PSWE was defined for human recordings as an MPF of <6 Hz for more than five consecutive seconds and that for animals was defined as an MPF of <5 Hz for more than 10 consecutive seconds (see Results). Analysis was performed separately for each recorded channel. Power spectrum for multichannel EEG data was calculated for the averaged channel, as well as for animals ECOG data over a single channel. FFT was also applied for the entire recording period buffered into 8-s epoch with 4-s overlap to analyze relative power across the frequency spectrum of 1 to 20 Hz. Processing was performed by home-developed MATLAB scripts. Analyzers were blind to patients/subjects/animals’ details.

Locomotor activity

Locomotor activity quantification in freely moving animals was performed as previously described (21). Briefly, signal strength (sampling rate, 250 Hz) derivative received from the implanted transmitter is correlated with the animal’s movement. Data were buffered into 2-s-long epochs with 1-s overlap, and mean value was calculated per epoch. Thus, locomotor activity indication was obtained for every second of the recorded period.

Immunohistochemistry

For rat samples, animals were anesthetized with 5% isoflurane and transcardially perfused with ice-cold heparinized physiological saline (10 U of heparin/ml of physiological saline), followed by 4% paraformaldehyde (PFA; #AC416785000, Thermo Fisher Scientific) in 0.1 M phosphate-buffered saline (PBS). Brains were removed, postfixed in 4% PFA for 24 hours at 4°C, and cryoprotected in 30% sucrose in PBS. Brains were then embedded in Tissue-Tek O.C.T. compound (Sakura), frozen and sliced on a cryostat into 20-μm coronal sections, and mounted on slides. Samples were stained according to the following protocol. Slides were treated for antigen retrieval, incubated for 15 min at 65°C in tris-EDTA buffer [10 mM tris base, 1 mM EDTA solution, and 0.05% Tween 20 (pH 9.0)], and then incubated in blocking solution (5% normal donkey serum in 0.1% Triton X-100/tris-buffered saline) for 1 hour at room temperature. Samples were then stained with primary antibody at 4°C, followed by fluorescent-conjugated secondary antibody for 1 hour at room temperature, and then incubated with 4′,6-diamidino-2-phenylindole (900 nM; Sigma-Aldrich) to label nuclei. Secondary antibodies were anti-rabbit Alexa Fluor 568, anti-chicken Alexa Fluor 647, anti-goat Alexa Fluor 488, anti-mouse Alexa Fluor 488 (1:500; Jackson ImmunoResearch), and anti-goat Alexa Fluor 647 (1:500; ab150131, Abcam). All antibody dilutions were in blocking solution.

Epilepsy induction by SE in rats

SE was induced in Sprague-Dawley rats using paraoxon [0.45 mg/kg (equivalent to 1.4 median lethal dose) intramuscularly (IM)] as previously reported (60). Paraoxon injection was followed 1 min later by atropine (3 mg/kg IM) and toxogonin (20 mg/kg IM) to reduce peripheral effects of paraoxon and to lower mortality rate. Rats were monitored for mortality and sickness during the next 48 hours.

Animal MRI BBB imaging

Animals (n = 12 naïve and n = 10 epileptic Sprague-Dawley rats) were scanned 1 month after induction of SE (for epileptic group) using the Aspect M2 system (Aspect Imaging Technologies) as previously described (36). Briefly, postcontrast signal intensity changes were measured in the whole brain. To measure leakage of the contrast agent, a linear curve was fitted to the dynamic scan intensities of the six consecutive postcontrast T1 scans. That is, a signal s(t) is fitted such that: s(t) = A × t + B, where the slope A is the rate of wash-in or wash-out of the contrast agent from the brain parenchyma. The BBB score represents the percentage of brain voxels that have a slope value >0. A threshold score was defined as the means + SD of the BBB scores of the control group (12 rats), and BBBd was defined as a score higher than the threshold.

Human BBB imaging

Imaging protocol was approved by the institutional review board of Soroka University Medical Center. All participants signed on an informed consent form before undergoing the scan. BBB status was measured using DCE-MRI in 59 nonepileptic individuals and in n = 12 patients followed at the epilepsy outpatient clinic. See tables S4 and S5 for epidemiological data. BBB permeability was calculated in each brain voxel using an in-house MATLAB script (MathWorks), as previously described (3032). Briefly, a linear regression is applied to the later part of the concentration curve of each voxel (6 to 20 min); the derived slope is divided by the slope at the superior sagittal sinus, to compensate for physiological (heart rate and blood flow) and technical (contrast agent injection rate) variability. Analyzers were blind to patients/subjects’ clinical data.

Statistics

Statistical tests were performed by Prism (GraphPad) unless mentioned otherwise. Mann-Whitney U test was used to compare PSWE occurrence per minute between groups, relative power frequency bands, PSWE durations, and median frequency of PSWEs. When PSWE occurrences were compared between hemispheres (Fig. 5), Wilcoxon test was implemented. Comparison of PSWE or relative power in more than two groups was made by Kruskal-Wallis test, followed by Mann-Whitney. Distributions of PSWE duration or median frequency were compared by Kolmogorov-Smirnov test. Comparing prevalence of AD or epilepsy between groups was done by χ2 test. MMSE correlation with PSWE was tested by Pearson’s correlation coefficient. Comparison of PSWEs throughout day and night by hour (Fig. 3F) was done by multiple t test comparisons with Holm-Sidak correction. PSWE correlation with BBBd per lobe was performed by Spearman coefficient correlation. All tests are two-sided. GMM for blind clustering was performed by MATLAB.

SUPPLEMENTARY MATERIALS

stm.sciencemag.org/cgi/content/full/11/521/eaaw8954/DC1

Fig. S1. PSWE occurrence per minute with respect to age among epilepsy patients.

Fig. S2. Locomotor activity of old animals during PSWE.

Table S1. Patients of Fig. 1.

Table S2. ROC analysis values of different PSWE thresholds.

Table S3. Patients with EEG recordings in Fig. 2.

Table S4. Clinical data of patients with epilepsy.

Table S5. Patients with DCE-MRI scan in Fig. 2.

Table S6. Spearman’s r values and P values per patient as demonstrated in Fig. 2G.

Table S7. PSWE occurrence per day in aged versus young mice in Fig. 3A.

Table S8. PSWE occurrence per day in control versus 5xFAD mice in Fig. 3G.

Table S9. Percentage of BBBd in naïve and epileptic rats in Fig. 4A.

Table S10. PSWE occurrence per day in naïve and epileptic rats in Fig. 4D.

Table S11. PSWE occurrence per day in ACSF-treated versus albumin-treated rats in Fig. 5E.

Table S12. EEG electrode locations of patients recorded in Rabin and Wolfson Medical Centers.

Table S13. EEG electrode locations of patients recorded in Soroka Medical Center.

Data file S1. Raw data.

REFERENCES AND NOTES

Funding: This research was supported by the European Union’s Seventh Framework Program (FP7/2007-2013, grant agreement 602102, EPITARGET) to A.F., the Israel Science Foundation (717/15) (to A.F.), the Binational Israel-USA Science Foundation (to D.K. and A.F.), a Bakar Foundation Fellowship and the Archer Foundation Award (to D.K.), NSF GRFP fellowships (to V.V.S. and A.R.F.), and the Canadian Institute of Health Research (CIHR 366355) (to A.F.). Author contributions: D.Z.M. devised the study, performed experiments and analysis, and wrote the manuscript. J.O. analyzed ECOG and DCE-MRI data. V.V.S. and A.R.F. recorded ECOG from old and young mice. O.P., N.E., and E.S. recorded ECOG from rats after SE. L.S. recorded EEG from patients with epilepsy. R.V. developed DCE-MRI methods. I.W. and D.Z. recorded ECOG from pump-implanted mice and 5xFAD mice. G.B.-K. and E.H. performed MRI scans of SE rats guided by M.H.S. G.B.-A. and I.S. clinically analyzed MRI scans of patients. O.S. developed tools for human EEG analysis. L.K. developed tools for ECOG analysis. R.S.-A. performed human MRI and analyzed data. I.G., A.G., and F.B. analyzed clinical data. F.B., A.F., and D.K. directed the project. D.Z.M. and A.F. wrote the manuscript. Competing interests: D.K. is a member of the advisory board of Minerva Technologies Ltd. D.K. and A.F. are co-founders and shareholders of Mend Neuroscience. A.F. is the founder of EMAGIX Inc. A.R.F. and V.V.S. have stock options in a company working in an area related to the subject matter of this manuscript. Data and materials availability: All data associated with this study are present in the paper or the Supplementary Materials.

Stay Connected to Science Translational Medicine

Navigate This Article