The authors have declared that no competing interests exist.
Conceived and designed the experiments: CLP CL PB. Performed the experiments: CL. Analyzed the data: CL MAK CLP. Contributed reagents/materials/analysis tools: DK LK. Wrote the paper: CL CLP. Revision of the manuscript: MA PB CP LK.
Acute alcohol intake is known to enhance inhibition through facilitation of GABAA receptors, which are present in 40% of the synapses all over the brain. Evidence suggests that enhanced GABAergic transmission leads to increased large-scale brain connectivity. Our hypothesis is that acute alcohol intake would increase the functional connectivity of the human brain resting-state network (RSN). To test our hypothesis, electroencephalographic (EEG) measurements were recorded from healthy social drinkers at rest, during eyes-open and eyes-closed sessions, after administering to them an alcoholic beverage or placebo respectively. Salivary alcohol and cortisol served to measure the inebriation and stress levels. By calculating Magnitude Square Coherence (MSC) on standardized Low Resolution Electromagnetic Tomography (sLORETA) solutions, we formed cortical networks over several frequency bands, which were then analyzed in the context of functional connectivity and graph theory. MSC was increased (p<0.05, corrected with False Discovery Rate, FDR corrected) in alpha, beta (eyes-open) and theta bands (eyes-closed) following acute alcohol intake. Graph parameters were accordingly altered in these bands quantifying the effect of alcohol on the structure of brain networks; global efficiency and density were higher and path length was lower during alcohol (vs. placebo, p<0.05). Salivary alcohol concentration was positively correlated with the density of the network in beta band. The degree of specific nodes was elevated following alcohol (vs. placebo). Our findings support the hypothesis that short-term inebriation considerably increases large-scale connectivity in the RSN. The increased baseline functional connectivity can -at least partially- be attributed to the alcohol-induced disruption of the delicate balance between inhibitory and excitatory neurotransmission in favor of inhibitory influences. Thus, it is suggested that short-term inebriation is associated, as expected, to increased GABA transmission and functional connectivity, while long-term alcohol consumption may be linked to exactly the opposite effect.
Social drinking, for most people, is an inseparable part of every-day life. Alcohol is used and abused for its ability to modify emotional states, and more precisely, to reduce anxiety
Acute alcohol administration affects behavior
An important finding is that short-term alcohol consumption disrupts the delicate balance between inhibitory and excitatory neurotransmission in favor of inhibitory influences
Alcohol's effect on neurotransmitters may, at least partially, explain some of the complex behavioral manifestations of intoxication in both animals and humans. GABA induction has been associated with the observed sedation effects
The GABA neurotransmission is of particular interest in alcohol intoxication, since (i) it has been associated with a person's susceptibility to develop alcohol abuse or dependence
The disrupted balance between inhibitory and excitatory neurotransmission at the microscopic level caused by alcohol intoxication, may be reflected at the macroscopic level in the brain activity recorded even from outside the human scalp. Recent electroencephalographic (EEG) findings suggest that neurotransmission inhibition, induced by GABAergic agonist lorazepam, can cause a widespread increase in the inter-area functional connectivity and an increase in the strength of functional long-range and inter-hemispheric connections
Since alcohol admission affects significantly the inhibitory function of GABAA receptors and enhanced inhibition increases the inter-area functional connectivity, we hypothesize that short-term alcohol consumption will increase the functional connectivity of the human brain at rest. We expect the enhancement of the GABA inhibition activity to potentially affect the brain as a whole system rather than specific brain regions in particular, since GABAA receptors are widely distributed in the human brain.
In the current piece of work, we examine how acute alcohol intake modulates the human brain functional coupling of the resting-state network (RSN). The study of RSN is of particular importance for studying the GABA neurotransmission mechanism, since GABA concentration in specific regions has been recently found to be positively correlated with their activation at rest
Twenty-six healthy right-handed individuals (12 females, 14 males), with a mean age of 26.1±1.8 (mean ± SD), and free of any neurological or hepatic disorders, participated in the study. Female participants were in the first half of their menstrual cycle. All participants were free of any medication and completed a questionnaire on their drinking habits. The selected participants were characterized as light drinkers
Each participant was tested in two experimental sessions and received a real alcohol dose (alcohol session) and a placebo beverage (placebo session). The order of the sessions was counterbalanced, while the doses were administered double-blindly. Thus, half of the participants had an alcoholic beverage in the first session and a non-alcoholic placebo beverage in the second, while the order was reversed for the other half. The two sessions were performed at least six weeks apart to ascertain that the participants had no clear memory of the first session inebriation level. All experiments started at the same time of the day (16.00) in order to simulate as close as possible real life situations (people consume alcohol usually late in the afternoon and/or evening) and to avoid differences in the circadian rhythm. EEG and electrooculography (EOG) were collected for five minutes during an eyes-open as well as an eyes-closed session. During eyes-open session participants were asked to fixate on a white cross on a black screen placed 80 cm in front of their eyes, relax, and refrain from blinking and moving their eyes to the extent possible. They were also advised to keep their head still and refrain from falling asleep.
During the alcohol session, participants drank a mixture of two parts of orange beverage and one part of vodka (vol 37.5%). The mixture was divided in four doses, each consumed within 5 min. As a result, alcohol administration lasted twenty minutes and was followed by an absorption period of 25 min. The alcohol dose was adjusted (for males 0.48 g/kg and for females 0.38 g/kg) in order to achieve a Blood Alcohol Concentration of 0.7 g/L
In order to assess inebriation following acute alcohol intake, saliva samples were collected
Two extra saliva samples were collected from each participant during each session in order to measure salivary cortisol. The first one was collected upon arrival, prior to the procedure (pre-intake), while the second one 20 min after beverage consumption (post-intake), just before the recording, when alcohol is expected to reach its peak concentration in blood
The Salivette device (Sarstedt, Numbrucht) was used to collect and measure the salivary cortisol. The sample was moved into the laboratory's cooler and it was centrifuged in a cylindrical plastic tube to isolate the saliva, approximately 1.5–2 mL. Saliva samples were then stored at −20°C. The participant must abstain from food, and coffee at least two hours prior to sampling, as these factors change the environment of the oral cavity and the activation of the Hypothalamic Pituitary Adrenal (HPA) axis.
The measure of the cortisol in the saliva samples was performed with the Enzyme-Linked immunosorbent test (ELISA), also known as immunoassay, which is mainly used in immunology to detect the presence of an antibody or an antigen in a biological sample. The lower limit of sensitivity was determined by interpolating the mean minus 2 standard deviations for 18 zero standards. The minimal concentration of cortisol that can be distinguished from zero, is 2.5 pg/mL.
The difference in the stress levels caused by alcohol or placebo intake was assessed indirectly, that is, by means of the difference between the post-intake and the pre-intake cortisol values. The difference in the stress levels of the participants was the same for alcohol and placebo administration (p = .5, F (1, 25) = .046), meaning that the effect of alcohol and placebo drink on the stress levels was the same, as in previous literature
EEG signals were recorded (Nihon Kohden) from 57 scalp sites with a sampling rate of 500 Hz. The scalp electrodes were placed on a cap (EASYCAP) according to the 10/10 system. The electrodes were commonly referenced to the average of the two linked mastoid electrodes. EOG signals were also recorded simultaneously by means of four electrodes placed around the eyes (one above and one below the right eye, and another two placed at the outer canthi of each eye). The vertical EOG (VEOG) was calculated as the difference between the two signals recorded above and below the right eye, while the horizontal EOG (HEOG) as the difference between the signals recorded from the left and the right electrodes respectively. EEG signals were band-pass filtered between 0.5 and 90 Hz, while EOG signals between 0.5 and 8 Hz. Ocular artifacts were removed from the EEG signals by means of the REG-ICA methodology applying the technique proposed in
Cortical source time series were estimated with sLORETA
Functional connectivity was estimated across cortical sources, rather than across EEG sensors, because source topographies largely overlap at the EEG sensor level. We used MSC as a pair-wise connectivity measure.
MSC is a function of the Power Spectrum Density (PSD) of two signals x and y (Pxx and Pyy), and the cross PSD (Pxy) of signals x and y in a certain frequency f.
P was estimated using the Welch method
Finally, the
The networks consisting of nodes and connections were, therefore, mathematically represented as graphs. This information was stored in the form of a 259×259 Adjacency Matrix (AM), which in this case is symmetric, since MSCxy = MSCyx. For each subject, two weighted AM matrices were calculated, one for the alcohol and another for the placebo session. To reduce the complexity of the AMs, the original MSC values ranging from 0 to 1 were subjected to subsequent thresholds converting the weighted graphs into binary ones.
For each threshold and frequency band, for each of the two binary networks (alcohol and placebo sessions), and for each participant, the following parameters were calculated: density, characteristic path length, global efficiency, clustering coefficient, small-worldness and degree of each node.
Density (K) is the simplest parameter and is defined as the number of existing edges (E) divided by the number of the possible edges among nodes (N).
Characteristic path length (L) is an integration measure and is defined as the average of the shortest paths Li between each pair of connected nodes.
Global efficiency (E) is also an integration measure and is defined as the arithmetical mean of the inverse of the shortest paths between each pair of nodes.
Clustering coefficient (C) is a segregation measure quantifying the trend of the network to form clusters. It is defined as the mean of the C of each node i. The Ci of each node is the proportion of links between the vertices within its neighborhood divided by the number of links that could possibly exist between them.
Characteristic path length, global efficiency and clustering coefficient were normalized (divided) by the corresponding parameters of random surrogate graphs
The small-worldness (S) of a network is defined as the ratio of the clustering coefficient (C) divided by the characteristic path length (L), both normalized by a set of random graphs with the same number of nodes and connections
Finally, in an undirected network, the node degree is defined as the number of links connected to this node. This is not a large-scale parameter, since it is calculated for each node separately (herein 259) and it is not one single value globally characterizing the network.
For each connection during both eyes-open and eyes-closed sessions and in all frequency bands, repeated measures Analysis of Variance (ANOVA) was performed on the MSC values to reveal differences (alcohol vs. placebo intake) in the connectivity. The results were corrected for multiple comparisons (False Discovery Rate (FDR)
After constructing the binary graphs for multiple thresholds, repeated measures ANOVA (alcohol vs. placebo) was performed for each threshold on the globally estimated network parameters (density, characteristic path length, global efficiency and clustering coefficient, p<.05) and on the degree of each node (p<.05, FDR corrected similarly to
Correlations of the inebriation level, measured through SAC, with the concurrent network parameters during alcohol sessions were calculated for each threshold and only for the frequency bands where significant alcohol effects were observed. The post-intake cortisol values, reflecting the stress levels at the time of EEG recordings, were also correlated with the concurrent network parameters in all frequency bands for both the alcohol and placebo sessions. Pearson's correlation was used and the significance level was set to .05.
Finally, it was tested for all thresholds whether the functional networks obtained during alcohol and placebo sessions contrast significantly to ‘random’ graphs or not. A set of 1000 random undirected graphs with the same number of nodes and links as our experimental graphs were generated (similarly to
All statistical analyses were performed in Matlab R2010a (MathWorksInc, USA).
Differences during eyes-open and during eyes-closed sessions are presented in the upper and lower panel respectively. Significantly (p<.05, FDR corrected) increased connectivity is shown in the upper rows (red connections), decreased in the lower rows (blue connections). The number of significantly affected connections is also presented for each frequency band.
The graph density, path length, global efficiency and clustering coefficient were compared across participants between alcohol and placebo sessions.
Double stars (**) indicate significant p-values (p<0.03), single stars (*) indicate marginally significant effects (p<0.06). To maintain the high resolution and quality of the figures, we did not include all the results for the many different thresholds, but only for the first seven. Characteristic path length, clustering coefficient, and consequently small-worldness are normalized by random surrogate graphs
In alpha band, density and global efficiency were higher and characteristic path length was lower during the alcohol session as compared to placebo (see
Averaged values are grouped for alcohol (red symbols) and placebo (blue symbols) for both eyes-open (left panel) and eyes-closed (right panel) sessions. Greek letters designate the frequency band studied accordingly. The corresponding values of 100 random graphs are represented with black dots.
Eyes-open session | ||||||||||
z-scores | Alcohol | Placebo | ||||||||
δ | θ | α | β | γ | δ | θ | α | β | γ | |
Global Efficiency | 145.82 | 141.23 | 152.7 | 128.56 | 130.32 | 138.13 | 138.83 | 167.29 | 125.15 | 167.28 |
Local Efficiency | 223.17 | 217.42 | 231.78 | 201.56 | 203.76 | 213.53 | 214.41 | 250.07 | 197.28 | 250.06 |
z-values of the 20 contrasts performed with a z-test with significance level at 0.05 (FDR corrected). All contrasts were significant (p<0.001).
Eyes-closed session | ||||||||||
z-scores | Alcohol | Placebo | ||||||||
δ | θ | α | β | γ | δ | θ | α | β | γ | |
Global Efficiency | 145.96 | 141.36 | 152.84 | 128.7 | 130.46 | 138.26 | 138.96 | 167.44 | 125.29 | 167.43 |
Local Efficiency | 212.94 | 207.45 | 221.16 | 192.32 | 194.42 | 203.75 | 204.58 | 238.61 | 188.24 | 238.6 |
z-values of the 20 contrasts performed with a z-test with significance level at 0.05 (FDR corrected). All contrasts were significant (p<0.001).
Left: The highlighted nodes have a degree significantly altered by alcohol intake (vs. placebo) in alpha band during eyes-open and in theta band during eyes-closed. The size of red nodes is inversely proportional to the significant (< .05) p-values: the larger the node the more significant the effect is. Right: Mean node degree values of the nodes significantly affected (p<.05) by alcohol for alpha band (upper row) and theta band (lower row) networks averaged across participants (± SD).
Since alcohol affected only the network parameters on theta (eyes-closed), alpha and beta (eyes-open) frequency bands, possible correlations of network parameters with SAC and cortisol levels were examined selectively for these bands across different thresholds. The correlation between the network parameters and SAC was estimated only during the alcohol sessions, since no alcohol was detected in the saliva samples in placebo. Significant (p<0.01) positive correlations between SAC and density for thresholds from 0.2 till 0.5 were observed in beta band.
(A) Scatter plot of SAC values versus beta band density during the alcohol session; (B) To illustrate this correlation, the averaged functional networks of the 7 participants with the lowest and 7 participants with the highest SAC values are presented as weighted AMs. The weights of these AMs are the MSC values; (C) connections of the networks in (B), after being converted to binary graphs using a representative threshold of 0.35.
In the present study, the neurophysiology of short-term inebriation was studied for the first time under the context of functional connectivity and graph theory. Based on the already reported (i) facilitated large-scale functional connectivity due to enhanced GABAergic transmission
Our hypothesis was confirmed by the elevated functional connectivity in alpha, beta and theta bands after alcohol as compared to placebo. Increased undirected functional connectivity (herein estimated with MSC) is indicative of increased functional coupling between distributed brain regions exhibiting synchronized neuronal activity
The increased functional connectivity after short-term alcohol intake is further supported if one considers the reported reduced functional connectivity observed in chronic alcoholic patients
We should state, however, that since acute alcohol intake affects multiple neurotransmitter systems, like the dopaminergic and the opioids'
The majority of the significant alterations in MSC values were reported as MSC increases (vs. MSC decreases) observed almost in all frequency bands; however, they were more prominent in the theta, alpha and beta (see
The role of theta band at rest is also prominent; sharper topography of the DMN was obtained in alpha and theta bands in MEG recordings
The core graph parameters reflected also the elevated functional connectivity by detecting consistent alterations in the structure of brain complex networks. During eyes-open sessions, density and global efficiency were higher, while path length was lower following alcohol as compared to placebo in alpha band (see
Moreover, higher density accompanied with lower small-worldness was observed in beta band during eyes-open (see
During eyes-closed sessions, the global efficiency of the networks in theta band was marginally higher as compared to placebo (see
Core graph parameters were selected for this study since the main types of graph structure have been defined on the basis of their values
The most interesting finding of the present study is probably the correlation of alcohol concentration measured in saliva with the human RSN parameters. SAC is an accurate estimator of Blood Alcohol Concentration (BAC)
Neither the correlation with SAC, nor the main effect of alcohol was significant on global efficiency or characteristic path length. This fact indicates an alcohol-induced meaningless increase of the total wiring cost and the absence of a corresponding enhancement of the communication in the network. Our study provides evidence for an obstructed functional connectivity pattern due to alcohol in beta band, a band that has been also linked to emotion and cognition
From
The random graphs used as surrogates herein, were generated keeping the same number of nodes and connections. There are also other approaches that preserve also the same degree distribution
Effects of alcohol on the Central Nervous System (CNS) are primarily mediated via GABAA receptors, which are expressed across the brain. Different parts of the brain have, however, a higher affinity for different subunits
Following our findings on increased large-scale connectivity, the degree of specific nodes was higher following alcohol (vs. placebo). Our hypothesis for increased connectivity was reflected on the higher strength of connections, which in turn resulted to a larger number of connections on specific nodes. During eyes-open this was the case for nodes located on the frontal and medial sites in alpha band, while during eyes-closed this effect was observed for nodes located mainly on medial and posterior sites (see
The spatial pattern of the DMN consists of the regions which are deactivated during attention demanding tasks independently of the task, implying significant functionality during resting state (see
From a methodological point of view, our study presents some limitations. Our hypothesis, based on the effects of alcohol on GABA neurotransmission is supported by our findings. However, since EEG does not allow measuring the expected alcohol-induced alteration on the concentration of GABA in the brain, we cannot safely deduce a causality relationship between the observed pronounced functional connectivity and the neurotransmission effects. Moreover, alcohol studies face the problem of individual susceptibility and tolerance developed to the effects of alcohol. This effect can be critical when studying the intoxication process
Since individual MRI structural scans were not available, we used a BEM model in a standard brain for all EEG reconstructions. Combined with the inherently low spatial resolution of EEG, we cannot make detailed inferences on the anatomical origin of the recorded oscillations. The ideal experimental design for this study would also need the recording of the individual subject's MRIs and/or additionally magnetoencephalography (MEG) recordings at rest. MEG, as a technique, fundamentally avoids the limitations of EEG in functional connectivity analysis (i.e. possible leakage due to common reference for all channels) thereby providing also very good localization accuracy for the active sources
Regarding the connectivity methodology, different methods have been proposed to select the threshold to be applied on weighted AMs to convert them into binary ones. We followed the approach of arbitrary threshold to test if the significant results were preserved across multiple connection densities. There is still a lot of debate around this issue and, as a consequence, there is no optimal method for the selection of the threshold to construct graphs, although different approaches have been suggested (see
The current study provides evidence that acute moderate alcohol doses in healthy young social drinkers increases the functional connectivity of the RSN. Such increase is evident in terms of MSC values and it is also reflected on graph parameters that describe its internal organization and structure. Interestingly, the inebriation level was positively correlated with the RSN density. The findings support our hypothesis for increased large-scale connectivity due to the GABAergic inhibition induced by alcohol. The elevated connectivity is pronounced in alpha, beta and theta bands in line with the distinct role of these frequencies at rest. Thus, it is suggested that despite the fact that reduced GABA transmission is linked to reduced functional connectivity in alcoholics, facilitated GABAergic inhibition following short-term inebriation is associated to pronounced brain functional connectivity. Finally, all networks under study exhibited small-world properties and differed significantly from ‘random’ ones computationally produced with the same characteristics.
The authors gratefully thank Anastasia Semertzidou and Christos Frantzidis for the data collection, Domniki Fragou for the SAC analysis and Kalli Tzioti for participants' recruitment.