Sixty-eight young, healthy adults (average age 23.3 ± 3.3 yrs., age range: 18–30 yrs., 34 women) were recruited from the local community through flyers and ads. They were all right-handed and reported no illegal drug, cannabis, or tobacco use at least 1 month prior to the study, no history of seizures, brain injury, neurological or neuropsychiatric disorders, no vision, hearing, or learning problems, and no medication use at the time of the study. This information was obtained in an initial online screening survey, and was queried in greater detail in a follow-up phone interview. Based on questionnaires querying their current and recent drinking patterns, they were assigned to Binge Drinking (BD) and Light Drinking (LD) groups (Table 1). The BD group comprised 34 participants (17 women) who reported at least five binge episodes in the past 6 months and at least one binge episode in the previous month, with 13.2 ± 8.9 binge episodes on average. A binge episode was defined as consuming at least 6 (men) or 5 (women) drinks within a two-hour time span. This criterion was adopted based on the evidence suggesting that this level of drinking is likely to result in blood alcohol concentration (BAC) of 0.08% or above (Lange and Voas, 2001). The remaining 34 participants (17 women) who reported no more than one binge episode in the past 6 months were assigned to a LD group. No abstainers were recruited as all LDs reported consuming at least 1 drink per week on average. The two groups were matched on age, sex, education, ethnicity/race, and family history of AUD (Table 1). They took part in an encoding session (ENCODE), followed by a recognition session scheduled 48 h later. In addition, 19 BDs and 25 LDs participated in a third session probing recognition after a 6-mos long retention interval.

Participants completed a battery of questionnaires including handedness (Oldfield, 1971), medical history, drinking habits including the frequency and quantity of alcohol consumption (modified from Cahalan et al., 1969), the severity of alcoholism-related symptoms (Self-Administered Short Michigan Alcoholism Screening Test, SMAST; Selzer et al., 1975), and motives for engaging in alcohol use (Drinking Motive Questionnaire Revised Short Form, DMQ-R SF; Kuntsche and Kuntsche, 2009). A modified version of the Family History Assessment Module (FHAM; Rice et al., 1995) was administered to assess family history of AUD. Family history positive (FHP) participants were those who reported at least one first-degree and one first-or second-degree relative, or at least three second-degree relatives. Prospective participants with a maternal history of alcohol misuse were excluded from the study to avoid possible fetal alcohol exposure confounds. Family history negative (FHN) participants reported no first-or second-degree biological relatives with problem drinking or AUD. A subset of participants (6 BDs and 4 LDs) did not meet the criteria for either negative or positive family history. In addition, participants completed questionnaires measuring depression (9-item Patient Health Questionnaire, PHQ-9; Kroenke and Spitzer, 2002) and anxiety (7-item anxiety scale, GAD-7; Spitzer et al., 2006). The participants also completed the NIH-Toolbox Cognitive Battery comprising the List Sorting Working Memory Test to assess working memory capacity, Dimensional Change Card Sort (DCCS) Test to assess cognitive flexibility, Pattern Comparison Processing Speed Test to measure processing speed, and Picture Sequence Memory (PSM) Test which probed episodic memory (Gershon et al., 2013).

Participants came to the lab on three occasions. The first (ENCODE) and the second (48-h) experimental sessions were scheduled exactly 48 h apart and the third session followed after 6 mos. The participants were asked to refrain from consuming any alcohol at least 48 h prior to each experimental session. Upon arrival at the lab, they provided a urine sample and all tested negative on a 12-panel multidrug test (American Screening Corporation, United States). In the ENCODE session, participants completed an emotional rating task the results of which have been reported in a separate manuscript (Huang et al., 2018). This rating task served as an implicit encoding of the pictorial stimuli while the EEG signals were recorded from the electrodes placed on the scalp. In the two subsequent sessions participants took part in recognition tasks probing their recent (48-h) and remote (6-mos) memory of these pictures. Participants were monetarily compensated for taking part in the study.

During all three sessions participants were presented with color pictures depicting scenes with negative, neutral, or positive valence, which were selected mostly from the International Affective Picture System (IAPS; Lang et al., 2008). In the ENCODE session, 264 pictures were included in the emotional rating task. EEG analyses were carried out only on these initial 264 pictures. They were used as “old” (previously seen) items in both 48-h and 6-mos recognition sessions (Figure 1). In each recognition session, additional 96 pictures were presented as “new,” not previously seen items. Importantly, the “new” pictures included in the 48-h recognition session were used as lures in the 6-mos recognition session, but were excluded from the behavioral analysis of that session. The “old” and “new” sets were randomly selected from a larger picture set and were equated in valence, arousal ratings, and the presence of human faces. The stimulus set contained an equal number of pictures with positive, neutral, and negative emotional valence. For more details on picture selection and characteristics please see a related article (Huang et al., 2018).

In all three experiments, each trial started with a fixation cross shown on the screen for 1,000 ± 250 ms, followed by a picture presented for 1,000 ms in the center of a 24-inch color monitor subtending a visual angle of 6.0° × 4.5° at a viewing distance of 180 cm. Pictures were presented in a randomized order in eight blocks. After the offset of each picture, a scale was shown on the screen for 2,700 ms. In the emotional rating task during ENCODE, participants were instructed to judge how each image made them feel on a 9-point visual Likert scale ranging from 1 (very negative) to 9 (very positive) by using a joystick. In the subsequent 48-h recognition session, participants were asked to judge whether they had seen each picture during ENCODE (remembered/old) or not (new) and to indicate the confidence level on a 4-point rating scale (1 = high-confidence new, 2 = low-confidence new, 3 = low-confidence old, 4 = high-confidence old) with a joystick. Similarly, in the 6-mos delay recognition session, they judged whether they had seen the picture before and the confidence level on the same scale. The experiment was conducted with a PC using the Presentation software (Neurobehavioral Systems Inc.). Before each recording, participants practiced the task with additional 20 images that were excluded from the actual experiment.

EEG signals were recorded from a 64-channel actiCap DC Brain Vision system (Brain Products GmbH, Germany) and were continuously sampled at 500 Hz. The nose served as the reference and an electrode attached to the forehead as ground. Eyeblinks and eye movements were monitored with bipolarly referred electrodes attached above and below the left eye. The electrode impedance was maintained below 5 kΩ.

EEG data were analyzed with customized MATLAB (Mathworks, Natick, MA) routines that incorporated modules from publicly available packages including Fieldtrip (Oostenveld et al., 2011) and EEGLAB (Delorme and Makeig, 2004). The continuous EEG data were bandpass filtered from 0.1 to 100 Hz, and segmented into epochs extending from-300 to 1,000 ms relative to each stimulus onset. A 300 ms long padding was added to each end of the epoch to account for edge artifacts during the wavelet analysis. The data were carefully inspected and the trials that contained obvious artifacts were rejected. An independent component analysis was used to detect and remove artifacts caused by eyeblinks and heartbeat (Delorme and Makeig, 2004). Complex power spectra were calculated for each trial using Morlet wavelets (Oostenveld et al., 2011) in 1 Hz increments with 500 ms cycle length and 2–4 cycles across all frequencies in theta range (4–7 Hz) (Marinkovic et al., 2012). Theta band wavelet results were visually inspected and any additional trials that were contaminated by artifacts were rejected. Analysis of the raw theta power in the baseline showed no group or condition differences, indicating that any observed stimulus-related differences were due to event-related changes in theta power and not to the overall differences in the baseline. Event-related theta power was averaged across theta band frequencies (4–7 Hz) and across trials for each condition and expressed as percent signal change from the average theta power during the-300 to 0 ms prestimulus baseline. To examine signal distribution across the scalp, event-related theta indices were averaged into five electrode clusters representing the frontal (comprising AFz, AF3, AF4, Fz, F1, F2, F3, F4, F5, F6 electrodes), central (FCz, FC1, FC2, FC3, FC4, FC5, FC6, Cz, C1, C2, C3, C4, C5, C6), parietal (CPz, CP1, CP2, CP3, CP4, CP5, CP6, Pz, P1, P2, P3, P4, P5, P6), left temporal (FT7, T7, TP7, TP9, P7), and right temporal (FT8, T8, TP8, TP10, P8) montage areas (López-Caneda et al., 2013, 2014). Co-oscillations between the midline frontal (Fz) and the electrodes positioned over the left and right temporal areas were estimated by calculating phase-locking values (PLV) which reflect the consistency of phase differences in theta frequency band irrespective of the amplitudes of the neural activity (Lachaux et al., 1999; Tallon-Baudry and Bertrand, 1999; Beaton et al., 2018; Correas et al., 2019; Marinkovic et al., 2019). PLVs were expressed as percent signal change from the baseline. All event-related changes were quantified by analyzing a time window capturing the peak activity.

The subsequent memory effects (SMEs) for the encoding EEG signals were analyzed as a function of retrieval after two retention delays. Specifically, the EEG trials recorded during the ENCODE session were divided into the trials that were later remembered with high-confidence, and those that were later forgotten, as indicated by recognition performance after 48 h and 6 mos, respectively. Event-related theta power and PLV indices were analyzed with mixed-design analyses of variance (ANOVAs) with Group (BD vs. LD) as a between-subject factor and SME (Later Remembered and Later Forgotten) as a within-subject factor.

To examine the behavioral indices of changes in memory retrieval as a function of delay, mixed-design ANOVAs with Group (BD vs. LD) and Delay (48-h and 6-mos) were carried out on the d-prime (d’) derivation based on recognition with high (H-d’) and low (L-d’) confidence. D′ was calculated from hit rate (HIT) and false-alarm rate (FA) using the formula d’ = Z_HIT – Z_FA where Z represents a transformation of the two distributions allowing for comparison of measures with different ranges of absolute values (Macmillan and Creelman, 1990). Trials were aggregated across all emotional categories to ensure optimal power for the EEG analyses. Moreover, the Emotion × SME interaction effects on behavioral HIT and FA rates were comparable for BD and LD groups, all ps > 0.12. The factor of Sex was included initially in an overall analysis model for both the EEG and behavioral data. However, there were no main effects or interactions including Sex, so it was omitted from the reported analyses. Moreover, the factor of Brain Region (frontal, central, parietal, left temporal, right temporal) was included in an initial analysis of the theta activity, but no effects on SMEs were found, ps > 0.10 (Supplementary Table S1). However, the analyses focused on the frontal electrodes where the memory modulations of theta activity appeared most prominent. To estimate fronto-temporal interactions, PLVs were primarily calculated for the Fz-C5 and Fz-C6 electrode pairs, which permitted laterality comparisons. PLV values for all other calculations between Fz and other lateral electrodes are available in the Supplementary Table S2.

To investigate the changes in participant self-reports across the 6-month time period, the questionnaire scores were analyzed with mixed model ANOVAs with Group as a between-group factor and Delay (at enrollment vs. 6 months later) as a within-subject factor. Group differences in dispositional variables were evaluated with the Mann–Whitney U tests for independent sample comparisons given that many variables violated the assumption of normal distribution. A Chi-square test was used for categorical variables such as sex and race/ethnicity. Spearman’s rank correlation analyses were performed on the intensity of drinking behaviors/symptoms and the strength of the EEG-based brain activity. Of note, one BD participant who reported a number of binge episodes beyond three standard deviations of the mean was excluded from all correlation analyses.

As expected, group differences were observed on all variables associated with alcohol consumption (Table 1), as BDs reported consuming more alcohol, engaging in more binge episodes, and experiencing more blackouts and other alcohol-related consequences than LDs. However, the BD and LD groups did not differ on NIH-Toolbox tests of cognitive functions including working and episodic memory. A follow-up after 6 months confirmed all group differences in drinking habits.

Figure 2 displays d’ for both groups based on high-confidence (H-d’) or low-confidence judgments (L-d’) during the recognition task. As expected, participants showed higher overall H-d’ after a 48-h, compared to a 6-mos delay, F(1, 42) = 184.43, p < 0.001, indicating a decrease in H-d’ across time. There was no main effect of Group, F(1, 42) = 0.42, p = 0.52, or an interaction effect of Group × Delay on H-d’, F(1, 42) = 0.59, p = 0.45. Similarly, while a reduction of L-d’ was observed after a 6-mos relative to a 48-h delay, F(1, 42) = 116.58, p < 0.001, there was no group difference, F(1, 42) = 0.11, p = 0.74, and no Group × Delay interaction, F(1, 42) = 0.11, p = 0.74.

In the present study, the BD and LD groups differed in SMEs reflected in theta power and co-oscillations. However, we found no group differences in behavioral performance, which is broadly aligned with extant evidence. Binge drinking seems to exert a subtle impact on verbal memory performance, and only a small proportion of studies have reported impaired performance on visual memory tasks in young binge drinkers (see Carbia et al., 2018 for review; Scaife and Duka, 2009). Of note, an EEG study that demonstrated deficits in SME-associated event-related potentials (ERPs) during encoding in binge drinkers did not identify group differences in behavioral performance, either (Folgueira-Ares et al., 2017). Moreover, a recent review of EEG studies on binge drinking (Almeida-Antunes et al., 2021) reported that behavioral differences between BDs and non/low drinkers were observed in fewer than 25% of studies that employed cognitive tasks. Indeed, group differences are typically observed in studies using neural measures, often in the absence of behavioral deficits (Crego et al., 2012; Petit et al., 2014; Huang et al., 2018; Holcomb et al., 2019; Lannoy et al., 2019), suggesting the subtlety of deficits at the behavioral level. Moreover, the BD group comprised highly functional individuals whose performance did not differ from LDs on neuropsychological tests of episodic memory, working memory, processing speed, or cognitive flexibility, corroborating that EEG indices are more sensitive to neural alterations in young BDs than behavioral measures (Maurage et al., 2009; López-Caneda et al., 2017; Huang et al., 2018; Correas et al., 2019; Holcomb et al., 2019).

In the present study, SMEs were reflected in greater event-related theta power during encoding which predicted better recognition performance after a long retention delay. This finding is consistent with recent reports of increased frontal midline theta recorded during encoding of items that are subsequently remembered (see Hsieh and Ranganath, 2014 for review). Intracranial EEG (iEEG) recordings in humans indicate that theta oscillations are primarily generated in superficial cortical layers and may represent widespread integration across different cortical areas (Wang et al., 2005; Halgren et al., 2015, 2018; Solomon et al., 2019). Intracranial EEG evidence also suggests that the coherent theta-band activity in the hippocampus supports successful encoding of new items by coordinating cortical rhythmic activity (Hasselmo and Eichenbaum, 2005; Lega et al., 2012; Berens and Horner, 2017; Zheng et al., 2019). Consequently, it has been proposed that theta oscillations recorded from neocortical areas during memory formation reflect activity of the hippocampo-cortical feedback loops (Klimesch et al., 1997; Jones and Wilson, 2005; Eichenbaum, 2017). In support of this idea, our results indicate that theta activity in the neocortex, which is likely coordinated by hippocampal theta, is important for creating the integrated representations of novel items in the memory system (Siapas et al., 2005; Sauseng et al., 2007; Benchenane et al., 2010; Nyhus and Curran, 2010).

Consistent with the integrative role of theta during encoding, our PLV results indicate elevated fronto-posterior theta phase-locking during SME in the LD group. Specifically, the LD group showed greater theta-entrained PLVs between the frontal and the left-dominant posterior brain regions to the pictures that were subsequently remembered with high-confidence, relative to those that were subsequently forgotten after both retention intervals. These PLV findings align with other scalp EEG studies documenting increased theta phase synchronization between the frontal and posterior cortices during episodic memory formation (Schack and Weiss, 2005; Summerfield and Mangels, 2005; Staudigl and Hanslmayr, 2013). Such enhanced fronto-posterior theta synchrony during encoding adds to the evidence that formation of episodic memories is subserved by neural synchrony integrating diverse brain regions including the frontal and the lateral and medial temporal lobes (Paller and Wagner, 2002). Further, the stronger SMEs on theta oscillations between frontal and left posterior locations are aligned with the prior evidence of left-lateralized SME on theta oscillatory activity (Staudigl and Hanslmayr, 2013; Miller et al., 2018) and the importance of the left entorhinal cortex for successful encoding (Solomon et al., 2019). Even though the great majority of studies have probed verbal memory (Staudigl and Hanslmayr, 2013), left-lateralized theta during pictorial encoding is also sensitive to successful memory (Pu and Yu, 2019).

While there were no group differences in SMEs on event-related theta power for a short retention interval, the SME theta modulations were attenuated in BDs when considered for the 6-mos interval. It points to selective deficits in oscillatory neural networks subserving encoding processes that dissipate over time and are not maintained over a longer delay. Similarly, the PLV data unveiled the absence of SMEs on the theta phase-locking between the frontal lobe and the posterior regions in BDs. Indeed, greater functional connectivity (Sneve et al., 2015) between the hippocampus and other areas, as well as more robust DTI connectivity with the prefrontal cortex (Cohen, 2011) result in stronger and longer lasting memory. The present finding of weaker or absent SMEs for both, event-related theta and synchronous co-oscillations in BDs compared to LDs, mirrors prior evidence of the deficient neural synchrony subserving integrative cognitive processing following acute alcohol consumption or among young binge drinkers (Beaton et al., 2018; Correas et al., 2019; Marinkovic et al., 2019).

These observations are consistent with the convergent evidence of alcohol-induced disturbances in the brain areas critical for memory formation such as the hippocampus and the prefrontal cortex (Oscar-Berman and Marinkovic, 2007; Oscar-Berman et al., 2014; Sullivan, 2017; Fama et al., 2021). Relatedly, fMRI evidence has documented atypical activation patterns in the hippocampus mediating novel encoding in teenage binge drinkers, suggesting that binge drinking may alter the neural substrate of the encoding processes in the developing brains (Schweinsburg et al., 2010). Studies in rodents have confirmed morphological changes in the hippocampus such as decreased numbers of pyramidal and dentate gyrus granule neurons (Nixon et al., 2002; Herrera et al., 2003) as well as suppressed induction of long-term potentiation (Roberto et al., 2002) following chronic exposure to ethanol. Neurodegeneration and inhibition of neurogenesis following long-term exposure to ethanol have also been documented in frontal regions in animal model studies (Crews et al., 2000; Crews and Nixon, 2009). These extensive findings from animal models provide substantial explanatory evidence for the neurophysiological underpinnings of the impairments in long-term episodic memory reported in humans with AUD (Oscar-Berman and Marinkovic, 2007; Oscar-Berman, 2012; Stavro et al., 2013).

These alterations in theta activity and fronto-posterior theta phase synchrony during encoding for long-term memory among young BDs are suggestive of a selective dysregulation of excitation/inhibition (E/I) balance that underlies the long-range co-oscillatory synchrony between the principal cortical and limbic nodes implicated in long-term memory as function of binge drinking (Klimesch et al., 2001; Siapas et al., 2005). Indeed it has been well established that neural inhibition, as effectuated by the gamma amino butyric acid (GABA), the primary inhibitory neurotransmitter, plays an essential role in stabilizing neural networks and memory consolidation (Barron, 2021). However, alcohol misuse is associated with reduced inhibitory function (Koob and Le Moal, 2008; Roberto and Varodayan, 2017). Indeed, recent evidence indicates that binge drinking is associated with lower GABA concentration (Marinkovic et al., 2022) and neural hyperexcitability (Correas et al., 2021). GABA reduction in the hippocampus is associated with neural hyperactivity and memory impairments in animal (Li et al., 2021) and human studies (Jiménez-Balado et al., 2021). Over time, alcohol-induced neurochemical changes may contribute to the allostatic neuroadaptations in limbic brain structures that are critical for memory consolidation (Koob, 2003; Koob and Le Moal, 2008; Wise and Koob, 2014). Such neuroadaptive effects observed in young BDs tip the E/I balance towards excitation, making it more difficult to encode information and retain it over longer time intervals. The altered theta-mediated memory processes may underpin the more frequent alcohol-induced blackouts that contribute to memory loss for the events occurring during intoxication (Read et al., 2013; Hingson et al., 2016), and lower academic achievement (Miller et al., 2007; Pascarella et al., 2007; Huang et al., 2018; Holcomb et al., 2019), which have been reported in college binge drinkers.

While these findings provide evidence for disrupted theta activity during memory encoding in young BDs, these deficits are broadly consistent with dysregulated theta observed in BDs during tasks probing attention (Correas et al., 2019), inhibitory control (Holcomb et al., 2019) and emotional processing (Huang et al., 2018). Furthermore, theta dysfunction during cognitive tasks has also been observed in individuals with AUD (Kamarajan et al., 2004, 2012; Jones et al., 2006) and adolescents at high risk for alcohol addiction (Porjesz and Rangaswamy, 2007). This convergent evidence suggests that theta alterations during memory processing is one aspect of a more general deficit in neurocognitive functioning in relation to binge drinking. Furthermore, binge drinking is theorized to be a transitional phase towards alcohol dependence (McCarty et al., 2004; Enoch, 2006). Therefore, these theta disturbances during mnemonic processes among BDs may be the precursor to memory impairments characterizing AUD (Oscar-Berman and Marinkovic, 2007). In addition, our results are aligned with proposals that altered theta synchrony during cognitive processes may serve as an effective neurophysiological marker of a predisposition towards the development of AUD (Porjesz et al., 2005; Rangaswamy and Porjesz, 2014). However, due to the relatively small sample size impeding the statistical power especially for the recognition assessment after a 6-mos delay, the observed group differences should be interpreted with caution. Moreover, to ensure sufficient power for the EEG analyses, trials were aggregated across emotional picture categories. Given the previous reports of altered EEG indices of emotional processing in binge drinkers (e.g., Huang et al., 2018), future studies could be designed to allow comparisons between emotional categories to examine the impact of emotional processing on SME-related EEG outcomes. Another limitation involves the potential confounding effect of volume conduction on the estimate of the theta synchronization between different electrode locations. It is recommended that future researchers endeavor to apply advanced methods (e.g., Bruña et al., 2018) to mitigate such possible effects.

In conclusion, in the absence of BD vs. LD group differences in pictorial memory performance, SME theta power associated with long-term (6 mos) memory retention was attenuated in BDs compared to LDs. This observation suggests that during encoding, LDs were able to engage a distributed neural network reflected in increased theta power, supporting item retention in remote memory. In contrast, the BD group was characterized by an inefficient network-level interactive engagement of the brain areas that mediate memory formation, particularly for the items that were prospectively retained over 6 months in remote memory. This SME deficit is further substantiated by dysregulated long-range synchronous co-oscillations in BDs. The importance of excitation/inhibition (E/I) balance for the long-range corticolimbic neural synchrony that mediates memory encoding and retention is well established. At the same time, convergent animal and human mechanistic evidence indicates that E/I dysregulation is associated with alcohol misuse represented by binge drinking. Thus, the observed SEM deficits in BDs are consistent with suboptimal neural activity in the corticolimbic circuitry during encoding. Aligned with other evidence, the divergence between the behavioral and EEG results endorses the argument that direct neural measures are selectively sensitive to deficits in young BDs that are otherwise too subtle to be detected with behavioral tests (Lannoy et al., 2019; Almeida-Antunes et al., 2021). These findings address a gap in the memory literature on binge drinking and expand our current understanding of possible neural underpinnings of the early stages of alcohol use disorder. Furthermore, these results may have clinically relevant implications for the development of diagnostic and prevention strategies for problematic alcohol use by underscoring the importance of the elements that focus on memory disturbances.