Reduced theta power during memory retrieval in depressed adults

Abstract

Background

Major Depressive Disorder (MDD) is associated with poor recollection, but the neural mechanisms responsible for this deficit are unclear. Recollection is supported by interactions between the hippocampus and cortex that appear to be mediated by oscillatory activity in the theta band (4–7Hz), and it is elicited during source memory retrieval. Therefore, we tested the hypothesis that evoked theta power during source memory retrieval would be reduced in MDD, as this would provide a physiological basis for deficient recollection in depressed adults.

Methods

Morlet wavelets were applied to event-related potentials collected from 24 unmedicated adults with MDD and 24 healthy controls during the retrieval of source and semantic memories. Whole-scalp analyses focused on group differences in evoked theta power.

Results

There were no group differences in behavior. Nevertheless, from 400–799 ms theta power was broadly reduced in depressed vs. healthy adults. This reduction was observed during source and semantic retrieval. However, parietal midline electrodes showed significantly reduced theta power during source—but not semantic—retrieval in depressed vs. healthy adults in this interval. Furthermore, theta power over parietal midline sites from 400–799 ms was more strongly related to source memory accuracy in healthy vs. depressed adults.

Conclusions

Relative to healthy controls, depressed adult showed reduced theta power during memory retrieval and a weaker relationship between parietal midline theta power and source memory accuracy. These findings indicate that abnormal theta signals may contribute to memory deficits in adults with MDD.

Introduction

Behavioral research has revealed poor recollection in depressed adults (1–5). Recollection refers to the retrieval of spatiotemporal details that define an encoding episode (6; 7), and it depends on cortico-hippocampal interactions (8–11). The memory deficits seen in depression may reflect disrupted communication between the hippocampus and cortex (12), such that depressed adults cannot retrieve information as effectively as healthy controls. We investigated this issue by applying time-frequency analysis to event-related potentials (ERPs) collected from adults with Major Depressive Disorder (MDD) and healthy controls during retrieval; the ERPs and extensive behavioral analyses have been described previously (13).

We focused on evoked theta (4–7Hz) power (14) for three reasons. First, hippocampal theta signals are robust in humans (15) and non-human animals (16). Because the hippocampus is critical for encoding and retrieval, this suggests that memory-related signals may be carried at the theta frequency. Second, there is cross-species evidence that communication between the hippocampus and cortex, and between different cortical sectors, involves theta synchrony (17; 18). This may be important because successful retrieval involves the coordinated recovery of information from diverse cortical and subcortical territories. Third, human electroencephalogram (EEG) studies have linked increased theta power to successful retrieval (19–21). This relationship appears particularly strong when recollection is elicited, such as when participants must specify the source of an item (the context in which it was studied) rather than simply indicating that it was previously encountered (22). Tying these strands together, we reasoned that impaired recollection in MDD could reflect disrupted theta signals during retrieval.

To test this hypothesis, we re-analyzed the ERP data from our recent study of source memory (13). During encoding, the participants viewed neutral words shown on the left or right side of a monitor (perceptual source) in the context of animacy or mobility judgments (conceptual source). At retrieval, the words were shown again under cues prompting perceptual retrieval (Side cue; “Was this word shown on the left or right?”) or conceptual retrieval (Question cue: “What judgment did you make for this word?”). On Number trials, a numeral (e.g., “ninety-six”) was shown and the participant indicated whether it was odd or even. Number trials controlled for the sensorimotor demands present on Question and Side trials, but because they required participants to draw on general knowledge rather than their memory for a recent event, the Number trials were intended to elicit retrieval from semantic rather than episodic memory. By contrasting Question and Side trials with Number trials, we hoped to detect brain activity associated with episodic memory. We assumed that such activity would primarily reflect recollection due to the need to retrieve encoding details. However, participants sometimes make accurate source judgments even when recollection fails (23), thus we could not rule out the possibility that such contrasts might also reveal activity associated with familiarity.

Replicating prior studies (24–27), successful source retrieval was associated with positive ERPs over left parietal scalp from 400–800 ms, as well as a long-lasting negative deflection from 800–2000 ms called the late posterior negativity (LPN). The LPN was centered over posterior midline electrodes and extended over left frontal scalp during retrieval of conceptual source memories (13). Surprisingly, however, these contrasts did not reveal group differences, which only emerged when we considered interactions between encoding and retrieval. Briefly, the combination of deep encoding plus conceptual source retrieval enhanced memory accuracy and boosted the amplitude of left centro-parietal ERPs in the depressed adults more than in controls (13). These results are consistent with Hertel’s cognitive initiative framework (28) and the ERPs provided insight into the underlying cognitive operations. Nevertheless, the results were complex and the failure to find stronger group differences was surprising.

We undertook the current analysis to see if time-frequency decomposition would yield additional information. In particular, ERP analyses yield no data about spectral power and thus a negative effect of MDD on theta could be overlooked. To circumvent this problem, we applied Morlet wavelets to the ERPs to assess evoked theta power. We made three predictions. First, we expected reduced theta power in depressed vs. healthy participants during source retrieval. Second, given limited evidence for disrupted semantic memory in depression, we did not anticipate group differences on Number trials. Third, within each group we expected higher theta power during source vs. semantic retrieval, because only source retrieval should involve recollection (20).

Methods and Materials

Participants and self-report

The data were obtained from 24 unmedicated adults who met criteria for current MDD based on the M.I.N.I. (28) and 24 controls with no current or past Axis I psychopathology. In the MDD group, no current or past history of any other DSM-IV Axis I diagnosis was allowed with the exception of generalized anxiety, social anxiety, and specific phobias because these are often comorbid with MDD. Participants were right-handed, reported no current or past neurological conditions, and consented to a protocol approved by the Partners HealthCare Human Research Committee. The groups did not differ on gender (MDD: 15 females, 9 males; controls: 13 females, 11 males; χ² < 1), age (mean±SD, MDD: 29.79±10.85; controls: 30.58±11.32; t < 1), or years of education (MDD: 16.29±2.49; controls: 16.92±2.20; t < 1), all ps > 0.34. Immediately after the EEG session, participants completed the Beck Depression Inventory-II (BDI-II; 29), the Mood and Anxiety Symptom Questionnaire (MASQ; 30), the Ruminative Responses Scale (RRS; 31), and the Pittsburgh Sleep Quality Index (PSQI; 32). The Wechsler Test of Adult Reading (WTAR; 33) was used to estimate IQ. One control did not complete the MASQ, one depressed participant did not complete the PSQI, and WTAR data from non-native English speakers (2 controls, 2 MDD) were not analyzed, as WTAR results may be invalid in such cases.

The results from these measures are shown in Table 1. Based on BDI-II scores, the depressed adults were experiencing moderately severe depressive symptoms. Relative to controls, they reported more anxiety, more rumination, and poorer sleep over the last month. There was no group difference in IQ as estimated by WTAR scores. Two depressed adults met criteria for generalized anxiety disorder in the past six months, seven reported one or more lifetime panic attacks, and a small number reported sub-threshold symptoms of anxiety disorders (social anxiety, n = 2; agoraphobia, n = 2; panic attacks, n = 2).

Table 1.

Mean (SD) Self-Report Data

Variable	Controls n = 24	Depressed n = 24	P
BDI-II	1.29 (2.22)	25.38 (8.69)	< .001
MASQ-GDA	13.04 (2.10)	21.38 (7.04)	< .001
MASQ-AA	17.65 (0.98)	24.00 (8.24)	.001
MASQ-GDD	13.65 (2.08)	38.46 (10.0)	< .001
MASQ-AD	45.61 (12.29)	86.54 (8.74)	< .001
RRS-Dep	17.96 (4.73)	32.96 (4.51)	< .001
RRS-Brood	7.75 (2.38)	12.54 (2.99)	< .001
RRS-Reflect	9.04 (3.80)	12.25 (2.97)	0.002
PSQI*	3.00 (2.00)	8.48 (2.73)	< 0.001
WTAR	116.73 (11.58)	117.09 (7.84)	0.90

Note. BDI-II = Beck Depression Inventory II; MASQ = Mood and Anxiety Symptoms Questionnaire (GDD = General Distress: Depressive symptoms, AD = Anhedonic Depression, GDA = General Distress: Anxious symptoms, AA = Anxious Arousal); RRS = Ruminative Response Scale (Dep = depression subscale, Brood = brooding subscale, Reflect = reflection subscale); PSQI = Pittsburgh Sleep Quality Index; WTAR = Wechsler Test of Adult Reading. Statistics reflect between-group t-tests.

^*PSQI scores <= 5 indicate good sleep quality, scores > 5 indicate poor sleep quality.

Task and stimuli

The task was programmed in PsychoPy (34) and consisted of six encoding-retrieval cycles. Each encoding run involved displaying 16 neutral words on the left or right of fixation for 3500 ms. Words were presented above a question: “Living or non-living?” (animacy task) or “Mobile or immobile?” (mobility task). Participants responded by pressing a button (c or m) on a keyboard; a jittered inter-trial interval (ITI: 500–2000 ms) separated the trials. After encoding, a three-digit number (e.g., “557”) was shown and the participant counted down in steps of three for 30 seconds, to clear working memory (35). No EEG data were collected in these task phases.

Each retrieval run included 48 trials consisting of a cue (1,000 ms), a word (3,000 ms), and a response screen (up to 10,000 ms). The cue was “Question”, “Side”, or “Odd/Even”. On Question and Side trials, the words were drawn from the most recent encoding list and the task was to indicate whether each word had been shown with the animacy or mobility task (Question cue) or on the left vs. right (Side cue). On Number trials (Odd/Even cue), the words were numerals (e.g., “ninety-six”) and the participant had to indicate whether each one was odd or even. The EEG analysis focused on neural activity when the words were onscreen as this was when retrieval should have occurred.

To avoid overlap between EEG signals elicited by memory vs. motor activity, participants could not respond until the response screen appeared, 3000 ms after words were shown. The response screen listed the options for each cue (Question: living/non-living vs. mobile/immobile; Side: left vs. right; Number: odd vs. even) with two levels of confidence (high/low) for each option. Following prior studies (36), a “guess” option was provided should participants be unable to recover information favoring either source. Participants responded by pressing the c and v keys with the first two fingers of their left hands, and the b, n, and m keys with the first three fingers of their right hands. The options remained onscreen until the participant pressed a button or 10 s elapsed. A jittered ITI (500–2000 ms) separated the trials. Trial order was randomized and a fixation cross was presented throughout.

The stimuli were 100 neutral words from the MRC Psycholinguistic Database (37), 25 from each of four categories: “living/immobile” (e.g., elm), “non-living/immobile” (e.g., hill), “living/mobile” (e.g., fox), and “non-living/mobile” (e.g., car) (see 13 for a complete list). Neutral words were used to identify effects of MDD on memory that do not depend on mood congruency. Four words were used in practice trials, with the rest shown in the experiment.

Behavioral analysis

As described previously (13), encoding accuracy was high (≥ 92% correct) for both groups. For simplicity, and to reduce overlap with our prior publication, we do not consider the encoding data further. We cleaned the retrieval data by dropping trials with no response or where log(RT) exceeded the participant’s mean±3SD (< 1% of retrieval trials). Next, we excluded “guess” responses (< 7% of trials) before computing Group x Cue (Question, Side, Number) ANOVAs on accuracy (percent correct), confidence, and correct RT. For RT, ANOVAs were computed on log transformed data, but results are described using untransformed RT data for interpretability. ANOVAs were implemented in the R software (38) library afex (39). Post-hoc Tukey tests were computed with the lsmeans package (40). Alpha was set to 0.05. All tests were two-tailed unless stated otherwise.

EEG recording and analysis

Recording

The EEG was recorded with a 128-sensor HydroCel GSN Electrical Geodesics Inc. (EGI) net (sample rate: 1000 Hz, 0.02–100 Hz). Data were referenced to vertex and impedances were kept below 45 kΩ when possible (maximum: 75 kΩ).

Pre-processing

EEG pre-processing was conducted with the EEGLAB (41) and ERPLAB (42) toolboxes for MATLAB (MathWorks, Natick). EEG data were merged, re-referenced to the average of all electrodes, and filtered (0.1–30 Hz). Bad channels were interpolated, ICA was used to remove activity reflecting blinks, HEOG, and EKG, and the cleaned data were time-locked to word onsets and segmented (−200 to 2000 ms). The pre-stimulus interval was used for baseline correction, and segments where any raw value or the maximum-minimum voltage difference (200 ms intervals, 100 ms sliding window) exceeded 100 μV were rejected. Data from “guess” trials were excluded and there were too few clean segments to analyze misses, thus we focused on correct responses (24; 43–45). There were no group differences in the number of clean segments available (Question: MDD, 49.5±11.6; controls, 47.63±13; Side: MDD, 48.21±11.5; controls, 48.86±14.02; Number: MDD, 73.79±10.43; controls, 69.54±11.42; ps > 0.18). Finally, segments were averaged to form ERPs.

Time-frequency analysis

Spectral decomposition of ERPs was achieved by convolution with complex Morlet wavelets (CMW; 46), which were structured as follows:

CMW = A*exp(−t^2 / (2*s^2) )*exp(2*pi*f *i)

Here, t is time, f is frequency, and i is the square root of −1. s = n/(2*pi*f), where n is the number of cycles in the wavelet. We used n=3 to increase our ability to detect transient activity (46). We set A =1/( sqrt(s)*pi^(¼)*2).

The theta band was divided into four frequencies (4, 5, 6, and 7 Hz). For each center frequency, the complex Morlet wavelet was convolved with each participant’s ERPs. The absolute value of the resulting signal was squared to give the instantaneous power, and results from all frequencies in the theta band were then averaged to give evoked theta power (14).

We took two steps to avoid edge effects. First, we applied a Hanning taper to rapidly dampen the first and last 2.5% of the ERPs. Second, we analyzed data from 0–1600 ms post-stimulus, as edge effects should be minimal in this window (given the 200 ms pre-stimulus baseline and the fact that epochs ended at 2000 ms). The 400–800 ms period is when ERP effects related to recollection are commonly detected (27), and the LPN typically emerges at about 800 ms (47). Thus, this analysis window should capture neural signals related to source memory.

Statistical analysis

The primary analysis consisted of between-groups t-tests on evoked theta power at every electrode on Question, Side, and Number trials. To estimate the false positive rate, we ran 10,000 simulations in which we generated two 24×128 matrices of random values, computed 128 “between-group” t-tests (comparing 24 vs. 24 values at each “electrode”), and then identified the largest cluster of contiguous significant (p < 0.05) electrodes. We saved the largest cluster from each simulation to create a distribution of maximal cluster sizes. The 95^th percentile corresponded to a 3-electrode cluster, so only clusters at least this large were considered significant. In exploratory analyses, we used the same approach to examine between-group differences in evoked delta (0.5–2.5Hz), alpha (8–12Hz), and beta (13–31Hz) power. We do not report on gamma power because the ERPs were low-pass filtered at 30Hz.

In secondary analysis we computed within-group t-tests on theta power for Question vs. Number and Side vs. Number contrasts. These were expected to reveal increased theta power for source vs. semantic retrieval (20). To explore brain-behavior relationships, we computed correlations between Question, Number, and Source accuracy and theta power at every electrode. To estimate the false positive rate for this analysis, we conducted another simulation in which we correlated the controls’ Question accuracy data with a 24×128 matrix of random values, repeated this process 10,000 times, and each time identified the largest cluster of contiguous electrodes that showed a significant (p < 0.05) correlation. This simulation indicated that a 5-electrode extent would be expected by chance 5% of the time, thus a 5-electrode cluster threshold was set for our correlational analysis.

Results

Behavior

The retrieval accuracy, confidence, and RT data are shown in Figure 1 (for mean±SD values, see Table S1). Analysis of these data revealed only main effects of Cue, Fs > 114, ps < 0.001, η_p² values > 0.7; neither the Group effect nor the Group x Cue interaction was significant in any analysis, ps > 0.08. On Number trials, participants were highly accurate, confident, and responded quickly. Question and Side trials were characterized by lower accuracy and confidence, and slower RTs. Responses on Question trials were less accurate and slower than on Side trials, but they were more confidently made. All pairwise comparisons were significant for accuracy, confidence, and RT (ps < 0.001).

Figure 1.

Retrieval accuracy plus confidence and RT on correct trials. Analysis of these data did not reveal group differences, but results are shown separately for depressed and healthy adults to permit visual comparison. Q = Question, S = Side, N = Number. Error bars show SEM.

Group differences in theta power

Figure 2 depicts electrodes that showed significantly stronger theta power in healthy vs. depressed adults for Question (top), Side (middle), and Number (bottom) hits. No electrodes showed greater theta power in depressed adults. Electrodes shown in gray are not in a cluster of at least three electrodes and may be false positives. By contrast, electrodes shown in blue, green, and red formed distinct clusters whose size exceeded the correction for multiple comparisons.

Figure 2.

Electrodes showing a significant (p < 0.05) control > MDD difference in evoked theta power on Question (top), Side (middle), and Number (bottom) trials. Electrodes shown in grey are not part of a cluster of at least three electrodes and so may be false positives. Electrodes shown in the other colors (blue, green, or red) are part of distinct clusters that are sufficiently large to survive correction for multiple comparisons. No electrodes showed a significant MDD > control difference.

Two results are noteworthy. First, the control > MDD difference was evident for source retrieval (Question, Side conditions) but it also extended to semantic retrieval (Number condition). Second, the effect of MDD was clearest in the 400–799 ms window. Specifically, although a group difference was evident in every interval for at least one condition, the 400–799 ms interval was the only one in which the group difference was present in every condition.

Exploratory analysis revealed control > MDD differences—but never the opposite—in the delta (Figure S1), alpha (Figure S2), and beta (Figure S3) bands. The group differences in these bands were qualitatively weaker than in the theta band, however, especially on Side and Number trials.

Condition effects on theta power

In both groups, theta power was greater on Question vs. Number trials (controls, Figure S4; MDD, Figure S5) and on Side vs. Number trials (controls, Figure S6; MDD, Figure S7). These condition effects were broadly distributed and at least one significant cluster appeared in each interval, except for the 0–399 ms interval for the Question vs. Number contrast in controls.

Brain-behavior correlations

Figure 3A shows that, in controls, source memory accuracy was positively correlated with theta power at 17 electrodes during the 400–799 ms interval. Side accuracy was correlated with 2 electrodes over left frontal scalp, and Question accuracy was correlated with 2 electrodes over right central scalp; these clusters did not meet the 5-electrode threshold. By contrast, 13 electrodes were bunched over the parietal midline. In this region, clusters of 10 and 8 electrodes showed correlations with accuracy in the Question (Figure 3; red) and Side (Figure 3; blue) conditions, respectively; 5 electrodes showed correlations with both conditions (Figure 3; purple). Of the 13 electrodes, 6 showed a significantly stronger correlation in healthy vs. depressed adults (Figure 3; larger circles). Moreover, as shown in Figure 3B, this cluster was characterized by a Group x Cue interaction, F(1.81, 83.38) = 5.37, p = 0.008, η_p² = 0.10, driven by group differences on Question (p < 0.03) and Side (p < 0.001) trials, but not Number (p = 0.79) trials, from 400–799 ms. No significant clusters emerged in the other intervals in controls, or in any interval in the MDD group. Number accuracy was not significantly correlated with any electrode clusters in either group, in any interval.

Figure 3.

(A) Topography shows electrodes in which evoked theta power was positively correlated with accuracy on Question (red) trials, Side (blue) trials, or both (purple) in the healthy controls; no positive correlations were seen in the MDD group. Enlarged electrodes indicate sites at which the strength of the theta/accuracy correlation differed significantly between groups by Fisher’s Z-test (ps < 0.05). (B) At the 13 electrodes over the parietal midline (circled in panel A), theta power was significantly higher in controls vs. depressed adults on Question and Side trials, but not Number trials (Group x Cue interaction, p = 0.008). (C) In controls, average theta power over the 13 parietal midline electrodes (circled in panel A) was significantly positively correlated with accuracy on Question (left) and Side (right) trials. No such correlations were observed in the MDD group.

Figure 3C shows correlations averaged over all 13 electrodes. Theta power was positively related to accuracy in healthy but not depressed adults on Question (controls: r = 0.52, p = 0.01; MDD: r = −0.11, p = 0.62) and Side (controls: r = 0.48, p = 0.02; MDD: r = 0.00, p = 1.0) trials. These correlations differed significantly by one-tailed Z-test (Question: Z = 2.24, p = 0.013; Side: Z = 1.67, p = 0.047). As an alternative approach to these data, we regressed average theta power at the 13 posterior midline electrodes on Condition (Question, Side), Group, and Accuracy. This returned an effect of Accuracy (β = 0.54, p < 0.001) and a Group x Accuracy interaction (β = 0.04, p = 0.024), because accuracy on Question and Side trials was positively related to theta power in healthy but not depressed adults.

Discussion

This analysis yielded four findings. First, evoked theta power was broadly reduced in depressed vs. healthy adults during retrieval of source but also semantic memories, with this effect being strongest from 400–799 ms post-stimulus. Second, in both groups theta power was higher during source vs. semantic retrieval. Third, theta power over the parietal midline was higher during source retrieval—but not semantic retrieval—in healthy vs. depressed adults from 400–799 ms. Fourth, source accuracy was positively correlated with parietal midline theta power from 400–799 ms in controls, but not depressed adults. Overall, these data point to disrupted theta signals as a possible mechanism underlying deficient recollection in MDD.

It is unclear why depression was associated with reduced theta power, but disrupted sleep may play a role. In non-human animals, sleep deprivation decreases the excitability of hippocampal neurons (48), disrupts long-term potentiation (49), and reduces the generation of the hippocampal theta rhythm (50). In this study, depressed adults reported poorer sleep over the past month than did the controls. Thus, we speculate that sleep deprivation may have contributed to reduced theta power in MDD.

Midline parietal electrodes were characterized by group differences in theta power on Question and Side (but not Number) trials, and in the strength of correlations between theta power and source memory accuracy. These results ar econsistent with the fact that the parietal memory network (PMN; 51) makes critical contributions to episodic retrieval (24; 52–55). For example, intervention studies have established a causal role for the PMN in retrieval success, which appears to reflect its connectivity with the hippocampus. When the PMN is stimulated, functional connectivity with the hippocampus increases and recollection precision improves (56; 57). By contrast, inhibition of the PMN, especially the precuneus, impairs memory performance (58; 59). We cannot be certain of the generators of our scalp-recorded effects. Nonetheless, a prior study used a similar paradigm in conjunction with functional magnetic resonance imaging and magnetoencephalography to show that successful source retrieval strongly recruited the PMN, especially the precuneus (24). In short, one would expect memory-related abnormalities in theta to be evident over parietal midline electrodes, as was observed in the MDD group; this may reflect dysfunction of the PMN.

This interpretation raises questions. For instance, if MDD affects episodic but not semantic memory, why was a group difference in theta power observed on Number trials? This result was unexpected and so our interpretation is speculative, but two considerations are important. First, our characterization of the Number trials may be flawed. We conceptualized the Number condition as requiring semantic retrieval because judging whether numerals are odd or even depends on general knowledge rather than memory for a specific event. However, the Number trials involved numerical cognition and that is atypical of semantic memory tasks. Numerical cognition depends on parietal cortex (60) and can induce fronto-parietal theta synchrony (61). Consequently, the group difference may have extended to Number trials because these trials drove a stronger theta response than a more standard semantic memory task would have. Second, MDD may involve a trait-like reduction in theta power that can be observed even when evoked theta signals are relatively weak. Indeed, empirical studies (62) and a recent review (63) emphasize that resting-state EEG data show reduced parieto-occipital theta power in depressed vs. healthy adults. Therefore, it may be possible to detect reduced parieto-occipital theta power in depressed adults even if evoked theta power is weak (e.g., during semantic retrieval).

A second question is: if theta is important for memory, and if theta power was reduced in MDD, then why were there no group differences in behavior? The answer to this question is unknown. Other researchers have suggested that increased functional connectivity may help compensate for reduced EEG power in MDD (64), but whether such an account applies in this case is unclear. Thus, better characterizing the relationships between behavioral and EEG markers of memory retrieval in MDD remains an important goal.

Finally, it is interesting that the strongest effects consistently emerged from 400–799 ms. This is in line with prior ERP research (27), and with an intracranial study that found greater high gamma power for hits vs. correct rejections in the left intraparietal sulcus from 300–700 ms (65). Intracranial recordings from hippocampus indicate that signals distinguishing old vs. new items emerge at 250 ms post-stimulus (66). The PMN may accumulate such memory signals, sent from the hippocampus, in order to make memory-based decisions. Therefore, it may be that the temporal difference in the onset of memory signals across these two regions (~250 ms for hippocampus vs. ~300–400 ms for parietal cortex) reflects conduction delay between them (65).

It is important to acknowledge limitations. First, the design did not include new items. Therefore, while we assume that our paradigm elicited recollection due to the need to retrieve encoding details, we cannot rule out the possibility that familiarity may have contributed to our results. Second, low trial counts precluded examination of misses and comparisons of retrieval success vs. failure. Third, the lack of group differences in behavior was unexpected. In our prior report, we found relatively subtle effects of MDD by examining the retrieval of words from each encoding task separately (13). Using emotional stimuli to drive stronger group differences in behavior that can then be related to EEG dynamics is a priority.

In conclusion, theta power was reduced in depressed adults during retrieval of source but also semantic memories. This effect may reflect PMN dysfunction. Given the centrality of memory to cognitive function and the enormous burden placed on society by depression, we propose that future studies continue testing the hypothesis that abnormal theta signaling contributes to memory deficits in MDD.