Background: Negative neurocognitive bias is a core feature of depression that is reversed by antidepressant drug treatment. However, it is unclear whether modulation of neurocognitive bias is a common mechanism of distinct biological treatments. This randomized controlled functional magnetic resonance imaging study explored the effects of a single electroconvulsive therapy session on self-referent emotional processing. Methods: Twenty-nine patients with treatment-resistant major depressive disorder were randomized to one active or sham electroconvulsive therapy session at the beginning of their electroconvulsive therapy course in a double-blind, between- groups design. The following day, patients were given a self-referential emotional word categorization test and a free recall test. This was followed by an incidental word recognition task during whole-brain functional magnetic resonance imaging at 3T. Mood was assessed at baseline, on the functional magnetic resonance imaging day, and after 6 electroconvulsive therapy sessions. Data were complete and analyzed for 25 patients (electroconvulsive therapy: n = 14, sham: n = 11). The functional magnetic resonance imaging data were analyzed using the FMRIB Software Library randomize algorithm, and the Threshold- Free Cluster Enhancement method was used to identify significant clusters (corrected at P < .05). Received: June 15, 2017; Revised: August 18, 2017; Accepted: September 29, 2017 © The Author(s) 2017. Published by Oxford University Press on behalf of CINP. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http:// creativecommons.org/licenses/by-nc/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, 226 provided the original work is properly cited. For commercial re-use, please contact email@example.com Downloaded from https://academic.oup.com/ijnp/article-abstract/21/3/226/4372287 by Ed 'DeepDyve' Gillespie user on 16 March 2018 Miskowiak et al. | 227 Significance Statement Negative neurocognitive bias is a core feature of depression that is reversed by antidepressant drug treatment. This randomized, double-blind, sham-controlled fMRI study shows for the first time that a single ECT session influences neuronal response in the prefrontal cortex during retrieval of emotional self-referent words. The effect may reflect increased memory efficiency for positive self-referent information and highlights modulation of negative neurocognitive bias as a putative common mechanism of distinct biological treatments for depression. Results: A single electroconvulsive therapy session had no effect on hippocampal activity during retrieval of emotional words. However, electroconvulsive therapy reduced the retrieval-specific neural response for positive words in the left frontopolar cortex. This effect occurred in the absence of differences between groups in behavioral performance or mood symptoms. Conclusions: The observed effect of electroconvulsive therapy on prefrontal response may reflect early facilitation of memory for positive self-referent information, which could contribute to improvements in depressive symptoms including feelings of self-worth with repeated treatments. Keywords: ECT, fMRI, depression, self-referent memory, emotional bias Introduction According to the cognitive theory of depression, negative bias Indeed, modulation of negative bias in the processing and in the cognitive processing of emotional information is a cen- recall of self-referent information may play a key role in the tral component in the development and maintenance of depres- clinical efficacy of antidepressant drugs (Harmer and Cowen, sion (Beck and Alford, 2014) and increases the risk of depressive 2013). Specifically, a single dose of the selective noradrenaline relapse (Bouhuys et al., 1999Mathe ; ws and MacLeod, 2005). This reuptake inhibitor reboxetine reverses negative bias in the negative bias is particularly evident in the processing of person- categorization and recall of self-referent words in depressed ally relevant information such as self-referent emotional words patients prior to observable changes in mood symptoms (Harmer (Miskowiak and Carvalho, 2014). Specifically, patients with major et al., 2009). Similarly, a single dose of reboxetine facilitated rec- depressive disorder (MDD) display slowed responses to and ognition of positive vs negative self-referent words in healthy impaired recall of positive self-referent words and increased individuals, which was accompanied by reduced neuronal acti- memory for negative compared with positive and neutral words vation in a fronto-parietal network during recognition of posi- during acute episodes (Bradley and Mathews, 1983; Watkins tive target words vs matched distractors (Miskowiak et al., 2007). et al., 1992; Neshat-Doost et al., 1998Dozois and Dobson, ; 2001; Finally, short-term treatment with citalopram reduced ventro- Harmer et al., 2009). Functional magnetic resonance imaging medial prefrontal response to negative self-referential words (fMRI) studies show that this negative bias is accompanied by in a population of individuals at risk for depression, indicated hyper-activity in the medial prefrontal cortex, rostral anterior by high neuroticism (Di Simplicio et al., 2012). Interestingly, a cingulate cortex, and insula during encoding of negative self- case report of vagus nerve stimulation in a depressed patient referent words and hippocampal hypo-activity during encoding showed treatment-related modulation of ventral prefrontal cor - of positive words (van Tol et al., 2012 Y ; oshimura et al., 2010). tex response during encoding of negative self-referent words, Studies of whether negative self-referent memory bias per - which was accompanied by reduced recognition of negative sists after clinical remission have produced equivocal results. (but not positive or neutral) words (Critchley et al., 2007). Aside In one study, remitted patients endorsed and recalled more from this single case report study, it is unknown whether other negative self-referent words compared with healthy controls biological treatments for depression produce similar effects on (Romero et al., 2014). However, other studies showed either no self-referent emotional memory that could be mechanistically negative bias for self-referent words (Dobson and Shaw, 1987) important for their therapeutic effects. or a necessity for induction of sad mood state to detect reduced Electroconvulsive therapy (ECT) is the most effective and recall of positive self-referent words in remitted patients (Ramel fast-acting treatment for severe depression (The UK ECT Review et al., 2007). We recently found that monozygotic and dizygotic Group, 2003). Putative neurobiological mechanisms of ECT are twins at familial risk of depression display subtle negative recall increase in brain-derived neurotrophic factor, neurogenesis bias for self-referent emotional words (more false recollections (Madsen et al., 2000; Altar et al., 2004 Bol ; wig and Madsen, 2007), of negative words) than low-risk twins (unpublished observa- and hippocampal volume (Jorgensen et al., 2016). However, it is tions). In contrast, other studies observed no negative self- unclear whether ECT modulates neural and behavioral response referent memory bias in unaffected individuals at familial risk to emotional information in the same manner as antidepressant of depression (Mannie et al., 2007; Wolfensberger et al., 2008). pharmacotherapy. We recently demonstrated in a randomized, Nevertheless, aberrant neural response in the inferior frontal sham-controlled, double-blind study that a single ECT session gyrus was detected in these at-risk individuals during encoding modulates the neural (but not behavioral) response to emo- and recognition of negative words (Wolfensberger et al., 2008). tional faces in patients with MDD (Miskowiak et al., 2017). The The consistent evidence for negative self-referent memory bias present study of the same patient cohort aimed to investigate in symptomatic patients and less consistent findings in remit- whether a single ECT vs sham session modulates neuronal and ted patients and at-risk individuals point to negative memory cognitive measures of memory for self-referent emotional infor - bias as a state marker of depression that is at least partially mation in patients with MDD. We hypothesized that ECT would resolved after successful treatment. facilitate the retrieval of positive vs negative self-referent words, Downloaded from https://academic.oup.com/ijnp/article-abstract/21/3/226/4372287 by Ed 'DeepDyve' Gillespie user on 16 March 2018 228 | International Journal of Neuropsychopharmacology, 2018 as indicated by: (1) reduced neural response in frontal and par - One day after the initial ECT/sham session (day 1), patients ietal regions during recognition of positive words vs matched attended a test session of approximately 2 hours at the Danish distractor words (i.e., greater efficiency of memory retrieval Research Center for Magnetic Resonance. During this time, for positive words) (primary outcome), as well as (2) increased they performed a set of computerized emotional processing retrieval-specific hippocampal (but not amygdala) (Miskowiak tests from the Emotional Test Battery (Oxford, P1 Vital) before et al., 2007) response to positive than negative words, and (3) and during fMRI and a nonemotional verbal memory test after improved recall of positive vs negative words (secondary out- the scan. For clarity, this report includes only the results of the comes). These effects were expected to occur prior to any dif- self-referent emotional word categorization, recall, and recog- ferential change between ECT and sham groups in depressive nition tests from the Emotional Test Battery and the nonemo- symptoms or subjective state. tional verbal memory test. Mood symptoms and subjective state were assessed using the HDRS-17 and relevant questionnaires. From day 3 and onwards, all patients received active ECT treat- Methods ment according to the standard protocol of the Capital Region (3 times/wk). For these subsequent ECT sessions, a titration Study Design and Patients based on seizure quality (configuration and length of the EEG seizure that should exceed 25 seconds) was applied with 50% Patients scheduled for ECT were recruited from Psychiatric dosage increase in case of threshold seizure. Mood symptoms Centre Copenhagen (Rigshospitalet and Bispebjerg) from were rated with the HDRS-17 after 6 (active) ECT sessions. End of November 2009 to July 2015. Personnel at the respective depart- treatment was decided by the treating psychiatrist. ments carried out eligibility assessments in terms of a diagno- sis of depression and established degree of depressive symptom severity using the Hamilton Depression Rating Scale 17-items Neurocognitive Tests (HDRS-17) (Hamilton, 1960). The inclusion criterion was current Patients were given a self-referent emotional categorization moderate to severe depression, defined by a HDRS-score ≥18. task followed by a free recall tests before entering the scanner Exclusion criteria were current substance- or alcohol misuse (Anderson, 1968). A total of 90 unambiguously negative and posi- disorders, neurological disorders, bipolar depression, schizo- tive personality trait words from the Anderson’s list of personal- typal disorder or schizophrenia, pregnancy, major somatic ill- ity trait words (Anderson, 1968), matched for length, frequency, ness contraindicating ECT, and (for fMRI scanning) having a and meaningfulness, were displayed on the screen of a laptop pacemaker or other metal implants in the body. The study was computer. The words were presented one at a time for 500 milli- approved by the local ethics committee (H-3-2009-074) and the seconds in a random order with each word occurring only once. Danish Data Protection Agency (2009-41-3676) and was carried Patients were instructed to press 1 of the 2 keys on a keyboard out in accordance with the Declaration of Helsinki. All patients as quickly and accurately as possible to categorize the words as provided written informed consent. likeable or unlikeable in a self-referential manner. Specifically, they were asked to imagine that they overheard someone talk- Randomization and Masking ing about them and decide if they would be happy or sad if these words were used to describe them. Accuracy and response time Block randomization was performed by Pharma Consulting were recorded using Superlab software. Fifteen minutes later, Group (Uppsala, Sweden) with stratification for age (< patients were given a free recall test, which involved recall of as or ≥ 45 years) and gender. Pharma Consulting Group created a many positive and negative words from the task as possible for randomization list and envelopes with information about treat- a maximum period of 5 minutes. The total numbers of correctly ment allocation (ECT/sham) for each patient ID number. Along recalled positive and negative words and of false recollections of with date of birth and gender, treatment allocation was assigned positive and negative words (memory intrusions) were recorded. to each patient upon inclusion. Numbering was consecutive, and Inside the scanner, patients were given an incidental emo- the ID number was recorded in each patient’s individual file. The tional recognition task. A total of 180 words was projected onto randomization list was kept in a locked cabinet in the ECT room, an opaque screen viewed by patients through angled mirrors. Of to which only personnel involved in ECT/sham treatment had these, 90 words had been presented in the categorization test access. Patients, personnel, and outcome assessors were blinded (old words) and 90 were new matched distractors (45 positive to group assignment and blinding was maintained throughout and 45 negative) from the Anderson’s list (new words). Patients the study, data management, and outcome assessment. were instructed to indicate as quickly and accurately as possible whether the words displayed were old or new by pressing corre- Procedure sponding keys on a key pad. The paradigm had an event-related Baseline mood ratings were performed during the screen- design with each trial consisting of a fixation cross shown ing interview (1–4 days prior to the first ECT or sham session). for 500 ms followed by a personality trait words displayed for Patients were then randomized to either a single ECT or sham 500 ms. Words were presented in a random order with an inter- treatment at the beginning of their ECT series. For this first ECT/ trial interval between 4000 and 9000 milliseconds, resulting in a sham treatment, electrodes were placed bilaterally and patients total task time of 12 minutes. Accuracy and response times for were put under full anesthesia using thiopental and given suc- correctly recognized words and misclassifications were recorded cinylcholine for muscle relaxation. After this, an envelope with with e-prime software version 1.2 (Psychology Software Tools information about the particular patient’s treatment allocation Inv.). In addition, a visual stimulation control task (with minimal was opened and either active or sham ECT was administered cognitive demands) was implemented to investigate whether accordingly with a Thymatron ECT Machine. The ECT pulse any potential effects of ECT on recognition-related neural activ- width range was 0.5 to 1.0 milliseconds and the initial dose ity were confounded by global changes in cerebral blood flow. based on the patient’s age was used (charge [per cent of 500 mil- A flashing checkerboard (frequency of 8 Hz) was presented licoulomb] = 50% of the age). in blocks of 14 seconds interleaved by a 14-second fixation cross Downloaded from https://academic.oup.com/ijnp/article-abstract/21/3/226/4372287 by Ed 'DeepDyve' Gillespie user on 16 March 2018 Miskowiak et al. | 229 for a total of 6 cycles. Patients were instructed to keep their eyes were performed with the Statistical Package for Social Sciences open during this time. (SPSS) (version 22.0) (IBM Corporation). After the scan, patients were given the Rey Auditory Verbal Learning Test (RAVLT) (Schmidt, 1996) for assessment of non- fMRI Data Analysis emotional verbal memory. Total recall across trials I to V, imme- Functional MRI data processing was performed with the FMRI diate recall following interference, delayed free recall (after 30 Expert Analysis Tool (version 6.00) part of FMRIB’s Software minutes), and recognition were recorded. Library (www.fmrib.ox.ac.uk/fsl). Preprocessing involved image realignment, non-brain removal, spatial normalization to an Mood and Subjective State MNI (Montreal Neurologic Institute) template, and spatially Depressive symptoms were assessed with the HDRS-17 and the smoothing (Gaussian kernel, 5 mm full-width-half-maximum). Beck Depression Inventory (BDI) (Beck et al., 1961). Anxiety and The time series in each session were high pass-filtered (to max subjective state were assessed at the time of fMRI scanning with 0.008 Hz). the State Trait Anxiety Inventory (STAI) (Spielberger, 1983) and Events of interest were defined according to whether or not visual analogues scales (VAS) for happiness, sadness, alertness, the word was presented in the previous categorization task anxiety, dizziness and nausea. (old vs new), whether the word was positive or negative, and whether the word was correctly identified as old (hit) or new (correct rejection). At the individual subject level, we there- MRI fore set up a general linear model (GLM) that included 8 event MRI data were collected with a 3 T Siemens Trio MR scanner types: (1) positive/hit, (2) positive/correct rejection, (3) nega- using an 8-channel head array coil. Blood oxygen-level depend- tive/hit, (4) negative/correct rejection, (5) incorrect positive/hit ent (BOLD)-sensitive fMRI used a T2*-weighted gradient echo (i.e., incorrect identification of new positive words as “old”), (6) spiral echo-planar imaging sequence with an echo time of 30 incorrect negative/hit, (7) incorrect positive/rejection (incorrect milliseconds, repetition time of 2.49 ms, and a low flip angle of identification of old positive words as “new”), and (8) incorrect 20° to minimize physiological noise (Gonzalez-Castillo et al., negative/rejection. The events were convolved with a canonical 2011). A total of 128 brain volumes were acquired in a single hemodynamic response function (Boynton et al., 1996), and the fMRI session, each consisting of 42 slices with a slice thickness GLM model included local autocorrelation correction (Woolrich of 3 mm and a field of view of 192 × 192 mm using a 64 × 64 grid. et al., 2001). High-resolution 3D structural T1-weighted spin echo images To examine our hypothesis that ECT would influence neural were obtained after the first session of BOLD fMRI (TI= 800, echo activity in areas involved in retrieval-specific processes for posi- time = 3.93, repetition time = 1540 ms, flip angle 9°; 256 × 256 tive and negative words rather than valence per se, we defined field of view; 192 slices). 3 contrasts of interest: (1) positive/hits vs positive/correct rejec- tions and (2) negative/hits vs negative/correct rejections (i.e., treatment by task interactions). Finally, to investigate the effects Statistical Analysis of Behavioral Data and Mood of ECT on neural activity on valence-specific memory processes, Symptoms we contrasted (3) negative/hits with positive/hits. At the group Behavioral data from the emotional categorization task (speed level, data for the 3 contrasts of interest were included in sep- and accuracy), free recall (total recall and memory intru- arate GLM models estimated using nonparametric permuta- sions), and recognition (speed, accuracy and response bias) tion-based inference (n= 5000) using the randomize algorithm were analyzed using repeated measures ANCOVA with valence implemented in FMRIB’s Software Library (Winkler et al., 2014). (positive, negative) as the within-subjects factor and treat- We first investigated the primary hypothesis that ECT ment (ECT, sham) as the between-subjects factor and adjust- would modulate the response in fronto-parietal regions dur - ment for potential differences in baseline characteristics. For ing retrieval of positive words, specifically. We therefore defined the emotional recognition task, we used signal detection the- a volume of interest (VOI) mask on a standard MNI template ory to obtain a measure of memory accuracy corrected for the that included the bilateral prefrontal and parietal cortical maps patients’ response tendency. The proportion of correctly rec- provided by the Harvard-Oxford cortical structural Atlas thresh- ognized words and of falsely recognized words constitute the olded at 5%. The prefrontal cortex included the cortical regions parametric sensitivity measure: d’ = (number of hits+ 0.5/num- anterior to the precentral sulcus, that is, superior, middle, and ber of targets + 1) - (number of false alarms + 0.5/n umber of dis- inferior frontal gyri; the frontal medial cortex and subgenual tractors + 1). In addition, response bias was computed according cortices; and the frontal poles. The parietal regions included the to false-alarm scores: (number of false alarms + 0.5/n umber of precuneus and superior parietal lobe. We then investigated the distractors + 1)/(1 - Pr). This measure reflects the tendency of secondary hypothesis that ECT would increase hippocampal participants, when uncertain about the category to which a response during retrieval of positive relative to negative words. word should belong, to categorize the word as old rather than Bilateral hippocampal formations were therefore defined using new. We applied Sidak correction for the number of statistical maps included in the Harvard-Oxford Subcortical Structural comparisons across the behavioral tests (n = 6), leading to an Atlas and thresholded at 5%. The statistical inference at group adjusted threshold for significance of P = .009. Significant inter- level was restricted within the defined VOIs. We further carried actions were followed up by independent samples tests for t out a whole-brain analysis to explore possible effects of ECT in normally distributed data or Mann-Whitney U tests for nonnor - the 3 contrasts of interest in other brain regions. mally distributed data. Performance on the RAVLT, mood symp- Significant clusters were identified using the Threshold-Free toms (HDRS-17 and BDI scores), and anxiety (STAI scores) were Cluster Enhancement method at corrected P < .05. Upon signifi- analyzed with independent samples tests. t Finally, VAS ratings cant findings, we first conducted a posthoc analysis to explore of subjective state were examined with repeated-measures whether the effects of the ECT intervention were dependent ANOVA. Analyses of behavioral data, mood, and subjective state on the individual task performance. To control for the effect of Downloaded from https://academic.oup.com/ijnp/article-abstract/21/3/226/4372287 by Ed 'DeepDyve' Gillespie user on 16 March 2018 230 | International Journal of Neuropsychopharmacology, 2018 performance, we set up an analogue GLM model and added an n = 14, sham: n = 11) (see Figure 1 for CONSORT flow diagram). accuracy covariate calculated as the individual average correct Self-reported mood and subjective state ratings were missing recognitions across positive and negative old and new words. for 5 patients (for details, see Table 1), RAVLT data were miss- Secondly, we explored whether the mean percent signal change ing for one patient (sham), data from the emotional categor - in the region responding to the ECT intervention was correlated ization task were lost for one participant (sham), and data with the speed or accuracy in the recognition for positive words, from the emotional recognition were missing for one partici- or change in mood symptoms (HRDS-17 scores) from baseline to pant (sham) (see details in Table 3). the sixth ECT session. Patients with very low recognition accur - The ECT and sham groups were comparable for gender com- acy, defined a priori as ≥2 SD below from the group mean were position, age, indexes of illness chronicity, and medication sta- excluded from fMRI analyses. tus (P ≥ .3). However, ECT-treated patients had somewhat shorter education than patients allocated to sham treatment (mean ± SD: ECT: 14 ± 2; sham: 16 ± 2; t = 2.2, df = 23, P = .04) (Table 1). There Results were no differences between the 2 groups in depressive symp- tom severity at baseline or on the day of the scan (P ≥ .2). There Participant Characteristics and Mood were also no differences between the 2 groups in subjective state or state anxiety on the day of the scan ( ≥ P 0.6). There was Of the 29 patients randomized to either ECT or sham treat- a nonsignificant trend towards lower trait anxiety in ECT-treated ment (ECT: n = 16, sham: n = 13), one participant (ECT) fell vs sham-treated patients (P = .06). As expected, all patients dis- asleep during the fMRI scan, the image quality for the struc- played substantial reduction in depressive symptom severity tural scan was inadequate for another participant (sham), and from baseline to the assessment following 6 active ECT sessions one patient (ECT) had a task performance ≥2 SD below the (mean score ± SD after 6 ECT sessions: HDRS-17: 15± 7; BDI: 14 ± 7, group mean and was therefore excluded from the fMRI ana- P < .01). Of the 25 included patients, 11 (44%) showed treatment lysis. Two patients failed to complete the emotional recogni- response (i.e., a ≥50% symptom reduction from baseline), with 5 tion task (ECT: n= 1, sham: n = 1). Finally, data for one patient (20%) achieving clinical remission (i.e., HDRS-17 ≤ 7) after 6 active (ECT) from the control visual stimulation paradigm was lost ECTs, which is somewhat lower than the response and remission due to technical difficulties, but the patient’s fMRI data from rates after complete ECT series in unipolar depression (Dierckx the recognition paradigm were intact and included in the et al., 2012). analyses. Data were therefore analyzed for 25 patients (ECT: Figure 1. CONSORT flow diagram. Downloaded from https://academic.oup.com/ijnp/article-abstract/21/3/226/4372287 by Ed 'DeepDyve' Gillespie user on 16 March 2018 Miskowiak et al. | 231 Table 1. Demographic and Clinical Characteristics at Baseline vs negatives words across the entire cohort (F(1,22) = 14.4, P < .01; posthoc Wilcoxon Signed Ranks: Z= -3.0, P < .01) but showed no Sham ECT difference between the 2 groups (P ≥ .3). (n = 11) (n = 14) P value Finally, there were no differences between the treatment groups in nonemotional memory as reflected by RAVLT total Gender, n/% males 7/63 8/53 .7 recall, recall following a distractor list, delayed recall, and rec- Age, median (IQR) 37 (32) 40 (29) .9 ognition (all P ≥ .7). Age at onset, median (IQR) 22 (24) 18 (30) .5 First hospitalization, median (IQR) 24 (30) 39 (30) .9 Number of episodes, median (IQR) 5 (8) 9 (20) .6 fMRI Results Education (yrs.), mean (SD) 16 (2) 14 (2) .04 Effects of ECT vs Sham Treatment STAI-trait, mean (SD) 63 (9) 52 (13) .06 A single ECT session reduced the retrieval-specific neural Medications Antidepressants, n (%) 8 (73) 11 (85) .5 response to positive words compared with sham in the left fron- Antipsychotics, n (%) 4 (36) 5 (39) .9 topolar cortex within the defined VOI (for peak cluster activa- Anticonvulsants, n (%) 1 (9) 1 (8) .9 tion, see Table 2; for visualization of the frontopolar cluster, see Benzodiazepines, n (%) 4 (36) 6 (46) .6 Figure 2). In contrast with our hypothesis, no effects of ECT vs Lithium, n (%) 1 (9) 0 (0) .3 sham treatment were observed in retrieval-specific hippocam- Thyidea, n (%) 0 (0) 1 (8) .3 pal activity for positive or negative words or in valence-specific Combined prescriptions, n 1.6 1.8 .6 activity to old negative vs positive words. Further, no differences Baseline were observed between the treatment groups in retrieval-spe- HDRS-17, mean (SD) 28 (6) 26 (5) .6 cific response to negative words or in valence-specific activity BDI, median (IQR) 25 (20) 28 (10) .5 to old positive vs old negative words. Exploratory whole brain Day 1 analyses also revealed no effect of ECT vs sham in other brain HDRS-17, mean (SD) 26 (7) 23 (5) .2 regions. BDI, median (IQR) 17 (17) 25 (10) .2 The visual stimulation paradigm revealed no differential STAI-state, median (IQR) 59 (23) 51 (21) .7 BOLD response to photic stimulation in the occipital cortex VAS .6 between ECT and sham groups (P = .8). Happiness, median (IQR) 1 (3) 2 (4) - Sadness, mean (SD) 7 (2) 7 (3) - Alertness, median (IQR) 4 (5) 5 (6) - Task-Related Activations Across All Participants Anxiety, mean (SD) 5 (3) 4 (3) - A broad distributed neural network including several bilateral Dizziness, median (IQR) 2 (4) 0 (2) - prefrontal and parietal regions was activated during recognition Nausea, median (IQR) 1 (3) 0 (1) - of old (positive and negative) words vs matched distractor words Abbreviations: IQP, interquartile range; SD, standard deviation; STAI, State- across all patients (for peak cluster activations, see Table 2). Trait Anxiety Inventory; Yrs, years. Further, contrasting old positive target words with matched distractors revealed significant retrieval-specific activity in a Behavioral Data distributed occipito-parietal network (peak cluster activations Across the entire cohort, patients showed greater accuracy dur - in Table 2). However, the hippocampi were not significantly ing categorization of negative vs positive words (F(1,22) = 5.2, activated during retrieval of old positive or negative words vs P = .03) in the absence of differences in speed (P ≥ .6) (Table 3). new distractor words (retrieval-specific activity) and showed no However, accuracy and speed in the categorization of positive differential activation to negative vs positive words across the vs negative words did not differ between the ECT and sham entire cohort. groups (P≥ 0.5). In the free recall test, patients generally recalled an equal number of positive vs negative words, and this did Exploratory Correlation Analyses not differ between the 2 treatment groups (P ≥ 0.2). There was a Exploratory postdoc correlation analyses showed no correlation trend towards more positive than negative false recollections in between retrieval-specific left frontopolar activity for positive general (F(1,21) = 3.9, P = .06), but this did not differ between the words and recognition speed for positive words (≥ P .2). There was groups (P ≥ 0.5). a trend toward a correlation between higher frontopolar activity Across all patients, there was greater recognition accuracy and greater accuracy across all participants (P = .08), suggesting for positive vs negative words (F (1,22) = 7.4, P = .01; posthoc that the region may play a role in retrieval success. Across the t test: t = 2.6, df = 23, P = .02), but there was no difference between entire cohort, there was no association between retrieval-spe- the groups (P≥ .03). Analysis of recognition speed including all cific frontopolar activity for positive words and patients’ mood patients demonstrated faster response times for positive vs improvement from baseline to their sixth ECT session ( ≥P .7). negative words (F(1,22) = 6.5, P = .02; posthoc t test: t = -2.3,df = 23, P = .03). At the unadjusted statistical threshold for significance, there was a trend towards an interaction between group and Discussion speed during emotional word recognition indicating faster response times to both positive and negative words in the sham This is the first randomized sham-controlled fMRI-study of the effects of a single ECT session on self-referent emotional vs ECT groups (F(1,22) = 3.8, P = .06; t tests nonsignificant; P ≥ .4). However, given our a priori Sidak correction for multiple test- memory in depression. We found that a single ECT session reduced the neural activity in the left frontopolar cortex dur - ing (adjusted threshold: P < .009), this finding rendered nonsig- nificant. Discrimination accuracy (d-prime) was unaffected by ing retrieval of positive self-referent words. The effect occurred in the absence of treatment effects on memory-relevant hip- valence and revealed no differences between ECT and sham groups (P≥ .2). Response bias was generally higher for positive pocampal response. There were no differences between groups Downloaded from https://academic.oup.com/ijnp/article-abstract/21/3/226/4372287 by Ed 'DeepDyve' Gillespie user on 16 March 2018 232 | International Journal of Neuropsychopharmacology, 2018 Table 2. Peak Cluster Activations for Main Effects and Between-Group Differences in Brain Regions during the Recognition Trial MNI P value Number of voxels X Y Z Positive/hits vs positive/correct rejections Sham>ECT Left middle frontal gyrus (BA 10) <.01 1 065 -40 58 0 Left inferior frontal gyrus (BA 45) .01 25 -58 22 6 Main effect of task Right precuneus (BA 7) <.01 797 4 -64 44 Left superior parietal lobe (BA 7) .03 43 -22 -76 52 Left cuneus (BA 18) .04 39 -10 -90 14 Left superior occipital gyrus (BA 18) .04 19 -20 -86 20 All correct hits (old target words) >baseline Left inferior parietal lobe (BA 7/40) .015 96 -40 -32 44 Right inferior occipital gyrus (BA 18) <.01 667 24 -86 -10 Left middle occipital gyrus (BA 18) <.01 239 -18 -92 4 Left inferior occipital gyrus (BA 19) .01 115 -44 -70 -8 Left fusiform gyrus (BA 19) .04 28 -28 -68 -16 Right fusiform gyrus (BA 37) .03 28 30 -54 -20 Abbreviations: BA, Broadmann area; MNI, Montreal Neurological Institute; PFC, prefrontal cortex; PPC, prefrontal parietal cortex; VOI, volume of Interest. Within the PFC+ PPC VOI analysis. Figure 2. Brain images: retrieval-specific response in the left frontopolar cortex for positive words (old positive minus new positive words) in electroconvulsive therapy (ECT) and sham groups. Chart: mean percent signal change in the left frontopolar cortex in the ECT and sham groups. The error bars represent the SEM. in performance on the self-referent emotional or nonemotional circuitries may thus be a more sensitive marker of treatment memory tests or in mood symptoms. As expected, all patients efficacy than change in overt behavioral measures. Given this, displayed improved depressive symptom severity following it is conceivable that the observed effects of ECT on frontopolar 6 active ECT sessions, but this was not predicted by the early activity indicates lesser need for prefrontal resources to retrieve modulatory effects of ECT on memory-related neural activity. positive self-referent words, that is, a relative ease of memory The reduced retrieval-specific neural activity in the fron- for positive self-referent material (Fletcher et al., 1996 Gould ; topolar cortex for positive words in ECT-treated patients is et al., 2003). Such an effect could be mechanistically important remarkably similar to our previous finding that a single dose of for the efficacy of ECT on depressive symptoms including low reboxetine decreases retrieval-specific fronto-parietal activity self-esteem and excessive self-blame. for positive words compared with placebo in healthy volunteers The absence of effects of ECT on hippocampal activity during (Miskowiak et al., 2007). While a single ECT session did not influ- retrieval of positive or negative words is consistent with no ECT- ence memory performance in our severely depressed patients, related change in limbic activity to emotional faces (Miskowiak the effect of reboxetine on neural activity in healthy volunteers et al., 2017). This contrasts with the robust evidence for effects of was accompanied by increased speed to recognize positive vs antidepressant drugs on limbic response to emotional informa- negative words (Miskowiak et al., 2007). It is possible that this tion (Harmer et al., 2006Norbur ; y et al., 2007; Godlewska et al., discrepancy in behavioral findings reflects lower sensitivity of 2016). Given this, it is conceivable that ECT primarily affects treatment effects on behavioral measures in this depressed sam- higher cortical processing of emotional information; in this ple due to confounding effects of medication and depressive case, increasing the efficiency of prefrontal retrieval-specific symptoms (Miskowiak et al., 2007). Modulation of key neural processes for positive information. Indeed, such differential Downloaded from https://academic.oup.com/ijnp/article-abstract/21/3/226/4372287 by Ed 'DeepDyve' Gillespie user on 16 March 2018 Miskowiak et al. | 233 Table 3. Overview of the Behavioral Data from the Categorization, Recall, and Recognition Trials Sham (n = 11) ECT (n = 14) Analysis Task Mean (SD) Mean (SD) [task by group] Task and performance measure P df Emotional categorization Accuracy (% correct responses) Positive words 90 (8) 89 (16) .03 1,22 Negative words 94 (5) 93 (6) [.5] 1,22 Response times (ms) Positive words 1 161 (376) 1 007 (328) .6 1,22 Negative words 1 270 (413) 1 077 (359) [.5] 1,22 Emotional recall Positive recalled 3 (1) 3 (3) .2 1,21 Negative recalled 3 (2) 2 (3) [.2] 1,21 Positive invented 2 (2) 2 (2) .06 1,21 Negative invented 1 (1) 1 (2) [.5] 1,21 Emotional recognition Accuracy (percent correct responses) Positive/hits 72 (18) 68 (8) .01 1,22 Negative/hits 61 (20) 63 (11) [.3] 1,22 Response times (ms) Correct trials Positive/hits 1 187 (450) 1 328 (296) .02 1,22 Negative/hits 1 274 (449) 1 340 (315) [.06] 1,22 Positive/correct rejections 1 347 (385) 1 205 (426) Negative/correct 1 331 (387) 1 193 (443) rejections Incorrect trials Incorrect positive/ 1 182 (279) 1 179 (297) rejection Incorrect negative/ 1 085 (211) 1 212 (299) rejection Incorrect positive/hit 1 007 (216) 1 202 (287) Incorrect negative/hit 1 187 (388) 1 219 (341) Discrimination accuracy Positive words 0.2 (0.2) 0.1 (0.3) .4 1,22 Negative words 0.2 (0.2) 0.1 (0.2) [.9] 1,22 Response bias Positive words 0.1 (0.2) 0.0 (0.2) .1 1,22 Negative words 0.0 (0.2) -0.1 (0.2) [.1] 1,22 Adjusted for years of education. Data missing for 1 participant (sham). Analyses performed n= 24 patients. Data missing for 3 participants (sham: n = 1, ECT: n = 2). actions of ECT and antidepressant drugs within a key emotion- The randomized, sham-controlled, parallel-group design processing circuitry may contribute to the well-established syn- was a major strength of the study, which is the first of its kind. ergy between these treatments for depression. The assessment of the effects of a single ECT vs sham ses- Notably, the lack of side-effects of a single ECT session on sion also enabled insight into the early neural activity change emotional and nonemotional verbal memory contrasts with that precedes, and may be mechanistically important for, the patients’ well-documented, burdensome memory impairments antidepressant effects of ECT. At the same time, this design after a complete series of ECT that lasts for weeks to months was a limitation, since a single ECT session may be insufficient (Semkovska and McLoughlin, 2010; Nordanskog et al., 2014; to modulate behavioral assays of negative self-referential Bodnar et al., 2016; Mohn and Rund, 2016). This suggests that memory bias in this heterogeneous patient cohort. However, the side-effects of ECT on verbal memory and hippocampal cir - it would have been unethical to conduct a longer-term ran- cuitry result from cumulative rather than acute actions of ECT domized controlled study of repeated ECT sessions, since on neurobiological processes involved in cognition, such as patients were severely ill and 50% received sham. A longitu- seizure-related excitotoxicity of N-methyl-D-aspartate receptor dinal study of the same patients scanned before and after ECT/ activation, alterations in cortisol and neurogenesis (Nobler and sham could have supported more robust inferences about the Sackeim, 2008), and upregulation of oxidative stress (Jorgensen effects of ECT on self-referent memory bias. Nevertheless, the et al., 2013). cross-sectional study design was chosen, because repeated Downloaded from https://academic.oup.com/ijnp/article-abstract/21/3/226/4372287 by Ed 'DeepDyve' Gillespie user on 16 March 2018 234 | International Journal of Neuropsychopharmacology, 2018 testing would have hampered assessment of incidental mem- References ory for self-referent words at a second scanning session Altar CA, Laeng P, Jurata LW, Brockman JA, Lemire A, Bullard J, and is consistent with the approach in our previous studies Bukhman YV, Young TA, Charles V, Palfreyman MG (2004) (Miskowiak et al., 2007). The emotional word recognition task Electroconvulsive seizures regulate gene expression of failed to activate the hippocampi. A possible explanation is distinct neurotrophic signalling pathways. J Neurosci that patients with mood disorders show deficient recruitment 24:2667–2677. of the hippocampus during memory retrieval (Fairhall et al., Anderson NH (1968) Likeableness of ratings of 555 personality- 2010; Kelley et al., 2013Dietsc ; he et al., 2014). Nevertheless, an trait words. J Pers Soc Psychol 9:272–279. almost identical self-referent word recognition task also failed Beck AT, Alford BA (2014) Depression causes and treatment. to activate the hippocampi in our healthy volunteer rebox- Philadelphia: University of Pennsylvania Press. etine study (Miskowiak et al., 2007), suggesting that this test Beck AT, Ward CH, Mendelson M, Mock J, Erbaugh J (1961) An might not sufficiently probe hippocampal function. Finally, it inventory for measuring depression. Arch Gen Psychiatry is worth noting that approximately 200 patients with nonpsy- 4:561–571. chotic unipolar depression between 18 and 60 years of age Bodnar A, Krzywotulski M, Lewandowska A, Chlopocka-Wozniak (i.e., potentially eligible patients) received ECT at Psychiatric M, Bartkowska-Sniatkowska A, Michalak M, Rybakowski JK Centre Copenhagen during the 6-year study period (Hundrup (2016) Electroconvulsive therapy and cognitive functions et al., 2016). The slow recruitment was mainly a result of in treatment-resistant depression. World J Biol Psychiatry lack of manpower (so many patients were randomly missed). 17:159–164. Another contributor was the sham-controlled study design Bolwig TG, Madsen TM (2007) Electroconvulsive therapy in since patients scheduled for ECT typically required immediate melancholia: the role of hippocampal neurogenesis. Acta treatment. This raises the possibility that less acutely unwell Psychiatr Scand 115:130–135. patients were selected, although this was not the main mech- Bouhuys A, Geerts E, Gordijn M (1999) Depressed patients’ per - anism of the slow recruitment. ceptions of facial emotions in depressed and remitted states In conclusion, this randomized, sham-controlled, double- are associated with relapse: a longitudinal study. J Nerv Ment blind fMRI study showed for the first time that a single ECT Dis 187:595–602. session reduces the retrieval-specific frontopolar response for Boynton GM, Engel SA, Glover GH, Heeger DJ (1996) Linear sys- positive words in the absence of changes in behavioral per - tems analysis of functional magnetic resonance imaging in formance or mood symptoms. This effect is compatible with human V1. J Neurosci Off J Soc Neurosci 16:4207–4221. early facilitation of memory for positive self-referent informa- Bradley B, Mathews A (1983) Negative self-schemata in clinical tion. The finding provides preliminary evidence for modula- depression. Br J Clin Psychol Br Psychol Soc 22:173–181. tion of negative self-referent bias in depression as a common Critchley HD, Lewis PA, Orth M, Josephs O, Deichmann R, Trimble mechanism of distinct biological treatments with efficacy on MR, Dolan RJ (2007) Vagus nerve stimulation for treatment- depression. resistant depression: behavioral and neural effects on encod- ing negative material. Psychosom Med 69:17–22. Acknowledgments Di Simplicio M, Norbury R, Harmer CJ (2012) Short-term anti- depressant administration reduces negative self-referential This work was supported by the Lundbeck Foundation (R32- processing in the medial prefrontal cortex in subjects at risk A2904). The funder had no role in study design, data collec- for depression. Mol Psychiatry 17:503–510. tion and analysis, decision to publish, or preparation of the Dierckx B, Heijnen WT, van den Broek WW, Birkenhäger TK (2012) manuscript. Efficacy of electroconvulsive therapy in bipolar versus unipolar major depression: a meta-analysis. Bipolar Disord 14:146–150. Dietsche B, Backes H, Stratmann M, Konrad C, Kircher T, Krug Statement of Interest A (2014) Altered neural function during episodic memory K.W.M. reports having received consultancy fees from Lundbeck encoding and retrieval in major depression: neural func- and Allergan in the past 3 years. The Lundbeck Foundation and tion during episodic memory in MDD. Hum Brain Mapp Weimann Foundation are acknowledged for providing half of 35:4293–4302. K.W.M.’s salary between 2012 and 2018 for her to do full-time Dobson KS, Shaw BF (1987) Specificity and stability of self- research during this time. O.B.P. was a member of the board of referent encoding in clinical depression. J Abnorm Psychol directors of the Elsass Foundation until 2016. C.J.H. has received 96:34–40. consultancy fees from P1vital ltd, Lundbeck, Servier, and Eli-Lilly Dozois DJ, Dobson KS (2001) Information processing and cog- and is a company director of Oxford Psychologists ltd. C.J.H. has nitive organization in unipolar depression: specificity and also received grant income from GSK, UCB, Janssen Inc, comorbidity issues. J Abnorm Psychol 110:236–246. Lundbeck, Servier, and Astra Zeneca. Within the past 3 years, Fairhall SL, Sharma S, Magnusson J, Murphy B (2010) Memory M.B.J. has received speaker fees for Lundbeck and consultant related dysregulation of hippocampal function in major fees for Shire. All other authors report no biomedical financial depressive disorder. Biol Psychol 85:499–503. interests or potential conflicts of interest. H.R.S. discloses hono- Fletcher PC, Shallice T, Frith CD, Frackowiak RS, Dolan RJ (1996) raria as journal editor from Elsevier Publishers and book editor Brain activity during memory retrieval. The influence of from Springer Publishing as well as honoraria as speaker from imagery and semantic cueing. Brain J Neurol 119:1587–1596. Genzyme and MerckSerono, and grant support from Biogen-idec Godlewska BR, Browning M, Norbury R, Cowen PJ, Harmer CJ within the past 3 years. L.V.K. reports having been a consult- (2016) Early changes in emotional processing as a marker ant for Lundbeck, AstraZeneca, and Sunovion within the last of clinical response to SSRI treatment in depression. Transl 3 years. Psychiatry 6:e957. Downloaded from https://academic.oup.com/ijnp/article-abstract/21/3/226/4372287 by Ed 'DeepDyve' Gillespie user on 16 March 2018 Miskowiak et al. | 235 Gonzalez-Castillo J, Roopchansingh V, Bandettini PA, Bodurka J Neshat-Doost HT, Taghavi MR, Moradi AR, Yule W, Dalgleish T (2011) Physiological noise effects on the flip angle selection (1998) Memory for emotional trait adjectives in clinically in BOLD fMRI. Neuroimage 54:2764–2778. depressed youth. J Abnorm Psychol 107:642–650. Gould R, Brown R, Owen A, ffytche D, Howard R (2003) FMRI Nobler MS, Sackeim HA (2008) Neurobiological correlates of BOLD response to increasing task difficulty during successful the cognitive side effects of electroconvulsive therapy. J ECT paired associates learning. Neuroimage 20:1006–1019. 24:40–45. Hamilton M (1960) A rating scale for depression. J Neurol Norbury R, Mackay CE, Cowen PJ, Goodwin GM, Harmer CJ (2007) Neurosurg Psychiatry 23:56–62. Short-term antidepressant treatment and facial processing: Harmer CJ, Cowen PJ (2013) ‘It’s the way that you look at it’- a functional magnetic resonance imaging study. Br J Psychiatry cognitive neuropsychological account of SSRI action in 190:531–532. depression. 368(1615):20120407. doi: 10.1098/rstb.2012.0407. Nordanskog P, Larsson MR, Larsson E- M, Johanson A (2014) Harmer CJ, Mackay CE, Reid CB, Cowen PJ, Goodwin GM (2006) Hippocampal volume in relation to clinical and cognitive Antidepressant drug treatment modifies the neural process- outcome after electroconvulsive therapy in depression. Acta ing of nonconscious threat cues. Biol Psychiatry 59:816–820. Psychiatr Scand 129:303–311. Harmer CJ, O’Sullivan U, Favaron E, Massey-Chase R, Ayres R, Ramel W, Goldin PR, Eyler LT, Brown GG, Gotlib IH, McQuaid JR Reinecke A, Goodwin GM, Cowen PJ (2009) Effect of acute (2007) Amygdala reactivity and mood-congruent memory antidepressant administration on negative affective bias in in individuals at risk for depressive relapse. Biol Psychiatry depressed patients. Am J Psychiatry 166:1178–1184. 61:231–239. Hundrup E, Osler M, Jørgensen MB (2016) Time trends and vari- Romero N, Sanchez A, Vazquez C (2014) Memory biases in remit- ations in electroconvulsive treatment in Denmark 2008 to ted depression: the role of negative cognitions at explicit 2014: a nationwide register-based study. J ECT doi: 10.1097/ and automatic processing levels. J Behav Ther Exp Psychiatry YCT.0000000000000381 [Epub ahead of print]. 45:128–135. Jorgensen A, Krogh J, Miskowiak K, Bolwig TG, Kessing LV, Fink-Jensen Schmidt M (1996) Rey auditory verbal learning test: a handbook. A, Nordentoft M, Henriksen T, Weimann A, Poulsen HE, Jorgensen Western Psychological Services Los Angeles. Available at: http://v- MB (2013) Systemic oxidatively generated DNA/RNA damage psyche.com/doc/Clinical%20Test/Rey%20Auditory%20Verbal%20 in clinical depression: associations to symptom severity and Learning%20Test.docx Retrieved 29 May, 2016. response to electroconvulsive therapy. J Affect Disord 149:355–362. Semkovska M, McLoughlin DM (2010) Objective cognitive per - Jorgensen A, Magnusson P, Hanson LG, Kirkegaard T, Benveniste formance associated with electroconvulsive therapy for H, Lee H, Svarer C, Mikkelsen JD, Fink-Jensen A, Knudsen GM, depression: a systematic review and meta-analysis. Biol Paulson OB, Bolwig TG, Jorgensen MB (2016) Regional brain Psychiatry 68:568–577. volumes, diffusivity, and metabolite changes after electro- Spielberger C (1983) Manual for the state-trait anxiety inventory. convulsive therapy for severe depression. Acta Psychiatr Palo Alto, CA: Consulting Psychologist Press. Scand 133:154–164. The UK ECT Review Group (2003) Efficacy and safety of elec- Kelley R, Garrett A, Cohen J, Gomez R, Lembke A, Keller J, Reiss troconvulsive therapy in depressive disorders: a systematic AL, Schatzberg A (2013) Altered brain function underlying review and meta-analysis. The Lancet 361:799–808. verbal memory encoding and retrieval in psychotic major van Tol M-J, Demenescu LR, van der Wee NJA, Kortekaas R, depression. Psychiatry Res Neuroimaging 211:119–126. Marjan MAN, Boer JAD, Renken RJ, van Buchem MA, Zitman Madsen TM, Treschow A, Bengzon J, Bolwig TG, Lindvall O, FG, Aleman A, Veltman DJ (2012) Functional magnetic reson- Tingström A (2000) Increased neurogenesis in a model of ance imaging correlates of emotional word encoding and rec- electroconvulsive therapy. Biol Psychiatry 47:1043–1049. ognition in depression and anxiety disorders. Biol Psychiatry Mannie ZN, Bristow GC, Harmer CJ, Cowen PJ (2007) Impaired 71:593–602. emotional categorisation in young people at increased famil- Watkins PC, Mathews A, Williamson DA, Fuller RD (1992) Mood- ial risk of depression. Neuropsychologia 45:2975–2980. congruent memory in depression: emotional priming or elab- Mathews A, MacLeod C (2005) Cognitive vulnerability to emo- oration? J Abnorm Psychol 101:581–586. tional disorders. Annu Rev Clin Psychol 1:167–195. Winkler AM, Ridgway GR, Webster MA, Smith SM, Nichols TE Miskowiak K, Papadatou-Pastou M, Cowen PJ, Goodwin GM, (2014) Permutation inference for the general linear model. Norbury R, Harmer CJ (2007) Single dose antidepressant Neuroimage 92:381–397. administration modulates the neural processing of self-ref- Wolfensberger SPA, Veltman DJ, Hoogendijk WJG, De Ruiter erent personality trait words. Neuroimage 37:904–911. MB, Boomsma DI, de Geus EJC (2008) The neural correlates Miskowiak KW, Carvalho AF (2014) “Hot” cognition in major of verbal encoding and retrieval in monozygotic twins at depressive disorder: a systematic review. CNS Neurol Disord low or high risk for depression and anxiety. Biol Psychol 79: Drug Targets 13:1787–1803. 80–90. Miskowiak KW, Kessing LV, Ott CV, Macoveanu J, Harmer CJ, Woolrich MW, Ripley BD, Brady M, Smith SM (2001) Temporal Jørgensen A, Revsbech R, Jensen HM, Paulson OB, Siebner HR, autocorrelation in univariate linear modeling of FMRI data. Jørgensen MB (2017) Does a single session of ECT alter the neural Neuroimage 14:1370–1386. response to emotional faces in depression? A randomised Yoshimura S, Okamoto Y, Onoda K, Matsunaga M, Ueda K, Suzuki sham-controlled fMRI study. J Psychopharmacol 31:1215–1224. S, Shigetoyamawaki (2010) Rostral anterior cingulate cortex Mohn C, Rund BR (2016) Significantly improved neurocognitive activity mediates the relationship between the depressive function in major depressive disorders 6 weeks after ECT. J symptoms and the medial prefrontal cortex activity. J Affect Affect Disord 202:10–15. Disord 122:76–85. Downloaded from https://academic.oup.com/ijnp/article-abstract/21/3/226/4372287 by Ed 'DeepDyve' Gillespie user on 16 March 2018
International Journal of Neuropsychopharmacology – Oxford University Press
Published: Mar 1, 2018
It’s your single place to instantly
discover and read the research
that matters to you.
Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.
All for just $49/month
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
All the latest content is available, no embargo periods.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud