Introduction
Major depressive disorder (MDD) is associated with clinically significant deficits in cognitive function and executive control (McIntyre et al. Reference McIntyre, Cha, Soczynska, Woldeyohannes, Gallaugher, Kudlow, Alsuwaidan and Baskaran2013). In particular, deficits in working memory (i.e. the maintenance and manipulation of relevant information in a temporary buffer that is constantly updated to guide behaviour) have been widely reported in MDD (Marazziti et al. Reference Marazziti, Consoli, Picchetti, Carlini and Faravelli2010). At a clinical level, such deficits could be consistent with MDD patients subjectively reporting problems with demanding cognitive tasks such as planning and organization (Millan et al. Reference Millan, Agid, Brune, Bullmore, Carter, Clayton, Connor, Davis, Deakin, DeRubeis, Dubois, Geyer, Goodwin, Gorwood, Jay, Joels, Mansuy, Meyer-Lindenberg, Murphy, Rolls, Saletu, Spedding, Sweeney, Whittington and Young2012).
The neural basis of working memory has been well characterized in imaging studies (Owen et al. Reference Owen, McMillan, Laird and Bullmore2005). For example the n-back task, where subjects are asked to monitor the identity or location of a series of verbal or non-verbal stimuli and to indicate when the currently presented stimulus is the same as the one presented n-trials previously, is associated with increased activation in a fronto-parietal network and concomitant deactivation of the default-mode network (DMN), including the medial temporal lobe (MTL) (Cousijn et al. Reference Cousijn, Rijpkema, Qin, van Wingen and Fernandez2012). This shift from the task-negative network to the diametrically opposed task-positive network is presumed to support reallocation of available neuronal resources to regions required for task-specific processing, thereby optimizing performance (Wirth et al. Reference Wirth, Jann, Dierks, Federspiel, Wiest and Horn2011).
Neuroimaging studies of the n-back task in the acute phase of MDD have reported both increased (Harvey et al. Reference Harvey, Fossati, Pochon, Levy, Lebastard, Lehericy, Allilaire and Dubois2005; Matsuo et al. Reference Matsuo, Glahn, Peluso, Hatch, Monkul, Najt, Sanches, Zamarripa, Li, Lancaster, Fox, Gao and Soares2007; Walsh et al. Reference Walsh, Williams, Brammer, Bullmore, Kim, Suckling, Mitterschiffthaler, Cleare, Pich, Mehta and Fu2007) and decreased activation (Elliott et al. Reference Elliott, Baker, Rogers, O'Leary, Paykel, Frith, Dolan and Sahakian1997; Okada et al. Reference Okada, Okamoto, Morinobu, Yamawaki and Yokota2003) in frontal regions and reduced deactivation in the DMN (Harvey et al. Reference Harvey, Fossati, Pochon, Levy, Lebastard, Lehericy, Allilaire and Dubois2005). Direct comparison of these studies is confounded by imaging technique [functional magnetic resonance imaging (fMRI) versus positron emission tomography (PET)], task (e.g. verbal working memory versus complex planning) and, importantly, behavioural performance. Where patients and controls are matched in terms of performance (Harvey et al. Reference Harvey, Fossati, Pochon, Levy, Lebastard, Lehericy, Allilaire and Dubois2005; Matsuo et al. Reference Matsuo, Glahn, Peluso, Hatch, Monkul, Najt, Sanches, Zamarripa, Li, Lancaster, Fox, Gao and Soares2007; Walsh et al. Reference Walsh, Williams, Brammer, Bullmore, Kim, Suckling, Mitterschiffthaler, Cleare, Pich, Mehta and Fu2007) increased frontal activation is consistently reported and together these data have led to the hyperfrontality hypothesis of working memory in MDD (Harvey et al. Reference Harvey, Fossati, Pochon, Levy, Lebastard, Lehericy, Allilaire and Dubois2005). In this theoretical model increasing task complexity is associated with increasing activation in task-related areas and increasing deactivation in the DMN. In MDD, hyperfrontality occurs to compensate for a lack of deactivation in the counterproductive task-negative network. It is unclear, however, if similar resource allocation abnormalities persist in remitted MDD patients. Schöning et al. (Reference Schöning, Zwitserlood, Engelien, Behnken, Kugel, Schiffbauer, Lipina, Pachur, Kersting, Dannlowski, Baune, Zwanzger, Reker, Heindel, Arolt and Konrad2008), using an n-back task in remitted, antidepressant-treated patients, reported increased activation in the anterior cingulate cortex, though frontal and DMN response in these subjects did not differ from controls.
It is possible, therefore, that hyperfrontality during working memory reflects a state marker of acute depression whereas an increased cingulate response may signal medication effects. Equally, it is possible that medication may act to dampen hyperfrontality in remitted patients. This makes it important to study the neural basis of working memory in unmedicated patients who have recovered from depression. The present study therefore employed fMRI, to study neural responses to a simple verbal n-back task, in a cohort of unmedicated remitted MDD patients.
Method
Participants
We included 15 remitted MDD patients (rMDD; eight females), mean age 39.5 years (range 21–61 years), recruited from a number of sources including newspaper advertisements and referrals from other researchers. All participants were assessed for the presence of current and past psychiatric disorder with the Structured Clinical Interview for DSM-IV (SCID; APA, 1995). Remitted depressed patients were included only if they met criteria for a primary diagnosis of MDD at least twice in the past, had no current Axis I disorders and were unmedicated for at least 6 months prior to inclusion into the study. Mean number of previous episodes was 2.7 (range 2–6), mean period since the last episode was 33.8 months (range 2–72 months), mean time since last medication was 25.1 months (range 8–108 months); five rMDD were never medicated; two rMDD (both females) suffered from panic attacks when depressed; there were no other co-morbidities. We also recruited by advertisement 15 controls (five females), mean age 38.3 years (range 23–61 years), who were determined by the same instruments to have no current or past history of major depression and no history of depression in a biological parent or other first-degree relative (healthy controls; HC). Current mood was assessed with the Beck Depression Inventory (BDI; Beck et al. Reference Beck, Ward, Mendelson, Mock and Erbaugh1961). All participants were right-handed and had normal or corrected-to-normal vision. Participants with any personal history of Axis I or neurological disorders were excluded.
Working memory task design
During fMRI scanning, subjects completed a letter variant of the n-back task (Harvey et al. Reference Harvey, Fossati, Pochon, Levy, Lebastard, Lehericy, Allilaire and Dubois2005). Working memory load was manipulated by using three levels of complexity: 1-, 2- and 3-back tasks. Briefly, subjects were requested to indicate whether a letter presented on the screen (the ‘target’ stimulus) matched a previously presented letter (the ‘cue’ stimulus). To minimize visual and phonological strategies, we used phonologically closed letters presented in upper and lower case. Thus, only the following characters were presented: b, B, d, D, g, G, p, P, t, T, v, V. Subjects were instructed to ignore the case of letters and respond by pressing a button with their right or left thumb if the target was identical or different from the cue, respectively. Subjects also performed a sensorimotor control task (0-back) during which they were required to respond to a prespecified letter (‘x, X’). All blocks consisted of a sequence of 10 consonants varying in case. Letters were presented for 500 ms with a fixed interstimulus interval of 1500 ms. Prior to each task block an instruction screen (0-, 1-, 2-, 3-back) was presented for 2000 ms. A 4000 ms blank screen separated the instruction from the onset of the first letter. Task blocks were separated by 8000 ms of fixation cross. Four blocks of each condition were presented in a fixed pseudo-random order (0-, 1-, 2-, 3-, 1-, 3-, 2-, 0-, 2-, 1-, 0-, 3-, 1-, 0-, 3-, 2-back). All conditions were matched for the number of target and upper-/lower-case letters presented. Stimuli were presented on a personal computer using E-Prime (version 1.0; Psychology Software Tools Inc., USA) and projected onto an opaque screen at the foot of the scanner bore, which subjects viewed using angled mirrors. Subject responses were made via an MRI-compatible keypad. Both accuracy and response latency were recorded by E-Prime. Immediately prior to scanning all subjects received training with another set of stimuli to ensure that they understood fully the task requirements. During this training period participants were exposed to single blocks of the 0-back, 1-back and 2-back conditions. The same blocks were repeated, as required, until the participant indicated that they understood the nature of the task.
fMRI data acquisition
All imaging data were collected using a Siemens TIM Trio 3T scanner located at the Oxford Centre for Magnetic Resonance (OCMR), University of Oxford. Functional imaging consisted of 45 T2*-weighted echo-planar image (EPI) axial oblique slices that began at the cerebral vertex and encompassed the entire cerebrum and the majority of the cerebellum. Acquisition parameters were: repetition time (TR) = 3000 ms; echo time (TE) = 50 ms; flip angle = 90°; field of view = 192 × 192 mm; matrix size = 64 × 64; slice thickness = 3 mm. These parameters were selected to optimize the signal across the entire volume of acquisition. The first two EPI volumes in each session were discarded to avoid T1 equilibration effects. To facilitate later co-registration of the fMRI data into standard space, we also acquired a magnetization prepared rapid acquisition gradient echo (MPRAGE) sequence (TR = 12 ms, TE = 5.65 ms, voxel size = 1 mm3).
Grey-matter probability maps
Each individual subject's T1-weighted high-resolution anatomic image was registered to a standard template [Montreal Neurological Institute (MNI) 152 stereotactic template] using an affine procedure with a 12-parameter fit (Jenkinson & Smith, Reference Jenkinson and Smith2001). The MRI images were then segmented into three tissue classes (cerebrospinal fluid, white matter and grey matter) (Zhang et al. Reference Zhang, Brady and Smith2001). Grey-matter probability maps were masked (i.e. non-brain voxels set to zero) and smoothed to yield images with similar smoothness to the corresponding functional data, in order to minimize partial volume effects (Oakes et al. Reference Oakes, Fox, Johnstone, Chung, Kalin and Davidson2007). Lastly, the resulting smoothed grey-matter probability maps were demeaned prior to inclusion in the between-groups analysis.
fMRI data analysis
fMRI data were pre-processed and analysed using Functional MRI of the Brain (FMRIB) Software Library (FSL) (version 4.1.8; Smith et al. Reference Smith, Jenkinson, Woolrich, Beckman, Behrens, Johansen-Berg, Bannister, De Luca, Drobnjak, Flitney, Niazy, Saunders, Vickers, Zhang, De Stefano, Brady and Mathews2004). Pre-processing included within-subject image realignment (Jenkinson et al. Reference Jenkinson, Bannister, Brady and Smith2002), correction for geometric EPI distortions based on an acquired B0 fieldmap, non-brain removal (Smith, Reference Smith2002) and spatial normalization to a standard template. During registration, signal loss (resulting from through-slice field gradients) was calculated and used as a cost function mask to exclude voxels where signal loss was greatest. Finally, images were spatially smoothed using a Gaussian kernel (5 mm full-width-half-maximum) and high pass-filtered (to a maximum of 0.004 Hz).
Analyses of data from individual subjects were computed using the general linear model with local autocorrelation correction (Woolrich et al. Reference Woolrich, Ripley, Brady and Smith2001). Four explanatory variables were modelled: ‘0-, 1-, 2- and 3-back’. These explanatory variables were modelled by convolving each trial block with a haemodynamic response function, using a variant of a γ function (i.e. a normalization of the probability density function of the γ function) with a standard deviation of 3 s and a mean lag of 6 s. In addition, temporal derivatives, estimated motion parameters (three translation and three rotation), task instruction and anticipation period were included in the model as regressors of no interest to increase statistical sensitivity. Regression analyses modelled two mutually orthogonal characteristics of brain activation at each voxel: (1) mean overall capacity (i.e. 1-, 2-, 3-back versus 0-back); and (2) quadratic load response to each level of task difficulty (i.e. modelled with the following contrast of 1 −1 −1 1).
Individual subject data were combined at the group level using a full mixed-effects analyses (Woolrich et al. Reference Woolrich, Behrens, Beckmann, Jenkinson and Smith2004). This mixed-effects approach enables generalization of the results beyond the sample of subjects tested. We included age, sex and grey-matter probability maps at a given voxel as covariates (nuisance variables) to minimize the potential impact of these variables on group comparisons (Oakes et al. Reference Oakes, Fox, Johnstone, Chung, Kalin and Davidson2007). Significant activations across the whole brain were identified using cluster-based thresholding of statistical images with a height threshold of Z = 2.3 and a (whole-brain-corrected) spatial extent threshold of p < 0.05. In a priori regions of interest (ROI) (i.e. the working memory network, anterior cingulate and medial prefrontal cortex) appropriate small volume corrections were applied.
Behavioural and demographic data analysis
Demographic and neuropsychological data were compared using Student's t tests and χ 2 tests for independence were appropriate. N-back response accuracy and latency were analysed using split-plot analysis of variance (ANOVA) with group as the between-subjects factor (two levels) and complexity as the within-subjects factor (three levels). Student's t tests were used to compare groups for the 0-back condition.
Results
Demographic and neuropsychological measures
Groups did not differ in terms of sex ratio (χ 2 = 1.22, p = 0.53) or age (t 28 = −0.24, p = 0.81). BDI scores were marginally greater in rMDD (1.87 versus 0.8, t 28 = 1.85, p = 0.08). The mean educational level of controls and recovered depressed patients did not differ significantly (data not shown).
N-back performance
Groups were similar in terms of 0-back response accuracy (t 28 = 0.47, p = 0.65) and latency (t 28 = −0.50, p = 0.62). Similarly, repeated-measures ANOVAs did not reveal significant between-group differences for either response accuracy (main effect of group: F 1,28 = 3.29, p = 0.08) or latency (main effect of group: F 1,28 < 1). As expected, increasing complexity was associated with reduced accuracy (main effect of complexity: F 2,56 = 57.13, p < 0.001) and increased response latency (main effect of complexity: F 2,56 = 26.02, p < 0.001). There was, however, no significant group × complexity interaction for response accuracy or latency (both F 2,56 < 1) [see online Supplementary Table S1 for a summary of accuracy and response latency data].
fMRI data
Overall capacity (n-back versus 0-back)
For each group, activations were observed in the anterior cingulate, parietal, medial frontal, temporal and occipital gyri. The regions of activation observed here are consistent with those previously reported (Fig. 1 and online Supplementary Table S2). There were no significant between-group differences in terms of overall capacity that reached statistical significance either across the whole brain or following small volume correction (i.e. masked with the main effect of overall capacity).
Fig. 1. Mean overall capacity (n-back versus 0-back) in healthy controls (top row) and unmedicated remitted depressed patients (bottom row). The left hemisphere is depicted on the right side of the image. The colour bar reflects minimum and maximum Z scores.
Quadratic load response activity
Remitted depression was associated with greater quadratic load response activity in the left hippocampus [cluster size (voxels) = 662, p = 0.002, MNI coordinates: x = −28, y = −10, z = −20] as compared with HC (Fig. 2). Post-hoc analyses restricted to the right hippocampus also revealed greater load response activity in rMDD as compared with HC [cluster size (voxels) = 78, p = 0.02, MNI coordinates: x = 28, y = −8, z = −22].
Fig. 2. Left hippocampus blood oxygen level-dependent (BOLD) response (expressed as percentage signal change) as a function of cognitive load in unmedicated remitted depressed patients (○, □) and healthy controls (●, ■). Values are means, with standard errors represented by vertical bars.
Discussion
Our findings indicate that rMDD is associated with reduced deactivation in the bilateral hippocampus. We found no evidence of hyperfrontality or increased cingulate activation in this group of unmedicated rMDD. We speculate, therefore, that a failure to dampen MTL activation (as reported here) reflects a trait vulnerability factor for depression unconfounded by pharmacotherapy.
There are a number of limitations associated with this study and these should be taken into consideration when interpreting the results. For example, we used a pragmatic cross-sectional approach rather than a prospective design and we did not include a medicated rMDD group. To fully assess if increased hippocampal reactivity, as reported here, represents a scar effect that persists into periods of remission requires longitudinal studies. Similarly, studies that include a medicated comparison group are required to directly assay pharmacotherapy effects. To probe the question of hyperfrontality we used an ROI approach based on the main effect of overall activation, which may have left us exposed to type II error. Again, a medicated comparison group would aid ROI definition and improve statistical sensitivity. Finally, we have limited demographic information for both patients and controls. There is likely to be considerable heterogeneity in the neurobiology of depression and careful delineation of patients and control groups in future studies will increase the probability of successful replication of imaging observations.
Functional imaging studies in healthy controls have demonstrated that working memory processes are accompanied by increased fronto-parietal activity and concomitant deactivations in the DMN, including the MTL/hippocampus (Cousijn et al. Reference Cousijn, Rijpkema, Qin, van Wingen and Fernandez2012). Clearly, optimal performance during cognitively demanding tasks requires a delicate balance between these two antagonistic networks (Wirth et al. Reference Wirth, Jann, Dierks, Federspiel, Wiest and Horn2011). Data from neuroimaging studies in acutely depressed patients suggest that this balance may be disrupted. For example, a converging corpus indicates that, when maintaining normal working memory performance, depressed patients show increased activation in the cingulate (Harvey et al. Reference Harvey, Fossati, Pochon, Levy, Lebastard, Lehericy, Allilaire and Dubois2005; Matsuo et al. Reference Matsuo, Glahn, Peluso, Hatch, Monkul, Najt, Sanches, Zamarripa, Li, Lancaster, Fox, Gao and Soares2007) and frontal cortex (Harvey et al. Reference Harvey, Fossati, Pochon, Levy, Lebastard, Lehericy, Allilaire and Dubois2005; Matsuo et al. Reference Matsuo, Glahn, Peluso, Hatch, Monkul, Najt, Sanches, Zamarripa, Li, Lancaster, Fox, Gao and Soares2007; Walter et al. Reference Walter, Wolf, Spitzer and Vasic2007; Fitzgerald et al. Reference Fitzgerald, Srithiran, Benitez, Daskalakis, Oxley, Kulkarni and Egan2008) and attenuated deactivation in the limbic system/DMN (Harvey et al. Reference Harvey, Fossati, Pochon, Levy, Lebastard, Lehericy, Allilaire and Dubois2005; Rose et al. Reference Rose, Simonotto and Ebmeier2006) and together these data have led to the hyperfrontality hypothesis of working memory in MDD (Harvey et al. Reference Harvey, Fossati, Pochon, Levy, Lebastard, Lehericy, Allilaire and Dubois2005). Alternative support for this hypothesis comes from a number of neuroimaging studies reporting hyperactivity within a number of regions encompassed by the DMN in the depressed ‘resting’ brain (Greicius, Reference Greicius2008). For example, Grecius et al. (Reference Greicius, Flores, Menon, Glover, Solvason, Kenna, Reiss and Schatzberg2007) reported increased connectivity of the subgenual cingulate and orbitofrontal cortex, thalamus and precuneus. In addition, subgenual cingulate connectivity was positively associated with depression refractoriness (as measured by duration of the current depressive episode) (Greicius et al. Reference Greicius, Flores, Menon, Glover, Solvason, Kenna, Reiss and Schatzberg2007). Cao et al. (Reference Cao, Liu, Xu, Li, Gao, Sun, Xu, Ren, Yang and Zhang2012) observed aberrant hippocampo-frontal connectivity in medication-naive MDD patients. Similarly, Monkul et al. (Reference Monkul, Silva, Narayana, Peluso, Zamarripa, Nery, Najt, Lancaster, Fox, Lafer and Soares2012) reported increased left hippocampus regional cerebral blood flow (as measured by H2 15O PET) in unmedicated MDD patients as compared with healthy controls. In addition, we have found reduced fronto-hippocampal connectivity following short-term antidepressant treatment in healthy controls (McCabe et al. Reference McCabe, Mishor, Filippini, Cowen, Taylor and Harmer2011). Together these data suggest that: (1) MDD is associated with hyperactivation and/or hyperconnectivity of the DMN; (2) hippocampal abnormalities may be more prevalent in unmedicated patients; and (3) treatment for depression modulates hippocampal connectivity.
To our best knowledge only one study has examined working memory in combination with fMRI in rMDD (Schöning et al. Reference Schöning, Zwitserlood, Engelien, Behnken, Kugel, Schiffbauer, Lipina, Pachur, Kersting, Dannlowski, Baune, Zwanzger, Reker, Heindel, Arolt and Konrad2008). In that study, Schöning et al. (Reference Schöning, Zwitserlood, Engelien, Behnken, Kugel, Schiffbauer, Lipina, Pachur, Kersting, Dannlowski, Baune, Zwanzger, Reker, Heindel, Arolt and Konrad2008) reported increased activation in the anterior cingulate cortex, though lateral prefrontal and hippocampal activation did not differ from controls. We did not observe increased cingulate activation (either at the whole-brain level or using a cingulate mask) in our group of unmedicated patients. We acknowledge, however, that a direct comparison between the current findings and those of Schöning et al. (Reference Schöning, Zwitserlood, Engelien, Behnken, Kugel, Schiffbauer, Lipina, Pachur, Kersting, Dannlowski, Baune, Zwanzger, Reker, Heindel, Arolt and Konrad2008) is difficult due to the differences in the nature of the working memory task and the analysis techniques employed. Of note, we have previously reported increased quadratic load response activity in the parietal cortex in young adults with increased familial risk of depression but no evidence of altered hippocampal reactivity (Mannie et al. Reference Mannie, Harmer, Cowen and Norbury2010). Together, these data suggest that perturbed parietal activity may form part of the familial risk of experiencing depression, abnormal hippocampal activity reflects a ‘scar’ effect that persists into periods of unmedicated remission and an increased cingulate response may signal medication effects in remitted patients.
Here we have proposed that rMDD is associated with increased DMN ‘noise’ which should, according to the hyperfrontality model of working memory in depression, lead to increased frontal activation. However, despite no significant between-group differences in accuracy or response latency, we found no evidence of hyperfrontality in this group of unmedicated rMDD. It is possible that our relatively small sample size (15 HC and 15 rMDD) may have left us exposed to a Type II error in detecting behavioural differences between the two groups (rMDD participants showed numerically increased response latency and reduced accuracy, although between-group differences were non-significant). Alternatively, DMN noise may need to reach a critical threshold before such interference has an adverse impact on task performance [as indexed by behavioural metrics or blood oxygen level-dependent (BOLD) response in task-dependent regions]. Future studies that include healthy controls, unmedicated acutely depressed and remitted depressed patients would be required to fully test this hypothesis.
In summary, unmedicated rMDD was associated with a pattern of increased activation in the bilateral hippocampus during a verbal working memory task. It is clearly of interest that remitted depressed patients who are off medication and essentially asymptomatic continue to manifest neural abnormalities similar to those observed in acute depression. We suggest, therefore, that a reduced ability to dampen task-irrelevant activity may reflect a neurobiological risk factor for recurrent depression.
Supplementary material
For supplementary material accompanying this paper visit http://dx.doi.org/10.1017/S0033291713001682.
Acknowledgements
The study was supported by the Medical Research Council (MRC). P.J.C. is an MRC Clinical Scientist.
Declaration of Interest
P.J.C. has been a paid member of advisory boards of DSM, Eli Lilly, Servier and Wyeth, and has been a paid lecturer for Eli Lilly, Servier and GlaxoSmithKline. P.J.C. has also received remuneration for scientific advice given to legal representatives of GlaxoSmithKline.
