Introduction
Mounting evidence suggests the presence of state-related abnormalities in resting state connectivity in active major depressive disorder (MDD), particularly within subsystems of the default-mode network (Anand et al., Reference Anand, Li, Wang, Wu, Gao, Bukhari and Lowe2005; Connolly et al., Reference Connolly, Wu, Ho, Hoeft, Wolkowitz, Eisendrath and Yang2013; Frodl et al., Reference Frodl, Bokde, Scheuerecker, Lisiecka, Schoepf, Hampel and Meisenzahl2010; Greicius et al., Reference Greicius, Flores, Menon, Glover, Solvason, Kenna and Schatzberg2007; Jacobs et al., Reference Jacobs, Jenkins, Gabriel, Barba, Ryan, Weisenbach and Welsh2014; Liu et al., Reference Liu, Wu, Qiu, Yang, Kuang, Chan and Gong2011; Sheline, Price, Yan, & Mintun, Reference Sheline, Price, Yan and Mintun2010). To a large extent, this work has focused on emotion-related phenomena, such as negative memory biases, sadness mood inductions, and rumination (Connolly et al., Reference Connolly, Wu, Ho, Hoeft, Wolkowitz, Eisendrath and Yang2013; Hamilton, Chen, Thomason, Schwartz, & Gottlib, Reference Hamilton, Chen, Thomason, Schwartz and Gottlib2011; Jacobs et al., Reference Jacobs, Jenkins, Gabriel, Barba, Ryan, Weisenbach and Welsh2014). More recently, differential resting-state patterns have also been observed in the remitted MDD state; specifically, hyperconnectivity from the default mode and salience networks to the cognitive control network (Jacobs et al., Reference Jacobs, Jenkins, Gabriel, Barba, Ryan, Weisenbach and Welsh2014). Even more intriguing, these differential patterns in MDD appear to be quite robust even early in the course of illness, and are related to persisting dysfunction, both in rumination and cognitive control (Connolly et al., Reference Connolly, Wu, Ho, Hoeft, Wolkowitz, Eisendrath and Yang2013; Jacobs et al., Reference Jacobs, Jenkins, Gabriel, Barba, Ryan, Weisenbach and Welsh2014; Zhang et al., Reference Zhang, Wang, Wu, Kuang, Huang, He and Gong2011). Other patterns of increased connectivity in early state illness appear to potentially offer some degree of resilience; such as in Zhang and colleagues (Reference Zhang, Wang, Wu, Kuang, Huang, He and Gong2011), who reported a negative correlation between left hippocampal connectivity and disease duration and severity in first-episode, medication-naïve MDD patients. Investigating and observing resting state connectivity patterns, especially as they relate to dysfunctional behaviors, performance, and affective responses collected at separate time points, increase confidence in the stability of the connectivity patterns and the relationships observed.
To date, these patterns of abnormal connectivity in both acute and remitted MDD have not yet been investigated in relation to one of the most common abnormalities observed in acute and chronic MDD—memory difficulties. Indeed, memory difficulties are extensively studied in MDD (see Burt, Zembar, Niederehe, Reference Burt, Zembar and Niederehe1995, for review); however, attention and executive functioning are more frequently examined in remitted MDD, as young adults in remission report these deficits to be the most prominent (Hasselbalch, Knor, & Kessing, Reference Hasselbalch, Knorr and Kessing2011; Paelecke-Habermann, Pohl, & Leplow, Reference Paelecke-Habermann, Pohl and Leplow2005; Peters et al., Reference Peters, Jacobs, Crane, Ryan, Weisenbach, Ajilore and Langenecker2015; Weiland-Fiedler et al., Reference Weiland-Fiedler, Erickson, Waldeck, Luckenbaugh, Pike, Bonne and Neumeister2004). A review of remitted MDD literature examining memory, specifically, has demonstrated that remitted MDD individuals show deficits in nonverbal memory, largely due to difficulties organizing visual information (Behnken et al., Reference Behnken, Schöning, Gerβ, Konrad, de Jong-Meyer, Zwanzger and Arolt2010), and middle-aged recovered and medicated melancholic patients display declines in delayed logical and visual memory (Marcos et al., Reference Marcos, Salamero, Gutierrez, Catalan, Gasto and Lazaro1994).
The association between memory and MDD remains controversial, despite thousands of articles and several meta-analyses showing memory difficulties (e.g., Bora, Harrison, Yücel, & Pantelis, Reference Bora, Harrison, Yücel and Pantelis2013; Considine, et al., Reference Considine, Weisenbach, Walker, McFadden, Franti, Bieliauskas and Langenecker2011; Elderkin-Thompson, Moody, Knowlton, Hellemann, & Kumar, Reference Elderkin-Thompson, Moody, Knowlton, Hellemann and Kumar2011; Hermens, Naismith, Redoblado-Hodge, Scott, & Hickie, Reference Hermens, Naismith, Redoblado-Hodge, Scott and Hickie2010; Liu, Li, Xiao, Yang, & Jiang, Reference Liu, Li, Xiao, Yang and Jiang2013; Mulligan, Reference Mulligan2011). In part, this controversy relates to whether the observed memory difficulties persist outside of active episodes (Snyder, Reference Snyder2013), whether they predate and comprise a risk factor for depression (Porter, Gallagher, Thompson, & Young, Reference Porter, Gallagher, Thompson and Young2003), whether they are a consequence of slowed information processing (Butters et al., Reference Butters, Whyte, Nebes, Begley, Dew, Mulsant and Becker2004), or whether they are an epiphenomenon related to poor effort (Considine et al., Reference Considine, Weisenbach, Walker, McFadden, Franti, Bieliauskas and Langenecker2011; Rohling, Green, Allen, & Iverson, Reference Rohling, Green, Allen and Iverson2002), easy distractibility, or ascertainment biases in the patients who actually arrive for studies and evaluations (e.g., those with MDD and no overt cognitive difficulties may never show up for studies, disability claims, or neuropsychological evaluations).
Studies of memory functioning in MDD have been further hampered by the use of tools that do not always distinguish between specific aspects of memory or less severe memory impairment (Burt et al., Reference Burt, Zembar and Niederehe1995). List learning tasks with multiple learning trials allow for a dissociation of amnestic impairments from less severe difficulties with attention, distraction, encoding, and recognition, suitable for dementia and severe brain injury evaluations. Memory difficulties in MDD may still lead to disruptions in work, family, and social functioning, if relatively less severe. Furthermore, a great majority of memory demands in real life involve one-trial learning in distracting contexts. We designed an episodic memory task, the Semantic List Learning Test (SLLT) to engage one-trial learning and for use with fMRI (Langenecker, Caveney, Persad, & Giordani, Reference Langenecker, Caveney, Persad and Giordani2004). The SLLT includes a distraction condition and targeted learning strategies (semantic cues) to reduce individual variance that might be attributed to short-term memory or memory organization strategies. To this end, we observed memory difficulties in adults with active MDD relative to age-matched controls (Kassel et al., 2015), and equivalent memory performance when comparing depressed elders with older healthy controls (Weisenbach et al., Reference Weisenbach, Kassel, Rao, Weldon, Avery, Briceño and Langenecker2014).
Despite differences in memory performance on the SLLT between these two MDD age cohorts, both studies observed decreased temporal and frontal activation in active MDD, with the exception of increased left inferior frontal gyrus activation in elders with MDD. Inferior frontal activation was greater in elders with MDD relative to healthy controls, but this pattern was not observed in adults with MDD. These studies form an important nexus for understanding the nature, extent, and nuances in the neural bases of memory difficulties in MDD. However, a couple of limitations exist. First, the majority of these studies have used a wide age range, and it is not clear if there may be changes in memory performance in MDD that might be easier to observe earlier in the course of illness, before age associated memory decline occurs. Furthermore, study of active MDD may obscure persisting stable memory difficulties, or could even result in greater noise in measurement at the behavioral and neural circuit levels.
A final challenge to studies of this type is concern about the nature of activation differences when performance is not equivalent between groups. In this instance, it is difficult to fully attribute any network or region differences in activation solely to the disease, as it may be driven by performance differences. In our work, we highlight how specific performance differences should be present, such that the disease process is reflected in a performance difference and the underlying network abnormalities that drive performance (Briceño et al., Reference Briceño, Weisenbach, Rapport, Hazlett, Bieliauskas, Haase and Langenecker2013; Kassel et al., 2015; Langenecker et al., Reference Langenecker, Caveney, Persad and Giordani2004, Reference Langenecker, Weisenbach, Giordani, Briceño, Guidotti Breting, Schallmo and Starkman2012). However, the challenge of parsimony remains, that is, are these differences related to disease, performance, or disease and performance?
To address some of the existing weaknesses within the literature, some highlighted further by our own recent work, we undertook an investigation of memory performance in remitted MDD, using the SLLT while in the scanner. But in this case, we did not use the activation-derived data during this task, only the performance-derived data. We hypothesized that resting state connectivity differences, using the hippocampus as a seed, would be present in remitted MDD, and these abnormalities would be related to both aspects of disease and memory performance. Through use of resting state connectivity, we can directly address the performance and network activation patterns separately, and then determine the degree to which they are related. Furthermore, to control for variance in age, as well as duration, severity, and number of episodes of illness, we elected to conduct the most conservative test of the hypothesis of memory difficulties in MDD that we could determine. Namely, we restricted the study to only remitted MDD adults who were not taking any psychotropic medications and studied young adults to restrict developmental variance. We also carefully considered the role of sex on performance, given that known differences exist (i.e., females display enhanced verbal memory) in verbal memory performance (Kramer, Yaffe, Lengenfelder, & Delis, Reference Kramer, Yaffe, Lengenfelder and Delis2003; Lewin, Wolgers, & Herlitz, Reference Lewin, Wolgers and Herlitz2001).
Methods
Participants
The current study was approved by the University of Michigan (UM) and the University of Illinois at Chicago (UIC) Institutional Review Boards, and all participants provided written informed consent. Recruitment for participation was conducted in Ann Arbor, MI, and Chicago, IL, and involved community flyers and Internet advertisements. All participants who met criteria via phone screen were invited for a baseline diagnostic interview, neuropsychological testing, and fMRI. Remitted MDD criteria include a history of at least one major depressive episode and a score less than 7 on the Hamilton Depression Rating Scale (HDRS; Hamilton, Reference Hamilton1960). The Beck Depression Inventory-II (Beck, Steer, & Brown, Reference Beck, Steer and Brown1996) was administered before neuropsychological testing and the day of scanning to confirm remitted status. Diagnosis of MDD history, as well as mean age of depression onset, number of depressive episodes, and a history of psychiatric hospitalization, were determined with the Diagnostic Interview for Genetic Studies (DIGS), and remitted MDD (rMDD) participants were required to be medication free for 30 days before their fMRI scan.
Healthy control (HC) exclusion criteria included a history of MDD or any other Axis I or II psychiatric disorder, or a first-degree relative with a history of psychiatric illness. Participants in either group with a history of substance abuse or dependence within the previous 6 months were also excluded. The final sample included 34 rMDDs (17 UM, 17 UIC) and 23 HCs (14 UM, 9 UIC) participants ranging in age from 18 to 23 years old. There were no significant differences between groups in age, sex, education, or estimated verbal IQ. The two groups did statistically differ in their HDRS scores. Residual symptoms may be in part related to any effects observed. Notably, though, a two-point mean difference in HDRS at this low of a score is not clinically meaningful (see Table 1 for demographic and clinical information).
Demographic and clinical information by group
*This mean difference was statistically significant, although clinically these are very low symptom levels (below 7 is consider full remission).
a Estimated from the Shipley-2 Verbal subtest administered in Superlab.
Semantic List Learning Test (SLLT) and the Neuropsychological Battery
This SLLT has been previously described by Weisenbach and colleagues (Reference Weisenbach, Kassel, Rao, Weldon, Avery, Briceño and Langenecker2014) and Langenecker et al. (Reference Langenecker, Caveney, Persad and Giordani2004). In brief, the SLLT is a learning and memory task composed of three blocks—encoding, distraction, and silent rehearsal—with 12 lists completed over four different scanning runs (3 lists per run). There is an additional recall and recognition phase conducted immediately post-scan, outside of the scanner. During encoding, participants were presented with a semantically related word list (e.g., tools, vegetables, etc.) containing 14 words. Participants were prompted with the name of the semantic category for 3.5 s at the start and then words were presented one at a time for 1 s each with a 1- to 4-s jittered inter-stimulus interval.
They were instructed to read each word silently to themselves without moving their lips and to try to remember each word, using the semantic cue to aid in encoding and in recall phases. During the interval, a fixation cross was displayed and each encoding block lasted 58.25 s. Following each encoding block, participants completed the distractor task for 14 s, which involved pressing a key every time an x, y, or z was presented in a serial stream of letters. This distractor block was included to prevent recency effects during delayed recall/recognition by inhibiting rehearsal of items in short-term memory (Brown, Reference Brown1958, Peterson & Peterson, Reference Peterson and Peterson1959, Schallmo et al., Reference Schallmo, Kassel, Weisenbach, Walker, Guidotti-Breting, Rao and Langenecker2015). This distraction prevents initial consolidation, which then carries over to any later recall or recognition prompts.
All participants received the same word lists, although the presentation of the words within a list and the order of the three lists within each run were randomized. At the end of each of the four runs, there was a 32-s rest period. Immediately following the scan, participants were given sheets of paper with all of the category cues and told to write down all the words they could recall from each semantic category (delayed recall). Finally, they were given a written list of all of the previously presented words (168 words in all) and 210 foil words, and told to circle those words they had seen before (recognition). Performance dependent variables for all analyses were number of cued recalled words minus false positive recalled words, and number of recognized words minus recognition false positive words. The d’ was also calculated for recall and recognition.
The neuropsychological battery consisted of the Shipley-2 Verbal subtest (Shipley, Gruber, Martin, & Klein, Reference Shipley, Gruber, Martin and Klein2009), California Verbal Learning Test-2nd edition (Delis, Kramer, Kaplan, & Ober, Reference Delis, Kramer, Kaplan and Ober2000) semantic and phonemic fluency, Benton Visual Form Discrimination (Benton, Hamsher, Varney, & Spreen, Reference Benton, Hamsher, Varney and Spreen1978), Purdue Pegboard (Lezak, Reference Lezak1995), Digit Symbol Coding (Wechsler, Reference Wechsler2008), Trail Making Test (Reitan, Reference Reitan1992), Stroop Color/Word Test (Stroop, Reference Stroop1938), and the Wisconsin Card Sorting Test (Grant & Berg, Reference Grant and Berg1948).
fcMRI Acquisition
The resting state scan was acquired following the administration of the SLLT. MR data were acquired at two imaging sites: UM and UIC. Scans taking place at UM involved an eyes-open resting state scan acquired over 8 min on a 3.0 Tesla (T) GE Signa scanner using a T2*-weighted single shot reverse spiral sequence ]29 four-mm-thick slices, echo time (TE)=30 ms; repetition time (TR)=2000 ms; flip angle=90°; field of view (FOV)=20; matrix=64×64]. Eyes-open resting scans at UIC were also acquired over 8 min on a 3.0T GE Discovery scanner using parallel imaging with ASSET and T2* gradient-echo axial echo planar imaging sequence (44 three-mm-thick slices, TE=22.2 ms; TR=2000 ms; flip angle=90°; FOV=22; matrix=64×64). At both sites, participants were told to focus on a fixation cross, and high-resolution T1 anatomic scans were obtained for spatial normalization. Due to potential artifacts of using two different scanners with different acquisition parameters, we previously examined between-site differences in resting state connectivity and found that the influence of site on fMRI findings was minimal (see Jacobs et al., Reference Jacobs, Jenkins, Gabriel, Barba, Ryan, Weisenbach and Welsh2014, for details).
fcMRI Preprocessing
Slice timing was completed with SPM8, and motion detection algorithms were applied using FSL. After realignment, volumes with movement or artifacts greater than 1.5 mm over three volumes or less were interpolated. Only one subject required interpolation. Then, frame to frame (FTF) values in pitch, roll, and yaw deviations (as well as SDs of these mcflirt adjustments in FSL) were examined and compared between groups (ns, ps<.31). Five HCs and two rMDDs had more than one FTF movement exceeding 0.5 mm, although the mean FTF for both groups was less than one. Analyses were conducted with and without these seven individuals, with no notable differences in results. As such, these seven individuals were included in all analyses. Those excluded due to excessive movement in our prior report were not included for any part of this experiment (Jacobs et al., Reference Jacobs, Jenkins, Gabriel, Barba, Ryan, Weisenbach and Welsh2014). Of note, mean of standard deviation in pitch was significantly correlated with FTF values exceeding 0.5 (r=0.52; p<.05). Structural images were co-registered to functional images and then the co-registered T1-SPGR underwent spatial normalization to the Montreal Neurological Institute (MNI) template. The resulting normalization matrix was then applied to the slice-time-corrected, movement corrected, time series data and smoothed with a 5-mm Gaussian kernel. Resulting T2* images were 2 mm on a side with isotropic voxels.
Cross-Correlation Analysis
The seeds of interest were the left and right anterior hippocampus, using the following coordinates: −30 −12 −18 and 30 −12 −18, respectively. These two spherical regions of interest (ROIs) (2.9 mm radius, 19 voxels) were created in the MarsBaR toolbox (Brett, Anton, Valabregue, & Poline, Reference Brett, Anton, Valabregue and Poline2002) and were defined in MNI space. These coordinates were confirmed on the MNI brain visually, as well as upon an average T1 brain of all subjects, and were visually confirmed to have minimal overlap with the amygdala on the average brain. Spatially averaged time course data were extracted from these ROIs for each participant. Correlation coefficients between mean time course for the two seed regions and all other voxels of the brain were calculated, resulting in a three-dimensional correlation coefficient image (r image). These r images were transformed to Z-scores using a Fisher transformation and were used in independent samples t tests conducted in SPM8. AlphaSim correction (1000 iterations) was used for analysis, balancing height (p<.005) and extent (440 mm3) thresholds to achieve a whole brain correction of p<.05. All analyses included use of sex as a covariate.
In-network masks were derived from the HC correlational image for both seeds and were applied to each contrast (HC>rMDD, in network; HC<rMDD, out of network) to determine in-network versus out-of-network activation regions for each seed. Importantly, there are functional implications about whether group differences might be observed in memory supportive regions/networks, or “in-network,” or in other, potentially extraneous regions that may or may not be supportive of memory performance, or “out of network.” MarsBaR was used to extract mean Z values from each cluster of significant differences between groups separately for each seed and person. These values were then correlated with SLLT delayed recall and recognition performance for each group and the Hamilton Depression Rating Scale (rMDD only). Regression analysis was also used to examine the relationship between connectivity in both groups using both the left and right hippocampal seeds and SLLT cued recall and recognition performance. Due to the limitations of the SPM software, the direction of the analyses used SLLT memory performance to predict hippocampal connectivity; however, the nature of the correlation remains the same within a cross-sectional study like this. It is assumed that connectivity is in fact predicting performance.
Results
Behavioral Analysis of SLLT and Neuropsychological Performance
When examining both delayed recall and recognition performance on the SLLT, the scores were calculated by subtracting the number of false positives from the number of correct target hits. In doing so, the two groups significantly differed in their performance, with the HC group demonstrating a greater number of targets recalled (F(1,55)=5.6; p=0.021; d=0.65) and recognized (F(1,55)=6.0; p=.018; d=0.67). However, when calculating the sensitivity index for both SLLT delayed cued recall and recognition, d’ was not significantly different between groups. When comparing the HC and rMDD groups on performance while controlling for sex, the diagnostic groups statistically differed in the total number of both recall [F(1,55)=24.5; p<.001; d=0.73] and recognition [F(1,55)=53.4; p<.001; d=0.73 ], with the HC group having better delayed recall and recognition scores. Similarly, d’ was significantly different between groups for both SLLT delayed cued recall [F(1,54)=4.59; p=.037; d=0.58] and recognition [F(1,54)=5.33; p=.025; d=0.63] while controlling for sex.
There were no significant sex differences in SLLT delayed recall [F(1,54)=0.037; p=.849; HC: d=0.4; rMDD: d=0.4] or recognition [F(1,54)=0.002; p=.966; HC: d=0.3, rMDD: d=0.3) performance, covarying for group, see Table 2. The HC and the rMDD groups did not statistically differ in proportion of individuals by sex, however, there was still an imbalance of males (fewer) and females in both the rMDD and HC groups, as seen in Table 1. Therefore, these behavioral data were transformed into sex-corrected Z-scores, using separate male and female means and standard deviations. The following results all used these sex-corrected Z-scores when assessing SLLT performance predicting connectivity differences between groups. There were no significant differences in performance between groups on all neuropsychological tests, see Table 2. We also present the serial position curve values for each group and each sex in Figure 4.
SLLT and neuropsychological test performance raw data by group, means and standard deviations
a Where p value is significant.
Within Network fMRI Connectivity: Differences between Groups and Relationship to Memory
The network was derived from the HC pattern of connectivity for both the left and right hippocampal seeds, to highlight those regions that should be most directly related to memory performance. Within the network, there were no areas that demonstrated increased connectivity in the rMDD group compared to HC. However, there were numerous areas of increased within-network connectivity in the HC group versus the rMDD group (see Figures 1 and 2). The HC group demonstrated significantly greater connectivity between the left hippocampus and the right superior frontal region compared to rMDD. Additionally, there was greater connectivity between the right hippocampus and the left middle and superior frontal regions, the left inferior temporal region, the left insula, and the right posterior cingulate in the HC group compared to rMDD. Only one of these differential regions (HC>rMDD) was significantly related to SLLT performance. Specifically, in only the HC group, the connectivity between the right hippocampus and the left superior frontal region was positively correlated with both SLLT delayed recall (r=0.46; p=.029) and recognition performance (r=0.48; p=.02) (see Table 3).
Whole group baseline connectivity covaried by sex, left hippocampal seed. Combined HC and rMDD connectivity (aqua); HCs demonstrating greater connectivity than rMDD within network (lime green); rMDD demonstrating greater connectivity than HCs extra-network (red). Images are in ascending order: z = −39, −31, −23, −15, −7, 1, 9, 17, 25, 33, 41, 49.
Whole group baseline connectivity covaried by sex, right hippocampal seed. Combined HC and rMDD connectivity (aqua); HCs demonstrating greater connectivity than rMDD within network (lime green); rMDD demonstrating greater connectivity than HCs extra-network (red). Images are in ascending order: z = −38, −31, −24, −17, −10, −3, 4, 11, 18, 25, 32, 39.
Within network differences in sex-corrected resting state connectivity between remitted MDD and healthy controls and relationship to sex-corrected SLLT delayed cued recall (dcr) and recognition (recog) performance
*Denotes p<.05.
Extra-Network fMRI Connectivity: Differences between Groups and Relationship to Memory
Likely due to using HC connectivity as the network mask, there were no regions outside the network that demonstrated greater connectivity in the HC group than in rMDD. Extra, or out-of-network areas are not as strongly hypothesized to support memory processes. There were multiple extra-network ROIs demonstrating significantly greater connectivity to the left and right hippocampal seeds in the rMDD group (reported in Table 4, seen in Figures 1 and 2). Specifically, the left hippocampus exhibited greater connectivity to bilateral anterior cingulate, left paracentral lobule, middle frontal, and superior temporal regions, as well as to the right medial frontal, posterior cingulate, and supramarginal regions.
Extra-network differences in sex-corrected resting state connectivity between remitted MDD and healthy control and relationship to sex-corrected SLLT delayed cued recall (DCR) and recognition (Recog) performance
*Denotes p<.05.
When correlating the strength of connectivity between these left hippocampal connections (rMDD>HC) and SLLT performance and the HDRS, the rMDD group demonstrated a positive relationship with the right anterior cingulate and both SLLT delayed recall and recognition (r=0.35, p=.041 and r=0.36, p=.035, respectively). There was also a positive correlation between left hippocampus to right supramarginal connectivity with SLLT recognition performance (r=0.35; p=.043). Additionally, in rMDD greater connectivity of the left hippocampus to the left anterior cingulate was related to higher HDRS scores (i.e., greater depression, r=0.41; p=.016). The right hippocampus was more highly connected to the left middle frontal region in the rMDD group, and this relationship was not correlated with SLLT performance or HDRS scores.
Regression Analysis Using SLLT Delayed Recall Performance to Predict Hippocampal Connectivity
There were numerous significant ROIs that demonstrated a negative relationship with SLLT delayed recall performance in the HC group, covarying sex. That is, SLLT delayed recall performance was negatively associated with increasing connectivity in the right lingual gyrus and culmen of the cerebellum to the left hippocampal seed. Using the right hippocampus as the seed, a negative effect was demonstrated in multiple left medial frontal regions and in the left inferior frontal, middle, and inferior temporal, and postcentral regions, as well as the right precentral and paracentral lobule, superior temporal gyrus, and uncus. Significance was also reached in bilateral putamen and the left anterior lingual gyrus of the cerebellum. Additionally, the strength of connectivity between these significant ROIs in both the HC and rMDD groups did not significantly differ in direct Fischer’s t tests (see Table 5). There were no positive relationships between SLLT delayed recall performance and HC connectivity using either the left or right hippocampal seeds (see Table 5, Figure 3).
Predicting connectivity from SLLT performance by group using sex corrected cued recall-false positive scores. Panel A represents HC resting state connectivity pattern for negative cued recall for the (i) left hippocampal seed and (ii) the right hippocampal seed. Panel B represents resting state connectivity pattern in rMDD for positive cued recall for the (i) left hippocampal seed and (ii) the right hippocampal seed.
Regression analysis predicting connectivity from SLLT delayed cued recall scores (sex-corrected), using an in-network (HC) mask
In rMDD, sex-corrected SLLT delayed recall performance was positively associated with connectivity in the left middle temporal region when using the left hippocampal seed. Additionally, positive relationships were also demonstrated with the right hippocampal seed: the right parahippocampus/amygdala, the right claustrum, and the left putamen, as well as the right declive of the cerebellum. Similar to the above, the strength of connectivity between these significant ROIs in both the HC and rMDD groups did not significantly differ in direct Fischer’s t tests (see Table 5). There were no significant negative relationships in rMDD between SLLT delayed recall performance and connectivity using either the left or right hippocampal seeds (see Table 5, Figure 4).
Recall serial position curve of the SLLT by group and gender.
Finally, additional regressions using the significant connectivity ROIs from Table 5 to predict SLLT delayed cued recall performance in both groups were conducted to examine the effect of scanner site. Findings reveal that site was not a significant covariate in the regression model (see Supplementary Table 1).
Additionally, a principal components analysis was conducted for each group for all of the significant ROIs from Table 5, for data reduction purposes. For the HCs, three factors emerged: (PC1) loadings included connectivity from the right hippocampus to the left inferior and medial frontal regions, left inferior and middle temporal regions, and right putamen; (PC2) loadings included connectivity from the right hippocampus to the left putamen and postcentral regions, and to the right uncus, paracentral, precentral, and superior temporal regions; and (PC3) included loadings from the left hippocampal seed to the right culmen and lingual gyrus of the cerebellum and from the right hippocampal seed to the left anterior lingual gyrus of the cerebellum and the left medial frontal gyrus. For the rMDD, only one factor emerged from the PCA. These factors were then correlated with other neuropsychological measures, and results found that for HCs, PC1 negatively correlated with CVLT recognition, and Digit Symbol Coding, PC2 negatively correlated with the number of perseverative errors on the WCST, and PC3 positively correlated with Trail Making Test-Part B. In rMDD, connectivity data positively correlated with verbal fluency performance and negatively correlated with Trail Making Test-Part A (see Supplementary Table 2).
Regression Analysis Using SLLT Recognition Performance to Predict Hippocampal Connectivity
When correcting for sex, there were numerous significant ROIs that demonstrated a negative relationship with SLLT recognition performance in the HC group. That is, SLLT recognition performance was negatively associated with increasing connectivity in the right fusiform gyrus and bilateral posterior cingulate to the left hippocampal seed. Using the right hippocampus as the seed, a negative effect was demonstrated in the bilateral postcentral gyrus and putamen, as well as the right precentral, inferior frontal, superior temporal gyri, insula, precuneus. Additionally, the right hippocampal seed demonstrated a negative effect in connectivity to numerous left middle temporal regions, and to the left medial frontal and parahippocampal gyri. Of note, the strength of connectivity between the right hippocampal seed and the right inferior frontal gyrus, right insula, and left parahippocampal gyrus was significantly different between HC and rMDD groups in direct Fischer’s t tests (see Table 6). There were no positive relationships between SLLT recognition performance and HC connectivity using either the left or right hippocampal seeds (see Table 6).
Regression analysis predicting connectivity from SLLT recognition scores (sex-corrected), using an in-network (HC) mask
a Regions of significant overlap with the regions reported in Table 5 (recall).
In rMDD, sex-corrected SLLT recognition performance was positively associated with connectivity in bilateral fusiform gyrus when using the left hippocampal seed. Additionally, positive relationships were also demonstrated with the right hippocampal seed: the left medial frontal, fusiform, and superior temporal gyri, as well as the left amygdala and putamen, right parahippocampus and claustrum, and the right anterior culmen of the cerebellum. The strength of connectivity between these significant ROIs in both the HC and rMDD groups did not significantly differ in direct Fischer’s t tests (see Table 6). There were no significant negative relationships in rMDD between SLLT recognition performance and connectivity using either the left or right hippocampal seeds (see Table 6).
Discussion
This exploratory study sought to examine the relationship between resting state functional connectivity and semantically cued list learning performance in remitted MDD. Our results indicate that, in relation to disrupted performance on a semantically cued episodic memory task, individuals in remission from MDD demonstrate different resting state connectivity patterns when compared to their healthy age-matched counterparts. Specifically in remitted MDD, there are hypoconnectivities between the left hippocampus and right superior frontal gyrus and between the right hippocampus and multiple fronto-temporal regions, such as the left inferior and superior frontal, left inferior temporal, and left insula, as well as the right posterior cingulate. When examining extra-network connectivity, rMDD demonstrated hyperconnectivities between the left hippocampus and multiple fronto-parietal regions and the right caudate, as well as hyperconnectivity between the right hippocampus and the left middle frontal region. By and large, these decreased/increased connectivity levels in rMDD were not related to performance.
Finally, regression analyses revealed that memory performance was a significant predictor of connectivity: in HCs, memory performance was a negative predictor of connectivity between the right hippocampus and multiple “in-network” fronto-temporo-parietal regions, and in rMDD memory performance positively predicted connectivity between the right hippocampus and “in-network” subcortical structures including the amygdala. So there appears to be differential connectivity patterns between in-network regions that aid performance in MDD and, surprisingly detract from performance in HC. In direct comparisons, none of these correlations for cued recall (and only two for recognition), although in opposite directions for rMDD (positive) and HC (negative), were significantly different between groups. Yet, there are also “in network” hippocampal connectivity regions that are of decreased connectivity in rMDD and “out of network” regions that are positively correlated with performance. This suggests that those rMDD who have the pattern of connectivity most similar to extra-network compensation have retained memory performance, and that lower connectivity within network is linked to poorer memory, but not in a linear manner.
These findings are consistent with previous research by Tahmasian et al. (Reference Tahmasian, Knight, Manoliu, Schwerthöffer, Scherr, Chun and Sorg2013) who found reduced connectivity between the hippocampus and amygdala to the dorsomedial-prefrontal cortex and fronto-insular operculum in actively depressed individuals. In addition to these weaker connections interfering with learning and memory, they speculated that these weaker connections may also provide a mechanism for the emergence of depressive symptoms, such as increased self-focus, that are mediated by these particular functional networks (Tahmasian et al., Reference Tahmasian, Knight, Manoliu, Schwerthöffer, Scherr, Chun and Sorg2013).
Despite experiencing minimal active depressive symptoms, those with a history of depression continue to demonstrate worse semantically cued episodic memory than healthy controls, and show persistent disruptions of network integrity, particularly in connections to the hippocampus, after controlling for sex. Specifically, remitted MDD is associated with greater extra-network connectivities with the hippocampus when compared to controls. Of interest, the majority of the connections that differed between groups, largely connections to frontal and cingulate regions, were not related to SLLT performance or symptom severity (Tables 3 and 4). Additionally, the in-network connections in the HC group that are negatively predictive of performance (Table 5) are not the same regions that demonstrate greater connectivity in HC relative to rMDD (Table 3), which is consistent with a stronger SLLT performance in the HC group. This would suggest a disease-specific effect on connectivity that is consistent with the literature demonstrating that depression is associated with aberrant functional connectivity in the prefrontal cortex (Sheline et al., Reference Sheline, Price, Yan and Mintun2010).
Additionally, network connectivity is differentially related to memory performance in rMDD in comparison to HC. That is, resting state network connectivity in rMDD was positively predicted by episodic memory performance, though episodic memory performance was a negative predictor of network hippocampal connectivity in HCs. Given this positive relationship between network connectivity and episodic memory performance in rMDD, it is possible that this widespread pattern of hippocampal connections in rMDD may reflect compensation. In other words, those with rMDD must develop greater hippocampal connections to enhance their performance. Our results are consistent with previous work indicating that, in early disease course, hyperconnectivity is associated with disease resilience, or that there is a negative relationship between hippocampal connectivity and disease duration and severity (Zhang et al., Reference Zhang, Wang, Wu, Kuang, Huang, He and Gong2011). Nonetheless, individuals with rMDD, when controlled for sex, continued to perform more poorly on the task than HCs. Therefore, it remains unclear why hippocampal network integrity is differentially associated with memory performance in comparison to healthy individuals. These findings highlight how not all connectivity relationships will necessarily be additive or helpful in facilitating performance on a specific task across a backdrop of a myriad of disease and non-disease baseline functions.
In addition, sex had a small to medium effect on semantically cued episodic memory performance. It is important that the effect of sex on verbal memory performance is not overlooked, as many previous studies have demonstrated superior verbal memory abilities in females when compared to males (e.g., Kramer et al., Reference Kramer, Yaffe, Lengenfelder and Delis2003; Lewin et al., Reference Lewin, Wolgers and Herlitz2001). Additionally, previous studies have also shown sex differences in semantic categorization, which our semantically cued memory test is reliant upon (Pasterski, Zwierzynska, & Estes, Reference Pasterski, Zwierzynska and Estes2011). Therefore, we determined that sex should be a covariate when examining behavioral performance. If males and females were combined, the increased variance would have masked important group differences. Given the exploratory nature of this study, replication of these findings is needed, with continued scrutiny of sex differences in semantically cued episodic memory.
One limitation of this study is that it examined resting state functional connectivity using only two seeds of interest, the left and right hippocampus; however, it is clear that memory performance, especially in MDD, is mediated by multiple other brain regions in the Papez circuit (Papez, Reference Papez1937). Additionally, given the cross-sectional nature of this study, we are unable to make specific predictions about whether resting state connectivity patterns during remission are associated with future depressive episodes or memory impairment. Therefore, following these at-risk individuals over time will likely provide greater insight into protective versus risk factors. Finally, it is worth noting that, although our remitted MDD participants were not actively depressed according to the diagnostic interview, their scores on the self-administered HDRS did reveal significantly higher scores than the healthy control group. Although this two-point elevation is not clinically significant, it is possible that these remitted MDD individuals may have demonstrated different connectivity to performance relationships while actively depressed, and may still exhibit residual differences in these relationships at the time of this evaluation. Further analyses with a larger sample may help to reduce the discrepancy in symptom severity.
Of note, the memory test used for this study, the SLLT, was selected for its potential clinical application, seeing as how the one-trial learning is a format that more closely follows that observed in real life (i.e., single verbal instruction for a list of items to purchase at the grocery story). It is atypical to more traditional memory probes of repeated list learning, tests designed to test for impaired consolidation and rapid forgetting in the context of dementia (Delis et al., Reference Delis, Kramer, Kaplan and Ober2000; Schmidt, Reference Schmidt1996).
Overall, this exploratory study suggests that even when no overt depressive symptoms are present, and when controlling for sex, individuals with a history of depression demonstrate disruptions in their semantically cued episodic memory, as well as disruptions in hippocampal connectivity. Additionally, there is a differential relationship between hippocampal network integrity and semantically cued episodic memory performance in those with rMDD when compared to healthy individuals, which is intriguing and requires replication and further investigation.
Acknowledgments
There are no competing interests to declare for all authors. This work was supported by the BRAINS RO1 (to S.A.L.; MH 091811).
Supplementary Material
To view supplementary material for this article, please visit 10.1017/S1355617716000011
