Abstract

Neural systems may be characterized by measuring functional interactions in the healthy brain, but it is unclear whether components of systems defined in this way share functional properties. For instance, within the medial temporal lobes (MTL), different subregions show different patterns of cortical connectivity. It is unknown, however, whether these intrinsic connections predict similarities in how these regions respond during memory encoding. Here, we defined brain networks using resting state functional connectivity (RSFC) then quantified the functional similarity of regions within each network during an associative memory encoding task. Results showed that anterior MTL regions affiliated with a network of anterior temporal cortical regions, whereas posterior MTL regions affiliated with a network of posterior medial cortical regions. Importantly, these connectivity relationships also predicted similarities among regions during the associative memory task. Both in terms of task-evoked activation and trial-specific information carried in multivoxel patterns, regions within each network were more similar to one another than were regions in different networks. These findings suggest that functional heterogeneity among MTL subregions may be related to their participation in distinct large-scale cortical systems involved in memory. At a more general level, the results suggest that components of neural systems defined on the basis of RSFC share similar functional properties in terms of recruitment during cognitive tasks and information carried in voxel patterns.

INTRODUCTION

A major development in cognitive neuroscience research has been the identification of neural systems on the basis of functional interactions in the healthy brain. For example, fMRI data have been used to identify networks of brain regions showing spontaneous fluctuations in hemodynamic activation that are correlated over time, referred to as “resting state functional connectivity” (RSFC; Power et al., 2011; Yeo et al., 2011; Fox & Raichle, 2007). This approach stands in contrast to traditional patient- or task-fMRI-based approaches, which have aimed to link particular brain regions, usually in isolation, to specific cognitive functions. It has been proposed that cortical function is intimately related to connectivity with the rest of the brain: that is, functional differences between regions are related to differences in connectivity, and regions with similar “connectional fingerprints” have related cognitive functions (Passingham, Stephan, & Kötter, 2002). However, in many cognitive domains, there are no clear data linking RSFC-defined networks, which are consistent with known connectivity patterns (Yeo et al., 2011; Damoiseaux & Greicius, 2009), to task-fMRI results, which can ascertain whether regions are involved in similar aspects of cognition.

For instance, research on the neural organization of memory has predominately focused on neighboring structures in the medial temporal lobes (MTL), including the hippocampal formation (HF), parahippocampal cortex (PHC), and perirhinal cortex (PRC; Eichenbaum, Yonelinas, & Ranganath, 2007; Squire & Zola-Morgan, 1991). Although RSFC profiles of MTL subregions have not been extensively characterized, some recent evidence suggests that these regions may be situated within large-scale neural systems that encompass cortical areas both within and outside the MTL. For instance, the PHC and PRC differentially affiliate with separate sets of cortical structures (Libby, Ekstrom, Ragland, & Ranganath, 2012; Kahn, Andrews-Hanna, Vincent, Snyder, & Buckner, 2008) and even with different subregions of the HF (Libby et al., 2012). One possibility is that differences in connectivity are related to differences in memory function among MTL structures (Montaldi & Mayes, 2010; Diana, Yonelinas, & Ranganath, 2007; Eichenbaum et al., 2007; Davachi, 2006), but at present, it is unknown whether functional connectivity relationships revealed by RSFC are indicative of similar functional roles in memory processes. We hypothesize that differences in RSFC profiles among MTL subregions reflect their participation in large-scale cortical systems that play separable roles in memory.

Here, we used fMRI to test this hypothesis by identifying networks of regions during rest and then investigating the functional properties of these networks during associative memory encoding. Associative memory encoding has previously been shown to drive activity in the MTL as well as in connected cortical regions (Kim, 2011). We incorporated three complementary analysis approaches to characterize relationships between brain regions: RSFC, univariate task activation analyses, and multivoxel pattern information analyses. The RSFC analyses provided an estimate of the intrinsic functional connectivity between regions. The task activation analyses quantified task-related similarities between regions, as measured by univariate activation estimates aggregated across many trials (similar to a conventional general linear model approach). The pattern information analyses also quantified task-related similarities between regions, but did so by measuring trial-by-trial changes in multivariate spatial patterns. Whereas the task activation analysis identified regions that would be likely to be coactivated during a particular task condition, the pattern information analysis was sensitive to information specific to individual trials. Because we were interested in the similarities between connectivity and task relationships, rather than the differences, it was beneficial to use analytic approaches that are distinct from one another. This aim can be contrasted with that of prior investigations comparing intrinsic and task-related functional connectivity measures to highlight task modulation of network dynamics (Fornito, Harrison, Zalesky, & Simons, 2012; Nir, Hasson, Levy, Yeshurun, & Malach, 2006).

First, we delineated neural systems on the basis of RSFC by applying a data-driven community detection algorithm. Next, we investigated whether the relationships characterized in the RSFC analysis could account for the roles of different brain regions in memory encoding. Specifically, we tested the prediction that regions within the same RSFC-defined system should exhibit more similar profiles of task activation and multivoxel pattern information than regions in different systems.

METHODS

Participants

Data from 19 young adults (11 women; ages 19–30) were included in the final analyses. An additional two participants completed the study but were excluded from analysis because of insufficient variability in memory performance (i.e., fewer than 10 remembered or forgotten items in any condition). Participants reported that they were native English speakers, free of neurological and psychiatric disorders, and eligible for MRI.

Experimental Design

Participants completed a 7-min resting state scan, three 10-min task scans, and a postscan behavioral test. During the resting state scan, the computer screen was black with a white fixation cross at center, and participants were instructed to stay awake with their eyes open. The task scans consisted of an incidental associative encoding task. During each of the three task scans, 84 sentences appeared on-screen for 4 sec each, separated by jittered fixation intervals (mean = 4 sec, range = 2–10 sec). Each sentence included a concrete noun referring to an object and a fact about the object, which could describe its appearance (turquoise, purple, soft, or bumpy), its situational context (contest prize, birthday present, new purchase, or rental), or its spatial location (yogurt shop, pizzeria, science lab, lecture hall). For example, one possible sentence might read, “The apple is in the lecture hall.” These three conditions were selected to maximize the likelihood that we would observe differential engagement of regions involved in encoding different aspects of memory, including spatial context and item features. Object nouns were trial-unique, but facts were drawn from the same set of 12, repeated throughout the experiment. Participants were instructed to rate on a continuum the extent to which the object–fact pairing was unusual or common (i.e., 1 = unusual, 2 = neither unusual nor common, 3 = common).

The postscan behavioral test consisted of an associative memory test for each of the sentences studied in the scanner. For each trial, the object word was presented at the center of the screen along with four options, which corresponded to the four possible facts associated with the studied task condition. Participants were instructed to press the button corresponding to the associated fact or to press the space bar if they did not know the answer. They were explicitly instructed not to guess, and thus, we limited our analyses to binary comparisons of correct and “do not know” responses. This part of the experiment was self-paced.

Image Acquisition and Preprocessing

Scanning was performed on a Siemens Skyra 3T scanner system with a 32-channel head coil. High-resolution T1-weighted structural images were acquired using a magnetization prepared rapid acquisition gradient-echo pulse sequence (field of view = 25.6 cm, image matrix = 256 × 256, 208 axial slices with 1.0 mm thickness). Functional images were acquired using a gradient EPI sequence (repetition time = 2000 msec, echo time [TE] = 25 msec, field of view = 20.5 × 21.14 cm, image matrix = 64 × 66, flip angle = 90, 34 interleaved axial slices, voxel size = 3.20 × 3.20 × 3.20 mm). Field maps were also collected using the Siemens field map sequence with short TE = 4.92 msec and long TE = 7.38 msec and used to correct for geometric distortions because of magnetic field inhomogeneities.

SPM8 (www.fil.ion.ucl.ac.uk/spm/software/spm8/) was used to preprocess the images, including slice-timing correction, realignment, field map correction, normalization, and smoothing. The high-resolution T1 image was skull-stripped via segmentation. Functional images were corrected for slice timing, realigned (motion correction), and unwarped (field map correction). The mean functional was coregistered to the skull-stripped anatomical image, moving all of the functional images in register with the anatomical image. At this point, the anatomical and functional images were warped to a group-derived template generated using diffeomorphic registration (DARTEL) and normalized to MNI space. Functional images were smoothed with a 6-mm Gaussian kernel for RSFC analyses and univariate task analyses. Skull-stripped anatomical images were also warped and smoothed for use as an explicit mask for subsequent functional analyses. Quality assurance included the identification of “suspect” time-points via the Artifact Detection Tools (ART; www.nitrc.org/projects/artifact_detect), defined as time points marked by greater than 1 mm in movement or 2% global mean signal change. Runs with more than 3 mm total movement in any direction were also excluded from analysis (resulting in two participants who each had one task run excluded).

Data Analysis

RSFC Analysis

RSFC was calculated for a set of ROIs using in-house scripts in MATLAB (The MathWorks, Inc., Natick, MA). ROIs were defined as nonoverlapping spheres with a 6-mm radius centered on predefined coordinates identified from the comparison of PRC and PHC RSFC reported in Libby et al. (2012), including regions showing greater PRC than PHC connectivity and vice versa, and restricted to peaks spaced at least 12 mm apart (Table 1). For each participant, functional time series from the first resting state scan were extracted from each of the 56 a priori ROIs, as well as from masks of white matter and CSF. Time series were corrected for linear trends, bandpass filtered for frequencies of 0.01–0.1 Hz, and mean-centered. ART suspects were scrubbed from the time series, along with time points that deviated more than 3 SD from the ROI mean. Pairwise correlations (Pearson's r) of ROI time series were computed controlling for white matter and CSF time series and six motion parameters. This procedure resulted in a 56 × 56 RSFC matrix reporting the RSFC among the ROIs, which served as the basis for community detection. Community detection is a graph analytic technique designed to subdivide a set of brain regions into discrete systems (or “modules”), such that within-module correlations are stronger than between-module correlations (Rubinov & Sporns, 2010; Newman, 2006; Girvan & Newman, 2002). Modules were identified via the Louvain modularity algorithm (Blondel, Guillaume, Lambiotte, & Lefebvre, 2008) implemented in the Brain Connectivity Toolbox (Rubinov & Sporns, 2010; www.brain-connectivity-toolbox.net). This algorithm determined module assignments such that within-module connections were stronger than what would be expected by chance if connections were arbitrarily distributed (Blondel et al., 2008). Community detection was run on the group-averaged, weighted RSFC matrix thresholded to exclude negative correlation values (34.9% of possible connections). A range of other thresholds was explored and yielded similar results. Correlations between ROIs placed less than 20 mm apart were excluded to minimize the influence of short-range connections (2.2% of possible connections), which might be especially susceptible to motion artifacts. Because the output of this algorithm can vary slightly from run to run, it was run for 500 iterations to ensure stability. Solutions were summarized by assigning two regions to the same module if they had been coclassified on at least half of the iterations producing that solution, thus identifying the most frequent coclassifications of ROIs.

Table 1. 

Information about the Regions Included in the Analysis

Matrix Row Index
Label for Figures
MNI Coordinates
Module Assignment
x
y
z
3-Module
4-Module
PTHAL4 22 −30 1a 
MOCC1 −2 −78 −2 1a 
OCCP1 −16 −96 22 1a 
OCC2 16 −98 20 1a 
PREC3 18 −68 24 1a 
MOCC3 14 −72 1a 
OCCP2 14 −88 1a 
PHIPP1 −20 −30 −8 1a 
PHIPP2 20 −30 −6 1a 
10 PHC1 −14 −50 −6 1a 
11 PHC2 18 −46 −4 1a 
12 PREC1 −14 −60 18 1b 
13 MOCC2 −58 14 1b 
14 OCC1 −38 −82 28 1b 
15 ANG2 52 −48 28 1b 
16 PREC5 −2 −60 34 1b 
17 RSC1 −8 −50 14 1b 
18 PREC2 −12 −50 40 1b 
19 RSC2 −46 16 1b 
20 PREC4 18 −52 36 1b 
21 DLPFC1 −24 60 24 1b 
22 MPFC −2 60 34 1b 
23 DLPFC2 18 58 24 1b 
24 TPC1 −44 −42 
25 PMTG2 54 −2 −32 
26 AHIPP2 22 −4 −26 
27 OFC2 −6 16 −22 
28 AHIPP1 −24 −12 −30 
29 OFC3 24 12 −24 
30 TPC2 38 20 −40 
31 OFC1 −16 24 −20 
32 PMTG1 −64 −36 −10 
33 AITG2 64 −14 −28 
34 FPC2 38 60 −10 
35 FUS2 40 −18 −28 
36 FPC1 −44 58 −18 
37 PRC 30 −12 −36 
38 FUS1 −42 −14 −30 
39 OFC4 22 −20 
40 PMTG3 70 −38 −10 
41 ANG1 −40 −48 50 
42 AITG2 −56 −32 −24 
43 POST2 42 −46 42 
44 OCCP3 −8 −58 −30 
45 IFG2 54 38 
46 PSTG2 66 −44 
47 TPJ 60 −40 34 
48 VLPFC1 54 18 12 
49 IFG3 56 26 −8 
50 IFG1 −58 20 −2 
51 PSTG1 −66 −36 24 
52 PRE3 60 
53 POST3 62 −10 16 
54 POST1 −44 −36 52 
55 PRE1 −34 −20 54 
56 PRE2 −52 12 
Matrix Row Index
Label for Figures
MNI Coordinates
Module Assignment
x
y
z
3-Module
4-Module
PTHAL4 22 −30 1a 
MOCC1 −2 −78 −2 1a 
OCCP1 −16 −96 22 1a 
OCC2 16 −98 20 1a 
PREC3 18 −68 24 1a 
MOCC3 14 −72 1a 
OCCP2 14 −88 1a 
PHIPP1 −20 −30 −8 1a 
PHIPP2 20 −30 −6 1a 
10 PHC1 −14 −50 −6 1a 
11 PHC2 18 −46 −4 1a 
12 PREC1 −14 −60 18 1b 
13 MOCC2 −58 14 1b 
14 OCC1 −38 −82 28 1b 
15 ANG2 52 −48 28 1b 
16 PREC5 −2 −60 34 1b 
17 RSC1 −8 −50 14 1b 
18 PREC2 −12 −50 40 1b 
19 RSC2 −46 16 1b 
20 PREC4 18 −52 36 1b 
21 DLPFC1 −24 60 24 1b 
22 MPFC −2 60 34 1b 
23 DLPFC2 18 58 24 1b 
24 TPC1 −44 −42 
25 PMTG2 54 −2 −32 
26 AHIPP2 22 −4 −26 
27 OFC2 −6 16 −22 
28 AHIPP1 −24 −12 −30 
29 OFC3 24 12 −24 
30 TPC2 38 20 −40 
31 OFC1 −16 24 −20 
32 PMTG1 −64 −36 −10 
33 AITG2 64 −14 −28 
34 FPC2 38 60 −10 
35 FUS2 40 −18 −28 
36 FPC1 −44 58 −18 
37 PRC 30 −12 −36 
38 FUS1 −42 −14 −30 
39 OFC4 22 −20 
40 PMTG3 70 −38 −10 
41 ANG1 −40 −48 50 
42 AITG2 −56 −32 −24 
43 POST2 42 −46 42 
44 OCCP3 −8 −58 −30 
45 IFG2 54 38 
46 PSTG2 66 −44 
47 TPJ 60 −40 34 
48 VLPFC1 54 18 12 
49 IFG3 56 26 −8 
50 IFG1 −58 20 −2 
51 PSTG1 −66 −36 24 
52 PRE3 60 
53 POST3 62 −10 16 
54 POST1 −44 −36 52 
55 PRE1 −34 −20 54 
56 PRE2 −52 12 

Univariate Activation Analyses

Task activation was evaluated through a univariate analysis approach implemented in SPM8. Univariate analyses were conducted on normalized and smoothed functional images from the three task scans. Event-related stick function regressors were used to model trials corresponding to one of seven conditions: appearance hits, situational hits, spatial hits, appearance misses, situational misses, spatial misses, and others. Hits, or “remembered” items, were defined as trials receiving a correct response during associative memory retrieval. Misses, or “forgotten” items, were defined as trials receiving either the “don't know” or no response during associative memory retrieval. The “others” condition included any trial receiving no response during encoding or an incorrect response during associative memory retrieval. Six motion parameter regressors were included in the model. Nuisance regressors were also included to model time-points identified as ART suspects.

Subject-level contrasts corresponding to each of the six conditions of interest (relative to baseline activity during the intertrial interval) were estimated for each participant. For ROI-based analyses, mean contrast estimates were extracted from each ROI, yielding a six-element activation vector for each ROI. To test whether ROIs in the same module shared similar activation profiles, these activation vectors were correlated (using Pearson's r) between ROIs, resulting in a 56 × 56 activation similarity matrix for each participant. Activation similarity matrices were further analyzed by averaging activation similarity for pairs of ROIs within the same RSFC module (within-module) versus pairs of ROIs in different RSFC modules (between-module). Correlation values were Fisher-transformed and compared via paired t tests.

To further characterize the function of each module, activation vectors were averaged across all ROIs in the same module, resulting in a summary activation profile for each module and each participant. Separately for each module, summary activation profiles were submitted to a group-level repeated-measures ANOVA testing the effects of Association Type (appearance, situational, spatial) and Subsequent Memory Success (remembered, forgotten) on task activation. Finally, for whole-brain group analysis, contrast images comparing the conditions were entered into one-sample t tests. Resulting t maps were thresholded at p < .001, one-tailed, uncorrected, with a 10-voxel cluster extent threshold.

Pattern Information Analysis

Pattern information was evaluated through a multivariate analysis approach using SPM8 and in-house MATLAB scripts. These analyses were conducted on normalized but otherwise unsmoothed functional images. Single trial models were generated to estimate the response to each individual trial (n = 252 per participant), resulting in a beta image for every trial. Similar to the procedure described by Mumford, Turner, Ashby, and Poldrack (2012), a separate general linear model was run for each individual trial in SPM8, with the first regressor containing a stick function mapped to the onset of the individual trial and the second regressor containing stick functions modeling all of the other trials, with additional motion and nuisance regressors as described above. For each participant, the voxel-wise pattern of hemodynamic activity within each ROI was extracted from each of the 252 single-trial beta images. Separately for each ROI, trial patterns were correlated with each other using Pearson's r, resulting in a 252 × 252 pattern information profile for each ROI. Thus, pattern information profiles reflected the correlation of trial-specific multivoxel patterns with the patterns associated with every other trial. Only trials resulting in successful associative memory were included to limit analysis to trials in which participants were most likely to be fully engaged in the encoding task. In a secondary control analysis, pattern information profiles were evaluated separately for each task condition (appearance, situational, spatial), and in this analysis, only trials from that condition were included. Trials with an average signal that deviated more than 3 SD from the mean beta series in an ROI were excluded as outliers.

For each participant, the correlation between each pair of pattern information profiles was computed, separately for each run then averaged across the three runs, yielding a 56 × 56 pattern connectivity matrix. Pattern connectivity matrices were further analyzed by averaging pattern connectivity for pairs of ROIs within the same RSFC module (within-module) versus pairs of ROIs in different RSFC modules (between-module). Correlation values were Fisher-transformed and compared via paired t tests.

Model Fit Evaluation

We evaluated how well the task activation data were fit by two alternative frameworks: one in which regions were assigned according to the RSFC-derived modules and one in which MTL regions were removed from these modules and grouped together, consistent with the traditional MTL memory system framework. Multiple regression models were fit to the activation similarity matrices and pattern connectivity matrices for each participant. For the RSFC multiple regression models, there were three categorical regressors indexing comembership within each of the three modules. For the RSFC + MTL models, connections with MTL regions were removed from the three RSFC-derived regressors, and instead, a fourth nonoverlapping regressor coded comembership within the MTL. Adjusted R2 values were taken as a measure of fit for each model in each participant, Fisher-transformed, then compared via paired t tests.

Finally, we investigated similarity solely within the MTL ROIs, including regions in the HF and parahippocampal gyrus. The similarity of activation and pattern information profiles was evaluated for MTL regions assigned to the same RSFC module (within-module) versus different RSFC modules (between-module). Correlation values were Fisher-transformed and compared via paired t tests.

RESULTS

Definition of Resting State Networks via Community Detection

To determine the organization of relationships among MTL structures and other cortical areas, we first applied a data-driven community detection procedure to the resting state data. Community detection is a graph analytic technique designed to subdivide a set of brain regions into discrete systems (or “modules”), such that within-module correlations are stronger than between-module correlations (Rubinov & Sporns, 2010; Newman, 2006; Girvan & Newman, 2002). We defined 56 ROIs both within and outside the MTL based on an independent comparison of PHC and PRC RSFC (Libby et al., 2012). Mean filtered resting state time series were extracted from each region, and the full correlation matrix linking each pair of regions was computed, controlling for motion and other nuisance covariates. This matrix served as the basis for community detection. Results of the community detection algorithm can vary somewhat across iterations, because regions are sorted into modules in a random order until maximal modularity is achieved. Thus, we ran the algorithm 500 times; 47.2% of the iterations returned solutions containing three modules (mean modularity Q = .253), and all others contained four modules (mean modularity Q = .249). For the most part, the fourth module reflected the splitting of one module from the three-module solution into two. For simplicity, we focus here on the three-module solution, which was more consistent with connectivity relationships that have been reported (Libby et al., 2012; Kahn et al., 2008), and briefly describe the four-module results at the end of the Results section. Although the three-module solution is more parsimonious, both solutions support the same general conclusions.

In contrast to the traditional view that emphasizes connections among MTL regions, we found that MTL subregions sorted into separate modules, as shown in Figure 1. Notably, Module 1 contained the PHC and posterior HF along with the retrosplenial cortex, posterior cingulate, and precuneus, whereas Module 2 contained the PRC and anterior HF along with the anterior temporal cortex, amygdala, and OFC. Module 3 did not include any MTL regions but instead included ventral frontal and parietal regions. Modules 1 and 2, in particular, accord with connections identified via anatomical studies of rodents and monkeys (Ranganath & Ritchey, 2012), as well as the results from a prior study in which cortical networks were parcellated using RSFC data from over a thousand participants (Yeo et al., 2011). The presence of Module 3 was an unexpected result but bears some resemblance to a network of frontoparietal regions that has been shown to cooperate with posterior midline regions during recollection (Fornito et al., 2012). Furthermore, its presence suggests that the outcome of the community detection approach was not entirely predicted by direct comparisons of PHC versus PRC connectivity. These results support the hypothesis that the MTL subregions like the PHC and PRC affiliate with distinct neural systems during rest.

Figure 1. 

RSFC and community detection. Resting state time series were extracted from each ROI and correlated with one another, controlling for nuisance covariates such as motion parameters and white matter/CSF signal. Graph analyses were used to determine whether communities existed within the data such that regions were more connected with other regions in the same module than with regions in other modules. (A) This analysis revealed three modules: Module 1, including posterior medial regions (blue); Module 2, including anterior temporal regions (red); and Module 3, including ventral frontal and parietal regions (green). (B) The correlation matrix of region-by-region RSFC entered into the graph analysis, here shown sorted by module assignment. (C) Depiction of the relationships between regions, with nodes (regions) and edges (relationships between regions) colored according to module assignment. Here only graph edges with correlation value of r > .2 are included for visualization purposes, and node layout was determined by a force-directed algorithm implemented in Gephi.

Figure 1. 

RSFC and community detection. Resting state time series were extracted from each ROI and correlated with one another, controlling for nuisance covariates such as motion parameters and white matter/CSF signal. Graph analyses were used to determine whether communities existed within the data such that regions were more connected with other regions in the same module than with regions in other modules. (A) This analysis revealed three modules: Module 1, including posterior medial regions (blue); Module 2, including anterior temporal regions (red); and Module 3, including ventral frontal and parietal regions (green). (B) The correlation matrix of region-by-region RSFC entered into the graph analysis, here shown sorted by module assignment. (C) Depiction of the relationships between regions, with nodes (regions) and edges (relationships between regions) colored according to module assignment. Here only graph edges with correlation value of r > .2 are included for visualization purposes, and node layout was determined by a force-directed algorithm implemented in Gephi.

Evidence for Greater Within- versus Between-network Similarity during Associative Memory Encoding

To test the hypothesis that resting state relationships predict functional similarities during associative memory encoding, we also measured hemodynamic activity while participants completed three runs of an associative encoding task, immediately after the resting state scan. During each task run, participants encoded 84 sentences linking an object to a fact about the object's appearance, situational context, or spatial location (Figure 2A). Immediately after the scan session, participants were asked to remember the fact that was previously associated with each object or to indicate that they did not know. Participants demonstrated above-chance memory performance, selecting more correct (M = .45 ± .12) than incorrect (M = .14 ± .06) facts, F(1, 18) = 87.48, p < .001. They responded “do not know” on the remainder of responses, M = .41 ± .12. Memory accuracy (correct rate minus incorrect rate) was greater for the appearance condition than the situational, F(1, 18) = 13.95, p = .002, or spatial, F(1, 18) = 12.79, p = .002, context conditions, which did not differ, F(1, 18) = .27, p = .61.

Figure 2. 

Activation profiles during associative memory encoding. (A) Depiction of the experimental design. Participants encoded sentences linking an object to an appearance detail (app), situational context (sit), or spatial context (spat). After the scan, they completed an associative recognition task. (B) Activation profiles were measured by modeling the hemodynamic response of each region in response to six task conditions of interest, with factors for task condition (app, sit, spat) and subsequent memory (remembered, forgotten). Region activation profiles were correlated with each other, yielding an activation similarity matrix. (C) Activation similarity matrices were then summarized according to the RSFC module assignment (right). Within-module correlations were greater than between-module correlations, a pattern evident in each participant (left). In the bar plot, error bars denote SEM, and in the box plot, the gray-shaded box denotes standard deviation and the light red-shaded box denotes the 95% confidence interval. (D) Contrast estimates for three of the included regions, each of which were assigned to a different RSFC-defined module: the PHC, PRC, and ventrolateral pFC (VLPFC). (E) Whole-brain results corroborated the finding that many regions in Module 1 were preferentially activated during subsequently remembered spatial context trials compared with appearance trials, whereas many regions in Module 3 showed the reverse effect.

Figure 2. 

Activation profiles during associative memory encoding. (A) Depiction of the experimental design. Participants encoded sentences linking an object to an appearance detail (app), situational context (sit), or spatial context (spat). After the scan, they completed an associative recognition task. (B) Activation profiles were measured by modeling the hemodynamic response of each region in response to six task conditions of interest, with factors for task condition (app, sit, spat) and subsequent memory (remembered, forgotten). Region activation profiles were correlated with each other, yielding an activation similarity matrix. (C) Activation similarity matrices were then summarized according to the RSFC module assignment (right). Within-module correlations were greater than between-module correlations, a pattern evident in each participant (left). In the bar plot, error bars denote SEM, and in the box plot, the gray-shaded box denotes standard deviation and the light red-shaded box denotes the 95% confidence interval. (D) Contrast estimates for three of the included regions, each of which were assigned to a different RSFC-defined module: the PHC, PRC, and ventrolateral pFC (VLPFC). (E) Whole-brain results corroborated the finding that many regions in Module 1 were preferentially activated during subsequently remembered spatial context trials compared with appearance trials, whereas many regions in Module 3 showed the reverse effect.

The next set of analyses relied on task data to determine whether neural systems defined by RSFC are informative with respect to functional engagement during memory encoding. We predicted that, if networks defined by RSFC correspond to functional distinctions, then regions within the same module should be more functionally similar to each other than to regions in different modules. To test this prediction, we used two independent techniques for quantifying the functional similarity of regions during associative encoding, one based on task activation and one based on multivoxel pattern information.

Task Activation Similarity

The task activation analysis investigated between-region similarity of “activation profiles.” Activation profiles were designed to summarize the response of a region across task conditions, as they would be considered in a conventional task-fMRI analysis. Mean activation estimates for each of the six experimental conditions (appearance, situational, and spatial facts for both remembered and forgotten sentences) were extracted from each ROI, resulting in a six-element activation profile vector for each region. Activation profile vectors were correlated for each pair of regions to obtain an activation similarity matrix linking each region to every other region (Figure 2B). In this analysis, two regions would be strongly correlated if they displayed similar profiles of relative activation across experimental conditions, such as greater activity for remembered than forgotten items (or any other comparison of interest). The resulting correlations were then summarized according to whether they were from regions within the same RSFC-defined module versus between different modules. This analysis revealed that that regions that were affiliated during rest also tended to show similar activation profiles during associative encoding, whereas correlations between regions in different modules were relatively weak (Figure 2B). As shown in Figure 2C, within-module correlations were stronger than between-module correlations, t(18) = 9.52, p < .001, a pattern that was apparent for every individual participant. The within- versus between-module difference was also significant when considering each module separately, all ts > 3.4, all ps < .003.

Interestingly, three regions that have been consistently associated with successful memory encoding—the PHC, PRC, and ventrolateral pFC—were assigned to separate modules, and the activation profiles of these regions appeared to be distinct from each other (Figure 2D). To further characterize the function of these networks, activation profiles were averaged across all regions within a module, separately for each participant. Module activation profiles were then submitted to repeated-measures ANOVAs to evaluate the influence of Association Type (appearance, situational, spatial) and Memory Success (remembered, forgotten) on task activation. Module 1 was more active while learning about spatial location relative to appearance or situational context, F(2, 36) = 52.88, p < .001, especially when spatial location was later remembered, F(2, 36) = 6.55, p = .004. Although Module 2 activation was not significantly modulated by either factor, ps > .05, Module 3 activation was sensitive to association type, F(2, 36) = 10.62, p < .001, with stronger responses to the appearance than situational or spatial conditions. Whole-brain, voxel-wise comparisons of successfully encoded spatial location and appearance trials revealed similar dissociations, in that regions in Module 1 (including the PHC, posterior HF, retrosplenial cortex, and posterior midline) responded more to spatial than appearance trials, whereas regions in Module 3 (including ventrolateral prefrontal and parietal cortex) showed the reverse effect (Figure 2E).

Pattern Information Similarity

The next analysis investigated between-region similarity of “pattern information profiles,” testing whether brain regions within the same module carry similar information in the trial-to-trial variability of spatial (voxel-wise) activation patterns. In contrast to task activation analyses, which rely on univariate activation estimates averaged across several trials, pattern information analyses are based on the multivariate spatial patterns evoked by each individual trial, independent of response magnitude (Kriegeskorte, Mur, & Bandettini, 2008). Thus, although task activation and pattern information profiles rely on the same fMRI data, they are mathematically independent. The pattern information technique results in enhanced sensitivity to trial-specific information, and previous studies have shown that voxel patterns carry information about stimuli that are encoded or retrieved on a given trial (Ritchey, Wing, Labar, & Cabeza, 2013; Jenkins & Ranganath, 2010; Xue et al., 2010; Polyn, Natu, Cohen, & Norman, 2005). For every region, multivoxel patterns of activation in response to every trial were correlated with each other, resulting in a pattern information profile indexing the relationship of each trial to every other trial, controlling for activation magnitude and between-run differences. To assess similarities in information encoded by each region, pattern information profiles were compared for every pair of regions (Figure 3B), yielding a “pattern connectivity” matrix indicating the similarity of pattern information carried by each pair of regions (Figure 3C). If two regions have a strong correlation between pattern information profiles, then their multivoxel spatial patterns are thought to code similar types of information across trials (Kriegeskorte et al., 2008). As in the task activation analysis, resulting pattern connectivity estimates were summarized according to whether they were within- versus between-module.

Figure 3. 

Pattern information profiles. (A) Patterns of single-trial parameter estimates were extracted from each region for every trial and then correlated with each other, yielding a trial similarity matrix or pattern information profile per region. (B) Pattern information profiles for the PHC and retrosplenial cortex from a single run in a representative participant. Pattern information profiles were then correlated with each other, resulting in a pattern connectivity matrix (C) that relates regions on the basis of their multivariate pattern information. The pattern connectivity matrix, as shown here, is organized by RSFC module assignment. (D) Pattern connectivity was summarized according to RSFC module assignment (right), showing that within-module similarity was greater than between-module similarity, evident in each participant (left). In the bar plot, error bars denote SEM, and in the box plot, the gray-shaded box denotes standard deviation and the light red-shaded box denotes the 95% confidence interval.

Figure 3. 

Pattern information profiles. (A) Patterns of single-trial parameter estimates were extracted from each region for every trial and then correlated with each other, yielding a trial similarity matrix or pattern information profile per region. (B) Pattern information profiles for the PHC and retrosplenial cortex from a single run in a representative participant. Pattern information profiles were then correlated with each other, resulting in a pattern connectivity matrix (C) that relates regions on the basis of their multivariate pattern information. The pattern connectivity matrix, as shown here, is organized by RSFC module assignment. (D) Pattern connectivity was summarized according to RSFC module assignment (right), showing that within-module similarity was greater than between-module similarity, evident in each participant (left). In the bar plot, error bars denote SEM, and in the box plot, the gray-shaded box denotes standard deviation and the light red-shaded box denotes the 95% confidence interval.

Complementing the task activation analyses, regions within the same module had comparable pattern information profiles during associative encoding (Figure 3C). As shown in Figure 3D, within-module correlations were stronger than between-module correlations, t(18) = 20.45, p < .001, a pattern that was again apparent in each participant. The within- versus between-module difference was also significant when considering each module separately, all ts > 10, all ps < .001. To verify that these patterns were not somehow driven by overall differences between task conditions, we re-ran the analysis including trials from only one task condition at a time. Within-module correlations remained stronger than between-module correlations for each comparison, all ts > 9, all ps < .001. This suggests that in addition to showing comparable activation across experimental conditions, as shown by the activation analysis, regions within the same module carried similar information in trial-specific voxel patterns.

Correspondence between Resting State Relationships and Functional Similarities during Memory Encoding

Correlations between RSFC and Task Matrices

The convergence of each of the task-based methods (task activation and pattern information) with the RSFC data was quantified by correlating the region-by-region similarity matrices that emerged from each analysis. The RSFC matrix correlated with both the task activation similarity (mean r = .256, range = .067–.478) and pattern similarity (mean r = .265, range = .140–.373) matrices, suggesting that the relationships identified during rest were present in the task profiles. Although it is unclear whether these correlations reflect the intrinsic nature of these networks or the engagement of memory-related processes during rest, this convergence suggests that RSFC networks can provide insight into interregional similarities during memory encoding.

Evaluating the Fit of RSFC-derived Systems to Task Data

Although these data suggest that RSFC-derived systems can successfully predict regional similarities in memory-related activation, they do not necessarily indicate an improvement over frameworks that group the MTL as a separate memory system. That is, one could hypothesize that the activation and pattern similarity matrices calculated from memory encoding task data might be better explained by a model in which MTL subregions were excluded from the three RSFC-defined modules and assigned instead to a separate module, consistent with the traditional views of the MTL. Importantly, however, grouping the MTL regions together resulted in a worse fit to the activation similarity matrix than splitting them apart, t(18) = 2.58, p = .019, and made no significant improvement for the fit to the pattern connectivity matrix, t(18) = .85, p = .41 (Figure 4A). These findings suggest that treating the MTL as a separate system does not better account for the observed correlations, as compared with the RSFC-derived module assignment.

Figure 4. 

Fit of RSFC and RSFC + MTL memory system solutions to memory task data. (A) Multiple regression was used to assess the degree to which activation similarity and pattern connectivity matrices were explained by the three RSFC-defined modules (“RSFC”) or a model that included the three RSFC-defined modules, but with MTL regions excluded and grouped as a fourth module (“RSFC + MTL”). The percentage variance explained by the RSFC modules was significantly higher than that for the RSFC + MTL regressors (top), and there was no difference in fit for the pattern connectivity data (bottom). (B) Within the MTL, there was greater similarity among subregions within the same RSFC-defined module than between subregions in different modules, for both task activation (top) and pattern information (bottom) profiles. In all box plots, the gray-shaded box denotes standard deviation and the light red-shaded box denotes the 95% confidence interval.

Figure 4. 

Fit of RSFC and RSFC + MTL memory system solutions to memory task data. (A) Multiple regression was used to assess the degree to which activation similarity and pattern connectivity matrices were explained by the three RSFC-defined modules (“RSFC”) or a model that included the three RSFC-defined modules, but with MTL regions excluded and grouped as a fourth module (“RSFC + MTL”). The percentage variance explained by the RSFC modules was significantly higher than that for the RSFC + MTL regressors (top), and there was no difference in fit for the pattern connectivity data (bottom). (B) Within the MTL, there was greater similarity among subregions within the same RSFC-defined module than between subregions in different modules, for both task activation (top) and pattern information (bottom) profiles. In all box plots, the gray-shaded box denotes standard deviation and the light red-shaded box denotes the 95% confidence interval.

To further examine the utility of the RSFC-derived module assignment in understanding the organization of the MTL, we next investigated similarity solely within the MTL ROIs. We predicted that, if the module assignment accurately reflected heterogeneity within the MTL, then we should see greater similarity among MTL regions that were assigned to the same module (correlations among PRC and anterior HF, PHC and posterior HF, etc.) than among regions that were assigned to different modules (PRC and PHC, anterior and posterior HF, etc.). Consistent with these predictions, we found that both activation profiles, t(18) = 4.16, p < .001, and pattern information profiles, t(18) = 8.35, p < .001, were more highly correlated between MTL regions within the same RSFC-derived module than between MTL regions that were in different modules (Figure 4B). These findings indicate that RSFC-derived networks account for functional heterogeneity within the MTL.

Consistency of Findings across Both Community Detection Solutions

The RSFC community detection procedure identified two parcellation schemes: one containing three modules (detailed above) and one containing four modules. Critically, the results from the four-module solution do not change any of the major conclusions drawn from the three-module solution. As shown in Table 1, we found that MTL subregions sorted into separate modules for both solutions. For the most part, the fourth module reflected the splitting of Module 1 from the three-module solution into two modules: one including the PHC, posterior HF, and medial occipital cortex and another including the retrosplenial cortex, posterior cingulate, and precuneus. The dorsomedial frontal cortex regions from Module 2 also joined this latter module. Otherwise module assignments remained the same as in the three-module solution.

For both the task activation and pattern information analyses, grouping regions according to the four-module solution did not qualitatively change the results: within-module correlations were greater than between-module correlations for measures of task activation similarity, t(18) = 9.78, p < .001, and pattern connectivity, t(18) = 30.32, p < .001. Like the three-module solution, the four-module solution revealed that regions were more functionally similar within a system than between systems, both in terms of memory-related activation and multivoxel pattern information. Likewise, the results of the regression analysis remained qualitatively the same, in that grouping the MTL regions together resulted in a worse fit to the activation similarity matrix than splitting them apart, t(18) = 2.17, p = .043, and made no significant improvement for the fit to the pattern connectivity matrix, t(18) = 1.29, p = .21. Thus, regardless of which solution was adopted, RSFC-derived systems were able to account for regional similarities in both task activation and pattern information.

DISCUSSION

RSFC is often used to identify neural systems in humans, but the identification of a network during rest does not necessarily imply that its components exhibit similar properties during memory encoding or that different RSFC-defined networks exhibit different properties. Although anatomists have argued that the connectivity and function of cortical regions are intimately related (Passingham et al., 2002), this idea has not been extensively tested, particularly in cortical networks thought to contribute to memory. Here, we combined RSFC and task-based neuroimaging approaches, which are usually employed in isolation, to empirically characterize memory systems in the human brain. Despite inherent differences in these approaches, the convergence across methods was striking: RSFC-defined networks were marked by task-related functional similarities within a network and dissimilarities between networks, patterns that were observed for every participant in our sample.

Establishing the correspondence between functional connectivity and task-evoked activation is an essential step for translating what we have learned from data-driven approaches in systems neuroscience to task-oriented approaches in cognitive neuroscience. RSFC-defined networks are being increasingly used to identify ROIs (e.g., Wig et al., 2013) and mark changes in neural processing because of experimental manipulations (e.g., Tambini, Ketz, & Davachi, 2010) or disease (Greicius & Kimmel, 2012; Greicius, 2008). Prior attempts to ascertain the cognitive relevance of RSFC-defined networks have focused on measuring the spatial overlap between these networks and task-related activation patterns collected in separate samples (e.g., Crossley et al., 2013; Andrews-Hanna, Reidler, Sepulcre, Poulin, & Buckner, 2010; Nelson et al., 2010; Smith et al., 2009; Vincent et al., 2006). This approach has been useful for generating predictions about the function of RSFC-defined networks but does not directly address the extent to which intrinsic connectivity can account for interregional similarity in functional contributions to cognitive tasks. The present results indicate that functional connectivity relationships among regions identified during rest are highly informative with respect to the functional properties of these regions identified during memory encoding. Regions in the same RSFC-defined network had similar univariate activation profiles, indicating that regions that are connected during rest are likely to be identified by the same contrast in a conventional general linear model approach. Multivoxel patterns within connected regions are also likely to be affected similarly across trials. Taken together, we suggest that task-related fMRI data can be seen and interpreted through the lens of functional networks, in that task differences may reflect the differential deployment of large-scale cortical systems.

The combination of RSFC and task approaches might help to resolve some discontinuities in the literature on the neural bases of memory. Traditionally, memory has been linked to the function of the MTL, such that the PHC and PRC are grouped together with the HF and treated as being functionally distinct from cortical areas outside the MTL (Diana et al., 2007; Eichenbaum et al., 2007; Davachi, 2006; Squire & Zola-Morgan, 1991). Recent work has highlighted functional differences between different MTL subregions, especially the PHC and PRC (Montaldi & Mayes, 2010; Diana et al., 2007; Eichenbaum et al., 2007; Davachi, 2006). However, although it has been noted that these regions receive different visual inputs, little work has been done to directly relate task dissociations within the MTL to activation patterns in cortical areas outside the MTL. Within the neuroimaging literature, it has become apparent that many other cortical areas are heavily involved in memory (Kim, 2011; Spaniol et al., 2009) and that their task-related connectivity with the MTL is modulated by memory function (O'Neil et al., 2012; Ranganath, Heller, Cohen, Brozinsky, & Rissman, 2005). We propose that the present methods and results might help to integrate these literatures by constructing a framework in which task-related similarities can be directly compared and shown to correspond to functional connectivity relationships.

In the present data set, ROIs were selected to test specific hypotheses about networks that were expected to differentially affiliate with the PHC and PRC, and thus, we focused only on regions previously shown to differ in their connections to these regions (Libby et al., 2012), rather than regions across the entire brain. However, the community detection analysis assigned these modules entirely on the basis of relative RSFC strength as assessed independently in this study. There was some variance in whether the community detection approach identified three or four modules (based on the random, iterative nature of the algorithm), but both solutions parsed anterior and posterior portions of the MTL into separate systems. The anterior–posterior MTL distinction is consistent with parcellations derived in recent studies that have more comprehensively identified functional networks across the entire brain (Power et al., 2011; Yeo et al., 2011). Also, in both solutions, the within-module similarities outweighed the between-module similarities in terms of task activation and multivoxel pattern information. These results provide corroborating evidence for the idea that anterior and posterior MTL subregions may be part of two different memory systems, consistent with prior lesion work (Alvarado & Bachevalier, 2005; Norman & Eacott, 2005), but also that these systems extend beyond the MTL to include other cortical areas. Indeed, the present results suggest that the systems defined on the basis of RSFC, rather than an a priori grouping of MTL regions, might better account for how these regions are recruited during associative memory encoding.

Altogether, we take this as evidence that functional differences observed within the MTL are shared with large-scale cortical systems. This idea is compatible with hierarchical models that characterize MTL structures as extensions of perceptual pathways devoted to object and scene processing (Graham, Barense, & Lee, 2010; Murray, Bussey, & Saksida, 2007). The relationships observed here may be best understood according to a framework that emphasizes the cortical connections of MTL subregions, rather than their membership within the MTL (Ranganath & Ritchey, 2012; Aggleton, 2011; Kravitz, Saleem, Baker, & Mishkin, 2011; Kondo, Saleem, & Price, 2005). One such framework describes these relationships as a “posterior medial system” and an “anterior temporal system,” respectively (Ranganath & Ritchey, 2012), which correspond closely to Modules 1 and 2 observed here.

The present findings raise many questions that can be addressed in future studies. First, one important question concerns individual differences. In this study, we defined modules on the basis of the group-averaged RSFC matrix and then applied the same set of module assignments to each individual's task data. Although our results demonstrate that group-derived networks are sufficient to account for patterns of task-related similarities, future investigations should assess the stability of these systems across participants and their relationship to individual differences in behavior.

Second, future studies will be needed to better characterize the function of RSFC-defined networks in memory. One possibility is that, during an episode, the posterior medial system is involved in processing spatial and temporal context information (Kravitz et al., 2011; Vann, Aggleton, & Maguire, 2009; Epstein, 2008) and linking that information to internal models of the environment supported by regions in the default network (Buckner, Andrews-Hanna, & Schacter, 2008). The anterior temporal system, in contrast, may be involved in processing information about entities (animate and inanimate objects), including information about the specific properties of a particular entity (Graham et al., 2010; Bussey, Saksida, & Murray, 2005), related semantic concepts (Patterson, Nestor, & Rogers, 2007), and its salience or value (Rushworth, Noonan, Boorman, Walton, & Behrens, 2011; Adolphs, 2010). Here, we observed some correspondence between regions in Module 1 and activation in the spatial condition, especially during successful encoding, consistent with a role for this network in spatial context processing (Ranganath & Ritchey, 2012). Although we originally hypothesized that regions in Module 2 might be involved in encoding information about an item's appearance, they instead showed more complex patterns of recruitment across the different task conditions, precluding a straightforward interpretation of their functional roles. Our experimental design might have contributed to this complexity, in that each trial mixed both item and source information and required associative encoding. As demonstrated in the pattern connectivity analyses, the systems were evident even when only one task condition was considered at a time, suggesting that these systems map onto processes that were shared across the encoding conditions used here. Future studies should employ more distinct experimental tasks (e.g., episodic memory vs. semantic memory) or stimuli (e.g., object vs. scene encoding), which might better elicit consistent differences between these systems.

Third, an ongoing challenge for fMRI studies is to account for differences in cerebral vascularization, which could inflate or obscure true relationships between regions or between analytic measures. In the present data set, vascularization patterns are insufficient to explain the observed relationships between regions, in that module assignment did not accord with patterns of arterial supply, even at the grossest level. Each observed network included regions that are contained within different arterial territories (Duvernoy, 1999), and the standard vasculature of the MTL, in which both anterior and posterior MTL regions are supplied by the posterior cerebral artery (Duvernoy, 2005), is incompatible with the separation we observed here. Vascularization patterns are also unlikely to explain the correspondence among analytic measures, which vary in terms of their sensitivity to changes in blood flow. That is, whereas the RSFC and activation similarity analyses incorporate univariate activation estimates that are sensitive to increases in blood flow, the pattern connectivity analysis relies on multivoxel patterns that are insensitive to changes in activation magnitude. Thus, a simple vascularization account is inconsistent with both the observed relationships among regions and the degree of convergence among methods. Nonetheless, we cannot be certain that vascularization did not partially influence the present results, and as such, future work could incorporate both hemodynamic and/or electrophysiological approaches such as intracranial recordings (e.g., as in Keller et al., 2013).

Finally, although here we focused on similarities between the task activation and pattern connectivity results, there were some differences between these results. For instance, the off-diagonal relationships were stronger in the pattern connectivity analysis than in the task activation analysis, and there was more within-module variability in the pattern connectivity analysis than in the task activation analysis. Because the pattern connectivity approach relies on relationships among individual trials, it may be more sensitive to subtle trial-wise differences that would distinguish among regions in the same module. Future work should clarify the relationship between task activation and pattern information, as well as the circumstances under which regions might have similar activation profiles but carry different pattern information and vice versa.

Recent evidence has suggested that the identification of RSFC-defined neural systems may be relevant to understanding patterns of cortical pathology and memory deficits in neurodegenerative disorders (Greicius & Kimmel, 2012). For instance, semantic dementia and Alzheimer disease are both degenerative disorders that are associated with MTL pathology, but semantic dementia largely impacts semantic knowledge whereas Alzheimer disease disproportionately impacts episodic memory (Nestor, Fryer, & Hodges, 2006). Interestingly, semantic dementia has more severe effects on anterior temporal regions that correspond closely to Module 2, such as the perirhinal and temporopolar cortex (Boxer et al., 2003), whereas Alzheimer disease has more severe effects on regions in Module 1, such as the posterior cingulate and precuneus (Seeley, Crawford, Zhou, Miller, & Greicius, 2009; Boxer et al., 2003). Relating functional connectivity results to cognitive models of memory might be key to identifying memory systems in the healthy brain and understanding their roles in the diseased brain.

Acknowledgments

Funding was provided by the National Institutes of Mental Health grant R01MH083734 to C. R. and A. P. Y. We thank Maria Montchal and Manoj Doss for assistance with data collection and Laura Libby for advice and comments.

Reprint requests should be sent to Maureen Ritchey, Center for Neuroscience, University of California-Davis, 1544 Newton Court, Davis, CA 95618, or via e-mail: meritchey@ucdavis.edu.

REFERENCES

REFERENCES
Adolphs
,
R.
(
2010
).
What does the amygdala contribute to social cognition?
Annals of the New York Academy of Sciences
,
1191
,
42
61
.
Aggleton
,
J. P.
(
2011
).
Multiple anatomical systems embedded within the primate medial temporal lobe: Implications for hippocampal function.
Neuroscience and Biobehavioral Reviews
,
36
,
1579
1596
.
Alvarado
,
M. C.
, &
Bachevalier
,
J.
(
2005
).
Comparison of the effects of damage to the perirhinal and parahippocampal cortex on transverse patterning and location memory in rhesus macaques.
The Journal of Neuroscience: The Official Journal of the Society for Neuroscience
,
25
,
1599
1609
.
Andrews-Hanna
,
J. R.
,
Reidler
,
J. S.
,
Sepulcre
,
J.
,
Poulin
,
R.
, &
Buckner
,
R. L.
(
2010
).
Functional-anatomic fractionation of the brain's default network.
Neuron
,
65
,
550
562
.
Blondel
,
V. D.
,
Guillaume
,
J.-L.
,
Lambiotte
,
R.
, &
Lefebvre
,
E.
(
2008
).
Fast unfolding of communities in large networks.
Journal of Statistical Mechanics: Theory and Experiment
,
2008
,
P10008
.
Boxer
,
A. L.
,
Rankin
,
K. P.
,
Miller
,
B. L.
,
Schuff
,
N.
,
Weiner
,
M.
,
Gorno-Tempini
,
M. L.
,
et al
(
2003
).
Cinguloparietal atrophy distinguishes Alzheimer disease from semantic dementia.
Archives of Neurology
,
60
,
949
956
.
Buckner
,
R. L.
,
Andrews-Hanna
,
J. R.
, &
Schacter
,
D. L.
(
2008
).
The brain's default network: Anatomy, function, and relevance to disease.
Annals of the New York Academy of Sciences
,
1124
,
1
38
.
Bussey
,
T. J.
,
Saksida
,
L. M.
, &
Murray
,
E. A.
(
2005
).
The perceptual-mnemonic/feature conjunction model of perirhinal cortex function.
The Quarterly Journal of Experimental Psychology
,
58
,
269
282
.
Crossley
,
N. A.
,
Mechelli
,
A.
,
Vértes
,
P. E.
,
Winton-Brown
,
T. T.
,
Patel
,
A. X.
,
Ginestet
,
C. E.
,
et al
(
2013
).
Cognitive relevance of the community structure of the human brain functional coactivation network.
Proceedings of the National Academy of Sciences, U.S.A.
,
110
,
11583
11588
.
Damoiseaux
,
J.
, &
Greicius
,
M.
(
2009
).
Greater than the sum of its parts: A review of studies combining structural connectivity and resting-state functional connectivity.
Brain Structure & Function
,
213
,
525
533
.
Davachi
,
L.
(
2006
).
Item, context and relational episodic encoding in humans.
Current Opinion in Neurobiology
,
16
,
693
700
.
Diana
,
R. A.
,
Yonelinas
,
A. P.
, &
Ranganath
,
C.
(
2007
).
Imaging recollection and familiarity in the medial temporal lobe: A three-component model.
Trends in Cognitive Sciences
,
11
,
379
386
.
Duvernoy
,
H. M.
(
1999
).
The human brain: Surface, three-dimensional sectional anatomy with MRI, and blood supply
(2nd ed.).
New York
:
Springer
.
Duvernoy
,
H. M.
(
2005
).
The human hippocampus: Functional anatomy, vascularization, and serial sections with MRI.
New York
:
Springer
.
Eichenbaum
,
H.
,
Yonelinas
,
A. P.
, &
Ranganath
,
C.
(
2007
).
The medial temporal lobe and recognition memory.
Annual Review of Neuroscience
,
30
,
123
152
.
Epstein
,
R.
(
2008
).
Parahippocampal and retrosplenial contributions to human spatial navigation.
Trends in Cognitive Sciences
,
12
,
388
396
.
Fornito
,
A.
,
Harrison
,
B. J.
,
Zalesky
,
A.
, &
Simons
,
J. S.
(
2012
).
Competitive and cooperative dynamics of large-scale brain functional networks supporting recollection.
Proceedings of the National Academy of Sciences, U.S.A.
,
109
,
12788
12793
.
Fox
,
M. D.
, &
Raichle
,
M. E.
(
2007
).
Spontaneous fluctuations in brain activity observed with functional magnetic resonance imaging.
Nature Reviews Neuroscience
,
8
,
700
711
.
Girvan
,
M.
, &
Newman
,
M. E. J.
(
2002
).
Community structure in social and biological networks.
Proceedings of the National Academy of Sciences
,
99
,
7821
7826
.
Graham
,
K. S.
,
Barense
,
M. D.
, &
Lee
,
A. C.
(
2010
).
Going beyond LTM in the MTL: A synthesis of neuropsychological and neuroimaging findings on the role of the medial temporal lobe in memory and perception.
Neuropsychologia
,
48
,
831
853
.
Greicius
,
M.
(
2008
).
Resting-state functional connectivity in neuropsychiatric disorders.
Current Opinion in Neurology
,
21
,
424
430
.
Greicius
,
M. D.
, &
Kimmel
,
D. L.
(
2012
).
Neuroimaging insights into network-based neurodegeneration.
Current Opinion in Neurology
,
25
,
727
734
.
Jenkins
,
L. J.
, &
Ranganath
,
C.
(
2010
).
Prefrontal and medial temporal lobe activity at encoding predicts temporal context memory.
The Journal of Neuroscience: The Official Journal of the Society for Neuroscience
,
30
,
15558
15565
.
Kahn
,
I.
,
Andrews-Hanna
,
J. R.
,
Vincent
,
J. L.
,
Snyder
,
A. Z.
, &
Buckner
,
R. L.
(
2008
).
Distinct cortical anatomy linked to subregions of the medial temporal lobe revealed by intrinsic functional connectivity.
Journal of Neurophysiology
,
100
,
129
139
.
Keller
,
C. J.
,
Bickel
,
S.
,
Honey
,
C. J.
,
Groppe
,
D. M.
,
Entz
,
L.
,
Craddock
,
R. C.
,
et al
(
2013
).
Neurophysiological investigation of spontaneous correlated and anticorrelated fluctuations of the BOLD signal.
The Journal of Neuroscience: The Official Journal of the Society for Neuroscience
,
33
,
6333
6342
.
Kim
,
H.
(
2011
).
Neural activity that predicts subsequent memory and forgetting: A meta-analysis of 74 fMRI studies.
Neuroimage
,
54
,
2446
2461
.
Kondo
,
H.
,
Saleem
,
K. S.
, &
Price
,
J. L.
(
2005
).
Differential connections of the perirhinal and parahippocampal cortex with the orbital and medial prefrontal networks in macaque monkeys.
Journal of Comparative Neurology
,
493
,
479
509
.
Kravitz
,
D. J.
,
Saleem
,
K. S.
,
Baker
,
C. I.
, &
Mishkin
,
M.
(
2011
).
A new neural framework for visuospatial processing.
Nature Reviews Neuroscience
,
12
,
217
230
.
Kriegeskorte
,
N.
,
Mur
,
M.
, &
Bandettini
,
P.
(
2008
).
Representational similarity analysis—Connecting the branches of systems neuroscience.
Frontiers in Systems Neuroscience
,
2
,
4
.
Libby
,
L. A.
,
Ekstrom
,
A. D.
,
Ragland
,
J. D.
, &
Ranganath
,
C.
(
2012
).
Differential connectivity of perirhinal and parahippocampal cortices within human hippocampal subregions revealed by high-resolution functional imaging.
The Journal of Neuroscience: The Official Journal of the Society for Neuroscience
,
32
,
6550
6560
.
Montaldi
,
D.
, &
Mayes
,
A. R.
(
2010
).
The role of recollection and familiarity in the functional differentiation of the medial temporal lobes.
Hippocampus
,
20
,
1291
1314
.
Mumford
,
J. A.
,
Turner
,
B. O.
,
Ashby
,
F. G.
, &
Poldrack
,
R. A.
(
2012
).
Deconvolving BOLD activation in event-related designs for multivoxel pattern classification analyses.
Neuroimage
,
59
,
2636
2643
.
Murray
,
E. A.
,
Bussey
,
T. J.
, &
Saksida
,
L. M.
(
2007
).
Visual perception and memory: A new view of medial temporal lobe function in primates and rodents.
Annual Review of Neuroscience
,
30
,
99
122
.
Nelson
,
S. M.
,
Cohen
,
A. L.
,
Power
,
J. D.
,
Wig
,
G. S.
,
Miezin
,
F. M.
,
Wheeler
,
M. E.
,
et al
(
2010
).
A parcellation scheme for human left lateral parietal cortex.
Neuron
,
67
,
156
170
.
Nestor
,
P. J.
,
Fryer
,
T. D.
, &
Hodges
,
J. R.
(
2006
).
Declarative memory impairments in Alzheimer's disease and semantic dementia.
Neuroimage
,
30
,
1010
1020
.
Newman
,
M. E. J.
(
2006
).
Finding community structure in networks using the eigenvectors of matrices.
Physical Review E
,
74
,
036104
.
Nir
,
Y.
,
Hasson
,
U.
,
Levy
,
I.
,
Yeshurun
,
Y.
, &
Malach
,
R.
(
2006
).
Widespread functional connectivity and fMRI fluctuations in human visual cortex in the absence of visual stimulation.
Neuroimage
,
30
,
1313
1324
.
Norman
,
G.
, &
Eacott
,
M. J.
(
2005
).
Dissociable effects of lesions to the perirhinal cortex and the postrhinal cortex on memory for context and objects in rats.
Behavioral Neuroscience
,
119
,
557
566
.
O'Neil
,
E. B.
,
Protzner
,
A. B.
,
McCormick
,
C.
,
McLean
,
D. A.
,
Poppenk
,
J.
,
Cate
,
A. D.
,
et al
(
2012
).
Distinct patterns of functional and effective connectivity between perirhinal cortex and other cortical regions in recognition memory and perceptual discrimination.
Cerebral Cortex
,
22
,
74
85
.
Passingham
,
R. E.
,
Stephan
,
K. E.
, &
Kötter
,
R.
(
2002
).
The anatomical basis of functional localization in the cortex.
Nature Reviews Neuroscience
,
3
,
606
616
.
Patterson
,
K.
,
Nestor
,
P. J.
, &
Rogers
,
T. T.
(
2007
).
Where do you know what you know? The representation of semantic knowledge in the human brain.
Nature Reviews Neuroscience
,
8
,
976
987
.
Polyn
,
S. M.
,
Natu
,
V. S.
,
Cohen
,
J. D.
, &
Norman
,
K. A.
(
2005
).
Category-specific cortical activity precedes retrieval during memory search.
Science
,
310
,
1963
1966
.
Power
,
J. D.
,
Cohen
,
A. L.
,
Nelson
,
S. M.
,
Wig
,
G. S.
,
Barnes
,
K. A.
,
Church
,
J. A.
,
et al
(
2011
).
Functional network organization of the human brain.
Neuron
,
72
,
665
678
.
Ranganath
,
C.
,
Heller
,
A.
,
Cohen
,
M. X.
,
Brozinsky
,
C. J.
, &
Rissman
,
J.
(
2005
).
Functional connectivity with the hippocampus during successful memory formation.
Hippocampus
,
15
,
997
1005
.
Ranganath
,
C.
, &
Ritchey
,
M.
(
2012
).
Two cortical systems for memory-guided behaviour.
Nature Reviews Neuroscience
,
13
,
713
726
.
Ritchey
,
M.
,
Wing
,
E. A.
,
Labar
,
K. S.
, &
Cabeza
,
R.
(
2013
).
Neural similarity between encoding and retrieval is related to memory via hippocampal interactions.
Cerebral Cortex
,
23
,
2818
2828
.
Rubinov
,
M.
, &
Sporns
,
O.
(
2010
).
Complex network measures of brain connectivity: Uses and interpretations.
Neuroimage
,
52
,
1059
1069
.
Rushworth
,
M. F. S.
,
Noonan
,
M. P.
,
Boorman
,
E. D.
,
Walton
,
M. E.
, &
Behrens
,
T. E.
(
2011
).
Frontal cortex and reward-guided learning and decision-making.
Neuron
,
70
,
1054
1069
.
Seeley
,
W. W.
,
Crawford
,
R. K.
,
Zhou
,
J.
,
Miller
,
B. L.
, &
Greicius
,
M. D.
(
2009
).
Neurodegenerative diseases target large-scale human brain networks.
Neuron
,
62
,
42
52
.
Smith
,
S. M.
,
Fox
,
P. T.
,
Miller
,
K. L.
,
Glahn
,
D. C.
,
Fox
,
P. M.
,
Mackay
,
C. E.
,
et al
(
2009
).
Correspondence of the brain's functional architecture during activation and rest.
Proceedings of the National Academy of Sciences, U.S.A.
,
106
,
13040
13045
.
Spaniol
,
J.
,
Davidson
,
P. S.
,
Kim
,
A. S.
,
Han
,
H.
,
Moscovitch
,
M.
, &
Grady
,
C. L.
(
2009
).
Event-related fMRI studies of episodic encoding and retrieval: Meta-analyses using activation likelihood estimation.
Neuropsychologia
,
47
,
1765
1779
.
Squire
,
L. R.
, &
Zola-Morgan
,
S.
(
1991
).
The medial temporal lobe memory system.
Science
,
253
,
1380
1386
.
Tambini
,
A.
,
Ketz
,
N.
, &
Davachi
,
L.
(
2010
).
Enhanced brain correlations during rest are related to memory for recent experiences.
Neuron
,
65
,
280
290
.
Vann
,
S. D.
,
Aggleton
,
J. P.
, &
Maguire
,
E. A.
(
2009
).
What does the retrosplenial cortex do?
Nature Reviews Neuroscience
,
10
,
792
802
.
Vincent
,
J. L.
,
Snyder
,
A. Z.
,
Fox
,
M. D.
,
Shannon
,
B. J.
,
Andrews
,
J. R.
,
Raichle
,
M. E.
,
et al
(
2006
).
Coherent spontaneous activity identifies a hippocampal-parietal memory network.
Journal of Neurophysiology
,
96
,
3517
3531
.
Wig
,
G. S.
,
Laumann
,
T. O.
,
Cohen
,
A. L.
,
Power
,
J. D.
,
Nelson
,
S. M.
,
Glasser
,
M. F.
,
et al
(
2013
).
Parcellating an individual subject's cortical and subcortical brain structures using snowball sampling of resting-state correlations.
Cerebral Cortex
.
Xue
,
G.
,
Dong
,
Q.
,
Chen
,
C.
,
Lu
,
Z.
,
Mumford
,
J. A.
, &
Poldrack
,
R. A.
(
2010
).
Greater neural pattern similarity across repetitions is associated with better memory.
Science
,
330
,
97
101
.
Yeo
,
B. T.
,
Krienen
,
F. M.
,
Sepulcre
,
J.
,
Sabuncu
,
M. R.
,
Lashkari
,
D.
,
Hollinshead
,
M.
,
et al
(
2011
).
The organization of the human cerebral cortex estimated by intrinsic functional connectivity.
Journal of Neurophysiology
,
106
,
1125
1165
.