What Goes Down Must Come Up: Role of the Posteromedial Cortices in Encoding and Retrieval

Abstract

The hypothesis that the neural network supporting successful episodic memory retrieval overlaps with the regions involved in episodic encoding has garnered much interest; however, the role of the posteromedial regions remains to be fully elucidated. Functional magnetic resonance imaging (fMRI) studies during successful encoding typically demonstrate deactivation of posteromedial cortices, whereas successful retrieval of previously encoded information has been associated with activation of these regions. Here, we performed an event-related fMRI experiment during an associative face–name encoding and retrieval task to investigate the topography and functional relationship of the brain regions involved in successful memory processes. A conjunction analysis of novel encoding and subsequent successful retrieval of names revealed an anatomical overlap in bilateral posteromedial cortices. In this region, a significant negative correlation was found: Greater deactivation during encoding was related to greater activation during successful retrieval. In contrast, the hippocampus and prefrontal cortex demonstrated positive activation during both encoding and retrieval. Our results provide further evidence that posteromedial regions constitute critical nodes in the large-scale cortical network subserving episodic memory. These results are discussed in relation to the default mode hypothesis, the involvement of posteromedial cortices in successful memory formation and retention, as well as potential implications for aging and neurodegenerative disease.

Introduction

Episodic memory, the conscious mental process of forming, retaining, and recalling personally experienced events within a particular spatio-temporal context (Tulving 1983; Eichenbaum and Cohen 2001), has long been thought to depend on the medial temporal lobe memory system (Scoville and Milner 1957; Squire et al. 1992; Cohen and Eichenbaum 1993) as well as on the prefrontal cortices (Stuss and Benson 1984; Shimamura 1995). Although current models of episodic memory generally agree that these regions support the encoding of novel information as well as the retrieval of previously learned information (Alvarez and Squire 1994; Moscovitch et al. 2005), an emerging body of functional imaging evidence suggests that parietal regions, including the lateral posterior parietal cortex and precuneus extending into the posterior cingulate and retrosplenial cortices, may also have a pivotal role in episodic memory function (Rugg et al. 2002; Shannon and Buckner 2004; Cabeza et al. 2008; Miller et al. 2008; Spaniol et al. 2009; Uncapher and Wagner 2009). Until now, relatively few functional neuroimaging studies have directly compared encoding and retrieval processes within the same study. The majority of these studies have looked at similarities in activation patterns during these 2 processes. According to an influential cognitive theory, “the reinstatement hypothesis,” successful retrieval of episodic memory requires the reactivation of the same region activated during encoding of that information (Craik et al. 1996). To date, several neuroimaging studies have provided evidence in support for this hypothesis (e.g., Nyberg et al. 2000; Wheeler et al. 2000). However, a recent publication has demonstrated an exception of this hypothesis by showing an opposing relationship between encoding and retrieval (Daselaar et al. 2009). That is, using functional magnetic resonance imaging (fMRI) the authors found an overlap of regions, including posteromedial cortex, that is deactivated during encoding and subsequently activated during retrieval (Daselaar et al. 2009). These and other recent findings suggest the involvement of several different brain regions, involving both task-positive and task-negative networks, during successful episodic memory processes (Kim et al. 2009). In this study, we wanted to investigate whether brain regions known to be involved in episodic memory may have different functional contribution during encoding versus retrieval. In particular, we sought to determine whether the pattern of brain activity in the posteromedial cortices during encoding compared with retrieval processes differed from the brain activity during these processes in more traditional components of the memory network, specifically hippocampal and prefrontal regions.

The importance of medial and lateral parietal contributions to multiple cognitive processes is a rapidly emerging avenue of investigation in cognitive neuroscience, particularly in human neuroimaging studies. These parietal cortices, as well as the medial temporal and medial prefrontal regions, and other midline cortical regions have been implicated in the “default mode network,” whose function seems to be maximally engaged during rest and suspended during performance of focused cognitive processing (Shulman et al. 1997; Mazoyer et al. 2001; Raichle et al. 2001; McKiernan et al. 2003; Greicius and Menon 2004; Buckner et al. 2008). Interestingly, the frequent involvement of these regions reported in numerous cognitive tasks has lead to the belief that they may constitute a single “core network,” which underlies a number of brain functions including remembering, prospection, spatial navigation, and theory of mind (Buckner and Carroll 2007; Spreng et al. 2009). With regard to episodic memory in particular, several fMRI studies have shown deactivation in the parietal cortex, especially the posteromedial cortex extending into the posterior cingulate during memory encoding. Subsequently, in accordance with the default mode hypothesis, the observed deactivations during these experiments have been interpreted as a result of reallocation of neuronal resources needed for efficient cognitive processing, which contribute to successful task performance. To date, multiple fMRI studies have provided support for this idea by demonstrating decreased activity in the posteromedial cortices for stimuli that are subsequently remembered (successful encoding) than for those that are subsequently forgotten (Otten and Rugg 2001; Wagner and Davachi 2001; Daselaar et al. 2004; Kao et al. 2005; Miller et al. 2008; Shrager et al. 2008). These results have been interpreted as reflecting “beneficial” deactivation during successful encoding (Daselaar et al. 2004) as well as representing processes detrimental during encoding of forgotten items (Otten and Rugg 2001; Wagner and Davachi 2001). Additional support for the idea that deactivation is pivotal for successful memory formation has also been shown in studies comparing groups of individuals with different level of memory performance. A consistent finding in these studies is the observation of less deactivation in the posteromedial regions in healthy elderly and in patients with early Alzheimer’s disease (AD) compared with younger subjects (Lustig et al. 2003; Sperling et al. 2003; Greicius et al. 2004; Celone et al. 2006; Grady et al. 2009), and there is also evidence suggesting that failure of deactivation in these groups is related to worse memory performance (Miller et al. 2008; Pihlajamaki et al. 2009).

Interestingly, as mentioned above, the posteromedial cortices have not only been implicated in the default mode network but also has been shown to be involved in self-referential and reflective activity (Gusnard and Raichle 2001; Greicius et al. 2003; Johnson et al. 2006, 2009), which specifically includes episodic memory retrieval, prospective thinking, autobiographical memory, mental images, emotions, and inner speech (Mazoyer et al. 2001; Addis et al. 2004, 2007; Greicius et al. 2004; Svoboda et al. 2006; Burianova et al. 2010). Importantly, with regard to successful retrieval, event-related fMRI studies have shown greater activation in the posteromedial cortices in proportion to the strength or certainty of the memory decision (Wagner et al. 2005; Svoboda et al. 2006), indicating that this region seems to be particularly involved in subjective judgments about the memory retrieval, including “remember vs. know” assessments and metamemory processes, such as postretrieval confidence ratings (Wheeler and Buckner 2004; Chua et al. 2006). In addition, this region has also been shown to exhibit greater activation for stimuli that are correctly classified as “old” (retrieval hit) than when they are incorrectly classified as “new” (retrieval miss) (Henson et al. 2005; Prince et al. 2005; Wagner et al. 2005), providing evidence that these regions are also engaged in the processes supporting successful episodic memory retrieval.

Taken collectively, these previous findings of dissociation between encoding and retrieval studies in the posteromedial cortices suggest that the brain may be switching between 2 distinct functions supported by the same set of brain regions: one that is task negative and supports encoding processes and one that is task positive and is engaged during successful retrieval processes (Fox et al. 2005; Buckner and Carroll 2007; Spreng et al. 2009). The relationship between beneficial deactivation during encoding and activation during successful retrieval in this region has not been fully defined. In particular, there is a need to investigate the inverse functional contribution of the posteromedial cortices within subjects and within the same experiment to determine whether deactivation during encoding is related to subsequent activity during retrieval for successful memory processes. In addition, we wanted to compare the pattern of activity in the posteromedial cortices with that of other brain regions thought to be critical for successful memory formation and retention but which, interestingly, may not demonstrate a differential pattern of activity during encoding versus retrieval processes.

Here, we sought to further increase our knowledge of the potential role of the posteromedial cortices with respect to episodic memory by jointly investigating the brain activity across 2 event-related fMRI tasks, namely task-related deactivation during encoding and task-related activation during cued recall. Specifically, at encoding, a cross-modal associative memory task was employed involving forming novel face–name associations (Sperling et al. 2001). We have previously reported that this task reliably activates the hippocampus and deactivates a specific set of brain regions during successful encoding in young and cognitively intact elderly adults, particularly in posteromedial cortex (Sperling et al. 2003; Rand-Giovannetti et al. 2006; Miller et al. 2008). Successful memory retrieval was defined here as a combination of objective accuracy and subjective judgment. In a previous experiment, isolating confidence assessment from recognition processes, we reported that the metamemory processes involved in the sense of knowing engage bilateral medial and lateral parietal regions (Chua et al. 2006). In this experiment, we focused on the neural activity engaged during a cued recall task in which the subjects indicated whether they specifically “remembered” or “forgot” the name associated with that face for each face–name pair.

In this study, our first objective was to investigate to what extent associative face–name encoding and retrieval processes share anatomically overlapping engagement of the same brain regions. Given the emerging literature demonstrating that specific regions within the posteromedial cortices play an important role during both encoding and retrieval, we predicted that cortical “activation” associated with successful retrieval would be, at least partially, overlapping with “deactivation” patterns during encoding in these regions. Second, given that anatomical consistency exists, we sought to investigate the functional relationship between these 2 memory processes in the posteromedial cortex. In particular, how does the degree of deactivation during successful encoding relate to the degree of activation that would be observed during the successful retrieval of that information? Based on the current belief that deactivation in the posteromedial cortices may be the result of reallocation of neuronal resources for efficient cognitive processing, in addition to the previous findings demonstrating that activation in regions part of the parietal cortex correlates with successful memory retrieval, we expected that greater deactivation during encoding would translate into greater activation during successful retrieval of this information. We believe that these questions will serve to clarify the role of the posteromedial cortex in episodic memory and will also yield further insight as to why this region may be particularly vulnerable to functional alterations reported in aging (Miller et al. 2008; Duverne et al. 2009; Goh and Park 2009; Park and Reuter-Lorenz 2009) and in memory impairment associated with neurodegenerative diseases, such as AD (Lustig et al. 2003; Celone et al. 2006; Pihlajamaki et al. 2009).

In addition to looking at the dissociation between successful encoding and retrieval processes, the present work also sought to investigate the anatomical overlap of the whole-brain patterns of activity supporting these 2 memory processes. That is, according to a fundamental principle of memory, encoding and retrieval processes are strongly interdependent, in that successful retrieval of episodic information is seen as depending on “reactivation” of parts of the neural patterns or networks associated with encoding (Alvarez and Squire 1994; McClelland et al. 1995; Rolls 2000; Shastri 2002; Norman and O’Reilly 2003; Rugg et al. 2008). Evidence in support for this idea comes from functional neuroimaging studies demonstrating reinstatement of encoding activity in several areas supporting memory during successful retrieval (Wheeler et al. 2000, 2006; Vaidya et al. 2002; Wheeler and Buckner 2003, 2004; Kahn et al. 2004; Khader et al. 2005; Woodruff et al. 2005; Johnson and Rugg 2007). Thus, with regard to the more traditional view of episodic memory that has been focused on the medial temporal lobe and the prefrontal cortices, we predicted that these areas would demonstrate an overlap of activation in encoding and retrieval, as opposed to “toggling” between deactivation and activation during these processes as the predicted observation in the posteromedial cortices. As both the medial temporal lobe and the posteromedial cortices are emerging as critical nodes in the large-scale network supporting memory function and have been previously shown to be functionally connected, even during the resting state (Greicius et al. 2003; Vincent et al. 2006), it is important to understand how the activity in each of these regions contributes to both successful encoding and retrieval. In addition, both these regions are known to be affected early in age-related amnestic dementias, such as AD, but, interestingly, have predilection for different pathologies: Deposition of amyloid-beta occurs early in the posteromedial cortices, whereas tangle pathology and neuronal loss predominate in the medial temporal lobe early in AD (Jack et al. 2009). Thus, our study may yield important insights into the differential contributions of these regions during successful memory formation and retrieval that may have important implications for understanding late-life amnestic disorders.

Materials and Methods

Participants

Twenty healthy young individuals participated in this study (11 females and 9 males; mean age: 23.4 years, range: 20–29 years). All subjects were right-handed, were English native speakers, and had normal or corrected-to-normal vision. The subjects were screened for history of psychiatric or neurological disorders, history of head trauma, as well as medication use with significant effects on the central nervous system. The study procedures were approved according to the Human Research Committee at the Brigham and Women's Hospital and Massachusetts General Hospital (Boston, MA), and informed written consent was obtained from every subject prior to participation. The subjects participating in this study were recruited specifically for this study and have not appeared in any other previous publication.

Experimental Task

The face–name association task consisted of color pictures of faces displayed against a black background paired with fictional first names (Times New Roman, 36 point font) printed underneath in white (see Fig. 1). The experiment consisted of 4 encoding runs alternating with 4 retrieval runs. The encoding part of this paradigm is a modification of a previous published event-related paradigm (for details, see Sperling et al. 2003; Rand-Giovannetti et al. 2006). The retrieval portion of the paradigm was developed specifically for this study, building on previous work using recognition and confidence rating (Chua et al. 2006, 2009a, 2009b). In each encoding run, subjects viewed 20 face–name pair stimuli, each shown for 2.75 s, which were presented in a pseudorandom order in groups of 4 face–name pairs (see caption of Fig. 1 for additional details of the experimental setup). Each of the 20 face–name pairs was presented a total of 3 times during the encoding run. To investigate the pattern of brain activity during the encoding of novel stimuli compared with retrieval, in this study, we focused only on the first encoding trial for each face–name pair. We developed the repetitive encoding paradigm for eventual use in older and cognitively impaired populations; however, we postulate based on previous research that most of the encoding activity occurs during the first presentation of each face–name pair in young subjects. Thus, the effect of stimulus repetition during encoding runs is not analyzed or reported in this paper but will appear in future reports comparing young and older subjects.

Figure 1.

fMRI paradigm. (A) The paradigm consisted of 4 encoding runs alternating with 4 retrieval runs. In each encoding run, 20 face–name pairs were presented to the subjects 3 times (EN1, EN2, and EN3). Each encoding run was immediately followed by a retrieval run, consisting of a cued recall (CR) task followed by a forced-choice recognition (FCR) task for each of the 20 face–name pairs in the preceding encoding run. (B) The experimental setup of the first encoding run followed by the first retrieval run. In encoding run 1, the first group of 4 face–name pairs (5 groups of 4 face–name pairs comprised the total of 20 stimuli for each encoding run) is displayed (EN1_a) and how the stimuli were repeated over the run (EN2_a and EN3_a), each time presented in a pseudorandomised order within the group (see numbers on the top of the figure). For each face–name pair, the subjects were asked to press a button indicating a purely subjective decision about whether the name was a good “fit” for the face or not. The figure also displays the same 4 faces (RET_a) in the first retrieval run. In the CR task, the subjects had to respond with a button press whether he/she “remembered” the name associated with the face or had “forgotten” it. During the FCR task, the subjects were instructed to indicate the correct name associated with that face by pressing 1 of the 2 buttons (correct name was presented in counterbalanced order across all runs).

During the presentation of each face–name pair, subjects were asked to press a button indicating a purely subjective decision about whether the name was a good “fit” for the face or not, which has been previously shown to enhance successful encoding (Sperling et al. 2003). Before each run, the instructions were repeated and the subjects were explicitly told to try to remember the name associated with the face. Face–name stimuli were randomly intermixed with trials of visual fixation (a white crosshair centered on a black background). During the presentation of the fixation cross, subjects were told to focus their attention on the cross.

After each encoding run, a retrieval run was performed consisting of cued recall stimuli followed by forced-choice recognition stimuli for each face–name pair learned during the preceding encoding run. In the cued recall task, a face was presented for 5.25 s and the subject had to respond with a button press to indicate whether he or she remembered or forgot the name associated with the face. During the forced-choice recognition task, each face was shown (for 3.25 s) with 2 names printed underneath: the correct name that was paired with the face during encoding, and an incorrect name that was previously paired with a different face during encoding. Subjects were instructed to indicate the correct name by pressing 1 of the 2 buttons. The experiment was designed and generated on an external personal computer using MacStim 2.5 software (WhiteAnt Occasional Publishing) and projected by means of a magnetic resonance (MR) compatible goggle system (VisuaStim XGA; Resonance Technology Inc.). Subjects received detailed oral instructions prior to each run and completed a practice session both inside and outside the MR scanner. Subjects responded with an MR compatible optical key press device with 2 buttons held in their right hand, and responses (accuracy and reaction time) were recorded by a computer interfaced with the optical switch using MacStim software outside the scanner room.

fMRI Acquisition

All measurements were carried out using a General Electric 3.0 Tesla MR system (GE Signa), equipped with an 8 channel head coil, and single-shot echo-planar imaging (EPI) software was used. Functional imaging was performed using a T2*-weighted gradient-echo EPI sequence sensitive to blood oxygen level–dependent (BOLD) signal (time repetition, 2000 ms; time echo, 30 ms; flip angle, 90°) within a field of view of 220 cm, with a 64 × 64 pixel matrix and a slice thickness of 5 mm (interslice distance, 1 mm). Thirty oblique coronal slices, perpendicular to the anterior–posterior commissural line, were acquired to cover the whole brain. Eight functional runs were acquired for each subject with 145 time points per run. The first 5 (additional) images in each run were discarded to allow the magnetization to reach equilibrium. Scanning time for each functional run was 5 min resulting in a total functional scanning time of approximately 40 min.

Image Preprocessing

Imaging data were transferred and processed on a Linux platform running MATLAB version 7.1 (The Mathworks, Inc. Sherborn, MA, USA). Data preprocessing and statistical analyses were performed with Statistical Parametric Mapping (SPM 2; Wellcome Department of Cognitive Neurology: http://www.fil.ion.ucl.ac.uk). The data were motion corrected using sinc interpolation, by aligning (within subject) each time series to the first image volume using a least-square minimization a 6-parameter (rigid body) spatial transformation. No subject exceeded head movement over 3 mm (in the z axis translation). Data were normalized to the standard SPM2 EPI template and resliced into 3 × 3 × 3 mm³ resolution in Montreal Neurological Institute (MNI) space. Smoothing was accomplished using an isotropic Gaussian kernel of 8 mm full-width half-maximum. No scaling was implemented for global effects. The coordinates were later converted to Talairach and Tournoux’s (1988) space using a software available online http://imaging.mrc-cbu.cam.ac.uk/imaging/MniTalairach.

fMRI Statistical Analysis

The face–name stimuli were categorized based on the subjects’ responses during the cued recall (remember vs. forgotten) and forced-choice recognition (hit vs. miss) tasks, allowing 4 possible conditions: remembered hit (RHIT), forgotten hit (FHIT), remembered miss, and forgotten miss (FMISS). Given our goal to investigate successful encoding and retrieval, and the high performance overall in this task among young subjects, the current paper will focus on the results based on the RHIT variable. For each subject, all runs were concatenated and regressors added, in lieu of global scaling, to account for signal differences between runs.

To identify the brain regions involved in successful episodic encoding, second level t-statistics (using random effects in SPM2) were computed for all subjects. This was done by creating contrasts for each subject comparing event-related fMRI activity during encoding of face–name pairs that were subsequently remembered (RHITenc) to a control condition (fixation cross) (RHITenc > Fix). To investigate the areas that deactivate during this condition, the opposite contrast was used (Fix > RHITenc). Brain regions involved in successful episodic retrieval of face–name association were identified by using t-statistics, comparing fMRI data during successful cued recall with the control condition (RHITret > Fix). All whole-brain statistical maps were threshold at P_uncorrected < 0.001 (extent threshold 5 voxels) and using a P_FDR ≤ 0.05 at the cluster level.

A conjunction analysis (Price and Friston 1997) was performed to investigate the topographical overlap of brain activity during the face–name encoding and retrieval tasks. A conjunction is defined as the intersection of the 2 statistical maps (random-effects group analysis) for the 2 contrasts by using a conjunction null test. Thus, it makes it possible to identify regions that are involved in both contrasts since regions that are differentially engaged in 1 of the 2 contrasts are excluded. Two different conjunction analyses were performed: 1) overlapping brain areas activated during encoding and retrieval (RHITenc > Fix) AND (RHITret > Fix) and 2) overlapping areas deactivated during encoding and activated during retrieval (Fix > RHITenc) AND (RHITret > Fix). The conjunction maps were threshold at P_uncorrected < 0.001 (extent threshold 5 voxels) and using a P_FDR ≤ 0.05 at the cluster level.

Functional Comparison of Overlapping Regions

A volume of interest (VOI) analysis was performed to test the magnitude and relation of the BOLD fMRI signals observed during the encoding and retrieval task. VOI placements were determined based on the regions that showed an overlap in the conjunction analysis and created using MarsBar (Brett et al. 2002). A VOI size of 5 mm radial spheres (10 mm diameter), corresponding to 20 contiguous resampled voxels, was used in order to obtain a close measurement of the peak activity value observed within each cluster’s maxima. For each subject and VOI, the mean beta weights for each contrast were extracted. These values were entered in a Pearson’s correlation analysis (using STATISTICA 8; StatSoft Inc.) to assess the extent to which deactivation during encoding might predict their task-related activity during the subsequent retrieval of that information.

The hemodynamic response function was reconstructed using the general linear model to estimate the finite impulse response (FIR) (Ollinger et al. 2001). This procedure is equivalent to the method of selective averaging (Dale 1999). The FIR model has no a priori assumption about the shape of the hemodynamic response and no requirement of a uniform transition matrix. Thus, this set of basis functions is able to capture any shape of response up to a given frequency limit (Henson et al. 2001). We used a model with 10 FIR time bins, allowing us to extract the event-related data for the 0–18 s after stimulus onset.

Results

Task Performance

Behavioral results on the retrieval task during the fMRI experiment are presented in Table 1. There was a high percentage of RHIT during the task (median: 78.8%), indicating that the subjects in overall encoded the face–name stimuli successfully and confirmed a high attentional engagement during both tasks.

Table 1.

Behavioral results

Performance measure	Mean (SD)
Percentage of trials
Total hits	94.1 (6.2)
RHIT	76.1 (18.8)
FHIT	18.0 (15.2)
RMISS	2.1 (2.4)
FMISS	2.9 (5.6)
Reaction time (ms)
Cued recall	1523 (447)
Force choice recognition	1186 (233)

Notes: Mean percentage and standard deviation (SD) of face–name pairs that were classified as remembered hits (RHIT), forgotten hits (FHIT), remembered misses (RMISS), and forgotten misses (FMISS) on the cued recalled and forced-choice recognition task for young adults. Mean reaction time (i.e., seconds until button press) during cued recall (the subjects are asked if they remember or have forgotten the name that goes with a particular face) and forced-choice recognition task (the subjects are asked to choose between 2 names).

Identification of fMRI Activition during Successful Encoding

We first examined regions that were involved in successful face–name encoding by investigating the pattern of activity during the encoding for face–name pairs of names that were subsequently indicated as “remembered” during cued recall and correctly chosen during the recognition trials versus fixation (RHITenc > Fix). Significant activation was observed in the hippocampus (see Fig. 2D) bilaterally as well as in bilateral parahippocampal gyrus (BA 28), and left fusiform gyrus (BAs 37 and 20). Activation was also found in the occipital lobe including left primary visual cortex (BA 17). Several significant clusters of activation were also revealed in the frontal lobes in both hemispheres (e.g., inferior frontal gyrus [BAs 47, 46 and 45] and left middle frontal gyrus).

Figure 2.

Brain regions involved in the face–name association task. Statistical parametric maps (random effects) from the one sample t-test demonstrating the deactivation in encoding (A) and activation during cued recall (B) as well as the conjunction (C) of these 2 statistical maps, demonstrating a significant overlap in the precuneus and posterior cingulate regions. In the lower row, the statistical parametric maps from the one sample t-test illustrates the activation in encoding (D) and during cued recall (E) as well as the conjunction (F) between these 2, demonstrating significant overlap in activation in the hippocampus. The maps are threshold at P < 0.001, minimal extent threshold 5 adjacent voxels and superimposed on a single-subject high-resolution T1 structural images, A–C on Talairach sagittal planes, x = 12 and D–F on Talairach coronal planes, y = 15. Lighter color scale indicates more significant activation (orange/red) or deactivation (blue).

Identification of fMRI Deactivition during Successful Encoding

To investigate which regions were deactivated during successful encoding of the face–name stimuli, we investigated the opposite contrast as presented above (Fix > RHITenc). This analysis revealed significant deactivation in brain regions part of the default mode network (e.g., Buckner et al. 2008). This included extensive significant deactivation in bilateral parietal lobe, superior parietal lobe (SPL), precuneus (BA 7), and inferior parietal lobe (IPL) (BA 40), which extended into bilateral posterior cingulate and retrosplenial cortex (BA 31) (see Fig. 2A). Significant bilateral clusters of deactivation were also found in bilateral frontal lobes, including dorsolateral medial prefrontal cortex (e.g., BAs 10 and 46) as well as motor regions (BAs 2 and 4). In addition, areas in the middle temporal cortex (BA 21) were also deactivated. Specific Talairach coordinates from these activation clusters can be given upon request from the authors.

Identification of fMRI Activition during Successful Retrieval

Regions involved in successful episodic retrieval of items were identified by comparing the contrast successful face–name stimuli versus fixation (RHITret > Fix) (see Fig. 2 B,E and Fig. 5A). This generated several activated areas, including bilateral parietal lobe (SPL, precuneus [BA 7] and IPL [BA 40]) extending into the retrosplenial cortex as well as bilateral occipital lobe (BA 18). Activation was also found in bilateral frontal lobes (e.g., BA 46) as well as in the temporal lobe, including bilateral hippocampus. Specific Talairach coordinates from these activation clusters can be given upon request from the authors.

Figure 5.

(A) One sample t-test of Remember Hit > Fix (left) and Forgotten Hit > Fix (right) contrasts. The latter was made in order to investigate which areas were activated when the subjects said that they had forgotten the name during the cued recall test but choose the right name during the forced choice recognition test. The activation maps are projected on an inflated (200 iterations) template brain surface. A surface based smoothing (10 mm) was applied. Statistical values are displayed on the surface in color scale, thresholded at p = 0.001 with an extent threshold of 5 voxels. (B) Anatomical overlap in fMRI activity from exploratory analysis. Conjunction map of (A) RHITenc > FORGET AND RHITret > FORGET demonstrating overlap of activation in hippocampus and (B) FORGET > RHITenc AND RHITret > FORGET showing overlap of deactivation and activation in posteromedial cortex. The maps are threshold at P < 0.05, minimal extent threshold 5 adjacent voxels and superimposed on a single-subject high-resolution T1 images. Lightness in center of colored area indicates more significant activity. Extracted beta weights during successful encoding (light blue) and FORGET (i.e., weighted mean for FHIT and FMISS) (dark blue) and successful cued recall (light orange) and FORGET (dark orange) from the VOIs in the left precuneus (global maximum found in Tal coordinates [−15 −68 42]) and right hippocampus (21 −9 −15).

To further explore whether the activation elicited in the posteromedial cortex is a result of the process of subjective judgment about memory retrieval rather than merely successful memory retrieval, we performed several additional analyses. First, we performed a one sample t-test looking at the contrast FHITret > Fix (P_uncorrected = 0.001, extent threshold 5 voxels). This analysis demonstrated small areas of activation in the posteromedial cortex (see Fig. 5A right), both areas situated in precuneus (BA 7), implying that these areas might be involved in successful memory recognition regardless of whether the subject recalled or had forgotten the right name in the cued recall part of the experiment. However, when directly contrasting RHIT and FHIT, we could demonstrate increased activation in bilateral posteromedial cortex (P_uncorrected = 0.001, extent threshold 5 voxels) only for RHIT > FHIT (the opposite contrast did not reveal any activation at all in the parietal cortex), suggesting that although these areas seem to be involved in recognition, they are involved to a greater extent in the process of subjective sense of remembering the right name (i.e., knowing you know) during cued recall. In addition, as the recognition trials were forced choice between only 2 alternatives, this is a much easier retrieval task. In a subsequent analysis, we investigated the activation elicited during items when the subject picked the right name in the forced-choice recognition task versus all the stimuli where the subject picked the wrong name (i.e., all hits > all misses). This analysis did not reveal any significant activation in the posteromedial cortices at the prespecified threshold, and even at the very liberal threshold of uncorrected P_uncorrected = 0.05, we observed only small clusters of bilateral activation in the superior parietal lobules, BA 7.

Anatomical Comparison of Identified Activation Maps

A group-level conjunction analysis was performed (RHITenc > Fix AND RHITret > Fix) to probe the anatomic overlap of regional activity across the 2 tasks. Regions with overlapping activation patterns included bilateral hippocampus, occipital lobes, as well as several areas in the frontal lobes, including the inferior and superior frontal gyrus (Fig. 2F). The specific location of the peak activations as well as the mean beta weights during encoding and retrieval are presented in Table 2B. To illustrate the relationship between encoding and retrieval, the average time courses (percent [%] MR signal) were extracted from a priori region of interest in the hippocampus (Fig. 3B).

Table 2.

Overlapping anatomical regions during successful encoding and retrieval of associative face–name stimuli

Anatomical region	BA	Talairach coordinates			T value	PFDR	Beta weight
		x	y	z			Encoding	Retrieval
(A) Deactivation encoding AND activation cued recall
Parietal lobe
Precuneus	7	12	−62	34	6.03	0.006	−1.5 ± 0.23	1.4 ± 0.24
	7	−6	−68	45	5.09	0.011	−1.6 ± 0.34	1.8 ± 0.25
	7	−6	−65	36	4.96	0.011	−1.4 ± 0.29	1.3 ± 0.24
	7	12	−68	48	4.96	0.011	−1.7 ± 0.39	1.7 ± 0.27
Inferior parietal lobe	40	−39	−50	41	3.68	0.050	−1.1 ± 0.31	1.04 ± 0.23
Retrosplenial cortex	31	21	−54	28	5.05	0.011	−0.89 ± 0.18	0.93 ± 0.17
Posterior cingulate	23	−3	−16	26	4.84	0.011	−0.82 ± 0.18	1.18 ± 0.16
	23	9	−11	23	3.81	0.041	−0.54 ± 0.01	0.54 ± 0.12
(B) Activation encoding AND activation cued recall
Occipital lobe
Inferior occipital gyrus	18	36	−79	−1	12.99	<0.001	2.2 ± 0.15	2.9 ± 0.15
Middle occipital gyrus	19	42	−84	7	12.17	<0.001	2.0 ± 0.18	2.2 ± 0.13
Temporal gyrus
Hippocampus		24	−24	−6	5.55	<0.001	0.39 ± 0.08	0.71 ± 0.09
		−27	−15	−12	4.67	<0.001	0.91 ± 0.21	0.94 ± 0.19
Parahippocampal gyrus	28	−24	−24	−6	4.78	<0.001	0.36 ± 0.08	0.37 ± 0.08
Frontal gyrus
Middle frontal gyrus	46	−42	16	24	8.05	<0.001	1.7 ± 0.19	1.8 ± 0.23
Inferior frontal gyrus	13	−39	32	7	4.32	<0.001	0.96 ± 0.17	0.96 ± 0.24
	13	42	33	9	5.21	<0.001	1.0 ± 0.18	0.98 ± 0.19
	44	39	10	22	4.67	<0.001	0.45 ± 0.09	0.76 ± 0.11
	45	48	21	18	4.76	<0.001	1.1 ± 0.23	1.1 ± 0.22
	47	−33	29	−6	5.16	<0.001	0.90 ± 0.13	1.0 ± 0.21
Superior frontal gyrus	8	3	20	52	5.83	<0.001	1.2 ± 0.19	1.7 ± 0.22
Precentral gyrus	6	−39	−12	61	5.02	<0.001	1.1 ± 0.21	1.1. ± 0.21
Cingulate gyrus	32	−6	22	40	4.82	<0.001	0.93 ± 0.17	2.3. ± 0.23
		12	25	37	3.92	0.002	0.42 ± 0.12	0.65 ± 0.11
Insula	13	−18	9	2	3.51	0.006	0.42 ± 0.12	0.53 ± 0.15
Cerebellum		39	−48	−18	10.40	<0.001	2.1 ± 0.19	3.3 ± 0.24
		9	−77	−29	4.36	<0.001	0.37 ± 0.08	0.69 ± 0.11
		3	−60	−32	4.53	<0.001	0.62 ± 0.12	0.84 ± 0.17

Notes: Data derived from conjunction analysis of the deactivation in the face–name association task during successful encoding AND activation during retrieval (A) and activation in the face–name association task during successful encoding and retrieval (B). Brodmann areas (BA), Talairach coordinates (x, y, and z values in millimeters) showing 3 local maxima more than 8.0 mm apart in each cluster, statistical threshold (t value) for voxel definition and the P value (FDR corrected at the voxel level), and mean beta weight and standard error during the encoding and retrieval condition.

Figure 3.

MR signal time courses from overlapping brain regions during successful encoding and retrieval. Extracted % MR signal time courses (time repetition = 2 s) during successful encoding (blue) and successful cued recall (orange) from region of interest in posteromedial cortex (A) left precuneus (x = −6, y = −65, z = 36) and right retrosplenial cortex (x = 21, y = −54, z = 28) and hippocampus (B) left (x = −27, y = −15, z = 12) and right (x = 24, y = −24, z = −6). Error bars represent standard error.

To specifically examine whether the same areas that were deactivating during encoding were activating during retrieval, an additional conjunction analysis was performed (Fix > RHITenc AND RHITret > Fix). Overlapping regions were found in bilateral parietal lobe including left and right SPL and precuneus (BA 7) and left IPL (BA 40) and extending down to retrosplenial cortex (Fig. 2C). An additional overlapping cluster was found in the posterior cingulate. The specific location of the peak activations as well as the mean beta weights during encoding and retrieval are presented in Table 2A. To illustrate the relationship between encoding and retrieval, the average time courses (% MR signal) were extracted from an a priori region of interest in the posteromedial cortex (Fig. 3A).

Functional Relationship of Overlapping Regions

To determine whether the degree of deactivation during encoding was related to the degree of activation during retrieval, we performed a product-moment correlation analysis between the beta weights extracted from all the VOIs in the posteromedial cortex (see Table 2). A significant negative correlation was found between the 2 conditions only in the right retrosplenial cortex (x = 21, y = −54, z = 28; R= −0.44, P = 0.05), as shown in Figure 4.

Figure 4.

Relationship between the beta weights during encoding and retrieval in the posteromedial cortices. Extracted beta weights during successful encoding (blue) and successful cued recall (orange) from the right retrosplenial VOI (x = 21, y = −54, z = 28). Error bars represent standard error. A negative correlation was found between the 2 variables (r = −0.44, P = 0.05).

We also investigated whether the degree of activation during encoding was related to the degree of activation during retrieval in the hippocampus (see VOI details in Table 2). No significant correlation could be found for any of the VOIs.

Exploratory Analysis to Look at Anatomical Comparison of Identified Activation Maps

In order to explore these processes using a more constrained contrast than using the fixation condition, we performed 2 additional conjunction analyses using RHIT items compared with FMISS + FHIT items combined (termed here as FORGET). Although we have limited power in this contrast, as there were relatively few stimuli in the FORGET condition, we found a similar pattern of results to those presented above using a more liberal threshold (P_uncorrected = 0.05, 5 voxels extent). The contrast RHITenc > FORGET AND RHITret > FORGET revealed an overlap of activation in the hippocampus and bilateral, and the contrast FORGET > RHITenc AND RHITret > FORGET demonstrated overlapping activation in the left posteromedial cortex (Fig. 5B).

Discussion

Successful episodic memory function requires both encoding and retrieval processes, which can be studied in isolation using functional neuroimaging techniques to identify patterns of activity and specific neuroanatomical structures engaged during each process. Here, we used event-related fMRI to address the question to what extent associative encoding and retrieval operations share overlapping anatomical distributions. Given that anatomical consistency exists, we were further interested in investigating the functional relationship between neural activity during these 2 memory processes. Our results demonstrate that the anatomy of the brain regions activated during successful retrieval overlapped several areas (including hippocampus and prefrontal cortex) involved in encoding the same information. Of particular interest was the finding in the posteromedial cortex, which demonstrates a shift or toggling between a deactivated and activated state during encoding and retrieval processes. Furthermore, we observed a significant negative correlation in the activity between these 2 processes in this region, such that greater deactivation during successful encoding was related to greater activation during successful retrieval. These results provide further evidence that posteromedial cortices are important for both encoding and retrieval processes and may serve as critical nodes in the large-scale cortical network subserving episodic memory.

Several cognitive theories of episodic memory hold that successful memory performance is based on a match between encoding and retrieval (e.g., the encoding specificity principle [Tulving and Thomson 1974] and the transfer appropriate processing theory [Morris et al. 1977]). Accordingly, the likelihood of successful retrieval is seen as a function of the extent to which the processes engaged by a retrieval cue overlaps with that engaged at encoding. Later on, Craik et al. (1996) suggested that this overlap at a cognitive level could translate into overlap at the neural level. To date, evidence in support of a “reinstatement or cortical reactivation hypothesis” has been provided by several functional neuroimaging studies using positron emission tomography (Nyberg et al. 2000, 2001; Persson and Nyberg 2000) and fMRI (Wheeler et al. 2000, 2006; Vaidya et al. 2002; Wheeler and Buckner 2003, 2004; Kahn et al. 2004; Khader et al. 2005; Woodruff et al. 2005; Johnson and Rugg 2007). Results from these studies have shown that successful retrieval, using a wide range of tasks, is associated with reinstatement of neural circuits that were originally involved in processing that information (for review and discussion, see Buckner and Wheeler 2001; Rugg et al. 2008). Our finding that young individuals demonstrate overlapping activation patterns during both encoding and retrieval in bilateral hippocampi and adjacent medial temporal regions as well as prefrontal regions is in accordance with these previous findings and supports the proposed mechanism of reinstatement. While our findings and multiple previous studies demonstrate the commonalities in activation patterns during encoding and retrieval, the role of regions demonstrating deactivation during encoding but subsequent activation during retrieval may be of particular interest.

The present findings of posteromedial cortical involvement during encoding and retrieval are consistent with recent neuroimaging data demonstrating beneficial deactivations during successful encoding (Daselaar et al. 2004; Kao et al. 2005; Miller et al. 2008) and activation during successful retrieval of that information (Addis et al. 2004; Wheeler and Buckner 2004; Chua et al. 2006), suggesting an important role of the posteromedial cortex in episodic learning and memory. With regard to deactivation, these posteromedial regions have been hypothesized to be part of a “default mode network” (Raichle et al. 2001; Greicius et al. 2004; Buckner et al. 2008), referring to the phenomenon that these brain areas are active during rest and that BOLD activity is decreased during performance of focused cognitive processing. Although the exact functional relevance of this deactivation is still a focus of debate, it has been implicated in the reallocation of neuronal resources, in order for the individual to focus more on the task at hand and subsequently be more successful during periods of encoding (Buckner et al. 2008). To date, several neuroimaging studies have provided support in favor of the view that the observed deactivation pattern is involved in ongoing information processing and contribute to successful task performance. For example, using a parametric auditory target detection task, McKiernan et al. (2003) were able to demonstrate that task-induced deactivation increased as task demand increased in a subsequently administered cognitive probe task. In a similar manner, our laboratory (Miller et al. 2008), using a similar face–name encoding paradigm to the one presented in this study, demonstrated greater deactivation in the parietal cortex during successful encoding in both young and healthy elderly subjects. However, only the elderly subjects demonstrated a significant relationship between amount of deactivation and overall memory performance. Similarly, in the present study, the correlation analysis in young subjects did not reveal a significant relationship between memory performance and deactivation magnitudes for any of the region of interest investigated (data not presented). We hypothesize that although deactivation of the posteromedial cortices is an important component of successful memory encoding, that this critical node of the network is generally intact in young subjects but may be particularly challenged in the context of aging and/or early AD pathology (see Lustig et al. 2003; Petrella et al. 2007; Miller et al. 2008; Pihlajamaki et al. 2009; Sperling et al. 2009). Of particular interest is the finding that posteromedial cortices are also selectively vulnerable to early amyloid deposition (Buckner et al. 2005; Mintun et al. 2006; Sperling et al. 2009) and early amyloid pathology associated with failure of deactivation in this region may contribute to the hallmark symptom of episodic memory impairment in early AD (Sperling et al. 2009).

With regard to the current findings of the anatomic overlap between deactivation and activation, it is also important to take into account the involvement of posterior parietal cortex in the context of successful episodic retrieval. The role of this region, sometimes characterized as part of a “retrosplenial memory system”(Buckner et al. 2005; Wagner et al. 2005; Cabeza 2008; Cabeza et al. 2008; Ciaramelli et al. 2008; Vilberg and Rugg 2008; Uncapher and Wagner 2009), has been suggested to be that of evaluating information, once that information has been retrieved from memory (see Wagner et al. 2005), although several other theories have been proposed as well (see Spaniol et al. 2009). In particular, Wheeler and Buckner (2004) reported that regions in the posterior parietal cortices and lateral parietal lobe were more active when subjects report that they “remember” a previously encoded item compared with when they say they “know” the previously encoded item. Further support for this theory is the finding that activation in some parietal regions correlates with successful memory retrieval and with false memory errors (Wheeler and Buckner 2003; Kahn et al. 2004; Svoboda et al. 2006), “raising the possibility that the strength of activation contributes to the eventual decision” (Wagner et al. 2005). Our results see Fig 5. A are in line with these findings and suggest that activation in the posteromedial cortex is not driven merely by successful objective memory retrieval, but rather is a result of processes accompanied by subjective judgment about the memory retrieval.

The present findings complement recent functional neuroimaging studies in demonstrating the dissociation in the pattern of neural activity between successful encoding and retrieval in the posteromedial cortex and provide further evidence that within the same individual and same experiment, this region must modulate its activity between these memory processes. Our results are very consonant with the findings from Daselaar et al. (2009) who in an elegant study, which included five separate fMRI experiments (each including encoding and retrieval) of faces, scenes, and single words, reported a similar encoding/retrieval toggling phenomenon. Daselaar et al. were able to demonstrate that greater activity for hits compared with misses in areas in the posteromedial cortex as well as in the ventral part of lateral posterior parietal cortex during retrieval, whereas during encoding, these same areas showed greater activation during misses than hits. The authors were also able to present evidence of an overlap of these regions with the ones that show activations during conscious rest, concluding that retrieval processes requires engagement of the default mode network, whereas encoding does not, but rather benefits from its suppression. Our results are in line with these findings, and we were further able to provide the first evidence of a direct functional relationship between these 2 memory processes, demonstrating a correlation between the degree of deactivation during encoding and subsequent activation during retrieval.

With regard to the default mode network, we also want to acknowledge the emerging literature focusing on the relationship between this network, as measured by resting state fMRI, and activation patterns across a wide range of cognitive tasks. Based on the discovery of a strong negative correlations between default network and other systems coined “dynamic equilibrium” and “anticorrelations” (Greicius et al. 2003; Fox et al. 2005; Fransson 2005; Golland et al. 2007; Tian et al. 2007) the idea that the brain’s default mode network may work in direct opposition to other systems has recently been proposed. Kim et al. (2009) recently suggested that the “task-negative and task-positive networks” may have opposing roles in encoding and retrieval success. Buckner et al. (2008) discussed the implications for this relationship and suggested that the brain may shift between 2 distinct modes of information processing; the first characterized by mental explorations based on past memories and detached from focused attention on the external environment, and the second associated with focused information extraction from sensory channels. In line with this, recent findings suggests the existence of a core brain network, involving the same regions observed during default mode processing, that has been proposed to underlie a number of internally oriented tasks, including autobiographical memory, prospection, navigation, theory of mind (Buckner and Carroll 2007; Buckner et al. 2008; Spreng et al. 2009), as well as envisioning the future (Szpunar et al. 2007) and self-referential and reflective activity (Gusnard and Raichle 2001; Kelley et al. 2002; Greicius et al. 2003; Fransson 2005). Common among these processes is the finding of an increase of activation in these regions, implied to be the result of internally oriented attention (Buckner et al. 2008). In contrast, the deactivation observed during a wide range of tasks, including visual (Shulman et al. 1997) and auditory (McKiernan et al. 2003) tasks, has been implied to be the result of external oriented attention. These findings are intriguing and could be extended to the interpretation of the observed dissociation found in the posteromedial cortex. Given the above-mentioned idea that the deactivation patterns might reflect more focused external attention of information processing, we speculate that the successful encoding process might be the result of the ability to efficiently suppress internally generated cognitive processes. The observed increased activation in the same area during successful retrieval would then reflect the ability to orient the attention to internal representations of the encoded memory, most likely through pathways connecting posteromedial cortex and regions in the medial temporal lobe including hippocampus (Wagner et al. 2005). As a consequence, one could speculate that the overlap seen in the current data could in fact represent functionally collaborating brain systems that work together to increase the likelihood of successful information processing. Thus, it is likely that the observed encoding/retrieval “flip” may represent evidence of the ability to focus attention on novel stimuli, which enhances encoding and sets the stage for activity during subsequent retrieval. The relationship of the toggling of neural activity in the posteromedial cortices to the consistent activation observed in other functionally connected nodes of the network, specifically the hippocampus, remains to be elucidated. Our previous work has suggested that failure of deactivation during encoding may be associated with paradoxical hyperactivity within the hippocampus in older individuals with mild memory impairment, particularly in those with evidence of amyloid deposition in the posteromedial cortices (Miller et al. 2008; Sperling et al. 2009).

In summary, these findings support the hypothesis that process of retrieving episodic information from memory engages the same network of regions involved in encoding that information. In addition to our findings supporting the reactivation or reinstatement hypothesis, as observed in the hippocampus and prefrontal cortex, we report here an important regional exception to that theory. The posteromedial cortex demonstrated an interesting toggling phenomenon between deactivation during successful encoding and activation during successful retrieval. Furthermore, we found a significant negative correlation between the negative task–related activity during encoding and positive task–related activity during retrieval, suggesting that encoding and retrieval of associative memories are subserved by reciprocal activity in this region. Although much remains to be understood regarding the physiological basis of deactivation during episodic memory, the relationship of this phenomenon during encoding to activity during retrieval processes demonstrates that “what goes down evidently does come up.” Our finding suggests that these are not isolated processes, but rather constitute carefully orchestrated activity within an integrated memory system. In conclusion, our observations, together with those of other groups, point to a key functional specialization within the posteromedial cortices and provide support for the ongoing investigation of the contribution of these regions to age and neurodegenerative disease–related alterations in episodic memory encoding and retrieval.

Funding

Karolinska Institutet foundations (Fobi0794) to P.V.; Swedish Research Council to P.V.; Royal Science Academy in Sweden to P.V.; Marie Curie Fellowship (FP7-PEOPLE-2007-4-1-IOF) from the European Union to P.V. National Institutes of Health (R01 AG027435, P01AG036694 to R.A.S.); Alzheimer’s Association to R.A.S.