Cerebral Cortex Advance Access originally published online on January 4, 2007
Cerebral Cortex 2007 17(11):2507-2515; doi:10.1093/cercor/bhl156
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Recollection and the Reinstatement of Encoding-Related Cortical Activity
Center for the Neurobiology of Learning and Memory, and Department of Neurobiology and Behavior, University of California—Irvine, Irvine, CA 92697-3800, USA
Address correspondence to Jeffrey D. Johnson, Center for the Neurobiology of Learning and Memory, University of California—Irvine, Irvine, CA 92697-3800, USA. Email: jeff.johnson{at}uci.edu.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The neural correlates of episodic memory retrieval ("recollection") differ according to the type of information contained in the recollected episode. Such content-specific recollection effects have been hypothesized to reflect the reinstatement of processes or representations active during encoding. Using event-related functional magnetic resonance imaging, we evaluated this hypothesis by directly contrasting the neural activity elicited during the encoding and subsequent recollection of words studied with one of 2 encoding tasks. Study words appearing on pictures of scenes required imagining the word's referent at any location within the scene, whereas words appearing on a blank background required generating a sentence that incorporated the word. On a later memory test, the neural correlates of recollection were operationalized by contrasting the activity elicited during correct "remember" versus "know" responses. Recollected words from the "scene" task elicited activity in regions of left occipital cortex and anterior fusiform gyrus that overlapped regions where encoding-related activity was greater for the scene than sentence task. Conversely, activity elicited by words recollected from the "sentence" task overlapped with a region of ventromedial frontal cortex where encoding-related activity was greater for the sentence task. These content-specific associations between encoding- and recollection-related neural activity strongly support the reinstatement hypothesis of episodic retrieval.
Key Words: episodic memory • fMRI • know • recognition • remember
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
A long-standing idea in memory research is that the retrieval of a prior episode ("recollection") involves the reinstatement of processes or representations that were active when the episode was encoded. This idea is incorporated into several neurally inspired models of memory retrieval (e.g., Alvarez and Squire 1994
; Rolls 2000; Shastri 2002; Norman and O'Reilly 2003). According to such models, recollection of a recent episode occurs when a pattern of cortical activity corresponding to the episode is reinstated via activation of a hippocampally stored representation of that pattern.
Whereas there is only limited evidence for the "reinstatement hypothesis" from neuropsychological studies (Rubin and Greenberg 1998), functional neuroimaging studies have provided more extensive support. In several studies employing blocked experimental designs, cortical regions where activity differed during encoding of one stimulus class relative to another (e.g., words vs. pictures) were also differentially active during later memory tests for that stimulus class (Nyberg et al. 2000, 2001; Persson and Nyberg 2000; Vaidya et al. 2002). Such blocked designs, however, leave open the possibility that content-specific retrieval effects reflect processes engaged in attempting to retrieve targeted memories, rather than processes associated specifically with recollection. A similar issue arises with respect to a recent functional magnetic resonance imaging (fMRI) study that employed a pattern-classification analysis to assess the similarity between neural activity associated with the encoding and subsequent recall of stimuli from different categories (Polyn et al. 2005).
More recently, other fMRI studies employing event-related, randomized designs have also reported findings suggesting that the neural correlates of recollection reflect reinstatement of encoding-related activity (Wheeler et al. 2000, 2006; Wheeler and Buckner 2003, 2004; Gottfried et al. 2004; Kahn et al. 2004; Khader et al. 2005; Woodruff et al. 2005). However, findings from 4 of these studies (Wheeler et al. 2000, 2006; Wheeler and Buckner 2003; Khader et al. 2005) were based on the retrieval of items studied on multiple occasions, raising the possibility that the effects reflect a form of learning distinct from that supporting memory for unique episodes. Further, in 3 of the studies (Gottfried et al. 2004; Wheeler and Buckner 2004; Wheeler et al. 2006), inferences about the content-sensitivity of the neural correlates of recollection were based on single dissociations between cortical activity and retrieved content. Only 2 studies employing trial-unique study presentations reported content-specific double dissociations of recollection-related activity, such that 2 different regions exhibited a cross-over pattern of activity levels according to the nature of the recollected information (Kahn et al. 2004; Woodruff et al. 2005). However, in neither of these studies was the relation between recollection- and encoding-related activity assessed within subjects.
The aim of the present study was to provide a strong test of the cortical reinstatement hypothesis. Subjects studied a series of words, each presented in one of 2 distinct encoding tasks that differentially engaged multiple cortical regions. Memory for the words was later tested with a "remember/know" procedure (Tulving 1985), permitting identification of items where recognition was accompanied by recollection. Event-related fMRI data were obtained at both study and test, allowing for direct comparison between encoding- and recollection-related activity. According to the reinstatement hypothesis, differential encoding-related activity should overlap with regions where recollection of words from a given encoding task gives rise to greater recollection-related activity than do words from the alternate task.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Subjects
Twenty-six volunteers between 18 and 35 years of age were recruited from the University of California—Irvine (UCI) community and remunerated for their participation. All subjects reported themselves to be right-handed, native English speakers with normal or corrected-to-normal vision, no history of neurological disease, and no other contraindications for MRI. Informed consent was obtained from all subjects in accordance with UCI Institutional Review Board guidelines. The data from 10 subjects were excluded from all analyses: 3 for inadequate memory performance, one for excessive head movement, 2 for failing to complete the study phase within the allotted time (see Scanning parameters), and 4 for responding "know" (K; see Procedure) to fewer than 10 previously studied test items. The remaining 16 subjects (10 males) had a mean age of 20 years.
Stimuli
Stimuli were drawn from a pool of 218 words (mean length = 6 letters; mean written frequency = 17/million; Kucera and Francis 1967) and a pool of 73 color pictures. The words were names of single objects from several categories, including tools, furniture, animals, and food. The pictures were of natural landscapes that did not include any buildings, animals, or people, and were selected to minimize overlap in content. Ten of the words were used as study buffers (2 at the beginning and 2 at the end of the study list) and test buffers (2 at the beginning, and 2 immediately following each of 2 breaks), and 4 pictures were used as backgrounds for the study buffers. For each subject, 180 words from the pool were randomly assigned to 3 groups of 60 items. The words from 2 of the groups were presented in the study phase and subsequently served as old test words, whereas the words from the other group served as new test words. For the study phase, each of the words from one of the groups was randomly paired with a landscape background. The remaining 28 words and 9 pictures were used in practice study and test phases to familiarize subjects with the experimental tasks.
All stimuli were displayed via VisuaStim XGA (Resonance Technology, Inc., Northridge, CA) MRI-compatible head-mounted goggles with a resolution of 800 x 600 pixels. The field of display subtended visual angles of 30° x 23° at a virtual viewing distance of 1.2 m. All words were presented in black uppercase 30-point Helvetica font (subtending a vertical visual angle of 0.5° and a maximum horizontal visual angle of 4°) on a yellow patch (4.5° x 1°) that was slightly larger than the longest word. Words in the study phase were presented near one of the 4 corners of either a centrally presented landscape or solid gray background, whereas words in the test phase were presented centrally on a single textured background that was constructed by heavily blurring and pixelating an unused landscape picture. The backgrounds in both phases subtended 6.5° x 7° and were displayed against a black background. During the interstimulus intervals, a white or red (see Procedure) central fixation character ("+"; 0.5° x 0.5°) was presented.
Procedure
Instructions and practice were administered outside the scanner. The experiment proper consisted of a study phase followed by a test phase. Scanning took place during both phases, which were separated by a break of around 3 min. Each phase began and ended with the presentation of a white fixation character for 12.5 s.
For the study phase, subjects were informed that they would see a series of words that would be presented within either a landscape or a plain gray background. For words presented on landscape backgrounds (the "scene" condition), subjects were instructed to form a mental image of the object corresponding to the word appearing somewhere in the natural scene. For words presented on a gray background (the "sentence" condition), instructions were to generate a meaningful sentence that incorporated the word. During the initial practice phase, examples of study phase stimuli were provided, and subjects were required to vocalize their responses to ensure that they were complying with the task instructions. In addition, subjects were encouraged to complete the 2 tasks to the best of their ability throughout the study phase proper, so that they could better remember the words on the later test. Study trials were self-paced. Subjects were instructed to begin the task as soon as the word appeared and to press a button with their right index finger when they had finished generating an appropriate image or sentence, depending on the word's background.
Study phase trials consisted of a red fixation character for 500 ms, followed by a word presented on either a landscape or gray background until the response button was pressed. After the response, a white fixation character was displayed for 1 s until the next trial began. The study phase also contained 60 "null" trials, during which the white fixation character was continuously displayed for 2950 ms and no response was required. The null trials were interspersed with the test trials to stochastically vary stimulus onset, thereby allowing for more efficient estimation of event-related responses with respect to the interstimulus baseline (Josephs and Henson 1999). The order of study trials was chosen randomly for each subject, with a limit of 3 consecutive trials of a given type (including null trials). The study phase was divided into 3 blocks (60 trials each) by 2, 20-s rest periods that consisted of a visually presented instruction ("take a short break"). There was no communication between the subject and the experimenter during the breaks.
For the test phase, subjects were informed that they would see a series of words, each of which had either been presented in the previous study phase ("old") or was not studied ("new"). Each test word was presented on a single pixelated background (see Stimuli) that was constructed to be roughly equal in similarity to the landscape and gray backgrounds used at study. Subjects were informed that the background on which each test word was presented was irrelevant to their task, providing no indication as to whether the word had appeared in the study phase, or that it had appeared on a landscape or gray background. Instructions for the test phase followed standard instructions for the "remember/know" (R/K) procedure (e.g., Rajaram 1993), and examples were provided during the practice phase. Subjects were to make one of 3 responses to each test word, according to whether 1) any details about the word's study presentation could be recollected ("remember," R), 2) the word was judged to have appeared at study but no details could be recollected ("know," K), or 3) the word was judged not to have been presented in the study phase ("new," N). To minimize the influence of guessing on R and K responses, subjects were further instructed to use these 2 responses only when they were highly confident that a test word appeared in the study phase, and to use the N response in the event that they were unsure of a word's study status. The R, K, and N responses were made with the index, middle, and ring fingers, respectively, of the right hand. Accuracy and speed of responding were given equal emphasis in the test instructions.
Test-phase trials consisted of a red fixation character for 500 ms, followed by a word on the pixelated background for 500 ms, and then a white fixation character for 1950 ms. This trial sequence resulted in a stimulus-onset asynchrony (SOA) of 2950 ms. Sixty null trials were randomly interspersed with the test stimuli in the same manner as the study trials. The test phase was divided into 3 equal blocks (80 trials each) separated by 2, 20-s rest periods.
Scanning Parameters
A Philips Eclipse 1.5-T MR scanner (Philips Medical Systems, Bothell, WA) was used to acquire 27 T2*-weighted echoplanar images (EPIs; 64 x 92, 2.6 x 3.9-mm2 pixels; time echo = 40 ms) per volume, with blood oxygenation level–dependent (BOLD) contrast. EPIs comprised 3-mm-thick axial slices separated by 1.5-mm interslice gaps, and were positioned to give full coverage of the cerebrum and most of the cerebellum. EPIs were acquired sequentially in descending order, with a repetition time (TR) of 2500 ms. Because of the self-paced nature of the trials in the study phase, scanning continued until the subject finished the study phase (mean number of volumes = 390). (Scanning during the study phase was set to collect a maximum of 480 volumes for each subject but as noted earlier 2 subjects did not finish the study phase within this time). The test phase comprised 322 volumes. The ratio of SOA to TR in the test phase resulted in a sampling of the impulse response at a rate of approximately 2 Hz over trials. The first 5 volumes from each phase were discarded to allow for T1 equilibration effects. A T1-weighted anatomical volume (1-mm3 voxels) covering the whole brain was acquired for each subject following the test phase.
fMRI Analysis
Processing and analysis of the fMRI data were performed with Statistical Parametric Mapping (SPM2; Wellcome Department of Cognitive Neurology, London, UK; www.fil.ion.ucl.ac.uk/spm). Volumes were spatially realigned to the first volume of the study phase, and the data in each volume were temporally realigned to the acquisition time of its middle slice. The resulting images were spatially normalized to a standard EPI template based on the Montreal Neurological Institute (MNI) reference brain (Cocosco et al. 1997) and resampled into 3-mm3 voxels using nonlinear basis functions (Asburner and Friston 1999). The normalized data were smoothed with an isotropic 8-mm full-width half-maximum Gaussian kernel. The time series in each voxel was high-pass filtered at 1/128 Hz to remove low-frequency noise and scaled to a grand mean of 100, averaged over all voxels and volumes within a session.
Statistical analysis was performed on the study and test-phase data using a 2-stage mixed effects model. In the first stage, the neural activity elicited by each study word was modeled by a trial-specific boxcar function extending from stimulus onset to 300 ms prior to response. The neural activity elicited by each test word was modeled by a delta function (impulse event) at stimulus onset. The breaks during both phases were modeled as boxcar functions (length = 8 TRs). The ensuing BOLD response was modeled by convolving these functions with a canonical hemodynamic response function (HRF) and its temporal and dispersion derivatives (Friston et al. 1998). This convolution was performed in high-resolution time space and downsampled at the midpoint of each scan to form covariates in a General Linear Model. Results from the HRF derivatives (available upon request from the corresponding author) did not add any theoretically meaningful information to those from the canonical HRF and are not reported.
For the study phase, 4 covariates based on event types of interest were defined: words from the scene condition that were later given R responses, words from the sentence condition that were later given R responses, words that were later given K responses, and words that were later given N responses. Because of the low numbers of study words in each of the scene and sentence conditions that were later given K or N response, each of these 2 event types was collapsed over the 2 study conditions. For the test phase, 7 covariates corresponding to events of interest were defined: Words from the scene condition that were given R responses, words from the sentence condition that were given R responses, old words that were given K responses, old words that were given N responses, new words that were given R responses, new words that were given K responses, and new words that were given N responses. As in the study phase, the old words given K and N responses were collapsed over the scene and sentence conditions, due to low numbers of trials in the separate conditions for some subjects. This collapsing allowed for adequate estimation of the BOLD activity corresponding to old items given K responses, which were used in the recollection contrast (R > K) that was crucial to our hypotheses. In addition, 9 covariates of no interest were defined for each phase, representing the breaks, the mean (constant) for the phase, a single covariate for all other stimulus-related activity (buffer items and omitted or multiple responses), and 6 covariates corresponding to the movement parameters (3 rigid-body translation and 3 rotation) determined during realignment. The parameters for each covariate and the hyperparameters governing the error covariance were estimated using a restricted maximum likelihood method. Nonsphericity of the error covariance was accommodated by an AR(1) model, in which the temporal autocorrelation was estimated by pooling over suprathreshold voxels (Friston et al. 2002).
The second stage of analysis utilized linear contrasts of the aforementioned parameter estimates, treating subjects as a random effect. These contrasts consisted of one-sample t-tests, and unless otherwise noted, were thresholded for 5 or more contiguous voxels surviving P < 0.001 (uncorrected for multiple voxel-wise comparisons). Regions exhibiting activity that overlapped across multiple contrasts were identified by inclusively masking the relevant SPMs. Regions where activity was specific to one of multiple contrasts were identified by exclusive masking.
An across-subjects mean (N = 16) T1-weighted anatomical scan was created, after normalizing each scan to a standard T1 template of the MNI brain (Cocosco et al. 1997), and resampling into 2-mm3 voxels (Asburner and Friston 1999). Regions showing significant effects were localized either on sections of this mean image or on the rendered surface of a canonical normalized brain (with the cerebellum artificially removed). (We thank Rik Henson for providing the rendered brain.) The peak voxels of clusters exhibiting reliable effects are reported in stereotactic coordinates of MNI space.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Behavioral Results
The mean response times (RTs) during the study phase were slightly shorter for words presented in the sentence condition (4008 ms, standard deviation [SD] = 1581 ms) than in the scene condition (4290 ms, SD = 1342 ms) but this difference was not reliable.
The mean proportions and RTs corresponding to "remember" (R), "know" (K), and "new" (N) responses in the test phase are provided in Table 1. Memory performance—as measured by the difference in probabilities of making an R response to old versus new words (pRold – pRnew)—was higher for words from the sentence (0.67) than the scene condition (0.50; t[15] = 6.42, P < 0.001), although performance for both word types was reliably above chance (Ps < 0.001). Under the assumption that R and K responses depend upon independent processes (Yonelinas and Jacoby 1995), an analogous measure of memory performance based on K responses ([pKold/1 – pRold] – [pKnew/1 – pRnew]) was computed. This measure was also higher for words from the sentence (0.49) than the scene condition (0.38; t[15] = 2.62, P < 0.025). In addition, both of these measures were significantly greater than zero (both Ps < 0.001), confirming that the K responses to old items were not merely guesses.
|
Analysis of the mean RTs (collapsed across the 2 classes of old word) from the test phase indicated that correct K responses (1614 ms, SD = 392 ms) were slower than both correct R (1224 ms, SD = 288 ms; t[15] = 8.09, P < 0.001) and correct N responses (1344 ms, SD = 321 ms; t[15] = 4.22, P < 0.001), whereas the difference between the latter 2 only approached significance (t[15] = 2.12, P < 0.06). An additional analysis revealed that correct R responses were faster for words in the sentence than in the scene condition (t[15] = 3.51, P < 0.005). However, the numbers of incorrect responses and K responses for each class of old word were deemed insufficient for further RT analysis.
fMRI Results
Analysis of the fMRI data was conducted in 3 parts. In the first part of the analysis, the data from the study phase were used to identify regions that were differentially involved in the online processing of items studied in each of the 2 encoding tasks. In the second part of the analysis, the reinstatement hypothesis was evaluated by testing for overlap between regions exhibiting activity elicited during the recollection of items studied with a given encoding task and regions that were implicated in the "content-specific" processing of those items during encoding (from the first analysis). Regions where encoding- and recollection-related activity overlapped were restricted to those areas in which different levels of recollection-related activity were evident for items from one encoding condition rather than the other. Because a key prediction of the reinstatement hypothesis is that patterns of encoding- and recollection-related effects should mirror one another, this additional analysis step is critical because it excludes regions where recollection-related activity does not differ according to encoding condition. In a final analysis, the test-phase data were also used to identify regions exhibiting recollection-related activity that was invariant with respect to the encoding history of the items (i.e., "content-independent" recollection effects).
Content-Specific Study Activity
Regions exhibiting differential activity at study were identified by contrasting the activity elicited by words presented in the scene and sentence encoding tasks. To maximize the likelihood of overlap between study and test activity, these contrasts were based on activity elicited by only those words that were subsequently endorsed as recollected (given an R response) at test. (Additional contrasts based on all study trials, regardless of whether the corresponding words were subsequently recollected [endorsed with an R response] or not [endorsed with either a K or N response], were also conducted. The outcomes of these contrasts, including the regions demonstrating overlap with content-specific recollection-related activity, were qualitatively similar to those based on study trials associated with later recollected words.) The results of these contrasts are detailed in Table 2 and illustrated in Figure 1. As shown in the figure, the scene condition was associated with relatively greater activity in bilateral occipital cortex, extending dorsally to superior parietal cortex, and ventrally along the fusiform and parahippocampal gyri. The reverse contrast revealed greater activity for the sentence condition in medial frontal cortex, posterior cingulate, and several bilateral temporal regions (see Table 2 and Fig. 1).
|
|
Overlap of Content-Specific Study Activity and Recollection-Related Activity
Regions associated with recollection of words from each study condition were identified by contrasting the test-phase activity elicited by studied words endorsed as recollected with the activity elicited by studied words accorded K responses (collapsed over the 2 study conditions, as described in Materials and Methods). To identify regions that exhibited overlapping encoding- and recollection-related activity, each recollection contrast was inclusively masked with the corresponding study contrast. Thus, the contrast identifying regions associated with recollection of items studied with the scene task was masked with the scene > sentence encoding contrast, and the recollection contrast for items from the sentence task was masked with the reverse encoding contrast. In each case, both contrasts were thresholded at P < 0.001, giving a conjoint threshold of P < 10–5 (Fisher 1950; Lazar et al. 2002). For the reason noted previously, a further inclusive mask was then applied to identify the voxels where recollection-related activity was significantly greater for items studied with a given encoding task rather than the alternative task. In each case, this final mask comprised the outcome of the relevant directional contrast (thresholded at P < 0.05) between test-phase activity elicited by the 2 classes of recollected word (e.g., for the overlap between encoding- and recollection-related activity pertaining to the scene task, the mask consisted of the scene > sentence contrast for the recollected items). The outcomes of these 2-step masking procedures are listed in Table 3 and illustrated in Figure 2.
|
|
As is illustrated in Figure 2, overlap between encoding- and recollection-related activity for the items associated with the scene task was found in left superior and lateral occipital cortex (Brodmann area [BA] 19), along with left anterior fusiform gyrus (BA 37). As is also illustrated in the figure, one region—ventromedial frontal cortex (BA 10)—demonstrated overlap between encoding- and recollection-related activity for items from the sentence task.
The foregoing results support the reinstatement hypothesis in as much as they indicate that there are regions where activity associated with the encoding and recollection of items from a given study task overlaps. This support would be undermined, however, if regions also exist where activity overlaps between the encoding of items from one study task and recollection of items from the alternative task. To investigate this possibility, masking procedures analogous to those employed in the foregoing analyses were also used to test for overlap between regions active during the encoding of items in the scene task and the recollection of items from the sentence task, and vice versa. These additional analyses identified no overlapping voxels in either case.
Content-Independent Recollection-Related Activity
A final analysis was directed at identifying regions associated with recollection-related activity common to the 2 classes of recollected words. This was achieved by first computing a "global" recollection contrast that identified where activity elicited by recollected words from both encoding tasks was greater than the activity elicited by words accorded K responses. The outcome of this contrast was then exclusively masked with the 2 directional contrasts of test-phase activity that identified differences between the 2 classes of recollected item (i.e., recollection-related activity that was greater for items from the scene than the sentence task, and vice versa, each thresholded at P < 0.05). The outcome of this procedure is detailed in Table 4 and illustrated in Figure 3. Among the regions identified were left lateral posterior parietal cortex (BA 39), left parahippocampal gyrus (in the vicinity of entorhinal cortex), and retrosplenial cortex (BA 30).
|
|
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
According to the reinstatement hypothesis outlined in the introduction, neural activity associated with the processing of an episode as it is encoded should be "reactivated" when the episode is later recollected. The present study employed a design in which the neural correlates of encoding and later recollection could be directly compared within subjects. Consistent with the reinstatement hypothesis, regions where recollection-related activity differentiated test items according to their encoding history overlapped with some of the regions in which activity differed at the time of encoding. Moreover, this pattern of study-test overlap was selective; no regions were identified where recollection-related activity associated with items from a particular task overlapped with encoding-related activity for items from the alternative task. An additional network of regions exhibited recollection-related activity that was invariant with respect to the encoding history of the test item.
For items studied in the scene encoding task, recollection- and encoding-related activity overlapped in 3 left posterior regions, all of which have been implicated previously in the online processing of visual information. Superior occipital cortex has been reported to be engaged during scene processing (Nakamura et al. 2000), whereas the more inferior occipital region (part of the "lateral occipital complex") has been implicated in visual object processing (Malach et al. 1995; Goh et al. 2004; for review, see Grill-Spector et al. 2001). The left anterior fusiform area is in the vicinity of a region—the "parahippocampal place area"—that is reported to be selectively engaged during scene processing (Epstein and Kanwisher 1998; Epstein et al. 1999), as well as during the recollection of visual objects (Wheeler and Buckner 2003, 2004; Woodruff et al. 2005; Wheeler et al. 2006). An obvious interpretation of these findings is that recollection of words encoded within the scene task was associated with reinstatement of the visual processing engaged in service of the task, which required integration of scenes and internally generated object representations derived from the superimposed words. Interpretation of the single region (ventromedial prefrontal cortex) that exhibited overlap between encoding- and recollection-related activity for items studied in the sentence encoding task is less straightforward. This region has been implicated in recollection in 2 previous studies (Cansino et al. 2002; Yonelinas et al. 2005) that employed different experimental materials (pictures in the case of Cansino et al.; words in Yonelinas et al.) and encoding tasks (artificial/natural judgment in Cansino et al.; abstract/concrete judgment in Yonelinas et al.). The study and retrieval processes that might have been shared between these studies and the present study are obscure. Activity in the vicinity of this region has also been reported in a variety of tasks that involve "self-referential" processing or "mentalizing" (for reviews, see Amodio and Frith 2006; Northoff et al. 2006). It is unclear, however, whether covert sentence generation would engage such processes to a greater extent than the scene encoding task.
A noteworthy feature of the present findings is that, for both study conditions, the overlap between encoding- and recollection-related activity was confined to only a small fraction of the voxels that demonstrated differential activity at encoding. One factor that might have contributed to this disparity was the employment of the contrast between items endorsed as remembered versus known to identify recollection-related activity. If items endorsed with "know" responses are also associated with some degree of cortical reinstatement (as might be expected under the proposal that the memory signals underlying "remember" and "know" responses differ quantitatively rather than qualitatively; e.g., Dunn 2004), the foregoing contrast will likely underestimate the overlap between study- and retrieval-related activity. Accordingly, we repeated our original analyses, this time employing the contrast between items correctly endorsed with "remember" versus "new" responses, to generate a less restrictive estimate of retrieval-related activity. These secondary analyses yielded results that were both quantitatively and qualitatively very similar to our original findings. (We thank an anonymous reviewer for pointing out the need for these further analyses, the detailed results of which are available from the first author on request.) Another factor that might have contributed to the limited overlap between encoding- and recollection-related activity was our requirement that, to qualify as evidence of cortical reinstatement, a recollection effect had to both overlap with the appropriate content-specific encoding-related effect, and also be of greater magnitude than the recollection effect for the alternate condition. As discussed previously, the latter requirement is critical to the reinstatement hypothesis because it permits the identification of regions that are differentially sensitive to the retrieval of only one class of episode. Relaxation of this latter requirement, however, led to only a single additional overlapping cluster for each class of study episode (scene: right superior occipital cortex [BA 19]; sentence: posterior cingulate [BA 31]).
It would appear, therefore, that the limited study-test overlap observed here is due to the fact that only a small fraction of the processes and representations activated during the course of the study episodes was reinstated upon later recollection. As already discussed, this limited overlap is not a consequence of the contrasts employed to identify content-specific recollection-related activity. Likewise, it is also unlikely that it reflects the employment of an overly strict statistical threshold for these contrasts: Whereas relaxation of this threshold led to an expansion of regions demonstrating encoding–retrieval overlap, voxels demonstrating overlap remained only a small fraction of those exhibiting differential encoding activity. That said, it is of course possible that potentially more sensitive methods for detecting similarity between study- and test-related activity, such as those based on multivoxel pattern classification (e.g., Polyn et al. 2005; for review, see Norman et al. 2006), might reveal more extensive patterns of overlap. Assuming that the present findings are not just a consequence of a lack of statistical power, one possible explanation for the limited overlap between encoding- and recollection-related activity is that it reflects the fact that only a subset of the information comprising an episode—those aspects of the episode that enter awareness, for example (Moscovitch 1992)—is actually encoded into memory. Another possibility is that only a small portion of encoded information was actually recollected at test, perhaps because the test items served as relatively weak retrieval cues, or because the test task did not necessitate retrieval of highly differentiated episodic information (but rather only information sufficient to support a "remember" response). The relative importance of these factors in accounting for the present findings is unclear but deserves attention in future research.
In addition to regions putatively demonstrating content-specific reinstatement during recollection, regions where recollection-related activity was invariant with respect to encoding history were also identified. These "generic" recollection effects were evident in left lateral posterior parietal cortex, left entorhinal cortex, and retrosplenial cortex (see Fig. 3). One possible interpretation of these findings is that they reflect the reinstatement of specific processes or classes of information (e.g., conceptual representations) that were common to the 2 encoding conditions. However, the regions reported here as exhibiting generic recollection effects correspond well with regions identified as recollection-sensitive in several prior studies that, between them, employed a wide range of materials and study tasks (Henson et al. 1999; Eldridge et al. 2000; Cansino et al. 2002; Wheeler and Buckner 2004; Woodruff et al. 2005; Yonelinas et al. 2005). It is possible that the identification of these regions across all of these studies reflects the recollection of a common class of information. Alternatively, these seemingly content-independent recollection effects may indicate the existence of a "core" recollection network that is engaged regardless of the nature of the information recollected.
The co-occurrence of content-specific and (putative) generic recollection effects raises the question of how the processes associated with these 2 types of effect might interact to support recollection. The general framework for episodic retrieval outlined in the introduction, whereby the cortical reinstatement of encoding-related activity is hippocampally mediated, offers a possible account of the role of the content-independent entorhinal and retrosplenial recollection effects. In the animal literature, these 2 regions have been shown to be densely interconnected (for review, see Wyss and Van Groen 1992), and, in addition, each projects to a wide range of other cortical regions (Burwell and Amaral 1998; Kobayashi and Amaral 2003). Further, entorhinal cortex is the primary input and output relay between the hippocampus and the rest of the forebrain (Suzuki and Amaral 1994; Amaral and Witter 1995). Under the assumption that recollection depends upon the hippocampally mediated reinstatement of cortical activity, the present findings may indicate that entorhinal and retrosplenial cortices serve as key conduits for afferent information driving cortical reinstatement. A role for lateral parietal cortex is less obviously accommodated within this framework. Whereas this region is among the cortical areas most consistently identified with recollection in neuroimaging studies (Henson et al. 1999; Eldridge et al. 2000; Cansino et al. 2002; Wheeler and Buckner 2004; Woodruff et al. 2005; Yonelinas et al. 2005), the functional significance of recollection-related parietal activity remains enigmatic (for further discussion, see Wagner et al. 2005). Beyond providing additional support for the association of this region with recollection, the present findings do nothing to elucidate this issue.
The foregoing discussion is predicated on the assumption that the present findings of overlap between encoding- and recollection-related activity do indeed reflect the fact that recollection is mediated through cortical reinstatement, and that the content of what is recollected is derived in large part from the pattern of reinstated activity. One important caveat to this account is in order, however (for similar caveats, see Kahn et al. 2004; Woodruff et al. 2005). The present findings cannot rule out the alternative possibility that "reinstatement effects" are a consequence, rather than a cause, of successful recollection. By this argument, recollection depends on processes associated with activity in regions distinct from those exhibiting content-specific effects; these effects, in turn, reflect postrecollective processes—such as the deployment of attention toward particular types of recollected information or the maintenance of that information in working memory—that operate in service of task goals.
To conclude, the present study constitutes arguably the strictest test of the cortical reinstatement hypothesis thus far conducted with functional neuroimaging methods. Encoding- and retrieval-related activity was characterized within subjects for 2 different types of study episode, permitting a search for regional double dissociations in encoding–retrieval overlap. In addition, the retrieval task ensured that recollection-related activity could be separated both from more general neural correlates of retrieval, and from effects due to differential retrieval cue processing. These findings suggest that the neural correlates of recollection comprise both content-independent and content-specific activity. The overlap between the content-specific activity and the activity engaged when the same content was originally encoded lends strong support to the reinstatement hypothesis.
|
Acknowledgments |
|---|
This research was supported by a National Institutes of Heath grant (MH-072966). We thank Brian Minton for assistance with stimulus construction, the members of the UCI Research Imaging Center for their assistance with fMRI data acquisition, and Ken Norman for helpful comments on a previous version of this manuscript. Conflict of Interest: None declared.
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Alvarez P, Squire LR. Memory consolidation and the medial temporal lobe: a simple network model. Proc Natl Acad Sci USA (1994) 91:7041–7045.
Amaral DG, Witter MP. The hippocampal formation. In: The rat nervous system—Paxinos G, ed. (1995) San Diego, CA: Academic Press. 443–493.
Amodio DM, Frith CD. Meeting of minds: the medial frontal cortex and social cognition. Nat Rev Neurosci (2006) 7:268–277.[CrossRef][Web of Science][Medline]
Asburner J, Friston K. Non linear spatial normalization using basis functions. Hum Brain Mapp (1999) 7:254–266.[CrossRef][Web of Science][Medline]
Burwell RD, Amaral DG. Cortical afferents of the perirhinal, postrhinal, and entorhinal cortices of the rat. J Comp Neurol (1998) 398:179–205.[CrossRef][Web of Science][Medline]
Cansino S, Maquet P, Dolan RJ, Rugg MD. Brain activity underlying encoding and retrieval of source memory. Cereb Cortex (2002) 12:1048–1056.
Cocosco C, Kollokian V, Kwan RS, Evans A. Brainweb: online interface to a 3D MRI simulated brain database. Neuroimage (1997) 5:S425.
Dunn JC. Remember-know: a matter of confidence. Psychol Rev (2004) 111:524–542.[CrossRef][Web of Science][Medline]
Eldridge LL, Knowlton BJ, Furmanski CS, Bookheimer SY, Engel SA. Remembering episodes: a selective role for the hippocampus during retrieval. Nat Neurosci (2000) 3:1149–1152.[CrossRef][Web of Science][Medline]
Epstein R, Harris A, Stanley D, Kanwisher N. The parahippocampal place area: recognition, navigation, or encoding? Neuron (1999) 23:115–125.[CrossRef][Web of Science][Medline]
Epstein R, Kanwisher N. A cortical representation of the cortical visual environment. Nature (1998) 392:598–601.[CrossRef][Medline]
Fisher RA. Statistical methods for research workers. (1950) 11th ed. London: Oliver and Boyd.
Friston KJ, Fletcher P, Josephs O, Holmes A, Rugg MD, Turner R. Event-related fMRI: characterizing differential responses. Neuroimage (1998) 7:30–40.[CrossRef][Web of Science][Medline]
Friston KJ, Glaser DE, Henson RNA, Kiebel S, Phillips C, Ashburner J. Classical and Bayesian inference in neuroimaging: applications. Neuroimage (2002) 16:484–512.[CrossRef][Web of Science][Medline]
Goh JO, Siong SC, Park D, Gutchess A, Hebrank A, Chee MW. Cortical areas involved in object, background, and object-background processing revealed with functional magnetic resonance adaptation. J Neurosci (2004) 24:10223–10228.
Gottfried JA, Smith APR, Rugg MD, Dolan RJ. Remembrance of odors past: human olfactory cortex in cross-modal recognition memory. Neuron (2004) 42:687–695.[CrossRef][Web of Science][Medline]
Grill-Spector K, Kourtzi Z, Kanwisher N. The lateral occipital complex and its role in object recognition. Vision Res (2001) 41:1409–1422.[CrossRef][Web of Science][Medline]
Henson RN, Rugg MD, Shallice T, Josephs O, Dolan RJ. Recollection and familiarity in recognition memory: an event-related fMRI study. J Neurosci (1999) 19:3962–3972.
Josephs O, Henson RN. Event-related functional magnetic resonance imaging: modelling, inference and optimization. Philos Trans R Soc Lond B Biol Sci (1999) 354:1215–1228.
Kahn I, Davachi L, Wagner AD. Functional-neuroanatomic correlates of recollection: implications for models of recognition memory. J Neurosci (2004) 28:4172–4180.
Khader P, Burke M, Bien S, Ranganath C, Rosler F. Content-specific activation during associative long-term memory retrieval. Neuroimage (2005) 27:805–816.[CrossRef][Web of Science][Medline]
Kobayashi Y, Amaral DG. Macaque monkey retrosplenial cortex: II. Cortical afferents. J Comp Neurol (2003) 466:48–79.[CrossRef][Web of Science][Medline]
Kucera H, Francis WN. Computational analysis of present-day American English (1967) Providence, RI: Brown University Press.
Lazar NA, Luna B, Sweeney JA, Eddy WF. Combining brains: a survey of methods for statistical pooling of information. Neuroimage (2002) 16:538–550.[CrossRef][Web of Science][Medline]
Malach R, Reppas JB, Benson RR, Kwong KK, Jiang H, Kennedy WA, Ledden PJ, Brady TJ, Rosen BR, Tootell RB. Object-related activity revealed by functional magnetic resonance imaging in human occipital cortex. Proc Natl Acad Sci USA (1995) 92:8135–8139.
Moscovitch M. Memory and working with memory: a component process model based on modules and central systems. J Cogn Neurosci (1992) 4:257–267.[CrossRef][Web of Science]
Nakamura K, Kawashima R, Sato N, Nakamura A, Sugiura M, Kato T, Hatano K, Ito K, Fukuda H, Schormann T, et al. Functional delineation of the human occipito-temporal areas related to face and scene processing: a PET study. Brain (2000) 123:1903–1912.
Norman KA, O'Reilly RC. Modeling hippocampal and neocortical contributions to recognition memory: a complementary-learning-systems approach. Psychol Rev (2003) 110:611–646.[CrossRef][Web of Science][Medline]
Norman KA, Polyn SM, Detre GJ, Haxby JV. Beyond mind-reading: multi-voxel pattern analysis of fMRI data. Trends Cogn Sci (2006) 10:424–430.[CrossRef][Web of Science][Medline]
Northoff G, Heinzel A, de Greck M, Bermpohl F, Dobrowolny H, Panksepp J. Self-referential processing in our brain—a meta-analysis of imaging studies on the self. Neuroimage (2006) 31:440–457.[CrossRef][Web of Science][Medline]
Nyberg L, Habib R, McIntosh AR, Tulving E. Reactivation of encoding-related brain activity during memory retrieval. Proc Natl Acad Sci USA (2000) 97:11120–11124.
Nyberg L, Petersson KM, Nilsson LG, Sandblom J, Aberg C, Ingvar M. Reactivation of motor brain areas during explicit memory for actions. Neuroimage (2001) 14:521–528.[CrossRef][Web of Science][Medline]
Persson J, Nyberg L. Conjunction analyses of cortical activations common to encoding and retrieval. Microsc Res Tech (2000) 51:39–44.[CrossRef][Web of Science][Medline]
Polyn SM, Natu VS, Cohen JD, Norman KA. Category-specific cortical activity precedes retrieval during memory search. Science (2005) 310:1963–1966.
Rajaram S. Remembering and knowing: two means of access to the personal past. Mem Cognit (1993) 21:89–102.[Web of Science][Medline]
Rolls ET. Hippocampo-cortical and cortico-cortical backprojections. Hippocampus (2000) 10:380–388.[CrossRef][Web of Science][Medline]
Rubin DC, Greenberg DL. Visual memory-deficit amnesia: a distinct amnesic presentation and etiology. Proc Natl Acad Sci USA (1998) 95:5413–5416.
Shastri L. Episodic memory and cortico-hippocampal interactions. Trends Cogn Sci (2002) 6:162–168.[CrossRef][Web of Science][Medline]
Suzuki WA, Amaral DG. Topographic organization of the reciprocal connections between the monkey entorhinal cortex and the perirhinal and parahippocampal cortices. J Neurosci (1994) 14:1856–1877.[Abstract]
Tulving E. Memory and consciousness. Can Psychol (1985) 26:1–12.[Web of Science]
Vaidya CJ, Zhao M, Desmond JE, Gabrieli JDE. Evidence for cortical encoding specificity in episodic memory: memory-induced re-activation of picture processing areas. Neuropsychologia (2002) 40:2136–2143.[CrossRef][Web of Science][Medline]
Wagner AD, Shannon BJ, Kahn I, Buckner RL. Parietal lobe contributions to episodic memory retrieval. Trends Cogn Sci (2005) 9:445–453.[Web of Science][Medline]
Wheeler ME, Buckner RL. Functional dissociation among components of remembering: control, perceived oldness, and content. J Neurosci (2003) 23:3869–3880.
Wheeler ME, Buckner RL. Functional-anatomic correlates of remembering and knowing. Neuroimage (2004) 21:1337–1349.[CrossRef][Web of Science][Medline]
Wheeler ME, Petersen SE, Buckner RL. Memory's echo: vivid remembering reactivates sensory-specific cortex. Proc Natl Acad Sci USA (2000) 97:11125–11129.
Wheeler ME, Shulman GL, Buckner RL, Miezin FM, Velanova K, Petersen SE. Evidence for separate perceptual reactivation and search processes during remembering. Cereb Cortex (2006) 16:949–959.
Woodruff CC, Johnson JD, Uncapher MR, Rugg MD. Content-specificity of the neural correlates of recollection. Neuropsychologia (2005) 43:1022–1032.[CrossRef][Web of Science][Medline]
Wyss JM, Van Groen T. Connections between the retrosplenial cortex and the hippocampal formation in the rat: a review. Hippocampus (1992) 2:1–11.[CrossRef][Web of Science][Medline]
Yonelinas AP, Jacoby LL. The relation between remembering and knowing as bases for recognition: effects of size congruency. J Mem Lang (1995) 34:622–643.[CrossRef][Web of Science]
Yonelinas AP, Otten LJ, Shaw KN, Rugg MD. Separating the brain regions involved in recollection and familiarity in recognition memory. J Neurosci (2005) 25:3002–3008.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  R. A. Epstein and E. J. Ward How Reliable Are Visual Context Effects in the Parahippocampal Place Area? Cereb Cortex, June 16, 2009; (2009) bhp099v2. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. G. R. McDuff, H. C. Frankel, and K. A. Norman Multivoxel Pattern Analysis Reveals Increased Memory Targeting and Reduced Use of Retrieved Details during Single-Agenda Source Monitoring J. Neurosci., January 14, 2009; 29(2): 508 - 516. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
N. Abe, J. Okuda, M. Suzuki, H. Sasaki, T. Matsuda, E. Mori, M. Tsukada, and T. Fujii Neural Correlates of True Memory, False Memory, and Deception Cereb Cortex, December 1, 2008; 18(12): 2811 - 2819. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Duarte, R. N. Henson, and K. S. Graham The Effects of Aging on the Neural Correlates of Subjective and Objective Recollection Cereb Cortex, September 1, 2008; 18(9): 2169 - 2180. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||