Elsevier

NeuroImage

Volume 54, Issue 3, 1 February 2011, Pages 2446-2461
NeuroImage

Neural activity that predicts subsequent memory and forgetting: A meta-analysis of 74 fMRI studies

https://doi.org/10.1016/j.neuroimage.2010.09.045Get rights and content

Abstract

The present study performed a quantitative meta-analysis of functional MRI studies that used a subsequent memory approach. The meta-analysis considered both subsequent memory (SM; remembered > forgotten) and subsequent forgetting (SF; forgotten > remembered) effects, restricting the data used to that concerning visual information encoding in healthy young adults. The meta-analysis of SM effects indicated that they most consistently associated with five neural regions: left inferior frontal cortex (IFC), bilateral fusiform cortex, bilateral hippocampal formation, bilateral premotor cortex (PMC), and bilateral posterior parietal cortex (PPC). Direct comparisons of the SM effects between the studies using verbal versus pictorial material and item-memory versus associative-memory tasks yielded three main sets of findings. First, the left IFC exhibited greater SM effects during verbal material than pictorial material encoding, whereas the fusiform cortex exhibited greater SM effects during pictorial material rather than verbal material encoding. Second, bilateral hippocampal regions showed greater SM effects during pictorial material encoding compared to verbal material encoding. Furthermore, the left hippocampal region showed greater SM effects during pictorial-associative versus pictorial-item encoding. Third, bilateral PMC and PPC regions, which may support attention during encoding, exhibited greater SM effects during item encoding than during associative encoding. The meta-analysis of SF effects indicated they associated mostly with default-mode network regions, including the anterior and posterior midline cortex, the bilateral temporoparietal junction, and the bilateral superior frontal cortex. Recurrent activity oscillations between the task-positive and task-negative/default-mode networks may account for trial-to-trial variability in participants' encoding performances, which is a fundamental source of both SM and SF effects. Taken together, these findings clarify the neural activity that supports successful encoding, as well as the neural activity that leads to encoding failure.

Research Highlights

► SM effects associated mostly with content processing, storage, and attention regions. ► SF effects associated mostly with default-mode network regions. ► The left IFC and fusiform cortex support content processing. ► The MTL supports storage operations ► The PMC and PPC support attention during encoding.

Introduction

On any given day, we encounter and experience many events. Only some of these experiences are transformed into memories and can be subsequently remembered. A key line of inquiry for students of memory concerns the neural activity predicting events that will be remembered, as opposed to those that will be forgotten. The advent of event-related functional magnetic resonance imaging (fMRI) made an extremely powerful paradigm for addressing this question possible (Brewer et al., 1998, Wagner et al., 1998, Wagner et al., 1998). In this paradigm, participants are presented with a series of encoding stimuli (trials), and encoding stimuli are sorted into those that would be remembered versus those that would be forgotten, based on participants' performances on a subsequent memory test. The fMRI signal that is greater for the stimuli later remembered than for those later forgotten (called a subsequent memory [SM] effect) indicates the presence of neural activity that supports successful encoding. The reverse situation, i.e., greater fMRI signal for the stimuli later forgotten than for those later remembered (called a subsequent forgetting (SF) effect), indicates neural activity that interferes with successful encoding. The SM paradigm, with its “compelling” operational definition of successful encoding activity, was very popular throughout the past decade. To date, over 100 fMRI studies have used the SM approach (Uncapher and Wagner, 2009). The present study aimed to provide the first comprehensive meta-analysis of such literature. This meta-analysis, though restricted to the literature examining visual information encoding in healthy young adults, considered both SM and SF effects.

In addition to the generic purpose of integrating results across studies, the present meta-analysis investigated several specific hypotheses regarding SM effects. After reviewing the relevant literature, comprising all published fMRI studies using an SM approach, we observed the emergence of two major study divisions. In one group of studies, the to-be-remembered material was verbal (e.g., words), while in the other group, it was pictorial (e.g., pictures, scenes, or faces). Further, one group of studies used item-memory tasks, while the other group used associative-memory tasks. In an item-memory task, participants try to remember an item with no other associated information, whereas in an associative-memory task, they must remember both an item and some information associated with that item, e.g., the context in which the item was presented (item–context association) or the fact that two items were presented as a pair (item–item association). Based on the reviewed studies' two divisions, the present meta-analysis compared (i) the SM effects of those studies using verbal versus pictorial material and (ii) the SM effects of those studies using item-memory versus associative-memory tasks. These comparisons cut across several critical issues relating to episodic encoding activity (see below).

Neuroimaging studies examining SM effects have traditionally focused on the prefrontal cortex (PFC) and the medial temporal lobe (MTL; Buckner et al., 2001, Fernandez and Tendolkar, 2001, Simons and Spiers, 2003). However, other brain regions, such as the fusiform cortex (Dickerson et al., 2007, Garoff et al., 2005, Kim and Cabeza, 2007), posterior parietal cortex (PPC; Sommer et al., 2005a, Sommer et al., 2005b, Uncapher and Rugg, 2009), and premotor cortex (PMC; Kao et al., 2005, Morcom et al., 2003) have been associated with SM effects. The multiple brain regions associated with SM effects may be broadly divided into three types, (i) content processing, (ii) storage, and (iii) attention. First, content processing regions mediate “the transformation of sensory input into internal representations that are interpreted or comprehended” (Paller and Wagner, 2002). The PFC, particularly the left inferior frontal cortex (IFC) and fusiform cortex are canonical examples of such regions (Kirchhoff et al., 2000). Second, storage regions bind the content representations into a durable memory representation, which the individual can access and retrieve into consciousness later. The MTL, and, in particular, the hippocampal formation, is widely recognized as the key structure in this function (Diana et al., 2007, Squire et al., 2004). Finally, attention during encoding selects an event among competing inputs and biases its “online” processing. A leading model of visual attention (Corbetta et al., 2002, Corbetta et al., 2008) has implicated a frontoparietal network, which includes the PMC and PPC, as a critical attention-coding structure. This three-component model does not represent a strict categorization of encoding-related activity, but rather some useful heuristics that can guide further research. For example, all regions displaying SM effects may have at least some relevance to storage.

As stated above, the present meta-analysis also compared SM effects between studies using verbal versus pictorial material. Historically, researchers have framed such comparisons and associated results in terms of hemispheric specialization or laterality effects (Golby et al., 2001, Kelley et al., 1998). This presupposes that verbal material encoding depends more on left- than right-hemispheric processing, whereas pictorial material encoding involves more right- than left-hemispheric processing. However, in the present meta-analysis, the primary purposes for this comparison were to address the following two hypotheses.

The first hypothesis states processing and successfully encoding either verbal or pictorial materials, which are canonically different in content, will emphasize different content-processing regions. Specifically, the left IFC, known to support controlled semantic/phonological retrieval and analysis (Badre and Wagner, 2007, Buckner et al., 1999), may be critical for successfully encoding verbal material, whereas the fusiform cortex, known to mediate visuoperceptual analysis and differentiation (Garoff et al., 2005, Kanwisher et al., 2001), may be critical for successfully encoding pictorial material.

Second, many SM studies have failed to find significant effects in the MTL (for a review, see Henson, 2005), despite the MTL's widely accepted critical role in storage operations. This failure remains largely unexplained, though researchers have attributed it to a low signal-to-noise ratio, susceptibility to MRI artifacts, or other nuisance factors. One relevant factor may be the widespread use of high frequency words, which participants had encountered numerous times prior to the studies, as encoding stimuli. Thus, participants had high pre-experimental familiarity with the verbal material, but not the pictorial material, used in SM studies. One influential view of the encoding process, known as the novelty-encoding hypothesis (Tulving et al., 1996), suggested the encoding system is biased toward processing novel, as opposed to familiar, information, because the system evolved to register information having high survival value. Thus, the second hypothesis addressed in this comparison is a novelty-encoding hypothesis predicting greater SM effects in the MTL during the encoding of pictorial material as compared to verbal material.

The present meta-analysis also compared SM effects between those studies using an item-memory task versus those with an associative-memory task. Historically, the distinction between item- versus associative-memory related to the distinction between familiarity (i.e., a feeling of “oldness” in the absence of contextual details) and recollection (i.e., vividly remembering specific contextual details). For example, a dual-process model of recognition memory suggested that associative recognition reflects recollection-based responses, whereas item recognition reflects both recollection- and familiarity-based responses (Yonelinas, 1997). The primary purposes for the present comparison of item- versus recognition-memory were to test the following two hypotheses.

First, prior discussions of the neural substrates for item- versus associative-memory predominantly focused on the MTL (Brown and Aggleton, 2001, Eichenbaum et al., 2007). A critical MTL function in episodic encoding is binding or associating multiple internal representations linked to an event, so the individual can retrieve the resultant representation as a whole (Davachi, 2006, Diana et al., 2007, Squire et al., 2004). Even though an item-memory task implicitly involves associating an item with spatiotemporal characteristics of the study episode, an associative-memory task makes stronger demands on associations, by requiring explicit item-context or item-item associations. Thus, the first hypothesis states MTL regions will show more robust SM effects during an associative-encoding task than during an item-encoding task.

Second, the distinction between item- and associative-memory, though traditionally focused on the MTL, may also involve differential SM effects in other brain regions (Kirwan et al., 2008). An associative-encoding task, as compared to an item-encoding task, typically makes greater content-processing demands, presenting multiple pieces of information and requiring relational processing among them. For example, an item–item association task may ask participants to rate how well two members of a pair (e.g., word–word) fit together (e.g., Qin et al., 2007) or to form a mental image incorporating both members of a pair and rate the quality of the image (e.g., Jackson and Schacter, 2004). By contrast, an item encoding task typically involves a relatively simple semantic (e.g., Is the word concrete or abstract?) or visual judgment (e.g., Is the face male or female?). Thus, the second hypothesis states content processing regions, such as left IFC and fusiform cortex, will exhibit stronger SM effects during associative versus item encoding.

Finally, encoding makes demands on attention, as shown by extensive behavioral evidence demonstrating that divided attention had negative effects on encoding (e.g., Craik et al., 1996). However, SM studies' explicit documentation of attention-related effects are relatively recent (Kensinger et al., 2003, Uncapher and Rugg, 2009, Uncapher and Rugg, 2008). A recent meta-analysis (Uncapher and Wagner, 2009) of relevant literature focused on the PPC as mediating attention during encoding. However, mounting evidence suggests attention does not depend on a single region but rather on a network of regions that interact with each other (for a review, see Raz and Buhle, 2006). Thus, attention during encoding is unlikely to involve a single region, but rather multiple regions, which include the frontal as well as the parietal cortex. Both a leading visual attention model (Corbetta et al., 2002, Corbetta et al., 2008) and meta-analyses of attention and working memory studies (Owen et al., 2005, Wager et al., 2004) suggest perhaps a frontoparietal network, including both the PMC and PPC, supports attention during encoding. Thus, the present meta-analysis investigated whether the PMC and PPC regions showed significant SM effects, and, if so, whether the nature of the material (verbal versus pictorial) and/or the type of encoding (item versus associative) modulated these SM effects.

Otten and Rugg (2001b) were the first to describe regions that showed SF effects, also called “reversed” or “negative” SM effects (Duverne et al., 2009, Weis et al., 2004). Their findings indicated SF effects were associated with widespread cortical regions, including the inferior parietal, medial parietal, posterior cingulate, and superior frontal cortices. At a minimum, these findings indicated that, to understand encoding, researchers must pay attention, not only to the positive correlates of remembering (SM effects), but also to the negative correlates of remembering (SF effects). Though relatively few fMRI studies have focused on SF effects, researchers generally accept the existence of cortical regions associated with SF effects (Park and Rugg, 2008, Shrager et al., 2008). Given the relatively limited number of available studies, the present meta-analysis focused mainly on general SF effects, involving the whole group, rather than specific SF effects restricted to a subgroup.

Multiple prior studies (Daselaar et al., 2004, Kim et al., 2010, Park and Rugg, 2008, Shrager et al., 2008, Turk-Browne et al., 2006) noted that the regions associated with SF effects tended to be components of what has been termed the default-mode network, which consists of the anterior and posterior midline cortex, the temporoparietal junction (TPJ), and the superior frontal cortex (Raichle et al., 2001). Based on this evidence, the present study examined whether, and to what extent, SF effects associate with default-mode network regions. The default-mode network was originally defined as the set of regions that are more active during the passive resting state than during attention-demanding cognitive tasks (Raichle et al., 2001, Shulman et al., 1997, Laird et al., 2009a). Researchers are currently debating these regions' functions, but increasing evidence suggests they mediate self-referential processing, or, more generally, internally oriented processing, as indicated by higher activations (or less deactivations) of these regions during imagining the future, conceiving the viewpoint of others (theory of mind), and autobiographical memory (for reviews, see Buckner and Carroll, 2007, Gusnard and Raichle, 2001, Spreng et al., 2009). Of greater direct relevance to SF effects, activation of these regions during an exogenous task may signal a wandering mind or momentary lapse of attention (Christoff et al., 2009, Mason et al., 2007, McKiernan et al., 2006, Weissman et al., 2006). For example, Christoff et al. (2009), using experience sampling during an fMRI task, found direct evidence for an association between activation of default-mode network regions and mind-wandering. Thus, activation of these regions during encoding may take neural resources away from the processes that lead to effective remembering.

The present study's principal methodology was a quantitative (i.e., statistical) meta-analysis of the relevant literature. A primary use for meta-analyses in neuroimaging is identifying significant concordances in brain activity patterns across a set of independent studies using a specific paradigm (Wager et al., 2007). The discernment of findings' convergences and divergences among studies is becoming increasingly important, albeit more difficult, as neuroimaging data continue to accumulate at a rapidly accelerating pace (Laird et al., 2009b). A quantitative meta-analysis provides an efficient and bias-free means of accomplishing this. The results of the present meta-analysis identify the brain regions most reliably associated with SM or SF effects and those most consistently exhibiting modulation of SM effects by the nature of the material and/or by the type of encoding. Of four recent meta-analyses of neuroimaging data that included SM studies, only one (Spaniol et al., 2009) used a quantitative approach; however, it employed a limited database (26 studies). The other three used a tabular method and focused exclusively on the MTL (Diana et al., 2007), PFC (Blumenfeld and Ranganath, 2007), or PPC (Uncapher and Wagner, 2009). Thus, the present study is the first quantitative meta-analysis of SM studies based on a comprehensive database and a whole-brain approach.

Section snippets

Study selection

Multiple literature searches via Pubmed were completed in order to isolate all fMRI studies reporting SM or SF effects. Additionally, a reference list check of recent neuroimaging memory study reviews (Diana et al., 2007, Spaniol et al., 2009, Uncapher and Wagner, 2009) was done to identify relevant studies not found by the online database search. These search results were filtered to include only studies that (i) included healthy, young participants, (ii) presented encoding material via visual

SM effects

Table 2 and Fig. 1 show the ALE meta-analysis results for all included studies (see supplementary material available online (Supplementary Fig. 1) for a series of coronal views). The results indicated SM effects were mainly associated with five regions: the left IFC, bilateral fusiform cortex, bilateral MTL, bilateral PMC, and bilateral PPC, in approximate order of decreasing spatial extent. The left IFC cluster (Brodmann area [BA] 44, 45, 46, 47) included both the anterior and posterior extent

SM effects

These results revealed that SM effects associated most consistently with five neural regions: left IFC, bilateral fusiform cortex, bilateral hippocampal formation, bilateral PMC, and bilateral PPC. Moreover, the SM effects' magnitude in these regions showed reliable modulation by the nature of the material (verbal or pictorial) and/or by encoding type (item or associative). As per the three-component model outlined in Introduction section, activation of the left IFC and fusiform cortex regions

Conclusions

The present study performed a meta-analysis of functional MRI studies using an SM approach. The meta-analysis of SM effects indicated they most consistently associated with five neural regions: the left IFC, bilateral fusiform cortex, bilateral hippocampal formation, bilateral PMC, and bilateral PPC. Comparisons of SM effects among the four subgroups of studies, formed by crossing two major study divisions (verbal versus pictorial material and item versus associative encoding), yielded three

Acknowledgments

This work was supported by a Daegu University research grant in 2010.

References (144)

  • N.A. Dennis et al.

    Effects of aging on transient and sustained successful memory encoding activity

    Neurobiol. Aging

    (2007)
  • R. Diana et al.

    Imaging recollection and familiarity in the medial temporal lobe: a three-component model

    Trends Cogn. Sci.

    (2007)
  • S. Erk et al.

    Emotional context modulates subsequent memory effect

    NeuroImage

    (2003)
  • S. Erk et al.

    Emotional context during encoding of neutral items modulates brain activation not only during encoding but also during recognition

    NeuroImage

    (2005)
  • G. Fernandez et al.

    Integrated brain activity in medial temporal and prefrontal areas predicts subsequent memory performance: human declarative memory formation at the system level

    Brain Res. Bull.

    (2001)
  • P.C. Fletcher et al.

    Regional brain activations predicting subsequent memory success: an event-related fMRI study of the influence of encoding tasks

    Cortex

    (2003)
  • K. Fliessbach et al.

    The effect of word concreteness on recognition memory

    NeuroImage

    (2006)
  • K. Fliessbach et al.

    Cerebellar contributions to episodic memory encoding as revealed by fMRI

    NeuroImage

    (2007)
  • M. Fox et al.

    Intrinsic fluctuations within cortical systems account for intertrial variability in human behavior

    Neuron

    (2007)
  • P. Fransson

    How default is the default mode of brain function?: further evidence from intrinsic BOLD signal fluctuations

    Neuropsychologia

    (2006)
  • R.J. Garoff et al.

    The neural origins of specific and general memory: the role of the fusiform cortex

    Neuropsychologia

    (2005)
  • C. Genovese et al.

    Thresholding of statistical maps in functional neuroimaging using the false discovery rate

    NeuroImage

    (2002)
  • Y. Golland et al.

    Data-driven clustering reveals a fundamental subdivision of the human cortex into two global systems

    Neuropsychologia

    (2008)
  • B. Gonsalves et al.

    Memory strength and repetition suppression: multimodal imaging of medial temporal cortical contributions to recognition

    Neuron

    (2005)
  • L.J. Gottlieb et al.

    Dissociation of the neural correlates of visual and auditory contextual encoding

    Neuropsychologia

    (2010)
  • A.L. Haskins et al.

    Perirhinal cortex supports encoding and familiarity-based recognition of novel associations

    Neuron

    (2008)
  • O. Jackson et al.

    Encoding activity in anterior medial temporal lobe supports subsequent associative recognition

    NeuroImage

    (2004)
  • W. Kelley et al.

    Hemispheric specialization in human dorsal frontal cortex and medial temporal lobe for verbal and nonverbal memory encoding

    Neuron

    (1998)
  • H. Kim et al.

    Overlapping brain activity between episodic memory encoding and retrieval: roles of the task-positive and task-negative networks

    NeuroImage

    (2010)
  • A.R. Laird et al.

    Lost in localization? The focus is meta-analysis

    NeuroImage

    (2009)
  • C. Li et al.

    Greater activation of the “default” brain regions predicts stop signal errors

    NeuroImage

    (2007)
  • A. Maril et al.

    Feeling-of-knowing in episodic memory: an event-related fMRI study

    NeuroImage

    (2003)
  • B. McCandliss et al.

    The visual word form area: expertise for reading in the fusiform gyrus

    Trends Cogn. Sci.

    (2003)
  • K. McKiernan et al.

    Interrupting the “stream of consciousness”: an fMRI investigation

    NeuroImage

    (2006)
  • G. Northoff et al.

    Rest–stimulus interaction in the brain: a review

    Trends Neurosci.

    (2010)
  • L.J. Otten et al.

    When more means less: neural activity related to unsuccessful memory encoding

    Curr. Biol.

    (2001)
  • K. Paller et al.

    Observing the transformation of experience into memory

    Trends Cogn. Sci.

    (2002)
  • H. Park et al.

    Neural correlates of successful encoding of semantically and phonologically mediated inter-item associations

    NeuroImage

    (2008)
  • N. Axmacher et al.

    Interaction of working memory and long-term memory in the medial temporal lobe

    Cereb. Cortex

    (2008)
  • J.T. Baker et al.

    Neural correlates of verbal memory encoding during semantic and structural processing tasks

    NeuroReport

    (2001)
  • R. Blumenfeld et al.

    Prefrontal cortex and long-term memory encoding: an integrative review of findings from neuropsychology and neuroimaging

    Neuroscientist

    (2007)
  • M. Boly et al.

    Baseline brain activity fluctuations predict somatosensory perception in humans

    Proc. Natl. Acad. Sci. U. S. A.

    (2007)
  • J. Brewer et al.

    Making memories: brain activity that predicts how well visual experience will be remembered

    Science

    (1998)
  • M. Brown et al.

    Recognition memory: what are the roles of the perirhinal cortex and hippocampus?

    Nat. Rev. Neurosci.

    (2001)
  • R. Buckner et al.

    Frontal cortex contributes to human memory formation

    Nat. Neurosci.

    (1999)
  • R.L. Buckner et al.

    Encoding processes during retrieval tasks

    J. Cogn. Neurosci.

    (2001)
  • T. Canli et al.

    Sex differences in the neural basis of emotional memories

    Proc. Natl. Acad. Sci. U. S. A.

    (2002)
  • S. Cansino et al.

    Brain activity underlying encoding and retrieval of source memory

    Cereb. Cortex

    (2002)
  • K. Christoff et al.

    Experience sampling during fMRI reveals default network and executive system contributions to mind wandering

    Proc. Natl. Acad. Sci. U. S. A.

    (2009)
  • M. Corbetta et al.

    Neural systems for visual orienting and their relationships to spatial working memory

    J. Cogn. Neurosci.

    (2002)
  • Cited by (446)

    View all citing articles on Scopus
    View full text