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

Abstract

The notion that the brain is organized into two complementary networks, one that is task-positive and supports externally-oriented processing, and the other that is task-negative and supports internally-oriented processing, has recently attracted increasing attention. The goal of the present study was to investigate involvement of the task-positive and task-negative networks in overlapping activity between episodic memory encoding and retrieval. To this end, we performed a functional MRI study that included both encoding and retrieval tasks. We hypothesized that during the study phase, encoding success activity (remembered > forgotten) involves mainly the task-positive network, whereas encoding failure activity (forgotten > remembered) involves mainly the task-negative network. We also hypothesized that during the test phase, retrieval success activity (old > new) involves mainly the task-negative network, whereas novelty detection activity (new > old) involves mainly the task-positive network. Based on these hypotheses, we made 3 predictions regarding study-test overlap. First, there would be relatively high level of overlap between encoding success and novelty detection activity involving the task-positive network. Second, there would be relatively high level of overlap between encoding failure and retrieval success activity involving the task-negative network. Third, there would be relatively low level of overlap between encoding success and retrieval success activity as well as between encoding failure and novelty detection activity. The results fully confirmed our 3 predictions. Taken together, the present findings clarify roles of the task-positive and task-negative networks in encoding and retrieval and the function of overlapping brain activity between encoding and retrieval.

Introduction

Episodic memory consists of multiple subprocesses, and a fundamental distinction can be drawn between encoding and retrieval processes. Most prior functional neuroimaging studies of episodic memory have focused either on encoding or retrieval processes, whereas relatively few have directly compared the two processes within the same study. An influential view for relating encoding and retrieval activations has been the “reinstatement” hypothesis, which postulates that successful retrieval of episodic information involves reactivation of several of the brain regions that were activated during encoding of that information (e.g., Nyberg et al., 2000, Wheeler et al., 2000). Consistent with this hypothesis, several studies (e.g., Persson and Nyberg, 2000, Johnson and Rugg, 2007) have found significant activation overlaps between encoding and retrieval. However, in practically all studies that have examined this issue, the extent of overlap was highly limited, involving only a small fraction of sensory/perceptual regions activated during encoding. Thus, the reinstatement hypothesis, while helpful for interpreting memory activity in parts of sensory/perceptual regions, is of limited value for understanding memory activity in other brain regions.

Here, we propose a more global, system-wide model to relate brain activity elicited during study (encoding) and test (retrieval) phases, and we test 3 predictions derived from the model. A fundamental characteristic of the model is that it considers not only encoding and retrieval activations but also encoding and retrieval deactivations. This feature allows the model to link the contribution of brain regions to memory processes to their roles within distributed networks that consistently activate or deactivate during cognitive tasks. It has become well established in recent years that during attention-demanding cognitive tasks, a set of wide-spread regions routinely show activity increases, whereas a different set of wide-spread regions routinely show activity decreases (Binder et al., 1999, Fransson, 2005, Fox et al., 2005, Golland et al., 2008). These two sets of regions have been termed the task-positive and the task-negative network, respectively (Fox et al., 2005). The task-positive network includes lateral prefrontal cortex (PFC), dorsal parietal cortex, sensory–motor cortices, subcortical areas, and the cerebellum (Cabeza and Nyberg, 2000, Naghavi and Nyberg, 2005, Shulman et al., 1997a), whereas the task-negative network, also known as the default-mode network (Raichle et al., 2001), consists of anterior and medial PFC, the precuneus, and the angular gyrus (Shulman et al., 1997b, Gusnard and Raichle, 2001). There is increasing evidence that the task-positive network is activated for processing externally presented information, including stimulus processing, task-execution, and monitoring external environments (e.g., Cabeza and Nyberg, 2000, Naghavi and Nyberg, 2005, Golland et al., 2008), whereas the task-negative network is activated (or less deactivated) for processing internally generated information, including self-referential processing, task-unrelated thoughts, and theory of mind (e.g., Gusnard et al., 2001, Fransson, 2006, McKiernan et al., 2006, Mason et al., 2007, Buckner et al., 2008). In short, externally-oriented processing is associated with increased activity in the task-positive network and decreased activity in the task-negative network, whereas internally-oriented processing is associated with increased activity in the task-negative network and decreased activity in the task-positive network. This opposing (i.e., anticorrelated) relationship between the networks can be also observed in spontaneous low-frequency fluctuations of blood oxygenation level-dependent signal (Fransson, 2005, Fox et al., 2005, Golland et al., 2008). On the basis of hypothesized associations of the task-positive and task-negative networks with encoding and retrieval activity, we made 3 specific predictions regarding study-test overlap (see below).

To directly compare brain activity during study and test phases, we performed an fMRI study that included both encoding and retrieval tasks. In each encoding trial of the present study, subjects studied a “mini word list” comprising 4 instances (e.g., horse, chicken, sheep, goat) of a semantic category (e.g., farm animal). At test, they performed an old/new recognition test with confidence ratings that included studied words as well as unrelated, new words. Previous studies (Otten and Rugg, 2001, Wagner and Davachi, 2001, Daselaar et al., 2004) have shown that some activity at study phase is positively correlated with subsequent remembering (i.e., remembered > forgotten), whereas other activity is negatively correlated (i.e., forgotten > remembered), and attention must be given to both effects to fully account for subsequent memory effects. Thus, using the subsequent memory procedure we coded the two forms of activity in the study phase: (i) activity positively correlated with subsequent hit rates, or encoding success activity; (ii) activity negatively correlated with subsequent hit rates, or encoding failure activity. Activity elicited during test phase includes not only processes associated with successful retrieval (i.e., old > new), but also processes associated with detection of novel information (i.e., new > old; Tulving et al., 1996, Buckner et al., 2001). Both behavioral and neuroimaging evidence indicates that novelty is an important determinant of memory processing (Tulving and Kroll, 1995, Ranganath and Rainer, 2003). Thus, we also coded two forms of activity in the test phase: (i) greater activity for high-confidence hits than for high-confidence correct rejections, or retrieval success activity; (ii) greater activity for high-confidence correct rejections than for high-confidence hits, or novelty detection activity. We used high-confidence responses in the contrasts because we were interested in recollection rather than familiarity.

Our model linking activity during study and test phases to the task-positive and task-negative networks consist of four hypotheses, which correspond to the four cells of Table 1. We assume that during the study phase, encoding stimulus information into long-term memory is supported by externally-oriented attention to the study item, but interfered by internally-oriented processing, such as stimulus-independent thoughts (McKiernan et al., 2006, Mason et al., 2007, Uncapher and Wagner, 2009). In other words, activity in the task-positive network facilitates encoding by directing attention appropriately to the word list, whereas activity in the task-negative network interferes with encoding by taking processing resources away from the word list. Thus, we hypothesize that encoding success activity involves mainly the task-positive network, whereas encoding failure activity involves mainly the task-negative network. Additionally, we assume that during the test phase, vivid remembering of specific contextual details, or recollection, redirects attention to internal mnemonic associations, whereas detection of novel information is supported by externally-oriented attention. In other words, activity in the task-positive network facilitates novelty detection by directing attention externally to the word cue, whereas activity in the task-negative network is associated with internal re-experience of episodes associated with the word cue. Thus, we hypothesize that retrieval success activity involves mainly the task-negative network, whereas novelty detection activity involves mainly the task-positive network. The hypothetical associations of encoding success and novelty detection activity with the task-positive network and encoding failure and retrieval success activity with the task-negative network are meant to be relative rather than absolute. This is particularly true for retrieval success activity, because access to the internal mnemonic associations is dependent upon externally-oriented attention toward the memory cue. Thus, retrieval success activity should include at least some subcomponents of the task-positive network, possibly centered in sensory/perceptual region.

On the basis of our model (i.e., the 2 × 2 matrix in Table 1), we made the following 3 predictions regarding study-test overlap. First, we predicted relatively high level of overlap between encoding success and novelty detection activity involving the task-positive network. Second, we predicted relatively high level of overlap between encoding failure and retrieval success activity involving the task-negative network. Third, we predicted relatively low level of overlap between encoding success and retrieval success activity, as well as between encoding failure and novelty detection activity. In other words, we predict that overlaps between study-phase and test-phase activity will occur along the columns of the matrix (the task-positive and task-negative networks) rather than along the diagonals of the matrix.