Spontaneous intrinsic neural activity represents a significant part of the brain's activity dynamics (Fox and Raichle 2007
) and has been studied extensively in both humans and animals across a broad range of temporal and spatial domains using various neuroimaging modalities (for review, see Buzsaki and Draguhn 2004
). A particularly robust example of spontaneous activity involves the low-frequency (LF: <0.1 Hz) fluctuations that can be measured indirectly using blood oxygen level–dependent (BOLD) functional magnetic resonance imaging
(fMRI). Initially observed during passive “rest states” (Biswal et al. 1995
), these LF-BOLD fluctuations are correlated among functionally related brain regions.
However, despite the recent expansion of research on the topic, the nature and function of LF-BOLD correlations remains unclear (Fox and Raichle 2007
; Bandettini 2009
; Van Dijk et al. 2009
). They persist during a variety of rest states (Fox et al. 2005
; Fair et al. 2007
), task states (Arfanakis et al. 2000
; Greicius et al. 2004
; Fransson 2006
; Buckner et al. 2009
), sleep (Fukunaga et al. 2006
; Dang-Vu et al. 2008
; Horovitz et al. 2008
), under anesthesia (Kiviniemi et al. 2005
; Peltier et al. 2005
; Greicius et al. 2008
), and during various other altered states of consciousness (Boly et al. 2008
). The persistence of these correlation patterns across various states suggests that they reflect, in part, intrinsic properties of neuroanatomical networks. Furthermore, recent analysis of connectivity measured using diffusion-based magnetic resonance imaging
(MRI) techniques suggests that LF-BOLD correlations are constrained by anatomic connectivity but are more pervasive, reflecting polysynaptic projections and common driving inputs (Greicius et al. 2009
; Honey et al. 2009
; for review, see Van Dijk et al. 2009
Two lines of evidence would support the hypothesis that LF-BOLD correlations during rest are involved in learning and/or memory consolidation. First, performance of a given task should modulate the pattern of LF-BOLD correlations during subsequent rest periods, specifically within brain regions or networks that are functionally relevant to the task recently performed. Second, the magnitude of observed changes of the FC of these regions could predict subsequent behavior, such as performance on the same or a related task (e.g., improved motor performance, improved subsequent memory, etc.).
An early study that compared LF-BOLD correlations during prolonged (>5 min) periods of rest before and after a simple language task (orthographic lexical retrieval) showed differences in 6 individual subjects but very modest consistent effects at the group level, providing tentative evidence of increased FC within a purported language network following a language task (Waites et al. 2005
). However, a more recent study argued that task induced changes in LF-BOLD correlations during subsequent rest occur exclusively following learning (Albert et al. 2009
). Using probabilistic independent components analysis, Albert et al. (2009)
identified 2 resting-state networks—a frontoparietal network and a cerebellar network, thought to be involved in motor performance—which showed increased component strength following a novel motor learning task but not following a simple motor performance task. The authors stressed that because changes in resting-state activity occurred only following the more difficult motor learning task, this modulation was attributable to learning per se. However, the possibility cannot be ruled out that the observed differences in resting-state activity could be attributed to the increased difficulty or attentional demand of the novel motor learning task relative to the simple motor performance task. Further, despite evident learning demonstrated by the participants during the motor learning task, analyses of potential brain-behavior correlations failed to show any relationship between LF-BOLD activity at rest and previous or subsequent task performance across a number of performance measures.
While the latter study provided evidence to support the hypothesis that LF-BOLD correlations may reflect learning, by satisfying the first of 2 lines of evidence outlined above, a direct link with measures of actual performance was not observed. Indeed, while recent evidence suggests that FC during task performance is related to individual differences in subsequent memory (Hasson et al. 2009
) and intrinsic BOLD fluctuations interact with ongoing task performance (Fox et al. 2007
), to date there is no direct evidence that modulation of LF-BOLD fluctuations during sustained periods of “rest” is predictive of subsequent memory or cognition (but see Lewis et al. 2009
for effects on perceptual learning). Further, the task that presumably modulated subsequent LF-BOLD correlations in the latter study (Hasson et al. 2009
) was a passive listening task without any active learning component. Thus, it seems plausible that modulation of LF-BOLD correlations during rest may occur automatically within networks previously engaged during prior task performance, akin to a passive “echo” or “ripple effect” within recently coactivated brain regions. Given these competing hypotheses, questions that remain are 1) whether resting-state LF-BOLD correlations among brain regions engaged by a particular task can be modulated by simple prior task performance per se (Waites et al. 2005
; Hasson et al. 2009
), or whether these modulations are directly related to a prior learning episode (Albert et al. 2009
; Lewis et al. 2009
); and 2) to what extent modulations of LF-BOLD activity during rest might impact subsequent memory or cognition.
Here, we explored the possibility that LF-BOLD correlations during sustained periods of rest are modulated by recent experience in functionally relevant brain regions by varying the content of stimulus exposure and examining the dynamics of activity in subsequent rest states. We specifically probed the activity of functionally well-characterized brain regions known to be preferentially associated with processing of specific categories of visual stimuli (faces and complex scenes). We considered it to be an open question as to whether LF-BOLD correlations are involved in mnemonic processes, but we reasoned that if modulation of LF-BOLD correlations is directly related to learning, then this modulation should be related to subsequent performance or cognition in some way.
Exposure to particular categories of visual stimuli has long been known to differentially engage dissociable “category-preferential” regions within the human ventral visual cortex, for example, a scene-preferential (SP) region in bilateral parahippocampal cortex (PHC) for complex visual scenes (Epstein and Kanwisher 1998
) and a face-preferential (FP) region in the right fusiform gyrus (FG) for faces (Kanwisher et al. 1997
). Furthermore, processing these categories of visual stimuli involves functional interactions between nodes of distributed networks of brain regions, depending on the type of cognitive task being performed (Haxby et al. 2000
; Epstein et al. 2007
). Previous studies have shown that during a memory task involving faces and scenes, top-down modulation of category-preferential visual regions associated with memory performance involves not only the enhancement of “category-relevant” visual regions but also the simultaneous suppression of “category-irrelevant” visual regions (Gazzaley, Cooney, McEvoy, et al. 2005
; Gazzaley, Cooney, Rissman, and D'Esposito 2005
); this top-down modulation involves long-range functional interactions with prefrontal brain regions (Gazzaley et al. 2007
). Interestingly, age-related memory deficits may be attributable to a failure to suppress activity in category-irrelevant visual regions, despite preserved enhancement of activity in category-relevant regions (Gazzaley, Cooney, Rissman, and D'Esposito 2005
A significant body of research highlights the importance of long-range functional interactions between category-preferential visual regions and the right inferior frontal gyrus (rIFG) in particular. A recent study of patients suffering from congenital prosopagnosia showed that disruption of long-range structural connectivity between the rIFG and posterior FP visual cortex in the right FG underlies impaired face processing (Thomas et al. 2009
). Indeed, evidence of an important role of the rIFG in processing of both faces and scenes across a range of cognitive tasks is abundant. For example, a number of fMRI studies have demonstrated that the rIFG is involved in both encoding and retrieval of nonverbal stimuli including faces and visuospatial scenes specifically (e.g., Wagner et al. 1998
; Rajah et al. 1999
; Golby et al. 2001
; Wig et al. 2004
). Thus, regions such as the rIFG that play an important role in both face and scene processing may be candidate regions likely to show modulation of LF-BOLD correlations with FP and SP visual regions during rest, depending on the stimulus category previously experienced.
We hypothesized that if intrinsic activity events captured by resting-state LF-BOLD fluctuations play a role in memory consolidation or preparation for future behavior, then recent stimulus exposure should modulate patterns of LF-BOLD correlations among functionally relevant brain regions during subsequent rest periods. Furthermore, if modulation of LF-BOLD correlations is related to learning per se, rather than being an automatic or passive consequence of previous task performance, then variation in these fluctuations during rest should predict subsequent memory performance. To test our hypotheses, we used FC analyses to compare fMRI LF-BOLD fluctuations (<0.08 Hz) of category-preferential visual brain regions (FP and SP regions) during sustained periods of rest following 2 distinct cognitive tasks using different categories of visual stimuli—faces and complex scenes. We then assessed the degree to which the magnitude of LF-BOLD modulation predicted subsequent memory for these visual images.