- Top of page
Summary: Purpose: Previous research suggests that the hippocampus is modulated both by stimulus novelty and by the extent to which relational processing (formation of associations) occurs during episodic encoding. The aim of this study was to compare hippocampal activation patterns measured by functional magnetic resonance imaging (fMRI) during encoding protocols emphasizing either novelty or relational processing.
Methods:fMRI was performed on 32 healthy volunteers while they encoded complex visual scenes or unrecognizable scrambled versions of the same scenes. In the Novelty contrast, encoding of novel scenes was compared with encoding of a repeated pair of scenes. In the Relational Processing contrast, semantic encoding of novel scenes was compared with structural encoding of scrambled scenes.
Results: Both protocols elicited bilateral hippocampal activation. Overall mean activation values were similar for the two protocols, but the Relational Processing protocol resulted in a larger volume of hippocampal activation. The pattern of activation along the longitudinal hippocampal axis differed for the two protocols. The Novelty contrast produced stronger activation in the posterior hippocampus, whereas the Relational Processing contrast produced stronger activation in the anterior hippocampus.
Conclusions: Hippocampal activation is determined by both stimulus novelty and degree of relational processing during encoding. Given the importance of anterior hippocampal pathology in temporal lobe epilepsy, an approach emphasizing modulation of relational processing may be preferable for clinical fMRI of the medial temporal lobes.
Assessment of the morphologic and functional status of the medial temporal lobes (MTL) is a clinically important aspect of the presurgical evaluation of patients with temporal lobe epilepsy (TLE). Asymmetry of MTL anatomy and function has been used to predict the side of seizure focus (1–7), the likelihood of seizure control (3–5,8–11), and the probability of memory decline after temporal lobe resection (12–18). Techniques currently in common use for MTL assessment include EEG, structural MRI, interictal positron emission tomography (PET), single-photon emission tomography, Wada memory testing, and psychometric testing. Several initial studies using functional magnetic resonance imaging (fMRI) suggest that this technique may also provide helpful adjunctive information about MTL function in TLE (11,19–24).
In developing fMRI for this application, it will be critical to evaluate a range of alternative imaging protocols to determine an optimal approach. For the purpose of clinical applications, a procedure is needed that robustly and reliably modulates the blood oxygenation level–dependent (BOLD) signal in targeted MTL structures of clinical interest. Because the majority of temporal lobe seizure foci arise in the hippocampus (25), a prime concern should be activation of the hippocampal formation itself rather than surrounding parahippocampal areas. Investigations in normal participants using episodic memory encoding and retrieval tasks have been notable for a relative lack of observed activation in the hippocampus, yet many reports of successful hippocampal activation exist (26–39). A number of factors might account for these variable results, including the fact that the hippocampus is a relatively small structure, the visualization of which can be adversely affected by macroscopic susceptibility artifacts in fMRI (29,40). What is clear from existing data, however, is that the particular characteristics of the memory activation task and baseline state used during imaging also play a large role in determining the extent and location of hippocampal activation [for excellent reviews, see (41–44)].
Because of the complexity of episodic memory processes, the number of these potential task variables is large, yet all need to be considered in designing an optimal fMRI protocol for hippocampal activation. The hippocampus is likely to play a role in both encoding (initial memory storage) and retrieval (recall from storage) aspects of episodic memory, so one basic issue is whether an encoding or a retrieval task is preferable. Although a few studies have shown hippocampal activation during retrieval (30,32), encoding tasks have generally produced more robust results (41–44). Encoding tasks can differ along a number of dimensions, such as the type of material being encoded, the novelty or familiarity of the material, and the type of task (if any) performed during encoding. For example, several PET and fMRI studies have shown that MTL activation is left-lateralized for word stimuli and more symmetric for pictorial stimuli (28,33,45,46), probably reflecting the fact that pictures can be encoded both verbally and nonverbally. If the aim, therefore, is to design a task that produces bilateral hippocampal activation in the normal brain, a reasonable first choice might be a task involving encoding of pictures.
Stimulus novelty is an important variable that has been manipulated in a number of MTL studies. Electrophysiological studies show that the hippocampus responds more strongly to novel than to repeated stimuli (47–50), thus the contrast between novel and repeated stimuli is expected to show hippocampal activation. Although hippocampal activation has been observed in a few of the imaging studies by using a novelty contrast, the activation associated with novelty occurs more often in the posterior parahippocampus and adjacent fusiform gyrus than in the hippocampus proper (26,31,35,40,51–53). When observed, hippocampal activation is typically located in more posterior aspects of the hippocampus (26,31).
Another very important factor is the degree to which the encoding task encourages associative or “relational” processing. Recent models of the MTL propose that the hippocampus “binds” distributed cortical activity during perception, comprehension, and response to a stimulus event or “episode.” Binding of activity in these systems creates a complex, unique spatiotemporal representation of the event “configuration,” composed of the salient stimulus elements (including the environment or context in which the stimulus occurs), stored knowledge (e.g., semantic or spatial information) associated with these elements, and behavioral (including emotional) responses by the participant to the stimulus, creating a unitary representation of the episode for later retrieval from long-term memory (54–57). According to this model, hippocampal activity depends on how much co-occurring neural activity is elicited by an episode (i.e., how complex the episode is in terms of evoked sensory and associative processing). Stimuli that evoke elaborative associative processing by virtue of being recognizable and meaningful should therefore elicit greater hippocampal activation than nonsense stimuli, and tasks that require activation of such associations (e.g., conceptual, associative, or semantic tasks) should elicit stronger hippocampal activation than tasks that do not.
Support for this model of hippocampus function comes from a number of imaging studies showing stronger hippocampal activation for meaningful relative to meaningless stimuli and associative/semantic relative to nonsemantic tasks (19,33,34,36–39,58–68). Examples of task contrasts used in these studies include encoding of complex visual scenes versus unrecognizable “scrambled” versions of the same scenes, processing object pictures vs. meaningless shapes, processing words versus nonwords, performing semantic judgments versus phonologic or orthographic judgments, and learning new associations between stimuli. Although the precise location of these activations is not yet clear, results from our laboratory (19,59,68) and others (33,37–39,58,60,61,65–67) suggest greater involvement of the anterior compared to the posterior MTL. In a meta-analysis of episodic encoding studies, Schacter and Wagner (41) also suggested that MTL activations tend to be more anterior when stimulus and task contrasts emphasize differences in the degree of relational processing.
A final consideration involves the use of a “resting” or “passive” baseline in hippocampal activation studies (the latter term refers to conditions in which sensory stimuli are presented, but no task is required). Some authors have expressed the view that the conscious “resting” state may be associated with conceptual-processing, memory-encoding, and memory-retrieval processes that activate the hippocampus (28,35,64,69). Detection of hippocampal activation, according to this view, would require a baseline state that engages the participant in an active “structural” task (i.e., a task that does not involve associative or semantic processing) to “interrupt” this ongoing memory encoding and retrieval. In an important empirical test of this notion, Stark and Squire (35) demonstrated that the hippocampus and parahippocampus both show higher BOLD signals during “rest” than during active perceptual discrimination tasks. Activation of these MTL regions during encoding of pictures was detected by using the perceptual discrimination tasks as a baseline, but not when “rest” was used as a baseline.
Table 1 summarizes some of these general effects of stimulus and task factors on hippocampal activation. A consideration of these factors suggests at least two rather different task contrasts that would be expected to (and have been reported to) produce hippocampal modulation. The first of these emphasizes stimulus novelty, using a contrast between novel and repeating pictures. For both novel and repeating conditions, the stimuli are meaningful, and an associative task is required; thus the conditions differ only in terms of stimulus novelty. The second approach emphasizes relational processing, by using a contrast between associative processing of pictures and structural processing of nonsense stimuli. In both conditions, the stimuli are novel; thus the conditions differ in terms of the extent to which relational processing occurs. The aim of the current study was to compare these two approaches to hippocampal activation in the same participants, using a large enough sample to ensure stable activation patterns, and focusing specifically on activation in the hippocampus proper. Our larger goal was to assist in the eventual design of an optimal strategy for hippocampal assessment with fMRI in the presurgical evaluation of patients with TLE.
Table 1. Stimulus and task characteristics affecting hippocampal activation
| ||High, Bilateral Activation|| Low Activation|
|Stimulus Novelty|| Novel||Familiar, repeating|
|Stimulus Type|| Meaningful, pictorial||Nonsense|
|Task Type|| Associative (semantic)||Active structural|
- Top of page
In this study, we evaluated two task contrasts designed to produce modulation of the BOLD signal in the hippocampus. Although both produced robust bilateral activation, the pattern of activation differed along the anteroposterior hippocampal axis. One contrast emphasized stimulus novelty through a comparison of novel versus repeated pictures, all of which were encoded by using a semantic classification task (“Is the picture indoor or outdoor?”). Hippocampal activation for this contrast was stronger posteriorly. The second contrast emphasized relational processing through a comparison of semantically encoded meaningful pictures versus structurally encoded (“Do the two halves match?”) nonsense stimuli. As expected, identifiable pictures that could be recognized and categorized were better remembered than the spatially scrambled, unidentifiable versions of these images, suggesting more extensive encoding of the meaningful pictures. Activation for this contrast was much stronger in anterior portions of the hippocampus.
Results for the novelty contrast are very consistent with previous PET and fMRI studies showing novelty effects in the posterior MTL, particularly in the posterior parahippocampus, and in surrounding visual extrastriate regions (26,31,35,40,51–53). Novelty effects in extrastriate regions have been attributed to perceptual priming, a process by which repeated stimuli are processed more efficiently and therefore require less neural activity (73–75). In addition to perceptual priming, differences in “top-down” attentional modulation could explain some of the novelty effects, because processing of repeated stimuli (especially stimuli that are repeated many times) should require less attention than processing of novel stimuli. Our data provide further evidence that novelty also modulates activity in the hippocampus proper, although this modulation primarily involved the posterior hippocampus.
Results for the relational processing contrast also are consistent with previous imaging studies (19,33,34,36–39,58–68). In designing a contrast to elicit relational processing effects during encoding, we did not attempt to distinguish task effects from stimulus effects. Consequently, the activation differences observed in the relational processing contrast could be due to differences between the normal and scrambled images in the degree of stimulus meaningfulness, to differences in explicit processing demands between the classification (indoor/outdoor judgment) and discrimination (hemifield-matching) tasks, or to both. According to the relational processing account of hippocampal function, these manipulations should have similar effects, producing more elaborate associative processing either when the stimuli are meaningful or when the explicit task requires retrieval of meaning. If anterior hippocampus activation is dependent on such relational processing, both stimulus and task variables should modulate activity in this region. In this experiment, we manipulated both the stimuli and task to produce as strong a relational processing contrast as possible.
The neurophysiologic explanation for these differences in anterior and posterior hippocampal activation patterns is not yet clear. Some evidence from animal studies supports the general notion of functional differences along the anteroposterior (or, in rats, ventral-dorsal) longitudinal axis in the hippocampus. For example, anatomic studies in cat and monkey indicate a topographic organization of inputs to the dentate gyrus from entorhinal cortex, with anterior dentate gyrus receiving input from anteromedial regions of entorhinal cortex, and posterior dentate gyrus receiving input from posterolateral entorhinal cortex (76). Related studies suggest a relatively greater input from unimodal visual cortex to lateral entorhinal and perirhinal cortices (which have stronger projections to posterior dentate gyrus) and more widespread input from polymodal, temporoparietal, olfactory, and prefrontal areas to medial and anterior entorhinal cortex (which project more strongly to anterior dentate gyrus) (76–78). These connectivity patterns suggest a preponderance of input to posterior hippocampus from the visual system and a more widespread, multimodal input to the anterior hippocampus. Studies in both monkey and rat show that the posterior/dorsal hippocampus is more involved in encoding spatial information than is the anterior/ventral hippocampus (79–81). To our knowledge, however, no studies have yet examined hippocampal functional heterogeneity in experimental animals from the standpoint of nonspatial relational processing.
Several of the results are relevant to clinical fMRI of the medial temporal lobe. Important variables to consider in developing a clinical protocol include the robustness, the hemispheric lateralization, and the intrahemispheric pattern of hippocampal activation. Both of the protocols tested here produced robust activation in the hippocampus proper. The voxel count analysis showed a clear difference between the protocols [i.e., a larger volume of activation for the relational processing contrast (68.7 μl for relational processing vs. 29.6 μl for novelty)]. This difference did not appear in the mean voxel value analysis, suggesting that activation in the relational processing contrast tended to be more focal and of higher intensity, whereas activation in the novelty contrast tended to be of lower intensity and more evenly distributed. The voxel count analysis also showed a larger volume of activation in the left hippocampus, although this difference was relatively small and was not present in the mean voxel value analysis or evident in the group maps. This point is important because the capacity to detect a unilateral abnormality of hippocampal activation is greatly enhanced by computing an asymmetry index (11,20–22,24), which is feasible only when activation is bilateral. Finally, the choice of a relational processing contrast or a novelty contrast depends on whether the anterior or posterior hippocampus is of greater clinical interest. Pathologic, electrophysiologic, and imaging studies suggest that although all regions of the hippocampus can be affected in TLE, the anterior portions are the most consistently involved (82–85). The relational processing contrast, which produces more robust activation and is more sensitive to functioning of the anterior hippocampus, may therefore be the more sensitive protocol for detecting functional abnormalities in TLE.
It should be emphasized that many other variables require careful study before an optimal fMRI protocol for visualizing hippocampal activation is determined. For example, it may be important to include a subsequent item-recognition test as part of the protocol to identify which items presented during scanning are later remembered and which are forgotten. This information can then be used in the analysis of the fMRI response to account for additional variance in MTL activation. This “subsequent memory effect” has been shown in numerous studies to correlate with activation in the MTL, including, in some cases, the hippocampus (27,29,31,33,36,63,86–88). Another potentially critical variable is the specific task used to elicit relational processing. The picture-classification task used here is relatively easy and does not require participants to learn new associations (although implicit formation of new associations between stimuli and context is a defining feature of all episodic encoding, including the task used here). Tasks that make greater demands on semantic retrieval or that require explicit learning of new associations might elicit even stronger hippocampal activation (33,37,39,60,65,67). Finally, although bilateral activation of the hippocampus is a desirable feature for predicting side of seizure focus and seizure outcome, protocols for eliciting material-specific, unilateral activation may be preferable for other applications (22,23).