This study provides evidence that the successful encoding of novel cross-modal associations is subserved by a specific set of brain regions, in particular, anterior portions of the hippocampal formation bilaterally and the left inferior prefrontal cortex. Furthermore, our data suggest that the degree of coordination between activity in the hippocampal formation and prefrontal cortex may contribute to the likelihood of successful subsequent associative memory.
Our findings are consistent with previous event-related fMRI studies reporting that activation in the medial temporal lobe and prefrontal cortex during encoding is predictive of subsequent memory performance (Brewer et al., 1998
; Wagner et al., 1998
; Kirchhoff et al., 2000
; Otten et al., 2001
). More specifically, our results suggest that the location of activation within the medial temporal lobe may be dependent on the specific demands of the memory task (i.e., whether the formation of a novel association is required). Several previous “block design” neuroimaging studies (Henke et al., 1999
; Schacter and Wagner, 1999
; Mitchell et al., 2000
; Yonelinas et al., 2001
, Zeineh et al. 2003
), including our own (Sperling et al., 2001
), have suggested that associative memory tasks preferentially activate anterior regions of the hippocampal formation. Zeineh et al. (2003)
recently published a block design fMRI experiment during the encoding and retrieval of face-name pairs. Their study used high resolution MR acquisition and “flat mapping techniques” to examine activation within specific subregions of the hippocampal formation. They demonstrated activation in the anterior CA fields 2 and 3, and the dentate gyrus during blocks of encoding novel face-name pairs compared to fixation. Consistent with our findings, Zeineh et al. (2003)
reported that the activation was maximal in these regions during the first of the four blocks of encoding, when the group of subjects learned the greatest number of face-name pairs. Our results using an event-related design, which allows the separation of the MR signal during encoding of face-name pairs that were subsequently remembered from those that were subsequently forgotten for each individual subject, furthermore suggests that activation of these anterior hippocampal regions is directly related to the likelihood of successfully forming these cross-modal associations.
In this study, we did not directly compare associative encoding (i.e., binding together previously unrelated items of information) to nonassociative encoding (single items). However, the location of the activation within the hippocampus for successful encoding of associative stimuli in our study is more anterior than what has now been reported in multiple studies for successful encoding of single item stimuli (Brewer et al., 1998
; Wagner et al., 1998
; Kirchhoff et al., 2000
). For example, Wagner et al. (1998)
reported that increased left posterior parahippocampal activation predicted successful memory for single words, and Brewer et al. (1998)
reported that increased right posterior parahippocampal activation predicted subsequent memory for single pictures. Kirchoff et al. (2000)
found that activation in the posterior hippocampus (28, −30, −6) and parahippocampus (−28, −37, −9) predicted subsequent memory for single pictures and single words, respectively. In our study, the analysis comparing each of the conditions (HC-Correct, LC-Correct, and Incorrect) vs. Fixation as shown in suggests that, while attempted encoding activates more posterior regions of the hippocampal formation, it is only the successful encoding of novel associations that activated the anterior portions of the hippocampal formation.
A recent event-related fMRI study by Davachi and Wagner (2002)
compared relational and item based learning of word triplets. They found greater hippocampal activation for relational processing of items, and greater parahippocampal activation for item based processing. Consistent with our results, Davachi and Wagner (2002)
also found that hippocampal activation predicted subsequent memory for items encoded with relational processing, but not with item-based processing. The relational aspect in the Davachi and Wagner (2002)
study differs somewhat from the associative encoding task in our study. They used an “elaborative” encoding condition, which required the rating of relative ordering of words according to desirability, compared to a “rote” learning condition. An additional difference is that the subsequent memory testing in the Davachi and Wagner study was performed for each individual word rather than for the triplet associations, whereas in our study, the subsequent memory test required subjects to indicate which of two names, both seen during the experiment, was associated with the face during encoding. Another study from Davachi et al. (2003)
demonstrated greater activation in the hippocampus bilaterally for words that were encoded through associative mental imagery compared to words that were read. Furthermore, greater hippocampal activation was seen during encoding when the subject subsequently remembered whether they had encoded the word during imagery or reading, suggesting that the hippocampus is involved in encoding “source” associations. Also consistent with our results is the finding of Otten et al. (2001)
, who reported greater activation in the anterior hippocampus for “semantically” encoded words that were subsequently remembered, which the authors speculated might represent the binding together of item and contextual information.
A block design fMRI study by Small et al. (2001)
examined associative vs. single item encoding for faces and names. This study reported that the encoding of faces alone activated posterior hippocampal regions, while hearing names alone activated more anterior hippocampal regions, consistent with previous studies using auditory encoding paradigms. The encoding of face-name pairs activated a unique spatial distribution, which included mid-hippocampal regions. Because the Small et al. (2001)
study was performed with a block design paradigm, it did not address the issue of whether activation was dependent on successful encoding of associations between the stimuli. Our data examining attempted encoding, using a visual presentation of face-name pairs, is consistent with the results of Small et al. (2001)
, but our data comparing successful vs. failed encoding suggests that it is the anterior hippocampal formation, which is important for creating enduring associative memories. Further study directly comparing the activation pattern during the successful and failed encoding of novel cross-modal associations to successful and failed encoding of single items should clarify whether the anterior hippocampal formation is selectively activated only during successful associative memory processes.
It is also possible, however, that the anterior-posterior differential in activation is an artifact of functional mapping with limited spatial resolution. The difference in results seen with single-item encoding versus associative encoding could represent a distinction between activation in the hippocampus proper versus activation in the posterior parahippocampus. Several of the fMRI studies using single-item encoding parahippocampal and posterior hippocampal activation have used complex scenes (Stern et al., 1996
; Brewer et al., 1998
; Kirchhoff et al., 2000
). The concept of a specific “parahippocampal place area” has thus been suggested (Epstein et al., 1999
). This explanation is unlikely to fully explain the anterior-posterior distinction in all of these studies, because posterior parahippocampal activation has also been reported with single-item word encoding (Wagner et al., 1998
; Fernandez et al., 1998
). It is also possible that, in our study, amygdala activation is contributing to the activation at the border of the anterior hippocampus. It is difficult to accurately distinguish the anterior hippocampus, entorhinal cortex, and amygdala using relative thick slice acquisition parameters (5 mm, skip 1 mm) and additional “smoothing” (8 mm) in the fMRI analyses. However, in a previous study using a similar face-name association task with block design, we anatomically defined the hippocampus, parahippocampus, and amygdala using each individual subject's anatomy, and the activation was clearly centered in the anterior hippocampus (Sperling et al., 2002
). In the current study, the activation peaks were also centered in the anterior hippocampus, although we cannot rule out some contribution from amygdalar activation, which is also heavily connected with the entorhinal cortex (Van Hoesen, 1982
An alternative theory for the role of the hippocampus that has been suggested by several neuroimaging studies (Tulving et al., 1994
; Strange et al., 1999
) involves the detection of novel stimuli. For example, a recent subsequent memory paper by Strange et al. (2002)
reported anterior hippocampal activation, in addition to perirhinal and para-hippocampal activation, for subsequent memory of word lists. In their study, the anterior hippocampal activation was found to predict subsequent memory only for the initial words on each list, which the authors postulated might represent a “primacy” or novelty effect. Our findings in mid- and posterior hippocampal regions are somewhat supportive of the novelty detection theory, since all of the face-name pairs were novel to the subjects and showed a similar increase in the MR signal in these more posterior hippocampal regions compared to baseline. However, novelty detection is unlikely to fully explain our results in the anterior hippocampal formation, as it was only the face-name pairs that were subsequently remembered correctly with high confidence that showed significantly higher MR signal response in these anterior regions. Furthermore, each run included a very large number of face-name pairs, and the distribution of HC-Correct, LC-Correct, and Incorrect responses was independent of the presentation order of the stimuli within each run. Additionally, each subject remembered a unique subset of the face-name pairs; thus, it is unlikely that any other features of specific stimuli explain our results.
There is some neuroanatomical and clinical support for a differential role of the anterior and posterior regions of the hippocampal formation in associative memory function. The most distinguishing features of the anterior hippocampal formation lie in its connections to neocortex (for review, see Moser and Moser, 1998
). The primary input from multimodal association neocortical areas into the hippocampus occurs anteriorly, via the entorhinal cortex, and is known as the perforant pathway (Van Hoesen, 1982
). The perforant pathway is thought to be of critical importance in memory function, and disruption of this pathway, as is known to occur early in the pathophysiologic progression of Alzheimer's disease (AD), produces profound memory impairment (Hyman et al., 1984
). Of particular relevance to this study is the finding that AD patients exhibit significant deficits in associative memory function very early in the clinical course of their disease (Morris et al., 1991
; Fowler et al., 2002
), and difficulty remembering proper names continues to be the most common complaint of older individuals visiting memory clinics (Zelinski and Gilewski, 1988
; Leirer et al., 1990
Our findings are also consistent with evidence from both functional imaging and animal studies suggesting that the longitudinal axis of the hippocampus may function as a series of interconnected “memory circuits,” a feature that may be particularly important for associative encoding (Small, 2002
; Nakazawa et al., 2002
). In this context, activation of the anterior hippocampus may be particularly important when it occurs simultaneously with activation of other hippocampal regions, thus effectively activating the entire longitudinal axis of the hippocampus. Our data comparing each of the conditions (HC-Correct, LC-Correct, and Incorrect vs. Fixation as shown in ) suggest that while activation in mid- and posterior hippocampus occurs during “attempted” encoding, it may be the addition of the anterior hippocampal activation, which completes the “circuit” along the longitudinal axis of the hippocampus, that might be important for creating successful associations.
In addition to activating anterior regions of the hippocampal formation, successful associative encoding also produced significant activation in the left inferior prefrontal cortex. This finding is consistent with multiple previous subsequent memory event-related designs examining non-associative encoding that have reported that activation in left inferior prefrontal cortex predicts the likelihood of subsequent memory (Brewer et al., 1998
; Wagner et al., 1998
; Kirchhoff et al., 2000
; Otten et al., 2001
; Buckner et al., 2001
). It remains unclear what the precise role the prefrontal cortex plays during encoding, but this region is likely important for modulating complex attentional networks (Rees and Lavie, 2001
; Cabeza et al., 2003
). We did not observe any differences in the reaction time during the encoding task between face-name pairs that were subsequently remembered vs. forgotten, but it is possible that the increased left inferior prefrontal cortex activation reflects increased vigilance during the encoding of stimuli that will be later remembered. We also found that a small region in the left inferior prefrontal cortex was activated in highly confident responses for both correct and incorrect answers. Previous studies have suggested that specific regions in the left pre-frontal cortex may be involved in the subjective experience accompanying memory formation and retrieval (Henson et al., 1999
; Dobbins et al., 2002
). Interestingly, although there was significant activation within the hippocampal formation for HC-Correct vs. LC-Correct contrast, there was no activation within the hippocampal formation for the HC-Incorrect vs. LC-Incorrect contrast. This finding suggests that the primary role of the anterior hippocampus in successful associative memory formation may be in accurately “binding” together items of information rather than in imparting the subjective feeling of confidence for subsequent memory.
As one of the unique attributes of the anterior hippocampal formation is its strong neuroanatomical connectivity with neocortex, we hypothesized that successful associative encoding may depend not only on activation of the anterior hippocampal formation, but also on the synchronization of neural activity with neocortical regions. We used random effects functional connectivity analyses to examine which other brain regions showed evidence of correlated MR activity with the hippocampal formation. These analyses demonstrated that activity in the hippocampal formation was highly correlated with activity in several neocortical regions during successful encoding but not in attempted encoding. Interestingly, only a subset of the regions from the main successful vs. failed encoding contrast showed evidence of significant functional connectivity, namely the left and right hippocampal peaks. In addition, the functional connectivity analyses revealed evidence of correlated activity between the hippocampal formation and additional neocortical regions that were not significantly activated in the main contrast. In successful encoding, the regions showing significant positive correlations with the hippocampus included the fusiform cortices bilaterally, a region which is known to be important for the processing of human faces in the right hemisphere (Kanwisher et al., 1997
) and in the successful encoding of faces (Kuskowski and Pardo, 1999
). A similar fusiform region in the left hemisphere has also been shown to activate with processing of famous names (Tempini et al., 1998
), as well as in other naming paradigms (Martin et al., 1996
). A recent PET study of learning face-name associations also demonstrated that fusiform and neighboring inferior temporal cortices were activated during successful encoding (Herholz et al., 2001
). Our previous block design studies have also indicated that the anterior regions of the fusiform in particular are involved in the encoding of novel face-name associations (Sperling et al., 2002
). The functional connectivity analyses in this study furthermore suggest that correlated activity between the hippocampus and fusiform cortices may be particularly important for successful encoding of face-name associations.
The hippocampal formation bilaterally also showed evidence of significantly correlated activity with the prefrontal cortex during successful encoding. The prefrontal cortex is known to have both afferent and efferent projections from the hippocampal formation (Goldman-Rakic et al., 1984
; Insausti et al., 1987
), including specific prefrontal connections projections to the anterior hippocampal formation (Barbas and Blatt, 1995
). Interestingly, although only the left inferior prefrontal cortex was significantly activated in the main successful vs. failed encoding contrast, both left and right prefrontal cortices demonstrated significant correlated activity with the hippocampal formation. Our connectivity analyses suggest that, in addition to activation of these regions, the degree of coordination in the activity between the hippocampal formation and neocortical regions may predict successful memory formation. New multivariate methods for image analyses that are being adapted for functional MRI, such as singular value decomposition, may prove particularly useful in investigating the pattern of co-variance between these regions.
Although the prefrontal activation was lateralized to the left during the main successful vs. failed encoding contrast, we observed activation that was quite symmetric in the left and right anterior hippocampal formation in the main contrast. This finding may be related to the fact that our face-name association task requires integration across both verbal and visual domains, since previous studies have suggested that there may be lateralization of hippocampal activation in verbal vs. nonverbal memory tasks (Kelley et al., 1998
). In addition, the functional connectivity analyses suggested that activity in the left and right anterior hippocampal formations was strongly intercorrelated during successful encoding, and both left and right anterior hippocampal formations showed correlated activity with inferior prefrontal cortices bilaterally during successful encoding. It is unknown whether the coordination of right and left hippocampal activity is mediated through some other neocortical region, but one might speculate that it is the prefrontal cortex that serves in this capacity.